Selection of carbonic anhydrase variants displayed on phage. Aromatic residues in zinc binding site enhance metal affinity and equilibration kinetics.

In all metalloenzymes, hydrophobic residues surround the metal binding site. In carbonic anhydrase II (CAII) residues Phe93, Phe95, and Trp97 flank two of the three histidines that coordinate zinc to form a hydrophobic cluster beneath the zinc binding site. A library of CAII variants differing in these hydrophobic amino acids was prepared using cassette mutagenesis, then displayed on filamentous phage, and screened for proteins retaining high zinc affinity. Wild-type CAII was enriched 20-fold by selection, and consensus residues at each position were identified from the enriched CAII variants (Ile, Phe, Leu, and Met at position 93; Ile, Leu, and Met at position 95; and Trp and Val at position 97). Highly selected variants have zinc affinity and catalytic activity nearly equal to that of wild-type CAII, indicating that the aromatic residues are not absolutely essential. However, the zinc dissociation rate constant and catalytic activity of the variants correlate with the volume of the amino acids at positions 93, 95, and 97. In summary, metalloenzyme variants displayed on phage can be selected on the basis of metal affinity; such methods will be useful for optimization of metal ion biosensors.

Understanding the structural determinants of metal ion affinity in proteins is an important step in the design of protein metal sites. Such sites may be used to stabilize proteins, to regulate the activity of proteins, or for use in biosensors to quantify trace metal ions (see reviews in Refs. [1][2][3]. Because of its high affinity and specificity for zinc, the His 3 metal polyhedron of carbonic anhydrase (CAII) 1 often has been used as a model for designing metal sites in existing proteins (4 -6) and in de novo proteins such as the minibody (7) and four helical bundle protein (8,9). While the protein metal ligands of CAII have been incorporated into these proteins with coordination geometry very similar to that observed in CAII, these designed metal sites lack the zinc avidity and catalytic activity of biological zinc sites, suggesting that further protein structural factors in CAII contribute to the metal affinity and reactivity of this enzyme.
To probe the relationship between protein structure and metal ion affinity and specificity, many of the conserved structural features near the zinc binding site of CAII have been investigated (review see Ref. 10). The refined x-ray crystal structure of CAII (11) reveals that the zinc ion is bound at the bottom of a deep active site cleft, coordinated by protein residues His 94 , His 96 , and His 119 . At physiological pH, a solvent hydroxide ion completes the tetrahedral coordination geometry of the bound metal. The functional and structural effects of altering the protein zinc ligands have been examined (12)(13)(14); substitution of any one of the three histidine ligands by alanine decreases the affinity for zinc by 10 4 -fold. Additionally, the direct zinc ligands are fully saturated by hydrogen bond networks with second shell residues as follows: His 94 donates a hydrogen bond to the carboxamide side chain of Gln 92 ; His 119 donates a hydrogen bond to the carboxylate side chain of Glu 117 ; His 96 donates a hydrogen bond to the backbone carbonyl oxygen of Asn 244 ; and zinc-bound hydroxide donates a hydrogen bond to the hydroxyl side chain of Thr 199 (11). This conserved network of residues that form hydrogen bonds with the His ligands enhances zinc affinity (15)(16)(17); the Q92A and E117A substitutions both decrease zinc affinity 5-10-fold and increase the zinc dissociation rate constant. Therefore, the protein structure surrounding the zinc binding site of CAII is finely tuned for high zinc affinity and slow rate constants of metal dissociation.
A further structural feature observed in all metalloproteins is that the hydrophilic direct metal ligands are embedded within a larger shell of hydrophobic groups (18). In CAII, residues Phe 93 , Phe 95 , and Trp 97 flank the zinc ligands His 94 and His 96 in a ␤-strand structure (Fig. 1), and this aromatic cluster forms packing interactions with the hydrophobic core of the protein, as visualized by x-ray crystallography (11). These bulky aromatic residues may increase metal binding and/or reactivity by sterically restricting the flexibility of the site, controlling the range of movement allowed for the residues coordinating the metal ion and affecting both K D and k off . Alternatively, the hydrophobicity of these amino acids may be important for providing an interior region with low dielectric constant to enhance electrostatic interactions (18). In this study we demonstrate that this conserved hydrophobic region is essential for increasing zinc affinity and decreasing the rate constant for zinc dissociation but has modest importance for the reactivity of the zinc-bound hydroxide. Furthermore, a correlation between the zinc dissociation rate constants and the volume of the substituted amino acid suggests that these bulky residues function mainly by decreasing the mobility of the site.
A second goal of this work is to explore the utility of phage display technology as a means of selecting metalloenzyme vari-ants on the basis of metal affinity. Large libraries (ϳ10 9 ; for review, see Ref. 19) of variants can be screened using phage display; since the genetic information coding the selected protein is present in the attached phage particle, very rare variants can be identified through amplification and DNA sequencing. Phage display has proven to be a very useful tool for the in vitro development of antibodies (20). Furthermore, several enzymes, including ␤-lactamase (21), trypsin (22), and alkaline phosphatase (23), have been expressed on the surface of phage and screened on the basis of their affinity for various ligands. Recently, we have demonstrated that CAII can be expressed as a fusion protein on the surface of filamentous phage in a correctly folded, active form and enriched using the affinity of the holo-enzyme for sulfonamides. 2 In this study, we screen a library of CAII variants displayed on phage based on their affinity for a catalytic zinc ion and demonstrate that this method is very useful for probing the structural determinants of metal affinity and for preparing CAII variants with altered metal affinity and specificity.

EXPERIMENTAL PROCEDURES
Mutagenesis of pCANTAB-CA Phage Display Vector-In the plasmid pCANTAB-CA, the CAII gene is inserted 5Ј to the gene coding for the M13 minor coat protein (g3p) to create a CAII-g3p fusion protein. In this vector, the amino acid sequence Ala-Ala-Gln-Leu is inserted between the g3p signal sequence and the N terminus of CAII. At the junction between the C terminus of CAII and the N terminus of g3p, the stop codon of CAII and the lysine immediately preceding it are replaced by three Ala residues (Pharmacia Biotech Inc.). 2 To facilitate cassette mutagenesis, silent mutations were added to the CAII gene producing unique KpnI and MscI sites at codons 86/87 (ggcact-ggtacc) and 102 (tggaca-tggcca), respectively, to form pCANTAB-CAII. To produce the library of CAII variants, a mutagenic cassette flanked by KpnI and MscI sites was prepared by annealing synthetic oligonucleotides composed of the coding and non-coding strands of CAII DNA from codon 87 to codon 102. This cassette contained the nucleotide sequence DBK (D ϭ G/A/T; B ϭ G/T/C; K ϭ G/T) at residues 93, 95, and 97 to generate a library of 18 3 , or 5832, DNA sequences that encode 1728 different CAII variants. Both the pCANTAB-CAII vector and the synthetic oligonucleotide were digested with the restriction enzymes KpnI and MscI, and then the cassette was ligated to the large KpnI/MscI fragment of pCANTAB-CAII using standard techniques (25,26). To ensure that affinity enrichments were not due to contamination of the library or loss of rare variants, duplicate libraries were prepared, transformed, and screened separately.
Production of CA Phage-The library of pCANTAB-CAII plasmids was transformed into XL1-Blue cells (26), and colonies were grown for 20 h at 30°C on SOB/agar plates (26) containing 100 g/ml ampicillin, 12.5 g/ml tetracycline, and 2% glucose to estimate the number of individual transformants and to prevent growth competition among the variants. In the first preparation of CAII phage library, 3,000 colonies were obtained, and 14,000 colonies were produced in a duplicate library. All of the individual transformants from both libraries were combined, suspended in 2 ϫ yeast tryptone media (26) containing 100 g/ml ampicillin, 12.5 g/ml tetracycline, and 2% glucose and grown in shaker flasks at 37°C for two doublings to reach an A 600 of 0.3-0.4 before superinfection by the helper phage M13KO7 (27) at a multiplicity of infection of 100 infective colony-forming units (cfu) per cell (as estimated by A 600 of 0.4 Ϸ10 8 cells/ml). After shaking at 100 rpm for 1 h, cells were transferred to glucose-free Luria Broth (LB) media containing 100 g/ml ampicillin, 12.5 g/ml tetracycline, and 50 g/ml kanamycin and shaken at 300 rpm, 37°C, for 5-6 h to produce phagemid particles. The solution was centrifuged (16,000 ϫ g, 30 min) to remove cells. The phagemid in the supernatant was precipitated by polyethylene glycol using standard procedures (27) and resuspended in a 1/200 volume of 10 mM Tris, pH 7, 1 mM EDTA buffer (TE). The concentration of phagemid particles was estimated by infecting log-phase XL1-Blue cells with serial dilutions of the phagemid solution and plated on LB agar containing 100 g/ml ampicillin to quantify colony-forming units. Typical phagemid yields were 2-3 ϫ 10 8 cfu/ml original growth volume. DNA from individual phagemid colonies was isolated by the alkaline lysis method (26) and sequenced by the dideoxy method (28) to deter-mine the diversity of the variant CA phage library.
Preparation of Affinity Resin and Determination of Binding Enrichments-p-Aminomethyl benzenesulfonamide was coupled to Ultralink Carboxy resin (Pierce) using carbodiimide as described previously (29). 2 Affinity resin was pre-blocked by incubation with 10 mg/ml bovine serum albumin in 2 volumes of wash buffer (0.1 M Tris-SO 4 , pH 7.0, 0.5% Tween 20) for 1 h at 22°C and then washed 4 -5 times with 2-3 volumes of wash buffer to remove any excess bovine serum albumin. For determination of the binding enrichment, solutions of CA phagemid (10 9 -10 12 cfu) were added to 400 l of resin in a final volume of 1 ml of wash buffer containing 1 M ZnCl 2 and incubated overnight at room temperature with rotary mixing at 7 rpm. The eluate was collected by centrifugation of the resin in a Spin-X filter tube (Costar) (1-2 min, 2,000 ϫ g). The resin was resuspended in wash buffer and incubated for about 5 h. After 4 -5 washes, CA phagemid bound to the sulfonamide resin were eluted by incubation for 16 -24 h with wash buffer containing either 10 mM acetazolamide or varying concentrations of the metal ion chelator, dipicolinate (DPA). To amplify eluted phagemid for a second round of affinity selection, XL1-blue cells were infected with 10 3 -10 5 cfu of phage and plated onto LB agar containing 100 g/ml ampicillin. Fifty colonies were randomly picked; their DNA was isolated (26) and the CAII gene sequenced (28). The remainder of the transformants (about 9,000) was resuspended in liquid media and superinfected with M13KO7 phage, as described above, to generate CA phagemid.
Expression and Purification of CAII Variants-To facilitate the subcloning of CAII variants, unique KpnI and MscI sites were added to the CAII gene in the CAII expression vector pACA (30) to create the expression vector pMACA. The DNA of selected CAII variants was subcloned into pMACA by ligating the small KpnI/MscI fragment from the pCANTAB-CAII vector containing the variant sequence to the large fragment of KpnI/MscI-digested pMACA. The CAII genes of these constructed plasmids were then sequenced through both ligation sites. The plasmids encoding CAII variants were transformed into BL21(DE3) cells (31), and CAII was induced by the addition of 0.25 mM isopropyl-␤-D-thiogalactopyranoside to late log Escherichia coli BL21(DE3)pMACA cells followed by incubation at 30°C for 5 h (32). CAII variants were purified using sequential ion exchange chromatography on DEAE-Sephacel and SP-Sepharose Fast Flow, as described previously (33).
Catalytic Activity-The k cat /K M for CAII-catalyzed p-nitrophenyl acetate (pNPA) hydrolysis was measured at 25°C in either 50 mM MES (pH 5.5-7.0) or 50 mM Tris-SO 4 (pH 7.0 -9.0), 0.5 mM pNPA, 0.1-0.3 M CAII and the ionic strength maintained at 0.1 by the addition of Na 2 SO 4 . The initial rate of pNPA hydrolysis was monitored at 348 nm (⌬⑀ ϭ 5,000 M Ϫ1 cm Ϫ1 ) (34) and background rates of hydrolysis, measured in the presence of 50 M acetazolamide, were subtracted to obtain the CAII-dependent rates. For some variants, the pK a was determined by fitting the data to Equation 1 using k cat /K M values measured over the pH range 5.5 to 9. For other variants, k cat /K M was measured at pH 6.5 (k 6.5 ) and pH 8.0 (k 8 ), and the ratio of these values was used to estimate the pK a using Equation 2.
Initial rates of CO 2 hydration were measured at 25°C using a Kin-Tek stopped-flow apparatus by the changing pH indicator (35). The reaction was monitored at 578 nm in 50 mM TAPS buffer, pH 8.5, containing 28 M m-cresol purple, 0.1 mM EDTA, ionic strength adjusted to 0.1 with Na 2 SO 4 . The CO 2 concentration was varied from 3.3 to 24 mM. For those CAII variants with rapid zinc dissociation, 0.1 mM ZnSO 4 was included in the assay. Background rates measured in the absence of CAII were subtracted from the observed rates.
Zinc Dissociation Rate Constant and Dissociation Constant-All solutions were prepared in plasticware using deionized water. The rate constant for zinc dissociation was determined by diluting CAII variants (10 -20 M) into 15 mM potassium phosphate, pH 7.0, containing 35 mM EDTA at 25°C to chelate zinc dissociating from the enzyme. Dissociation of zinc from CAII was monitored by measuring the decrease in pNPA hydrolysis activity since the apoenzyme has low catalytic activity. At various times, an aliquot of CAII was diluted 100 -200-fold into assay buffer, and the esterase activity was assayed as described above. Data were fit to Equation 3 using the curve-fitting program Kaleida-Graph (Synergy Software) where A 0 and A t are the catalytic activity at dilution and at various times after dilution, respectively. The standard errors were determined from these fits.
To measure the zinc dissociation constant, apo-CAII variants were prepared using Amicon diaflow filtration (36) against first 50 mM dipicolinate (DPA), pH 7.0, and then 15 mM potassium phosphate buffer, pH 7, followed by chromatography on a PD-10 column (Pharmacia) to remove excess DPA. Apo-CAII (50 -80 M) was dialyzed against varying concentrations of zinc (0 -1 mM) in 0.2-3 mM DPA, 15 mM potassium phosphate buffer, pH 7.0, for 18 -22 h at 30°C (13). After equilibrium was achieved, the fraction of CAII containing a bound zinc ion ([E⅐Zn]) was quantified by measuring the specific pNPA hydrolysis activity, or by removing free zinc by chromatography on a PD-10 column and quantifying bound zinc using a colorimetric 4-(2-pyridylazo)resorcinol method (37). The concentration of free zinc in the dialysis buffer was calculated from the DPA-zinc stability constants at 30°C (38). The dissociation constant and standard error were calculated using the KaleidaGraph curve-fitting program with Equation 4, varying both C and K D , where C ranged from 0.9 to 1.1.

Library of CAII Variants Displayed on
Phage-To investigate the functional importance of a conserved hydrophobic cluster beneath the zinc binding site in CAII, we used cassette mutagenesis (25) to prepare a library of CAII variants with alterations in three residues (Phe 93 , Phe 95 , and Trp 97 ) flanking two of the histidines (His 94 and His 96 ) that coordinate zinc (11; Fig. 1). Rather than completely randomizing the DNA sequence of each of these codons, we chose to use a semi-random mutagenesis scheme in which the codons encoding residues 93, 95, and 97 in CAII were simultaneously replaced with a mixture of codons encoding the following amino acids: Ala, Cys, Phe, Gly, Ile, Leu, Met, Arg, Ser, Thr, Val, and Trp. This scheme was chosen because it eliminates all stop codons and most charged residues (39), which are predicted to significantly destabilize the folded structure of CAII. This method creates a library of 5832 DNA sequences encoding 1728 unique CAII variants. The most rare variant is encoded by only one sequence in this coding scheme, and the most common variant is encoded by 27 sequences. These plasmids were transformed into E. coli and a library of 17,000 individual colonies were combined, as described under "Experimental Procedures." The DNA sequence of the CAII gene from 42 randomly chosen transformants was determined to assess the accuracy of the ligation reaction and the diversity of the library; 62% of the sequences contained one correctly oriented copy of the cassette. Furthermore, little nucleotide bias in residues 93, 95, and 97 was observed (first nucleotide: 42% A, 29% T, and 29% G; second nucleotide: 32% C, 32% G, and 36% T; and third nucleotide: 51% G and 49% T). One sequence encoding wild-type CAII was also observed (3.8%). This is higher than the 0.02% predicted from the randomized sequence in the starting oligonucleotide and suggests wild-type contamination from re-ligation of wild-type pCANTAB-CA when the original library was created. In the future, this problem could be alleviated by positioning a stop codon at residue 93 to prevent production of wild-type CAII. Using these data, we estimated that the library contains 10,500 (62% of 17,000) viable transformants. Based on Poisson distribution, this gives 90% confidence that all 5,832 sequences are represented in the library (40), indicating that the starting library has high diversity.
Enrichment of CAII Variants-For the purposes of selecting variants with high zinc affinity, the CAII variants were expressed as CAII-g3p fusion proteins on the external surface of phagemid, as described by Hunt and Fierke. 2 These CAII phagemid were then fractionated by their ability to bind to the sulfonamide resin (described under "Experimental Procedures") where an essential interaction is the direct coordination of the sulfonamide nitrogen with the active site zinc in CAII (41). This procedure enriches CA phagemid 10 3 -fold compared with M13 phage 2 and can potentially fractionate CA variants based on their sulfonamide affinity, zinc affinity, and/or protein stability. Under the conditions of the selection, CA phagemid were added in great excess over affinity binding sites; however, the total number of binding sites (3-8 ϫ 10 6 ) 2 is much greater than the diversity of the variant library to ensure that multiple copies of desired variants are bound.
In the first round of screening, CA variants were eluted by the addition of a high affinity sulfonamide inhibitor, acetazolamide (42). This first step was designed to eliminate variants with greatly reduced zinc or sulfonamide affinity and/or severe defects in the structural integrity or stability. From this pool, 79 individual transformants were chosen at random, and the CAII gene was sequenced. In this selection step, wild-type CAII was greatly enriched (Ն10-fold) since 44% of the sequenced transformants (35 out of 79) contained the wild-type sequence compared with Ϸ4% wild-type sequence in the starting library. These data clearly indicate that the wild-type sequence of F 93 F 95 W 97 confers significant advantages for expression of CA-g3p fusion protein and/or binding to the sulfonamide resin under the selection conditions. Additionally, the results of the first round of selection clearly indicate that positively charged residues in these positions are discriminated against in this selection. In the original variant library, arginine is the only charged amino acid possible and it comprises 5.6% of the amino acids encoded at position 93, 95, or 97. Of the variants sequenced after this first round of selection, only two contain arginine at either position 93, 95, or 97 (Ͻ1.5%), suggesting that positioning this charged residue in the hydrophobic protein core prevents correct functioning of the protein.
CAII phage from the first selection pool were amplified for a second round of screening, incubated with the sulfonamide resin, and eluted using either acetazolamide or DPA, which catalyzes the dissociation of zinc from CAII (43). Twenty-four individual transformants eluted using acetazolamide and 56 transformants eluted using dipicolinate were chosen at random, and the CAII gene of each was sequenced ( Table I). The percentage of wild-type sequences was high (71%) but varied somewhat with the method of elution; 87% of the phagemid eluted with acetazolamide displayed wild-type CAII compared with 64% of the phagemid eluted with dipicolinate. These data further indicate the critical importance of the wild-type F 93 F 95 W 97 sequence. In addition to wild type, several variants were identified multiple times; the sequence encoding F 93 M 95 V 97 CAII was identified three times, whereas the CAII variant F 93 L 95 V 97 was observed twice in the selected pool.
Enrichment of several amino acids compared with the starting population at each randomized position is evident (Fig. 2) and the general trend is that hydrophobic residues are enriched and hydrophilic residues are not observed. The largest enrichments (Ն3-fold) are observed for Phe, Ile, Leu, and Met at position 93, Ile, Leu, and Met at position 95, and Val and Trp at position 97. The lack of enrichment of Phe at position 95 is curious since this is the wild-type amino acid, suggesting that specific contacts between the aromatic amino acids at 93, 95, and 97 are important since we observed such a large enrichment of the wild-type sequence (20-fold). Furthermore, several amino acid residues are strongly disfavored; no Arg, Cys, Ser, or Thr residues are observed in any of the randomized positions of the selected variants, and Gly and Trp are not observed at position 93; Phe and Val are absent from position 95; and Ala, Gly, and Leu are not detected at position 97. Nonetheless, these results indicate that several similar amino acid residues can be substituted at each position without great loss of protein function, including zinc affinity and catalytic activity, as demonstrated below.
Zinc Dissociation Constants of Selected CAII Variants-To evaluate the effectiveness of our selection for identifying variants with high zinc affinity and to further explore determinants of metal affinity, we used equilibrium dialysis to measure the zinc dissociation constants (K Zn ) for variants from the pre-selection pool and variants eluted after one and two rounds of selection (Fig. 3). After two rounds of screening, 20 CAII variants other than wild type were identified (Table I). The CAII variant observed three times in the pool (F 93 M 95 V 97 CAII) has the highest zinc affinity (K Zn ϭ 1.6 pM) of all of the variants examined, comparable to the wild-type CAII zinc affinity (K Zn ϭ 0.8 pM). A similar CAII variant, F 93 L 95 V 97 , that is represented twice in this pool has a slightly larger K Zn value of 2.2 pM. The zinc affinity of other CAII variants selected through two rounds of screening is uniformly high, with K Zn values ranging from 1.6 to 12 pM (Table II). The CAII variant with the weakest measured zinc affinity, M 93 M 95 V 97 , contains substitutions at each of the three randomized positions. These data clearly demonstrate that aromatic residues at these positions are not absolutely required for high zinc avidity in CAII.
The zinc dissociation constants of CAII variants selected after only one round of screening are weaker, ranging from 3.2 to 35 pM with an average zinc affinity of 22 pM (Fig. 3 and Table  II). To evaluate the range of zinc affinities of CAII variants in the original library, three variants from the pre-selection pool containing residues that appear to be disfavored in selection were also characterized ( Fig. 3 and Table II). For these CAII variants, both the zinc affinity is weaker (K Zn ϭ 30 -90 pM) and the expression levels in E. coli are decreased to Ϸ10% that of wild-type CAII, perhaps due to lower protein stability. Both of these factors likely contribute to the negative selection of these variants using the phage display technology. Nonetheless, all  Table II. of the data indicate that sulfonamide resin selection of CAII phage identifies variants with high metal affinity (Table II); however, the additional requirements that the protein be stably expressed on the surface of phage and bind sulfonamides tightly means that not all of the variants with high zinc affinity will be observed. Therefore, this technique selects proteins with both high affinity for the ligand and reasonable expression levels and protein stability. Zinc Dissociation Rate Constants of Selected CAII Variants-In wild-type CAII, zinc not only binds tightly but the dissociation rate constant is also very slow (Table II). To probe the role of residues 93, 95, and 97 in modulating the slow zinc dissociation and association rate constants, the zinc dissociation rate constant (k off ) ( Fig. 5 and Table II) of a number of variants was measured. The zinc dissociation rate constants were determined from the time-dependent disappearance of CAII-bound zinc after dilution into high concentrations of EDTA to trap the dissociated zinc. Unlike dipicolinic acid which efficiently catalyzes the dissociation of zinc from wildtype CAII (43), EDTA is believed to act only by lowering the free metal ion concentration (44). Substitutions in the aromatic residues flanking the histidine ligands have much greater effects upon the zinc dissociation rate constant than upon K Zn . The k off values for zinc in these variants range from the CAII variant I 93 M 95 W 97 with a k off comparable to that of wild-type CAII to S 93 T 95 M 97 CAII with a k off that is 1000-fold faster than that of wild-type CAII ( Fig. 4 and Table II).
From these data several generalizations can be drawn. The four variants with the fastest dissociation rate constants contain a Ser residue at one of the three randomized positions. Replacing the large aromatic residues at positions 93, 95, or 97 with smaller residues may increase the conformational mobility of the ␤-strand, allowing the adjacent histidines that coordinate zinc (His 94 and His 96 ) additional flexibility. Furthermore, the variants with the slowest dissociation rate constant retain Trp at position 97, and, regardless of the nature of amino acid substitutions at positions 93 and 95, the half-time for zinc dissociation of the isolated variants that retain Trp 97 is greater than 5 days. In a comparison of the CAII variants I 93 M 95 W 97 and I 93 M 95 V 97 , the values of K Zn are comparable, whereas the dissociation rate constant increases 10-fold for substitution of Trp 97 with Val. These data suggest that Trp 97 may have an important role in anchoring the ␤-strand to reduce the confor-mational flexibility of His 94 and His 96 required for zinc dissociation.
The association rate constants for zinc can be estimated from the measured K Zn and k off values assuming a single association step where k on ϭ k off /K Zn (Table II). Several variants in which the wild-type residue Trp 97 is retained have zinc association rate constants that are 5-10-fold slower than that of wild type CAII; substituting Ser at position 93 or 97 leads to a 10-or 60-fold increase, respectively, in the zinc association rate constant. These data further indicate that residues 93, 95, and 97 may restrict the conformational flexibility of the histidine ligands by anchoring the ␤-strand on which they are located.
Catalysis and pK a of Zinc-bound Solvent-To delineate the role of the aromatic cluster beneath the zinc binding site on the reactivity and pK a of the zinc-bound solvent, we measured the pH dependence of the second-order rate constant for pNPA hydrolysis catalyzed by the CAII variants. As observed for wild-type CAII, the pH dependence of pNPA hydrolysis of all of the variants is consistent with the ionization of a single enzymic group. The pH dependence of the esterase activity of wildtype CAII, and likely these variants as well, directly reflects ionization of the zinc-water moiety to zinc-hydroxide, which acts as a nucleophile during catalysis (for review, see Ref. 45). The pK a values determined for all of the selected CAII variants examined (Table III) are the same, within experimental error, as that of wild-type CAII (6.8 Ϯ 0.1; 46). Since the pK a of the zinc-water is indicative of the ability of the zinc to stabilize a negative charge (10), these data suggest that the substituted amino acids at positions 93, 95, and 97 do not significantly affect the electrostatic environment of the zinc binding site. However, the pH-independent k cat /K M for pNPA hydrolysis catalyzed by these CAII variants decreases 2-30-fold compared with that of wild-type CAII (Table III). Furthermore, log k cat /K M for pNPA ester hydrolysis increases roughly proportionally to the additive volume (47) or overall hydrophobicity of the substituted side chains at positions 93, 95, and 97 as measured by the transfer free energy of amino acids between octanol and water (48,49). Similar dependence of the esterase activity on hydrophobicity has previously been observed for variations in either the substrate or amino acids Val 121 , Val 143, and Leu 198 in the active site of CAII (32,46,50,51). This dependence on hydrophobicity may be due mainly to increased affinity of the ester substrate, rather than enhanced reactivity of the nucleophilic zinc-bound hydroxide.
There is considerable evidence that hydration of CO 2 by wild-type CAII consists of two main steps (for review, see Ref. 45) as shown in Equations 5 and 6: 1) nucleophilic attack of zinc-bound hydroxide on CO 2 to form enzyme-bound HCO 3 Ϫ  Fig. 3. b Measured at pH 7.0, 25°C as described in the legend of Fig. 4. c Calculated from t 1/2 ϭ ln 2/k off . d Calculated from k on ϭ k off /K Zn .  Table II. followed by product dissociation resulting in the zinc-water form of the enzyme; 2) proton transfer to solvent via a proton shuttle (His 64 ) to regenerate the zinc-hydroxide species. For wild-type CAII, k cat /K M is a combination of the association and CO 2 hydration steps (k cat /K M Ϸ k 1 as k 2 Ͼ k Ϫ1 ). At high CO 2 and buffer concentrations, intramolecular proton transfer between zinc-water and His 64 becomes rate-limiting as indicated by a solvent isotope effect of 3-4 on k cat (k cat Ϸ k 3 as k 4 Ͼ k Ϫ3 ) (52). Therefore, K M reflects this change in rate-limiting steps, not CO 2 affinity which is estimated at 0.1 M (51,53).
The steady-state kinetic parameters for CO 2 hydration, catalyzed by the CAII variants with substitutions in the aromatic cluster, were measured at pH 8.5, 25°C (Table III). For 11 of the 14 variants studied, alterations in Phe 93 , Phe 95 , and Trp 97 decrease k cat /K M for CO 2 hydration less than 3-fold. These data clearly indicate that aromatic amino acids underneath the zinc binding pocket are not essential for high catalytic activity in CAII and that the majority of substitutions in these amino acids do not significantly alter the reactivity or position of the nucleophilic zinc-bound hydroxide. However, the selection procedure may enrich variants with high catalytic activity since the variants must retain affinity for sulfonamides which are proposed as transition state analogs (54).
Similarly, small decreases in k cat were measured for these CAII variants, indicating that the rate constant for proton transfer from zinc-water to His 64 is slightly perturbed by these substitutions. However, the effect on k cat does not parallel changes in the pK a of the zinc-solvent ligand, as observed for transfer of a proton between an acid to a base in model reactions and between His 64 and bases (55)(56)(57). Therefore, the decreases in k cat are likely caused by small changes in the positioning of the active site solvent/His 64 proton transfer pathway, as has been observed for amino acid substitutions at positions near His 64 (58,59).
Interestingly, the variants with the largest increase in the zinc dissociation rate constant (Ͼ300-fold relative to wild-type CAII) also display the biggest decreases in CO 2 hydration. For CAII variants F 93 I 95 S 97 , S 93 L 95 M 97 , and T 93 S 95 V 97 , k cat /K M is decreased 9 -74-fold relative to wild-type CAII. If k cat /K M reflects the association and hydration of CO 2 for these variants as it does in the case of wild-type CAII, then substrate association and/or reactivity is hindered in these variants, possibly due to increased conformational flexibility in the active site or repositioning of the zinc-hydroxide. The decreases in k cat /K M for these variants are accompanied by large increases in K M so k cat is only 2-20-fold lower than that of wild-type CAII. These results indicate that CO 2 association and hydration are impaired in these variants while effects on proton transfer are less severe.

DISCUSSION
Effectiveness of Phage Display-Recently, phage display technology has emerged as a powerful tool for the design and re-design of proteins (19,20,60). This method of rapidly screening large protein libraries for rare variants has been applied to diverse uses such as the selection of linkers for catalytic antibodies (61) and the selection of enzymes based on substrate affinity (21,22). In this study we describe the successful adaptation of phage display methods to include the selection of metalloenzyme variants based on metal affinity.
Our results show that carbonic anhydrase variants displayed on phage can be screened on the basis of metal affinity. Using this method we have investigated the functional role of three residues, Phe 93 , Phe 95 , and Trp 97 , flanking amino acids His 94 and His 96 that coordinate zinc. This hydrophobic cluster beneath the metal binding site is highly conserved among CA isozymes (62) and is proposed to modulate the metal affinity and specificity of this protein. Conserved hydrophobic clusters observed near metal binding sites in a number of other proteins may play similar roles (18).
To investigate the role of these residues a pool of variants in which Phe 93 , Phe 95 , and Trp 97 have been semi-randomly substituted was prepared and screened for the ability to bind zinc tightly. A striking result is that after two rounds of selection and elution with sulfonamides the majority of variants (87%) contained the wild-type amino acid sequence. These data indicate that the wild-type sequence is optimized to achieve a combination of high expression levels, tight zinc affinity, and catalytic activity (as indicated by high sulfonamide affinity (54)). Additional diversity in the selected pool was achieved by using a metal chelator to elute the CA phage from the column. This is understandable since this elution method should select for properties not optimized in wild-type CAII, including decreased sulfonamide affinity and increased zinc dissociation rate constants.
Additionally, non-wild-type sequences were also selected at each substituted position, pinpointing a set of consensus residues required for tight zinc binding; variants that contain the residues most highly enriched during selection were found to have zinc dissociation constants nearly identical to that of the wild-type protein. Variants isolated after only one round of selection have, on average, a lower affinity for zinc, whereas three variants that contain residues that were not enriched by selection had the lowest zinc affinity, with K Zn values 70-fold higher than that of wild-type CAII. Overall, phage display proved to be a very powerful tool in selecting variants with zinc affinity in the top 2% of the range of zinc affinities through only two rounds of selection.
Function of Hydrophobic Cluster: Zinc Affinity-The enormous selection of the wild-type F 93 F 95 W 97 CAII sequence observed by phage display indicates that this specific sequence confers a selective advantage over any other sequence included in this library of variants. This advantage is not due solely to high zinc affinity since variants with substitutions in residues 93, 95, and 97 are able to retain high zinc affinity; in particular, the free energy for binding zinc is decreased by Յ 0.5 kcal/mol at 25°C for variants containing Phe at position 93; Met, Ile, or Leu at position 95; and Val at position 97. The overall selection indicates a higher amino acid diversity (Fig. 2), but this includes variants with a free energy of binding that is decreased up to 1.6 kcal/mol. Of the three amino acid positions varied in this study, metal affinity is influenced most by alterations at position 93. Variants that retain the wild-type Phe 93 bind zinc more tightly than any of the variants with substitutions at position 93. Additionally, substitutions at positions 95 and 97 influence zinc affinity to a lesser extent; the substitutions Phe 95 3 Thr (F 93 T 95 W 97 versus wild-type CAII) and Trp 97 3 Ser (F 93 I 95 S 97 versus F 93 I 95 V 97 ) decrease zinc affinity only 4-fold.
The selection data (Fig. 2) clearly indicate that hydrophobic amino acids are preferred at all three positions. The hydrophobicity of amino acids surrounding metal sites is proposed to decrease the dielectric constant of the region to enhance electrostatic interactions (18). To investigate the characteristics of amino acids at positions 93, 95, and 97 important for high metal affinity in CAII, we have probed the dependence of the metal affinity on the combined hydrophobicity (as indicated by the free energy of transfer between octanol and water (48,49)) and volume (47) of the amino acids at these positions. For both these parameters, a low correlation coefficient (R Յ 0.74) is observed when the K Zn values for all of the variants are included. However, if the proteins containing a Trp at position 97 are excluded, the log K Zn shows a high correlation (R Ն 0.93) with both hydrophobicity and volume (Fig. 5A). These data clearly indicate that larger and more hydrophobic amino acids in this region increase the metal affinity. On the other hand, the metal affinity of the variants containing Trp 97 show a significantly decreased correlation with either hydrophobicity (R ϭ 0.83) or volume (R ϭ 0.86) and, furthermore, the metal affinity of these variants is lower than predicted (Fig. 5A).
These data suggest that changing the size of the amino acid from Val to Trp at position 97 does not significantly enhance zinc affinity. The hydrophobicity or size of the amino acid at positions 93 and 95 may increase metal affinity mainly by decreasing the mobility of the ␤-sheet to "preorganize" the protein side chains that coordinate metal and decrease the entropy loss accompanying metal binding. Similarly, calorimetric studies indicate that the hydrogen bonds between protein side chains and the histidine ligands in CAII decrease the entropy of zinc binding. 3 Zinc Equilibration-In wild-type CAII, zinc not only binds tightly (1 pM), but the dissociation rate constant is very slow, and the observed rate constant for zinc association with CAII (k on ; estimated from the measured K Zn and k off) ) is also slower than the diffusion-controlled limit (Table II). The slow zinc association rate constant suggests a two-step mechanism for the binding of zinc to wild-type CAII where the rate of formation of the metal complex is limited by an intramolecular step, such as the dissociation of inner-sphere water molecules (64 -66). One possible mechanism for the association of zinc with wild-type CAII that is consistent with the properties of first coordination sphere and second shell variants (13,15) is the rapid formation of an initial complex in which zinc is coordinated to two protein ligands, e.g. His 94 and His 96 , followed by the slow formation of the tetrahedral complex comprised of His 94 , His 96 , His 119 , and a solvent molecule. The rate constant for this interconversion increases as the mobility of the site increases; for example, deletion of a hydrogen bond between His 119 and Glu 117 greatly increases the rate constant of zinc dissociation (15)(16)(17).
Similarly, the data in this study indicate that substitution of any one of the three aromatic residues with a small residue, such as Ser or Ala, significantly increases the rate constant of zinc dissociation (Table II). Furthermore, for all of the variants without a Trp at position 97, the association rate constant increases 1.3-10-fold more than either the dissociation rate constant or the dissociation constant, indicative of an increase in the rate-limiting intramolecular step. Finally, a high correlation between the log k off and either the additive volume (R ϭ 0.92 excluding M 93 G 95 W 97 CAII; Fig. 5B) or hydrophobicity (Refs. 48 and 49; R ϭ 0.89) of the substituted amino acid indicates that the rate constant for zinc dissociation increases as the size or hydrophobicity of the amino acids underneath the zinc binding site decreases. In this case, 15 of the 16 CAII variants fit on a single line indicating that the Trp at position 97 is likely conserved mainly to decrease k off rather than to maintain high zinc affinity. The value of k off for one variant, M 93 G 95 W 97 CAII, is slower than predicted for unknown reasons. Taken together, these data suggest that residues 93, 95, and 97 play a steric role in reducing the conformational flexibility of the ␤-strand on which they are located and that flexibility of this strand containing His 94 and His 96 is a major factor controlling the rate constant of zinc dissociation.
Catalytic Activity-Amino acid substitutions in residues 93, 95, and 97 likely do not influence the electrostatic environment of the active site of CAII, as the pK a of the zinc-water is unaltered. However, the value of k cat /K M for the esterase activity decreases as the volume of the amino acids at these positions decreases. On the other hand, k cat /K M for CO 2 hydration catalyzed by CAII variants is unaffected (Յ1.6-fold decrease) by reducing the additive volume of these three amino acids from 608 Å 3 (wild-type CAII) to 466 Å 3 . As the volume of the amino acids at positions 93, 95, and 97 shrinks further, the FIG. 5. Correlation of zinc dissociation and dissociation rate constants for CAII variants with the volume of residues in the aromatic cluster. A, the value of log K Zn (Table II)  value of k cat /K M decreases (by as much as 74-fold for T 93 S 95 V 97 CAII, volume ϭ 346 Å 3 ). This break could reflect a change in the rate-limiting step of k cat /K M from CO 2 association to CO 2 hydration, indicating that the reactivity of zinc-hydroxide does depend on the volume or hydrophobicity of these residues. Alternatively, these data could indicate that the position of zinc and zinc-hydroxide is unaffected by substitutions in the aromatic cluster until the volume of these amino acids are significantly decreased; at this point structural changes occur that alter the ground state position of the active site nucleophile and decrease the observed reactivity.
Comparison of Phage Display Selection with Natural Selection-As deduced amino acid sequences have been determined for a number of CA proteins from different species (67), it is of interest to compare the results of the directed evolution methods of this study with data from evolutionary selection. At position 93, the wild-type residue Phe was clearly the most highly enriched residue, a result that correlates well with the finding that the six variants with highest zinc affinity retained Phe 93 (Table II). Leu, Ile, and Met were also enriched at this position. Interestingly, two of these three residues (Leu and Met) occur in this position in several different CA proteins, including human CAIV, and only residues Phe, Leu, and Met are present in this position in catalytically active CA (67). In this case, phage display has selected variants that closely mimic the structures allowed through evolution. Slightly different results were observed at position 95, where residues Leu, Ile, and Met were again enriched but Phe was not observed in the non-wild-type sequences. It is not known why the wild-type residue Phe was not enriched during selection. Again, Ile and Leu occur at this position in sequences of naturally occurring CA proteins (67), although Met has not yet been observed.
Phage display enrichment at position 97 provided a very interesting result: Val was greatly enriched, the wild-type Trp was less enriched, and no other residues were selected at this position. Interestingly, in all known CA sequences Trp is found at position 97 (67). Surprisingly, of the variants characterized in this study, those containing Val at position 97 bind zinc with affinity near that of wild type, demonstrating that this residue can substitute for Trp with little loss of zinc affinity. Even variants containing Ser at position 97 bind zinc with K Zn values less than 7 pM. Furthermore, the CO 2 hydrase activity of variants containing Val 97 is also near the wild-type value. However, as discussed above, altering position 97 greatly increases (7-1000-fold) the rate constant for zinc dissociation compared with that of wild-type CAII, which may explain why Trp 97 is invariant in all known CA proteins with catalytic zinc ions. These data suggest that while Trp 97 is not essential for high zinc affinity, this particular residue may have been preserved through evolution due to its role in maintaining a slow rate constant for metal dissociation from the protein. Alternatively, the stability of CAII may be significantly influenced by the amino acid at position 97 and this may be reflected in an increased zinc dissociation rate constant. These results demonstrate that phage display can be used to dissect the specific functional roles of conserved residues as well as to prepare variants with specific functions not selected by evolution.
Optimization of Metal Ion Biosensors-The ability to predict the properties of a given protein structure is necessary for rational protein design and redesign. Correlations between the zinc binding properties and residue size or hydrophobicity in CAII are of particular interest in the development of CAIIbased biosensors for the detection of zinc or other specific metal ions. A CAII sensor that accurately measures levels of zinc in complex media has been designed (24,68). However, to expand the zinc concentration range and speed of measurement of this biosensor, CAII variants with altered zinc affinities and zinc equilibration rates would be highly useful. The high correlations between the size or hydrophobicity of residues at positions 93, 95, and 97 in CAII and the zinc affinity and dissociation rate constants suggest the possibility of designing CAII variants with known zinc binding properties that can be used in a rational fashion to optimize the properties of this biosensor.