Roles of Calcium Ions in the Activation and Activity of the Transglutaminase 3 Enzyme*

The transglutaminase 3 enzyme is widely expressed in many tissues including epithelia. We have shown previously that it can bind three Ca2+ ions, which in site one is constitutively bound, while those in sites two and three are acquired during activation and are required for activity. In particular, binding at site three opens a channel through the enzyme and exposes two tryptophan residues near the active site that are thought to be important for enzyme reaction. In this study, we have solved the structures of three more forms of this enzyme by x-ray crystallography in the presence of Ca2+ and/or Mg2+, which provide new insights on the precise contribution of each Ca2+ ion to activation and activity. First, we found that Ca2+ ion in site one can be exchanged with difficulty, and it has a binding affinity of Kd = 0.3 μm (ΔH = –6.70 ± 0.52 kcal/mol), which suggests it is important for the stabilization of the enzyme. Site two can be occupied by some lanthanides but only Ca2+ of the Group 2 family of alkali earth metals, and its occupancy are required for activity. Site three can be occupied by some lanthanides, Ca2+,orMg2+; however, when Mg2+ is present, the enzyme is inactive, and the channel is closed. Thus Ca2+ binding in both sites two and three cooperate in opening the channel. We speculate that manipulation of the channel opening could be controlled by intracellular cation levels. Together, these data have important implications for reaction mechanism of the enzyme: the opening of a channel perhaps controls access to and manipulation of substrates at the active site.

Transglutaminases (TGases) 1 are ubiquitous enzymes that are used widely in biology for many different purposes. There are nine different genes for TGases in the human (1)(2)(3)(4)(5). Typically, TGases recognize and activate a protein-bound Gln res-idue by formation of a thiol-acyl intermediate form. The recognition of a Gln residue may be highly specific, such as is apparently the case of the factor XIIIa, TGase 3, and TGase 4 enzymes, or with rather low specificity as for the TGase 2 enzyme. Next, this acyl intermediate is approached by a nucleophilic second substrate, which transfers onto the Gln residue. Commonly, the nucleophile is water, resulting in the net deamidation of a target Gln residue. If the nucleophile is the ⑀-NH 2 of a protein-bound Lys residue, an isopeptide ⑀-(␥-glutamyl)lysine cross-link is formed. As this cannot be cleaved in animal cells, controlled TGase activity thereby provides an efficient way for the formation of stable, insoluble macromolecular complexes. Other nucleophiles used include polyamines (to form mono-or bi-substituted/cross-linked adducts) or -OH groups to form ester linkages (as in the case of the membranebound TGase 1 enzyme to link epidermis-specific ceramides required for barrier function) (6). The TGase 2 enzyme can bind GTP nucleotides, and there is a reciprocal relationship between this binding and transamidation reactivity (7)(8)(9), presumably because the GTP binds in the vicinity of where cross-linking substrates must gain access to the enzyme for reaction (10). It remains to be demonstrated whether other TGase isoforms can also manipulate nucleotides.
In addition, TGase enzyme transamidation reactions require Ca 2ϩ , in both in vitro assays and in vivo (1)(2)(3)(4)(5). In the few cases measured, the Ca 2ϩ concentration required to activate an enzyme isoform (Ͼ500 M) is far higher than net intracellular Ca 2ϩ ion concentrations (about 100 nM) (11). Also, Ca 2ϩ is not required for GTP binding (7)(8)(9). Thus manipulation of intracellular Ca 2ϩ concentrations could afford an effective way to control TGase functions, including cross-linking.
Despite the evident essential role of Ca 2ϩ ions in TGase cross-linking reactions, very little is known structurally/functionally why Ca 2ϩ ions are in fact required. Recently however, we solved the structure of the zymogen and the activated form of the TGase 3 enzyme system (12,13). Whereas the zymogen constitutively acquires one ion during expression (in baculovirus), it is insufficient for activity. Upon proteolytic cleavage of a loop segment connecting the active site domain to the ␤-barrel 1 domain, the cleaved enzyme can acquire two additional Ca 2ϩ ions, and becomes fully active. These events coincide with the opening of a channel, which passes through the enzyme. Moreover, this channel exposes two tryptophan residues, which are believed to be important in the enzyme reaction mechanism (13). The channel is formed upon the opening of an existing deep cavity by movement of a ␤-strand loop (residues Gly 322 -Ser 325 ) so that Asp 324 can coordinate with the Ca 2ϩ ion that occupies the site nominally termed site three. We also showed that certain trivalent lanthanide ions can occupy sites three and/or two with retention of activity and an open channel. However, key questions remain unanswered. What are the relative contributions of the occupancy of each Ca 2ϩ ion in the two sites, and why is there a site two? Does the proteolyzed form in the absence of metal ions in sites two and three possess a channel, or partially opened channel, or no channel? Nor of course are lanthanide ions physiologic. Accordingly, here we present the structures of three more forms of the proteolyzed TGase 3 by x-ray crystallography crystallized in the absence or presence of Ca 2ϩ and/or more biologically relevant Mg 2ϩ ions. Our new data reveal that Ca 2ϩ ions are essential in sites two and three to open the channel for full enzymic activity, and that a Mg 2ϩ ion in site three cannot open the channel to activate the enzyme. Given that the intracellular concentration of Mg 2ϩ ions is several orders of magnitude higher than for Ca 2ϩ ions, these data show that the opening of a channel and acquisition of activity can be exquisitely controlled by the ionic milieu and by differential coordination of metal ions. Moreover, these data strongly imply that this channel is indeed required for enzyme reactivity.

EXPERIMENTAL PROCEDURES
Protein Expression, Crystallization, and Data Collection-Recombinant human TGase 3 zymogen was expressed in the baculovirus system and purified as described (12). To prepare the proteolyzed form (form I, see Table I), a 12-ml solution of zymogen (17 mg/ml in a buffer of 20 mM Tris-HCl (pH 8.0) containing 1 mM EDTA and 125 mM NaCl) was treated with 3 mg of dispase (Roche Applied Science) at 37°C for 30 min. However, the dispase contains significant amounts of Ca 2ϩ ions. Accordingly, to recover the proteolyzed form devoid of Ca 2ϩ ions, protein was then passed through a mono Q column with a linear gradient of 0 -500 mM NaCl and eluted at 150 mM NaCl (12). This is defined here as form II. Empirically, we found that the addition of 0.5 mM CaCl 2 , 1.5 mM MgCl 2 , and 1 mM ATP salt to form II resulted in the form III, which possessed a Ca 2ϩ ion in site two and a Mg 2ϩ ion in site three, although the ATP salt did not bind. We were unable to find crystallization conditions in which only site two or site three alone contained an ion.
Crystallization trials of form I were performed using the hanging drop vapor diffusion method by adding the nonionic detergent ␤-octyl glucoside (0.1 mM) to 100 l of protein in 20 mM Tris-HCl (pH 8.0), 1 mM EDTA, and 125 mM NaCl at a concentration of 17 mg/ml. To a fresh 2 l of protein, 2 l of precipitant solution was added and equilibrated over a well solution containing 4% (w/v) PEG 20,000 and 100 mM Tris-HCl (pH 8.5) at 21°C. The crystallization of TGase 3 form (II) (15 mg/ml) was done using 0.1 mM ␤-octyl glucoside in 100 l of protein in 20 mM Tris-HCl (pH 8.0), 1 mM EDTA, and 125 mM NaCl. To a fresh 4 l of protein, 2 l of reservoir solution was added and equilibrated over a well solution containing 4 -8% (w/v) PEG 6,000 or PEG 20,000, 100 mM Tris-HCl (pH 8.5) at 21°C. The TGase 3 form III (10 mg/ml) crystals were obtained using 0.1 mM ␤-octyl glucoside in 100 l of protein in 20 mM Tris-HCl (pH 8.0), 1 mM EDTA, 125 mM NaCl, 0.5 mM CaCl 2 , 1.5 mM MgCl 2 , and 1 mM ATP salt. To a fresh 4 l of protein, 2 l of reservoir solution was added and equilibrated over a well solution containing 100 mM NaCl, 100 mM sodium HEPES (pH 7.5), 4 -8% (w/v) PEG 4,000 at 21°C.
Thin, plate-like crystals were harvested into mother liquors containing 20% (v/v) 2-methyl-2,4-pentanediol (MPD) as the cryoprotectant and were then flash-cooled into liquid N 2 with a rayon mounting loop (Oxford Cryosystems, Oxford, UK) prior to data collection. All data were collected at the National Synchrotron Light Source, Brookhaven National Laboratories, using beamline X9B, the ADSC Quantum 4 detector with a wavelength of 0.92 Å, crystal distance-to-detector of 205 mm, and oscillation range of 1.0°and exposure for 45 s. The diffraction data from all crystals were indexed, processed, scaled, and merged using the HKL2000 suite of programs (14). An attempt was made to collect multiwavelength anomalous dispersion data in which halide ions are introduced as anomalous scatterers in the cryoprotectant during a short soak (15). The crystal was soaked in 1 M NaBr for 45 s but no anomalous scatterers could be obtained. The data collection statistics of crystals are summarized in Table I.
Structure Solution and Refinement-The crystal structure of the TGase 3 proteolyzed form I was solved by molecular replacement using AmoRe (16) and a resolution range of 10 to 3.5 Å for the rotation search and between 10 to 3.0 Å for the translation search. There were two solutions to both the rotation and translation searches. These solutions gave rise to two molecules within each asymmetric unit in the crystal form in space group P2 1 . The model of the TGase 3-activated enzyme (PDB 1L9M, Ref. 13) was used as the initial search model. The initial R-factor after correctly orienting and positioning the two molecules was 35.4% for all the data from 10 to 3.0 Å resolution. The rigid body refinement using the program CNS (17), followed by simulated annealing applying strict 2-fold non-crystallographic symmetry (NCS) constraints (18) to 3.0 Å resolution, improved the solutions yielding a final R-factor of 31.8%. Initial phases were improved using solvent flattening (19) with 2-fold NCS molecular averaging (20). For final refinement, the NCS restraints were released. After many cycles of manual model rebuilding into SIGMAA-weighted 3 F o Ϫ2 F c map, the refinement converged at an R-factor of 19.9%, which included all low resolution data set from 20 Å resolution as bulk solvent correction. Water molecules were included in the model based on hydrogen bonding, and the presence of reproducible peaks in electron density maps based on their effects on the free R value. In order to avoid overfitting of the model, each step of the rebuilding procedure was monitored using the free R-factor and a residue real space correlation coefficient as a guide.
For the TGase 3 (II) data set, starting with one of the monomers of the zymogen form (PDB 1L9N, Ref. 13) model as the search probe, molecular replacement in AmoRe (16) was used to search for the location of monomer in this crystal form in P2 1 space group. A translation search yielded the correct solution (CC ϭ 39.5%). The model was refined in CNS (17) with the simulated annealing, positional, and B-factor refinement protocols using a maximum likelihood target (21). Anisotropic scaling and a bulk solvent correction were used, and the individual B-factors were refined isotropically. Except for the free R set (a random sampling consisting of 9.6% of the data set), all data between 20 and 2.7 Å (with no cutoff) were included in the refinement, which was converged at an R-factor of 18.6%.
The crystal structure of the TGase 3 form III was solved by molecular replacement using AmoRe (16). There were two solutions to both the rotation and translation searches. These solutions gave rise to two molecules within each asymmetric unit in the P1 space group. The model of the TGase 3-activated enzyme (PDB 1L9M, Ref. 13) was used as the initial search model without any metals. The rigid body refinement using CNS software (17) followed by simulated annealing applying 2-fold NCS molecular averaging (20). For final refinement, the NCS restraints were applied. After many cycles of manual rebuilding into SIGMAA-weighted 3 F o Ϫ2 F c and F o Ϫ F c maps, the refinement converged at an R-factor of 19.4%, which included all low resolution data set from 25 Å resolution as bulk solvent correction (Table I). In each cycle of refinement, model bias was avoided by calculating a composite simulated annealing omit map, which was regularly used to verify, correct, or build the model. The stereochemical quality of the structure was analyzed with the programs PROCHECK (22) and CNS (17). Figures were generated with Molscript (23) and Raster3D (24).
Biophysical Studies of TGase 3-The thermodynamic properties of Ca 2ϩ binding to the activated TGase 3 were measured by isothermal titration calorimetry using a MicroCal VP-ITC calorimeter as described previously (13). Data were fitted to appropriate binding models and thermodynamic parameters were determined from nonlinear leastsquares fits, using ORIGIN TM software. Quantitative measurement of metal ion contents was performed by the use of inductively coupled plasma-mass spectrometry dynamic reaction cell (ICPMS-DRC) using a PerkinElmer ELAN 6100 instrument. Prior to measurements, the TGase 3 (10 mg/ml) was treated with dispase at 37 o for 30 min and dialyzed five times against 20 mM Tris-HCl (pH 8.0) and 5 mM EGTA. The EGTA solution was removed by dialysis against 20 mM Tris-HCl (pH 8.0). The concentration of zymogen treated with dispase was 1.5 mg/ml. TGase 3 Activity Assay-TGase activity was assayed by incorporation of [ 14 C]putrescine into casein (25). Assays were done at 37°C in 0.5-ml portions of 100 mM Tris-HCl (pH 7.5), containing 1% N,Ndimethylated casein (Sigma), 0.5 Ci of putrescine (118 mCi/mmol), 1 mM dithiothreitol, 5 mM CaCl 2 , and 1 mM EDTA.
TGase 3 Ca 2ϩ Ion Binding Assay-Calcium binding was determined by equilibrium dialysis (26) allowing equilibration of the enzyme (0.5-0.8 mg/ml) against 100 volumes of buffer containing 20 mM Tris-HCl (pH 8.0) supplemented with increasing amounts of 45 CaCl 2 (300 cpm/ nmol) up to 10 mM. After 48 h at 4°C, aliquots of the enzyme solution and of the dialysate were withdrawn for liquid scintillation counting. The results were normalized to the protein concentration determined by absorbance at 280 nm.

RESULTS AND DISCUSSION
The purpose of this study is to better understand the differential structural/functional roles of cations in the three known positions of the TGase 3 enzyme system and to explore their roles in the activation process. We have therefore solved the structures of three additional forms, and selected features are summarized in Table II.
Briefly, the overall structure of TGase 3 consists of four folded domains that are common to the structures solved here and previously (13). The amino-terminal ␤-sandwich domain of TGase 3 (first 134 amino acids) consists of nine strands of ␤-sheets interspersed with three ␣-helices adopting two fourstranded antiparallel sheets twisted about 50°with respect to each other to form a distorted ␤-barrel. The catalytic core domain (residues 135-472) consists of 15 ␤-sheets interspersed with 15 ␣-helical segments and is ␣/␤-type. It contains a central twisted six-stranded antiparallel ␤-sheet motif, which separates two clusters of ␣-helices. The longest ␣-helix of 16 residues is located in the center of the molecule and harbors the active site Cys 272 residue, which is buried in this domain. Other members of the catalytic triad, His 330 and Asp 353 , are located on adjacent strands of ␤-sheets. The barrel 1 and barrel 2 domains consist of largely of ␤-sheets arranged in barrel-like conformations, span from residues 473-592 and 593-692, respectively. Residues 462 to 471 form a highly flexible solventexposed loop that links the last ␣-helical segment of the catalytic core domain to the first ␤-strand of the barrel 1 domain. This flexible hinge region harbors Ser 469 , the cleavage site used for proteolytic activation of the TGase 3 zymogen. All structures possess two non-proline cis peptide bonds: Arg 268 -Tyr 269 in the strand prior to the loop of ␣-helix that contains the active site Cys 272 residue; and Asn 382 -Phe 383 in a loop adjoining two ␣-helices of the core domain. Furthermore, the Gly 367 -Pro 368 cis-peptide bond is conserved in location in all TGases.
The Structure of Form I Is Essentially Identical to the Activated TGase 3-First, we wanted to know whether the channel is opened by cleavage of the enzyme alone, that is, before insertion of metal ions at sites two and three. Proteolysis of the TGase 3 zymogen with dispase cleaves at Ser 469 the flexible hinge of sequences that joins the active site domain to ␤-barrel 1 domains. This form I, measured in the presence of dispase but with no added Ca 2ϩ , is fully active (Table II). Its x-ray diffraction crystal structure is shown in Fig. 1a. The refinement crystallographic statistics are listed in Table I. The x-ray models consist of two crystallographically independent monomers of 692 residues. Both monomers have missing density for a flexible loop between residues 461-479. The r.m.s. difference between the 673 C␣ carbon atoms of the two monomers is 0.41 Å, indicating that the C␣ backbone structures are almost identical. Each monomer has three Ca 2ϩ ions, and an open channel (Fig. 1b). However, its structure and coordinations of the Ca 2ϩ ions (Fig. 2, form I and Table III) are essentially identical to those of the activated enyzme solved previously (Ref. 13 and fourth item of Table II). It turns out that the dispase used contains high levels of Ca 2ϩ salts, so that during proteolysis, the cleaved enzyme acquires the two additional Ca 2ϩ ions from the solution to occupy sites two and three. However, the crystal packing arrangement of the two monomers in the asymmetric unit of the structure is the P2 1 space group, different from the P1 space group seen previously (13). Accessible surface area calculations for interface between the monomers in the asymmetric unit give a value of 2694 Å 2 for the area buried upon formation of the two crystallographic monomers.
Form II: the Proteolyzed Enzyme Without Ca 2ϩ Ions at Sites Two and Three-We were able to quantitatively remove the two Ca 2ϩ ions by chromatography through a mono Q column (confirmed by use of ICPMS-DRC), and this form II was crystallized. It has one molecule of 692 residues with missing density for residues 462-478 in its asymmetric unit of the P2 1 space group (Fig. 1c and Table II). This enzyme is inactive, and are the observed and calculated structure factor amplitudes for reflection h. g R free was calculated against 10% of the complete data set excluded from refinement. the channel is closed (Fig. 1d). Comparisons of its structures and the coordination of its single Ca 2ϩ ion at site one reveal near identity to the uncleaved zymogen (Fig. 2, form II and Table III, and Ref. 13). The only significant structural change is that the loop Ile 223 -Val 231 containing Asp 228 has shifted away so that Asn 227 and Asn 229 instead coordinate with the Ca 2ϩ ion in site one. Both are solvent-exposed yielding direct coordination with the Ca 2ϩ ion, thereby binding the ion even more tightly. In summary, these data show that metal ions must occupy sites two and/or three to open the channel necessary for activity.
The Ca 2ϩ Ion in Site One Contributes Significant Stabilization to TGase 3-Previously, we were unable to assess the energetic consequences of binding a Ca 2ϩ ion at site one. However, by use of more rigorous procedures, we were able to remove 85% of this ion, as assessed by titration with 45 Ca 2ϩ and direct quantitative measurements employing ICPMS-DRC (Fig. 3a). By use of microcalorimetry, the binding data was obtained for the Ca 2ϩ ions (Fig. 3, b and c). At the lowest Ca 2ϩ ion concentrations, there is one high affinity binding site (average K d ϭ 0.3 M) with ⌬H ϭ Ϫ6.70 Ϯ 0.52 kcal/mol. Addition of further Ca 2ϩ indicated two other low affinity sites (average K d ϭ 3.0 M) with ⌬H ϭ Ϫ3.73 Ϯ 0.16 kcal/mol. The shape of the latter part of the curve and values are consistent with the previous reported data for Ca 2ϩ ion binding at sites two and three (13). Thus the first part of the curve corresponds to binding at site one. Such an exothermic reaction implies significant stabilization of the zymogen.
Form III: Structural Differences That Affect Activity and Function-Previously, we showed that lanthanides can occupy  1. Conformations of the forms I (a and b), II (c and d), and III (e and g) solved in this study. The upper row shows the solved structures of the three forms. This is nominally the front side of the enzyme. The amino-terminal ␤-sandwich (red), catalytic core (blue), ␤-barrel 1 (magenta), and ␤-barrel 2 (orange) domains are shown. The Ca 2ϩ ions are shown in yellow, the sole Mg 2ϩ ion in cyan. Below are shown the electrostatic surface potential images. The acidic and basic residues are colored red and blue, respectively. The electrostatic potentials, including Ca 2ϩ and Mg 2ϩ ions, have been mapped onto the surface plan from Ϫ15 kT (deep red) to ϩ15 kT (deep blue). The open channel is clearly evident in b. In g, the back side of the enzyme has a deep cavity; the front side (f) remains closed. sites two and/or three with retention of full activity (13). But the major multivalent metal cation in cells is Mg 2ϩ . Accordingly, we explored conditions to occupy sites two or three with Ca 2ϩ or Mg 2ϩ ions. First, we measured TGase 3 activity in the presence of MgCl 2 and/or CaCl 2 (Fig. 4a). With MgCl 2 alone, there were only traces (Ͻ10%) of activity. With CaCl 2 , maximal activity was acquired by about 0.6 mM in the absence and presence of MgCl 2 . Addition of higher concentrations of MgCl 2 inhibited activity by 10 -20%. Second, we titrated form II with different molar ratios of these ions and found by employing ICPMS-DRC that in the range of 0 -2.5 mM CaCl 2 or MgCl 2 , at low ratios (Յ0.1) of Ca 2ϩ /Mg 2ϩ the TGase 3 contains one of each ion only (Fig. 4b). Form II rapidly acquired a second Ca 2ϩ ion by ratios ϳ0.2, then gradually began to displace the sole Mg 2ϩ ion by ratios Ն0.3, and finally Mg 2ϩ ion was lost by ratios Յ0.7 (Fig. 4b). The one Ca 2ϩ /one Mg 2ϩ enzyme had no activity; the two Ca 2ϩ /one Mg 2ϩ enzyme had very low activity, and the three Ca 2ϩ enzyme had full activity (Fig. 4b).
Attempts were therefore made to crystallize form II decorated with one Ca 2ϩ /one Mg 2ϩ , and one Mg 2ϩ /two Ca 2ϩ . By use of MgCl 2 and an ATP salt, employed because they allowed introduction of Mg 2ϩ , and the ATP does not bind to TGases (27), we were able to recover only the second of these, termed form III. We reconfirmed by ICPMS-DRC in crystals the presence of one Mg 2ϩ and two Ca 2ϩ ions and no other metals. The x-ray diffraction crystal structure of form III model has two crystallographically independent monomers of 692 residues per asymmetric at 2.4 Å in the P1 space group (Table I and Fig. 1e). Both monomers have missing density for a flexible loop between residues 462 and 478 and the r.m.s. difference between the C␣ carbon atoms of the two monomers is 0.30 Å, indicating that the C␣ backbone structures are almost identical. Accessible surface area calculations for interface between the monomers in the asymmetric unit give a value of 2688 Å 2 for the area buried upon formation from each crystallographically independent monomers. As the protein is a monomer in solution (28), this interface may not be functionally relevant and is imposed by crystal contacts within the unit cells. Notably, the channel is closed (Fig. 1, f and g). There are major differences in the coordinations of the metal ions (Fig. 2, form III and Table III Table III. an octahedral geometry. Furthermore its coordinations were shorter and did not involve the conserved residue Asp 324 . To identify which of sites two or three possessed the Mg 2ϩ ion, we initially modeled both for Ca 2ϩ . After several rounds of building and refinement, a decrease in the ͗B͘ factor value for Ca 2ϩ relative to the surrounding side chain atoms and relatively short distances to nearby oxygen atoms (2.0 -2.2 Å), indicated that site three most likely contained the Mg 2ϩ ion. Second, we calculated the mean peak size in the F o Ϫ F c difference electron density maps at above 4 level (large positive difference peak) for each metal ion in sites two and three in each independent monomer and found that site two is most likely Ca 2ϩ and site three Mg 2ϩ . The clarity of the simulated annealing F o Ϫ F c omit electron density map, the observed bond distances, and the coordination geometry strongly support the assignment of Mg 2ϩ at site three. Finally, compared ͗B͘ values as well as bond valence calculations using parameters from Brese and O'Keeffe (29) provided support for this assignment as Mg 2ϩ ion. We note that the bond distances to the metal-oxygen ligand atom typically range from 2.0 -2.2 Å for Mg 2ϩ and 2.2-2.9 Å for Ca 2ϩ . In addition, the smaller size of Mg 2ϩ ions determines a preference for six coordinations, rather than an energetically more favorable seven for Ca 2ϩ (30). Furthermore, only the side chain carbonyl oxygen atom of Asn 305 is coordinated in a monodentate manner to Mg 2ϩ , whereas Ca 2ϩ coordinates in a bidentate manner with both the side chain atoms of Asn 305 in form I that contains a Ca 2ϩ ion at site three. Thus the Asn 305 residue plays an especially important role in discriminating between Ca 2ϩ and Mg 2ϩ ions. Nevertheless, superpositions of the Ca 2ϩ and Mg 2ϩ of forms I and III show that the locations of the metal ions are similar. Notably however, comparisons of their crystal structures reveals that the Mg 2ϩ ion at site three results in a contraction of the coordination sphere, and a less energetically favored conformation because of a weaker interaction with Asn 305 . Moreover, the Mg 2ϩ ion in site three is unable to coordinate with residue Asp 324 from the Gly 322 -Ser 325 loop segment motif. Thus this loop occupies the same position as in the zymogen and no channel forms. On the other hand, some lanthanide metal ion (Er 3ϩ or Tb 3ϩ ) forms were as active as the Ca 2ϩ form (13). The likely reason is that for example Ca 2ϩ and Tb 3ϩ both possess spherical filled outer electronic subshells and are of similar size; their effective ionic radii for 7-fold coordination are 0.98 and 1.06 Å, respectively.
In the sequence Asp 320 -Lys-Gly-Ser-Asp-Ser 325 involved in metal ion binding at site 3, we suggest that Gly 322 plays a pivotal role. Because the Gly residue lacks a side chain, it provides a greater range of backbone torsion angles than any other amino acid. Having a Gly 322 modulate the range of motion or pliability of this sequence motif by 9 Å thereby allows specific side chain interactions of Asp 324 with a Ca 2ϩ ion. Such flexibility could thus influence the affinity and kinetics of metal ion binding or release and simply control the ability to discriminate between metal ions of different size, as seen here for Ca 2ϩ and Mg 2ϩ ions. Indeed, there are numerous established precedents for this concept (31)(32)(33). In particular, mutagenesis experiments on EF-hand calcium-binding proteins has revealed that Gly residues can greatly affect the affinity, specificity, and/or stability of metal binding (31).
Accordingly, these data establish that the metal binding properties of site three determine the precise coordination with neighboring atoms, which in turn directly affects the folding of the peptide chain.
Cooperative Role of the Ca 2ϩ Ion in Site Two- Fig. 4 showed that in the presence of high relative concentrations of Mg 2ϩ ions, the proteolyzed enzyme possesses the one Ca 2ϩ ion at site one and a sole Mg 2ϩ ion, and the enzyme is inactive. The data for form II presented above indicate that the Mg 2ϩ ion should be bound at site 3 only, leaving site two vacant. As relative Ca 2ϩ ion concentrations rise, the enzyme rapidly acquires a second ion (Fig. 4b), but the two Ca 2ϩ /one Mg 2ϩ ion form III is still inactive. Our data indicate that this second Ca 2ϩ ion occupies site two. However, as relative Ca 2ϩ ion concentrations rise still further, the Mg 2ϩ ion at site three is displaced by Ca 2ϩ , and only when a net of Ն1.5 Ca 2ϩ ions have been acquired, does the enzyme begin to display activity (Fig. 4b). We assume this corresponds to opening of the front side of the enzyme. The shape of the activity curve of Fig. 4b suggests that site two should be filled with the Ca 2ϩ ion first, and that activity is recovered as soon as site three is at least half-filled. These data therefore strongly imply that sites two and three cooperate with each other.
We explored the reason why site two is not occupied by a Mg 2ϩ ion in a high Mg 2ϩ ion environment. Examination of 2 F o Ϫ F c and simulated annealing F o Ϫ F c omit electron density maps suggests that in order to comply with the obligatory 2.0 -2.2 Å bonding range, a Mg 2ϩ ion could only coordinate with one neighboring amino acid atom at a time. The smaller size of Mg 2ϩ ion determines its preference for a coordination number of six with an octahedral geometry of its complexes and is less comfortable than Ca 2ϩ in accepting larger multidentate and anionic ligand groups. An alternative possibility for a Mg 2ϩ ion to coordinate with several amino acid atoms would impose severe local distortion of the peptide backbone. Both situations are energetically unfavorable. Thus, site two provides a highly constrained binding cavity for metal ions, so that only a Ca 2ϩ (or lanthanide) ion with an appropriate ionic radius and multidentate property can comfortably occupy site two.
The fXIIIa enzyme has one Ca 2ϩ ion at a location generally similar to site two of TGase 3 and lies in an acidic pocket formed by Asp 438 , Glu 485 , and Glu 490 , but the Ca 2ϩ ion coordinates only with the main chain carbonyl oxygen of Ala 457 , and four water molecules (34). However this ion does not confer any significant change in fXIIIa structure, and the enzyme is inactive. It remains to be determined whether other Ca 2ϩ ions are utilized by an active fXIIIa enzyme form and whether they too cooperate with each other to allow reaction.
Structure of the Channel-We note from the structure of the zymogen (13) and form II (Fig. 1c) the presence of a deep cavity on one side (nominally termed the back side) of the TGase 3 enzyme. We have seen in form I that the binding of the Ca 2ϩ ion in site three laterally displaces a loop following the strand bearing residues Asp 320 to Ser 325 so that Asp 324 coordinates with the ion (Fig. 5a). This loop on the front side of the enzyme moves so as to open the deep cavity and thus a channel through the enzyme. However, when the Ca 2ϩ ion (weaker Lewis acid) is replaced by a Mg 2ϩ ion at this site three, the channel is closed (Fig. 1, f and g). The side chain of Ser 323 in the loop FIG. 5. Front (a) and back (b) views of the cavity/channel of the activated TGase 3 enzyme. The Ca 2ϩ ions in sites one, two, and three are colored lime green. In a, the transparent electrostatic surface potential shows that the active site triad residues Cys 272 , His 330 , Asp 353 are buried and inaccessible. The movement of the loop bearing residues Asp 320 -Ser 325 in the activated enzyme opens the cavity to form a channel through the enzyme and exposes the side chains Trp 236 and Trp 327 residues, thought to be important in manipulation of substrates. In b, the guanidinium group of Arg 396 participates in a salt bridge interaction with Glu 586 at the entrance of the cavity. Also shown are the side chains of Arg 420 and Arg 570 , which protrude into the cavity, thereby likely severely interfering with possible substrate access.
sequence Asp 320 -Lys-Gly-Ser-Asp-Ser 325 motif instead forms a direct bond via a water molecule to the main chain nitrogen of Leu 529 and the carbonyl oxygen of Asp 320 causing the front of the cavity/channel to be closed.
At the back side of the enzyme, the cavity is always open whether or not any metal ion occupies site three. The guanidinium side chain of Arg 396 near the opening of the cavity adjacent to the bulk solvent is about 17 Å from the C␣ atom of the active site Cys 272 and 13.5 Å from the C␣ atom of Tyr 525 . The length of the cavity is about 16 Å from the guanidinium side chain of Arg 396 to the metal ion at site three. At its widest point, the diameter of the cavity is about 13 Å from the C␣ atom of Arg 587 to the C␣ atom of Thr 417 , with a total volume of 90,414 Å 3 . The entrance of the cavity is controlled by the side chain movements of Arg 396 from the catalytic core domain and Glu 586 of the last ␤-strand of the ␤-barrel 1 domain (Fig. 5b). These latter residues appear to be coordinated with each other and participate in hydrogen bonding. The upper portion of the cavity is lined predominantly by hydrophilic side chains of Arg 420 , Arg 587 , Thr 167 , and Asn 168 and is filled with a chain of water molecules connected to each other through hydrogen bonds. The cavity itself is lined by the side chains of Arg 570 , Asp 588 , Met 568 (right interior), and Val 326 and Asn 328 (left interior). When viewing the back of the cavity, the guanidinium groups of three arginines (Arg 420 , upper portion; Arg 396 in the middle; and Arg 570 at the end of the cavity) intrude into the channel. The spatial conformation of the cavity does not significantly change in the structures solved here or previously (13), except for the opening of the front side.
Altogether, these data impose severe constraints on how this cavity/channel could be used by substrates: it is too deep for a Gln or Lys substrate residue side chain to penetrate to the active site Cys residue and it is too narrow for a peptide chain backbone bearing a reactive Gln or Lys residue to penetrate through the maze of Arg side chains to the active site Cys residue. Accordingly, it is difficult to see how the cavity at the back side of the enzyme could be utilized for substrate access. Alternatively, substrates might preferably approach the enzyme from the front side.

CONCLUSIONS
One of the most intriguing aspects of catalysis by the TGase 3 enzyme is its dependence on two divalent metal ions. They occupy distinct binding sites and differ greatly in their affinities for various divalent metal cations. The physiological concentration of free Mg 2ϩ ions in resting eukaryotic cells is 1-2 mM, whereas that for free Ca 2ϩ ions is ϳ100 nM, so that there is about a 10 4 molar ratio. This consideration allows the following hypothesis for the control of function of the TGase 3 enzyme. First, any TGase 3 protein present in a cell will likely be in the inactive zymogen form and have a sole Ca 2ϩ ion in site one. But the flexible loop of residues 462-471 of TGase 3 is physically easily broken leading to activation (13,28). Should any enzyme become proteolyzed or broken, we can conclude from the data of Fig. 4 that site three will readily acquire a Mg 2ϩ ion but the enzyme form will remain inactive. This therefore provides a simple but elegant mechanism to protect the cell from aberrant transamidation reactions by some proteolyzed TGase 3 enzyme. Only when global intracellular or microenvironmental Ca 2ϩ concentrations rises into the micromolar range will site two acquire a Ca 2ϩ ion, and the Mg 2ϩ ion from site three be displaced by a Ca 2ϩ ion (Fig. 4). The Ca 2ϩ ion occupancy at both sites cooperatively allow energetically favorable coordination with residues Asn 305 and Asp 324 , resulting in a change in the conformation of the loop Asp 320 -Ser 325 , the channel opens, and the enzyme becomes active. How can Ca 2ϩ concentrations rise to the required high level in living cells? We speculate that local microenvironmental Ca 2ϩ ion levels could be greatly changed in living cells by association of the TGase 3 enzyme with other cellular constituents, as indeed has been demonstrated previously for the TGase 1 enzyme (35). Further studies are now necessary to understand how this channel is employed by substrates to effect reaction.