Crystal structure of a dual activity IMPase/FBPase (AF2372) from Archaeoglobus fulgidus. The story of a mobile loop.

Several hyperthermophilic organisms contain an unusual phosphatase that has dual activity toward inositol monophosphates and fructose 1,6-bisphosphate. The structure of the second member of this family, an FBPase/IMPase from Archaeoglobus fulgidus (AF2372), has been solved. This enzyme shares many kinetic and structural similarities with that of a previously solved enzyme from Methanococcus jannaschii (MJ0109). It also shows some kinetic differences in divalent metal ion binding as well as structural variations at the dimer interface that correlate with decreased thermal stability. The availability of different crystal forms allowed us to investigate the effect of the presence of ligands on the conformation of a mobile catalytic loop independently of the crystal packing. This conformational variability in AF2372 is compared with that observed in other members of this structural family that are sensitive or insensitive to submillimolar concentrations of Li(+). This analysis provides support for the previously proposed mechanism of catalysis involving three metal ions. A direct correlation of the loop conformation with strength of Li(+) inhibition provides a useful system of classification for this extended family of enzymes.

In mammalian cells, inositol monophosphatases are abundant cytosolic enzymes necessary to regenerate the supply of myo-inositol for the synthesis of phosphatidylinositols. However, in archaeal hyperthermophilic organisms, e.g. Archaeoglobus fulgidus, an IMPase 1 ortholog is thought to be used in the biosynthesis of a unique osmolyte, di-myo-inositol 1,1-phosphate (1). The archaeal enzymes have another twist as well. Besides catalyzing the hydrolysis of Ins-1-P (needed in response to thermal and salt stress) they also very specifically dephosphorylate FBP at C-1 (2). Enzymes with these characteristics exist as two distinct gene products in mammalian systems and several bacteria (3). The dual specificity of the enzyme from primitive hyperthermophiles strongly suggest that IMPase and FBPase enzymes may have evolved from the same gene product (2).
Structural features of the IMPase/FBPase (MJ0109) from Methanococcus jannaschii (2) are similar to those of mammalian IMPase, inositol polyphosphase phosphatase (IPPase), FBPase, and the yeast enzyme Hal2. These enzymes form an extended family for which the general principles of chemical reactivity are the same (3). Two or three metal ions are necessary for catalysis to occur. From the structure of human IMPase with bound Ins-1-P and Gd 3ϩ , it was inferred that two metal ions were needed for catalysis (4). However, crystallographic data for pig kidney FBPase (5) and for MJ0109 (6) are consistent with a three-metal ion-assisted catalytic mechanism.
There is strong biochemical evidence that Li ϩ inhibits IMPase in mammalian cells where a dose response to this ion correlates with a decrease in the intracellular pool of myoinositol (7). Li ϩ , by inhibiting mammalian IMPases, is thought to cause a depletion of inositol that attenuates the synthesis of phosphatidylinositol necessary to regenerate the phospholipase C dependent phosphatidylinositol/IP 3 signaling pathway (8,9). There are two other enzymes that are very Li ϩ -sensitive with known three-dimensional structures: human IPPase (10) and yeast 3Ј-phosphoadenosine-5Ј-phosphate phosphatase, Hal2 (11). In contrast, the dual specificity thermophilic IMPase/FBPase enzymes from M. jannaschii and A. fulgidus and mammalian FBPase are not inhibited by Li ϩ in the submillimolar range (Table I). Thus, the IMPase superfamily can be functionally subdivided into those that are strongly inhibited by Li ϩ and those that are less sensitive to Li ϩ . Structural hints on the mechanism of Li ϩ inhibition of human IMPase are indirect. In pig kidney FBPase (13), which is inhibited by Li ϩ but activated by low concentrations of K ϩ , monovalent cations are thought to inhibit the enzyme by distorting the active site geometry or preventing turnover or product release. Despite extensive mutagenesis experiments carried out on several en-zymes in the family, the mechanism of Li ϩ inhibition is still not entirely understood.
This work presents the results of structural studies for another dual specificity IMPase/FBPase from the hyperthermophilic sulfate reducer A. fulgidus. Like MJ0109, this archaeal enzyme (AF2372) has sequence features close to human IMPase but key structural similarities more akin to pig kidney FBPase. A comparison of the structures of the two archaeal Li ϩ -insensitive enzymes (AF2372 and MJ0109), a phosphatase of intermediate sensitivity (pig kidney FBPase), and three very Li ϩ -sensitive members of this superfamily (human IMPase, IPPase and yeast Hal2) provides a strong correlation between the level of Li ϩ sensitivity and the conformation of a critical catalytic loop. The disposition of this catalytic loop may serve as a predictor of Li ϩ sensitivity for other IMPase family members. Furthermore, changes in the loop conformation of AF2372 induced by ligand binding shed some light into the possible mechanism of Li ϩ inhibition in the human IMPase and into how this phenomenon might have developed from the more primitive archaeal enzymes.

EXPERIMENTAL PROCEDURES
Chemicals-D,L-Ins-1-P, EDTA, sodium chloride, magnesium chloride, zinc chloride, ammonium chloride, malachite green oxalate, ammonium molybdate, isopropyl-1-thio-␤-D-galactopyranosideisopropyl-1thio-␤-D, SDS-PAGE molecular weight markers, gel filtration molecular weight markers Trizma base, and PEG 8000 were obtained from Sigma. Glycine, myo-inositol, and ammonium sulfate were obtained from Aldrich. Ampicillin was obtained from Fisher. Tryptone and yeast extract were obtained from Difco. Q-Sepharose fast flow, phenyl-Sepharose, and Sephacryl S-200HR resins were obtained from Pharmacia. L-Ins-1-P was synthesized enzymatically using an Ins-1-P synthase cloned from A. fulgidus as described previously (14). The Ins-1-P and the other small molecule fractions were separated from the synthase with Centricon concentrators. The filtrate was evaporated under high vacuum; a known amount of glucose 6-phosphate was used to quantify the concentration of purified L-Ins-1-P by 31 P NMR spectroscopy. Further purification of the sugar required chromatography on an AG1-X8 anion exchange column (elution with a gradient of 0.2-1.0 M formic acid); fractions containing Ins-1-P were collected and the pH value was adjusted to 7.5 with NH 4 OH. The purified L-Ins-1-P was stored at 4°C and used for enzymatic assays and crystal soaks. All other chemicals were reagent grade.
Overexpression and Purification of AF2372 from A. fulgidus-Cloning the AF2372 gene from A. fulgidus and overexpression of the gene product in Escherichia coli were carried out as described elsewhere (15). Key steps of the purification procedure included heating of crude protein at 85°C for 30 min, followed by chromatography on Q-Sepharose fast flow (elution with a linear gradient of 0 to 0.5 M KCl). Column fractions were monitored for IMPase activity using 15 mM pNPP and 30 mM MgCl 2 in 50 mM Tris and 1 mM EDTA, pH 7.5. Purity of the fractions was assessed with SDS-PAGE; fractions that were greater than 90% pure were collected, pooled, concentrated 10-fold, and dialyzed against two changes of 50 mM Tris, pH 7.5. The concentrated solution was then loaded onto a 1.6 ϫ 72-cm Sephacryl S-200HR gel filtration column pre-equilibrated with 50 mM Tris and 1 mM EDTA, pH 7.5. Fractions containing pure Ins-1-Pase activity (as measured by SDS-PAGE and silver staining of the gels) were concentrated to 12 mg/ml and stored in 20 mM Tris HCl and 1.0 mM EDTA, pH 8.0, until used for crystallization.
Activity Assays-Production of inorganic phosphate from FBP and Ins-1-P was monitored using a colorimetric assay (16). Substrate concentration was typically 2 mM (except for pNPP, which was 0.5 mM) in 50 mM Tris, pH 8.0, with varying amounts of divalent cations (Mg 2ϩ or Mn 2ϩ ) added. The solution with substrate and enzyme was heated to 85 or 80°C (the temperature is indicated with each assay set) for 1 min, then 900 l of dye was added, and the OD 660 was measured. A standard P i curve was used to convert OD 660 to mol of P i produced. Assays typically used 1.0 -2.0 g of enzyme.
Crystallization-AF2372 was crystallized by vapor diffusion using hanging drops of 5 l. The apoenzyme was crystallized in either 0.2 M ammonium nitrate and 30% PEG 3350 (P3 2 form) or 0.2 M dihydrogen ammonium phosphate and 30% PEG 3350 (P2 1 form). Enzyme solution (2.5 l of ϳ10 mg/ml) was mixed with an equal volume of mother liquor and equilibrated using the vapor diffusion hanging drop method. To monitor the catalytic loop conformation under various conditions, the apoenzyme crystals were soaked in solutions containing ligands and different divalent cations. In addition, the enzyme was co-crystallized with Ca 2ϩ (50 mM) using 5-7% v/v solution of PEG 8000 in 50 mM Tris, pH 8.0, with 100 mM NaCl. These crystals were then soaked with either D-Ins-1-P or FBP. The drops were equilibrated against the reservoir with the concentration of PEG 8000 between 5 and 7%. Crystals appeared in few days and grew to full size within a week.
X-ray Data Collection-Crystals having approximate dimensions of 0.5 ϫ 0.4 ϫ 0.4 mm were mounted in glass capillaries directly from the crystallization wells, or flash frozen in a stream of nitrogen gas. Diffraction data were collected both at cryo conditions and at room temperature. Data were collected at Boston University School of Medicine at room temperature on a Rigaku RU-300 rotating anode generator with an R-AXIS IIC imaging plate detector. Data were also collected at Rice University using a Rigaku H3R rotating anode generator with an R-AXIS IV imaging plate detector with the exception of the apo form in P2 1 space group that was collected at the Brookhaven Synchrotron beamline X8C. Data collected in Brookhaven and at Boston University was indexed and reduced using DENZO and Scalepack (see Ref. 17), while the data collected at Rice University was indexed and reduced using Crystal Clear software. In addition to the apoenzyme (P3 2 or P2 1 ), whose crystals diffracted to ϳ2 Å, several data sets were collected on crystals soaked with ligands or co-crystallized complexes. The data were collected to ϳ2.2 Å resolution for the ligand soaks.
Structural Solution and Refinement-The phase problem was solved by using the molecular replacement method with the MJ0109 model (PDB code 1DK4) as a probe structure. The rotational as well translational searches were performed using the program Amore (18). The searches succeeded with the monomer as a model. The initial solution was refined by rigid body refinement and simulated annealing (crystallography NMR software) with manual rebuilding using O (19). The apoenzyme in the crystal form characterized by the space group P3 2 , initially indexed as P3 2 21, was evaluated for merohedral twinning (20) and found to be nearly perfectly twinned with a twinning ratio of 0.499. The reflections for this trigonal form of the apoenzyme were then sorted and processed in space group P3 2 . Simulated annealing refinement in the program Xplor (21) lowered the R-factor significantly. Several macrocycles including manual rebuilding sessions done on the SGI work station with the program Xtalview (22) led to convergence. The final models were refined by the program Shelxl-97 (23). The models were evaluated using Procheck (24). Table II provides a summary of data collection and refinement parameters for both the apoenzyme crystal forms and the ligand-soaked ternary complexes.

AF2372 Affinities for Substrate and Metal
Ions-At the preparatory stage for crystallographic experiments, the substrate specificity as well as the metal ion affinity of AF2372 were evaluated kinetically (at 85°C). The substrate concentration, 2 mM, was chosen to be much greater than the K m for Ins-1-P or FBP (2), so that a comparison of activities is likely to reflect differences in V max . The substrate specificity of AF2372 is similar to that of the M. jannaschii ortholog (Fig. 1A) with Ins-1-P, FBP, ␤-glycerol phosphate, 2Ј-AMP, and pNPP as substrates. Except for FBP, no other -CH 2 OPO 3 2Ϫ phosphate of the sugar phosphates examined was removed by the enzyme. The K m values for L-Ins-1-P and FBP have previously been determined as 0.11 and 0.08 mM, respectively (2). With 2 mM L-Ins- 1-P or FBP (values well above K m ), the K d values for Mg 2ϩ and Mn 2ϩ with these substrates were found to be 15-30 mM (Table  III). There was a 2-fold difference in K d values for Mg 2ϩ using L-Ins-1-P or FBP as substrate. It is possible that the presence of two phosphate groups in FBP facilitate metal ion interactions within the active site. Similar to the phosphatase from M. jannaschii (25), Ca 2ϩ was an inhibitory metal ion for both Ins-1-P and FBP hydrolysis catalyzed by AF2372. The inhibition of AF2372 phosphatase activity (with L-Ins-1-P as substrate) by Li ϩ was very weak with an IC 50 ϳ290 mM (Table I). This very weak binding precludes an accurate determination of the mechanism of inhibition as uncompetitive (as many other IMPases) or noncompetitive. Regardless of which mode of inhibition is correct, the K i would be ϳ290 mM because the substrate concentration was well above V max . For comparison, 300 mM Na ϩ or K ϩ had no effect on AF2372 activity indicating that the inhibition by Li ϩ is specific to that monovalent cation. The AF2372 gene product, like MJ0109 (25) and a similar activity from Thermotoga maritima (26) is insensitive to Li ϩ . For comparison, Li ϩ -sensitive representatives in the IMPase superfamily like human IMPase (27) have much lower k i values (0.3 mM). Similarly, the Li ϩ k i for Hal2 was 0.07 mM (11).
General Description and Quality of Crystal Structures of AF2372, a Comparison to MJ0109 -The A. fulgidus IMPase/ FBPase crystallized in two distinct crystal forms: P2 1 and P3 2 . Although these two crystal forms of apoAF2372 had slight differences in orientation of one monomer with respect to the other, the r.m.s.d. calculated for C␣ positions of the dimers was ϳ0.6 Å indicating that the two forms are very similar. The crystal packing of the two forms in the crystal lattice was different. This finding is consistent with the P3 2 form having an unusually high solvent content (69% as measured by Ficoll gradient). In fact, this crystal form of the apoenzyme easily dissolved under slightly changed conditions in the well. The major crystal contacts in both crystal forms involved a mobile loop comprised of residues 32-43. In P2 1 and P3 2 lattices, the loop had contacts with the same region at the back of the symmetry-related molecule around residue 110. An additional contact was formed in the P2 1 lattice to the two N-terminal helices. This contact breaks the symmetrical contact of residues 37-110 that is most likely responsible for crystal twinning in the P3 2 lattice.
The overall fold of AF2372 is highly conserved in the entire IMPase superfamily from archaea to eukaryotes. IMPases from eukaryotes are organized as homodimers, while FBPases are tetrameric (a dimer of dimers). The AF2372 structure was organized in a similar fashion to that of dimeric MJ0109 with the same domain and global secondary structure organization, although there were some changes at the dimer interface. The dimer interface of AF2372 was large, ϳ1200 Å 2 . Hydrophobic contacts constituted 47% of the total interface, polar contacts 14.2%, and charged contacts 33.2% of the interface. There appeared to be 11 hydrogen bonds across the interface. Despite a very extensive nonpolar component coming from a cluster of hydrophobic residues (six phenylalanines), the charged and polar contacts appeared to be key stabilizing interactions for the dimer. Consistent with the structural observations, soaking crystals in high Ca 2ϩ (Ͼ100 mM) disrupted the dimer interface. More specific interactions in the P2 1 apoenzyme included hydrogen bonding at the interface as well as a possible salt bridge between positions Arg-27 and Glu-138.
The common architecture of each monomer consisted of an ␣/␤ structure organized in two domains of alternating layers of ␣-helices and ␤-sheets connected by variable length loops (Fig.  2). The structures of proteins in the IMPase superfamily differ mostly by the size of the connecting loops. Both domains in AF2372 were connected by a large hinge-like loop that divided the monomer into two structural fragments. Each domain was organized in an alternating ␣Ϫ␤Ϫ␣ motif with one possessing a seven-stranded sheet and the other a five-stranded sheet. The larger sheet was flanked by two helices on both sides, while the smaller one had only a single helix in the top layer ( Fig. 2A). The elongated loop that separated the monomer into two domains may provide enough flexibility to cause conformational changes within the monomer that affect ligand binding at the active site. That type of mechanism is used in inhibiting mammalian FBPases (28). Overall, the shape, volume, and even to the certain degree surface accessible area were conserved for both proteins. How-ever, despite having the same number of amino acids, there were significant conformational differences between AF2372 and MJ0109 (Fig. 2B). The N-terminal helix in the A. fulgidus IMPase was slightly elongated relative to that in MJ0109. The second N-terminal helix was also slightly elongated compared with MJ0109 (the MJ0109 helix is from residues 45-54; in AF2372 it comprises residues 44 -58). The shorter loop regions in MJ0109 might be one of the major factors that confer greater stability for the M. jannaschii homologue (T m ϭ 87°C for AF2372 versus 95°C for MJ0109 (15)). The remaining helices of AF2372 were very similar to both MJ0109 and human IMPase in length and positioning (6) with one exception. The first significant difference was at the elongated loop 105-112 in AF2372 that provides most of the crystal contacts in both lattices. There were some small differences at the loop around residue 148, but an important difference was at the first external helix. In MJ0109, this helix contributed to the dimer interface. A significant deletion in the sequence eliminated the interfacial helix present in MJ0109. The place of the helix was taken by a loosely packed coil. This structural element caused a much looser organization of the dimer that might also be associated with decreased thermal stability of AF2372 as compared with MJ0109.
Another interesting structural variation in the AF2372 structure was the presence of the cis-peptide bond between Tyr-155 and Tyr-156. The cis organization of this peptide bond allowed formation of an additional hydrogen bond between different strands of the second ␤-sheet. Both tyrosine residues seemed to play important roles in dimer formation; they also contributed ligands for binding of both substrates. When FBP was bound at the active site, it was coordinated by the hydroxyl group of Tyr-156, while when L-Ins-1-P was bound it was coordinated by the hydroxyl group of Tyr-155. Non prolyl cis-peptide bonds occur in proteins in relatively low amounts (ϳ0.15%). Their presence in an unfolded state of the protein can make a significant contribution to folding kinetics. A barrier for cis to trans transition can be the rate-limiting step for some slow folding molecules (29). However, it is relatively rare to observe cis-peptide bonds in the folded state. The cis-peptide bond between Tyr-155 and Tyr-156 may be an additional factor in the diminished thermal stability of the protein and in the irreversibility of the thermal denaturation (after heating at ϳ95°C AF2372 cannot refold into an active protein (15).
Structural Details of Two Apo Forms and Complexes with Ins-1-P/2 Ca 2ϩ , FBP/3 Ca 2ϩ , and F-6-p ϩ Pi/3 Mn 2ϩ -There were two structural forms of the apoenzyme. The first one was the P2 1 crystal form. For this form, the data were collected, and the structure determined to 1.8 Å resolution. The electron density at the first subunit was of exceptional quality (Fig. 3). The second subunit seemed to have more mobility, and the resulting map was of slightly lesser quality. Both subunits appeared to have an active site devoid of metal ligands. The second crystal form, refined at slightly lower resolution (ϳ2 Å), exhibited an apparent symmetry P3 1 21. After a closer examination of the resulting distribution of intensities, we determined that the crystal form was an ideal merohedrally twinned structure in the P3 2 space group. This crystal form also showed no significant occupancy of metal ions at the active site (Fig.  4A). However, the packing relations were sufficiently different in both crystal forms that it provided an independent estimate of conformational variability of the catalytic loop. The ordering of the catalytic loop is much higher in the P3 2 crystal form; that is best expressed by the ratio of the average temperature factors (B's) for the main chain atoms of the mobile loop divided by the average B's for the entire structure. This ratio drops from around 2.8 in the P2 1 crystal form to ϳ1.2 in P3 2 .
The third structure was for the enzyme with D-Ins-1-P and two Ca 2ϩ bound at the active site (Fig. 4B). This structure was derived by soaking the P2 1 crystals with 20 mM substrate and 50 mM Ca 2ϩ . The occupancy of metal ions as well as the substrate molecule refined around 60%. The precise description of the active site and coordination of the metal ions was hindered by the resolution and the quality of the data. However, general features were easily discernible, and they are described below. There were two Ca 2ϩ metal ions at each active site visible in the structure. Both calcium ions were coordinated by five ligands arranged in a distorted octahedral coordination with one ligand missing. The missing ligand could be easily complemented by a water molecule. The ligands of Ca1 included Asp-82, Asp-85, Asp-200, and two oxygens of the substrate molecule, while Ca2 was coordinated by Asp-82, Glu-67 OE1, and a single oxygen of the phosphate group. The inositol moiety was coordinated by the Asp-85, backbone nitrogen of Ala-172, guanidino group of Arg-191, and Tyr-155. The interactions of metal ions and substrates with the enzyme are summarized in Table IV.
The fourth structure was derived from the P3 2 crystals soaked with FBP and Ca 2ϩ at concentrations of 20 mM and 50 mM, respectively. The refined occupancy for the metals as well as the substrate was 100%. Three Ca 2ϩ ions were present in this structure (Fig. 4C). The coordination of two calcium ions was analogous to that observed in the D-Ins-1-P-enzyme complex. The third Ca3 metal ion was located more remotely from the active site than the third metal ion in MJ0109. In subunit A it was bound at a classical hepta-coordinated site formed by five water molecules and two protein ligands originating from the mobile loop (Thr-40 O␥1 and carbonyl oxygen of Pro-41). At subunit B two of the water molecules were missing. Ca3 was quite removed from the vicinity of other metal ions with distances ϳ4.5 Å from Glu-67 and ϳ4.5 Å from Asp-38. Both of these tentative ligands have a bridging water molecule coordinating Ca3. This implies that if it participates in catalysis, additional motion of the catalytic loop would be required to bring it to the immediate proximity of the tentative nucleophile. Additionally, the structure of MJ0109 suggested that activating and not inhibitory metal ions were necessary to observe a catalytically competent third metal ion binding site. Therefore, we have obtained a crystal structure of the product complex with activating metal ions (3 Mn 2ϩ , fructose-6phosphate (F-6-P), and inorganic phosphate) in P3 2 crystal form. The occupancy of the metal ions and the phosphate moiety refined to 100%, while the product F-6-P refined at ϳ60%. As expected, the loop moved toward the active site, and the direct coordination of the third metal ion changed (Figs. 4D and  5D). The Mn3 has an octahedral coordination with three protein ligands (side chains of Asp-38, Thr-40, and Glu-67) as well as three water molecules directly coordinating the metal ion. This is in marked difference with the FBP-calcium complex in which the third metal ion was lacking a direct protein bridge (Glu-67) to the second metal ion. Additionally, one of the inorganic phosphate oxygens was directly coordinated to Thr-84, a feature that clearly suggested the collapsed product complex. This oxygen position was previously suggested to be a site for the nucleophilic water molecule in a precatalytic complex. The detailed description of the ternary complexes will be published separately.

Structural Differences among Two Apo Forms and Complexes with Substrates and Metal ions, a Conformation of the Catalytic
Loop-The first loop that directly follows the N-terminal helix (residues 32-43) is a critical catalytic loop. This mobile loop was relatively disordered in the apoenzyme but became more a Substrate, 2.0 mM, was dissolved in 50 mM Tris, pH 8.0, at 80°C. This value is well above the K m for these two substrates (2) so that the V max was obtained by increasing Mg 2ϩ or Mn 2ϩ concentrations.
b The value extrapolated for the apparent K d for the metal ion depends on the substrate concentration because the phosphate ester will also chelate metal ions. ordered upon ligand binding. Thermal factors of the refined apoenzyme structures consistently showed elevated values in this region; this is an indicator of mobility and/or disorder. The structures of AF2372 in complexes with Ins-1-P/Ca 2ϩ or FBP indicated that, in analogy to the behavior of FBPase and MJ0109, the catalytic loop is stabilized by the presence of metal ions and ligands.
The refined conformation of the mobile loop (residues 32-43) was slightly different in the apo form and when a ligand was bound (Fig. 5, A, B, and C). Note the difference in the quality of the electron density comparing the apoenzyme to the liganded structures. The changes in B factors for this region when crystals were soaked in ligand solutions did fully reflect the increased "ordering" of the loop. There was a systematic drop in the B factor for the protein plus ligand structures. The ratio of the average B factor for residues 32-43 to the rest of the molecule approached 1 for both of the substrate/product com-plexes with three metal ions (Ca 2ϩ or Mn 2ϩ ).
Soaking in metal ions alone did not improve loop stability significantly. In fact, crystals soaked or co-crystallized with metal ions alone diffracted poorly (Ͼ3 Å). Under conditions that gave the P3 2 form of the enzyme, the combination of both substrate and metal ions was needed to stabilize the loop in these Tris/PEG 3350 conditions. When the AF2372 apoenzyme was soaked in D-Ins-1-P, as well as other substrate/metal complexes, a conformational change occurred that shifted the loop on average ϳ2.5 Å closer to the metal ions following the behavior observed in MJ0109 structures that contained ligands at the active.
Architectural Differences with Other Members of the Family, a Catalytic Loop Variability-The length of the amino acid peptide chain representing the members of the extended IMPase family differs substantially and extends from 252 residues for MJ109 and AF2372 to 400 for IPPase. Structural alignments carried out with SEQUOIA identified the common catalytic core despite relatively low levels of primary sequence homology. However, as inferred from the size difference, extensive external loops and even small domains were formed as additions to the core domain. The superposition in SEQUOIA indicated that AF2372 and MJ109 were the closest structural homologs and could be superimposed with 209 positions to an r.m.s.d. of 1.5 Å (Table V). There were fewer structurally conserved residues and higher r.m.s.d. when AF2372 was superimposed onto the other IMPase or FBPase structures (Table V). However, two clusters of similarities emerged. The structures in the individual clusters had closer similarities to each other than to proteins in the other cluster. AF2372, MJ0109, and  motion was similar to that observed for MJ0109 and amounts to a displacement of the tip of the loop by ϳ2.5 Å (Fig. 6A). dated unexpected features. While the common core superimposed quite well, the fragments containing the mobile loop formed two structural clusters. The loops of AF2372, MJ0109, and pig kidney FBPase were much closer to each other but further away from the loops of human IMPase, IPPase, and Hal2 that seemed to belong to a separate cluster (Fig 6, B and C).

DISCUSSION
There have been a number of studies directed at elucidating the details of Li ϩ effects on the IMPase superfamily. A series of combined mutational, spectroscopic, and kinetic experiments carried out on the human IMPase implicated two residues, Cys-218 and His-217, in Li ϩ inhibition (8). Both residues are remote from the metal ion binding site and, if confirmed as part of the Li ϩ binding site, they cannot be involved in catalysis. There was speculation that these residues could influence the binding of the metal ion at site 2, which would hinder the release of phosphate and result in inhibition of IMPase activity (8). This proposed mechanism was termed the "bumping hypothesis." Two similar mutations were made in one of the Li ϩ -insensitive archaeal IMPase/FBPase enzymes (MJ0109). Mutation of Arg-198 in the M. jannaschii enzyme (aligned with His-217 in the mammalian enzyme) to His led to complete loss of IMPase activity (25). However, introduction of cysteine in place of Ala-199 (which aligned with Cys-218 of the human IMPase) increased the sensitivity of MJ0109 to Li ϩ (IC 50 decreased from ϳ150 to ϳ 80 mM (25)). Such a relatively small change in Li ϩ sensitivity might be interpreted as stemming from other effects. It is more likely that elements conferring Li ϩ sensitivity involve secondary structural elements that require more than a point mutation.
Random mutagenesis studies carried out on the product of the yeast Hal2 gene provided other hypotheses. The Hal2 gene is an in vivo target of Na ϩ and Li ϩ toxicity, and its overexpression improves the tolerance of higher plants to salt. This enzyme is differentially sensitive to monovalent cations: tolerant of high sodium concentrations but inhibited by lithium. Mutating residues Val-170 and Trp-293 or one of the residues forming a salt bridge between Glu-238 and Arg-152 caused a marked decrease in Li ϩ sensitivity (11). It was speculated that the salt bridge between Arg-152 to Asp-263 might be stabilized by an ionic network involving monovalent ion binding. The hydrophobic interaction between Val-170 and Trp-292 was required for an effective formation of the catalytic core that, when disrupted, changed sensitivity to Na ϩ and Li ϩ more than catalytic activity. Mutations at these positions shifted the IC 50 for wild type Hal2 from 0.07 to 4 mM in E238K mutants and to 0.5 mM in the V70A mutant.
In the case of human IMPase, a two-metal ion-assisted catalytic mechanism was proposed. Li ϩ binding was proposed to interfere with the second metal ion binding site (4), in effect, preventing product release. In the case of archaeal IMPases, we have previously proposed that the catalysis should be described as a three-metal-assisted catalysis (6). Additionally, this proposal led to the hypothesis that the mobile loop was responsible for binding of the third metal ion. According to this hypothesis, the catalytic cycle is initiated when the second and the third metal ions are transiently localized in close proximity to create the water nucleophile. The nucleophile immediately reacts with an incoming phosphate group to effectively hydrolyze it with a reversal of phosphorus configuration. Subsequently, the much more mobile third metal ion facilitates release of the phosphate.
The direct consequence of this mechanism is the following hypothesis for Li ϩ inhibition. The mobile loop is important for localization and binding of the third metal ion, and this metal ion can be replaced by Li ϩ . However, Li ϩ at this site would not lead to the formation of the water-based nucleophile, and enzyme activity would be inhibited. This scenario offers a direct insight into why Li ϩ but not other more abundant monovalent metal ions like Na ϩ or K ϩ are inhibitors. To promote the nucleophile formation, the third metal ion needs to insert deep into the active site. The Mg 2ϩ ionic radius is 0.66 Å. Therefore, only Li ϩ is small enough (0.68 Å) to substitute structurally for the Mg 2ϩ ion. However, because of its charge it would be unable to create a nucleophile, thus inhibiting the enzyme. The much larger Na ϩ or K ϩ ions (0.97 Å and 1.33 Å, respectively) would be able to activate the enzyme by general electrostatic rearrangement but would be unable to substitute structurally for Mg 2ϩ . As is suggested by the difference in ionic radii they must have much lower affinity for the enzyme site. Only at exceedingly high concentrations would they cause significant inhibition. The existence of conformational classes for the mobile loop as found in this paper can be directly correlated with Li ϩ inhibition. The representative enzymes with a smaller and more compact mobile loop, like AF2372, respond to Li ϩ in a much more subdued manner than enzymes with longer and more flexible catalytic loop. Therefore, the IMPase superfamily can be subdivided into two distinct classes as defined by a key structural feature (mobile loop conformation) and its functional importance (Li ϩ inhibitory properties).
The use of "mobile loops" as sensors of monovalent ions is quite common in proteins. There are many examples of ubiquitous sodium/potassium transporters and ion channels that utilize mobile loops and differential cation sensitivity to operate. Na ϩ -and Cl Ϫ -dependent transporters GLYT1 and GLYT2 remove glycine from the synaptic cleft. Extracellular loop 1 in the GLYT2 transporter was identified as the source of "structural heterogeneity" that is involved in the specific effect of lithium on serotonin transport (30). Similarly, the glutamate transporter GLT-1 has a region identified as loop-like that is sodium-sensitive and coupled to drive glutamate transport. When Ser-440 in this loop was mutated to glycine, the GLT-1 mutate could use Li ϩ to drive glutamate transport (31). The Na ϩ /Ca 2ϩ exchangers NCX1 and NCX3 reveal structural domains (i.e. loops) important for differential sensitivity to exter- a The upper row indicates the r.m.s.d. and number of residues in final superposition (in brackets) with 3 Ȧ cutoff and 30°continuation. The lower row indicates the r.m.s.d. and number of residues in total superposition with 8 Ȧ cutoff and 80°continuation.
b Successful after weakening the gap penalty to 6. nal Ni 2ϩ or Li ϩ (32). Although the last of these is a totally separate biological system, this transporter has ion sensitivities that parallel those of Li ϩ -sensitive IMPases.
In the extended IMPase family the mobile loop appears to have a double role. Its mobility is associated with and controlled by the metal ion/substrate binding. The structures of archaeal dual function IMPase/FBPases suggest that this loop mobility is required for the efficient release of product (i.e. phosphate). The second role of the mobile loop is to differentiate the influence exerted by monovalent metal ions. In Li ϩ -sensitive enzymes, the loop appears to be longer and in a more open (farther away from the active site) conformation, while in Li ϩinsensitive enzymes, the mobile loop is shorter and in a more closed conformation. This structural differentiation appears to be useful in classifying IMPase-like proteins. It remains to be seen what is the precise role of the loop in Li ϩ inhibition and how universal is this structural classification. The predictive power of this hypothesis needs to be tested on more representatives of this family of proteins.
Recently, a new human IMPase (35,36) as well as new human BPNTase (37) were cloned. The new inositol monophosphatase termed A2 appears to have significant sequence homology to already characterized Li ϩ -sensitive IMPase. The amino acid sequence indicates a larger protein with N-and C-terminal extensions. However, the significant sequence homology in terms of the length and character is observed around the catalytic mobile loop. That homology would suggest that this enzyme also should operate by the same three-metal ionassisted catalysis, and it also should be Li ϩ -sensitive. The kinetic data (34) suggest that the same should hold true for IPPase and Hal2 (37). Additionally, during the submission process of this paper another enzyme from this extended protein family was characterized by x-ray crystallography (38). The rat 3Ј-phophoadenosione-5Ј-phosphate phosphatase (BPN-Tase) with significant homology to the corresponding human enzyme turned out to posses not only double function but also the ability to catalyze the hydpolysis of 1 and 4 inositol polyphosphate species. This protein is also Li ϩ -sensitive. The structural details appear to support fully our findings concerning the catalytic mechanism as well as inhibition by Li ϩ described previously (6) as well as observations reported in this paper.