Structural Basis for Autoinhibition of CTP:Phosphocholine Cytidylyltransferase (CCT), the Regulatory Enzyme in Phosphatidylcholine Synthesis, by Its Membrane-binding Amphipathic Helix*

Background: The CCT M-domain contains an autoinhibitory (AI) segment, but its mechanism was obscure. Results: The AI helix partly occludes active site access and engages a mobile loop (L2), impeding the dynamics of key catalytic lysine 122. Conclusion: Loop L2 is a key target of the AI clamp. Significance: This work reveals the nature of a regulatory off-switch in an enzyme regulated by membrane binding. CTP:phosphocholine cytidylyltransferase (CCT) interconverts between an inactive soluble and active membrane-bound form in response to changes in membrane lipid composition. Activation involves disruption of an inhibitory interaction between the αE helices at the base of the active site and an autoinhibitory (AI) segment in the regulatory M domain and membrane insertion of the M domain as an amphipathic helix. We show that in the CCT soluble form the AI segment functions to suppress kcat and elevate the Km for CTP. The crystal structure of a CCT dimer composed of the catalytic and AI segments reveals an AI-αE interaction as a cluster of four amphipathic helices (two αE and two AI helices) at the base of the active sites. This interaction corroborates mutagenesis implicating multiple hydrophobic residues within the AI segment that contribute to its silencing function. The AI-αE interaction directs the turn at the C-terminal end of the AI helix into backbone-to-backbone contact with a loop (L2) at the opening to the active site, which houses the key catalytic residue, lysine 122. Molecular dynamics simulations suggest that lysine 122 side-chain orientations are constrained by contacts with the AI helix-turn, which could obstruct its engagement with substrates. This work deciphers how the CCT regulatory amphipathic helix functions as a silencing device.

Amphitropic enzymes are a class of enzymes whose regulatory ligand takes the form of a membrane. How membrane binding activates catalytic function in these proteins is generally not well understood, although recent progress has been made for several lipid-activated kinases (1). CCT 3 is a prototypical amphitropic enzyme whose function is to control membrane phosphatidylcholine (PC) homeostasis (2). It catalyzes the rate-limiting step in PC synthesis, the formation of the headgroup donor, CDP-choline. In addition to its role in maintaining membrane phospholipid compositional balance, targeted gene deletions or siRNA-mediated suppression have revealed the involvement of the major isoform, CCT␣, in lung surfactant production (3), plasma lipoprotein balance (4), size regulation of lipid storage droplets (5), and control of diacylglycerol concentration in the Golgi, which in turn regulates vesicular traffic (6).
CCTs from yeast to humans contain a catalytic ␣/␤ fold (7) followed by a membrane binding region and a disordered C-terminal tail. The ␣/␤ fold catalyzes a nucleophilic attack of phosphocholine on the ␣-phosphate of CTP to generate CDP-choline and diphosphate. The reaction proceeds via a random order sequential mechanism (8). Catalysis is accomplished by surrounding the ␣-phosphates of CTP and phosphocholine with amine-containing groups supplied by several loops forming the active site that provide charge stabilization and adapt the geometry about the ␣-phosphate of CTP (7). Extensive mutagenesis of active site residues has highlighted the key importance of lysine 122 (Lys 122 ) in loop L2. Replacement with other residues (even conservative arginine), cripples k cat /K m by 5 orders of magnitude (9). 4 The membrane binding domain contains an inducible amphipathic helix (domain M) that detects physical properties of a PC-deficient membrane, such as high negative charge density and lipid packing stress (11)(12)(13)(14). An encounter with such a membrane surface triggers insertion of the hydrophobic face of the amphipathic helix (15) and transduces a conformational change in the catalytic domain associated with an increase in k cat /K m of nearly 3 orders of magnitude (16 -18). This process restores the membrane PC content. When not membrane-engaged, this same M domain functions as an autoinhibitory device. Deletion or mutagenesis of domain M results in a ϳ20fold increment in constitutive (lipid-independent) activity with a 10-fold increase in the maximum velocity and a ϳ2-fold decrease in the Michaelis constant (K m ) for CTP (16 -18). The K m for the other substrate, phosphocholine, is not regulated. However, membrane binding, which requires a functioning M domain, imparts an additional order of magnitude increase in catalytic efficiency above the increase observed upon domain M deletion (16). These data suggest that M has a dual role in CCT regulation: as a silencer in the soluble form and as an activator in its membrane-bound form. In this work, we have focused on the autoinhibitory role of M in the silenced, soluble form of CCT.
How does domain M impart catalytic silencing? Previous data suggest that most of domain M has a high degree of disorder in the CCT soluble form. Circular dichroism analyses revealed that the ␣-helical content increases by nearly 60 residues upon membrane binding mostly at the expense of disordered conformers (19). Protease accessibility, photocross-linking, and site-directed fluorescence anisotropy suggested a bipartite structure for domain M in the CCT soluble form: an N-terminal disordered segment (residues ϳ235-270) followed by a more structured segment (residues 272-301) (20,21). This bipartite structure is also predicted by many disorder prediction algorithms and is a feature of CCT M domains across phyla (16,22).
A recent analysis of chimeric enzymes revealed that the regulation of the catalytic domain of rat CCT␣ is tolerant of large variation in the sequence and length of the M domain, suggesting that physical features, such as hydrophobicity and amphipathy, are more important than sequence (16). However, deletion of a relatively conserved 22-mer from the C-terminal end generated a deregulated enzyme with catalytic properties similar to that of CCT-236, which is missing the entire M domain (16). We refer to this 22-mer segment as the AI motif (see Fig. 1A). Denaturation analysis of this segment in the CCT soluble form suggested a weakly folded conformation in contrast to an absence of structure in the rest of domain M (21).
Does the AI motif interact with the catalytic domain and, if so, where? Cysteine engineering of scattered sites within domain M followed by photocross-linking and mass spectrometry identified an association with the helix ␣E pair extending from the base of the active site of the catalytic dimer. This interaction was confirmed by deuterium exchange/mass spectrometry (21). Contacts were broken in the presence of activating lipid vesicles, suggesting a silencing function. Activation of CCT, according to the emerging model, involves the dissociation of the AI motif from ␣E and its association with the membrane surface. Essentially there is a competition between the ␣E helices and the membrane for an emergent AI helix. The flexible leash portion of domain M would serve to facilitate the transitions of the AI helix. Although this work identified three components of the CCT autoinhibitory interaction, the AI motif, the leash, and the helix ␣E docking site, the chemical nature of the docking interaction and the mechanism of silencing were not explained.
Here we present mutagenesis analysis of the AI segment to further define its role as a suppressor of catalysis in the CCT soluble form. We solved the crystal structure of a mammalian CCT containing both the catalytic and M domains, revealing the nature of the AI-␣E interaction and a plausible mechanism for catalytic silencing. Molecular dynamics simulations support that mechanism, which involves a clamp on an active site loop and restriction of the orientation of the key catalytic residue, Lys 122 .

Preparation of CCT Constructs
All CCTs were derived from rat CCT␣ (UniProtKB/Swiss-Prot accession number P19836). The sequence fidelity of all CCT constructs was confirmed by two-strand sequencing.
CCT-367-The preparation of wild-type, full-length CCT-367 plasmid and its expression and purification were described in Taneva et al. (19).
pAX142-CCT Full-length Constructs-The series of domain M variants with serine or alanine substitutions at 6, 8, or 10 hydrophobic residues in the AI motif were created in the COS cell expression vector pAX142. The pAX142 template used for mutagenesis contained a full-length CCT-367 with a single cysteine at residue 288 (21). This version was used to incorporate a fluorophore at one site in that work and has near-WT activity (21). This construct also has an N-terminal His tag that adds the sequence MAKH 6 IEGRSA before the first methionine of CCT. Using pAX142 HisCCT-367(S288C) as template, we created pAX142 versions of CCT-367(6S) and CCT-367 (6A) by making  serine or alanine substitutions at codons 285, 286, 289, 290, 292, and 293 using mutagenic primers and QuikChange mutagenesis. We made CCT-367(8S) using CCT-367(6S) as the template and engineering additional serine substitutions at residues 275 and 278. To make CCT-367(10S), we used the CCT-367(8S) template and another round of QuikChange to engineer additional serine mutations at residues 272 and 274.

Bacterial Expression and Purification of CCT-312, CCT-312(⌬18), and CCT-312(⌬32) for Crystallization
A single colony of Rosetta cells harboring the pET14b-CCT plasmid was grown in 1 liter of LB medium with 0.3 mM ampicillin and 0.1 mM chloramphenicol. After ϳ5 h at 37°C, CCT expression was induced with 1 mM isopropyl 1-thio-␤-D-galactopyranoside and incubation at 20°C for 3 h. Cells were harvested by centrifugation at 4500 rpm for 15 min and stored at Ϫ80°C. The frozen cell pellet (ϳ1.7 g) was resuspended in 30 ml of lysis buffer (10 mM Tris, pH 8.0, 150 mM NaCl, 2 mM DTT, 0.2 mg/ml lysozyme, 1% Triton X-100, and 25 units of the nuclease Benzonase). After incubation for 30 min at 4°C, lysis was completed by probe sonication with three maximum pulses for 30 s on ice. The crude cell lysate was centrifuged at 23,300 ϫ g for 30 min. The supernatant fraction (30 ml) was loaded onto 4 ml of nickel-agarose beads in a column equilibrated with EQ buffer (10 mM Tris, pH 8.0, 150 mM NaCl, 2 mM DTT, and 25 mM imidazole). The beads were washed with 10 ml of EQ buffer with 0.5 M NaCl and then with 10 ml of EQ buffer. The bound His-CCT was eluted with 10 ml of EQ buffer containing 100 mM imidazole followed by 4 ϫ 10-ml volumes of EQ buffer with 350 mM imidazole. Elution fractions with the highest CCT purity (assessed by gels) were pooled and further purified by ion exchange chromatography. HisCCT-312 was purified by SP-Sepharose, and HisCCT-312(⌬32) was purified by Q-Sepharose. Columns (4 ml) were washed with 10 ml of EQ buffer followed by 8 ml each of 200, 300, 400, and 500 mM NaCl in 10 mM Tris, pH 8.0 and 2 mM DTT. The fractions containing purified proteins were combined, concentrated, and buffer-exchanged with 10 mM Tris, pH 8.0, 300 mM NaCl, and 2 mM DTT using a 15-ml centrifugal filter device with 30-kDa cutoff (Millipore).

Bacterial Expression and Purification of CCT Truncation Variants
CCT-255, CCT-271, and CCT-293 were expressed in Rosetta cells as described (23) with minor modifications. The proteins were purified from the 20,000 ϫ g supernatant using nickel-agarose at 4°C (23). The peak elution fractions were dialyzed against 10 mM Tris, pH 7.4, 0.1 M NaCl, 0.15 mM Triton X-100, and 2 mM DTT and were frozen at Ϫ80°C. CCT-304 was insoluble and was isolated from the 20,000 ϫ g pellet fraction after urea denaturation, binding to nickel-agarose, on-resin refolding, and elution with imidazole (21). Peak fractions from the nickel-agarose column were dialyzed against the above buffer prior to storage.

Expression and Purification of Full-length His-tagged CCTs from the COS Cell System
CCT cDNAs incorporated into pAX142 plasmids were expressed in COS-1 cells using 48-h transfection (14). Cell pellets were lysed by sonication in 20 mM phosphate buffer, pH 7.4, 0.25 mM Triton X-100, 2 mM DTT, and a battery of protease inhibitors (24). NaCl was added to 0.5 M, and the lysates were sedimented at 20,000 ϫ g for 10 min. HisCCT-367(8S) was purified from this supernatant using nickel-agarose (23). His-CCT-367(6A), -(6S), and -(10S) were mostly insoluble and were purified from the 20,000 ϫ g pellet after washing the pellet by brief sonication in 10 mM Tris, pH 7.4, 0.5 M NaCl, 0.25 mM Triton X-100, protease inhibitors, and 2 mM DTT; collecting the insoluble material; and solubilizing it in 8 M urea. The ureadenatured HisCCT-367(6A) was purified and refolded using nickel-agarose chromatography (21). HisCCT-367(6S) and -(10S) were highly enriched in the supernatant after urea solubilization and were renatured slowly at 4°C over ϳ24 h by sequential dialysis against a 300-fold volume excess of 10 mM Tris, pH 7.4, 0.1 M NaCl, 0.15 mM Triton X-100, and 2 mM DTT containing 5 M urea followed by 2.5 M urea and finally 0.5 M urea. The dialyzed sample was centrifuged 10 min at 20,000 ϫ g, and the supernatant was saved for analysis. Although these concentrated protein preparations contained 0.5 M urea (to maintain solubility upon freeze-thaw), the urea concentration after dilution in the CCT activity assays was only ϳ4 mM. When assayed in the presence of saturating concentrations of PC/PG vesicles CCT-367(6S) and CCT-367(6A) had WT activity, suggesting complete refolding.

CCT Enzyme Activity Analysis
All CCT variants were analyzed using a standard assay (23,25) in which substrate concentrations are optimal for WT CCT (16) in the absence or presence of egg PC/egg PG (1:1) sonicated vesicles (26). Lipid concentrations were 0.25-0.5 mM for maximal activation depending on the variant. Kinetic constants were obtained in separate trials where the CTP concentration was varied from 0 to 36 mM while maintaining the phosphocholine concentration at 2 mM. The velocity versus CTP concentration curves were evaluated using Prism 4 software and were fit to the Michaelis equation and K m is the Michaelis constant. The constant k cat was obtained by dividing the V max by the molar concentration of the CCT.

Structure Determination and Refinement
Both structures were solved by molecular replacement using Phaser (30) in CCP4 6.1.3 (29). For the CCT-312 8.0-Å diffraction set, we initially used the CCT catalytic domain dimer excluding water (7) (Protein Data Bank code 3HL4) as the search model and found a molecular replacement solution that revealed the catalytic domain as well as additional tubular electron density adjacent to the ␣E helices. Subsequent model building and refinement resolved additional continuous and tubular electron densities that connected catalytic domains. These were modeled as extended polyalanine helices. The final model was tetrameric (dimer of dimers). After we had solved the 3-Å CCT-312(⌬32) dimer structure, we used it as a search model for the 8.0-Å data set instead of the CCT-236 structure. The structural outcomes of the molecular replacement and subsequent refinement were virtually the same. But there were significant statistical improvements. Both the log-likelihoodgain (LLG) score in Phaser and R work /R free values in REFMAC improved. The crystal structure had four dimers in the asymmetric unit, a very high calculated solvent content of 77.7%, and a Matthew's coefficient of 5.5 Å 3 /Da (31). For CCT-312(⌬32), the molecular replacement solution placed one dimer in the asymmetric unit, and the structure had a calculated solvent content of 52.4% and a Matthew's coefficient of 2.6 Å 3 /Da. For both structures, the membrane binding domain and AI motifs were built by several iterated cycles of manually fitting the corresponding electron density using the program Coot (32) followed by refinement using PHENIX (33) and REFMAC5 (34).

CD Analysis of the Secondary Structure of CCT-312 in Solution
CCT-312 (2 M) in 0.22 m-filtered 50 mM phosphate buffer, pH 7.4 was incubated for 2 min at 25°C without or with lyso-PC/egg PG (4:1, mol/mol) micelles (19). CD spectra were recorded using a Jasco J-810 spectropolarimeter at 25°C in a cell of 0.5-mm path length at a scanning rate of 100 nm/min and a bandwidth of 1 nm. Spectral resolution was 0.5 nm, and three scans were averaged per spectrum. All spectra were smoothed and corrected for the appropriate backgrounds (buffer or buffer ϩ micelles) and were converted to units of molar ellipticity. Protein concentrations were determined by absorbance at 280 nm and calculated extinction coefficients. Secondary structure content was deconvoluted from the CD spectra by CDPro and protein reference set 7, which includes 43 soluble and five denatured proteins (35).

Molecular Dynamics Simulations
Simulations with and without CDP-choline were based on rat CCT␣-236 (Protein Data Bank code 3HL4) (7) residues 40 -236. CDP-choline was parametrized in a two-step process. We used Protein Data Bank code 3HL4 as a starting structure for density functional theory calculations through the "RESP ESP charge Derive Server" (R.E.D. Server) (36). The results were ported to the general AMBER force field (37) for the CDPcholine molecule for use with AMBER protein force fields. For testing the effects of bound AI motifs, the starting structure was the 3-Å CCT-312(⌬32) in this study. Two systems were created, one with and one without the bound AI helices. Neither of these contained bound CDP-choline. AI peptides were embedded in the starting structure as separate chains, and their termini were charged.
All simulations used version 4.6 of the GROMACS MD package (38) with the AMBER99SB-ildn atomistic force field (39) and the TIP3P (transferable intermolecular potential 3P) (transferable intermolecular potential 3P) model for water (40). All systems were hydrated to ϳ1.2 nm around the dimer using dodecahedron-shaped periodic boundary conditions. The systems were neutralized with an excess of Na ϩ , and the water was ionized to create a 0.15 M NaCl solution. The simulations were performed in an NPT ensemble with T ϭ 310 K and P ϭ 1 atm. The temperature was held constant using the stochastic "v-rescale" thermostat with a coupling parameter of 2.0 ps (41), and the pressure was maintained by isotropic "Parrinello-Rahman" coupling (42) with a p of 5.5 ps and a compressibility of 4.5 ϫ 10 Ϫ5 bar Ϫ1 . Bonds were constrained using the LINCS algorithm (43) with a 2-fs time step. The van der Waals potential was cut off at 1.4 nm after being switched at 1.2 nm, and the electrostatic potential was also cut off at 1.4 nm to use the Verlet neighbor searching algorithm. Particle mesh Ewald (44) was used for long range electrostatics with a grid spacing of 0.1 nm. The CCT-236 simulations were equilibrated for 200 ps before analysis. The CCT-312(⌬32) systems were equilibrated for 10 ns to test the stability of the binding between the AI and ␣E helices. All systems were then evolved for a 200-ns production run, sampling every 5 ps. The finished trajectories were fitted (translated and rotated with solution removed) to the very stable ␤-sheets of both monomers. Trajectories were analyzed with VMD molecular visualization software (45) and GRO-MACS tools.
The structures reported here have been deposited with the Research Collaboratory for Structural Bioinformatics (RCSB) under Protein Data Bank codes 4MVC (CCT-312(⌬32) and 4MVD (CCT-312).

RESULTS
In the CCT Soluble Form (CCT sol ) the AI Segment Suppresses k cat /K m -We previously showed that deletion of the segment between residues 272 and 293 generated a desilenced CCT sol with a maximal velocity nearly identical to that of the deletion of the entire regulatory tail (16). This result suggests a primary autoinhibitory role for the motif. However, one could argue that a large deletion within a domain could disrupt the structure and interactions of the whole domain. To further define the boundaries of the AI motif, we constructed a series of systematic truncations to test the role of the positively charged N-terminal portion of the M domain (CCT-255), the non-conserved midsection of domain M (CCT-271), and the AI motif (CCT-293 and CCT-304) (Fig. 1A). We determined the k cat and CTP K m values for these CCTs in the absence of lipids.
Upon loss of the segment between residues 312 and 271, k cat values for the CCT soluble form transitioned from silenced (0.5 s Ϫ1 ) to the partially active rate (ϳ 3 s Ϫ1 ) characteristic of CCT-236 (Fig. 1B). The construct ending at residue 271 showed a higher k cat than did CCT-255 or CCT-236, perhaps reflecting different consequences of specific truncation sites on protein dynamics. Loss of the AI segment was linked to a transition in the CTP K m from ϳ9 to ϳ5.5 mM (Fig. 1C). Thus, the presence of the AI segment suppresses k cat /K m by about 10-fold. However, the kinetic parameters of the full-length soluble enzyme were achieved only when the segment between residues 293 and 312 was intact. These results strengthen previous results implicating a key silencing role for the AI motif (16) but suggest additional participation of the flanking segment between resi-dues 293 and 312. For comparison, fully active, membranebound CCT (CCT mem ) features a k cat value of ϳ13 s Ϫ1 and a K m of 0.5-1 mM.
The Hydrophobicity and Amphipathy of the AI Segment Are Important for Silencing Function-The most conserved feature of the AI segment is the spacing of hydrophobic residues, which tend to be aromatic and/or bulky (Fig. 1D). The effects of substitution of sets of these to alanine or serine on the lipid-independent specific activities (expressed as a fraction of maximal activity obtained with saturating concentrations of lipid) of fulllength CCT constructs (CCT-367) are shown in Fig. 1E. As the hydrophobicity of domain M was reduced by mutation, the fractional lipid-independent activity increased from 0.03 (WT CCT-367) to 0.18 (CCT-367(10S). Thus, silencing function clearly correlates with hydrophobicity of the AI. Silencing function also correlated with the amphipathy value (Fig. 1E, lower  panel), consistent with an amphipathic helix structure for the AI motif. The fractional lipid-independent activity for CCT-367(10S) was only 0.18 compared with 0.27 for a CCT with a deletion of the entire AI segment (16) or 0.44 for CCT-271 truncated just before the AI motif. These data suggest that although the hydrophobic residues within the AI segment con-tribute to catalytic silencing there are likely other interactions that participate.
The Crystal Structures of CCT312 and CCT-312(⌬32) Show a Four-helix Bundle Interaction Involving the AI and ␣E Helices-To understand the mechanism for silencing by the AI segment, we sought to obtain the structure of CCT-312. CCT-312, which is missing only the disordered phosphorylation (P) region, has catalytic and regulatory properties similar to full-length CCT  (Ref. 22 and Figs. 1, B and C, and 2). Extensive screens with CCT-312 generated only poorly diffracting crystals, one of which generated an 8-Å diffraction data set. To obtain crystals with better diffraction quality, we screened CCT-312 variants with various deletions from the disordered leash portion of domain M. Deletion of the 18-residue (residues 252-269) nonconserved portion of the leash (CCT-312(⌬18)) had minimal impact on silencing or lipid activation (Fig. 2), indicating that this portion of the leash is dispensable for regulation of activity. However, crystallization trials were unsuccessful. CCT-312(⌬32) missing 32 residues (residues 238 -269) from both the non-conserved and polybasic sections of domain M readily crystallized in the presence of CDP-choline. This construct showed strong silencing in the absence of lipid and partial activation in the presence of excess lipids (Fig. 2B). The lack of full activation is likely because the abbreviated M domain only weakly binds the membranes. Titrations with PC/PG (4:1) vesicles showed equivalence among CCT-367, CCT-312, and CCT-312(⌬18) but a 10-fold higher EC 50 for CCT-312(⌬32) (Fig. 2C). A CCT-312(⌬32) crystal generated a 3-Å resolution diffraction data set that was solved by molecular replacement using CCT-236 (Protein Data Bank code 3HL4) (see Table 1). The asymmetric unit was a dimer. The resolved portion of this dimer encompasses the catalytic domains and the AI segments of domain M (Fig. 3A). The segments linking the end of the ␣E helices to the AI helices are not resolved, which is indicative of their disorder. Because of this, there is uncertainty in assigning the two AI helices to chain A or B. We refer to them as AI-1 (resolved residues 275-295) and AI-2 (resolved resi-  dues 276 -295) (see Fig. 3B). Each AI helix interacts with both ␣E helices, and together they form an antiparallel fourhelix bundle near the opening of the active site. The two AI segments have very similar conformations with a bend at Lys 281 . AI-2 is unraveled between residues Gln 276 and Trp 278 . Each AI helix ends in a sharp turn at 294 GP (the AI-turn), which interacts with loop L2 at the mouth of the active site (discussed below).
The AI positioning in the CCT-312(⌬32) structure is not an artifact of the large domain M deletion. A very similar AI-␣E-L2 interaction is found in CCT-312 with no deletion. We initially used the 2.2-Å CCT-236 structure and subsequently the 3.0-Å CCT-312(⌬32) structure as a search model for the 8.0-Å data set of CCT-312. Both yielded very similar molecular replacement solutions and refined models ( Fig. 4 and Table 1). But there were significant statistical improvements using CCT-312(⌬32), indicating that the CCT-312(⌬32) structure was the more accurate search model. This suggests that the positioning of the AI segments in the CCT-312(⌬32) structure reflects the positioning of those of the CCT-312 with a complete M domain.
Unexpectedly, the refined CCT-312 structure revealed a tetrameric complex (a dimer of dimers) with the M domains in mostly full ␣-helical conformation extending directly from the ␣E helices of one catalytic dimer to the active site of another dimer (Fig. 4). The M helices of the two dimers interact to form a very long and loose antiparallel four-helix bundle. Gel filtration of this same preparation of CCT-312 showed a mixture of dimer and tetramer species. CD analysis showed a large increase in helical content (ϳ52 residues) induced by lipid  micelles (Fig. 5). These data suggest that in solution the dominant CCT-312 dimeric form has mostly disordered M domains and is refractory to crystallization and that only the tetrameric species with its ordered M domains was amenable to crystallization in the given condition. Despite the unexpected M domain arrangement, in both refined tetrameric models, the density for the C-terminal ϳ15 residues of each M helix rests against the ␣E helices and terminates at the L2 loop. We can assign these terminal residues to the AI segment as the density can be traced continuously from ␣E as an extended M helix of known sequence (Fig. 4).
The AI-␣E Helical Interaction Involves Hydrophobic and Polar Contacts-Two ␣E helices and loop L2 create a shallow elongated binding cavity for each AI segment. The residues contributing to the cavity are indicated in Fig. 6A and Table  2. The ␣E helices are longer than observed in the structure of CCT-236 where the visible density ended at residue 215 (chain A) or 216 (chain B) (7). This suggests stabilization of the ␣E helices by the interacting AI helices. The AI helix, an amphipathic helix with a conserved hydrophobic face and a less conserved polar face, uses the ␣E hydrophobic cavity to bury its aromatic residue-rich non-polar face. The mutagenesis results of Fig. 1E support a silencing function for these interactions. The specific contacts between residues in the AI and ␣E helices are described in Tables 2 and 3. The van der Waals interface between the helices is rather loose, featuring ridges-into-grooves packing involving mostly bulky side chains. There are no ␣E-AI packing interactions involving backbone contacts. In total, 30 hydrophobic side chains, including 12 aromatic side chains, are sequestered partially or fully away from water in the formation of the ␣E-AI four-helix bundle ( Table 2). Most of the charged side chains in the AI segment facing the solvent did not have strong densities, reflecting the absence of constraining interactions.
Although the interacting surfaces in the four-helix bundle that forms from two ␣E and two AI helices are mainly hydrophobic, there is a network of polar interactions near the base of the bundle (Fig. 6B). Both AI helices feature Trp 278 and Ser 282 contacts with the ␣E residues Asp 214 and Arg 218 , respectively. However, there is variation in the H-bonding network in AI-1 versus AI-2 due to different side chain rotamers for Asp 214 , Ser 282 , and Ser 288 . In AI-1, Ser 288 makes a novel H-bond to Tyr 216 in ␣E of chain B. In AI-2, Ser 282 makes a novel H-bond to Asp 214 in ␣E of chain B (Fig. 6B and Table 3). Thus, there may be more than one low energy binding interaction between the AI and the ␣E, and two have been captured in our structure.
The AI Segment Makes Direct Contact with Loop L2-The C-terminal end of the AI helix and AI-turn at 292 MFGP 295 is juxtaposed with residues 122 KGF 124 of loop L2 (Fig. 6C). The backbone of the AI helix-turn at Met 292 , Phe 293 , and Gly 294 are within H-bonding distance of backbone atoms at Phe 124 and Lys 122 in loop L2 (Table 2). At the C terminus of the AI helix, the interaction surface contributed by the catalytic domain is sparse. Two hydrophobic residues at the end of the AI helix, Met 292 and Phe 293 , interface only loosely with the side chain of Phe 124 in L2. The bulky side chain of Tyr 216 in helix ␣E may serve to guide the AI helix-turn to contact the L2 loop.
The Architectures of the CCT-312(⌬32) and CCT-236 Active Sites Are Similar-An overlay of the active sites of the partially active CCT-236 (Protein Data Bank code 3HL4) and the totally silenced CCT-312(⌬32) are surprisingly similar (Fig. 7A). We were anticipating that at a minimum in CCT-312(⌬32) the L2 loop would be displaced to realign Lys 122 . One explanation for the similarity is that the bound CDP-choline ligand can dictate the positions of the active site residues, essentially overriding the effects of the bound AI helix. Active site ordering by CDP-choline is supported by molecular dynamics simulations of CCT-236 with and without CDP-choline (Fig. 7, B-D). Root mean square fluctuation analysis of all backbone atoms during a 200-ns simulation revealed high mobility in active site elements L1, L2, L4/␣D, and L6/␣E (Fig. 7B). A CDP-choline stabilizing effect was observed for these mobile elements except for the C-terminal portion of helix ␣E, which unraveled in both simulations (Fig. 7, B and C). Stabilization was most pronounced for loop L2 (Fig. 7, C and D). During the simulation in the absence Residues in the interacting face of the AI-1 segment are listed in column 1, and the residues in the catalytic domain with atoms located within 4 Å of each AI-1 residue are listed in the other columns. For the AI-2 segment the residues on each chain would be the reverse. Except where noted (*,**), all interactions are side-chain van der Waals contacts. *, Hydrogen bonding side-chain interaction. **, Interaction between backbone atoms.    (Fig. 7D). If CDP-choline can dictate the arrangement of the active site, then a ligand-free structure might reveal a more accurate effect of the bound AI segment.

MD Simulations Reveal That the AI Helix Forges Interactions with Loop L2 That Constrain Lys 122 Orientations-Simulations
of the CCT-312(⌬32) dimer were performed using starting structures with the AI helices present or removed. The ligand CDP-choline was not included in the 210-ns simulations. For both simulations, fluctuation analysis of all backbone atoms revealed high mobility in the active site loops and helix ␣E as in the simulations of CCT-236. We focused on AI-dependent changes in the ␣E helices and L2 loop as these form the AI binding site in the solved structures. In the absence of the AI segments, portions of the ␣E helices unraveled locally within 20 ns and did not reform over the duration of the simulation. In the simulations with the AI segments included, the AI helices remained bound to the ␣E helices to maintain a stable bundle of four ␣-helices. As a result, the AI-turn remained in contact with loop L2 for the full duration. In the presence of the AI, the dynamics of the loop L2 backbone, monitored as root mean square deviation (r.m.s.d.) from the starting structure, showed a ϳ1-Å reduction in r.m.s.d. and a dramatic reduction in fluctuations (Fig. 8A). The restraining interaction of the AI-turn on L2 was mediated by backbone contacts as in the crystal structure and by side chain interactions between Phe 124 in L2 and Met 292 and Phe 293 in the AI helix.
This AI-turn-L2 interaction had profound consequences on the orientation of the Lys 122 side chain. During the simulation without the AI segment, the Lys 122 side chain sampled multiple, but preferred, orientations (Fig. 8B). In the first ϳ60 ns of the simulations with the AI helices, the Lys 122 side chains at both active sites forged intermittent interactions with the backbone of the AI-turn. Then between 60 -70 ns, the hydrophobic pair, Phe 293 in the AI helix-turn and Phe 124 in the L2 loop, assumed new rotameric states that resulted in a more intimate contact between them and positioned their carbonyl oxygens for interaction with the Lys 122 ⑀-amino (Fig. 8, C and D). This interaction remained intact for the duration of the simulation in chain B and for ϳ20 ns in chain A. The Lys 122 side chain positioning continued to be constrained in chain A by a fluctuating contact primarily with the Phe 293 carbonyl. Thus, the Lys 122 side chain became trapped in a more stable polar interaction with the   AI-turn. These effects of the AI on the dynamics and positioning of Lys 122 were also revealed in an analysis of the distance fluctuations over time between two backbone atoms located in the vicinity of the AI-turn: the Phe 124 carbonyl oxygen and Ser 203 amide nitrogen (Fig. 8E). The effects are consistent with a movement of the Lys 122 side chain and ⑀-amino group toward the AI-turn where it localizes between the carbonyls of Phe 293 in the AI-turn and Phe 124 in the L2 loop.

DISCUSSION
Of the many enzymes that have autoinhibitory domains that double as membrane-binding motifs, there are only a few for which the switching mechanisms between silenced and activated states are understood at a detailed structural level, including the protein kinases PKC and PKB and the GTPase exchange factors ␤2-chimerin and Grp1 (1, 46 -48). These proteins use modules that bind a lipid monomer within the membrane, such as C1 domains for diacylglycerol or pleckstrin homology domains for phosphatidylinositol 4,5-bisphosphate. Among the class of amphitropic proteins that rely exclusively on lipidinduced amphipathic helices for membrane binding, the switch mechanism has been elucidated at the atomic level only for pyruvate oxidase from Escherichia coli (49,50). In that enzyme, the autoinhibitory C-terminal 40 residues take on a mixed ␣/␤ ϩ extended tail structure that occludes most of the active site and dislocates a key active site residue, Phe 465 , that in the activated form likely mediates electron transfer between bound cofactors (49). This inhibitory structure reorganizes to form a linker and an amphipathic helix in the membranebound state (50). Our analysis of CCT suggests an autoinhibition mechanism, like that of pyruvate oxidase, involving substrate occlusion and dislocation of a key active site residue, but there are also substantial differences, such as the intrinsically disordered nature of a large portion of the regulatory M domain.
The mechanism of autoinhibition by the CCT domain M that is emerging involves an ordered, autoinhibitory amphipathic helix on a flexible leash. A structureless leash, suggested by previous spectroscopic evidence (19 -21), is supported by the lack of density for domain M residues outside of the AI helix in our 3.0-Å structure of the CCT-312(⌬32) dimer. Although the CCT-312 tetrameric structure showed an ordered helical conformation for the entire M domain, we have argued above that this represents a minor conformer in the ensemble that was more amenable to crystallization. The leash does not directly mediate catalytic silencing as deletion of 18 leash residues did not increase lipid-independent activity. The ␣E-AI silencing interaction, previously based on low resolution methods (16,21), is strengthened by the interactions revealed in both new x-ray-derived structures.
The major autoinhibition mechanisms that have been elucidated for allosteric enzymes include (a) direct steric block of active sites using pseudosubstrate motifs (51) or unoccupied ligand binding domains (46), (b) allosteric clamping of the active site in "closed" or non-productive configurations by unoccupied ligand binding domains (47,52), and (c) disabling of the active site by dislocating a key substrate-binding element (46,53). There is also evidence that allosteric mechanisms can impact protein dynamics without a change in structure (54). Our data are compatible with a combination of these strategies contributing to autoinhibition by the CCT AI helix.
The rate-limiting step in the overall catalytic mechanism is not known for CCT. The reduction in k cat /K m by the AI helix-turn implies an impact on the rate associated with the capture of substrate into a productive ES complex, but the product release step could also be inhibited by the AI helixturn. Fig. 9 illustrates how the opening to the active site is partly covered when the AI segment binds to the ␣E helices; thus, substrate access and/or product release could be restricted. The interaction of the AI helix-turn with loop L2 is also likely to contribute to its inhibitory mechanism by clamping L2 in a non-productive orientation and/or by constraining its dynamics. This interaction is backbone to backbone and may be stabilized by a cluster of hydrophobic interactions involving residues from the AI-turn and L2 in addition to the stable four-helix bundle. Molecular dynamics simulations probing the effect of the bound AI segment support a role for this hydrophobic cluster. We propose that this interaction impedes loop L2 movements and constrains the conformational space of the key catalytic residue on loop L2 during a catalytic cycle.
The evidence for loop L2 movement during a catalytic cycle is from a comparison of bacterial glycerol-phosphate cytidylyltransferase structures with CTP bound versus CDPglycerol product bound that revealed a ϳ3-Å displacement of loop L2 backbone into the active site to engage the product (55). This loop closure may facilitate the transition state by bringing the ⑀-amino group of the Lys 122 analog into contact with both the ␣-phosphate of CTP and the phosphate oxygen of the attacking nucleophile (55). It would also exclude water, optimizing the electrostatic interactions between the anionic substrates and basic active site groups (56). After the reaction, loop L2 would open to release product. The interaction of the AI-turn could hamper any of these loop movements.
Spectroscopic measurements of active site loop closure and opening dynamics have yielded correlation times ranging from ns to hundreds of s (56 -60). The time scale of the MD simulations we conducted (200 ns) did not capture the full L2 loop displacement. However, the simulations did reveal a profound and unanticipated effect of the AI-L2 interaction on the positioning of Lys 122 : the AI-L2 contact reorganizes a cluster of hydrophobic residues contributed from both segments that stabilizes a polar contact between the Lys 122 ⑀-amino and a backbone atom of the AI-turn. Whether this trap for the catalytic lysine would impact its interaction with substrates will be probed by additional simulations in the presence of individual substrates and products.
Our analysis thus far suggests a plausible silencing mechanism that explains the effect of the AI helix-turn on the development of a kinetically productive ES complex leading to product formation (k cat /K m ). Displacement of the AI segment upon engagement with membranes would disrupt the inhibitory interaction that blocks the opening to the active site and constrains the movements of loop L2 and Lys 122 .
CCT is up-regulated in ras-transformed epithelial cells and is required for maintaining transformed cells in an anoikis-resistant state (61), a prerequisite for metastasis. Establishing the mechanism of autoinhibition of CCT is important for developing inhibitors based on the AI motif that might be developed as potential suppressors of metastasis.