Allostery Is an Intrinsic Property of the Protease Domain of DegS

DegS is a periplasmic Escherichia coli protease, which functions as a trimer to catalyze the initial rate-limiting step in a proteolytic cascade that ultimately activates transcription of stress response genes in the cytoplasm. Each DegS subunit consists of a protease domain and a PDZ domain. During protein folding stress, DegS is allosterically activated by peptides exposed in misfolded outer membrane porins, which bind to the PDZ domain and stabilize the active protease. It is not known whether allostery is conferred by the PDZ domains or is an intrinsic feature of the trimeric protease domain. Here, we demonstrate that free DegSΔPDZ equilibrates between active and inactive trimers with the latter species predominating. Substrate binding stabilizes active DegSΔPDZ in a positively cooperative fashion. Mutations can also stabilize active DegSΔPDZ and produce an enzyme that displays hyperbolic kinetics and degrades substrate with a maximal velocity within error of that for fully activated, intact DegS. Crystal structures of multiple DegSΔPDZ variants, in functional and non-functional conformations, support a two-state model in which allosteric switching is mediated by changes in specific elements of tertiary structure in the context of an invariant trimeric base. Overall, our results indicate that protein substrates must bind sufficiently tightly and specifically to the functional conformation of DegSΔPDZ to assist their own degradation. Thus, substrate binding alone may have regulated the activities of ancestral DegS trimers with subsequent fusion of the protease domain to a PDZ domain, resulting in ligand-mediated regulation.

Proteases are carefully regulated to ensure that these destructive enzymes are only activated under the proper circumstances. For example, most extracellular proteases are synthesized as inactive proenzymes or zymogens (1). In the trypsin proenzyme, the Ser-His-Asp catalytic triad is properly formed, but the oxyanion hole of the active site is malformed and only rearranges to the active structure following the processing event that produces the mature enzyme (2). The activities of intracellular trypsin-like proteases are also regulated by changes in active site conformation. In the DegS protease, for instance, alternative conformations of the oxyanion hole result in an equilibrium distribution of inactive and active enzymes whose populations are controlled by the binding of substrates and allosteric effectors (3)(4)(5)(6)(7)(8).
Escherichia coli DegS is a trimeric HtrA family enzyme consisting of a trypsin-like protease domain, a PDZ domain, and a sequence that anchors the protease to the periplasmic surface of the inner membrane (3,9). The trimer is stabilized by interactions between the protease domains. Protein folding stress in the periplasm activates DegS cleavage of RseA, a transmembrane protein whose cytoplasmic domain binds and inhibits the E transcription factor (10,11). DegS cleavage of RseA initiates a cascade of subsequent cleavages by other proteases, resulting in release of E and transcriptional activation of genes whose products are needed to repair damage caused by envelope stress. The PDZ domain of DegS plays a critical role in signal transduction. Specifically, C-terminal peptides of misfolded outer membrane porins (OMPs), 2 which end with a conserved Tyr-Xaa-Phe motif, bind to the PDZ domain, and activate cleavage of RseA ( Fig. 1A and Ref. 3). Indeed, DegS cleavage of RseA in vitro can be accelerated more than 500-fold by OMP peptides (6,7).
OMP peptide activation of DegS appears to occur by relieving inhibitory contacts between the PDZ domain and the protease domain. For example, salt bridges between the PDZ domain and the protease domain are observed in crystal structures of peptide-free DegS, which has a malformed oxyanion hole, but these interactions are absent in peptide-bound DegS, which has a functional oxyanion hole (4,5). Importantly, disrupting inhibitory interactions between the PDZ domain and the protease domain by point mutations or by deleting the PDZ domain entirely (DegS ⌬PDZ ) activates cleavage of RseA to levels observed with some OMP peptides (6,7). Moreover, crystal structures of DegS ⌬PDZ closely resemble the active conformation of intact DegS (Fig. 1, B and C) and have a functional oxyanion hole (6,12). These results support an allosteric model in which the unliganded PDZ domain stabilizes the inactive protease domain conformation, whereas OMP peptide binding to the PDZ domain reduces or eliminates this inhibitory effect.
Although both OMP peptides and the RseA substrate bind more tightly to active DegS than to inactive DegS, neither sat-urating peptide nor saturating RseA is sufficient to shift the majority of enzymes into the active conformation (7). As a consequence, substrate cleavage is positively cooperative with respect to both RseA concentration and OMP peptide concentration. Moreover, because OMP peptides and RseA bind both to inactive and active DegS, it cannot be assumed that activity measured in the presence of saturating peptide and saturating substrate corresponds to the maximal cleavage rate that would be observed if all enzymes assumed the active conformation (7). In fact, different OMP peptides support very different maximal levels of RseA cleavage. Moreover, cleavage activity can be increased above the wild-type DegS level by mutations that disrupt inhibitory interactions and by mutations that stabilize the active conformation (6,8).
Is allostery an intrinsic property of the protease domain of DegS, or is it only conferred by the presence of the PDZ domains? The protease activity of the DegS ⌬PDZ trimer is substantially lower than that of full-length DegS stimulated by OMP peptides with the highest activities (6,7,12). There are two potential explanations for this fact. DegS ⌬PDZ itself may equilibrate between inactive and active conformations (Fig.  1D), and substrate binding may be insufficient to shift most enzymes into the active state. Alternatively, all DegS ⌬PDZ enzymes may assume the "functional" conformation, but this structure may have lower activity than peptide-activated fulllength DegS because of subtle changes in structure or dynamics. For example, based on differences observed in crystal structures, it was concluded that DegS ⌬PDZ has "a less ordered active site with reduced proteolytic activity" (12).
To probe mechanism, we characterized structure-activity relationships for variants of E. coli DegS ⌬PDZ . Importantly, we found that the H198P mutation decreased K m , increased V max , and essentially eliminated the positive cooperativity of substrate cleavage when compared with wild-type DegS ⌬PDZ . These results strongly support a model in which positive cooperativity arises for wild-type DegS ⌬PDZ because substrates bind more tightly to the active than inactive conformation and are inconsistent with a model in which positive cooperativity is a consequence of substrate-substrate interactions as was suggested previously (6). We also found that the protease activity of DegS ⌬PDZ could be decreased by many of the same mutations that destabilize the active conformation of full-length DegS. Moreover, crystal structures of DegS ⌬PDZ , H198P DegS ⌬PDZ , and inactive DegS ⌬PDZ mutants define two basic conformations with functional and non-functional oxyanion holes, respectively. Thus, our combined results support a two-state model in which allostery is an intrinsic property of the protease domain of DegS.

EXPERIMENTAL PROCEDURES
Variants of E. coli DegS ⌬PDZ (residues 27-256) with an N-terminal His 6 tag and lacking the wild-type membrane anchor were purified by Ni 2ϩ -nitrilotriacetic acid affinity chromatography and gel filtration on a Superdex 200 column (6,7). An 35 S-labeled variant of the periplasmic domain of E. coli RseA (residues 121-216) with a C-terminal His 6 tag and a His 6tagged variant of DegS (residues 27-355) were purified as described (6,7). Mutations were generated in plasmid-borne genes by the QuikChange method (Stratagene). The synthetic YYF peptide was purified by HPLC. Modification with diisopropylfluorophosphate (DFP) was performed as described (8).
Cleavage assays were performed as described (6,7). Enzymes were incubated with 35 S-labeled RseA for different times at room temperature (23 Ϯ 1°C), and cleavage was quantified by scintillation counting of acid-soluble radioactivity. Michaelis-Menten curves were fitted to the Hill equation using the nonlinear least square subroutine in KaleidaGraph (Synergy software).
DegS ⌬PDZ variants were crystallized at 20°C by hanging drop vapor diffusion crystallization using protein stocks at 10 -20 mg/ml in 50 mM NaHPO 4 (pH 8.0), 200 mM NaCl, 5 mM EDTA, and 10% glycerol. All crystals were obtained after 1:1 mixing with 50 mM sodium cacodylate (pH 6.5), 50 -150 mM sodium citrate, and 15-20% isopropanol and were cryoprotected by using 30% methylpentanediol. Diffraction data were collected using a Rigaku MicroMax008-HP rotating source or at the NE-CAT 24-ID-C beamline at the Argonne National Laboratory Advanced Photon Source. Initial phases were obtained by molecular replacement using PHASER (13). Structures were determined by cycles of model building using COOT and refinement using PHENIX (14,15). To obtain the best possible geometry, all structures were refined with riding hydrogen atoms, which were added using REDUCE (16). This procedure does not result in the introduction of additional parameters during refinement as the positions and B-factors of hydrogens are determined by the atoms to which they were attached. Refinement statistics for all final structures were also determined after stripping the hydrogens (without allowing changes in coordinates or B-factors of other atoms). As shown in Table 1, removing the hydrogens resulted in very small differences in R work and R free . Model building was guided by Mol-Probity (17), which was used to identify portions of the model with poor geometry or steric clashes for subsequent correction. In the final structures, fewer than 0.2% of residues had poor side-chain rotamers, none were Ramachandran outliers, and 98% or more were in the most favored regions of the Ramachandran plot (Table 1).
For the structural comparisons shown in Fig. 5, individual subunits were initially aligned to a single reference structure using PyMOL. For each pairwise combination of structures, a displacement between corresponding C␣ positions (e.g. residue 60) was calculated, and then these values were averaged over the entire set of comparisons involving that residue. If the C␣ atom was missing in one or both structures, a displacement between 5 and 7 Å was assigned using a random number algorithm.

H198P Mutation Increases DegS ⌬PDZ Activity and Active Site
Reactivity-Each wild-type or mutant variant of DegS (residues 27-355) or DegS ⌬PDZ (residues 27-256) lacked the wild-type membrane anchor and contained His 6 affinity tags as described (3,6). We previously found that the H198P substitution in DegS increased basal RseA cleavage in the absence of OMP peptide and enhanced peptide-stimulated cleavage under conditions of limiting substrate (8). These results were explained by crystal structures that revealed packing contacts between Pro 198 and Tyr 162 , which stabilize the active oxyanion hole and thus shift the conformational equilibrium to favor active DegS (8). As a result, the H198P mutation should only increase DegS ⌬PDZ activity if this truncated enzyme also exists as an equilibrium mixture of inactive and active species.
Earlier studies comparing DegS ⌬PDZ and peptide-activated DegS used non-quantitative assays or utilized OMP peptides that are now known to be suboptimal for activation (6,7,12). To eliminate these problems, we measured cleavage of a sub-K m concentration of 35 S-labeled RseA and used saturating concentrations of an YYF tripeptide, which is as potent as any known OMP peptide (7). Under these conditions, DegS ⌬PDZ cleaved RseA ϳ4.5-fold more slowly than peptide-activated DegS ( Fig.  2A). Importantly, however, H198P DegS ⌬PDZ cleaved RseA much faster than either wild-type DegS ⌬PDZ or YYF-activated DegS and at ϳ70% the rate of YYF-activated H198P DegS ( Fig.  2A). When we determined the steady-state rate of RseA cleavage at saturating substrate concentrations, peptide-activated DegS was slightly more active than peptide-activated H198P DegS, and H198P DegS ⌬PDZ had ϳ85% the activity of the latter enzyme (Fig. 2B). These results confirm that the H198P mutation increases DegS activity only by increasing the concentration of active relative to inactive species and also strongly support an allosteric model in which this mutation also stabilizes an active conformation of DegS ⌬PDZ .
If the H198P mutation stabilizes an active conformation relative to an inactive conformation of DegS ⌬PDZ , then both the Hill constant and apparent K m (substrate concentration at halfmaximal activity) should be lower for the mutant than for the wild-type truncated enzyme (6). To test this prediction, we assayed the steady-state rate of cleavage as a function of RseA concentration (Fig. 3A). Cleavage by DegS ⌬PDZ showed substantial positive cooperativity (Hill constant, 1.7 Ϯ 0.1), and half-maximal degradation required a substrate concentration of ϳ650 M as expected from previous studies (6). By contrast, cleavage by H198P DegS ⌬PDZ showed almost no cooperativity (Hill constant, 1.1 Ϯ 0.1), and half-maximal degradation was observed at a concentration of ϳ90 M. These results support a model in which the majority of unliganded DegS ⌬PDZ molecules assume an inactive conformation in solution with substrate binding and the H198P mutation acting to shift the equilibrium to favor the active structure. A model of Monod- Wyman-Changeux allostery (Fig. 3B) in which substrate binds 5-fold more tightly to active than inactive DegS ⌬PDZ with a conformational equilibrium constant of 115 for the unliganded wild-type species and 0.15 for the H198P mutant produced good fits of the experimental data in Fig. 3A     the substrate-bound acyl enzyme. In each of the 21 crystallographically independent subunits in these three structures, the oxyanion hole was in the active conformation (Fig. 4A). In the structure of DFP-treated DegS ⌬PDZ , the active site serine (Ser 201 ) was modified to monoisopropylphosphorylserine (Mis 201 ), and the O1P oxygen of this residue occupied the oxyanion hole and accepted hydrogen bonds from the backbone -NH groups of Gly 199 and Ser 201 (Fig. 4B). The side chain of Pro 198 in H198P DegS ⌬PDZ packed against the aromatic ring of Tyr 162 (Fig. 4C) as observed previously in the structure of a full-length DegS variant containing the H198P substitution (8). This interaction should increase the stability of the functional conformation of the oxyanion hole by helping to stabilize the hydrogen bond between the backbone -NH of Tyr 162 and the carbonyl oxygen of Pro 198 (Fig. 4C). Based on a DegS ⌬PDZ structure, Hasselblatt et al. (12) concluded that Asp 126 in the catalytic triad was too far from His 96 to allow efficient activation of Ser 201 for nucleophilic attack on the RseA substrate. Our results do not support this model. In the catalytic triad shown in Fig. 4D, for example, the hydrogen bond between the O␦2 oxygen of Asp 126 and N␦2 nitrogen of His 96 is 2.8 Å. This value is very similar to the distances observed in peptide-activated DegS and other serine proteases (4,8,12,18). Similarly, the hydrogen bond between the N⑀2 nitrogen of His 96 and O␥ oxygen of Ser 201 is 3.1 Å (Fig. 4D), which is typical of serine proteases and active DegS. We did find that the catalytic triads in many DegS ⌬PDZ subunits were disrupted by rotation of the His 96 side chain away from Ser 201 and Asp 126 . Importantly, however, there was no correlation between solution activity and the number of irregular active sites in various structures. For example, His 96 assumed a rotamer incompatible with catalysis in seven of nine subunits in the asymmetric unit of the H198P DegS ⌬PDZ crystal. However, this enzyme variant was substantially more active than wild-type DegS ⌬PDZ in which only two of 15 rotamers of His 96 were aberrant in the structures reported here and previously (Protein Data Bank codes 2QF0 and 2QF3; Ref. 6). Moreover, the irregular rotamer of His 96 was observed in six of nine subunits in DFP-modified DegS ⌬PDZ , and yet these active sites had to be functional for modification of Ser 201 to occur in the first place. His 96 probably equilibrates between active and inactive rotamer conformations in all of these structures with small changes in crystal packing influencing the rotamer observed.
Mutations That Reduce DegS ⌬PDZ Activity-Previously, we found that the R178A mutation severely reduced the activity of DegS and DegS ⌬PDZ (6). In our active DegS ⌬PDZ structures, the side chains of Arg 178 and Gln 191 were hydrogen-bonded to each other and to the side chain of Thr 167 from an adjacent subunit (Fig. 4E). The Thr 167 -Arg 178 -Gln 191 network is not present in the structures of DegS ⌬PDZ and peptide-bound DegS reported by Hasselblatt et al. (12) and Wilken et al. (4). However, this absence is likely to reflect model-building errors as the hydrogen bonds we observed could be introduced into those structures by changing the Thr 167 rotamer with good fits to the electron density. Formation of the hydrogen bond network also sets up a packing interaction between the side chain of Thr 167 and Pro 161 (Fig. 4E), which in concert with Tyr 162 helps stabilize the functional conformation of the oxyanion hole (Fig. 4C). In prior studies, the Y162A DegS ⌬PDZ variant was found to be inactive in RseA cleavage (6).
To test the importance of the Thr 167 -Arg 178 -Gln 191 network, we constructed, purified, and assayed the cleavage activity of DegS ⌬PDZ variants with T167A, T167V, R178K, R178D, Q191A, and Q191E mutations. All of these variants formed stable trimers as determined by gel filtration chromatography (data not shown). Nevertheless, only one of these mutants (T167A) had detectable activity, which was ϳ10% the value of

Allostery in DegS Protease Domain
OCTOBER 29, 2010 • VOLUME 285 • NUMBER 44 wild-type DegS ⌬PDZ (Fig. 5A). Importantly, an isosteric substitution at the same position (T167V) produced an inactive enzyme (Fig. 5A), suggesting that a water molecule may partially substitute for the missing side-chain hydroxyl in T167A DegS ⌬PDZ . We conclude that interactions between the side chains of Arg 178 , Gln 191 , and Thr 167 play important roles in maintaining DegS ⌬PDZ activity.
Structural Features of Inactive Mutants and Allosteric Transition-In principle, different mutations might inactivate DegS ⌬PDZ in diverse ways or by stabilizing a common inactive conformation. To evaluate these possibilities, we carried out crystallization trials using inactive DegS ⌬PDZ mutants and obtained structures for the Y162A, Q191A, T167V/H198P, and Y162A/H198P variants (Table 1). We also rerefined a previously reported structure of R178A DegS ⌬PDZ ( Table 1). As described below, these structures had malformed oxyanion holes and resembled the protease domains in peptide-free DegS (4,5).
Pairwise comparisons of aligned subunits from the five inactive DegS ⌬PDZ structures produced an average displacement of corresponding C␣ atoms of 0.18 Ϯ 0.11 Å for 168 residue positions but also revealed substantial structural variations at the N and C termini, in residues 133-137, and in parts of the LA, LD, L3, and L2 loops (Fig. 5B). Parts or all of these regions were disordered or assumed different structures in different subunits. Similar variations were observed in pairwise comparisons of the 21 active DegS ⌬PDZ subunits ( Fig. 5C) with the notable exception of the LD loop, which was well ordered in all active structures and showed little structural variation. It seems likely that the remaining highly variant regions play little if any role in stabilizing either the active or inactive proteolytic conformation. In pairwise alignments of trimers (Table 2), the average root mean square deviations were 0.32 Ϯ 0.08 Å between active trimers (21 comparisons), 0.21 Ϯ 0.09 Å between inactive trimers (10 comparisons), and 0.68 Ϯ 0.10 Å for active versus inactive trimers (35 comparisons).
Careful comparisons of the active and inactive DegS ⌬PDZ subunits revealed a common set of backbone positions, which aligned extremely well in individual subunits as well as in intact trimers. These amino acids (45-58, 82-84, 90 -93, 117-121, 139 -146, 151-159, 167-175, 189 -194, 204 -216, and 233-249) form a solid triangular base (Fig. 6, A and B), which remains essentially invariant in the conversion of the inactive to the active enzyme. Most of the remaining residues, which switch conformations between the active and inactive structures, project upward from this base (Fig. 6, C and D).
Importantly, the inactive structures showed a common set of differences when compared with active DegS ⌬PDZ structures. For example, in each inactive variant, residues 229 -231 in the L2 loop had different conformations than in active structures, the oxyanion hole assumed a non-functional conformation, residues 161-164 in the LD loop were disordered or assumed different conformations than in active structures, Gln 191 and Thr 167 were not hydrogen-bonded, and side-chain density for Arg 178 was absent. Except for the final feature, these differences are also observed in the protease domains of full-length DegS in active and inactive conformations (4,5,8,12). In active structures, by contrast, the side chains of Glu 230 form a network of polar interactions around the 3-fold axis, which is coupled via interactions with Asn 197 to the oxyanion hole of the active sites (Fig. 6E). The oxyanion hole, in turn, interacts with residues in the LD loop and the Thr 167 -Arg 178 -Gln 191 network (Fig. 4, C and E). These connections, which are lost in the non-functional structures, collectively link the active sites of different DegS ⌬PDZ subunits and are probably responsible for the apparent all-or-none character of the allosteric transition.
The foregoing results support a model in which trimers of DegS ⌬PDZ can assume two basic structures. Mutations such as H198P stabilize the active conformation, whereas mutations such as Y162A, T167V, R178A, and Q191A stabilize the inactive conformation. Combining stabilizing and destabilizing mutations in the T167V/H198P and Y162A/H198P DegS ⌬PDZ variants resulted in enzymes that were inactive in solution and assumed the non-functional conformation in crystals. Thus, the H198P mutation only stabilizes the active conformation of the protease domain in concert with interactions mediated by other residues.

DISCUSSION
The studies presented here provide strong evidence that the protease domain of DegS in the absence of the PDZ domain equilibrates between inactive and active conformations. Degradation of the RseA substrate by the DegS ⌬PDZ trimer is posi-tively cooperative (Hill constant, 1.7 Ϯ 0.1), supporting a model in which substrate binding stabilizes the active conformation relative to the inactive conformation. In principle, positive cooperativity could also be caused by favorable substrate-substrate interactions (see Ref. 6). However, a substrate interaction model does not account for our finding that the H198P mutation in DegS ⌬PDZ almost completely abolished positive cooperativity in RseA cleavage. The H198P mutation also reduced K m for RseA cleavage by DegS ⌬PDZ from ϳ650 M to less than 100 M. This result provides additional support for the allosteric model, which predicts that some of the energy of RseA substrate binding to wild-type DegS ⌬PDZ is used to drive conversion of inactive to active enzymes. If most molecules of H198P DegS ⌬PDZ are already active, then RseA should bind more ) that are highly conserved following structural alignment of seven active (blue) and five inactive (red) DegS ⌬PDZ trimers. B, surface representation of the trimer base from the wild-type structure (Protein Data Bank code 3LGI). C and D, ordered backbone positions that switch conformation between the active (blue) and inactive (red) structures are shown in ribbon representation. The conserved base is shown in surface representation. E, model and electron density (1.5 ) for residues Asn 197 , Glu 230 , and Gly 231 in the wild-type DegS ⌬PDZ trimer (Protein Data Bank code 3LGI): yellow carbons, chain A; cyan carbons, chain B; purple carbons, chain C. All active trimers contain the same network of hydrogen bonds (dashed lines), which link the oxyanion holes of different subunits.

TABLE 2 Pairwise alignments of active and inactive DegS ⌬PDZ trimers
The top number is the root mean square deviation (Å), and the bottom number is the number of C␣ comparisons. Italics mark comparisons of active with inactive structures. DFP mod, DFP-modified.
fore, that a variant like H198P DegS ⌬PDZ would have higher activity in vivo and thus could be useful in probing potential OMP-independent inputs into the envelope stress pathway (11).
Our finding that the isolated protease domain of DegS can function as an allosteric enzyme raises the possibility that an evolutionary precursor functioned in exactly this fashion. DegS can cleave peptide bonds with a wide variety of P1 and P1Ј residues flanking the scissile peptide bond (22), and its active sites are highly exposed. Thus, an intracellular variant that was constitutively active and unregulated might function as a rogue protease and damage numerous cellular proteins. By having the inactive conformation be the stable form, however, only substrates that bind sufficiently tightly and specifically to the active conformation would be cleaved. Strong positive cooperativity would also ensure that most active protease trimers were fully saturated with the proper substrate, minimizing the number of unoccupied but functional active sites. By fusing an allosteric protease of this type to a PDZ domain, an extra layer of ligandmediated regulation could then be added to generate enzymes of the current DegS family.
Most DegS homologs consist of the protease domain and one or more PDZ domains. Modern biology may also have found uses for the trimeric protease domain in the absence of the PDZ domain as human HtrA proteases have isoforms that lack parts or all of the PDZ domain due to alternative splicing (23). It remains to be seen whether the allosteric potential of these natural ⌬PDZ enzymes is utilized and, if so, whether allosteric effectors other than protein substrates are involved in regulation. The allosteric nature of proteases has long been recognized as an important component of many zymogen activation reactions. Reversible allosteric regulation of protease conformation is an essential component of DegS function, and a growing number of studies suggest that other mature proteolytic enzymes without additional regulatory domains can also be allosterically regulated (24 -26).