Molecular motion regulates the activity of the Mitochondrial Serine Protease HtrA2

HtrA2 (high-temperature requirement 2) is a human mitochondrial protease that has a role in apoptosis and Parkinson’s disease. The structure of HtrA2 with an intact catalytic triad was determined, revealing a conformational change in the active site loops, involving mainly the regulatory LD loop, which resulted in burial of the catalytic serine relative to the previously reported structure of the proteolytically inactive mutant. Mutations in the loops surrounding the active site that significantly restricted their mobility, reduced proteolytic activity both in vitro and in cells, suggesting that regulation of HtrA2 activity cannot be explained by a simple transition to an activated conformational state with enhanced active site accessibility. Manipulation of solvent viscosity highlighted an unusual bi-phasic behavior of the enzymatic activity, which together with MD calculations supports the importance of motion in the regulation of the activity of HtrA2. HtrA2 is an unusually thermostable enzyme (TM=97.3 °C), a trait often associated with structural rigidity, not dynamic motion. We suggest that this thermostability functions to provide a stable scaffold for the observed loop motions, allowing them a relatively free conformational search within a rather restricted volume.


SI Results
Structure of HtrA2 mutants. Several HtrA2 mutations have been linked with increased susceptibility for Parkinson's disease (PD) 1,2 . This association has not been without controversy 3,4 , although more detailed studies have clearly demonstrated a link with PD in patients carrying a G399S mutation 5 . Less controversially, the S276C mutation has been linked to neurodegeneration in mice and leads to loss of proteolytic activity against the common protease substrate, β-casein 2 .
Loss of proteolytic activity due to the S276C mutation can be attenuated by deletion of the PDZ domain 2 . The PD-linked mutations, found in both the protease and PDZ domains of HtrA2 are particularly distant from the protease active site. While it is not uncommon for distal mutations to play significant roles in enzyme function, the mechanism by which these mutations exert their influence is often difficult to assign 6 . It is known, however, that HtrA2 is phosphorylated at S142 and S400 7 , and the PD-associated mutations A141S and G399S are proximal to these locations suggesting an obvious causative role, but it is still uncertain how changes in phosphorylation would specifically lead to PD susceptibility.
In an effort to understand their role in modulating protease activity, the crystal structures of several HtrA2 mutants associated with PD (e.g., A141S and G399S) 1 , the phosphomimetic mutant S142D 7,8 , and the neurological disorder mutant S276C 2 were determined in this study (SI Table 1).
Mutations A141S and S142D are located at the N-terminus of the mature protease domain while S276C is at the border of the LD loop and G399S is in the PDZ domain. Overall the mutant structures are very similar to that of HtrA2/WT (SI Table 2). There was no interpretable electron density for the N-and C-termini, within the L3 loop (V275-I295), and for the linker region between the protease and PDZ domains (G345-S357) for all the structures. The structures of the A141S, S276C, and G399S mutants also lack electron density for a portion of the PDZ loop region between helix α1 and strand β1 (near P385) 9 . This is close to (< 10 Å) a ligand binding site in DegS where highly variable protein-ligand contacts relieve PDZ-based protease activity inhibition 10 . The structures of all mutants (with a serine residue at position 306) revealed a reorganization of loops L1 and LD close to the active site similar to that observed for HtrA2/WT. As observed in the structure of mature wild-type HtrA2, the catalytic site H198 presented a high degree of plasticity (SI Fig. 2B). In the structures of the A141S and G399S mutants, H198 could be observed in two distinct conformers, while in the structures of S142D and S276C it displayed an "active" conformation, closer to the one found in the S306A mutant structure 11 . In contrast, in the wildtype structure and in the structure of HtrA2/Open, the side-chain of H198 is predominantly found in an "inactive" or catalytically incompetent conformation. The presence of "active" or "inactive" geometries for the catalytic triad in our structures does not correlate with HtrA2 enzymatic activity (see Protease activity assays and SI Table 3).
Overall the three-dimensional structures of HtrA2/WT and point mutants are very similar, except for minor local changes required to accommodate the mutated residues, and no structural features were identified to justify the observed differences in proteolytic activity.
Umbrella sampling MD and the conformational plasticity of HtrA2 active site. The free energy profile for the rotation of the H198 side chain along the C-Cα-Cβ-Cγ dihedral was calculated with umbrella sampling MD. The calculations explain the prevalence of "inactive" conformers in the crystal structure of HtrA2/WT X-ray structure. The free energy barrier of 8 kcal . mol -1 on moving from the "inactive" to the "active" conformation (the slowest of the two directions) means that the transitions take place on the μs timescale. The shift of H198 from the "inactive" to the "active" rotamer (~ -65°) is associated with a free energy difference of +5 kcal . mol -1 , meaning that only the catalytically incompetent conformation (as seen in the HtrA2/WT structure) can be quantitatively observed, as the relative abundance of the two rotamers ("inactive"/"active") is 10 3 -10 4 at physiological temperature. The free energy penalty associated with changing from the "inactive" to the "active" conformer prior to catalysis is significant but not really problematic in terms of enzyme efficiency, as the typical free energy barriers for peptide hydrolysis amount to ~ 16.4 ± 1.4 kcal . mol -1 12 . In the case of the S306A mutant structure, the catalytically competent conformation of H198 was calculated to be clearly more stable than in the wild type. Quantitative agreement is not perfect here, as the "active" conformation is still less stable than the inactive conformation by ~1.5 kcal . mol -1 probably due to the inherent inaccuracies of the force fields, since this value is of the same magnitude as the computational uncertainty (see Materials and Methods). The largest free energy barrier separating the two conformers of H198 in the S306A mutant is around 9 kcal . mol -1 , and consequently the interconversion between catalytic and inactive conformation happens also on the μs timescale.
Protease Activity and Thermal stability assays: HtrA2/WT was found to be an alkaline protease whose hydrolytic activity (k cat /K M ) against a fluorescent peptide substrate (H2-Opt) 13 increased linearly with increasing temperature up to 45 ˚C (SI Fig. 3A, B, SI Table 3). The point mutants analyzed displayed slightly reduced proteolytic activity (A141S had near wild-type activity while S142D and G399S activities were only slightly reduced (3.4-and 2.1-fold respectively)) while the ΔPDZ mutant increased activity by 18-fold (SI Table 3) in agreement with previous reports 11,14 .
HtrA2/WT was also found to be particularly thermostable (SI Fig. 4A, SI Table 3) at or above pH 6.0 in the presence of 400 mM KCl, 20 % (v/v) glycerol, as judged by CD measurements. A ΔCp for HtrA2/WT was calculated to be 1.7 kcal/mol (7 kJ/mol) using the method of Privalov 15 in the presence of 2 M GdnHCl and was used for further data fitting. This value is much lower than that expected from the protein's amino acid sequence (4.7 ± 0.8 kcal/mol (19.7 ± 3.5 kJ/mol) 16 ) suggesting that the thermostability of HtrA2 is due to shallow ΔCp curvature 17,18 Table 3) respectively, representing an average ΔT M of 3.5 °C per mutated residue (SI Table 3). Estimation of the effect of these mutations on thermostability gives ΔΔG values of -2.3 and -0.5 kcal/mol for the HtrA2/Open and HtrA2/Closed, respectively 18 (SI Table 3, SI to the six presented here were analyzed to identify experimental parameters giving rise to a discernable pattern in the orientations between the protease and PDZ domains. An additional structure, published later 19 was also included and we found 26 structures (not including the six presented here) with an associated publication that contained a protease domain and at least one PDZ domain (SI Table 4). One structure of E. coli DegP (PDB entry 1ky9 20 ), which has two distinct PDZ orientations in the asymmetric unit was included twice in our analysis. The HtrA structures grouped into 5 sets (Fig. 4) when aligned by hierarchical clustering using Multidendograms 5.0 21 .
Structures were grouped at an arbitrary cut-off of 20 Å, but even at this level it was not clear if the DegQ (PDB entry 3pv4 22 ) and HtrA3 (PDB entry 4ri0 23 ) structures are either experimental outliers or singular members of novel groups. Group 1 (SI Table 4) comprises mainly human HtrA structures (HtrA2 and HtrA3) but also includes one structure from L. fallonii DegQ (PDB entry structures. No examined factor (e.g., pH of crystallization solution, species of origin, presence of ligands, or research group determining the structure) was found that clearly explained the observed groupings of the conformational states of the protein structures (SI Table 4). Therefore, the groupings (other than group 1, the HtrA2 set, and group 3, which is largely comprised of higher order oligomeric structures) appear to be largely stochastic (see SI Table 4). The stochastic nature of the groupings (Fig. 4) and their conservation between the prokaryotic and eukaryotic proteins suggest that all these conformations are likely accessible to all HtrAs in solution, while crystal formation and packing constraints restrict the proteins to a single conformation in these structures (E. coli DegP (PDB entry 1ky9 20 ) being the exception). The switching of the protein between these conformations would comprise some of the dynamic motions that contribute to proteolytic activity and are affected by solvent viscosity, consistent with the experimental and MD results. The range of motions that have been reported so far in HtrA proteins includes a displacement of over 100 Å for residue S142 (animated as SI movie 4).

SI Methods
Circular dichroism measurements: Circular dichroism (CD) measurements were performed on a Jasco J-815 spectropolarimeter with a Peltier thermostat for temperature control. Concentrations of guanidinium hydrochloride (GdnHCl) stocks were determined using refractive index as measured on an A. Kruss AR3 refractometer 30 . Melting curves were determined by change in CD signal at 224 nm with a 2 ˚C/min temperature change rate. A Lowess spline fit of the observed data generated by Prism (Graph Pad Software) was used to fit the thermodynamic parameters of the melting curves using EXAM 31  temperatures determined using this experimental approach are significantly higher than those previously reported for HtrA2 34 . However, those experiments were performed using Tris as buffer, which is unsuitable for melting temperature measurements both due to its high β value and its considerable UV absorption at 207 nm, which was the wavelength of the previously reported CD measurements. However, the melting temperatures determined here are in agreement with Zhang and Chang 35 who report no change in secondary structure up to 70 °C.

SI Figures
SI Figure 1 -Sequence alignment of mammalian HtrA2 proteins. The secondary structure elements from the HtrA2/WT structure are shown as cylinders (α-helices) and arrows (β-strands) above the alignment, with disordered residues in the crystal structure indicated by a discontinuous line. The boundaries of the protease and PDZ domains are indicated below the alignment by a green and a salmon box, respectively. The PD-associated point mutants are indicated by blue diamond-shaped boxes. Residues mutated to engineer the HtrA2/Open and HtrA2/Closed variants are highlighted by yellow and pink boxes, respectively. The standard loops are highlighted as in Figure 1A.  Figure 1A; Subunit B of HtrA2/WT (PDZ domain colored as subunit A, protease domain in light gray) is shown superposed to the structurally equivalent subunit B of HtrA2 S306A mutant (PDB entry 1lcy 11 ), shown with PDZ domain colored orange and protease domain blue. Residues in subunit B are indicated by an asterisk. B) Active site plasticity in HtrA2. Close view of HtrA2 catalytic triad from the wild-type (light green, "inactive" H198 conformer), G399S (cyan, two H198 conformers), A141S (green, two H198 conformers), S142D (yellow, "active" H198 conformer), S276C (pink, "active" H198 conformer), and HtrA2/Open (magenta, "inactive" H198 conformer) structures. Significant conformational variability is observed for H198 while the other two catalytic residues (S306 and D228) have much less positional variation across the set of structures. In all cases F303 blocks access to active site with the obvious exception of the HtrA2/Open structure due to the F303A mutation. C) The catalytic residue, S306, is more buried (as calculated by DSSP 36 ) in the catalytically active structures than it would be in the presumed "active" conformation of HtrA2 observed in a previous crystal structure of the enzymatically inactive S306A mutant 11 . The PDB entry used for each estimate is given in parenthesis. is no effect on enzyme activity due to solution viscosity. Note that for clarity this graph shows the ratio of (k cat /K M )/ k cat /K M ) 0 , which better indicates increased enzymatic activity than the standard (k cat /K M ) 0 /(k cat /K M ) that best represents suppressive viscosity effects, thus leading to the inverse form of the fit line. It is evident that all HtrA2 constructs either meet or exceed this level of viscosity response up to η ~ 3 indicating that many of these enzyme motions are activity reducing. At higher viscosities the normal suppressive effect of viscosity dominates and reduces enzymatic activity. All results are the mean values of triplicate measurements from one experiment. SI movie 1 -Morph animation illustrating dynamic changes in loops surrounding the HtrA2 catalytic site in the HtrA2/Open structure during the 200 ns MD simulation. L1 loop is represented in cyan, LD loop in yellow and the catalytic S306 as sticks with red dots. The mutation sites F303A and L266R are shown as sticks, as well as Q267 (because its side chain moves towards the L1 loop), H198 (displaced away from the active site S306 during the simulation) and N181. Snapshots from the simulations were visualized in PyMOL and then the resulting images were combined into a fluid movie (http://gifmaker.me and https://giphy.com).
SI movie 2 -Morph animation illustrating dynamic changes in loops surrounding the HtrA2 catalytic site in the HtrA2/Closed model during the 200 ns MD simulation. L1 loop is represented in cyan, LD loop in yellow and the catalytic S306 is shown as sticks with red dots. The mutation sites N181S, Q267R and L266R, as well as residues F303 and H198 are shown. The Q267 side chain moves towards L1 loop, and H198 is displaced away from the active site S306 during the simulation. Snapshots from the simulations were visualized in PyMOL and then the resulting images were combined into a fluid movie (http://gifmaker.me and https://giphy.com).
SI movie 3 -Morph animation illustrating dynamic changes in loops surrounding the HtrA2 catalytic site in the HtrA2/WT structure during the 200 ns MD simulation. L1 loop is represented in cyan, LD loop in yellow and the catalytic S306 is shown as sticks with red dots. Residues L266, Q267, F303, N181 and H198 are shown as sticks. Notice that the Q267 side chain moves away from the L1 loop during the simulation, while H198 moves towards the active site S306. The mobility of the L1 and LD loops and residue 303 is significantly higher in HtrA2/WT, when compared to the HtrA2/Open and HtrA2/Closed constructs. Snapshots from the simulations were visualized in PyMOL and then the resulting images were combined into a fluid movie (http://gifmaker.me and https://giphy.com).
SI movie 4 -Morph animation illustrating dynamic changes in orientations of HtrA proteins observed in various crystal structures. In this movie the PDZ domain is shown as a stationary salmon surface while the protease domain is shown in a green cartoon representation. The PDassociated phosphorylated residues S142 (blue spheres, protease domain) and S400 (red surface, PDZ domain) are also indicated. The Cα of S142 is displaced by 104.1 Å between the first and the last frame of this movie. The morph was generated by aligning the HtrA2/WT structure with representative structures from the other orientational groups from Figure 5 -HtrA2/WT (group 1) ! 3pv4 (group 1) ! 4rr0 (group 2) ! 3pv2 (group 3) ! 4rqy (group 4) ! 1ky9, monomer A (group 5) -using the morph function in PyMOL and then combining the resulting images into a fluid movie (http://gifmaker.me and https://giphy.com).

SI Tables
SI Table 1 -Diffraction data collection, processing and refinement statistics (part I) *Values in parentheses are for the high resolution shell.

WT HtrA2
HtrA2 Groupings are derived from the analyses in Figure 4. This table provides the raw data (extracted from the PDB or this work) for the categories that were examined in an attempt to discern any non-stochastic patterns to the observed groupings of HtrA structural orientations. The table lists the name of the protein and the PDB entry, whether the catalytic residue is a free serine or how it is mutated/modified, the host organism, the experimental conditions used in the structural study, expected quaternary structure from the literature, the experimental pH, space group (if applicable), number of proteins in the asymmetric unit, if an explicit ligand was present, and the research group that generated the structure. There was a grouping of the human HtrA2 structures into orientational group 1 and of the higher order oligomerization states of DegP into group 3. Otherwise, no obvious overall correlations were found although this may be due to the low number of experimental structures available relative to the total number of variable parameters. Both single and complete linkage gave the same grouping pattern, although there were what appeared to be insignificant ordering differences within the group. SI