Distinct 3D Architecture and Dynamics of the Human HtrA2(Omi) Protease and Its Mutated Variants

HtrA2(Omi) protease controls protein quality in mitochondria and plays a major role in apoptosis. Its HtrA2S306A mutant (with the catalytic serine routinely disabled for an X-ray study to avoid self-degradation) is a homotrimer whose subunits contain the serine protease domain (PD) and the regulatory PDZ domain. In the inactive state, a tight interdomain interface limits penetration of both PDZ-activating ligands and PD substrates into their respective target sites. We successfully crystalized HtrA2V226K/S306A, whose active counterpart HtrA2V226K has had higher proteolytic activity, suggesting higher propensity to opening the PD-PDZ interface than that of the wild type HtrA2. Yet, the crystal structure revealed the HtrA2V226K/S306A architecture typical of the inactive protein. To get a consistent interpretation of crystallographic data in the light of kinetic results, we employed molecular dynamics (MD). V325D inactivating mutant was used as a reference. Our simulations demonstrated that upon binding of a specific peptide ligand NH2-GWTMFWV-COOH, the PDZ domains open more dynamically in the wild type protease compared to the V226K mutant, whereas the movement is not observed in the V325D mutant. The movement relies on a PDZ vs. PD rotation which opens the PD-PDZ interface in a lid-like (budding flower-like in trimer) fashion. The noncovalent hinges A and B are provided by two clusters of interfacing residues, harboring V325D and V226K in the C- and N-terminal PD barrels, respectively. The opening of the subunit interfaces progresses in a sequential manner during the 50 ns MD simulation. In the systems without the ligand only minor PDZ shifts relative to PD are observed, but the interface does not open. Further activation-associated events, e.g. PDZ-L3 positional swap seen in any active HtrA protein (vs. HtrA2), were not observed. In summary, this study provides hints on the mechanism of activation of wtHtrA2, the dynamics of the inactive HtrA2V325D, but does not allow to explain an increased activity of HtrA2V226K.


Methods Principal Component Analysis (Essential Dynamics)
The Essential Dynamics introduced by Amadei et al. [1] turns up identical with the Principal Component Analysis, PCA, a statistical approach of a proven general use and efficiency [2].
The method is useful whenever major factors influencing/describing objects' behavior/properties have to be fished out from multitude of cross-correlated data [3,4]. The method relies on diagonalization of an appropriately chosen covariance matrix (Cov). To our aims, an MD trajectory matrix was initially translated to an origin, defined at the geometrical center of the protein (based only on C α atom coordinates) and averaged over the MD-time.
This gave MD matrix (size 3N • number of MD time-steps; N being the number of residues in a protein). Cov (of size 3Nx3N) resulted from multiplication of MD by its transpose MD T .
Cov, when normalized, becomes identical with the respective correlation matrix. The diagonalization of Cov, i.e. solving an eigenproblem, leads to a set of 3N eigenvalues λi,(normalized λi,/Σλi become weights; Σweights=1) and a 3Nx3N matrix of orthogonal eigenvectors, collectively with λi. The factors ordered by decreasing λi signify declining major contributions of orthogonal (non-correlated) motional modes to the total dynamics of the protein. The effort turns out valuable when 3-4 major orthogonal modes suffice to explain a majority (70-80%, as measured by the sum of λi) of total mobility of the protein.
The PCA was performed using ProDy package [5]. Only Cα atom coordinates were processed (3x320 total; ligand ignored) and 5000 points from the productive MD trajectories (every 10 ps) were selected, resulting in the Cov size equal to 960x960. Analyses and visualizations of the PCA results were prepared using plug-ins to ProDy own NMWIZ routine and/or to the VMD [6] program. 3

Principal Component Analysis
PCA confirms the observations of visual inspection of the MD results. Therefore, here we present only representative results on principal motional modes of HtrA2-ligand, HtrA2 S306Aligand and apo HtrA2 S306A (S1-S3 Tables and S1-S3 Figs., respectively). Scree plots (S1-S3  Table 4) the scree plots are steeper than for the other ones. Indeed, for these three units, 70-80% of the total variance is explained in the first three factors (i.e. orthogonal collective/segmental motional modes), contrary to the other units, in which the first three factors typically explain only about 50-70% of the total motional variance, see S1-S3 Tables. The ratios of the eigenvalues (weights) of the first to the second factors (λ1/λ2) support this conclusion. They are noticeably larger (λ1/λ2≈5-9) for the units experiencing specific movements than for the units moving more randomly (λ1/λ2≈2-5).
Steeper scree is associated with faster accumulation of variance through the first 30 factors (per total 3x320 C α atoms, see Methods). For the units moving specifically, 90-95% of the total variance is included in the first 30 factors while for the others only 85-90%. The factors above around the 30 th account for harmonic oscillations, Brownian motions and noise.
Three largest-eigenvalue factors (modes), evidently associated with segmental motions were analyzed in more detail. A bulk of segmental mobility of the HtrA2-ligand trimer is explained in the first factor (S1 Fig.), whereas the majority of segmental motions of the HtrA2 S306A -ligand and apo HtrA2 S306A trimers are explained in the first two factors (S2 and S3 Figs.). Significant mobility of LA, L3 and the PD-PDZ-linker and lower than the average mobility of the inter-barrel linker and α7 are common for all HtrA2 models (S1-S3 Figs.).
Mode 1 within units A and C of HtrA2-ligand trimer (S1 Fig.) and unit B of HtrA2 S306Aligand trimer (S2 Fig.) describes these units' common segmental motion. This observation is readily appreciable by comparison of the movements of linker and PDZ, especially if allowing for "mental scaling" between fluctuations of mode 1 in S1 and S2 Figs. For example, along with generally similar shapes of the segmental fluctuations in mode 1, in all three units α5 and α7 are in minimum while α6 and the β16-β17 and β18-β19 loops are in maximum. In S4 Fig., modes 1-3 of unit 3 of HtrA2-ligand-trimer are visualized as a representative example of a common motional behavior. Specifically, mode 1 (58,4% total variance) accounts for a major contribution of the segmental motion shown in Figure 5, referred to as a lid-opening ( Fig. 5 and S4 Fig.). Mode 2 ( 8,0% total variance) accounts for another type of lid-opening, however, with the "joints-rotation" axis running horizontally and perpendicularly to that in mode 1 behind the PD-PDZ interface ( Fig. 5 and S4 Fig.). Mode 3 (6,3% total variance) accounts for a minor unspecific motion of PDZ versus PD. As such, PCA results fully confirm all observations based solely on visual inspection of the MD results. Omnipresent superior LA, L3 and linker mobility (see above) is also reflected in S4 Fig.