Cryo-EM analysis of human sACE dimer.

(A) Diagram for the key features of domain on primary sequence of human ACE. The construct used in this study contains the first 1235 residues, which we refer to as the soluble region of ACE (sACE) however, the first 29 residues comprise a signal. By convention, sACE is labeled based on the mature, processed peptide (68). D1a and D3a domains, also called "lid" encompasses residues 1-98 and aa 616-696, respectively. D1b and D3b domains each have two discrete segments; residues 263-436 and 496-574 for D1b and residues 868-1031 and 1091-1171 for D3b. D2 encompasses three discrete segments, residues 99-262, 437-495, and 575-615 while D4 domain encompasses residues 697-867, 1032-1090, and 1172-1202. sp = signal peptide. asterisk = zinc binding motif. Colored by sub-domain: D1a blue, D1b gold, D2 magenta, D3a cyan, D3b green, D4 purple. (B) 2D classification of human sACE particles from grids made by vitrobot (360 pixels box size) and Chameleon (256 pixels box size). Clear four domain classes visible in the Chameleon-derived classification are boxed in red, similar views are lacking in the vitrobot dataset. White scale bar measures 100 Å. (C) Full-length sACE 3D volumes, colored by sub-domain as in (A). Chain B sub-domains are depicted as lighter tones of their chain A counterparts. Glycan density is shown in gray. See Figure S2 for vitrobot-prepared data processing details, Figure S4 for Chameleon-prepared data processing details and Table S2 for data refinement statistics.

Overall structure of human sACE.

(A) Overlay comparing sACE-N states highlighting the structure differences between the closed “C”, intermediate “I”, and open “O” states. (B) Overlay comparing sACE-C states highlighting the structure differences between the closed “C”, and intermediate “I” states. We define the state based on the distance between the edge of the D2/4 domain bordering the catalytic cleft (residues 121-126 or 721-726) and the tip of the D1/3a region (residues 41-51 or 647-657): closed <15 Å, intermediate >15 Å and <19 Å, open >19 Å. (C) Overall dimer comparisons. Black bracket depicts the D1/3a-D2/4 distance measurement used to define the state of each domain, as described above. (D) Table of openness measurements for each domain per structure.

sACE dimerization interfaces.

(A) Overlay comparing sACE-N/N interface (blue) and sACE-C/C (gold) interfaces. Interfaces adopt the same secondary structure but interacting residues vary between them. (B) Residue-specific interactions at the sACE-N/N interface, see text for details. (C) Unsharpened Coulomb potential density map (blue) showing density corresponding to glycan-glycan interaction from N82 as part of sACE-N/N interface. Sharpened map is shown in magenta for reference. (D) Residue-specific interactions in the sACE-C/C interface, see text for details.

Comparison of molecular dynamics simulations with cryo-EM data.

(A) Alpha carbon displacement values for each residue were calculated by comparing each of our cryo-EM structures against one another in a pairwise manner and averaged. Individual residues are colored by sub-domain as defined in figure 1. Alpha carbon displacement values were also mapped onto the structure of sACE to better visualize the mobile regions. Residues are colored by degree of displace from blue (no displacement) to red (high displacement, values in Angstrom, as indicated). Numbers denote regions of interest: 1, the top of the D1a sub-domain, 2, the bottom of the D1a sub-domain, 3, a flexible loop region within the D1b that interacts with the bottom of the D1a sub-domain near the catalytic cleft, 4, the bottom of the D3A sub-domain. (B) Alpha carbon RMSF values for a representative 100 ns subset from our non-glycosylated MD simulations that demonstrates the simulated dynamics correlate well with displacement calculated from the cryo-EM structures. (C) Alpha carbon RMSF values for the entirety of our non-glycosylated MD simulations. Mobile regions agree with predictions from the cryo-EM structures, yet the magnitude is dampened due to the protein rapidly transitioning from an open state to a closed state and remaining closed for most of the simulation time. (D) Alpha carbon RMSF values for a representative 100 ns subset from our glycosylated MD simulations. (E) Alpha carbon RMSF values for the entirety of our glycosylated MD simulations.

Structural mechanism of the sACE open/close transition.

(A) sACE-N overlay comparing the open (left panel) and closed (right panel) states in detail. The open state is stabilized by interaction between residues in the D1a (blue) and D2 (magenta) regions that, notably K73-D189. In the closed state, the K73-D189 interaction is broken. (B) Overlay of the sACE-N open (dark shades, sub-domains colored as above) and closed (light shades, sub-domains colored as above) states showing the range of motion. The D1a region rotates about a fulcrum region described in (A), while the D1b region moves as a rigid body. Front view arrows depict the “fulcrum” motion of the D1a subdomain, with the top and bottom of the subdomain moving in opposite directions. Side view arrows depict rigid body motion of the D1a and D1b subdomains moving together. (C) Overlay of sACE-C closed (light shades, sub-domains colored as above) and intermediate (dark shades, sub-domains colored as above) states. Unlike sACE-N, the “top” of the D3a region is constrained by its connection to sACE-N and largely immobile. The primary source of opening is only the motion of the D3a tip. We did not observe any open state structures of sACE-C, suggesting a smaller range of motion relative to sACE-N. Front view arrow depicts motion of D3a subdomain. Only the bottom of the subdomain moves, in contrast to the “fulcrum” motion observed in the D1a subdomain. Side view depicts the rigid body motion of the D3a and D3b subdomains moving together. (D) Comparison of the hydrophobic “latch” region formed in the closed state between residues of the D1/3a, D1/3b, and D2/D4 domains. V724 in sACE-C has been replaced by T124 in sACE-N, suggesting that the closed state in sACE-N may be less stabilized than the sACE-C closed state. (E) Example all-atom MD simulation tracking the openness of one sACE-N region (black line) and this distance between K73 and D189 (red line). These residues form a salt bridge early in the simulation when sACE-N is open (left inset) but the interaction breaks as sACE-N transitions to the closed state (right inset). Distance measurements for MD simulations were consistently greater than distance values in our static structures and cannot be directly compared to Figure 2.

Cryo-EM heterogeneity analysis.

(A) Visualization of the structural changes revealed by cryoSPARC 3DVA trajectories calculated along two principal components (PCs) of structural variance. Starting states are showing in cyan, ending states in gray. PC 0 reveals a large, inter-domain bending motion accompanied by the open/close transition in sACE-C. PC 1 and the remaining PCs are dominated by the open/close transition of individual regions. See Table S3, and Movie S9 for additional details. Arrows depict generalized motions. (B) Visualization of the structural changes revealed by the cryoDRGN trajectories calculated along two PCs of structural variance. Starting states are shown in cyan, ending states in gray. PCs are dominated by the open/close transition of individual regions. See Table S3, and Movie S10 for additional details. Arrows depict generalized motions. (C) Analysis in RECOVAR with a focus mask on sACE-N/N reveals that particles adopt roughly four clusters within the latent space (heat map of particle density) corresponding to the open-open (OO, white square), open-closed (OC, white dot), closed-open (CO, white dot) and closed-closed (CC, white star) states of sACE-N/N. A trajectory estimating the path through latent space corresponding to the structural transition from sACE-N/N CC state to the OO state (blue points) suggests that individual sACE regions transition at different rates, as indicated by the size of the transition arrows between states. See Movie S11 for trajectory. (D) Focused 3D classification was performed in cryoSPARC to explore evidence of coordinated motion between sACE-N regions. 3D classification focusing on one sACE-N region revealed 2 roughly equal classes of particles: open and closed. Subsequent 3D classification focused on the other sACE-N region again revealed 2 roughly equal classes, suggesting the lack of coordinated motion between sACE-N regions in the absence of substrate.

DSF optimization of sACE grid-making conditions.

48 buffer conditions were screened for sACE stability in the presence and absence of EDTA (see table S1). Citrate-phosphate buffer pH 5.5 (red) was chosen for the best combination of stability (temperature shift right) and homogeneity (steepness of transition slope). Two other citrate-phosphate conditions, pH 5 and pH 4.5 gave better stability, but a more gradual transition slope, and so were not chosen.

Processing workflow for dataset using Vitrobot-prepared grids.

3,653 micrographs were collected on a 300 keV Titan Krios at the University of Chicago. A full-length model of sACE was manually generated by combining available PDB structures of ACE-N and ACE-C. The resulting PDB file was converted to an EM map in Chimera and imported into cryoSPARC to generate particle picking templates. Using these templates, 1,171,502 particles were picked and extracted. Successive rounds of 2D classification were used to trim the particle number down to 577,123 particles. Ab initio reconstruction followed by heterogeneous refinement yielded a full sACE dimer map (214,893 particles), a map representing only the N/N dimer (268,471 particles), and 93,759 particles that mapped to an uninterpretable junk class. The particles mapping to the full length sACE structure were then transferred to PCA-based 3D classification. 103,507 particles mapped to low resolution partial structures that could not clearly be assigned structural significance and were discarded as junk. Of the remaining particles, 20,533 particles mapped to a three-domain sACE structure. The remaining 88,930 particles mapped to a full-length sACE structure that was refined to 3.65 Å (sACE-3.65). Scale bar in representative micrograph shows 200 Å, scale bars in 2D classes show 100 Å.

Assessment of map quality from Vitrobot prepared grids.

(A) Angular distribution. (B) Local resolution colored as blue-cyan-green-yellow-red spectrum from 3-5 Å. (C) Example residue fits to density map. (D) 3D-FSC results.

Processing workflow for dataset using Chameleon-prepared grids.

18,281 micrographs were collected at the National Center for CryoEM Access and Training. After motion correction and CTF estimation, the dataset was trimmed to include only micrographs with a max CTF value lower than 4 Angstroms. The sACE-3.65 structure was used to generate 2D templates for particle picking. 10,708,232 particles were originally picked. This number was trimmed to 2,580,990 based on quality metrics and visual observation. The resulting particles were used in successive rounds of 2D classification to eliminate junk particles, reducing the particle count to 662,219. Ab initio reconstruction followed by heterogeneous refinement revealed four classes: 165,395 particles mapped to a class we interpret as the sACE-C/C dimer, 140,038 particles mapped to a class we interpret as the sACE-N/N dimer, 50,647 particles mapped to a 3-domain sACE class, and the remaining 306,139 particles mapped to a full-length sACE class. PCA-based 3D classification was used to eliminate junk particles, and the remaining ∼170k particles were refined with C2 symmetry to 2.8 Angstroms. The D1 domains in the C2 reconstruction displayed poor density, likely resulting from molecular motion, so a final round of 3D classification and refinement with C1 symmetry was performed to finalize three states of full-length sACE. Scale bar in representative micrograph shows 150 Å, scale bar in 2D classes show 100 Å.

Assessment of map quality from Chameleon prepared grids.

Angular distribution, local resolution, example density fit and 3D-FSC curves for (A) sACE-3.05, (B) sACE-3.15, and (C) sACE-2.99.

B factor comparison of all structures.

Full-length sACE structures colored by B factor. Scale ranges from a B factor of 10 (dark blue) to 225 (red). Protein is depicted as cartoon, glycans are shown as sticks.

sACE-N/N dimers compared to previously published structures.

Previously identified sACE-N/N dimers 2XYD (green, which is similar to structures of following pdb codes: 3nxq, 4bxk, 4bzs, 4ca6, 4ufa, 6f9r, 6f9v, 6h5x, 6qs1, 6tt1, 6tt3, 6tt4, 7q25, 7q26, 7q27, 7q29, 7z6z), 2C6F (cyan, which is similar to structures of following pdb codes: 2c6n, 5amb, 5amc), 4UFB (magenta, which is similar to structures of following pdb codes: 5am8, 5am9, 5ama, 6en5, 6en6, 6zpq, 6zpt), and 7Q4D (orange) aligned to sACE-2.99 (gray).

sACE exhibits a continuous gradient of structural heterogeneity.

All 16 individual domains from our sACE structures were aligned to their respective D2/4 regions, revealing a continues gradient of “openness” (gray) from the most closed conformation (green) to the most open conformation (magenta).

RMSD analysis of non-glycosylated MD simulations.

RMSD values were calculated for the D1/3a (red), D1/3b (gray), and D2/4 (orange) regions and compared to the RMSD values for the global domain (blue) to identify the primary source of structural variance in the non-glycosylated sACE MD simulations. RMSD was calculated using backbone atoms only. For each panel, the Y-axis is RMSD value, the X-axis is simulation frame. Every frame represents a time step of 0.1 ns.

Principal Component Analysis.

PCA was performed using the alpha carbon coordinates for every residue in every frame of our MD simulations along with each of our cryo-EM structures. Red X’s denote data from cryo-EM structures. MD simulation data is colored by run as indicated. Non-glycosylated MD simulation data is shown on the left, glycosylated MD simulation data is shown on the right. The conformational space explored in our MD simulations is significantly greater than the range of conformations observed in our cryo-EM structures, as expected. The most significant source of conformational variation in our MD simulations relative to our cryo-EM structures is the reposition of sACE-C/C relative to sACE-N/N (Movies S7,8).

Clustering Analysis.

The conformational space sampled by each domain in our non-glycosylated (left) and glycosylated (right) MD simulations was assessed by tracking the angle formed by the centers of mass of the D2-D1a-D1b (or D4-D3a-D3b for sACE-C) relative to the openness of the domain (D1/3a-D2/4 measurement as shown in figure 2) for each frame of our simulations and plotted as density. The corresponding measurements from our cryo-EM structures are shown in red.

RMSF comparison of glycosylated vs non-glycosylated MD simulations.

(A) RMSF values for each residue throughout the glycosylated sACE MD simulations. Chain A residues shown in black, chain B residues shown in red. (B) RMSF values for each residue in the non-glycosylated MD simulations. Chain A residues shown in black, chain B residues shown in red. (C) Comparison of RMSF values from glycosylated and non—glycosylated simulations. RMSF values from the glycosylated simulations (panel A) were subtracted from RMSF values for the non-glycosylated simulations (panel B). Negative values indicate the residues are more dynamic when glycans are present, positive values indicate that the residues are more dynamic when glycans are not present. Chain A residues shown in blue, chain B residues shown in orange.

Shielding of glycans in MD simulations.

200 representative frames of glycan orientations (light blue) were overlaid the sACE structure (pink shade by chain) to visualize the amount of protein surface insulated from solvent access by glycans.

CryoDRGN clustering of particles.

Top panel shows the deconvolution of the latent space along specified principal component vectors and the path of each trajectory generated as movies (see Movie S10). Bottom panels show particle clustering resulting from UMAP deconvolution of the latent space with the path of each trajectory generated.

Artificially observed interdomain motion caused by the subtle yet noticeable differences between the two chains within the sACE dimer and the choice of chain pairs between sACE dimer structures.

A. Overall structure of the sACE dimer and domain assignment. B. Ribbon representation of sACE-2.99 structures where the assignment of chain A and chain B within the sACE dimer is swapped. C. Ribbon representation of the sACE-3.05 structure. D. Alignment of the sACE-2.99 structure with sACE-3.05. When sACE-2.99 A/B chains are aligned with sACE-3.05 A/B chains, one domain is well-aligned but the other three are not. There is an ∼5° rotation of D4/D4 relative to D2/D2 from the alignment. E. The alignment of two structures that are the same as D except chain A and chain B in sACE-2.99 is swapped. When sACE-2.99 B/A chains are aligned with sACE-3.05 A/B chains, three domains are well-aligned and the remaining one depicts the open-closed transition of sACE catalytic domain. In this case, D2/D2/D4/D4 of the two structures are well aligned and there is no rotation.

List of buffer conditions used in DSF analysis of sACE.

CryoEM data collection, refinement, and validation statistics.

Summary for the conformational states of individual sACE domains in our all-atom MD simulation.

Summary of heterogeneity analysis.