Structural basis of phosphodiesterase-5 conformational organization revealed by a PDE6/PDE5 chimera

Abstract

Phosphodiesterase 5 (PDE5) plays a critical role in the nitric oxide-cGMP signaling pathway. Consequently, PDE5 catalytic site inhibitors are widely used in the treatment of erectile dysfunction and pulmonary arterial hypertension. Despite a wealth of structural data on the individual PDE5 catalytic domain with bound drug molecules, understanding of the structural organization of the full-length enzyme and its allosteric activation by noncatalytic cGMP-binding is lacking. To begin to understand the structural organization of PDE5, we solved a cryo-EM structure of a chimeric PDE enzyme (PDE6C/5) comprised of the regulatory domains of cone PDE6C and the PDE5 catalytic domain. The PDE6C/5 structure revealed the protein in the open state conformation similar to that of PDE6, suggesting a comparable conformation for the cGMP-bound PDE5 molecule. The H- and M-loops outlying the catalytic pocket, which are conformationally variable in the structures of the isolated PDE5 catalytic domain, are immobilized in the PDE6/5 chimera via the interaction of the H-loop with a linker helix LH2. Decreased dynamics of these loops may underlie the higher catalytic activities of the full-length PDE5 and PDE6C/5 compared to those of the isolated PDE5 catalytic domain. Furthermore, the PDE6C/5 structure defines the folding requirement of the PDE6 catalytic domain for chaperone-dependent maturation that is important for vision.

Cyclic nucleotide phosphodiesterases (PDEs) play a key role in controlling intracellular levels of cAMP and cGMP, and thus, many important biological signaling pathways (1, 2, 3). Eleven families of PDEs have been identified that differ in substrate specificity (e.g. cAMP, cGMP), tissue expression profiles, and regulation. PDE enzymes are catalytic dimers comprised of regulatory N-terminal domains and conserved C-terminal catalytic domains. Several PDE families, including PDE2, 5, 6, and 10 contain tandem N-terminal GAF (cGMP-specific cGMP-stimulated PDE, adenylate cyclase, and FhlA) domains, GAF-A and GAF-B, that are critical for the allosteric modulation of the enzymes (1, 2, 3, 4, 5, 6, 7).

Despite the close evolutionary relationships between PDE5 and PDE6 (8), the physiological roles of the two enzyme families are remarkably different. PDE5 is widely expressed in the smooth muscle, brain, lung, kidney, urinary bladder, and platelets (2, 3, 5, 9, 10, 11). It modulates a variety of physiological processes, most notably vascular smooth muscle relaxation. Nitric oxide (NO) produced by endothelial nitric oxide synthase (eNOS) in endothelial cells diffuses into vascular smooth muscle cells, where it activates soluble guanylyl cyclase (GC). Subsequently, PKG, activated by cGMP, phosphorylates several targets, including myosin light chain phosphatase and Ca²⁺ channel proteins, leading to relaxation of smooth muscle (12, 13, 14, 15). By hydrolyzing cGMP, PDE5 counteracts and dampens the NO-cGMP signaling pathway. PDE5 catalytic site inhibitors such as sildenafil (Viagra), tadalafil (Cialis), vardenafil (Levitra), and avanafil (Stendra) are used in the treatment of erectile dysfunction and pulmonary arterial hypertension (5, 13, 15, 16). Furthermore, the GAF-A domains of PDE5 enclose noncatalytic cGMP-binding sites that are central to allosteric feedback regulation of PDE5 whereby elevated levels of cGMP activate the enzyme (17, 18). Increased occupancy of the noncatalytic cGMP binding sites activates PDE5 by as much as 10-fold compared to the basal unstimulated state of the enzyme (17). By comparison, PDE6 enzymes are expressed exclusively in rod and cone photoreceptor cells and are specialized as the principal effector enzymes in the vertebrate phototransduction cascade (4). Unique attributes of PDE6 include association of the catalytic subunits with small inhibitory γ-subunits (Pγ) and a diffusion-limited catalytic efficiency when Pγ is displaced by the visual G-protein transducin-α (Gα_t) during phototransduction (4). Similarly to PDE5, the GAF-A domains of PDE6 bind noncatalytic cGMP. However, noncatalytic cGMP does not directly control cGMP hydrolysis at the PDE6 catalytic sites; rather it modulates the affinity of the catalytic subunits for Pγ (19, 20). Another distinct feature of PDE6, in contrast to other PDE enzymes, is that rod and cone PDE6 require the complex of HSP90 with a specialized cochaperone aryl hydrocarbon receptor (AhR)-interacting protein (AIP)-like 1 (AIPL1) as well as the Pγ subunits for its proper folding and maturation (21, 22, 23). Failure of PDE6 to properly mature due to mutations in AIPL1 leads to a severe early-onset form of blindness in children (24).

Understanding of function and regulation of PDEs in general, and the GAF-domain containing PDEs in particular, requires structural information on the full-length enzymes. Despite the physiological significance of PDEs, the structures of nearly full-length proteins have been solved for only two PDE families, PDE2 and PDE6 (25, 26, 27). The structures revealed strikingly different conformations, PDE2 in the closed state and PDE6 in the open state (25, 26, 27). The sparsity of the full-length PDE structures contrasts the numerous structures of the individual catalytic domains available for nearly every PDE family (25, 28, 29, 30). The structures of the PDE catalytic domains revealed conserved closely superimposable folds except for two conformationally variable H- and M-loops at the opening of the catalytic pocket (28). The two loops have been proposed to play important roles in the catalytic properties, drug binding, and selectivity (29).

Solution of the structures of full-length PDE5 structures in the apo and cGMP-activated states is ultimately required for understanding the allosteric regulation of the enzyme. Our initial efforts indicated unique challenges for studies of PDE5 by cryo-EM linked to its dynamic nature. As a first step to understand the structural organization of PDE5, we solved the structure of chimeric PDE6/PDE5 protein by single-particle cryo-EM analysis. We expressed and characterized this chimera containing the GAF-A/GAF-B domains from cone PDE6C and the catalytic domain from PDE5 previously (31). The structure shows the chimeric protein in the open state, suggesting that cGMP-activated PDE5 assumes an open conformation similar to that of PDE6 (26, 27). Moreover, the H-loop is immobilized in the open state via the interaction with the long linker helix 2 (LH2) connecting the GAF-B domain and the catalytic domain of a partner catalytic subunit in the dimer. Also, the H-loop interfaces extensively with the M-loop, indicating that both loops are markedly less variable and dynamic in the context of the full-length enzymes. Finally, the structure of PDE6/5 chimera supports the idea that the folding deficiency of immature PDE6 in the absence of specialized chaperone machinery is confined to its catalytic domain or that the immature PDE6 is in the closed state akin to the closed state of PDE2 where the catalytic pockets are occluded at the dimer interface (25).

Results

Structure of PDE6C/5 chimera

For cryo-EM analysis, the PDE6C/5 chimera was purified following the protein expression in Sf9 insect cells. Analysis of the purified sample by mass photometry (MP) indicated a monodisperse species with predicted molecular weight (Fig. 1, A and B). Freshly purified PDE6C/5 samples were plunge frozen after addition of Pγ, cGMP, and sildenafil. Structural characterization of PDE6C/5 with omission of Pγ and/or cGMP was not attempted because the entire PDE6C molecule is severely destabilized in the absence of Pγ, whereas the PDE6C GAF-A domains lacking bound cGMP become too dynamic to be visualized by cryo-EM SPA (27). The particles were picked with the use of Topaz model obtained previously by training on selected PDE6C particles (Fig. S1) (27). 2D classification revealed various projections of the PDE6C/5 particles with high-resolution features (Fig. S2). The consensus map of PDE6C/5 based on 158.6 K particles was refined to an overall gold-standard FSC (GSFSC) resolution of 3.05 Å (Fig 1, C and D, Fig. S1, Table S1). As indicated by the local resolution map, the reconstruction featured well-resolved density for all main domains of PDE6C/5 including the GAF-A, GAF-B, and the catalytic domains (Fig. 1E). Consequently, these domains were modeled de novo with high confidence. In the structure, the GAF-A and GAF-B domains are connected by long linker helices LH1, whereas GAF-B domains and the catalytic domains are linked with long helices LH2 (Fig. 2A). The density for the N-terminal 43 aa residues was missing. In addition, the density was present for the GAF-B bound portion of both copies of the Pγ subunits (residues Pγ-25–40 and 52–60). Similarly to PDE2 and PDE6, catalytic subunits of PDE6C/5 are arranged in a dimer with a 2-fold symmetry (Fig. 2A). The LH2 linkers cross over at the symmetry axis placing the catalytic domains beneath the GAF-B domains from a partner subunit. Overall, the structure of PDE6/5 chimera is very similar to that of PDE6C (RMSD 2.1 Å) including the positions and orientations of the catalytic domains (Fig. 2B). Thus, like PDE6, PDE6C/5 chimera assumes an open state. The key difference between the structure of PDE6 and PDE6C/5 is the lack of extensive noncovalent allosteric contacts between the GAF-B β1-β2 loop and the catalytic domain that were identified in PDE6 (26, 27). In the structure of PDE6C/5 chimera, the GAF-B β1-β2 loop is largely disordered (Fig. 2A). This finding is not surprising considering that the surface of the PDE6 catalytic domain interacting with the GAF-B β1-β2 loop is significantly different from the corresponding surface of PDE5 catalytic domain. Also, in the cryo-EM structures of PDE6, the Pγ C-terminus bound to the opening of the catalytic pocket is well resolved (26, 27). In the PDE6C/5 map, the density for the C-terminus of Pγ is predictably lacking since, as we demonstrated earlier, the Pγ C-terminus does not interact with the catalytic domain of PDE5 (32).

Figure 1.

Cryo-EM analysis of the PDE6C/5 chimera. A, a Coomassie-stained SDS-PAGE gel showing the preparations of PDE5 and PDE6C/5 chimera. B, mass photometry (MP) analysis of the PDE6C/5 sample for cryo-EM studies. C, fourier shell correlation (FSC) curves of the final refinement map. No mask, Loose, Tight, and Corrected refer to four primary FSC curves based on the mask type used in calculation of Gold-standard FSC resolution (GSFSC). D, the final cryo-EM map for PDE6C/5. E, local resolution estimation map for PDE6C/5. Resolution keys are from 2.8 to 6.0 Å.

Figure 2.

Structure of the PDE6C/5 chimera. A, the structure of the PDE6C/5 chimera modeled into cryo-EM map is shown in cartoon. Dashed lines delimit the GAF-A, GAF-B, and catalytic domains. The linker helices connecting GAF-A with the GAF-B domains (LH1) and GAF-B with the catalytic domains (LH2) are indicated by arrows. cGMP bound to the GAF-A domains and sildenafil bound to the catalytic pockets are shown as sticks. The PDE6C part of the molecule encompassing the GAF-A/GAF-B domains is colored pale cyan, whereas the PDE5 portion including the two catalytic domains is colored in green and cyan. Pγ residues bound to the GAF-B domains are colored in magenta and hot pink. The H-loops are shown in wheat. Noncovalent allosteric contacts of the GAF-A domain with GAF-B involving the GAF-A β1-β2 hairpin are indicated by arrow. The GAF-B β1-β2 loop, making contacts with the catalytic domain in PDE6, is largely disordered in PDE6C/5 (arrow points to where the β1-β2 loop would be if it was structured). B, an overlay of the PDE6C/5 chimera structure (colored green and cyan) with the structure of PDE6C (PDB 9CXH) (grey). The Pγ subunits of PDE6C are omitted for clarity. RMSD 2.1 Å. Noncovalent allosteric contacts of the GAF-B domain with the catalytic domain in PDE6 involving the GAF-B β1-β2 loop are indicated by arrow.

Interaction of the H-loop with the LH2 helix stabilizes the enzyme in the open state

Two conformationally variable loops at the opening of the catalytic pocket, termed H- and M-loops, have been identified in the crystal structures of individual PDE catalytic domains (28, 29). The structural changes in the H- and M-loops are ligand specific, and the conformational variability of the H-loop is the most dramatic in the PDE5 structures (29). In contrast, in the structure of the full-length PDE6/5 chimera, the conformation of the H-loop is constrained by the interaction with the LH2 helix from the partner subunit. Specifically, the H-loop (aa 585–608 of bovine PDE5) forms two short α-helices, H8 and H9 according to PDB 1RKP (28), with the H9 residue H595 making a hydrogen bonding interaction with the LH2 residue E449, and the H9 L597 and L600 packing against the LH2 V446 (Fig. 3A). This interaction blocks any potential movement of the H-loop towards the catalytic pocket, which was observed in the structures of the PDE5 catalytic domain complexed with sildenafil and icarisid II (Fig. S3A) (29). Furthermore, stabilization of the H-loop allows the H8 helix to form an interface with the M-loop residues, thereby possibly constraining the conformation of the M-loop as well (Fig. 3B). In this interface, the sidechains of N586, Y589, R592, and S593 from the H-loop make H-bond interactions with the M-loop Q714 sidechain, R717 backbone, the sidechains of E721, E718, and E721, respectively (Fig. 3B). The conformations of the H- and M-loops in PDE6C/5 chimera closely resemble only the avanafil-bound structure among PDE5 structures (PDB 6L6E) (Fig. S3B) (33). Remarkably, the H- and M-loop conformations in PDE6C/5 are also very similar to those in the structures of rod and cone PDE6 regardless of the ligand presence in the catalytic pocket (Fig. S3C).

Figure 3.

Interaction of the H-loop with the LH2 helix stabilizes PDE6C/5 chimera in the open state. A, in the structure of PDE6C/5, the H9 loop residues H595, L597, and L600 interact with the LH2 residues E449 and V446. B, stabilization of the H-loop conformation allows it to form an interface with and stabilize the M-loop. The H-loop N586, Y589, R592, and S593 interact with Q714, R717, E721, E718, and E721 from the M loop, colored pale cyan.

To test if the interaction with LH2 in PDE6C/5 is affecting the H-loop mobility, we conducted molecular dynamics (MD) simulations of the PDE5 catalytic domains in the context of PDE6C/5 in comparison to the individual catalytic domain. The latter was represented by the structure from PDB 6L6E with a deleted avanafil molecule. These simulations demonstrated markedly increased dynamics of the H-loop in the individual PDE5 catalytic domain compared to that in PDE6C/5 (Fig. 4, A and B). In one of the simulations, the displacement of the H-loop residues from the original positions towards the opening of the catalytic pocket was particularly dramatic (Fig. 4, A and C). The average RMSF values for the M-loop residues were also greater in MD simulations of the individual PDE5 catalytic domain, albeit the difference was not as large as for the H-loop (Fig. 4, A and B). Large differences in the mobility of the H-loop were also evident from MD simulations of the sildenafil-bound individual PDE5 catalytic domain (PDB 2H42) and that from PDE6C/5 (Fig. S4).

Figure 4.

Molecular dynamics simulations of the PDE5 catalytic domain in the context of the PDE6C/5 structure. A, B, RMSF analyses of MD trajectories of the individual PDE5 catalytic domain structure (PDB ID 6L6E) (A) and the catalytic domain in the catalytic dimer from the PDE6C/5 structure as shown in inset (B) (four simulations each; trace colors – black, red, blue and green). One MD simulation of 6L6E (A, blue trace) shows particularly large displacements of the H-loop. A gray trace in (B) is an overlaid trace from (A, red) to illustrate the difference. Residues in (B) are numbered according to (A). C, D, overlay of the starting conformation of the PDE5 catalytic domain (cyan) with a trajectory frame (green) from an MD simulation (C, blue trace in A: D, red trace in A) showing large conformational variability of the H-loop (blue, dark green) in the structure of the individual domain.

Catalytic properties of the PDE6/5 chimera and the individual PDE5 catalytic domain

In contrast to the structured and constrained H- and M-loops in PDE6C/5, these loops are disordered in the structure of the individual apo PDE5 catalytic domain (PDB 2H40) (29). The loops, particularly the H-loop, have been shown to influence the catalytic properties of PDE5 (29). To test if constraining the H- and M-loops in PDE5 and PDE6C/5 affects the kinetic properties of the enzyme, we expressed and purified the individual PDE5 catalytic domain (PDE5CD) to compare its kinetic properties with those of the full-length proteins. Unexpectedly, we found that the specific cGMP hydrolytic activity (k_cat) of PDE5CD preparations was higher in diluted preparations (≤1 μM) and it significantly decreased at concentrations ≥2 μM (Fig. 5, A and B). To explain the effect of PDE5CD activation upon dilution, we considered the possibility that the protein undergoes concentration dependent dimerization that inhibits its activity. The fits of the PDE5CD concentration dependent cGMP hydrolysis (Fig. 5A) and its specific activity (Fig. 5B) to the simple monomer-dimer model indicate that the data, albeit with deviation, can be related to the free monomer concentration and the fraction of the monomer, respectively. The fits are improved when a model is used where the rate of cGMP hydrolysis is also proportional to the dimer concentration with a small negative coefficient, which may indicate some PDE5CD association beyond dimerization (Fig. 5AB). Moreover, analysis of the 3-Isobutyl-1-methylxanthine (IBMX)-bound structure of PDE5CD (PDB 1RKP) for potential dimerization interface using PDBePISA suggested a dimeric quaternary structure that is predicted to be stable in solution (Figs. 5C, S5A). Further analysis of the dimeric structure using PRODIGY server confirmed this prediction (34). Consistent with the concentration effect on the activity of PDE5CD, the catalytic pockets in the dimeric PDE5CD are occluded by the H-loops at the dimer interface (Fig. 5C). To provide more direct experimental support for dimerization of PDE5CD, we performed crosslinking of the protein with disuccinimidyl suberate (DSS) in the absence and presence of IBMX. This experiment demonstrated the formation of the crosslinked product corresponding to dimeric PDE5CD regardless of the presence of IBMX (Fig. S5B).

Figure 5.

Concentration-dependent activity of the isolated PDE5 catalytic domain. A, effect of the protein concentration on the cGMP hydrolysis by the PDE5CD. cGMP hydrolysis was measured in the presence of 0.5 mM [cGMP + [³H]cGMP] (100,000 cpm) and increasing concentrations of the purified PDE5CD. B, the data in A are replotted as k_cat values. The data in A and B were fit to a simple monomer-dimer model, which assumes dimerization-dependent inactivation of active PDE5CD monomers, and that the rate of cGMP hydrolysis and k_cat are proportional to monomer concentration and fraction of the monomer, respectively, according to the equation: Y = A∗(K_D/4)∗(sqrt(1 + (8∗X)/K_D) - 1) (blue lines; R² = 0.83 in A and 0.84 in B). Alternatively, the data were fit to a monomer-dimer model, which assumes that the rate of cGMP hydrolysis Y is proportional to monomer and dimer concentration with coefficients A and B, respectively, according to the equation: Y = (B/2)∗X + (A − B/2)∗(sqrt(K_Dˆ2 + 8∗K_D∗X) - K_D)/4 (redlines; R² = 0.97 in A and 0.94 in B), where X is total protein concentration X = [M] + 2[D], and the k_cat values calculated as Y/X. From the fitting curve (redline in A): A, 0.27; B, −0.13; K_D 2.5 μM. C, analysis of PDE5CD structure (PDB 1RKP) using PDBePISA indicates a stable dimerization interface whereby the catalytic pockets are occluded by the H-loops (highlighted in blue and dark green).

Michaelis-Menten analysis of PDE5 and PDE6C/5 revealed no significant differences in the K_M values for these proteins (Fig. 6A). However, the K_M value of 6.6 μM cGMP for PDE5CD was about 3-fold higher, indicating its reduced affinity for the substrate (Fig. 6A). A similar K_M for the catalytic fragment of PDE5 was reported earlier (35). Comparison of the maximal activities of PDE5 and PDE6C/5 revealed that it was ∼30% fold higher for PDE5, suggesting that additional intermolecular interaction(s) between the PDE5 catalytic domain and the rest of the molecule may increase the catalytic activity of the enzyme. Notably, the maximal measured activity of PDE5CD was ∼4-fold lower than the activity of the full-length PDE5 and ∼3-fold lower that of PDE6C/5 chimera (Fig. 6B).

Figure 6.

Kinetic properties of the PDE6/5 chimera, PDE5 and PDE5CD. A, Michaelis-Menten analysis of PDE6C/5, PDE5 and PDE5CD. Representative experiment is shown. From four similar experiments the K_M values (μM cGMP, Mean ± SD): PDE6C/5, 2.2 ± 0.4; PDE5, 2.6 ± 0.6; PDE5CD 6.6 ± 0.8. Unpaired t test PDE6C/5 versus PDE5CD, p∗∗ = 0.004; PDE5 versus PDE5CD, p∗∗ = 0.003. B, the calculated k_cat values (s^-1) of cGMP hydrolysis by PDE5, PDE6C/5, and PDE5CD were (Mean ± SD) 0.66 ± 0.11, 0.52 ± 0.06, and 0.17 ± 0.03, respectively. Unpaired t test PDE5 versus PDE6C/5, p∗ = 0.026; PDE5 versus PDE5CD, p∗∗∗∗<0.0001; PDE6C/5 versus PDE5CD, p∗∗∗∗<0.0001.

Discussion

The structure of the PDE6C/5 chimera reported in this study has important implications for understanding the structure and function of PDE5, a critical pharmacological target in treatment of various pathological conditions and diseases. The revealed structural similarity of PDE6C/5 to rod and cone PDE6 enzymes suggests that the wild-type PDE5 in the cGMP-activated state adopts an open conformation, whereby the catalytic pockets are fully accessible to cGMP. Multiple structural elements stabilize the open state in PDE6, including the H-loop interaction with the LH2 linker, the GAF-B β1-β2 loop interaction with the catalytic domain, and Pγ stapling contacts with the GAF-B and catalytic domains (26, 27). The allosteric communication pathway in PDE6 whereby the GAF-A β1-β2 loop makes contacts with the GAF-B domain, whereas the GAF-B β1-β2 loop interacts extensively with the catalytic domain, is thought to play a key role in maintaining the open state of PDE6 (Fig. S6) (26, 27). In contrast, only the H-loop/LH2 stabilizing contacts are retained in PDE6C/5 (Fig. S6). While probable allosteric contacts in PDE5 can alter the position of its catalytic domains relative to that in PDE6C/5, significant differences seem unlikely. This is because of the similar position of the PDE6C/5 catalytic domains to that in PDE6 with stable allosteric interactions. Thus, two possibilities might be envisioned for the conformation of apo PDE5: (a) it is in the intermediate state with low catalytic activity due to increased dynamics of the H- and M-loops, and (b) it is in the closed state similar to that of apo PDE2. Following the transition to the activated open state, new contacts are likely established between the GAF-B and the catalytic domain in PDE5 since its catalytic activity is higher than that of PDE6C/5.

The predicted dimerization of PDE5CD is not physiological, and there is no dimeric interface between the catalytic domains in the open-state structure of PDE6C/5. However, it points to the fact that the use of the individual PDE5 catalytic domain in biochemical analyses can introduce artefacts such as dimerization. Also, the dimer interface of PDE5CD is distinct from the one described in the closed state of apo PDE2A, suggesting the propensity of the H-loops to contribute to different dimerization interfaces (25).

The structure of the PDE6C/5 chimera is also the first structure of the PDE5 catalytic domain in the context of the full-length protein. As such, it highlights limitations of the numerous structural and biochemical studies of the isolated PDE5 catalytic domain. In the isolated PDE5 catalytic domain, the H- and M-loops at the opening of the catalytic pocket adjust their conformations to participate in inhibitor binding, and thus, they are thought to contribute to the inhibitor selectivity (28, 29). A critical consequence of the constraining interaction of the H-loop with the LH2 linker of a partner catalytic subunits is the immobilization of the H- and M-loops in a defined conformation that is similar across different families (PDE5 or PDE6) irrespective of the bound ligand (27). Comparison of the sildenafil-bound individual PDE5 catalytic domain with that in the sildenafil-bound PDE6C/5 structure is striking. In the structure of the isolated PDE5 catalytic domain (PDB 2H42) (29), the H-loop moves as much as 25 Å relative to its position in PDE6C/5 to cover the active site and interact with sildenafil (Fig. S3A). Evidently, such a conformational variability of the H-loop is not physiologically relevant. Subsequent to structural analyses of the PDE5 catalytic domain complexes with drugs, conformational dynamics of its apo and various liganded states have been analyzed by molecular dynamics simulations and other computational techniques (36, 37, 38, 39). The structure of PDE6C/5 suggests that these studies should be re-evaluated to provide an accurate description of the dynamics of the PDE5 catalytic site and its interactions with drugs. Furthermore, conformational stabilization of the H- and M-loops that are proximate to the catalytic site with a binuclear Zn²⁺ and Mg²⁺ metal center may also underly the markedly higher catalytic activity of PDE5 and PDE6C/5 compared to PDE5CD as demonstrated in this study.

Lastly, although the overall conformation of PDE6C/5 is remarkably similar to that of mature PDE6, the former was attained in the absence of a specialized PDE6 chaperone machinery comprised of HSP90, AIPL1, and Pγ (21, 31, 40). This suggests that the folding deficiency of immature PDE6 is confined to the catalytic domain. This notion combined with the reported structures of the maturation complexes involving HSP90 and AIP(XAP2)) (41, 42) favor a hypothetical model of the maturation complex whereby HSP90-bound AIPL1 cradles the catalytic domain of PDE6, whereas a segment of the LH2 linker connecting the GAF-B and catalytic domain is unfolded in the HSP90 lumen. Indeed, one of the AlphaFold3 models of the PDE6 maturation complex (Fig. S7) satisfies these predictions.

Experimental procedures

Recombinant bacmid generation and transfection

Sequences encoding bovine PDE5 and the chimeric PDE6C/5 enzyme were cloned into the pFastBacHTb vector for expression as His6-tagged proteins using the Bac-to-Bac system (Life Technologies, Inc.) as described previously for chimera 4 (31). In this N-terminally His6-tagged chimera, PDE6C sequence (aa 1–440) is fused to the PDE5 sequence (aa 506–865). Recombinant bacmids were obtained by transforming the constructs in E. coli DH10Bac and used for the transfection of Sf9 cells using PEI. The virus thus generated was amplified further and the Sf9 cells were infected at a density of 2.5 × 10⁶cell/ml with the P2 virus stock of PDE5 and chimera. The cells were harvested 96 h after infection by centrifuging at 1000g for 15 min.

Protein purification

The harvested Sf9 cells were resuspended in the lysis buffer containing 50 mM Tris-HCl (pH 7.5), 200 mM NaCl, 1 mM TCEP, 2 mM MgCl₂, 5% glycerol (buffer A) supplemented with 2 mM Phenylmethylsulfonyl Fluoride and cOmplete, EDTA-free Mini Protease inhibitor cocktail (Roche). The cells were lysed using sonication followed by centrifugation of the lysate at 20,000 rpm for 1 h at 4 °C. PDE5 and PDE6C/5 from cleared lysates were purified using Ni-NTA resin (Genscript) and further purified by gel filtration using Superose 6 10/300 Gl (Cytiva) equilibrated with buffer containing 10 mM HEPES (pH 7.5), 150 mM KCl, 1 mM TCEP, and 5 mM MgCl₂. The construct for bacterial expression of PDE5CD in the pET15b vector was prepared, and the protein was purified as described (43). The construct encoding bovine rod Pγ in the pET11a vector was transformed in Rosetta (DE3) cells and expressed in terrific broth media at 20 °C for 14 to 16 h. The cells were harvested and were lysed in a buffer containing 50 mM TRIS (pH 7.5), 200 mM NaCl, 1 mM TCEP, 5% glycerol, 2 mM PMSF, and complete Protease Inhibitor Cocktail (Roche). The cleared lysate was purified using NiNTA (Genscript) followed by cation exchange chromatography (Cytiva) with 800 mM NaCl elution buffer. The protein was further purified by reverse-phase HPLC as described (44).

Cryo-EM sample preparation, data collection, and processing

The sample of PDE6C/5 was concentrated to 0.6 mg/ml and incubated with 5 mM cGMP, 30 μM sildenafil, and 12 μM rod Pγ for 10 min at 25 °C (40). The sample (3 μl) was then applied on glow-discharged holey carbon grids (R 1.2/1.3, Quantifoil), and the grids were blotted in a Vitrobot (Thermo Fisher) (humidity 100%, 4 °C) for 3 s before plunge freezing in liquid ethane. Samples were imaged on a Titan Krios TEM (ThermoFisher Scientific) equipped with a Gatan K3 direct electron detector at a magnification of 105,000 and physical pixel sizes of 0.8267 Å (0.41335 Å super resolution mode). The movies were collected with 40 frames each, with a defocus range of −1.0 to −3.0 μm and a total exposure of 60 e−/Ų. 7078 movies were collected.

The cryo-EM data processing for all reconstructions was performed using cryoSPARC V4.7.1 (45). The movies were motion-corrected using patch motion correction with binning by 2 followed by patch CTF estimation. Particle picking was conducted using a Topaz model generated previously (27). After Topaz picking and extraction with additional binning by 2, several rounds of 2D classification were performed to select the 2D class averages with high resolution features. Four ab initio models were reconstructed with the selected particles followed by 2 rounds of heterogeneous refinement. For the final refinement steps, particles were re-extracted without binning. Homogeneous refinement, non-uniform, and local refinement were used to generate the final map. A flow chart for cryo-EM data processing is shown in Figure S1.

Model building and refinement

An AlphaFold3 model of PDE5 and the structure of PDE6C from PDB 9CXG were docked into the cryo-EM map using UCSF Chimera (46, 47). The coordinates of the GAF-A/GAF-B domain of PDE6C was joined with the coordinates of the catalytic domain of PDE5 using a text editor, and again flexible fitted into the map using NAMDinator (48). Subsequently, the PDE6C residues were mutated to the bovine sequence, and PDE6C/5 chimera model was iteratively refined by manual refinement in coot and automated refinement in phenix (49, 50).

PDE activity assays

The assays were conducted in 40 μl (final volume) of 20 mM Tris–HCl (pH 7.5) buffer containing 120 mM NaCl, 2 mM MgSO4, 1 mM BME, 0.1 U bacterial alkaline phosphatase, and various concentrations of [cGMP + [³H]cGMP] (100,000 cpm) (PerkinElmer) for 10 to 20 min at 25 °C. The reaction was stopped by adding AG1-X2 cation exchange resin (0.5 ml of 20% bed volume suspension) (Bio-Rad). Samples were incubated for 6 min at 25 °C with occasional mixing and spun at 10,000g for 3 min. 0.25 ml of the supernatant was removed for counting in a scintillation counter.

Protein concentrations of purified PDE5, PDE6C/5, and PDE5CD were determined using calculated ε₂₈₀ values. The K_M values were determined by fitting the data to the Michaelis-Menten equation. Assuming that the protein concentration dependence of cGMP hydrolysis by PDE5CD (Fig. 5A) arises from dimerization (inactivation) of active PDE5CD monomers, the data were fit to a simple monomer-dimer model 2M⇌D with K_d= [M²]/[D], total protein concentration X = [M] + 2[D], and the rate of cGMP hydrolysis Y proportional to [M] with a coefficient A, whereas calculated k_cat values (Y/X) are proportional to the fraction of the monomer present: Y = A∗(K_D/4)∗(sqrt(1 + (8∗X)/K_D) − 1). The data were also fit to a monomer-dimer model, which assumes that the rate of cGMP hydrolysis Y is proportional to monomer and dimer concentration with coefficients A and B, respectively, according to the equation: Y = (B/2)∗X + (A − B/2)∗(sqrt(K_Dˆ2 + 8∗K_D∗X) − K_D)/4. If the model correctly approximates the system, coefficient A can be interpreted as specific activity of the monomer, and B, if negative, as negative feedback from a dimer on the activity or the concentration of the monomer.

Cross-linking

Cross-linking reactions were initiated by adding disuccinimidyl suberate (DSS) (0.3 mM final

concentration) to PDE5CD (5 or 15 μM) in the absence and presence of 0.2 mM IBMX in 20 mM HEPES (pH 8.0) buffer containing 150 mM KCl, and 1 mM TCEP at 25 °C. The reactions were allowed to proceed for 15 min followed by quenching with the addition of Tris-HCl (pH 7.5) to a final concentration of 30 mM. Crosslinked products were resolved by SDS-PAGE.

Mass photometry

Mass Photometry experiments were performed on a Refeyn TwoMP (Refeyn Ltd). No.1.5, 24 mm × 50 mm microscope coverslips (Thorlabs Inc.) were cleaned by serial rinsing with Milli-Q water and HPLC-grade isopropanol (Sigma Aldrich), on which a CultureWell gasket (Grace Bio-labs) was then placed. All measurements were performed at 25 °C in Dulbecco's phosphate-buffered saline (DPBS) without calcium and magnesium (Thermo Fisher). For each measurement, 15 μl of DPBS buffer was placed in the well for focusing, after which 3 to 5 μl of 100 nM protein was introduced and mixed. Movies were recorded for 60 s at 50 fps under standard settings. MP measurements were calibrated using protein standard mixture: β-Amylase (56, 112 and 224 kDa), and Thyroglobulin (670 kDa). From the calibration, the estimated error of the MP measurements was 0.2%. MP data were processed using DiscoverMP (Refeyn Ltd).

Molecular dynamics simulations

MD simulations were performed with YASARA Structure 18.2.7 using the md_runfast macro. For the starting conformation of the individual PDE5 catalytic domain, avanafil was extracted from PDB 6L6E (26). For the starting conformation of the catalytic domains in the context of the PDE6C/5 protein, the PDE5 portion was extracted from the PDE6C/5 structure. The simulations were run using the AMBER14 force field in water at a temperature of 310°K, pH of 7.4, and NaCl concentration of 0.9%. Particle mesh Ewald summation was used to compute long-range columbic interactions with a periodic cell boundary and a cutoff of 8 Å. Four independent MD runs were performed on each of the starting models with simulation times ranging from 150 to 250 nsec. Analysis of MD trajectories for per residue fluctuation (RMSFs) was performed using the MD_analyzeres macro and by specifying superposed Cα atoms within the macro.

Statistical analysis

Unless otherwise indicated, measurements were taken from at least three independent experiments, and data are shown as mean value and standard deviation. Measurements between two groups were compared using t test. GraphPad Prism 10 (GraphPad Software Inc., La Jolla, CA, USA) was used for data fitting and analysis.

Data availability

The electron density map and structure file have been deposited with EMDB ID EMD-75366 and PDB ID 10OZ.

Supporting information

This article contains supporting information.

Conflict of interest

The authors declare that they do not have any conflicts of interest with the content of this article.

Reviewed by members of the JBC Editorial Board. Edited by Kirill Martemyanov