The molecular architecture of photoreceptor phosphodiesterase 6 (PDE6) with activated G protein elucidates the mechanism of visual excitation

Photoreceptor phosphodiesterase 6 (PDE6) is the central effector of the visual excitation pathway in both rod and cone photoreceptors, and PDE6 mutations that alter PDE6 structure or regulation can result in several human retinal diseases. The rod PDE6 holoenzyme consists of two catalytic subunits (Pαβ) whose activity is suppressed in the dark by binding of two inhibitory γ-subunits (Pγ). Upon photoactivation of rhodopsin, the heterotrimeric G protein (transducin) is activated, resulting in binding of the activated transducin α-subunit (Gtα) to PDE6, displacement of Pγ from the PDE6 active site, and enzyme activation. Although the biochemistry of this pathway is understood, a lack of detailed structural information about the PDE6 activation mechanism hampers efforts to develop therapeutic interventions for managing PDE6-associated retinal diseases. To address this gap, here we used a cross-linking MS-based approach to create a model of the entire interaction surface of Pγ with the regulatory and catalytic domains of Pαβ in its nonactivated state. Following reconstitution of PDE6 and activated Gtα with liposomes and identification of cross-links between Gtα and PDE6 subunits, we determined that the PDE6–Gtα protein complex consists of two Gtα-binding sites per holoenzyme. Each Gtα interacts with the catalytic domains of both catalytic subunits and induces major changes in the interaction sites of the Pγ subunit with the catalytic subunits. These results provide the first structural model for the activated state of the transducin–PDE6 complex during visual excitation, enhancing our understanding of the molecular etiology of inherited retinal diseases.

The photoreceptor cGMP phosphodiesterase (PDE6) 4 is the central effector enzyme of the G protein-coupled visual transduction pathway in vertebrate rod and cone photoreceptors. PDE6 is exquisitely regulated by a cascade of reactions beginning with photoactivation of the visual pigment, opsin, and subsequent activation of the heterotrimeric G protein, transducin, in the signal-transducing outer segment of the photoreceptor cell. The activated transducin ␣-subunit (Gt ␣ ) then binds to the membrane-associated PDE6 and accelerates its hydrolytic activity to transiently lower cGMP levels in the photoreceptor outer segment. This results in the closure of cGMP-gated ion channels and hyperpolarization of the membrane, leading to synaptic transmission to other retinal neurons (1).
PDE6 belongs to the 11-member phosphodiesterase enzyme superfamily that shares a highly conserved catalytic domain responsible for the hydrolysis of the intracellular messengers cAMP and cGMP (2,3). In addition to the C-terminal catalytic domain, the catalytic subunits of PDE6 consist of two N-terminal regulatory GAF (regulatory domain found in certain PDEs, bacterial adenylyl cyclases, and the bacterial transcription factor FhlaA) domains (GAFa and GAFb) that are also present in four other PDE families (4). However, PDE6 differs from the other 10 PDE families in several important respects: (a) unlike the other 10 homodimeric PDE families (as well as cone PDE6), rod PDE6 is composed of two different catalytic subunits, ␣ and ␤, that form a heterodimer (P␣␤); (b) PDE6 catalysis is uniquely regulated by an intrinsically disordered, 9.7-kDa inhibitory ␥-subunit (P␥) that interacts with both the regulatory and catalytic domains of each catalytic subunit to form the nonactivated rod PDE6 holoenzyme (stoichiometry ␣␤␥␥); (c) rod and cone PDE6 are the only PDEs whose activation directly results from binding of a G protein, specifically the activated Gt ␣ subunit; and (d) upon activation, PDE6 catalysis occurs at the diffusion-controlled limit, more than 2 orders of magnitude larger than the catalytic turnover rate of other PDE families (reviewed in Ref. 5).
Numerous biochemical approaches have been undertaken to understand the molecular mechanism by which Gt ␣ binds to the nonactivated PDE6 holoenzyme and relieves the inhibitory constraint of P␥ on PDE6 catalysis. It has been conclusively demonstrated that in the nonactivated state of the PDE6 holoenzyme, the C-terminal portion of P␥ binds to the catalytic domain and blocks access of substrate to the enzyme active site (6 -8). Catalytic activation of PDE6 is believed to result from interactions of the switch II-␣3-helix region of Gt ␣ with the C-terminal region of P␥ that displaces it from the catalytic pocket of PDE6 (9). This same region of P␥ also modulates the GTPase activity of Gt ␣ (10) by potentiating the activity of RGS9 (regulator of G protein signaling 9) that binds to Gt ␣ and P␥ in this inactivation complex (11). Additional sites of interaction between the Gt ␣ and the N-terminal, polycationic, and glycinerich regions of P␥ (reviewed in Refs. 12 and 13) have been implicated in regulating the efficacy with which Gt ␣ is able to activate PDE6 (8,(13)(14)(15), as well as modulating the affinity of cGMP for noncatalytic binding sites in the GAFa domain of the PDE6 catalytic subunits (16 -18).
Consistent with the demonstration of structural asymmetry in the binding interactions of the two P␥ subunits with the rod PDE6 P␣␤ heterodimer (19,20), complete activation of PDE6 by Gt ␣ requires the binding of Gt ␣ to two nonidentical binding sites on PDE6 (Ref. 21 and the references cited therein). Because of the limited information on structure-function relationships of PDE6 holoenzyme in its nonactivated and Gt ␣activated states, the molecular sequence of events by which Gt ␣ binds to PDE6 to relieve the inhibition of catalysis by P␥ at two different sites is not known.
Building on recent advances to determine the molecular architecture of the PDE6 holoenzyme at the atomic level using integrative structural modeling (20) and cryo-EM (22,23), we present here a structural model for the nonactivated PDE6 holoenzyme that includes the complete interaction surface of its inhibitory P␥ subunits. We also provide a refined structural model for the membrane-associated structure of Gt ␣ and its association with P␥, as well as the complex of the activated G protein ␣-subunit (Gt ␣ -GDP-ALF 4 Ϫ ) with PDE6. In addition to elucidating the mechanistic basis of the first steps in visual signaling, this work provides insights into the molecular etiology of retinal diseases associated with mutations in transducin and PDE6.

Solution structure of the PDE6 catalytic heterodimer
Upon comparing the 3.4 Å cryo-EM structure of the PDE6 holoenzyme (23) with our previous solution structure of PDE6 catalytic dimer determined by chemical cross-linking, identification of cross-linked peptides by mass spectrometric analysis, and integrative structural modeling (20), we observed that a number of distance restraints defined by our cross-linking results were inconsistent with the cryo-EM structure (e.g. cross-links in the ␤-subunit between residues 675 and 813 and between residues 675 and 815; Table 1). In addition, neither of the above-mentioned studies resolved the entire structure of the P␥ subunits that are tightly bound to the PDE6 catalytic dimer in its nonactivated state. We therefore performed integrative structural modeling of the bovine rod PDE6 holoen-zyme using the cryo-EM structure of Gulati et al. (23) as a template (PDB code 6MZB), and the previously reported (20) and new cross-linking data for the PDE6 holoenzyme (Table 1) as inputs into the Integrated Modeling Platform (IMP) and Modeler (see "Experimental procedures") to determine the complete structure for the tetrameric PDE6.
As shown in Fig. 1A, the cross-link-refined solution structure of the PDE6 holoenzyme fits well within the cryo-EM envelope (23), with the spatial restraints imposed by the crosslinking results generating a more compact arrangement of structural elements, as well as providing predicted structures for missing elements in the cryo-EM structure (Fig. S2A). Comparisons at the level of individual domains of our cross-linkrefined PDE6 solution structure with the cryo-EM structure (23) identified several significant differences in conformation ( Fig. S2 (structures) and Fig. S3 (root-mean-square deviation

intra-and intermolecular cross-linked peptides
Cross-linked peptides were identified following chemical cross-linking of 10 -50 pmol of purified rod PDE6 holoenzyme as described under "Experimental procedures." Except where indicated with superscripts, samples consisted of native PDE6 holoenzyme and were digested with trypsin prior to mass spectrometric analysis. Exp. m/z is the experimentally measured mass-to-charge ratio, z is the charge state of the peptide, and ⌬ is the accuracy measured in parts per million. The cross-linked peptides are defined as the protein subunit (pep1 and pep2) and amino acid residue number (aa1 and aa2) identified using the indicated cross-linker. In the aa1 column, the presence of a single-letter amino acid residue preceding the residue number indicates an amino acid substitution of the wild-type P␥ sequence at the site of cross-linking. In addition to the cross-links in this table, the PDE6 structural model included spatial constraints from cross-links reported previously for the PDE6 holoenzyme (20). BMH, 1,6-bismaleimidohexane; BMOE, bis-maleimidoethane; BS(PEG)9, PEGylated bis(sulfosuccinimidyl)suberate; Sulfo-MBS, m-maleimidobenzoyl-N-hydroxysulfosuccinimide ester; Sulfo-SDA, sulfosuccinimidyl 4,4Ј-azipentanoate.  Whereas the GAFa domains showed relatively small differences in secondary structure when compared with the cryo-EM structure (Fig. S2C), the GAFb domains of the PDE6 solution structure exhibited greater dissimilarity (Fig. S2D). Our crosslinking restraints identified conformational differences in several loop structures of GAFb, including the ␤1/␤2 loop that contains a novel ␣2/␣3 helix (Fig. S2D). This loop is in close proximity to the catalytic domain and as previously suggested may play a role in intersubunit allosteric communication (23,24). (c) the catalytic domains of our structural model also exhibited significant differences compared with the cryo-EM structure, particularly in the flexible H-loop and M-loop regions near the enzyme active site and in the ␣16 helix (Fig. S2F). Cross-links in the C-terminal region (Table 1 and Ref. 20) imposed spatial restraints to the conformation of the ␣15 and ␣16 helices in our model that displaced these two helices toward the center of the catalytic domain (and are likely to contribute to the observed conformation of the H-and M-loops), as well as defining additional ␣-helical segments (C␣1 and C␣2) in the C-terminal region. The fact that the C termini of the PDE6 catalytic subunits are prenylated (25) and membrane-associated under our experimental conditions likely accounts for the structural differences we observe in the catalytic domain and C terminus. Together, these observations emphasize the importance of relying on this lower-resolution, chemical cross-linking/MS approach to define both flexible structural elements (e.g. loops) and protein conformations unique to the membrane-associated state that are often chal- The structural model of rod PDE6 holoenzyme (␣␤␥␥) was determined by using the cryo-EM structure 6MZB (23) as a template and applying spatial restraints determined by chemical cross-linking of purified bovine rod PDE6 (Table 1 and Ref. 20). In the model, PDE6 subunits are colored as follows: ␣-subunit (P␣), cyan; ␤-subunit (P␤), green; P␥ subunit primarily associated with ␣-subunit (P␥(P␣)), red; and P␥ subunit primarily associated with ␤-subunit (P␥(P␤)), deep purple. A, superimposition of the template cryo-EM map (EMD-9297) with the cross-linkrefined structural model of nonactivated PDE6 holoenzyme. B, asymmetric interactions of P␥ with the P␣␤ catalytic dimer extending from the cGMP-binding GAFa domain to the GAFb domain and then crossing over to the catalytic domain to the site of inhibition of catalysis. Each P␥ subunit primarily interacts with one catalytic subunit. The two images are rotated 180°. C, interaction surface of the P␥(P␣) subunit with the PDE6 catalytic dimer. P␥(P␣) residues interacting with the catalytic dimer are shown as main-chain atom spheres: red, residues interacting with the ␣-subunit; pink, residues interacting with the ␤-subunit; and yellow, P␥ residues that interact with both catalytic subunits. Noninteracting P␥(P␣) residues are shown as red loops and ␣-helix. The catalytic subunit interacting residues are shown as a surface representation (␣-subunit, dark cyan; ␤-subunit, dark green). D, interaction surface of the P␥(P␤) subunit with P␣␤. The interaction surface of the P␥(P␤) subunit (180°rotation in C) is depicted in which the deep purple, light purple, and orange spheres represent interactions with the ␤-subunit, ␣-subunit, or both catalytic subunits, respectively.

Molecular architecture of G protein-PDE6 activation complex
lenging to obtain from high-resolution, unbiased structural methods such as X-ray crystallography and cryo-EM.

Each intrinsically disordered P␥ subunit forms multiple interactions with both PDE6 catalytic subunits
To map the entire interaction surface of P␥ with the PDE6 catalytic dimer, we performed cross-linking experiments with a variety of chemical cross-linkers, as well as using several sitedirected mutants of P␥ that were reconstituted with P␣␤. The 21 new intermolecular cross-links between P␥ and the ␣or ␤-subunits (Table 1) along with previously reported cross-links (20) and the cryo-EM structure of two fragments of P␥ (23) permitted visualization for the first time of the molecular architecture of the entire PDE6 holoenzyme. Fig. 1B shows that the overall topology of the each P␥ subunit is similar, originating at the noncatalytic cGMP binding pocket in the GAFa domain of one catalytic subunit and terminating at the enzyme active site of the same catalytic subunit. Interestingly, the C-terminal region of P␥consisting of an ␣-helix and a C-terminal "cap"-is similar overall to the crystal structure of a P␥ fragment complexed with a PDE5/6 chimera (7). Although the N-and C-terminal regions of P␥ assume a predominantly linearly extended conformation, the midregion of P␥ exists in a random coil conformation.
Analysis of the interaction surface of P␥ with the catalytic subunits ( Fig. 1, C and D) reveals marked differences in the number and types of interactions of each P␥ with the two catalytic subunits. One P␥ subunit (designated P␥(P␣)) follows the trajectory of the ␣-subunit (Fig. 1C), with approximately onehalf of its 87 residues forming an interaction surface in the GAFa, GAFb, and catalytic domains, ending at the active site of the ␣-subunit. Nine P␥(P␣) residues interact with the ␤-subunit in its GAFa and GABb domains, with four of the nine being in close proximity to both catalytic subunits. The second P␥ subunit (designated P␥(P␤)) has an even greater interaction surface with the catalytic dimer ( Fig. 1D), with 89% of its residues interacting with P␣␤. P␥(P␤) interactions with P␣␤ include 62 residues of the ␤-subunit and 30 residues of the ␣-subunit, with 15 of these residues being in close proximity to both catalytic subunits. The large number of P␥(P␤) interactions with the ␣-subunit is most evident in the GAFb domain where the P␥(P␤) subunit comes into contact with the ␣-subunit GAFb domain (leftward projection in Fig. 1D), as well as multiple interactions of P␥(P␤) with the central ␣-helical "backbone" of both catalytic subunits. This complex network of interactions of both P␥ subunits with both catalytic subunits localized predominantly in the GAFb domains of the catalytic dimer may represent the structural basis for allosteric communication between the ␣and ␤-subunits during transducin activation of PDE6 (see "Discussion").

Structure of membrane associated Gt ␣ and its interactions with soluble P␥
We first carried out cross-linking experiments with activated Gt ␣ attached to liposomes to determine the solution structure of membrane-associated Gt ␣ . Experiments were carried out with Gt ␣ -GDP-AlF 4 Ϫ for which a crystal structure is available (PDB code 1TAD). For the N-terminal ␣-helix (␣N), which is missing from this crystal structure (and proposed to have conformational flexibility) (26), we used as a template the structure of the ␣N helix that was determined for the inactive transducin heterotrimer (PDB code 1GOT). With the ␣N helix and the Gt ␣ -GDP-AlF 4 Ϫ structures as templates and the intramolecular Gt ␣ cross-links that we identified (Table 2), a model of the membrane-associated, activated Gt ␣ -GDP-AlF 4 Ϫ subunit of transducin was created ( Fig. 2A). Intramolecular cross-links (Lys 18 to Lys 267 and Glu 21 to Lys 275 ; Table 2) between the ␣N helix and the Ras-like GTPase subdomain of Gt ␣ imposed spatial constraints that are reflected in a major shift of the ␣N helix toward the ␣F/␤2 loop region that is part of the interface between the GTPase subdomain and the helical insertion subdomain. This shift brings the ␣N helix in proximity with the nucleotide-binding site. We conclude that the structural model shown in Fig. 2A better represents the membrane-associated, solution structure of Gt ␣ in that it takes into account the N-terminal acylation of Gt ␣ responsible for its association with rod outer segment membranes in vivo.
Previous biochemical studies have identified two major regions of P␥ that bind to activated Gt ␣ , namely the polycationic central region of P␥ and the C-terminal half of P␥ (8,(11)(12)(13)(27)(28)(29). To determine the topological relationship of Gt ␣ with P␥, we incubated liposome-associated Gt ␣ -GDP-AlF 4 Ϫ (see "Experimental procedures") with purified P␥ and conducted cross-linking analyses of the protein band migrating at the apparent molecular mass expected for a 1:1 complex of P␥ and Gt ␣ (ϳ50 kDa). We identified five intermolecular, cross-linked peptides spanning residues 25-45 of the central region of the P␥ molecule ( Table 2) that interact with both the Table 2 Intra-and intermolecular cross-linked peptides of membrane-associated Gt ␣ -GDP-AlF 4 ؊ and P␥ Cross-linked peptides were identified following chemical cross-linking of either lipobead-associated Gt ␣ -GDP-AlF 4 Ϫ or Gt ␣ -GDP-AlF 4 Ϫ incubated with a 2-fold stoichiometric excess of purified P␥ and analyzed as described under"Experimental procedures." The abbreviations are defined in the legend to

Molecular architecture of G protein-PDE6 activation complex
helical subdomain and the switch II region of the GTPase subdomain of Gt ␣ ( Fig. 2A). This 20-amino acid segment of P␥ interacts on the opposite face of the Gt ␣ subunit from the interface of Gt ␣ with the PDE6 catalytic domain (see below). As seen in Fig. 2B, P␥ assumes a highly extended linear structure when bound to Gt ␣ compared with the conformation of the same region of P␥ bound to the PDE6 ␣or ␤-subunits. (Although there is structural evidence that the C-terminal half of P␥ binds to the PDE6-facing side of Gt ␣ (11,30), our inability to observe cross-linked peptides between Gt ␣ and this region of P␥ arises from the absence of amino acid residues in the C-terminal half of P␥ capable of generating cross-linked peptides for mass spectrometric detection (20).) No significant changes in the tertiary structure of Gt ␣ were detected upon P␥ binding.

Molecular architecture of the G protein-effector activation complex
Full activation of PDE6 by Gt ␣ is greatly enhanced when both proteins are associated with either rod outer segment membranes or are reconstituted with phospholipid bilayers (31). To determine the structure of the transducin-PDE6 complex in its membrane-associated state, we therefore preincubated purified proteins with cationic phospholipid vesicles that have been shown to enhance PDE6 activation by Gt ␣ (32). To restrict our analysis to only membrane-associated Gt ␣ and PDE6, we prepared liposome-coated silica beads ("lipobeads"; see "Experimental procedures" and Fig. S5) that allowed for sedimentation of membrane-associated proteins for further analysis. Using this method, ϳ90% of the PDE6 holoenzyme was pulled down in the lipobeads pellet (Fig. S5C), and under these conditions we observed at least 80% of maximal activation of PDE6 catalysis by transducin (Fig. S5B).
We first assessed whether all of the cross-linked peptides we observed between Gt ␣ and the PDE6 catalytic subunits could be accounted for by a single Gt ␣ -binding site per P␣␤. Table  3 shows that three PDE6 ␣-subunit-specific and three ␤-subunit-specific cross-links with Gt ␣ ruled out a single binding site per P␣␤, consistent with biochemical studies (21). When 10 cross-linked peptides between Gt ␣ and the ␣or ␤-subunit of PDE6 in Table 3 were used as distance restraints for input into the IMP workflow, we found no single structural model that was able to accommodate all of the cross-link restraints. Instead, we observed two major classes of structural models with different cross-links that violated the distance restraints. The predominant set of structural models was generated by omitting the two cross-links between Gt ␣ and the GAFb domains (Gt ␣ 9-P␣442/P␤440 and Gt ␣ 24-P␣330/ P␤328; Table 3); the remaining eight cross-links permitted docking of Gt ␣ to two similar-but not identical-sites on the ␣and ␤-subunit catalytic domains (Fig. 3A).
Closer examination of the interface of Gt ␣ with the PDE6 ␣-subunit catalytic domain (Fig. 3B) revealed that the GTPase subdomain of this Gt ␣ molecule (including the switch II region A, the structural model of Gt ␣ was determined using the 1TAD crystal structure as the template (36) and refined with spatial restraints imposed from cross-linking results in the absence or presence of P␥ (Table 2). Structural elements that were unchanged in the cross-link-refined model are represented in green, with the conformational change of the ␣N helix shown in brown (for the crystal structure) and blue (for the cross-link modified solution structure). Also shown is the docked structure of P␥ (red) with Gt ␣ -GDP-AlF 4 Ϫ based on the observed cross-linking results when Gt ␣ associated with lipobeads was incubated with a 2-fold molar excess of P␥. Note that no significant changes in Gt ␣ conformation were observed upon P␥ binding. B, a comparison of the conformation of the central region of P␥ (residues 24 -44, depicted as a gradient from blue to red spheres) when bound to Gt ␣ or to the PDE6 catalytic subunits.

Molecular architecture of G protein-PDE6 activation complex
and the ␣N helix) interacts with the ␣14 helix, the M-loop region (implicated in regulating P␥ occlusion of the active site) (7), and the ␣15 and ␣16 helices of the ␣-subunit catalytic domain, in excellent agreement with previous mutagenesis studies (33). Interestingly, the ␣B helix of this Gt ␣ molecule interacts with the adjacent PDE6 ␤-subunit in the linker region between the GAFb and ␤-subunit catalytic domain (Fig. 3B).
The interaction surface of the second Gt ␣ with the ␤-subunit was generally similar to the ␣-subunit, but with a greater surface of interaction reflecting additional interactions with the region of the ␤-subunit linking the GAFb and catalytic domains. For both Gt ␣ subunits, the N-terminal ␣-helix has significant surface interactions with the catalytic domains. With the cross-links in Table 3, we were unable to observe any significant changes in Gt ␣ -GDP-AlF 4 Ϫ conformation upon interaction with P␣␤. The same was true for the PDE6 catalytic dimer where the overall root-mean-square deviations for each catalytic subunit was ϳ1.0 A when comparing the nonactivated and transducin-activated conformations of each PDE catalytic subunit (Fig. S4). Interestingly, the interaction surface of Gt ␣ with PDE6 catalytic subunits is similar to that observed for the complex of a membrane-bound adenylyl cyclase with the G s ␣-subunit, particularly in the switch II and ␣3 helix regions of the GTPase domain (34).
A second cluster of structural models of Gt ␣ -activated PDE6 docking with the GAFb domain ( Fig. 3C) was identified when three different cross-links at the bottom of the P␣␤ catalytic domains (Gt ␣ 10-P␣854, Gt ␣ 17-P␣551, and Gt ␣ 128-P␣807/ P␤805) were omitted during the structural modeling (Table 3). Although insufficient cross-linking data for the P␥ subunit precluded structural modeling of P␥ in the transducin-activated PDE6 complex, the same central region of P␥ that binds to purified Gt ␣ in an extended conformation ( Fig. 2A) is associated with the GAFb domain of nonactivated PDE6 (Fig. 1, C and  D) and likely promotes Gt ␣ binding to the GAFb domain shown in Fig. 3C. This second binding site for Gt ␣ is supported by biochemical studies indicating a role for the central region of P␥ in facilitating Gt ␣ activation of PDE6 catalysis (8), as well as Ϫ bound to lipobeads (see "Experimental procedures") were exposed to chemical cross-linkers, and the identified cross-linked peptides between Gt ␣ and PDE6 subunits (Table 3) were then used as spatial restraints for integrative structural modeling. Two predominant clusters of models of the Gt ␣ -P␣␤ complex were generated, one with Gt ␣ docked to the two catalytic domains (with distance violations for Gt ␣ 24-P␣330/P␤328, Gt ␣ 9-P␣442, and Gt ␣ 9-P␤440) and the other with Gt ␣ docked to the GAFb domains (with distance violations for Gt ␣ 10-P␣854, Gt ␣ 17-P␣551, and Gt ␣ 128-P␣807/P␤817). Because of insufficient cross-links for P␥ in the activated complex, the inhibitory subunit is not shown. A, structural model of association of Gt ␣ -GDP-AlF 4 Ϫ to the ␣-subunit (Gt ␣ (P␣), orange) and to the ␤-subunit (Gt ␣ (P␤), blue) catalytic domains. B, detailed view of the Gt ␣ GTPase subdomain interface with the ␣-subunit catalytic domain, with the interaction surface of Gt ␣ colored red and the ␣and ␤-subunit interacting residues colored magenta and brown, respectively. The black sphere indicates Gt ␣ Gln 200 . C, alternate docking of Gt ␣ to the GAFb domains of the P␣␤ catalytic dimer (with the same orientation as in A). Table 3 Intermolecular cross-linked peptides of the activated complex of Gt ␣ -GDP-AlF 4 ؊ with PDE6 holoenzyme Cross-linked peptides were identified following chemical cross-linking of a mixture of Gt ␣ -GDP-AlF 4 Ϫ and PDE6 holoenzyme attached to lipobeads and analyzed as described under "Experimental procedures." The abbreviations are defined in the legend to Table 1. All identified intra-and intermolecular cross-links involving PDE6 catalytic subunits were identical to those observed in the holoenzyme structure ( a Cross-links that were omitted from specific structural models during docking of Gt ␣ to catalytic domain. b Cross-links that were omitted from specific structural models during docking Gt ␣ to GAFb domain. c Cross-links that were omitted from specific structural models from computational modeling due to loop flexibility.

Molecular architecture of G protein-PDE6 activation complex
enhancing the dissociation of cGMP from GAFa noncatalytic binding sites (13).

Discussion
This paper reports the first complete structural models for the PDE6 holoenzyme ( Fig. 1), the activated ␣-subunit of transducin in a complex with the inhibitory ␥-subunit of PDE6 (Fig.  2), and the fully activated state of PDE6 in a complex with two transducin ␣-subunits (Fig. 3)-all in their membrane-associated state that mimics the localization of the transducin-PDE6 protein complex on photoreceptor outer segment disk membranes. Together, these structural models advance our understanding of the mechanism of visual excitation in rod photoreceptors by revealing the asymmetric surface of interaction between each P␥ subunit and the P␣␤ catalytic dimer, as well as the different sites of interaction of Gt ␣ with PDE6 and the major conformational changes that the P␥ subunits must undergo upon transducin activation of PDE6 in the phototransduction pathway.
Chemical cross-linking combined with mass spectrometric analysis (35) has enabled us to refine the secondary, tertiary, and quaternary structure of PDE6 in its nonactivated and transducin-activated states. It is important to emphasize that our ability to carry out integrative structural modeling of PDE6 in its nonactivated and activated states was enabled by having the atomic-level crystal structure for Gt ␣ -GDP-AlF 4 Ϫ (36) and a high-resolution cryo-EM structure for PDE6 holoenzyme (23). The distance restraints imposed by cross-linked residues within and between proteins comprising the nonactivated and activated states of PDE6 permitted us to dock Gt ␣ subunits to each catalytic subunit of PDE6, thereby providing a structural basis for the allosteric mechanism for G protein-coupled activation of PDE6 during visual excitation-including the functional asymmetry of the PDE6 holoenzyme that underlies the requirement for successive binding of two Gt ␣ molecules for full enzyme activation (Refs. 21 and 37 and references cited therein). This cross-linking/mass spectrometric approach also permitted visualization of flexible regions of the PDE6 catalytic and inhibitory subunits that were poorly resolved by cryo-EM (23), as well as structural elements not available in the existing crystal structures for Gt ␣ .
Our integrative structural modeling of PDE6 reveals the multiple intersubunit interactions that underlie the multifaceted allosteric regulation of this G protein-activated enzyme: (a) each P␥ subunit interacts with both PDE6 catalytic subunits, with lateral, cross-subunit communication likely transmitted through the GAFb domains where a number of P␥ residues are in close proximity to both catalytic subunits (Fig. 1, C and D); (b) the ␤-subunit exhibits greater interactions with P␥ than the ␣-subunit, consistent with two classes of binding sites for P␥ with P␣␤ (18); and (c) in addition to the extensive P␣␤ dimerization surface, direct allosteric communication may occur between the ␤1/␤2 loop in the GAFb domain of one catalytic subunit and the catalytic domain of the other subunit (23), as well as between the catalytic domains and C-terminal regions of the two subunits (Fig. 1B).
Defining the molecular architecture of the transducin-PDE6 -activated complex permitted structural verification of the stoichiometry of two Gt ␣ subunits bound to the PDE6 catalytic subunits in its fully activated state, as well as unexpectedly revealing two distinct sites of interaction of Gt ␣ with the GAFb (Fig. 3C) and catalytic domains (Fig. 3, A and B) of the PDE6 catalytic subunits. The observation that each Gt ␣ subunit has sites of interaction with both the ␣and ␤-subunits of PDE6 is consistent with a cooperative activation mechanism in which the binding of the first Gt ␣ induces conformational changes in P␣␤ that alter the ability of the second Gt ␣ subunit to bind to and trigger full enzyme activation (21,37).
For the model for G protein activation of the central effector enzyme of the visual signaling pathway, Fig. 4 presents a model consistent with our experimental results for the light-induced activation of PDE6 holoenzyme by transducin that involves the sequential binding of two Gt ␣ subunits that results in both Gt ␣ subunits releasing the inhibitory constraint of P␥ from its interactions with each PDE6 catalytic domain to cause full activation of PDE6 (37).
Upon light activation of the phototransduction cascade, nonactivated PDE6 holoenzyme (Fig. 4A) is proposed to form initial interactions between the central region of P␥ (associated with the GAFb domains) and an activated Gt ␣ subunit (Fig. 3C), resulting in the central region of P␥ becoming significantly more extended (Fig. 4B). In this model, the binding of Gt ␣ to this central region of P␥ does not require major displacement of either the N-or C-terminal regions of P␥ from its holoenzyme conformation. Upon binding of a second Gt ␣ , Fig. 4C depicts a A, in the dark-adapted condition, the PDE6 holoenzyme is inhibited by its P␥ subunits occluding the enzyme active site (Fig. 1B,  rotated 90°). B, the first light-activated Gt ␣ subunit is proposed to initially bind to the GAFb docking site (see Fig. 3C) without causing significant catalytic activation of PDE6 (21). The P␥ subunit was docked to this complex using the following information: (a) the central region of P␥ (gold) was docked using the cross-links obtained for the Gt ␣ -P␥ complex ( Table 2) in conjunction with the cross-links used to dock Gt ␣ to the GAFb domain (Table 3); (b) lacking cross-linking data for the N-terminal region of P␥ in the activated complex, this region of P␥ (purple) relied on PDE6 holoenzyme cross-links, and thus its topology only differs from Fig. 1 to the extent needed to accommodate crosslink spatial restraints imposed by the P␥ central region; and (c) in the absence of P␥ cross-links for its C-terminal region in the activated complex, we modeled this region of P␥ (purple) interacting with Gt ␣ using the crystal structure of P␥ (residues 50 -87) bound to a chimeric G protein (PDB code 1FQJ) (11). C, upon binding of a second Gt ␣ , PDE6 becomes fully activated as both Gt ␣ subunits dock to the catalytic domains and displace the C-terminal region of P␥ from the enzyme active sites. To accommodate the binding of the central region of P␥ to the helical face (Table 2) and the C-terminal region of P␥ to the GTPase face of Gt ␣ (Table 3 and Ref. 11), a major displacement of the N-terminal P␥ residues from the GAFa domains must occur.

Molecular architecture of G protein-PDE6 activation complex
relocation of the first Gt ␣ from the GAFb to the catalytic domain, along with binding of the second Gt ␣ to the catalytic domain of the other catalytic subunit (Fig. 4C)-resulting in full enzyme activation. As a consequence of Gt ␣ binding to the central and C-terminal regions of P␥ when docked to the catalytic domains, our model requires that the N-terminal region of P␥ dissociates from its interactions with the GAFa domain (Fig.  4C). This structural model for sequential activation of PDE6 is supported by prior biochemical and structural studies of Gt ␣ interactions with PDE6 subunits in the activated complex (11,33,(38)(39)(40). The required displacement of P␥ from the GAFa domains is also consistent with a lowered affinity of cGMP to its GAFa-binding sites upon transducin activation of rod PDE6 (8,13,17), as well as offering insights into differences in how rod and cone PDE6 may be activated by transducin (15). Experimental support for the model in Fig. 4 is currently under investigation, including validating the GAFb domain as an initial docking site for one or both Gt ␣ subunits, identifying whether the ␣or ␤-subunit preferentially binds the first Gt ␣ , the allosteric communication pathway leading to binding of the second Gt ␣ subunit, and the significance of cGMP occupancy of the GAFa-binding sites for the activation, recovery, and light adaptation stages of visual transduction.
In addition to advancing a structural basis for understanding the initial events in the visual signaling pathway, structural elucidation of PDE6 in its nonactivated and transducin-activated states offers insights into the molecular etiology of pathogenic mutations in these proteins and possible therapeutic interventions. For example, having characterized the interaction surface of Gt ␣ -PDE6, it is now evident that a missense mutation impairing GTPase activity (Q200E; black sphere in Fig. 3B) that is responsible for autosomal dominant congenital stationary night blindness (41) is located at the interface between the Gt ␣ switch II region and the PDE6 catalytic domain where P␥ regulation of catalytic activation occurs. Based on our structural model of Gt ␣ interactions with the central region of P␥ (Fig.  2), we hypothesize that the disease-causing D129G mutation in Gt ␣ (42) eliminates an ionic interaction with P␥ Lys 26 (Ͻ4 Å apart) that participates in the binding of activated transducin to PDE6 holoenyzme. Because somatic mutations in PDE6 catalytic subunit genes have been implicated in various cancers (Ref. 43) and references cited therein), knowledge of the atomic-level structure of PDE6 may become relevant should a causative link be established between PDE6 mutations and tumorigenesis.
Given that abnormal accumulation of cGMP is believed to be the causative factor in many retinal degenerative diseases (44), understanding the structural organization of PDE6 and the protein-protein interactions that regulate its activity may provide insights into development of allosteric regulators of PDE6 analogous to those being developed for other members of the PDE family of enzymes (45,46).

Materials
Bovine retinas were purchased from W. L. Lawson, Inc. The Mono Q, HiTrap SP Sepharose FF, HiTrap Blue HP, and Super-dex 200 columns were from GE Healthcare. The C18 reversephase column (Proto 300, 4.6 ϫ 250 mm) was from Thermo Fisher Scientific. The primers for P␥ mutagenesis and plasmid purification kits were from Invitrogen and Qiagen, respectively. The QuikChange II site-directed mutagenesis kit was from Agilent Technologies. Phospholipids and the Mini-Extruder were from Avanti Polar Lipids. Trypsin and Asp-N were purchased from Promega. Silica particles (70-nm diameter, plain) were obtained from Advance Scientific. Chemical cross-linkers were from Thermo Fisher, and all other reagents were from Millipore-Sigma, Thermo-Fisher, or VWR.

Preparation of purified PDE6
Rod PDE6 holoenzyme (subunit composition, ␣␤␥␥) was isolated from bovine rod outer segments and purified by anionexchange and gel-filtration chromatography as described previously (47). The P␣␤ catalytic dimer was prepared from purified PDE6 holoenzyme by limited trypsin proteolysis to selectively degrade the P␥ subunits; the time course of proteolytic activation of PDE6 catalysis was empirically determined to ensure that Ͼ90% of the P␥ subunit was degraded without altering the apparent molecular mass of the catalytic subunits on SDS-PAGE. P␣␤ was then repurified by Mono Q chromatography (47). Purified PDE6 preparations were stored in 20 mM HEPES, pH 7.5, 100 mM NaCl, 2 mM MgCl 2 (HNM buffer) plus 50% glycerol at Ϫ20°C until use. Just prior to the experiment, the protein was buffer-exchanged and adjusted to the indicated concentration for the cross-linking reaction.
PDE6 catalysis of cGMP hydrolysis was quantified using a coupled-enzyme assay with colorimetric detection of P i (48). The PDE6 concentration was estimated based on the rate of cGMP hydrolysis of trypsin-activated PDE6 and knowledge of the k cat of the enzyme (5600 mol cGMP hydrolyzed per mol P␣␤ per second) (49).

Preparation of persistently activated transducin ␣-subunit
Gt ␣ was selectively extracted from PDE6-depleted rod outer segment membranes by adding either 50 M GTP␥S or 100 M GTP to the ROS membranes and recovering the solubilized Gt ␣ following centrifugation of the membranes. Gt ␣ was subsequently purified by affinity chromatography on a HiTrap Blue HP column (50), followed by Superdex 200 gel-filtration chromatography to remove residual PDE6. The concentration of Gt ␣ was determined by a colorimetric protein assay (51) using bovine ␥-globulin as a standard. Purified Gt ␣ was stored at Ϫ20°C in 50% glycerol supplemented with 50 M of GTP␥S or GDP until use. Prior to a cross-linking experiment, the Gt ␣ -GTP␥S or Gt ␣ -GDP was buffer-exchanged into the appropriate cross-linking buffer. In the case of GDP-bound Gt ␣ , the Gt ␣ was incubated with 30 M AlCl 3 and 10 mM NaF for 15 min on ice to form the activated Gt ␣ -GDP-AlF 4 Ϫ complex (52).

Expression and purification of P␥ mutants
P␥ site-directed mutants were created with the codon-optimized WT bovine rod P␥ sequence as the template and the QuikChange II site-directed mutagenesis kit to introduce amino acid substitutions. The pET11a plasmids with the sequence-verified P␥ mutant sequences were transformed into

Molecular architecture of G protein-PDE6 activation complex
Escherichia coli BL21(DE3) cells and grown at 37°C in 2-YT medium to an A 600 of ϳ0. 6. Then 0.5 mM isopropyl ␤-D-1thiogalactopyranoside was added, and the cells were incubated at 30°C for 6 h. The recombinant P␥ protein was purified from the cell extract using a HiTrap SP column followed by C18 reverse-phase HPLC (53). The apparent molecular mass and purity (Ͼ95%) of the recombinant P␥ protein was verified by SDS-PAGE. P␥ cysteine mutants were prepared as described previously (20). All P␥ mutants were observed to inhibit P␣␤ catalysis over the same concentration range as WT P␥.

Preparation of liposomes and lipobeads to study interactions of transducin with PDE6
Large unilamellar vesicles and sucrose-loaded vesicles (consisting of an 80:20 molar ratio of 1,2-dioleoyl-sn-glycero-3-phosphocholine and 1,2-dioleoyl-3-trimethylammonium-propane) were initially utilized to improve the efficiency of transducin activation of PDE6 (32), closely following established procedures (54). To further improve the ability to quantitatively sediment PDE6 and Gt ␣ attached to the liposomes (and to eliminate soluble proteins), we adapted an existing method to prepare silica bead-supported liposomes (i.e. lipobeads) (55) for membrane association of PDE6 and Gt ␣ . The ability to pulldown PDE6 (ϳ90% PDE bound) and the enhancement of PDE6 activation by Gt ␣ (up to 95% of maximum activation) were equivalent for all of the above liposome preparations.
Lipobeads were prepared by first washing 5 mg of 70-nm silica beads several times with HNM buffer followed by centrifugation for 3 min at 15,000 ϫ g. The bead pellet was then resuspended in HNM buffer. 1,2-Dioleoyl-sn-glycero-3-phosphocholine and 1,2-dioleoyl-3-trimethylammonium-propane were mixed at a molar ratio of 80:20 in chloroform, evaporated, and resuspended in HNM buffer containing the lipobeads to a final phospholipid concentration of 500 M. Unilamellar vesicles coating the silica particles were formed by extruding the mixture 15 times through a 0.1-m polycarbonate membrane using a Mini-Extruder (Fig. S5A).

Chemical cross-linking, in-gel digestion, and MS analysis
Chemical cross-linking reactions were carried out following the manufacturer's protocols for each cross-linker. For crosslinking reactions with BS3, DSS, sulfosuccinimidyl 4,4Ј-azipentanoate, or m-maleimidobenzoyl-N-hydroxysulfosuccinimide ester, proteins were cross-linked in HNM buffer; for EDC cross-linking reactions, 100 mM MES buffer, pH 6.5, was used. After the cross-linking reaction was quenched, proteins were precipitated with TCA, separated by SDS-PAGE, and visualized with Coomassie Brilliant Blue G-250. For the case of the nonactivated PDE6 holoenzyme, a 50-fold molar excess of the cross-linker was used, closely following the protocol of our previous study (20).
To carry out cross-linking reactions with the complex of activated Gt ␣ and PDE6 holoenzyme, PDE6 holoenzyme (10 -50 pmol) was mixed with a 500-fold molar excess of Gt ␣ -GDP-AlF 4 Ϫ or Gt ␣ -GTP␥S along with 0.6 mg of lipobeads. The mixture was incubated at room temperature for 1 h and then spun at 10,000 ϫ g for 1.5 min (Fig. S5). Unbound proteins in the supernatant fraction (ϳ10% of the total PDE6 and ϳ50% of the Gt ␣ ) were discarded, and the lipobead-associated proteins were resuspended and cross-linked for 1 h with the following molar excess of cross-linker relative to PDE6: BS3 or DSS (500-fold), m-maleimidobenzoyl-N-hydroxysulfosuccinimide ester (100fold), sulfosuccinimidyl 4,4Ј-azipentanoate (100-fold), or EDC (1000-fold). Following quenching of the cross-linking reaction with 20 mM Tris, pH 7.5, the samples were spun at 5,000 ϫ g for 1.5 min, resuspended in SDS-PAGE sample buffer, and loaded onto NuPAGE 4 -12% Bis-Tris gels. Protein bands on the gel were visualized with Coomassie Brilliant Blue G-250.
Cross-linked products were in-gel digested and analyzed by LC-MS and LC-MS/MS as described previously (20), except that we also used Asp-N to generate peptide fragments. For Asp-N digestions, 3 ng of Asp-N were added to the gel pieces and incubated for 18 h at 37°C. For proteolytic digestions with both enzymes, 300 ng of trypsin was added to the gel pieces for 4 h, then 3 ng of Asp-N was added, and samples were incubated for an additional 18 h. The tryptic peptides were extracted as described (20), and Asp-N or double-digested peptide samples were extracted using 50% acetonitrile and 7% formic acid.
One-microliter aliquots of the concentrated peptides were injected into the Dionex Ultimate 3000 RSLC nano UHPLC system (Dionex Corporation, Sunnyvale, CA) and separated by a PepMap RSLC column (75 m ϫ 25 cm, 100 Å, 2 m) at a flow rate of 450 nl/min (mobile phase A: 0.1% formic acid in H 2 O, mobile phase B: 0.1% formic acid in 80% acetonitrile). The eluant was directed into the nano-electrospray ionization source of an LTQ Orbitrap XL mass spectrometer (Thermo Scientific). LC-MS data were acquired in an information-dependent acquisition mode. Full MS spectra were acquired in the Orbitrap (m/z 315-2000). The five most intense ions were selected for collision-induced dissociation in the linear ion trap for MS/MS data acquisition (24).

Identification of cross-linked peptides
Peak lists were created using RawConverter version 1. 1.0.19 for input into Protein Prospector (version 5.14.0 or newer). The data were initially searched against the full Swiss-Prot database (version 2013. 6.27 or newer) to verify the absence of contaminating proteins, and then the search was restricted to bovine PDE6 subunits (P11541, P23439, and P04972) and to bovine Gt␣ (P04695). In experiments in which mutant proteins were used, all protein sequences were input as user-defined proteins. Trypsin or a trypsin/Asp-N double digestion was selected with three to five maximum missed cleavages with precursor and fragment mass tolerances of 15 ppm and 0.7 Da, respectively. A maximum of two variable modifications were allowed, including: acetyl (N terminus); acetyl ϩ oxidation (N-terminal methionine); glutamine to pyro-glutamine; methionine-loss (N-terminal methionine); methionine loss ϩ acetyl (N-terminal methionine); oxidation (methionine); and specific cross-linker modifications (when appropriate). This search process also allowed for a single mass modification based on the cross-linker used: for cross-linkers selectable in Protein Prospector (bis-maleimidoethane, DSS/BS3, and EDC), the preset mass modifications were used; for user-defined cross-links, the default values were used.

Molecular architecture of G protein-PDE6 activation complex
Cross-linked peptides were identified using an integrated module in Protein Prospector, using a previously described strategy (20,56) in which a given score is credited for each fragment matched with the score weighting dependent on the type of ion type matched These scores were then converted to expectation values by determining the distribution of scores for random answers and calculating a probability and then an expectation value of a given score being in this distribution. Samples scoring above 10 for a trypsin digest or 5 for a double digestion were validated based on manual inspection of the spectra. Only results where the score difference was greater than zero confirmed that the cross-linked peptide match was better than a single peptide match alone and were therefore considered. Expectation values were calculated based on matches to single peptides and thus were treated as another score rather than as a statistical measure of reliability. False discovery rates were assessed by comparing the cross-linking data to available structures (57) (PDB codes 6MZB for PDE6 and 1TAD for Gt ␣ ).

Integrative structural modeling of PDE6, Gt ␣ , and the Gt ␣ -PDE6 -activated complex
Integrative structural modeling was performed using the open-source Integrated Modeling Platform (58) and Modeler (59) in an iterative manner. To perform rigid body docking of protein subunits, IMP was carried out in 2 ϫ 10 4 Metropolis Monte-Carlo sampling steps with a high temperature of 2.0, a low temperature of 0.5, and with a new system configuration following each step. The top 100 scoring models were generated and saved, and IMP was then used to perform clustering on the top 100 models to aid in model selection. The best fitting model was run in Modeler using the same cross-linking restraints to further refine the model, evaluate stereochemical quality, and fill in the missing atoms. Secondary structure identification was initially determined by PyMOL version 2.3 (Schrodinger) and further refined and validated with Coot (60).
The P␣␤ catalytic dimer refinement was performed using the PDE6 cryo-EM structure (23) as the template (PDB code 6MZB). Structural model refinements used the spatial restraints imposed by cross-linked peptides we identified in samples of native and reconstituted PDE6 catalytic subunits, as described previously (20). The domain boundaries and secondary structure assignments for the PDE6 catalytic subunits are given in the supporting information (Fig. S1). Analysis of the root-mean-square deviations of our structural model with other available structures was carried out using Visual Molecular Dynamics software version 1.9.3 (61).
To model the PDE6 holoenzyme ( Fig. 1), the refined P␣␤ model was used as a single, unchanging rigid body, and each P␥ subunit was treated as eight separate rigid bodies consisting of residues 2-30, 31, 38 -41, 44 -45, 52-53, 58 -62, 68, and 70 -87. This approach circumvented the lack of uniform crosslinking data for the entire P␥ subunit. Two of the rigid bodies (residues 2-30 and 70 -87) were based on the P␥ structure and topology obtained from the PDE6 cryo-EM structure (23). The remaining P␥ peptide fragments were generated in silico (http://www.arguslab.com/arguslab.com/ArgusLab.html), 5 assuming a linearly extended conformation. IMP was then used to dock the P␥ fragments. Subsequently, Modeler was used with the same cross-linking constraints to fill in the missing portions of P␥, as well as to add the missing atoms to each subunit.
The structure of membrane-associated Gt ␣ -GDP-AlF 4 Ϫ was obtained using the X-ray crystal structure of Gt ␣ -GDP-AlF 4 Ϫ (PDB code 1TAD) (36) as the primary template and imposing distance restraints from cross-linked peptides we identified, as described above. Because the cross-linking data of purified Gt ␣ included cross-links from the N-terminal ␣-helix of Gt ␣ that is not included in the 1TAD crystal structure, the Gt ␣ structure was refined by including the N-terminal helix (amino acids 1-27) obtained from the transducin heterotrimer structure (PDB code 1GOT) with the 1TAD structure as two rigid bodies for conducting integrative modeling.
The structure of P␥ docked to Gt ␣ -GDP-AlF 4 Ϫ (Fig. 2) was performed by treating Gt ␣ -GDP-AlF 4 Ϫ as a rigid body and dividing the central region of P␥ into three rigid bodies consisting of residues 25, 39 -41, and 45 in IMP. Modeler was used to refine the structure and add missing atoms to the model.
The structure of the activated complex of Gt ␣ and the P␣␤ catalytic dimer (Fig. 3) was docked using the previously described structures as templates. P␣␤ was treated as a single rigid body, and two Gt ␣ -GDP-AlF 4 Ϫ structures were included in the modeling in IMP, followed by refinement with Modeler.

Data availability
The input data files, modeling scripts, and output models can be accessed at https://github.com/rcotelab/Irwin-et-al-2019. 5 The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository (62) with the data set identifier PXD015989.