The Inhibitory γ Subunit of the Rod cGMP Phosphodiesterase Binds the Catalytic Subunits in an Extended Linear Structure*

The unique feature of rod photoreceptor cGMP phosphodiesterase (PDE6) is the presence of inhibitory subunits (Pγ), which interact with the catalytic heterodimer (Pαβ) to regulate its activity. This uniqueness results in an extremely high sensitivity and sophisticated modulations of rod visual signaling where the Pγ/Pαβ interactions play a critical role. The quaternary organization of the αβγγ heterotetramer is poorly understood and contradictory patterns of interaction have been previously suggested. Here we provide evidence that supports a specific interaction, by systematically and differentially analyzing the Pγ-binding regions on Pα and Pβ through photolabel transfer from various Pγ positions throughout the entire molecule. The Pγ N-terminal Val16–Phe30 region was found to interact with the Pαβ GAFa domain, whereas its C terminus (Phe73–Ile87) interacted with the Pαβ catalytic domain. The interactions of Pγ with these two domains were bridged by its central Ser40–Phe50 region through interactions with GAFb and the linker between GAFb and the catalytic domain, indicating a linear and extended interaction between Pγ and Pαβ. Furthermore, a photocross-linked product αβγ(γ) was specifically generated by the double derivatized Pγ, in which one photoprobe was located in the polycationic region and the other in the C terminus. Taken together the evidence supports the conclusion that each Pγ molecule binds Pαβ in an extended linear interaction and may even interact with both Pα and Pβ simultaneously.

Phototransduction in vertebrate rod photoreceptor cells features an extremely high sensitivity of "one photon" response and a quick recovery facilitated by complex regulatory mechanisms (1,2), and has thereby evoked enormous interest from researchers. In accordance with these features, rod phosphodiesterase (PDE6), 2 a nearly perfect effector enzyme in amplifying visual signal (1), is unique among the phosphodi-esterase families not only for its heterodimer (P␣␤) formed by the catalytic subunits, but also for its two identical inhibitory subunits (P␥) that strictly regulate the PDE6 activity (2)(3)(4). In the dark state P␥ binds P␣␤ tightly to keep the holoenzyme inactivated. Upon light activation P␥ is displaced from the catalytic sites of P␣␤ by the activated transducin ␣ subunit to relieve the cGMP hydrolysis activity of P␣␤. The lowered cGMP level leads to transfer of the visual signal to the brain and thus results in vision. Meanwhile, the displaced P␥ is an important component of the GTPase accelerating protein complex, which effectively terminates the visual signal (1). The Ca 2ϩ feedback loop that restores cGMP levels and the reciprocal action of cGMP/P␥ have been suggested to result in rebinding of P␥ to P␣␤ (5-7), thus preparing the phosphodiesterase for the next round of photoexcitation. Whereas the exceptionally tight binding of P␥ with P␣␤ (K d Ͻ 1 pM) maintains a very low visual background in rod cells, the efficient dissociation and re-association of P␥/P␣␤ is critical for a quick photoresponse and recovery (1,2,6). Interplay of these regulatory processes contributes to the superior sensitivity of rod phototransduction (1). Therefore, association and dissociation of P␥ with the P␣␤ heterodimer play a critical role in visual signal transduction, and the manner by which P␥ interacts with P␣␤ is a critically important issue for elucidating the visual signaling mechanisms. Furthermore, the disruption of P␥/P␣␤ interaction leads to night blindness and retinal degeneration (8 -12).
However, the structural aspects of P␥/P␣␤ interaction remain elusive so far due to limited structural information. Electron microscopy studies on bovine rod PDE6 have revealed the dimeric association of the two catalytic subunits and three distinct domains in each subunit, the tandem GAF domains (13) at the N-terminal side and the catalytic domain at the C terminus (14,15), yet how P␥ binds to P␣␤ remains unresolved. The recent crystal structure of GAF domains of a related PDE family member, mouse PDE2, revealed the atomic details of the GAF homodimer (16). The crystal structures of single catalytic domains of PDE4, PDE5, and PDE9 have also been solved (17)(18)(19), but no atomic structure is available for any intact PDE subunit that includes the tandem GAF domains and the catalytic domain. Whereas these electron microscopy derived and crystal structures provide useful information for producing a speculative structure for PDE6, none of these studies has revealed a quaternary organization of the two inhibitory ␥ subunits with the catalytic P␣␤ heterodimer of rod PDE6.
Previous studies through molecular biology and biochemistry have made remarkable contributions to understanding the P␣␤/P␥ interac-tions (2,4). Two major P␣␤-interacting regions on P␥ have been identified: the polycationic region that provides the major binding energy with P␣␤, and the C-terminal region that imposes inhibition on the catalytic site of P␣␤ (2, 4, 6, 20 -25). In stark contrast, however, only limited knowledge has been gained about the P␥-interacting sites on P␣␤ because this analysis is particularly difficult. Heterologous expression and mutagenesis of the ␣␤ catalytic heterodimer with high activity has not been successful thus far (26 -28), probably due to complex regulation of expression and assembly of the ␣␤ subunits (12,29,30). For this reason, PDE6 activity assays using native P␣␤ have been a popular approach for the study of P␥/P␣␤ interactions (4,6,10,24,25,31,32). Using functional assay approaches, however, it has been difficult to gain information about P␥-interacting regions on P␣␤, particularly differential interactions of P␥/P␣ and P␥/P␤, because the activity of P␣␤ is lost once P␣ and P␤ are either separated or dissected. We therefore employed an advantageous photolabel transfer approach in this study to map P␥-interacting regions on P␣ and P␤ thus providing insights into the quaternary organization of the PDE6 subunits.
A model of linear P␥/P␣␤ interaction has been previously suggested (4,33), which is supported by the studies that identified a site in the P␣ GAFa domain interacting with N-terminal position 23 on P␥ (34), and a binding region of the P␥ C terminus in the P␣ catalytic domain (35,36). However, this linear binding pattern was contended by other models, including a model in which one P␥ molecule binds to the P␣␤ GAF domains, whereas the second P␥ molecule binds to the catalytic domain(s) (31). Another layer of complexity of the P␥/P␣␤ interaction is added by the heterogeneity of P␣ and P␤ in the amino acid sequences that implies a structural difference between these two catalytic subunits (35). With little evidence available to differentiate the interactions of P␥/P␣ and P␥/P␤, it is not clear whether each of the two P␥ subunits binds only one of the P␣ and P␤ subunits or both of them simultaneously, although the conventional models suggest that each P␥ binds only one of P␣ and P␤ (1,7). A glimpse of differential interactions of P␥ with P␣ and P␤ was obtained from our recent work that indicated a preferential binding of the Phe 30 region to P␣ and the Ser 40 region to P␤, which led to a suggestion that each of the two identical P␥ subunits may bind P␣ and P␤ simultaneously (37). To elucidate a P␥/P␣␤ interaction pattern, we have carried out photolabel transfer to P␣␤ in a systematic manner from 10 positions evenly spread throughout the entire P␥ molecule. A successful P␣␤ domain dissection by thorough proteolytic analysis allowed us to identify the P␥-binding regions on P␣ and P␤ with a given P␥ position, and the use of P␥ constructs with dual photoprobes enabled us to observe a dually cross-linked holo-PDE6 complex. The combined data indicate an interaction pattern that each P␥ binds P␣ and P␤ linearly and, quite possibly, simultaneously.
Limited Proteolysis of nP␣␤ and Identification of Proteolytic Fragments-nP␣␤ was incubated with each protease under the conditions listed in Table 1. To terminate proteolysis, 0.5 l of protease inhibitor mixture (Sigma, reconstituted in 5 ml of H 2 O) was added, the reaction was then boiled for 5 min in the dark in sample buffer including 50 mM DTT and 1% SDS. The proteolytic fragments were separated by SDS-PAGE, and transferred in a tank system to an Immobilon-P PVDF membrane (0.45 micron, Millipore) at 75 volts for 1.5 h in transfer buffer containing 10 mM CAPS (pH 11) and 10% methanol. The blots were used either for Western blotting detection of the original N termini with anti-P␣ antibody, or microsequencing to identify fragments containing a new N terminus generated by proteolysis. For microsequencing, the blots were stained in Amido Black for 30 s, destained in methanol, rinsed intensely with double distilled water, and then air-dried. The desired bands were excised for microsequencing at the Macromolecular Structure Facility of Michigan State University.
Preparation of P␥ Photoprobes-The preparation of the construct for expressing the full-length wild type P␥ (single cysteine at position 68) in fusion with intein was described in Ref. 39. Using the QuikChange method (Stratagene), single cysteine mutants were prepared (37), which were then used as templates for generating dual cysteine mutants. The P␥ constructs were expressed in Escherichia coli and purified as previously described (40). ϳ99% pure P␥ was used for preparation of various P␥ photoprobes.
For preparation of mBP-P␥ photoprobes, 200 g of P␥ was incubated with 50 g of mBP under argon for 30 min (22°C, dark) in the buffer containing 40 mM NaH 2 PO 4 (pH 6.7), 10 mM EDTA, and 40% acetonitrile, with a total reaction volume of 250 l. For scavenging excess unreacted mBP, 200 mM DTT and 100 mM Tris-HCl (pH 8) were added at the end of the reaction. The reaction was then loaded to a self-packed POROS column for HPLC purification (37) of mBP-derivatized P␥, which was eluted at ϳ37% acetonitrile by a gradient of 0.7% per min in the presence of 0.1% trifluoroacetic acid.
Photocross-linking of P␥ Photoprobes with nP␣␤ or P␣␤-All the photocross-linking reactions were performed in Buffer UB (10 mM HEPES, pH 7.5, 120 mM NaCl, 5 mM MgCl 2 ) contained in 1.5-ml ultraclear polypropylene microcentrifuge tubes (Axygen). The reaction mixtures of nP␣␤ (or P␣␤) and P␥ photoprobes with the desired concentrations were dark incubated on ice for 15 min, and then subjected to UV light.
The reactions with FAzB-P␥ photoprobes sitting in ice water were exposed to the light generated by an AH-6 water-jacketed 1000-watt high pressure mercury lamp for 5 s at a distance of 10 cm. The UV light Ͻ300 nm was effectively filtered by a Pyrex glass water jacket of the lamp and the polypropylene microcentrifuge tubes. The reactions with a benzophenone photoprobe (BBM or mBP) were photolyzed at 5-10°C two times for 15 min with a 5-min dark interval on ice, in a RPR-100 Rayonet Photochemical Reactor equipped with 18 light bulbs of 350 nm (Southern New England Ultraviolet Co.). The sample buffer containing SDS and DTT was added to the reactions immediately after photolysis or subsequent proteolysis, the proteins (or proteolytic fragments) were then separated by SDS-PAGE.
Other  has the same domain arrangement except that the corresponding amino acid numbers are smaller by 2. To recognize the proteolytic fragments, the original N terminus was detected by Western blotting using anti-P␣ antibody (see B), whereas a new N terminus generated by proteolysis was identified by its N-terminal amino acid sequence (shown in parentheses, determined by microsequencing). Each proteolytic band is named according to its number on the gel in B (e.g. Clos-3). The proteolytic cleavage position at the C terminus of each fragment represents the most likely cleavage site, which was deduced based on the size of the proteolytic band and the properties of the corresponding protease (Table SI). *, the C termini are postulated according to Catty and Deterre (46). The molecular mass of P␣ is therefore calculated to be 97.9 kDa using PeptideMass; P␤, 96.3 kDa; tP␤, 79.9 kDa. a, the C terminus may be at 257. b, each band may contain fragments from both P␣ and P␤ because the sequences of P␣ and P␤ at either the N-or C-terminal cleavage sites are identical. c, one band contains two fragments (as revealed by two sequences) that migrated at the same position. B, Western blotting analysis of proteolytic bands of P␣ and P␤ generated by limited proteolysis. Western blotting was performed using anti-P␣. For each blot, the Amido Blackstained polyvinylidene difluoride membranes of the same blot are aligned on the right according to the exact positions of the corresponding bands. The molecular sizes of the M r standard (MW std) bands are shown on the right of AspN, and for other blots, only the molecular size of the lowest band is marked. In the blot of Thrombin, P␣␤ instead of nP␣␤ was used for lanes 3 and 4, in which band 2 should contain both thrombin itself and a P␣ (and P␤) fragment with an original N terminus, because this band is Western positive but thrombin is Western negative (Lane 1); no Western blotting was performed for lane 5. There must be only one cleavage site on each subunit of P␣ and P␤ because the sum of Throm-1 and Throm-2 is the size of P␣ or P␤. d, bovine carbonic anhydrase (29 kDa) band of the molecular mass standard, and e, Genenase I (51, 52) showed nonspecific Western signal.
mobile phase A was 0.4% formic acid/water and mobile phase B was 0.2% formic acid/acetonitrile. A standard gradient from 5 to 95% B was used.
Western blotting, far Western blotting, and protein quantitation were performed as previously described (37). Dilution of primary antibodies was: anti-P␣, 4,000-fold; anti-P␤, 1,000-fold; and anti-P␥, 10,000-fold. The secondary antibody, horseradish peroxidase-conjugated goat anti-rabbit IgG was diluted by 500,000-fold. The biotin-label transfer was detected by far Western blotting using streptavidin-conjugated horseradish peroxidase (Kirkegaard and Perry Laboratories) (100,000-fold dilution). 8% Laemmli (41) polyacrylamide gels were prepared with a supplement of 20% glycerol. Native PAGE was performed at 5-10°C with the buffers free of SDS. 16.5% gels were made by a modified Tricine gel protocol (42). The precast 7% NuPAGE gels were run according to the manufacturer's instructions (Invitrogen).

RESULTS
Domain Dissection of PDE6 ␣ and ␤ Subunits-Domain dissection of P␣␤ has not been previously reported, although the GAF domains and the catalytic domains of the homodimer of PDE2 were successfully separated by limited proteolysis combined with cyanogen bromide cleavage (43). Part of the difficulty for dissecting P␣␤ likely arises from the complexity of the heterodimer. In this study we have obtained separated domains of P␣ and P␤ through systematic proteolysis ( Fig. 1). It was determined that the key for successful conditions was use of SDS in the proteolysis reactions (Table 1). Low SDS (e.g. 0.1%) enhanced cleavage likely by improving access of the protease to substrate without denaturing the protease. To gain differentiated information from P␣ and P␤, nP␣␤ was used for these experiments. Although P␣ and P␤ in the P␣␤ sample are too similar in size to be separated by SDS-PAGE, P␤ of nP␣␤ is shortened by 16.4 kDa on a SDS gel due to the nicking (tP␤), and thereby produced proteolytic fragments of different lengths than P␣ (Fig. 1A). Taking advantage of this, we obtained different proteolytic cleavage patterns of P␤ than P␣ and thus were able to differentiate the regions on P␣ and P␤ that were photolabeled from a given position of P␥.
As shown in Fig. 1, five proteases were selected for domain dissection of P␣ and P␤ because they produced relatively simple and robust cleavage patterns that rendered identification of the proteolytic fragments practical. The proteolytic fragments of P␣ containing the original N terminus were recognized by Western blot using an antibody against the first 16 amino acids of P␣. The fragments with a proteolysis-generated new N terminus were unambiguously identified through microsequencing. The C terminus of each proteolytic fragment was determined from the most likely cleavage site based on the size of the proteolytic band (Fig. 1B) and the favored and unfavorable cleavage sequences for the protease (44) (supplemental Table SI). The maps of dissected P␣ and tP␤ have thus been obtained. As indicated in Fig. 1A, The GAF region (GAFa plus GAFb) and the catalytic domain of P␣ were readily separated by cleavage with either Genenase I or V8 protease, because these two enzymes both cleaved in the middle of the linker region between GAFb and the catalytic domain in P␣. Although a separate GAFb domain was not obtained, a proteolytic fragment, Clos-3, which was generated by clostripain cleavage at 257 (or 267), contained the entire GAFa but only a very little portion of GAFb. The other proteolytic fragments with a variety of lengths represented various regions of either P␣ or P␤. As a result, combining the information from the various identified proteolytic fragments narrowed down the P␥-binding regions. Whereas most of the identified proteolytic bands each represents a single fragment with a sequence from only P␣ or P␤, Clos-4 and Throm-3 may each contain two fragments of equal length from both P␣ and P␤ because the sequences of P␣ and P␤ at the same cleavage site are identical. A special case is band Gen-5, in which two distinct sequences from P␣ and P␤, respectively, were detected by microsequencing. These two fragments were differentiated by Asp-N, which produced two separate bands (AspN-2 and AspN-3) that are the proteolytic fragments corresponding to the two fragments in Gen-5, respectively. Nevertheless, these maps provide information not only for dissection of P␣␤ GAF and catalytic domains but also for differentiation of proteolytic fragments from P␣ and P␤.
Mapping of P␥-binding Regions on P␣ and P␤ by Photolabel Transfer-The domain dissection maps presented in Fig. 1A allowed us to systematically investigate the interaction interface between P␥ and P␣␤ through photocross-linking and label transfer. A cleavable photoprobe BBM was used for this purpose, because it contains an efficient benzophenone photophore and a biotin label that can be sensitively detected by far Western blotting, and the disulfide formed between BBM and a single cysteine on P␥ can be cleaved by DTT after photocross-linking thus transferring the biotin label from P␥ to P␣␤ (37). The preparation of BBM-P␥ photoprobes was verified by ESI-MS, for example, the measured mass of BBM40 (10,410) matched well with the prediction (10,406). To rule out any possible chemical selectivity of the benzophenone group toward certain amino acids (45) on P␣␤, the photoprobe FAzB that differs from BBM only by the tetrafluorophenylazide photophore was also used. Both of these photoprobes were shown to photolabel P␣␤ specifically (Fig. 2). BBM-derivatized P␥-photolabeled P␣␤ in UV light but not in the dark. The photolabeling did not occur on bovine serum albumin, which was included in the photocross-linking reaction as an internal control. BBM did not photolabel holo-PDE6 either, which had P␥ prebound ( Fig. 2A), indicating the specificity of the photolabeling. Accordingly, FAzB did not photolabel the P␣␤ dark control, bovine serum albumin, or holo-PDE6 (Fig. 2B), indicating that photolabeling of P␣ and P␤ reflected their true interactions with P␥. The photolabeled regions were therefore elucidated by far Western following photocross-  (35,46), and no difference with P␣␤ when tested for inhibition of their catalytic activity by P␥ (35). This is probably because the two P␤ fragments generated by the nicking are held together by intramolecular interactions, as is also evidenced by the fact that nP␣␤ and P␣␤ migrated at the same position on a native gel either in a P␥-free or P␥-bound form (Fig. 2C), indicating stable complexes. In addition, a preferential photolabeling on P␤ from P␥ position 40 occurred with both nP␣␤ (Fig. 3, to be discussed below) and P␣␤ (37), which was nicked after photolabeling. A low molar ratio of P␥/P␣␤ (0.3 P␥ per dimer) was used to avoid any possible nonspecific photolabeling, and to differentiate photolabeling on P␣ and P␤ because the polycationic region of P␥ binds P␣ and P␤ differentially at low P␥/P␣␤ ratios (37). Through a full spectrum analysis of P␥-binding regions on P␣ and P␤ mapped by Genenase I and photolabel transfer from various P␥ positions (Fig. 3A), it is evident that the P␥ N terminus including positions 16, 21, and 30 photolabeled the P␣ N-terminal region that contained GAFa and part of GAFb (1-370/P␣). In contrast, the C terminus of P␥ including positions 60, 68, 73, 76, and 87 interacted with the C-terminal region of P␣ (492-816/P␣) that contained the catalytic domain. This labeling pattern was confirmed using V8 protease (Fig. 3B) Fig. 1A) also supported this pattern (data not shown). A consistent photolabeling result was also observed with the dissection map of Asp-N (Fig. 3C), in which the C-terminal positions 73, 76, and 87 photolabeled a P␣ fragment 499-761/P␣ that corresponds to the P␣ catalytic domain, whereas position 40 photolabeled the P␤ fragments 436-667/P␤ and 153-335/P␤ but not the catalytic domain. This pattern of P␣␤ photolabeling was further confirmed by photolabeling of a lower band, Gen-5 (Fig. 3A), in which the major component is the fragment of the P␣ catalytic domain (492-691/P␣) and the minor is a fragment from the P␤ GAF region (148-353/P␤), as evidenced by the microsequencing signal (data not shown). This band was heavily photolabeled by the C-terminal half of P␥, and the labeling intensity ratio of Gen-5 versus Gen-4 is roughly the same as the ratio of the protein amounts in these two bands (ϳ3:1), indicating that it was fragment 492-691/P␣ rather than 148-353/P␤ that was photolabeled by the P␥ C-terminal positions. Therefore, modest labeling of the band Gen-5 by the N-terminal positions must be on fragment 148-353/P␤ rather than on 492-691/P␣. This is consistent with the fact that positions 16, 21, and 30 labeled the fragment 1-370/P␣, which is roughly an equivalent of 148-353/P␤ in Gen-5, but did not label the fragment 492-816/P␣, which is an equivalent of 492-691/P␣ in Gen-5. In good agreement with this observation, the fragment 148-350/P␤ in V8-5 was strongly labeled by the P␥ N-terminal positions (Fig. 3B). The modest labeling of V8-5 by the C-terminal positions was likely due to contamination of a fragment similar to 492-691/ P␣. This speculation is supported by the fact that Genenase I and V8 protease generated similar proteolytic fragments (Fig. 1A).
The central P␥ positions 40 and 50 preferentially photolabeled the P␤ fragment 369-789/P␤ in Gen-2 (Fig. 3A). In particular, position 40 labeled the P␤ bands such as V8-5 more strongly than the P␣ bands (such as V8-3). This preferential photolabeling of P␤ by positions 40 and 50 agrees with our previous observations (37).
For differentiation of the labeling from P␥ positions 30 and 40, the dissection map generated by clostripain was used. In comparison to position  A and B, respectively. After photolysis the reactions were incubated with the sample buffer containing SDS and DTT at room temperature for 1 h and subjected to 16.5% SDS-PAGE to separate the proteins, and far Western blotting was then performed as described under "Experimental Procedures" to detect photolabeled proteins. Bovine serum albumin (BSA) was used as an internal control in lane 2. Holo-PDE6 was used (lane 5) to show protection of photolabeling by P␥ that was prebound. C, nP␣␤ and P␣␤ (1.5 g each) showed similar migration on an 8% native gel, either in the absence of (lanes 1 and 3) or in the complex with (lanes 2 and 4) P␥, which was at a 2:1 molar ratio versus the heterodimers. The cathode buffer was pH 8.0 and the anode buffer pH 8.5. MW std, M r standard; BSA std, bovine serum albumin standard; PVDF, polyvinylidene difluoride. 30, which photolabeled P␣ fragments 1-274/P␣ and 1-306/P␣, position 40 photolabeled 1-306/P␣ but not 1-274/P␣ (Fig. 3D).
In Fig. 3E, a similar photolabeling pattern was observed by using BBM-and FazB-derivatized P␥ photoprobes. This result is consistent with the photolabeling pattern in Fig. 3B, although in this experiment V8 digestion was somewhat less vigorous so that tP␤ was incompletely digested. These data further confirm that the observed photolabeling of P␣ and P␤ was not a result of chemical selectivity of the benzopheone group. In addition, it is noteworthy that after proteolysis (using V8 or Genenase I) P␣ disappeared while the tP␤ band was still remaining (Figs. 1B and 3E). The different digestion rates of P␣ and P␤ suggest a structural heterogeneity of the two catalytic subunits in a heterodimer.
Dual Cross-linking of P␣␤ with Dually Derivatized mBP-P␥ Photoprobes-To investigate whether one P␥ could bind both P␣ and P␤ simultaneously, dual photocross-linking of P␣␤ was performed. For this purpose, P␥ constructs with double cysteines were dually derivatized with mBP, a benzophenone photoprobe that reacts with cysteine to form an uncleavable covalent bond (35) (Fig. 4A). The dually derivatized P␥ constructs were efficiently purified by HPLC (Fig. 4, B and C), and confirmed to have the exact masses by ESI-MS ( Fig. 4D; supplemental Table SII). Importantly, the mBP derivatization had only a modest effect on the P␥ function of P␣␤ inhibition (supplemental Table SII).
As discernable on the Coomassie-stained SDS gels, a dual cross-link band was substantially generated by the dually derivatized P␥ constructs 30/40 mBP, 30/73 mBP, and 30/76 mBP (Fig. 5). With HiMark as a A fragment refers to a single peptide generated by proteolysis. A band refers to an Amido Black-stained band on a PVDF membrane, which may contain two different proteolytic fragments that co-migrate at the same position. A similar photolabeling pattern of P␣␤ obtained by using FAzB and BBM is shown in E. The V8 digestion was less vigorous than that in B so that the tP␤ band was still remaining. The triangle on the right marks the position of uncleaved P␣. Lane 5 is a longer exposure of the condition in lane 4. *, lanes 50 and 73 were from a separate proteolysis that resulted in a weaker Amido Black-stained band of Gen-5 than the other lanes. **, unidentified band. Arrow, position 40 did not photolabel V8-4 but instead labeled a lower band that was not identified. standard on a precast 7% NuPAGE gel, which provides high resolution for large proteins, the dual cross-link band migrated at ϳ250 kDa (Fig.  5C). There was no additional band observed above the dual cross-link or at the top of the gel (Fig. 5, B and C), indicating that no oligomers of the cross-link complex were formed during photolysis. Therefore, the molecular size of the dual cross-linked protein complex was 15-20% greater than that predicted for the ␣␤␥␥ holoenzyme (ϳ214 kDa, based on the molecular mass calculations in the legend to Fig. 1: P␣ ϭ 97.9 kDa; P␤ ϭ 96.3 kDa; mBP-P␥ ϭ 10 kDa). The apparent higher migration position (ϳ250 kDa) of the dual cross-link band by SDS-PAGE analysis (Fig. 5C, lanes 1 and 3) was likely due to the covalent linkage of the cross-linked complex, which resists complete denaturation in SDS.
This dual cross-link was specifically generated by the dually derivatized P␥ photoprobes, because it did not occur when using the single derivatized P␥ constructs (Fig. 5, A and D). The fact that no cross-link was observed from the photolysis reaction using the underivatized dual-cysteine construct 30C/73C indicates that the dual cross-link was not a result of nonspecific UV-caused photooxidation (Fig. 5, A and D, lane  1). The low concentration reducing agent (DTT) included in the crosslinking reactions likely prevented nonspecific photooxidation. In addition, the dual cross-link was unlikely due to free mBP in the purified mBP derivatives, because the mBP-P␥ derivatives were readily separated by HPLC, and the MS data confirmed the exact mass of the single species of mBP-P␥ (Fig. 4).
Furthermore, the dual cross-link appeared to contain all the PDE6 subunits, as revealed by the Western blotting in Fig. 5D. The positive Western signal observed using anti-P␥ was unlikely from re-probing of the anti-P␣ or anti-P␤ remaining on the same blot, because a 30-min treatment of the blot with stripping buffer completely abolished the Western signal when re-probed with the secondary antibody (data not shown).
Consistent with the evidence from the Western blot, shortened P␤ (tP␤) in the nP␣␤ sample resulted in a smaller dual cross-link (Fig. 5C,  lane 5-7), further confirming the presence of P␤ in the dual cross-link. The photoprobe 30/40 mBP did not generate as strong a dual cross-link as did 30/73 mBP, possibly due to the fact that 30/40 mBP had a weaker interaction with P␣␤, as evidenced by its higher K i than of 30/73 mBP (Table SII). Or, the proximity of positions 30 and 40 that restricts flexibility of this part of P␥ likely suppressed the dual cross-linking.

DISCUSSION
Because of lack of information from atomic structure, it has been difficult to decipher the quaternary organization of the two identical P␥ molecules and the P␣␤ heterodimer. In this study, we have obtained evidence supporting the conclusion that interactions of the two P␥ molecules with P␣ and P␤ occur in a linear and extended conformation.
P␥ Binds P␣ and P␤ in an Extended Linear Structure-The systematic mapping of P␥-binding domains on P␣ and P␤ (Fig. 3) clearly indicates that the P␥ N-terminal positions interact with the P␣␤ GAF region, whereas the C-terminal positions interact with the P␣␤ catalytic domains, as summarized in Table 2. These observations support a headto-head/tail-to-tail extended linear interaction of P␥ with the P␣␤ heterodimer.
Furthermore, the P␥ central positions 40 and 50 appeared to be a connecting part on P␥ between the GAF-interacting region and the catalytic domain-interacting region of P␥. As shown in Fig. 3B Table 2). These results indicate that position 50 in the middle of P␥"bridges" the GAF-interacting N-terminal region and the catalytic domain-interacting C terminus. Position 40 appeared to be the N-terminal side of this "bridge," because it labeled GAF fragments 123-486/ P␣, 148-350/P␤, and 153-335/P␤ but not catalytic domain fragment 487-794/P␣ as did position 50 (Fig. 3, B, C, and E), instead it labeled a lower band that was not identified (see the double asterisk mark). Together with the labeling of AspN-2 (436-667/P␤) by position 40 (Fig.  3C), these data indicate that the 438 -486 region on P␣ (or 436 -484 on P␤) is a binding site for position 40 (Fig. 6). This binding site located within the linker between GAFb and the catalytic domain overlaps with the P␣-(461-553) region that was previously identified to interact with a P␥ peptide of the 24 -45 region as detected with a fluorescent label at its C-terminal end (33). To narrow down the binding region of position 40 in the GAF domains, photolabeling from positions 30 and 40 was compared using the clostripain map (Fig. 3D). It is obvious that frag-   (Fig. 6). Position 40 labeled more than one site probably because the arm length of the photoprobe conferred flexibility and these two sites may be close in a tertiary structure. Alternatively, binding of P␥ to P␣␤ may trigger a conformational change that tilts the P␣␤ catalytic domain toward GAFa/b thus bringing the two P␥-interacting sites closer.
Because P␣ and P␤ were equally photolabeled when the P␥ photoprobes were in molar excess (37), suggesting that the corresponding regions on P␣ and P␤ were labeled from the same position of the two identical P␥ molecules, it is thereby reasonable to extrapolate a P␥-binding region on one catalytic subunit from the labeling on the other. As discussed earlier, P␥ C-terminal positions 60 -87 photolabeled the P␣ catalytic domain fragments, such as 487-794/P␣ and 492-691/P␣ (Fig. 3,  A and B). In addition, detectable labeling of 369 -789/P␤ was produced from the probe at positions 60 and 68 on P␥ (see Fig. 3A and Table 2), indicating an interaction of the P␥ C terminus with P␤. Furthermore, P␥ C-terminal positions 73, 76, and 87 strongly labeled catalytic domain 499-761/P␣ but not 436-667/P␤, which is an equivalent to the 438-669/P␣ region (Fig. 3C). We thus have narrowed down the interaction site of the P␥ C terminus to the 669 -761 region on P␣ and 667-759 on P␤ (Fig. 6), which includes the P␥ binding pocket identified previously (35,36). Accordingly, because the P␥ N terminus (positions 16, 21, and 30) labeled 123-486/P␣ (Fig. 3B), but position 40 did not label the P␣ N-terminal region beyond residue 274 (Fig. 3D), it is most likely that the P␥ 16 -30 region on the N-terminal side of position 40 interacted with the GAFa 123-274 region on P␣, and the corresponding 121-272 region on P␤ if P␥ were in molar excess (Fig. 6). In accordance to this conclusion, a photoinsertion from P␥ position 23 identified at Met 138 -Gly 139 on P␣ lies within this region (34), and the other potential P␥ contact residues (47) are also included in this region. The full spectrum photolabel transfer from 10 P␥ positions throughout the entire molecule to various regions on P␣ and P␤ is thereby summarized in Fig. 6. It is evident that following the change of photoprobe positions from the N terminus throughout the C terminus of P␥, the photolabeled regions on P␣␤ shift systematically from the GAFa to the GAFb to the catalytic domains. This revealed a linear extended binding pattern of P␥ with P␣␤, in which the P␥ N-terminal 16 -30 region binds P␣␤ GAFa and the P␥ C terminus binds the catalytic domain while the middle 40 -50 region bridges these two regions by binding to GAFb and its C-terminal vicinity. In support of this idea, an overall linear structural arrangement of P␣ and P␤ domains has been revealed by the electron microscopy study (14) (Fig. 7A). These data strongly support a linear P␥/P␣␤ interaction pattern as modeled in Fig. 7B. Aside from the evidence mentioned above, this extended, linear binding pattern is also supported by a previous study using fluorescence resonance energy transfer, which showed that P␥ was elongated when bound to P␣␤ (48). P␥ May Bind P␣ and P␤ Simultaneously-In the case of the extended linear binding of P␥ to P␣␤, there may be two possible patterns: each P␥ interacts with either P␣ or P␤ separately, or each P␥ binds both P␣ and P␤ simultaneously (Fig. 7B). We sought to distinguish these two possibilities by photocross-linking using a P␥ construct with dual photoprobes at paired positions. The two P␣ and P␤ subunits should thus be "stapled" together if the latter is true. Indeed, a cross-linked complex with a molecular size consistent with the ␣␤␥(␥) product was specifically generated by the dually mBP-deri-vatized P␥ photoprobes and appeared to include ␣, ␤, and ␥ subunits (Fig. 5), as discussed earlier in this paper. One might argue that mixed species of ␣␣␥(␥) and ␤␤␥(␥) could co-migrate at the same position and thus result in the Western signals observed in Fig. 5D. However, crosslinking of 30/73 mBP with the nicked heterodimer (nP␣␤) led to a dual cross-link band at a position lower than that with P␣␤, which was consistent with the size of a cross-link complex ␣t␤␥(␥) calculated to be 17 kDa smaller than ␣␤␥(␥) (Fig. 5C). This evidence supports the conclusion that the dual cross-link complex was generated from the ␣␤ heterodimer rather than the putative ␣␣ and ␤␤ homodimers. Furthermore, if ␣␣␥(␥) and ␤␤␥(␥) were the major dual cross-link populations, in the dual cross-link of nP␣␤ one would expect to see a band of ␣␣␥(␥) at the ␣␤␥(␥) position (dashed line) and a band of t␤t␤␥(␥) at a lower position (ϳ170 kDa, dashed box) (Fig. 5C, lane 5). In fact, neither of these bands was generated by cross-linking with nP␣␤. Consistently, previous studies indicate that the P␣␤ heterodimer is the major population of PDE6 under physiological conditions even though minor populations of ␣␣ and ␤␤ homodimers may not be excluded (26,49).
Importantly, the functional relevance of the dual cross-link is reflected by its dependence on the mBP derivatization positions on P␥. It is obvious that substantial dual cross-links were generated only by the dual benzophenone probes at certain pairs of positions, such as 30/73, 30/40, and 30/76 (Fig. 5). This is in good agreement with previous observations that the P␥ polycationic and C-terminal regions are the major P␣␤-interacting domains (2,4,6,37). In particular, mutations of P␥ residues Lys 29 , Lys 31 , Arg 33 (23), Asn 74 , His 75 , and Leu 78 (25) disrupt P␥/P␣␤ interactions. In our study, however, the double cysteines for mBP derivatization were placed nearby but not at these residues and replaced uncharged hydrophobic residues (such as Phe 30 , Phe 73 , and Leu 76 ), so that mBP can still reach and efficiently photoinsert into P␣␤ by mimicking the wild type residues. The P␥ position dependence of the dual cross-link was further confirmed by the fact that the dual probes at  Fig. 3 and Table 2. The domain arrangement of P␣ is schemed as in Fig. 1A. P␤ has the same domain arrangement except that its amino acid numbers are smaller than those of P␣ by 2. P␥ is represented as a rod, with the photoprobe positions listed on top of it. The defined P␥/P␣␤ interaction interface is indicated by the shaded areas. The boundary of the labeled region on P␣␤ by position 50 was not clearly defined (dashed triangle).  (14) and the crystal structure of PDE2 GAF domains (16) were used as templates to propose the model in B. Each P␥ binds both P␣ and P␤ simultaneously in an extended linear structure. The P␥ shown in the front represents high affinity binding with P␣␤, with the P␥-(21-30) region binding P␣ GAFa, and with the 40 -50 region binding P␤ GAFb. The other P␥ is shown on the opposite side, representing a low affinity binding. The C-terminal region of P␥ is shown to bind the P␣␤ catalytic domain near the groove, and the dashed rod (the same for the P␥ on the reverse side) depicts a general possibility that some component of P␥/P␣␤ interaction may occur between one P␥ and one catalytic subunit. P␣, blue; P␤, red; P␥, gray. GAFa, GAFb, catalytic domain, and P␥ are represented by the egg, ball, column, and rod, respectively.

TABLE 2
Summary of a full spectrum mapping of the P␥-interacting regions on P␣ and P␤ a pair of positions at least one of which is not expected to have strong interaction with P␣␤, such as 10/73 and 3/87, did not cause noticeable dual cross-link (Fig. 5D). This was not an affinity issue because 10/73 mBP and 3/87 mBP actually had lower K i than 30/73 mBP and 30/76 mBP (Table SII).
It is noteworthy that 40/73 mBP and 50/73 mBP also caused dual cross-link, although these bands appeared to be doublets with the upper band migrating at the same position as 30/73 mBP and 30/76 mBP (Figs.  5, A, lane 4, B, lanes 3 and 6, and D, lane 4), which was further confirmed by the co-migration experiment (Fig. 5B, lanes 2 and 5). The upper and lower bands of the doublet could actually represent ␣␤␥␥ and ␣␤␥, respectively, which differ in size by one ␥ subunit, as is also consistent with the data that the Western signal of the upper band is more intense than the lower (Fig. 5D, lane 4, Anti-P␥). Such cross-linking may be explained by the model in Fig. 7B. If binding of the P␥ C terminus occurs near the "groove" between the two catalytic domains of P␣ and P␤ (Fig.  7A), photoinsertion could occur to both of the catalytic subunits. In addition, as depicted by the dashed rod in Fig. 7B, it is possible that some component of the interactions between the P␥ molecule with the dual photoprobes and P␣␤ involves covalent tethering of P␥ to the same catalytic subunit (P␣ or P␤). The possibility that the same photoprobe in the P␥ C terminus can react with both P␣ and P␤ precludes a definitive conclusion that P␥ binds both catalytic subunits simultaneously. Nonetheless, the dual cross-links of P␣ and P␤ are consistent with the notion of simultaneous binding supported by the previous study (37).
Taken together, evidence presented in this study demonstrates that in holo-PDE6 each P␥ binds P␣ and P␤ in an extended linear structure, most likely as illustrated in Fig. 7B. In this model, linearly extended binding of P␥ with P␣␤ is supported by the full spectrum mapping of P␥/P␣␤ interaction through label transfer ( Fig. 6; Table 2) as discussed above, and also by the evidence from previous studies (33)(34)(35)(36)48). Evidence consistent with the simultaneous binding of P␥ with P␣ and P␤ is the observed dual cross-linking of the ␣␤␥(␥) complex (Figs. 5), and the preferential photolabeling of P␣ GAFa from P␥ positions 21 and 30 and P␤ GAFb from position 40 (and 50) (Fig. 3) (37). A "crossover" GAF region of P␣ and P␤ conforms to the crystal structure of the PDE2 GAF dimer (16). It has been previously suggested that there are two classes of P␥ binding sites and two classes of cGMP binding sites in the P␣␤ GAF domains that positively cooperate with each other (2, 6), but the mechanism of this cooperation remains unknown. The two classes of P␥ binding sites may be actually each composed of interaction regions on both P␣ and P␤, as proposed in Fig. 7B. The high affinity P␥/P␣␤ interaction is proposed to be binding of the P␥-(21-30) region to P␣ GAFa and the 40 -50 region to P␤ GAFb (and/or the linker region between GAFb and the catalytic domain), as is supported by the data indicating preferential photolabeling of P␣ and P␤ from this study (Fig. 3) and our previous study (37). The low affinity interaction is thus proposed to be on the opposite side of the P␣␤ heterodimer, which is composed of the less preferred corresponding P␥ binding regions. This hypothesis is also supported by the fact that heterogeneous binding of P␥ and cGMP to the catalytic subunits has only been observed with the rod P␣␤ heterodimer but not the cone P␣Ј␣Ј homodimer (4).
The findings reported in this paper provide novel insights into understanding the visual transduction mechanisms in rod cells. The type of interaction depicted in Fig. 7B may add extra strength, beyond P␣␤ dimerization, for stabilizing the holo-PDE6 complex, and thus facilitate the extremely tight binding of P␥ to P␣␤ that is critical for maintaining a low visual background in the dark state. Furthermore, the preferential binding on P␣ and P␤ from different regions on P␥ may reasonably account for the previously observed heterogeneous P␥ binding sites.
The binding site with low affinity may be responsible for a rapid P␥ dissociation from P␣␤ upon photoexcitation, while the high affinity site may be important for a quick re-association of P␥ with P␣␤ during recovery and a tight binding in the dark-adapted state.