The GAFa domains of rod cGMP-phosphodiesterase 6 determine the selectivity of the enzyme dimerization.

Retinal rod cGMP phosphodiesterase (PDE6 family) is the effector enzyme in the vertebrate visual transduction cascade. Unlike other known PDEs that form catalytic homodimers, the rod PDE6 catalytic core is a heterodimer composed of alpha and beta subunits. A system for efficient expression of rod PDE6 is not available. Therefore, to elucidate the structural basis for specific dimerization of rod PDE6, we constructed a series of chimeric proteins between PDE6alphabeta and PDE5, which contain the N-terminal GAFa/GAFb domains, or portions thereof, of the rod enzyme. These chimeras were co-expressed in Sf9 cells in various combinations as His-, myc-, or FLAG-tagged proteins. Dimerization of chimeric PDEs was assessed using gel filtration and sucrose gradient centrifugation. The composition of formed dimeric enzymes was analyzed with Western blotting and immunoprecipitation. Consistent with the selectivity of PDE6 dimerization in vivo, efficient heterodimerization was observed between the GAF regions of PDE6alpha and PDE6beta with no significant homodimerization. In addition, PDE6alpha was able to form dimers with the cone PDE6alpha' subunit. Furthermore, our analysis indicated that the PDE6 GAFa domains contain major structural determinants for the affinity and selectivity of dimerization of PDE6 catalytic subunits. The key dimerization selectivity module of PDE6 has been localized to a small segment within the GAFa domains, PDE6alpha-59-74/PDE6beta-57-72. This study provides tools for the generation of the homodimeric alphaalpha and betabeta enzymes that will allow us to address the question of functional significance of the unique heterodimerization of rod PDE6.

Photoreceptor rod and cone cGMP phosphodiesterases (PDE6 1 family) are the effector enzymes in the vertebrate visual transduction cascade. The cascade is initiated by photoexcitation of the visual receptor rhodopsin and leads to hydrolysis of intracellular cGMP by transducin-activated PDE6 (1,2). PDE6 enzymes belong to a large superfamily of phosphodies-terases of cyclic nucleotides that are critical modulators of cellular levels of cAMP and cGMP. Currently, eleven PDE families have been identified in mammalian tissues based on primary sequence, substrate selectivity, and regulation (3,4). Rod PDE6 is composed of two homologous catalytic ␣and ␤-subunits of similar size and two copies of an inhibitory ␥-subunit (1,(5)(6)(7)(8). Cone PDE6 catalytic dimer is made up of two identical PDE␣Ј subunits (9). A cone-specific inhibitory P␥subunit is highly homologous to the rod P␥ (10). The ␦-subunit associates with soluble rod and cone PDEs. It interacts with the methylated prenylated C termini of PDE6 catalytic subunits and regulates the enzyme attachment to the membrane (11). The role of the PDE ␦-subunit in phototransduction is not well-defined, although it may modify the activity of the cascade by uncoupling transducin and PDE (12).
PDE6 enzymes have catalytic domains of about 280 aa residues in the C-terminal part of the molecule, which is highly conserved among all known cyclic nucleotide phosphodiesterases (3,4). The catalytic region of photoreceptor PDEs closely resembles the catalytic site of cGMP binding, cGMP-specific PDE (PDE5) (45-48% sequence identity) (13). Furthermore, PDE6 and PDE5 share a strong substrate preference for cGMP and have similar patterns of inhibition by competitive inhibitors, including zaprinast, dipyridamole, and sildenafil (13)(14)(15). In addition to the C-terminal catalytic domain, PDE6 contains two N-terminal GAF domains (GAFa and GAFb). GAF domains have been recognized as a large family of domain homologues and named for their presence in cGMP-regulated PDEs, adenylyl cyclases, and the E. coli protein Fh1A (16). Besides PDE6, several other PDE families possess GAF domains, including cGMP-stimulated PDE (PDE2), PDE5 (13,17,18), PDE10 (19), and PDE11 (20). At least one of the two GAF domains in the PDE2, PDE5, and PDE6 catalytic subunits serves as a site for noncatalytic binding of cGMP. Noncatalytic cGMP binding to GAF domains affects the catalytic properties of PDE2 and PDE5 (21)(22)(23)(24). In PDE6, noncatalytically bound cGMP appears to enhance the affinity of the inhibitory interaction between P␥ and the catalytic core (25,26). The second major function of the GAF domains is their role in dimerization of the PDE catalytic subunits. Earlier biochemical studies of PDE2 and PDE5 indicated that their dimerization occurs within the N-terminal parts of the molecules (18,27). Ultimate evidence on the intersubunit interface of PDE2 has been recently provided by a solution of the crystal structure of PDE2A GAFa-GAFb domains (28). The crystal structure revealed that the PDE2A regulatory region forms a dimer with the interface formed by the two GAFa domains (28). The role of PDE6 GAF domains in dimerization is supported by recent electron microscopy imaging of rod PDE6␣␤ (29). The imaging showed that the main intersubunit interaction occurs between the very N-terminal domains, presumably involving GAFa modules, and indicated that the GAFb domains may also contribute to the interface.
Almost all PDEs known to date are dimeric. However, the role of dimerization in enzyme function is not understood. Dimerization is not required for catalytic activity, because the isolated monomeric catalytic domain of PDE5 (30) and the monomeric short splice variant PDE4D2 (31) have been shown to be catalytically active. Furthermore, a majority of PDEs, with a notable exception of rod PDE6␣␤, form catalytic homodimers (3,4). Although the possibility of minor homodimeric species, ␣␣ and ␤␤, has not been ruled out, the dominant catalytic species of rod PDE6 is clearly a heterodimer ␣␤ (32,33). The functional significance of heterodimerization of rod PDE remains unclear. It is as yet unknown if the catalytic characteristics of PDE6␣ and PDE6␤ in the dimer are equivalent. To circumvent the problem of a lack of efficient expression of PDE6 in various cell types (34,35), we have previously developed a robust system for expression of PDE6␣Ј/PDE5 chimeras in insect cells (36). In this study, we extended this approach to chimeras between PDE6␣␤ and PDE5 to investigate the structural basis for specific dimerization of the rod PDE6 catalytic subunits. A series of chimeric proteins between PDE6 and PDE5 have been constructed that contain the GAFa and/or GAFb domains of PDE6 or PDE5. These chimeras were expressed in Sf9 cells in various combinations as His-, myc-, or FLAG-tagged proteins. The dimerization of chimeric PDE6 subunits was assessed using gel filtration, sucrose gradient centrifugation, Western blotting, and immunoprecipitation with anti-FLAG-and anti-myc-specific antibodies. The patterns of dimerization of chimeric PDE6/PDE5 subunits are consistent with the selectivity of PDE6 dimerization in rod photoreceptor cells. Our results indicate that the PDE6 GAFa domains contain major structural determinants for the affinity and selectivity of dimerization of PDE6 catalytic subunits.

EXPERIMENTAL PROCEDURES
Materials-cGMP was obtained from Roche Molecular Biochemicals. [ 3 H]cGMP was a product of Amersham Biosciences. All restriction enzymes were purchased from New England BioLabs. AmpliTaq DNA polymerase was a product of PerkinElmer Life Sciences, and Pfu DNA polymerase was a product of Stratagene. All other reagents were purchased from Sigma. Bovine holo-PDE6 was extracted from bleached rod outer segment membranes and purified as described previously (33).
Cloning, Expression, and Purification of PDE6/PDE5 Chimeras-To construct PDE6␣/PDE5 chimera FLAG-␣-␣-5 (see Fig. 1), the FLAG tag DNA sequence was first inserted into the pFastBacHTb vector (Invitrogen) replacing the His6 sequence. A DNA fragment was PCR-amplified using pFastBacHTb as a template, a 5Ј-primer containing an RsrII site and the FLAG tag sequence, and a 3Ј-primer containing a BamHI site. This fragment was then ligated with the large RsrII/BamHI fragment of pFastBacHTb to produce pFastBacFLAG. A bovine retinal cDNA library kindly provided by Dr. W. Baehr (University of Utah) was used as a template for PCR amplifications of PDE6␣␤ sequences. DNA coding for PDE6␣-1-443 was PCR-amplified using primers carrying BamHI and HindIII sites. DNA coding for PDE5-506-865 amino acids was PCR-amplified using the pFastBacHTb-PDE5 template (36) and primers with the HindIII and XhoI sites. The two PCR products were ligated into the pFastBacFLAG vector using the BamHI and XhoI sites. The myc tag DNA sequence was inserted into the pFastBacHTb vector using a PCR-directed cloning procedure similar to the insertion of the FLAG tag sequence. The construct for chimera myc-␤-␤-5 (see Fig. 1) was generated by ligation of the PCR-amplified PDE6␤-1-441 and PDE5-506-865 DNAs into the BamHI and XhoI sites of pFastBacmyc.
The His-tagged PDE6␣␤/PDE5 chimeras were constructed by amplifying appropriate regions of the PDE6␣␤ subunits with primers containing unique restriction sites. When unique restriction sites were not available at the desired location, first-round PCR products coding chimeric junctions were extended to the nearest unique sites in a secondround PCR amplification. To improve the recognition of the His 6 tag by commercial antibodies, the His 6 flanking sequence in the original pFastBacHTb vector was replaced by the His 6 flanking sequence from pET-15b (Novagen). The DNA sequences of all constructs were con-firmed by automated DNA sequencing at the University of Iowa DNA core facility.
Generation of the recombinant bacmids, transfection of Sf9 cells, and viral amplifications were carried out according to the manufacturer's recommendations (Invitrogen). For protein expression, Sf9 cell cultures (10 ml, 2 ϫ 10 6 cells/ml) were infected with one or two different viruses at a multiplicity of infection of 3-10. Sf9 cells were harvested at 48 h after infection by centrifugation and stored at Ϫ80°C until use. Sf9 cells were resuspended in 3 ml of 20 mM Tris-HCl buffer (pH 8.0) containing 2 mM MgSO 4 and Complete TM Mini protease inhibitor mixture (one-third tablet) (Roche Molecular Biochemicals) and sonicated with two 10-s pulses using a microtip attached to a 550 Sonic Dismembrator (Fisher Scientific). Sf9 cell lysates were cleared by centrifugation (100,000 ϫ g, 90 min, 2°C), dialyzed against 30 mM Tris-HCl buffer (pH 8.0) containing 130 mM NaCl, 2 mM MgSO 4 , and 50% glycerin, and then centrifuged again at 100,000 ϫ g for 1 h at 2°C. Dialysis against 50% glycerin allowed a concentration of PDE samples by ϳ4-fold and subsequent storage at Ϫ20°C without freezing. The presence of glycerin did not affect the behavior of PDEs in gel filtration.
Gel Filtration and Fraction Analysis-Aliquots of dialyzed PDE samples (50 -200 l) were injected into a Superose 12 10/30 column (Amersham Biosciences) equilibrated at 25°C with 30 mM Tris-HCl buffer (pH 8.0) containing 100 mM NaCl and 2 mM MgSO 4 . Proteins were eluted at 0.4 ml/min, and 0.4-ml fractions were collected starting at 17 min post-injection. Each fraction was assayed for PDE activity, protein concentration, and the presence of chimeric PDEs by Western blotting. PDE activity was measured using 10-to 20-l aliquots from fractions and 5 M [ 3 H]cGMP as described previously (37,38). Protein concentrations were determined by the method of Bradford using IgG as a standard (39). The column was calibrated with the following protein standards: bovine thyroglobulin (670 kDa, 85 Å), horse ferritin (440 kDa, 61 Å), sweet potato ␤-amylase (200 kDa), rabbit aldolase (158 kDa, 48.1 Å), bovine serum albumin (67 kDa, 35.5 Å), and chicken ovalbumin (45 kDa, 30.5 Å). The Stokes radii for PDEs were estimated using the correlation of elution volume with the Stokes radius proposed by Porath (40). Gel filtration analyses were performed two or more times with similar results for each PDE chimera combination from at least two different preparations of Sf9 cell extracts. Results of a typical analysis are shown.
Western blot analysis of the gel filtration fractions (20-l aliquots) was performed following SDS-PAGE in 10% gels (41). Monoclonal antipolyhistidine, M2 monoclonal anti-FLAG, and monoclonal anti-c-myc (clone 9E10) antibodies (Sigma) with the respective dilutions of 1:1500, 1:5000, and 1:5000 were utilized. The antibody⅐antigen complexes were detected using anti-mouse antibodies conjugated to horseradish peroxidase (Sigma) and ECL reagent (Amersham Biosciences). The compositions of PDE complexes in fractions corresponding to dimeric enzymes were examined by immunoprecipitation (IP). Aliquots of the gel filtration fractions (80 l) were incubated with or without anti-FLAG or myc-antibodies (1 l) for 30 min at 25°C followed by the addition of 5 l of protein G-Agarose (Sigma) and incubation for 40 min at 25°C. The agarose beads were washed four times with 300 l of phosphate-buffered saline (pH 7.1), and the bound proteins were eluted with an SDS-PAGE sample buffer. PDE complexes were separated on 10% gels and analyzed by Western blotting using appropriate antibodies.
Sucrose Gradient Centrifugation-Sucrose density gradients (5-35%) were prepared in 30 mM Tris-HCl (pH 8.0) buffer containing 100 mM NaCl, 2 mM MgSO 4 , and 4 mM 2-mercaptoethanol using a Gradi-Frac gradient former (Amersham Biosciences). Protein standards or aliquots of 200 l from peak PDE gel filtration fractions were loaded onto the gradients in 14-ϫ 89-mm centrifugation tubes and centrifuged for 24 h at 40,000 rpm in a Beckman SW41 rotor at 4°C. Fractions of 300 l were collected starting from the bottom of the tubes. Fractions from the tubes with protein standards were analyzed for protein concentration, whereas fractions from the tubes containing PDE samples were analyzed for PDE activity. Sucrose density centrifugations were performed two times with similar results for each PDE preparation. Results of a typical analysis are shown. The protein standards were: bovine liver catalase (250 kDa, 11.3 S), rabbit aldolase (158 kDa, 7.3 S), bovine serum albumin (67 kDa, 4.6 S), and chicken ovalbumin (45 kDa, 3.5 S). Sedimentation coefficients (s 20,w ) for chimeric PDEs were estimated using a linear plot of distances traveled by standards from meniscus versus the s 20,w values of standards (42). The molecular weights of PDEs were calculated using the estimated sedimentation coefficients, the Stokes radii obtained from the gel filtration data, and the following equation (43), where s is the sedimentation coefficient; N A is Avogardo's number; is the viscosity of the medium (0.01 g⅐cm Ϫ1 ⅐s Ϫ1 ), r is the Stokes radius; v is the partial specific volume of a protein (0.73 cm 3 ⅐g Ϫ1 ), and is the density of the medium (1 g⅐cm Ϫ3 ).

Selectivity of Dimerization of the GAFa-GAFb
Domains of Rod PDE6 -Dimerization of PDE6 catalytic subunits is very tight, and, apparently, there is no exchange of subunits between dimers once they are formed following the synthesis and folding of the polypeptide chains. The dimer formation between chimeric PDE6/PDE5 subunits was therefore assessed following co-expression of these chimeras in Sf9 cells. Chimeric PDEs ( Fig. 1) from soluble fractions of Sf9 cells as well as native PDE6 and recombinant wild-type PDE5 were examined by FPLC gel filtration on a calibrated Superose 12 HR 10/30 column and by sucrose gradient centrifugation. The elution profiles of native PDE6 and recombinant wild-type PDE5 on the gel filtration column were very similar and corresponded to an apparent molecular mass of ϳ210 kDa ( Fig. 2A), which is consistent with dimerization of the catalytic subunits. Molecular masses of rod holoPDE6 (PDE␣␤␥ 2 ) and the PDE5 dimer calculated from the sequences were ϳ217 and 197 kDa, respectively. Only relatively small fractions of the enzymes (ϳ15% PDE5 and ϳ10% PDE6) migrated as aggregates with high molecular mass. The previously developed chimeric PDE, Chi4 (termed hereafter His-␣Ј-␣Ј-5), containing the cone PDE6␣Ј GAFa-GAFb region and the catalytic domain of PDE5 (36), is predicted to form homodimers. The chromatographic behavior of His-␣Ј-␣Ј-5 on a Superose 12 HR 10/30 column indicated an apparent molecular mass of ϳ200 kDa ( Fig. 2A), which is in good agreement with the theoretical molecular mass of 183 kDa for the ␣Ј-␣Ј-5 homodimer. The fraction of aggregates in the chimeric PDE (ϳ10%) was similar to those in PDE5 and PDE6 preparations.
We first constructed PDE6/PDE5 chimeras containing both the GAFa and GAFb domains from either rod PDE6␣ (␣-␣-5) or PDE6␤ (␤-␤-5) and the catalytic domain from PDE5 (Fig. 1). As generally defined by the PDE sequence alignment and the crystal structure (17,28), the boundaries of the GAFa domains are PDE6␣-54 -220 and PDE6␤-52-218, and the boundaries of the GAFb domains are PDE6␣-255-443 and PDE6␤-253-441. The nonconserved N termini, PDE6␣-1-53 in ␣-␣-5 and PDE6␤-1-51 in ␤-␤-5, were from the respective catalytic subunits. Chimera ␣-␣-5 was constructed for expression as a FLAG-tagged protein, whereas ␤-␤-5 was generated as a myctagged polypeptide. Gel filtration fractions were analyzed by Western blotting for the presence of chimeric PDE proteins and for PDE activity with cGMP as the substrate. Individual expression of FLAG-␣-␣-5 or myc-␤-␤-5 resulted in the formation of only high molecular weight aggregates that migrated with the exclusion volume of the Superose 12 column (Fig. 2, B and C). Furthermore, the aggregates also migrated with the exclusion volume on a Superose 6 HR 10/30 column (not shown), which is capable of separating proteins weighing up to 1 ϫ 10 6 kDa. Notably, these aggregates were capable of hydrolyzing cGMP. When FLAG-␣-␣-5 and myc-␤-␤-5 were co-expressed in Sf9 cells, a peak of PDE activity appeared in the fractions corresponding to dimeric PDE species with an apparent molecular mass of ϳ200 kDa (Fig. 3A). Approximately 50% of the total soluble FLAG/myc-tagged protein formed the dimeric enzyme. In the peak activity fractions 11 and 12, the enzyme hydrolyzed cGMP with a K m value of 3.8 M and a V max of 21 nmol/min/mg of protein. The Western blot analysis of these fractions using anti-FLAG and anti-myc antibodies confirmed the presence of both FLAG-␣-␣-5 and myc-␤-␤-5 (Fig. 3A). A sizable peak of PDE activity was also present in fractions corresponding to aggregates of FLAG-␣-␣-5 and myc-␤-␤-5 (Fig.  3A). The aggregation appears to be irreversible, because rechromatography of the fractions with aggregates did not generate dimeric enzymes (not shown). Considering the intensity of immunostaining, the catalytic activity of these aggregated PDE species is 1.5-to 2-fold lower than that of the dimeric PDE species.
To verify the nature of the dimeric species, chimeric PDE was immunoprecipitated with anti-FLAG antibodies and then probed with anti-myc antibodies using Western blotting. The results of the IP experiments proved the formation of a heterodimer between FLAG-␣-␣-5 or myc-␤-␤-5 (Fig. 3A). The peak of dimeric PDE activity on the gel filtration column was relatively broad. To determine if this PDE might be heterogeneous, the combined concentrated front fractions 9 -11 and tail fractions 13-16 were reapplied onto the column. The resulting PDE activity profiles were nearly identical, suggesting the presence of a single dimeric form of PDE and trace amounts of aggregates (Fig. 3B).
Next, we examined the possibility of dimerization of PDE6␣ and PDE6␤ with cone PDE6␣Ј and PDE5. FLAG-␣-␣-5 and myc-␤-␤-5 each were co-expressed with His-␣Ј-␣Ј-5 or His-PDE5. Because PDE6␣Ј and PDE5 form catalytic homodimers, a peak of PDE activity in gel filtration fractions corresponding to dimeric PDE species cannot be used as evidence for heterodimerization. Instead, we relied on Western blot analysis of , and by Western blotting with anti-FLAG or anti-myc antibodies. IP, an aliquot (80 l) of the dimeric PDE peak fraction 12 was incubated with (ϩ) or without (Ϫ) anti-FLAG antibodies. Protein⅐antibody complexes were isolated using protein Gagarose and analyzed by Western blotting with anti-myc as described under "Experimental Procedures." B, combined PDE peak front fractions 9 -11 (solid line) and tail fractions 13-16 (dashed line) were concentrated by dialysis against 30 mM Tris-HCl buffer (pH 8.0) containing 130 mM NaCl, 2 mM MgSO 4 , and 50% glycerin and then reapplied onto the column. C, combined peak PDE gel filtration fractions 10 -13 were concentrated to a volume of 200 l using the YM-10 Microcon devices (Millipore Corp., Bedford, MA) and loaded onto the 5-35% sucrose density gradients. Following centrifugation for 24 h at 40,000 rpm in a Beckman SW41 rotor, fractions of 300 l were collected starting from the bottom of the tubes and analyzed for PDE activity. Arrows indicate sedimentation of protein standards. PDE5, holoPDE6, and ␣Ј-␣Ј-5 migrated at the same position of the gradient indicated as PDE*.  (Fig. 4A). The FLAG-␣-␣-5 signal appeared in fractions 10 -14 corresponding to dimeric PDE. The immunoprecipitates of these fractions with anti-FLAG antibodies contained His-␣Ј-␣Ј-5, demonstrating the heterodimeric composition of the dimers (Fig. 4A). Co-expression of myc-␤-␤-5 and His-␣Ј-␣Ј-5 has not led to any significant dimer formation between the two chimeric PDEs (Fig. 4B). Similarly, co-expression of FLAG-␣-␣-5 or myc-␤-␤-5 with PDE5 and examination of formed PDE species revealed no detectable dimerization between the GAF regions of rod PDE6 and PDE5 (not shown). The patterns of dimerization ␣-␣-5 or ␤-␤-5 were consistent with the selectivity of PDE6 dimerization in rod photoreceptor cells, allowing us to use them as templates to further probe the role of the PDE6 GAFa and GAFb domains.
The GAFb domains of PDE6␣ and PDE6␤ are more homologous than the GAFa domains and may contribute to the affinity of dimerization without influencing its selectivity. Chimera His-␤-5-5 containing the GAFa domain of PDE6␤ and the GAFb domain of PDE5 was generated to test this possibility. His-␤-5-5 did not exhibit any propensity for self-dimerization (not shown) but was able to efficiently form heterodimers with FLAG-␣-␣-5 when the two proteins were co-expressed in Sf9 cells (Fig. 6A). Sucrose density centrifugation of FLAG-␣-␣-5⅐His-␤-5-5 produced a s 20,w value of 8.1 confirming the dimeric nature of the complex (not shown). Therefore, the GAFb domains of PDE6 do not significantly contribute to the dimeric interface.

DISCUSSION
The structural basis and functional role of PDE dimerization are poorly understood because dimerization is not required for catalytic function. The first molecular insights into PDE dimerization have been revealed by the structure of the regulatory domains of PDE2A (28). This crystal structure demonstrates that the GAFa domain is responsible for the dimerization of PDE2A. One apparent implication from the PDE2A structure is that other GAF domain-containing PDEs, such as PDE5, PDE6, PDE10, and PDE11 utilize GAF modules for dimerization. Yet, it remains unclear how well the dimerization interfaces of PDE5 and PDE6 parallel that of PDE2. Although at a relatively low resolution, electron microscopy imaging of PDE5 and PDE6 showed molecular shapes that are somewhat different from the PDE2 structure (29). Furthermore, the electron microscopy study indicates that, in addition to GAFa, GAFb domains may contribute to the intersubunit interaction in PDE5 and PDE6. Analysis of the dimerization interface of rod PDE6 catalytic subunits permitted us to address two unresolved questions. What are the key structural determinants for dimerization of PDE6, and are they similar to those identified in the PDE2A structure? What are the selectivity determinants of rod PDE6 heterodimerization? The heterodimerization of rod PDE6 ␣ and ␤ subunits is unique among known PDEs. It may be critical in the forming of rod-specific photoresponses. However, a potential significance of PDE6␣␤ heterodimers and the properties of individual subunits cannot be assessed in the absence of ␣␣ and ␤␤ homodimers. Identification of the selectivity determinants would be a first major step toward generation of the homodimeric species.
Our analysis of dimerization of chimeric PDE6␣␤/PDE5 pro-teins using gel filtration, immunoprecipitation, and sucrose gradient centrifugation demonstrated selective heterodimerization between the GAF regions of PDE6 ␣and ␤-subunits with no significant homodimerization. This observation is in accordance with the established heterodimeric nature of the rod enzyme. Interestingly, the GAFa-GAFb region of PDE6␣, but not PDE6␤, was capable of forming a dimer with the GAFa-GAFb region of PDE6␣Ј. Although dimerization of PDE6␣ and PDE6␣Ј does not have physiological implications, as the subunits are expressed in different types of photoreceptor cells, it may provide additional clues to understanding PDE6 intersubunit interfaces. The patterns of dimerization (or lack thereof) (Table I) show that the GAFa domains are the major contributors to the dimer assembly of PDE. The finding that the PDE6 GAFa domains, similar to the PDE2 GAFa domains, are responsible for its dimerization suggests a common structural organization of the intersubunit interfaces of the GAF domain-containing PDEs. The subsequent identification of the PDE6␣␤ dimerization selectivity determinants was carried out using chimeras carrying complementing portions of the PDE6 ␣ and ␤ GAFa domains. The ability of His-␤␣-␤-5 to dimerize with FLAG-␣-␣-5 has implicated the N-terminal segment of the GAFa domains in the exclusive PDE6␣␤ association. Further evidence was provided by the analysis of the chimeric PDEs, His-␣ ␤ -␣-5 and His-␤ ␣ -␤-5, containing short replacements PDE6␤-57-72 and PDE6␣-59 -74 within the PDE6␣ and PDE6␤ GAF regions, respectively. PDE6␤-57-72 and PDE6␣-59 -74 correspond to a region of PDE2GAFa, ␣1 helix-␣1/␣2 loop, that is involved in the PDE2 dimer interface (28). Not only did the replacements prevent heterodimerization between His-␣ ␤ -␣-5 and myc-␤-␤-5, and between His-␤ ␣ -␤-5 and FLAG-␣-␣-5, but His-␣ ␤ -␣-5 gained the ability for self-dimerization and heterodimerization with FLAG-␣-␣-5. Dimerization of His-␣ ␤ -␣-5 with FLAG-␣-␣-5 suggests that the lack of homodimerization of PDE6␣ is caused by a defect in the interaction between the two PDE6␣-59 -74 segments (Fig. 7D). However, homodimerization of His-␣ ␤ -␣-5 coupled with the absence of association of His-␤ ␣ -␤-5 with myc-␤-␤-5 indicates that the lack of homodimerization of PDE6␤ cannot be accounted for by the defective interaction between the two PDE6␤-57-74 segments. Homology modeling of the PDE6␣␤ GAFa dimer using the structure of PDE2GAFa dimer as a template indicates that PDE6␤-57-72 may participate in two sets of interactions involving PDE6␣-59 -74 and PDE6␣-216 -224 (Fig. 7, A-C). PDE6␣-216 -224 resides at the start of the helix connecting GAFa and GAFb. Table I shows that no dimers had been formed that would entail the interaction between PDE6␤-57-72 and the beginning of the connecting helix from PDE6␤. Therefore, this structural constraint appears to disallow homodimerization of PDE6␤ (Fig. 7D). Additional negative dimerization determinants within the GAFa domain of PDE6␤ cannot be ruled out. A significant number of residues within PDE6␣-59 -74/PDE6␤-57-72 and PDE6␣-216 -224/PDE6␤-214 -222 are not conserved (Fig. 7B) and may play a role in the selective assembly of PDE6␣␤.
The finding that the selectivity determinants of PDE6 dimerization are confined to relatively short segments in the GAFa domains supports the feasibility of generating mutant PDE6 ␣ and ␤ subunits capable of homodimerization. Mutant homodimeric rod PDE6 expressed in transgenic animals would allow us to study the individual catalytic subunits and elucidate the functional significance of rod PDE6 heterodimerization.