Differential Expression of Rod Photoreceptor cGMP-Phosphodiesterase α and β Subunits

The catalytic core of photoreceptor-specific cGMP-phosphodiesterase (PDE) consists of two subunits, PDEα and PDEβ, that are homologous and have similar domain organization but are encoded by different genes. We have examined the PDEα and PDEβ mRNA steady-state and protein levels as well as the biosynthesis rate of these proteins in developing and fully differentiated retinas. We have also determined the translational efficiency of PDE subunits and the role of their mRNA structures in regulating protein synthesis. In mature retinas, PDEα and PDEβ are represented by ∼1.5 × 108 and 7.5 × 108 copies/μg retinal mRNA, respectively. The levels of these transcripts in developing photoreceptors (P10) are approximately 75% of those at P30. Quantification of protein concentration indicated that PDEα and PDEβ are equally expressed in developing and fully differentiated photoreceptors. Furthermore, the PDEα/PDEβ ratios obtained throughout a 2-h pulsechase period revealed a similar turnover rate for both subunits. The observed discordance between the mRNA and protein levels of PDEα and PDEβ suggested post-transcriptional regulation of their expression. We found that PDEα mRNA is translated more efficiently than either of the two PDEβ transcripts expressed in retina. Therefore, the lower level of PDEα mRNA is compensated by its more efficient translation to achieve equimolar expression with PDEβ. We also analyzed the effect of PDEα and PDEβ mRNA 5′- and 3′-untranslated regions as well as that of their coding regions on protein synthesis. We determined that the PDE-coding regions play a critical role in the differential translation of these subunits.

Signal transduction in the vertebrate rod and cone photoreceptors is mediated by a cascade of protein interactions/activations that lead to hydrolysis of cGMP by a photoreceptor-specific cGMP-phosphodiesterase (PDE) 1 followed by closure of the cGMP-gated channels on the plasma membrane (1). The catalytic core of cone PDE is composed of two identical ␣Ј subunits, whereas that of the rod enzyme is a heterodimer (PDE␣, 88 kDa and PDE␤, 84 kDa) (2,3). Although each of these three proteins is encoded by a different gene, they share significant homology in both amino acid and nucleotide sequences. Cone and rod PDEs have similar domain organization and contain two cGMP-binding site motifs, a catalytic site and a CAAX box signal for post-translational modification (4). Why the rod holoenzyme consists of two different subunits is still unknown. Based on classical biochemical studies carried out with the bovine rod cGMP-PDE, it has been accepted that PDE␣ and PDE␤ are present in a 1:1 ratio and that they are assembled in a heterodimeric catalytic core (2,3,5). This coordinated expression of PDE␣ and PDE␤ in rod photoreceptors must be regulated at the transcriptional and/or post-transcriptional level so that a final equimolar ratio of these proteins is achieved. However, there is some indirect evidence suggesting the possibility that the rod PDE catalytic core is not only present as a heterodimer but also as ␣␣ and ␤␤ homodimers (6), similar to cone ␣Ј␣Ј PDE. We have previously demonstrated that the individually expressed catalytic subunits have PDE enzymatic activity (7), supporting the idea of independent catalytic motifs on PDE␣ and PDE␤. Thus, PDE␣ and PDE␤ may be present in photoreceptor cells at any ratio depending on their expressional "strength." The proteins of the phototransduction cascade, PDE in particular, have been extensively studied for many years. However, very little is known regarding the regulation of PDE␣ and PDE␤ expression at either the transcriptional or translational level. Although the 5Ј-flanking regions of PDE␣ and PDE␤ genes have been cloned and sequenced, a detailed analysis that would allow us to speculate about the possible mechanisms of coordinated expression of these genes has not been carried out. A sequence analysis of the 5Ј-flanking regions of the PDE␣ and PDE␤ revealed two common potential regulatory elements: a Ret1-like site, the binding site for the retina-specific Ret1 transcription factor (8), and a GC box, the binding site for the SP family of transcription factors. Work coming out from our laboratory has implicated Sp4 in the transcriptional regulation of PDE␤ (9,10), but the effect of this factor on PDE␣ has not yet been established.
This study was initiated with the intention of: 1) determining the PDE␣ and PDE␤ mRNA steady-state levels as well as protein levels in the mouse retina at different stages of photoreceptor development; 2) examining the turnover of translated PDE␣ and PDE␤ by pulse-chase labeling and immunoprecipitation to determine whether the final stoichiometric amounts of the catalytic subunits are the result of differences in either efficiency of translation or post-translational protein stability; and 3) determining the PDE␣ and PDE␤ mRNA translational efficiency and the role of their 5Ј-UTRs, coding regions, and 3Ј-UTRs in protein synthesis.

EXPERIMENTAL PROCEDURES
RNA Isolation and in Vitro Transcription-Total RNA was extracted from 10-and 30-day-old normal mouse retinas using RNAzol B (Tel-Test, Friendswood, TX). mRNA was isolated using the mRNA purification kit (Amersham Biosciences). Concentrations of the RNA samples were determined by measuring their optical density.
For in vitro transcription, the cDNAs were subcloned into the pGEM3Z vector (Promega, Madison, WI) between BamHI and SalI. In vitro transcription was performed using 1 g of linearized DNA and the T7 mMessage mMachine kit (Ambion, Austin, TX) according to the manufacturer's protocol. The length and the concentration of the synthesized transcripts were determined by running an aliquot on a denaturing agarose gel as recommended by Ambion. Quantification of the samples was performed using the two-dimensional spot densitometry program of the Alpha Imager 2000 (Alpha Innotec, San Leandro, CA).
RT-PCR Analysis-First strand cDNA was synthesized using murine leukemia virus reverse transcriptase (PerkinElmer Life Sciences), 0.75 mM of antisense primer, 1 mM dNTPs, 10 units of RNAsin (Promega), 5 mM MgCl 2 , 10 mM Tris-HCl, pH 8.3, and 50 mM KCl. The reaction mixture was incubated at 42°C for 15 min. PCR was performed with AmpliTaq DNA polymerase (PerkinElmer Life Sciences), 0.75 mM antisense and sense primers, 2 mM MgCl 2, 10 mM Tris-HCl, pH 8.3, and 50 mM KCl. The PCR profile time and temperature were as follows: denaturation at 94°C for 2 min, annealing at 57°C for 1 min 20 s, and extension at 72°C for 1 min 30 s with a final elongation step at 72°C for 5 min.
Quantitative Northern Blot and Quantitative RT-PCR-mRNAs (2 g) were separated by electrophoresis in a 1.2% agarose gel containing 2.2 M formaldehyde, transferred to Hybond-N ϩ (Amersham Biosciences), cross-linked, and hybridized with random-labeled PDE␣ and PDE␤ cDNA fragments. Autoradiographs were analyzed using the twodimensional spot densitometry program of the Alpha Imager 2000 to determine the intensity of the hybridization signals.
Quantification of the PDE␤ transcripts differing by the length of their 5Ј-UTRs was based on the PCR method described by Riedy et al. (11). A competitive reference standard RNA (RNA-CRS) was obtained as follows: cDNA fragments with a 75-nucleotide deletion were synthesized by PCR using primers MB7 (5Ј-CGGGATCCTGCAGAATCCAT-GTGTACC-3Ј) and MB45 (5Ј-TGCGGTCGACAGGATCTTGAAGACCA-CACGTTCCAAGGCTGCCACAGTCCGCCAG-3Ј) and full-length PDE␤ as template. The resulting product was subcloned into the pGEM3Z vector between BamHI and SalI, and the RNA-CRS was synthesized using the T7 mMessage mMachine kit as described above.
Total retinal RNA and serial dilutions of RNA-CRS were used in RT-PCR with appropriate primer pairs (MB4 5Ј-TGCAGAATCCATGT-GTACCTGGGGG-3Ј and MB25 5Ј-AGGATCTTGAAGACCACACGT-TCCA-3Ј for amplification of the long PDE␤ transcript and MB5 5Ј-ACAGCAGCAGGAACACCATGAGCCT-3Ј and MB25 for both the long and short transcripts) to obtain data for mRNA quantification curves. Conditions for RT-PCR were the same as described above. Aliquots from each reaction were electrophoresed in a 2% agarose gel and stained with 0.5 mg/ml ethidium bromide, and the bands were quantified using the spot densitometry function of the Alpha Imager 2000. Since the RNA-CRS cDNA is smaller in size than the wild type cDNA, the decrease in ethidium bromide incorporation was taken into account by using a correction equation (12).
In Vitro Translation-Equal amounts of in vitro synthesized RNAs were used for all of the in vitro translation studies. In vitro translation reactions were performed in the presence of L-[ 35 S]methionine with the retic lysate IVT kit (Ambion) according to the manufacturer's protocol. Translation products were separated by SDS-polyacrylamide gel electrophoresis (6.5% acrylamide, 1.5% bis-acrylamide) using the Tris-Tricine buffer system as previously described (13). The gels were then fixed, stained with Coomassie Blue G-250, dried, and exposed to x-ray film. Proteins were quantified by densitometry of the autoradiograph or direct scintillation counting of the excised bands.
Preparation of Mouse Rod Outer Segments (ROS) and Immunoblot Analysis-An analysis of the stoichiometric amounts of PDE␣ and PDE␤ subunits was performed on retinas from 10-and 30-day-old animals and on sealed ROS isolated from 48 adult (greater than 30 days of age) C57/Bl retinas by sucrose gradient centrifugation according to the method of Papermaster (14). Proteins (50 g of total protein/lane from retinal homogenates and 8 g of total protein/lane from ROS) were separated by Tris-Tricine-buffered inverted gradient gel electrophoresis (13) and transferred to nitrocellulose membrane (15). Membranes were blocked in either 3% bovine serum albumin, 0.1% Tween 20 in Trisbuffered saline (500 mM NaCl, 20 mM Tris, pH 7.6) or 0.3% Tween 20 in Tris-buffered saline for colloidal gold staining. Immunoblots were incubated with a polyclonal peptide antiserum (␣T1) previously characterized (7), and the signal was amplified with a goat anti-rabbit biotinylated secondary antibody and AvidinD/biotin-coupled horseradish peroxidase (Vector Laboratories, Burlingame, CA). Blots were visualized with the ECL kit (Amersham Biosciences) on Hyperfilm-MP preflashed according to the Sensitize TM protocol (Amersham Biosciences) to increase the range of film exposure linearity. Films were developed and scanned densitometrically on a Pharmacia/LKB XL densitometer (Amersham Biosciences).
Labeling and Immunoprecipitation-Mice were sacrificed at 10 postnatal days (according to the guidelines of the UCLA Committee on Animal Research and the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research) and enucleated, and the eyes were hemisected. The resulting eyecups were incubated in HEPES-buffered Medium 199 with Hanks' salts without dl-methionine (Invitrogen) for 30 min at 37°C in a 95% O 2 , 5% CO 2 incubator to methionine-deplete the tissue. The eyecups were then incubated with L-[ 35 S]methionine (1000 Ci/mmol and 100 Ci/ml media, Amersham Biosciences) for 3 h and subsequently washed and incubated in fresh Medium 199 supplemented with cold l-methionine at 500 M final concentration. At the appropriate times, eyecups were removed and the retinas were dissected and immediately frozen at Ϫ80°C on dry ice to prevent further incorporation of label and protein degradation. Retinas were homogenized with hypotonic buffer, and aliquots were removed to determine trichloroacetic acid-precipitable counts and protein concentration by the method of Peterson (16) using bovine serum albumin as standard.
Immunoprecipitation was carried out as described previously (17) using a polyclonal rod cGMP-PDE-specific antiserum, MOE (18). Proteins were separated by a Tris-Tricine-buffered inverted gradient sodium dodecyl sulfate-polyacrylamide gel as described previously (13) for 22 h at 35 mA constant current. Proteins were then blotted to nitrocellulose membrane, stained with Aurodye total protein stain (Bio-Rad), and exposed to autoradiography on pre-flashed Hyperfilm-MP.

Quantification of PDE␣ and PDE␤ mRNAs by Northern Blot
Hybridization-This method for absolute quantification of specific mRNAs is based on the generation of a standard curve using known amounts of an external standard (synthetic sense strand RNA with identical sequence to the target mRNA) and interpolation of the concentration of the target mRNA from the curve. The external standards for PDE␣ and PDE␤ mRNA quantification were synthesized in vitro from constructs containing full-length PDE␣ and PDE␤ cDNAs. mRNA samples isolated from 10-and 30-day-old mouse retinas were hybridized with the random-labeled PDE␣ and PDE␤ cDNA fragments. Two transcripts with molecular masses of 3.0 and 2.8 kb corresponding to the PDE␣ and PDE␤ mRNAs, respectively, were hybridized. Using the external standard curve, we determined that in the 30-day-old mouse retina the level of PDE␣ transcript was 1.5 Ϯ 0.05 ng and that of PDE␤ transcript was 0.34 Ϯ 0.02 ng/1 g of retinal mRNA (Fig. 1). Based on the molecular masses of PDE␣ and PDE␤ transcripts (1.03 ϫ 10 6 and 0.87 ϫ 10 6 , respectively) we estimated that 1 g of retinal mRNA contains ϳ1.5 ϫ 10 8 (Ϯ0.2 ϫ 10 8 ) and 7.5 ϫ 10 8 (Ϯ0.5 ϫ 10 8 ) PDE␣ and PDE␤ transcripts, respectively. Similar quantification carried out with mRNA samples of 10-day-old mouse retina ( Fig. 1) showed that at this age the concentration of PDE␣ and PDE␤ transcripts is ϳ75% of that in the 30-day-old retina. The levels reported are the average of those determined in three different experiments.
Quantitative RT-PCR Analysis of mRNAs-It had been reported previously (19) that mouse PDE␤ has two different transcription start sites represented in the retina by two transcripts, one with a 59-nucleotide 5Ј-UTR and the other with a 34-nucleotide 5Ј-UTR. Because the 5Ј-UTR is a major element implicated in mRNA translation, the two PDE␤ transcripts could have different translation initiation strengths. Therefore, any changes in the molar ratio of these transcripts during the development of photoreceptor cells could be responsible for modulation of PDE␤ protein concentration.
Quantitative RT-PCR was used to determine the level of the PDE␤ transcript with long 5Ј-UTR and the total concentration of PDE␤ mRNA (the sum of the long and short transcripts) in the 10-and 30-day-old mouse retina. This technique involves the use of an RNA-CRS that is identical in sequence to the wild type mRNA with the exception of an internal nucleotide deletion. Primers MB4 and MB25 were used to amplify the long transcript, and primers MB5 and MB25 were used to amplify at the same time both the long and the short transcripts. PCR products were separated by gel electrophoresis, stained with ethidium bromide, and analyzed densitometrically (Fig. 2, Ia, Ib, IIa, and IIb). The concentration of the RNA-CRS (x axis) was plotted versus the ratio of the RNA-CRS to the wild type mRNA band densities (Fig. 2, Ic and IIc, y axis). The concentration of PDE␤ mRNA corresponds to the point on the x axis where the ratio RNA-CRS:PDE␤ mRNA is 1. The results of these experiments indicate that both long and short PDE␤ transcripts are represented at equimolar concentrations in the photoreceptor cells of the 10-and 30-day-old animals.
Western Blot Analysis-Given the 5-fold difference in transcript abundance observed by quantitative Northern blots, we had to establish the protein subunit stoichiometry in the mouse retina. We accomplished this by utilizing ␣T1, an antibody that recognizes the same epitope on both PDE␣ and PDE␤. The relative amounts of PDE␣ and PDE␤ were obtained by determining the corresponding integrated areas of the optical density peaks (IOD) and then the ratio of PDE␣ IOD to PDE␤ IOD (Fig. 3). Ratios of ϳ1:1 were obtained for all three of the samples analyzed, confirming equimolar expression of PDE␣ and PDE␤ subunits in mouse rod photoreceptors, similar to what is seen for the bovine enzyme.
Biosynthesis and Turnover-Using L-[ 35 S]methionine, we performed pulse-chase labeling of newly synthesized PDE subunits followed by immunoprecipitation with a specific polyclonal rod cGMP-PDE antiserum to confirm that the differences observed between protein and corresponding mRNA levels were the result of differential protein translation and/or turnover. The autoradiograph of the gel containing the immu-noprecipitated proteins (Fig. 4) was scanned densitometrically, and the IODs were obtained. The rate of L-[ 35 S]methionine incorporation was higher for PDE␣ than PDE␤ as indicated by values greater than 1.0 throughout the course of the experiment for the PDE␣/PDE␤-calculated IOD ratios. However, the range of observed IOD ratios from 1.20 to 1.41 suggests that the turnover rate for PDE␤ is not significantly greater than for PDE␣, particularly when the difference in the molar methionine content of PDE␣ and PDE␤, 34 residues and 29 residues, respectively, is taken into account.
Translational Efficiency of the Full-length PDE␣ and PDE␤ mRNAs-To determine the translational efficiency of the PDE␣ and PDE␤ mRNAs, three constructs were prepared by subcloning the PDE␣ and PDE␤ (B5s and B5l) cDNAs into the pGEM3Z vector (Fig. 5, Ia). The PDE␤ subunit is represented by two constructs since human photoreceptor cells express two PDE␤ transcripts that differ in the length of their 5Ј-UTRs. Capped mRNAs were in vitro transcribed from the A, B5s, and B5l templates, transcripts were quantified by both spectrophotometry and spot densitometry, and equal amounts of each mRNA were used for in vitro translation in the presence of L-[ 35 S]methionine. As a first step in the analysis of the translational efficiency of PDE␣ and PDE␤, we determined the approximate linear range of protein synthesis and the relative stability of the corresponding mRNAs in the reticulocyte lysate. In pilot studies, the in vitro translation reactions with A, B5s, and B5l mRNAs were incubated for 20, 40, or 60 min. The RNA from each reaction was then extracted and analyzed by slotblot hybridization. No noticeable mRNA degradation was observed during incubation for up to 60 min. Synthesized proteins were separated by PAGE, and their amount was determined by measuring L-[ 35 S]methionine incorporation. The results obtained showed that at 60 min but not after 20 or 40 min of incubation, protein synthesis reached saturation. Based on these observations, all of the reactions for quantitative analysis were performed for 30 min. mRNA synthesized from constructs A, B5s, and B5l yielded products with the expected molecular masses of 88 and 84 kDa, corresponding to PDE␣ and PDE␤, respectively, but exhibited differential translational efficiency (Fig. 5, Ib). The synthesized PDE␣ was ϳ4.5-fold more abundant than PDE␤ synthesized from the short transcript (B5s) and 5.0-fold more abundant than PDE␤ produced from the long transcript (B5l) (Fig. 5, Ib and Ic). A low level of protein synthesis from B5l was anticipated since this mRNA contains in the 5Ј-UTR an upstream AUG known to have a negative effect on translation. Thus, constructs A, B5s, and B5l enabled us to determine the differences in PDE␣ and PDE␤ protein synthesis efficiency resulting from the contribution of the full-length mRNAs. The input of the coding regions and 5Ј-and 3Ј-UTRs of each of these mRNAs in the differential translation of PDE␣ and PDE␤ was assessed in the experiments described below.
The Role of the 5Ј-UTR in PDE␣ and PDE␤ mRNA Translation-The 5Ј-UTR is known to be an essential factor in determining the rate of translation initiation (20, 21) and consequently the overall protein synthesis efficiency. We hypothesized that if the 5Ј-UTRs of PDE␣ or PDE␤ were solely responsible for the observed differences in efficiency of translation, the exchange of these regions between the subunits would lead to opposite results to those obtained with the original A, B5s, and B5l constructs. That is, the exchange of 5Ј-UTRs would produce approximately 5 times more PDE␤ than PDE␣. However, the chimeric constructs shown in Fig. 5, IIa, gave results different from what was expected. The amount of protein synthesized using construct A5B (containing PDE␣ 5Ј-UTR, PDE␤-coding region, and 3Ј-UTR) was ϳ5.5-fold lower than that produced by the native PDE␣ transcript (construct A) and slightly less than that produced by B5s or B5l (Fig. 5, IIb and IIc and Ia, Ib, and Ic). In contrast, the placement of the PDE␤ short 5Ј-UTR in front of the PDE␣-coding region and 3Ј-UTR (construct B5sA) resulted in greater protein synthesis than that produced by the original A (ϳ2-fold), B5s (ϳ8-fold), or B5l (ϳ10-fold) constructs. Similarly, the placement of the PDE␤ long 5Ј-UTR in front of the PDE␣-coding region and 3Ј-UTR (construct B5lA) also increased protein synthesis above the levels produced by the native mRNAs, although this construct was ϳ1.8-fold less efficient than B5sA with the PDE␤ short 5Ј-UTR. This reduction in efficiency of translation of construct B5lA when compared with B5sA is possibly due to the presence of the AUG upstream of the translation initiation site. These experiments clearly demonstrated the negative impact of the PDE␣ 5Ј-UTR on the rate of protein synthesis.
The Effect of a Non-consensus Translation Initiation Sequence and of the Upstream AUG in the 5Ј-UTR on PDE␤ mRNA Translation-It has been shown that both the absence of a "strong" translation initiation sequence (22)(23)(24) and the presence of an upstream AUG (25)(26)(27) can result in lower protein synthesis efficiency. In addition to the upstream AUG in its long 5Ј-UTR, the PDE␤ mRNA translation initiation sequence is not in absolute agreement with the Kozak consensus sequence (there is an A instead of G at the ϩ4 position). To find out whether these sequences have a negative effect on PDE␤ translation, three new constructs were created by sitedirected mutagenesis (Fig. 5, IIIa). Two of the constructs, B5sM and B5lM, were obtained by replacing the A at the ϩ4 position of PDE␤ mRNA with G, creating in this way the consensus translation initiation sequence. As expected, the amount of protein synthesized from the mutated short transcript was ϳ2.6-fold higher than that obtained from B5s mRNA (Fig. 5, IIIb and IIIc and Ib and Ic). Thus, the absence of a Kozak sequence reduces translational efficiency of PDE␤. However, the A to G mutation at the ϩ4 position in the transcript with the long 5Ј-UTR did not affect its translational rate. This may again be attributed to the presence of the upstream AUG in the long 5Ј-UTR. To test this possibility, we generated the third construct, B5lM2, by mutating the upstream AUG of the long PDE␤ transcript to AAG. Indeed, protein synthesis from this construct was 2.2 times more abundant than that from the wild type B5l transcript (Fig. 5, Ib and Ic and IIIb and IIIc).
The Role of the 3Ј-UTRs in PDE␣ and PDE␤ Translation Efficiency-Along with the 5Ј-UTR, the 3Ј-UTR is emerging as another important mRNA sequence controlling its translation (28 -30). To determine the contribution of the PDE␣ and PDE␤ mRNA 3Ј-UTR on protein synthesis, three chimeras were constructed by exchanging the 3Ј-UTRs of constructs A and B5s or B5l (Fig. 5, IVa). AB3 was translated only 1.3 times more efficiently than construct A. In contrast, the placement of PDE␣ 3Ј-UTR in the B5sA3 construct increased protein production by close to 3-fold (Fig. 5, IVb and IVc and Ib and Ic). However, protein synthesis by B5lA3 construct was similar to that produced by B5l, again suggesting a repressing effect of the upstream AUG in the long PDE␤ 5Ј-UTR. Swapping 3Ј-UTRs between PDE subunits did not reverse their translational efficiency, i.e. protein synthesis from PDE␣ mRNA was greater than that from PDE␤. A comparison of the levels of protein synthesized from all of the constructs described in this study demonstrates that those with the PDE␣-coding region, regardless of which flanking 5Јor 3Ј-UTR they may contain, produce more protein than any construct with the PDE␤-coding region. DISCUSSION As a first step toward understanding the mechanisms of PDE␣ and PDE␤ gene expression, we determined the steadystate levels of PDE␣ and PDE␤ mRNAs and examined the biosynthesis/turnover of the translated PDE␣ and PDE␤ in the developing and fully differentiated mouse retina. Our results indicate that the mRNA for PDE␤ at both 10-and 30-post-natal days is ϳ5-fold more abundant than the mRNA for PDE␣. Despite the fact that the PDE␣:PDE␤ mRNA ratio we observed by quantitative Northern analysis is larger than that observed by Phelan and Bok (31) in 10-day-old mouse retinas using quantitative RT-PCR (this variability could be due to the different techniques used to determine the mRNA levels), both studies as well as results from Y79 cells (32) indicate a higher number of transcripts for PDE␤ than for PDE␣. In contrast, the examination of protein levels showed equimolar expression and similar turnover rates of PDE␣ and PDE␤ subunits. This discordance between PDE␣ and PDE␤ steady-state mRNA and protein levels suggested that photoreceptor cells must have a mechanism for post-transcriptional regulation of PDE expression.
In this study, we were able to conclusively demonstrate that PDE␣ mRNA is translated ϳ5 times more efficiently than its PDE␤ counterpart. Thus, the low level of PDE␣ mRNA found in retina could be counterbalanced by its efficient translation if PDE␣ and PDE␤ mRNAs are translated in photoreceptor cells with a similar efficiency to that we observed in vitro. Our data also point to possible regulation of PDE expression in photoreceptor cells by the feedback mechanism, i.e. if one of the subunits is being translated in excess it could by itself or through interaction with other factors stimulate or inhibit transcription of the less or more abundant subunit mRNA, respectively. This regulatory mechanism would require translocation of transcription factors from the cytoplasm to the nucleus. Alternatively, signals could be delivered from the cytosol to the nucleus by protein kinases that would activate transcription factors by phosphorylation. It has been shown that proteins with a molecular size of Ͻ50 kDa can freely enter the nucleus but that larger proteins must have a nuclear localization sequence (NLS) that is recognized by the members of the importin/ karyopherin superfamily. The complex between importin/ karyopherin and the NLS-containing protein can then be actively transported through the nuclear pore complex (33,34). To check the possibility of PDE subunits (PDE␣, 88 kDa and PDE␤, 84 kDa) directly regulating their own transcription, we analyzed these proteins using Predict NLS: Prediction and Analysis of NLSs, PSORT II: Prediction of Protein Sorting Signals and Localization Sites in Amino Acid Sequences, and SubLoc (Version 1.0): Subcellular Localization Prediction of Eukaryotic Proteins. This analysis predicted no nuclear localization of PDE␣ and PDE␤ and revealed no potential NLS in their sequences, suggesting that the proposed feedback mechanism would require the involvement of other factors regulating their transcription.
After determining the translational efficiency of PDE␣ and PDE␤ mRNAs, we evaluated the roles of their 5Ј-and 3Ј-UTRs known to be involved in controlling protein synthesis as well as that of their coding sequences on this process. Since neither of the 5Ј-UTRs of PDE␣ and PDE␤ mRNAs contains stable secondary structures that can reduce the rate of protein synthesis, we first hypothesized that the presence of an upstream AUG and the absence of a strong initiation sequence in the 5Ј-UTR of PDE␤ mRNA could account for its lower protein synthesis efficiency. Indeed, when the upstream AUG was mutated and the consensus translation initiation sequence in the PDE␤ mRNA was restored, the protein production was increased. However, using several chimeric constructs, we demonstrated that in fact the 5Ј-UTR of PDE␣ leads to lower protein synthesis than the 5Ј-UTR of PDE␤ mRNA. Therefore, the differential translation of the PDE subunits cannot be solely explained based upon the primary or secondary structures of their 5Ј-UTRs.
The results of our studies undoubtedly implicate the involvement of the coding regions in the differential translation of PDE␣ and PDE␤ mRNAs. All of the eight constructs containing the PDE␤-coding region resulted in lower protein expression than the four constructs containing the PDE␣-coding region, regardless of the flanking 5Ј-or 3Ј-UTR they had. Regulation of protein synthesis by the mRNA-coding region is not a widespread phenomenon. One example of this type of regulation is provided by the response of HSP70 to a heat shock, which involves an increase of the elongation rate (35). It has also been shown that some mRNAs contain in the coding region secondary structures that lead to "ribosomal pause" (36). Liebhaber et al. (37) have shown that extensive secondary structures in the coding region suppresses translation to a minimal or to a substantial degree depending on the distance of these structures from the initiation codon. Coding region duplexes in close proximity to the AUG dramatically suppress translation, whereas duplexes further downstream in the coding region have a milder effect on protein synthesis. We have analyzed PDE␣ and PDE␤ mRNAs for the presence of stable secondary structures. The overall ⌬G of PDE␣ mRNA is Ϫ680.7 kcal/mol, and the most stable structure (⌬G ϭ Ϫ17.8 kcal/mol) proximal to the initiation codon is created by the coding region sequences 5Ј-GAGGAGGT-3Ј (position ϩ15 to ϩ22) and 5Ј-CTCCTCCA-3Ј (position ϩ107 to ϩ100). Structures with lower ⌬G (Ϫ20.4 and Ϫ20.3 kcal/mol) are located ϳ900 and 1800 nucleotides from the AUG start codon, respectively. The total free energy of the PDE␤ mRNA is Ϫ593.4 kcal/mol. It has six structures with ⌬Gs ranging from Ϫ19.0 to Ϫ23.5 kcal/mol. One of these structures (⌬G ϭ Ϫ19.0 kcal/mol) is in very close proximity to the AUG start codon and involves sequences 5Ј-GGCACAGC-3Ј (position Ϫ20 to Ϫ13 in the 5Ј-UTR) and 5Ј-CCGTGTCG-3Ј (ϩ164 to ϩ147 in the coding region). It will be interesting to determine the involvement of these predicted structures in the regulation of PDE translation.
In summary, our results indicate that the low level of PDE␣ mRNA found in retina can be compensated by its more efficient translation to achieve equimolar expression with PDE␤. Moreover, our results indicate that the PDE␣-and PDE␤-coding regions are responsible for the differential expression of these subunits.