Binding of Guanylyl Cyclase Activating Protein 1 (GCAP1) to Retinal Guanylyl Cyclase (RetGC1)

Guanylyl cyclase activating protein 1 (GCAP1), after substitution of Ca2+ by Mg2+ in its EF-hands, stimulates photoreceptor guanylyl cyclase, RetGC1, in response to light. We inactivated metal binding in individual EF-hands of GCAP1 tagged with green fluorescent protein to assess their role in GCAP1 binding to RetGC1 in co-transfected HEK293 cells. When expressed alone, GCAP1 was uniformly distributed throughout the cytoplasm and the nuclei of the cells, but when co-expressed with either fluorescently tagged or non-tagged RetGC1, it co-localized with the cyclase in the membranes. The co-localization did not occur when the C-terminal portion of RetGC1, containing its regulatory and catalytic domains, was removed. Mutations that preserved Mg2+ binding in all three metal-binding EF-hands did not affect GCAP1 association with the cyclase in live cells. Locking EF-hand 4 in its apo-conformation, incapable of binding either Ca2+ or Mg2+, had no effect on GCAP1 association with the cyclase. In contrast to EF-hand 4, inactivation of EF-hand 3 reduced the efficiency of the co-localization, and inactivation of EF-hand 2 drastically suppressed GCAP1 binding to the cyclase. These results directly demonstrate that metal binding in EF-hand 2 is crucial for GCAP1 attachment to RetGC1, and that in EF-hand 3 it is less critical, although it enhances the efficiency of the GCAP1 docking on the target enzyme. Metal binding in EF-hand 4 has no role in the primary attachment of GCAP1 to the cyclase, and it only triggers the activator-to-inhibitor functional switch in GCAP1.

enters outer segments of vertebrate photoreceptors through cGMP-gated Na ϩ /Ca 2ϩ channels in the outer segment plasma membrane, is continuously removed from the outer segment by a light-independent Na ϩ /K ϩ , Ca 2ϩ exchanger (for review, see Refs. [5][6][7][8]. In the dark, cGMP keeps a small percentage of the Na ϩ /Ca 2ϩ channels open, and the hydrolysis of cGMP by a light-activated phosphodiesterase, PDE6, generates photoresponses in rods and cones. When light triggers cGMP hydrolysis, it also, through the closure of the channels, lowers the intracellular concentration of Ca 2ϩ from ϳ250 nM in the dark to ϳ25 nM in the light (9 -12). At the same time, free concentrations of Mg 2ϩ in photoreceptors remain near 1 mM, regardless of illumination conditions (13). In response to the lightdependent decrease in free Ca 2ϩ concentrations, GCAPs exchange Ca 2ϩ in their EF-hands for Mg 2ϩ (14,15), which stimulates RetGC and thus prompts re-opening of the cGMPgated channels and accelerates the recovery. Of the four EFhand domains in GCAPs, only three are capable of Mg 2ϩ /Ca 2ϩ exchange, whereas the N-proximal EF-hand 1 domain deviates from the consensus sequence and cannot bind metal ions. Instead, this EF-hand is required for GCAPs interaction with their target enzyme, RetGC (16 -19). We previously described mutations that can inactivate only Ca 2ϩ or both Ca 2ϩ and Mg 2ϩ binding in all three metal-binding EF-hands of GCAP1 (15,20). We have demonstrated that the apo form of GCAP1 does not stimulate RetGC1. We have also found that when neither Ca 2ϩ nor Mg 2ϩ is present in EF-hands 2 and 3, activation of RetGC is suppressed. Inactivation of Ca 2ϩ coordination in EF-hand 4 prevented inhibition of RetGC1 by Ca 2ϩ but had no effect on the cyclase activation by Mg 2ϩ -liganded GCAP1 (15). Our previous data suggested that divalent cation binding in EF-hand 2 and, to a lesser extent, in EF-hand 3, was required for GCAP1 to stay in complex with RetGC1 in either Mg 2ϩ -or Ca 2ϩ -liganded forms. Here, we evaluated this hypothesis by monitoring the binding of a green fluorescent protein (GFP)tagged GCAP1 to functional RetGC1 in cultured cells. We demonstrate that elimination of both Ca 2ϩ and Mg 2ϩ binding in EF-hand 2 suppresses compartmentalization of GCAP1 with RetGC1, whereas inactivation of EF-hand 3 reduces the efficiency of the GCAP1 attachment to RetGC1, but does not prevent it. We also demonstrate that cation binding in EF-hand 4 has no role in GCAP1 association with the cyclase.
pQBI25-fN3 vector. The GCAP1 cDNA was inserted into the BamHI/EcoRI sites, in-frame with the Superglo GFP-coding sequence of the vector and downstream of the CMV promoter, as follows. The GCAP1 sequence was PCR-amplified by Pfu polymerase (Stratagene) from a DNA clone for transgenic expression of wild type bovine GCAP1 in mouse rods (9) using a forward primer, 5Ј-GGGGGGATCCCTCGAGAGCCGCA-GCCATGGGGAACATTATGAGCGGT-3Ј, and a reverse primer, 5Ј-CCCCCCGAATTCGCCGTCGGCCTCCGCGGC-CTCC-3Ј, thus adding the Kozak motif and the required restriction sites to the GCAP1 cDNA in the resultant plasmid, GCAP1-GFPpQBI25fN3. Expression constructs for EF-hand mutants were made by substitution of the BlpI/SfiI fragment in the GCAP1-GFPpQBI25fN3 construct with the corresponding fragment from cDNA coding for the EF-hand mutations described previously (15,20).
To produce GFP-tagged GCAP1 and its mutants in E. coli, the GCAP1-GFP cDNA fragment was inserted into the pET11d vector (Novagen/Calbiochem) in two steps. We first PCR amplified with Pfu polymerase two fragments from the GCAP1-GFP pQBI25fN3, and the expression pET11d vector for wild type GCAP1 (21,22). The two fragments were amplified separately, using two pairs of primers: 5Ј-GGGGCTA-GCGGCAGGGAAAC-3Ј and 5Ј-AGTGTTGGCCAGGGAAC-AGGCAGT-3Ј (pair 1), and 5Ј-ACTGCCTGTTCCCTGGCC-AACACT-3Ј and 5Ј-GAGAGAGGATCCTCAGTTAGTCAA-TCGATGTTGTACAGTTCATCCA-3Ј (pair 2), respectively. The two isolated fragments were spliced together by a second round of PCR, using the first and the last primers only, thus resulting in the BamHI site being placed near the 3Ј-end of the fusion cDNA. The resultant fusion cDNA was inserted into the SacII/BamHI sites of the modified pET11d vector harboring bovine GCAP1 cDNA, thus placing the GFP sequence in-frame with the GCAP1, followed by the stop codon and the transcription termination site of the pET11d vector. Wild type GCAP1 and its mutants all had a recognition site for a yeast N-myristoyl transferase (23). Recombinant GCAP1-GFP and its mutants were expressed in Escherichia coli and purified as previously described (20).
Recombinant dsRed-tagged RetGC1-Recombinant RetGC1 was expressed in HEK293 cells from a modified pRCCMV vector (Stratagene), as previously described (24). Fluorescently labeled RetGC1 was produced by inserting in RetGC1 cDNA a DNA fragment coding for a monomeric red fluorescent protein, dsRed (Clontech), PCR-amplified with the BstEII/KpnI restriction sites at the ends. The internal BstEII restriction site in the dsRed cDNA, GGTGACC, was inactivated via a silent substitution, GGTGACT, using PCR splicing by overlap extension. The KpnI site at the 3Ј end of the dsRed cDNA fragment was preceded by an additional nucleotide, which truncated the cyclase amino acid sequence by shifting the reading frame after the dsRed coding fragment and thus eliminated the entire intracellular domain of the cyclase. We used this construct, ⌬dsRed-RetGC1, as a control for the specificity in co-localization experiments. The reading frame in the ⌬dsRed-RetGC1 was shifted back to normal by restriction digest with KpnI, followed by blunting the ends by T4 polymerase and self-ligation. The resultant construct, dsRed-RetGC1, was used to express func-tional fluorescently tagged RetGC1 in HEK293 cells. All constructs were verified by automated DNA sequencing.
Transfection of RetGC1 and GCAP1 into HEK293 Cells-HEK293 cells were grown at 37°C, 5% CO 2 , in high glucose Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% fetal bovine serum (Invitrogen). To express RetGC1 for the functional assay in vitro, HEK293 cells were transfected with 40 g/100-mm culture dish of pRCCMV plasmid containing wild type RetGC1 or dsRed-RetGC1 using Ca 2ϩ -phosphate method (a Promega Profection protocol), and the membranes were harvested as previously described (24). We also developed a stable neomycin-resistant line expressing GCAP1-GFP selected in the presence of Geneticin (Invitrogen).
Co-expression of RetGC1 and GCAP1 in HEK293 Cells and Confocal Laser Scanning Microscopy-Cells were grown in standard glass coverslip chambers (four 2-cm 2 chambers per slide) in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, and were transfected with a mixture of expression constructs using the Ca 2ϩ -phosphate method. Routinely, a mixture of 3 g of pRCCMV plasmid containing wild type RetGC1 cDNA or dsRed-RetGC1 and about 0.02 g of the GCAP1-GFP pQBI25fN3 plasmid was used per chamber. In 24 -32 h, the live cells were either viewed directly or fixed for 15 min with freshly prepared 4% paraformaldehyde in a standard Tris-buffered saline (TBS) at room temperature for subsequent anti-RetGC1 staining with an antibody. The cells were gently washed twice with 1 ml of TBS, blocked by 1% goat serum, 1% bovine serum albumin in TBS containing 0.05% Triton X-100, for 30 min, incubated with anti-RetGC1 antibody in TBS, 0.3% Triton X-100 for 30 min, washed three times with the Triton X-100/TBS for 5 min, and then incubated with Alexa Fluor 568-conjugated goat anti-rabbit antibody (Invitrogen) for 30 min. The cells were then washed twice with TBS and covered with Vectashield solution (Vector Laboratories). Where indicated, the DNA in nuclei of the cells was counterstained for 10 min with 1 mM TO-PRO-3 iodide stain (Invitrogen) containing 20 g/ml RNase A (added to the first wash following the incubation with the secondary antibody), and then washed as described above. Where indicated, live cells expressing GCAP1-GFP were treated for 30 min with a fluorescent ER-Tracker TM Red marker (E34250, Invitrogen) to counterstain the endoplasmic reticulum or with a CellMask TM Deep Red stain (C10046, Invitrogen) to counterstain the plasma membrane, all according to the manufacturer's protocols.
The cells were viewed using an inverted Olympus IX81 microscope/FV1000 Spectral laser confocal system, and images were collected and analyzed using Olympus FluoView FV10-ASW software. Fluorescence from different spectral markers in the same sample was recorded in a sequential mode, at 40 s/pixel, typically averaged from three to four repeats, and where indicated, were superimposed on a confocal transmitted light differential interference contrast image of the same cells. The far-red fluorescence emitted from the TO-PRO-3 iodide or the CellMask Deep Red plasma membrane stain was pseudocolored blue or red, respectively.
Anti-RetGC1 Antibody-Anti-RetGC antibody was produced in a rabbit against recombinant fragments of a human RetGC1, Met 747 -Ser 1052 (antibody GC1Cat) or Arg 540 -Asn 815 (antibody GC1KHD), expressed in E. coli from pET15b vector (Novagen/ Calbiochem). The IgG fraction was purified using Protein A-Sepharose (GE Healthcare). The antibodies specifically stained the ϳ115-kDa band on immunoblots of HEK293 cells transfected with RetGC1 expressing constructs, but not from non-transfected control cells, and specifically stained membranes in RetGC1transfected, but not in non-transfected cells.
Immunoblotting-The HEK293 cells were washed in TBS, dissolved in a Laemmli SDS-PAGE sample buffer, and aliquots were separated in 4 -12% PAGE (Invitrogen). Following elctroblotting on Immobilon P membrane (Millipore), proteins were probed with the rabbit polyclonal anti-RetGC1 and anti-GCAP1 antibodies and developed using a Pierce Femto Supersignal luminescent peroxidase substrate, according to the manufacturer's protocol. The signal intensity on x-ray film was quantified by densitometry as previously described (9).
Statistics-In co-localization experiments, five groups of GCAP1-GFP (wild type and four mutants) expressed in HEK293 cells were compared by one-way analysis of variance and post hoc processed by Bonferroni all-pairs comparison test using a GraphPad QickCalcs calculator (GraphPad Software, Inc., San Diego, CA) and Synergy Kaleidagraph software. The same software was used for a non-paired t test to compare the data for the D64N mutant versus the wild GCAP1-GFP type expressed in the absence of the cyclase.

RESULTS
Study of direct binding of GCAPs to the cyclase is very challenging. In contrast to some other proteins, direct binding of GCAPs to RetGC could not be reliably assessed using the membrane binding in vitro assay due to the effect of nonspecific interactions (16,25,26), and a detergent-solubilized RetGC1 is not sensitive to GCAPs (27). Therefore, we decided to tag GCAP1 and its mutants with the enhanced GFP and test their co-localization with the target enzyme in live cells expressing either non-modified or fluorescently tagged dsRed-RetGC1.
The Activity of GFP-tagged GCAP1 and dsRed-tagged RetGC1-It was important for the entire line of the study to establish whether or not the fluorescent tags can affect the activity of GCAP1 or RetGC1. We found that the main regulatory properties of GCAP1 were unaltered by its fusion with GFP ( Fig. 1, A and B), nor were altered by its tag the properties of the dsRed-tagged RetGC1 (Fig. 1, C and D). There was no difference between the wild type RetGC1 activation by either nontagged or the GFP-tagged GCAP1 or dsRed-RetGC1 activation by the non-tagged or GFP-tagged GCAP1, except for a minor FIGURE 1. Properties of dsRed-RetGC1 and GCAP1-GFP. A, activation of non-tagged RetGC1 in HEK293 cell membranes by purified non-tagged (Ⅺ) or GFP-tagged (E) GCAP1; B, Ca 2ϩ sensitivity of the non-tagged RetGC1 reconstituted with the non-tagged or GFP-tagged GCAP1; C, activation of dsRed-RetGC1 in HEK293 cell membranes by the purified non-tagged or GFP-tagged GCAP1; D, Ca 2ϩ sensitivity of the dsRed-RetGC1 in HEK293 cell membranes reconstituted with the non-tagged or GFP-tagged GCAP1. RetGC1 activation by GCAP1 was measured in the presence of 2 mM EGTA, 1 mM free Mg 2ϩ , and increasing concentrations of wild type GCAP1 or GCAP1-GFP. The data were fitted by the equation, A is the activity of RetGC1 in the assay, A max is the maximal activity of RetGC1, [GCAP] is the concentration of GCAP1, K1 ⁄2 is the concentration of GCAP1 required for half-maximal activation of RetGC1, n is the cooperativity coefficient. Ca 2ϩ sensitivity of the recombinant RetGC1 regulation by GCAP1 was assayed at various free Ca 2ϩ concentrations in the presence of 10 M recombinant GCAP1. The data were fitted as fractional activity of RetGC1 by the function, ⁄2 is the free Ca 2ϩ concentration required for half-maximal inhibition of RetGC1, n is the cooperativity coefficient. The values from the fit are summarized in supplemental Table S1. For other conditions of the assay see "Experimental Procedures." decrease in the apparent affinity for dsRed-RetGC1 activation by both the tagged and non-tagged GCAP1. Ca 2ϩ sensitivity of both forms of RetGC1 reconstituted with either tagged or nontagged GCAP1 was the same (see the supplemental Table S1 for more detail).
Properties of GFP-tagged GCAP1 EF-hand Mutants-In our previous study, we identified mutations in individual EF-hands that disabled either Ca 2ϩ binding or both Ca 2ϩ and Mg 2ϩ binding altogether (20). Disabling of metal binding in different EFhands had profoundly different effects on RetGC1 activation and inhibition (15): inactivation of only Ca 2ϩ binding in all three EF-hands by E75Q/E111Q/E155Q substitutions prevented the cyclase inhibition by Ca 2ϩ , but did not affect activation of RetGC1 by the triple mutant in the presence of Mg 2ϩ . A complete inactivation of both Ca 2ϩ and Mg 2ϩ binding in EFhand 4 by a double substitution, D144N/D148G, did not affect it either. However, complete inactivation of Ca 2ϩ /Mg 2ϩ binding in EF-hands 2 and 3 by D64N and D100N/D102G mutations, respectively, suppressed activation of the cyclase. The effect was most dramatic in the case of EF-hand 2 (15). We have verified that the corresponding GFP-tagged GCAP1 mutants displayed virtually the same properties as their non-tagged counterparts ( Fig. 2 and supplemental Table S2): (a) disabling the EF-hand 2 drastically increased the K1 ⁄ 2 of GCAP1 for RetGC1; (b) inactivation of EF-hand 3 reduced the apparent affinity of GCAP1 for RetGC1, however, to a lesser extent than EF-hand 2; (c) prevention of both Ca 2ϩ and Mg 2ϩ binding in EF-hand 4 had no effect on RetGC1 activation. Therefore, it was justified to use the fluorescent protein-tagged GCAP1 mutants and RetGC1 in co-expression and co-localization analyses to directly assess their binding to each other.
Localization of GCAP1 and RetGC1 in HEK293 Cells-GCAPs have the ability to weakly bind to lipid membranes (16,25,26,28), but when expressed in HEK293 cells, GCAP1 was uniformly distributed throughout the cell, including the nucleus ( Fig. 3A and supplemental Fig. S1A). Only the nucleoli appeared to have lower density of the fluorescent GCAP1 in the nucleus, and only vacuoles were devoid of fluorescence in the cytoplasmic portion of the cells. This pattern remained in all GCAP1-GFP expressing cells, regardless of the overall intensity of the fluorescence (i.e. the levels of GCAP1 expression). We also found no difference in GCAP1-GFP distribution between the cells transiently expressing GCAP1-GFP versus a stable line of GCAP1-GFP expressing cell (data not shown). The uniform cellular distribution of GCAP1 drastically changed when the cells expressed both GCAP1-GFP and RetGC1 (Fig. 3A, panels b and c): GCAP1 fluorescence was depleted from the nuclei and was only observed in the cytoplasm of the cells, where it demonstrated a membrane association pattern, indistinguishable from that of RetGC1 (Fig. 3A, panels a-d, and supplemental Fig. S1B). The intrinsic fluorescence of GCAP1-GFP and immunofluorescence of RetGC1 both displayed a typical shape of "donuts" and "tennis rockets": the empty nucleus was surrounded by well defined membrane fluorescence, extended along the endoplasmic reticulum (ER) (Fig. 3, B and C, and supplemental Fig. S2). It should be noted that although RetGC1 expressed in HEK293 cells was active and responsive to GCAP in vitro (Figs. 1 and 2), most of the cyclase was localized to the membranes of the ER (Fig. 3A, panel c and supplemental Fig. S2), and so was GCAP1-GFP, co-expressed with the cyclase. Only a minor portion of the GCAP1 fluorescence appeared to coincide with the marker for the plasma membrane, whereas most of it followed the profile matching that of the ER marker (Fig. 3, B and C).
We have also observed co-localization of the GFP-tagged wild type GCAP1 with RetGC1 in non-fixed live cells expressing dsRed-tagged RetGC1 (Fig. 4A, panels a-c). Both fluorescent tags in the non-fixed cells co-localized in the same pattern that was observed in the fixed cells immunostained for RetGC1 (Fig. 3). That argued that co-localization in Fig. 3 was not an artifact of fixation. The distribution of both GCAP1-GFP and dsRed-RetGC1 had the characteristic absence of the signal in the nucleus and a sharp increase in the fluorescence of the surrounding ER membranes (Fig. 4A, panels a-c, the cell marked with an arrow). Such a pattern was not observed in control experiments when we co-expressed the GFP-tagged GCAP1 with the extracellular portion of the cyclase, incapable of binding GCAP1 (29 -32). The fluorescently tagged fragment, ⌬dsRed-RetGC1, retained the leader peptide sequence for guiding it to the ER (Fig. 4A, panel e), but the distribution of GCAP1-GFP and ⌬dsRed-RetGC1 was drastically different (panels d-f): the GCAP1-GFP fluorescence was spread throughout the cell, including the nucleus (same as GCAP1 expressed in the absence of RetGC1 in Fig. 3), and the cyclase fragment was only detected in the ER. Fig. 4B, panels a-d, further demonstrates that the diffuse distribution of GCAP1-GFP co-expressed with the truncated cyclase was not due to much  Table S2. higher GCAP1 expression relative to the ⌬dsRed-RetGC1 fragment. We could not use anti-retGC1 antibody for quantitative immunoblotting, because the truncated cyclase lacks the intracellular domains, against which the anti-RetGC1 antibodies were raised. Yet, we took advantage of the same fluorescent tag being present in both the truncated fragment and the non-truncated cyclase. We used the same excitation laser intensity and the photomultiplier gain settings to compare the fluorescence profiles of GCAP1-GFP co-expressed with either dsRed-RetGC1 or ⌬dsRed-RetGC1 in the neighboring chambers on the same coverslip. In that case, the relative intensities of the red versus green fluorescence in one group of the cells could be directly compared with the intensities of the corresponding fluorescent tags in the other group (Fig. 4B, panels b and d). The representative results shown in this figure illustrate that GCAP1-GFP failed to colocalize with the ⌬dsRed-RetGC1 fragment, even when GCAP1-GFP levels were lower, and the levels of ⌬dsRed-RetGC1 were higher than in the cells expressing the nontruncated cyclase. Therefore, the characteristic pattern of GCAP1 co-localization with RetGC1 in Figs. 3 and 4 could only be attributed to a direct binding of GCAP1 to the "intracellular" domains of RetGC1, missing in the truncated cyclase.
Differential Binding of GCAP1 Mutants to RetGC1 in HEK293 Cells (Fig. 5)-We were unable to modulate the binding of GCAP1 to RetGC in live cells by removing Mg 2ϩ from the cells by chelating agents, because that affected their shape and adhesion to the plate and thus hampered the analysis. Instead, we used the GFP-tagged mutants of GCAP1, which either lost their ability to bind Ca 2ϩ , yet retained their high-affinity Mg 2ϩ binding, or could bind neither of the two ions (15,20). All GCAP1-GFP mutants, expressed in the absence of the cyclase, displayed a diffuse pattern, the same as the wild type GCAP1-GFP in Fig. 3A (data not shown).
Consistent with its unchanged ability to activate RetGC1 in the presence of Mg 2ϩ (Fig. 2), GFP-tagged E75Q/E111Q/ E155Q GCAP1, which does not bind Ca 2ϩ , but binds Mg 2ϩ in all three metal binding EF-hands (15,20), co-localized with RetGC1, just like the wild type (Fig. 5A). The green fluorescence of GCAP1 matched immunofluorescence of RetGC1 in the ER and did not overlap with the TO-PRO-3-stained nuclei. We used an excess of the cyclase-expressing vector over that of each GCAP1 mutant (this will also be discussed in detail further below). Therefore, the cells, which would express GCAP1, but had no detectable expression of RetGC1, were present at low frequency. However, in Fig. 5A, for comparison with the other marked cells, there is shown a cell (cell number 3) that did not express RetGC1. That cell had a uniform GCAP1-GFP distribution, in sharp contrast to the typical "donuts and tennis rockets" pattern of the surrounding cells. Co-localization of the E75Q/E111Q/E155Q mutant with RetGC1 in HEK293 cells was consistent with the ability of the Mg 2ϩ -liganded GCAP1, the physiological form of GCAP1 in light-adapted photoreceptor cells (20), to activate the cyclase (Fig. 2). Inactivation of Mg 2ϩ binding in the individual EF-hands had profoundly different effects on the association of the GCAP1 mutants with the cyclase, depending on which particular EF-hand was inactivated.
In contrast to the wild type GCAP1, distribution of the D64N mutant in the presence of RetGC1 remained rather diffuse (Fig.  5B). The intensity of the GCAP1-GFP fluorescence was only slightly elevated outside the nuclei (panels a and c), regardless of the overall intensity of the GCAP-GFP expressed in the cells (compare cells 1-4 in panel a). In all cells of this group, the GCAP1 fluorescence was nearly the same in the nuclei as in the cytoplasm, strikingly different from the localization of RetGC1  (panels b and c).
When Mg 2ϩ binding in EF-hand 3 was abolished (D100N/ D102G) (Fig. 5C), co-localization of the mutant with the cyclase resembled the wild type, rather than the D64N mutant, except that the relative brightness of the GCAP1-GFP fluorescence in the ER versus the nucleus was typically lower than in wild type.
Although not as striking as in wild type, its pattern was still different from the diffuse fluorescence in the cells that did not express RetGC1 (compare cell 1 with 2). Complete inactivation of Ca 2ϩ and Mg 2ϩ binding in EF-hand 4 had no detectable effect on co-localization of GCAP1 with RetGC1 (Fig. 5D): the D144N/D148G GCAP1 fluorescence was depleted from the nucleus and co-localized with RetGC1 immunofluorescence in membrane structures, the same as in the wild type GCAP1-GFP.
Among all the mutants that we tested in Fig. 5, D64N showed the weakest co-localization with the cyclase. This pattern, however, was not due to an excessive level of the D64N expression compared with the wild type GCAP1. As an example, in Fig. 6 we compared the profiles of the D64N and the wild type GCAP1 versus that of RetGC1 directly, by their relative fluorescence intensities (Fig. 6). The overall GFP fluorescence integrated for the entire cell was almost twice as high as in the wild type GCAP1 than in the D64N mutant (Fig. 6, A and B, panels a), and the overall immunofluorescence of RetGC1 was the same in both cells (Fig. 6, A  and B, panels b). Yet, the distribution of the wild type GCAP1 exactly matched that of the RetGC1, whereas the distribution of the D64N remained much more diffuse (compare the profiles in Fig. 6, A and B, panels d).
To quantify the patterns of the GFP-GCAP1 co-localization with RetGC1, we measured the ratio between the maximal GFP fluorescence intensity in the whole cell versus its maximal fluorescence intensity inside the nucleus and plotted the data in Fig. 7A. We excluded from the analysis cells that were collapsing, dividing, detaching from the glass, or had exceedingly bright fluorescence of the entire cell, beyond the linear range of the photomultiplier. The distribution ratio varied widely between the cells within each group. That was not surprising, given the expected variability in the levels of GCAP1 and RetGC1 co-expression between different cells. However, after a sufficiently large number of randomly quantified cells, we were able to evaluate the statistical difference between the four GCAP1 mutants and the wild type by one-way analysis of variance (also see supplemental Fig. S3), post hoc processed by Bonferroni test for all pairs comparison. There was a significant difference from the wild type both for the D64N mutant and the D100N/D102G mutant (Cl ϭ 99%, p Ͻ 0.0001), as well as  RetGC1 (a-b) or ⌬dsRed-RetGC1 (c-d); a and b, representative fluorescence intensity profiles recorded from the two cells expressing GCAP1-GFP (green) and dsRed-RetGC1 (red), shown in the two left panels, along the lines superimposed on the merged fluorescence images shown as the insets in the right two panels; c and d, two fluorescence intensity profiles recorded from the cell expressing GCAP1-GFP (green) and ⌬dsRed-RetGC1 (red) shown in the two left panels, along the lines superimposed on the merged fluorescence images, shown as the insets in the right two panels. The images in panels a and c were recorded in the same experiment, at the same laser excitation and photomultiplier gain settings, from the cells in the neighboring chambers of the same coverslip; bar, 10 m. between the two mutants themselves (p ϭ 0.0003). The fluorescence distribution ratio (mean Ϯ S.E.) for the EF-hand 3 mutant fell nearly 2-fold (6. 6 Ϯ 0.4, n ϭ 115), and for the EF-hand 2 mutant at least 6-fold (2.2 Ϯ 0.2, n ϭ 59) compared with the wild type (13.5 Ϯ 1.6, n ϭ 65). The difference between the wild type and E75Q/ E111Q/E155Q (11.9 Ϯ 1.0, n ϭ 62) or the wild type and D144N/D148G (13.7 Ϯ 1.1, n ϭ 56) mutants was not significant (Fig. 7B). The D64N GCAP1 fluorescence distribution in the presence of RetGC1 (2.2 Ϯ 0.2, n ϭ 59) was also very close to the wild type expressed in the absence of RetGC1 (1.01 Ϯ 0.03, n ϭ 41, Fig.  7, A and B), yet the small difference between the two remained statistically significant based on a nonpaired t test (p Ͻ 0.05).
The more diffuse pattern of the D64N and D100N/D102G mutants could not result from saturation of the cyclase with the mutant GCAP1. In addition to the comparison of the fluorescence profiles in Fig. 6, such a possibility was also ruled out by the results shown in Fig. 7, c and d. The ER:nucleus GCAP1-GFP fluorescence ratio rises as the GCAP1-GFP expression levels drops relative to that of RetGC1 (Fig. 7C). To prevent RetGC1 from saturation by GCAP1-GFP, we used in our experiments the GCAP1-GFP:RetGC1 DNA ratio of 1:150 for the wild type and all GCAP1 mutants. The D100N/D102G mutant expression efficacy had to be at least 8 -16 times higher than the wild type to match the distribution shown in Fig.  7, A and B (the D64N mutant would have to be expressed at even higher levels, still). Yet, neither the D64N nor the D100N/D102G mutants were more abundant relative to the cyclase than the wild type GCAP1-GFP (Fig. 7D). We also estimated the average density of expression of RetGC1 and GCAP1-GFP in cell culture by immunoblot (Fig. 7D). Assuming that the immunoreactivity of the recombinant RetGC1 KHD fragment used for calibration was the same as the native cyclase (purified native human RetGC1 is unavailable), there was about 1.7, 0.7, 1.2, and 1.1 fmol of GCAP1 cm Ϫ2 of the wild type, D64N, D100N/D102G, and D144N/ D148G mutants, respectively, versus about 37 fmol of RetGC1 cm Ϫ2 . This analysis does not reflect cell to cell variations responsible for the data scattering in Fig. 7A, but it decisively argues that neither D64N nor the D144N/D148G mutants were expressed at higher levels than RetGC1 or the wild type GCAP1-GFP.

DISCUSSION
RetGC regulation by GCAPs is essential for recovery of photoreceptors from excitation. When the free Ca 2ϩ concentra-tions fall in the light, and Ca 2ϩ becomes replaced by Mg 2ϩ in EF-hands of GCAPs, they acquire their activator conformation and accelerate the cyclase (7,12,15,33). The three-dimensional structure of GCAPs in their inhibitory, Ca 2ϩ -loaded, state has been established (34 -36), but mechanistic details of the cyclase activation in the light by the Mg 2ϩ -liganded GCAPs remain elusive. It has also been established that both the activator and inhibitor forms of GCAPs interact with the cyclase in vitro and in vivo (24,30,(37)(38)(39), yet the mechanism, through which GCAPs change between their two functional states in response to the Ca 2ϩ /Mg 2ϩ exchange, remains unclear. Both processes must involve docking of GCAP on the cyclase (30) and a conformational change that likely translates into the cyclase dimerization and activation (40,41).
The replacement of Mg 2ϩ by Ca 2ϩ in the three metal-binding EF-hands, which occurs in rods and cones in the dark, switches the cyclase off, but does not prevent its interaction with the cyclase (14,15,20). Our previous study (15,20) indicated that binding of Mg 2ϩ in the light or Ca 2ϩ in the dark in two of the three metal ligand-binding EF-hands was likely required to maintain GCAP1 in complex with RetGC1. We also suggested that EF-hand 4 was required for turning RetGC1 off upon binding of Ca 2ϩ , rather than for the activation of cyclase in the light. Inactivation of Mg 2ϩ binding in EF-hands 2 and 3 drastically increases GCAP1 K1 ⁄ 2 for activation of RetGC1, measured in the conditions that mimic light-adapted rods (Ref. 15 and Fig. 2). Although other potential explanations for this phenomenon could not be ruled out at the time of the original observation, we hypothesized that the divalent cation binding in EFhand 2 and, to a lesser extent, in EF-hand 3, but not in EF-hand 4, was responsible for the docking of GCAP1 on the cyclase (15). Here, we present direct evidence in support of that hypothesis by studying the effects of the mutations in EF-hands on GCAP1 on its association with RetGC1 using live cultured cells.
Co-localization of GCAP1 with RetGC1 in vivo can be reliably assessed in a semi-quantitative fashion using co-transfection of HEK293 cells with fluorescently tagged GCAP1 and RetGC1. When RetGC1 is expressed in excess over GCAP1, this method reveals a clear co-localization pattern of RetGC1bound GCAP1 (Figs. 3 and 7).
The experiments, described in Figs. 3 and 4, demonstrate that GCAP1 changes its cellular distribution, from soluble to membrane-associated, when it is co-expressed with the cyclase. The robust shift of the GCAP1-GFP fluorescence from the nucleus to the ER is easily detectable by the confocal microscopy in either fixed cells stained with anti-RetGC1 antibody or in live cells expressing dsRed-labeled RetGC1. Importantly, co-localization of the fluorescently tagged GCAP1 with RetGC1 occurs only if the intracellular domain of the cyclase is present (Fig. 4), because this portion of the enzyme is required for its regulation by GCAPs (29 -32). The main disadvantage of using the tagged cyclase was a relatively weak fluorescence intensity of the dsRed tag. Therefore, we preferred to use the antibody staining for RetGC1 in most of the experiments.
In a standard Dulbecco's modified Eagle's cell culture medium, supplemented with 10% serum, HEK293 cells maintain relatively low (ϳ10 Ϫ7 M) free Ca 2ϩ , but have at least 0.7 mM free Mg 2ϩ in the cytoplasm (42,43), which resembles the conditions in photoreceptor cells exposed to light (10 -13). The free Mg 2ϩ concentration in HEK293 is therefore sufficient to maintain a cation-loaded state of GCAP1 (14,15,20). That allows GCAP1 to bind to RetGC1 in membranes of HEK293 cells when both proteins are co-expressed (Figs. [3][4][5]. Consistent with the high intracellular Mg 2ϩ concentrations in HEK293 cells, inactivation of Ca 2ϩ coordination in all three metal-binding EF-hands does not affect its co-localization with the cyclase, as long as Mg 2ϩ binding is preserved (E75Q/E111Q/E155Q mutant in Figs. 5 and 7).
The role of the individual EF-hands in regulation of RetGC1 activity is shown in Fig. 8). The D64N mutation in EF-hand 2, which blocks not only Ca 2ϩ , but also Mg 2ϩ coordination, ham- B, average distribution ratios (mean Ϯ S.E.) for the wild type and mutants of GCAP1-GFP expressed in the absence or presence of RetGC1, as indicated; *, values for the D64N and D100N/D102G were statistically different from the wild type (see "Results"). C, the effect of GCAP1-GFP DNA dilution on GCAP1-GFP and RetGC1 expression and co-localization in transfected cells. RetGC1 and the GCAP1-GFP DNA transfection mixture contained about 3 g of RetGC1 expression construct and various concentrations of the GCAP1-GFP expression construct to yield the required molar ratio, as indicated. The cell:nucleus fluorescence intensity ratio for GCAP1-GFP (f) was plotted as mean average Ϯ S.E. from the data points presented in supplemental Fig. S4. After the microscopy, the cells for each DNA ratio were harvested and analyzed by SDS-PAGE/immunoblot, probed with anti-RetGC1 or anti-GCAP1 antibody (shown above the plot); the retGC1 signal (E) in each case was normalized to the first dilution point (DNA ratio of 1:10), and so was the GCAP1-GFP signal (Ⅺ), accordingly. D, panel a, expression of RetGC1 (top) and GCAP1-GFP mutants (bottom) at the GCAP1-GFP:RetGC1 DNA ratio of 1:150, in the corresponding cell lysates were compared by immunoblotting; b, calibration for quantitation of retGC1 (top) and GCAP1-GFP (bottom) in cell lysates by immunoblotting; purified recombinant 31-kDa His-tagged fragment of the 115-kDa retGC1 and GCAP1-GFP expressed in E. coli (shown as insets in the corresponding plots) were used as standards for calibration. The sample of the cell lysate is shown on the right from each calibration series. The quasilinear intervals of calibration in both plots were used for calculating protein content, femtomole per square cm, in the coverslip chambers for the cytological analysis in panel A.
pers co-localization of GCAP1 and RetGC1 (Figs. 6 and 7). This result directly supports our hypothesis (15,20) that EF-hand 2, together with a non-cation binding EF-hand 1, creates the GCAP1 structure that fits into the docking site on the cyclase. The role of the metal ligand binding in EF-hand 2 is to maintain the proper conformation of this part of the GCAP1 molecule, rather than to provide a switch between the activator and inhibitor conformations of GCAP1.
The EF-hand 3 strongly contributes to the maximal level of RetGC1 activation (Fig. 2) and Ca 2ϩ sensitivity (15,16,20). In addition to that, cation binding in this EF-hand also contributes to the attachment of GCAP1 to the cyclase (Figs. 5 and 7). It contributes less than EF-hand 2, and not necessarily through a direct interaction with the cyclase. For example, ligand binding in EF-hand 3 may affect EF-hand 2, due to their proximity (36).
In contrast to the other EF-hands, metal ligand binding in EF-hand 4 has no major effect on GCAP1 association with the cyclase. Binding of Mg 2ϩ in this EF-hand is involved in neither the docking process (Figs. 5 and 7) nor the RetGC1 activation in the light (15,20). This EF-hand needs metal ligand binding only to function as a negative regulator of RetGC1 activity, and only in response to the binding of Ca 2ϩ in the dark (4,15,20).