PRIORITY PAPER Dimerization of mammalian adenylate cyclases Functional, biochemical and ¯uorescence resonance energy transfer (FRET) studies Chen Gu 1 , James J. Cali 2 and Dermot M. F. Cooper 1,3 1 Neuroscience Program University of Colorado Health Sciences Center, Denver, CO, USA; 2 Promega Corp., Madison, WI, USA; 3 Department of Pharmacology, University of Colorado Health Sciences Center, Denver, CO, USA 1 Mammalian adenylate cyclases are predicted to possess complex topologies, comprising two cassettes of six trans- membrane-spanning motifs followed by a cytosolic, catalytic ATP-binding domain. Recent studies have begun to provide insights on the tertiary assembly of these proteins; crystal- lographic analysis has revealed that the two cytosolic domains dimerize to form a catalytic core, while more recent biochemical and cell biological analysis shows that the t wo transmembrane c assettes also associate to facilitate the functional assembly and tracking of the enzyme. The older literature had suggest ed that adenyla te c yclases m ight form higher order aggregates, although the methods used did not necessarily provide convincing evidence of biologically relevant events. In the present study, we have pursued this question by a variety of approaches, including rescue or suppression of function by variously modi®ed molecules, coimmunoprecipitation and ¯uorescence resonance energy transfer (FRET) analysis between molecules in living cells. The results strongly suggest that adenylate cyclases dimerize (or oligomerize) via their hydrophobic domains. It is speculated that this divalent property may allow adenylate cyclases to participate in multimeric signaling assemblies. Keywords: adenylate cyclase; dimerization; ¯uorescence resonance energy transfer; green ¯uorescent protein; immunoprecipitation. The interjection of stimulatory and inhibitory G-protein modules between receptors and effector increased the complexity of the a denylate cyclase signaling system, while at the same time greatly expanding the perceived, regulatory responsiveness of these systems [1]. Coincident with the discovery of an increased number of signaling components, Rodbell and colleagues proposed that these elements occurred in higher order a ssemblies t han a simple mono- meric a rrangement of receptor, G protein and effector. Using radiation inactivation analysis, Schlegel et al.pro- posed that adenylate cyclase existed in dynamic, multimeric protein arrays of receptors, G proteins and ade nylate cyclases [2,3]. Independent, hydrodynamic analyses of detergent-solubilized adenylate c yclase preparations also indicated molecular m asses of a bout 2 20 kDa for the catalytic units [4±7], which, given the minimal protein molecular masses of  120 kDa, again sugge sted a higher order assembly of adenylate cyclases. Mammalian adenylate cyclases are in the family of ATP-binding cassette (ABC) transporters and share their overall structure [8] which, by analogy, further raises the possibility that they might multimerize. Many members of this family, such as the transporters for glutamate [9], glucose [10] and serotonin [11] are oligomeric. These proteins can f orm more complex, heterooligomeric structures with more elaborate functions. For instance, the cystic ®brosis transmembrane conduct- ance regulator (CFTR), forms a dynamic macromolecular complex, in which a PDZ domain-containing protein (CAP70) facilitates CFTR±CFTR interaction to potentiate chloride channel activity [12]. Another member of this superfamily, the sulfonylurea receptor (SUR) associates with inward ly rectifying K + (K ir ) channel subunits to form ATP-sensitive K + channel complexes, which contain four subunits each of SURs and K ir [13,14]. Adenylate cyclase is now known to be capable of intramolecular dimerization. The molecule is a twice- repeated motif of six-transmembrane segments followed by a cytosolic ATP-binding domain . These two ATP- binding domains are highly homologous, and they must associate for catalytic activity and regulation by G-proteins [15,16]. The crystal structure of t hese catalytic domains has been solved [17,18]. Dimerization between catalytic domains is even preserved in the much simpler trypanosomal adenylate cyclase, which possesses a single transmembrane spanning segment [19]. Recently, we showed by a variety of functional and imaging techniques that the two transmem- brane clusters, quite independently of the cytosolic compo- nents, interacted persistently, which dictated the traf®cking and functional assembly of adenylate cyclase, AC8 [20]. The interaction between the transmembrane domains was isoform speci®c, as the ®rst transmembrane domain o f Correspondence to D. M. F. Cooper, Department of Pharmacology, Box C-236, University of Colorado Health Science Center, 4200 East Ninth Ave, Denver, CO 80262, USA. Fax: + 303 315 7097, Tel.: + 303 315 8964, E-mail: dermot.cooper@uchsc.edu Abbreviations: CFTR, cystic ®brosis transmembrane conductance regulator; SUR, sulfonylurea receptor; FRET, ¯uorescence resonance energy transfer; PVDF, poly(vinylidene di¯uoride); GFP, green ¯uorescent protein; YFP, yellow ¯uorescent protein; CFP, cyan ¯uorescent protein; ABC, ATP-binding cassette; CCE, capacitative Ca 2+ -entry. (Received 12 November 2001, accepted 28 November 2001) Eur. J. Biochem. 269, 413±421 (2002) Ó FEBS 2002 AC8 did not cotraf®c to the plasma membrane with t he second transmembrane domains of AC2 and AC5. This latter conclusion was arrived at independently by functional assays [21]. I n our exper iments, we wer e also intrigued to ®nd that the second set of transmembrane segments homodimerized strongly, although they were retained in the ER. These observations, along with the earlier bio- chemical data, prompted us to consider the possibility that adenylate cyclases might dimerize . Here, we have used a variety of approaches ranging from either suppression or rescue of function by inactive or active partial molecules, respectively, intermolecular coimmuno- precipitation and ¯uorescence resonance e nergy t ransfer (FRET) between partial and full-length cyan ¯uorescent protein (CFP)- and yellow ¯uorescent protein (YFP)-tagged molecules i n live cells to search for persistent and intimate interactions. These studies lead us to conclu de that mam- malian adenylate cyclases do form dimers (or higher o rder assemblies) the regions responsible are the hydrophobic domains and this aggregation may contribute to the associ- ation of adenylate cyclases with cellular regulatory factors. MATERIALS AND METHODS cDNA plasmid constructs and cell culture Portions of AC8 were subcloned into N-terminal or C-terminal enhan ced green ¯uorescent protein (GFP) vectors (Clontech) using convenient restriction enzyme digestion sites or PCR-based strategies, as described previously [20]. In AC8 D582)594 , a region from Y582 to L594 in the C1 domain of AC8 was deleted; in AC8 D11 26)1248 [20], a region from R1126 to P1248 in the C-terminus of AC8 (in the C2 domain) was deleted; in AC6 D553)666 , a region from S553 to F666 in the C1 domain of AC6 was deleted. These three deletions were generated by a PCR-based strategy. GFP/ AC8, GFP/8Tm2C2, 8NTm1C1/GFP, GFP/8Tm2, CFP/ 8Tm2, YFP/8Tm2, GFP/C2, 8NTm1/GFP, GFP/8C1 and 8NTm1 were as described in [20]. CFP/AC8 and YFP/AC8 were obtained by switching the GFP of GFP/AC8 into CFP and YFP between the restriction enzyme sites Nhe1and BglII, from pECFP and pEYFP vector (Clontech). 8Tm1/ CFP/Tm2 w as obtained b y subcloning 8NTm1 of 8NTm1/ CFP between the restriction enzyme sites Nhe1andAge1, which are both located right before the CFP of CFP/8Tm2. Similarly, 8Tm1/YFP/Tm2 was obtained by subcloning 8NTm1 of 8NTM1/YFP into YFP/8Tm2. HEK 293 cells were maintained as described previously [22]. Measurement of cAMP accumulation In intact cells, cAMP accumulation was measured accord- ing to the method of Evans et al. [23], as described previously [22] with some modi®cations. Cells on 24-well plates were incubated (60 min at 37 °C) with [2- 3 H]adenine (1.5 lCi per well) to label the ATP pool. The cells were then washed once and incubated with a nominally Ca 2+ -free Krebs buffer (900 lL per well) containing 120 m M NaCl, 4.75 m M KCl, 1.44 m M MgCl 2 ,11m M glucose, 25 m M Hepes, and 0.1% bovine serum albumin (fraction V) adjusted to pH 7.4 with 2 M Tris base. The use of Ca 2+ - free Krebs buffer in experiments denotes the addition of 0.1 m M EGTA to the nominally Ca 2+ -free Krebs buffer. All experiments were carried out at 30 °C in the presence of phosphodiesterase inhibitors, 3-isobutyl-1-methylxanthine (500 l M ), and Ro 2 0±1724 (100 l M ), which were p reincu- bated with the cells for 10 m in prior to a 1-min assay. Cells were preincubated for 4 min with the Ca 2+ -ATPase inhib- itor, thapsigargin, at a ®nal concentration of 100 n M .This treatment passively empties intracellular Ca 2+ stores, establishing a low basal [Ca 2+ ] iand primes the cells for CCE [24]. Assays were terminated by addition o f 5% (w/v, ®nal concentration) trichloroacetic acid a nd the percent conversion of [ 3 H]ATP to [ 3 H]cAMP was m easured as previously described previously [22]. Means  SD of triplicate determination are indicated. GFP ¯uorescence imaging The procedure w as described p reviously [20]. T ransfected HEK 293 cells were plated on glass coverslips coated with E-C-L cell Attachment Matrix (Upstate, Lake Placid, NY, USA; 1 : 100 dilution, 2 h). Forty-eight hours after trans- fection, the coverslips were loaded onto an Atto¯uor cell chamber (Molecular Probes, Eugene, OR) and 0.5 mL NaCl/P i (137 m M NaCl, 2.7 m M KCl, 10 m M Na 2 HPO 4 and 1.8 m M KH 2 PO 4 , pH 7.4) was added. Images were captured at room temperature for GFP ¯uorescence (excitation, 480/20 nm; emission, 510/20 nm). The ¯uores- cence imaging workstation consisted of a Nikon Eclipse TE 300 microscope equipped with a 100 ´ 1.4 N.A. oil immer- sion objective lens, thermoelectrically cooled charged- coupled device Micromax 5 MHz camera (Princeton Instruments), z-step motor and dual ®lter wheels controlled by SLIDEBOOK 3.0 software (Intelligent Imaging Innovation, Denver, CO, USA). Binning 1 ´ 1 mode and 500 ms integration times were used. The criteria for imaging analysis was that only cells with medium and low expression levels were captured and counted. Co-immunoprecipitation and Western blotting HEK 293 cells transfected with various constructs were solubilized in 1 mL immunoprecipitation buffer ( 50 m M Tris/HCl (pH 7.4), 150 m M NaCl, 1% Triton X100 (or 1% Nonidet P-40) and protease inhibitor cocktails) for 1 h at 4 °C,andthencentrifuged(100000g;Optima TM TL ultracentrifuge, Beckman). The supernatant was incubated (2±4 h, 4 °C) with 5 lg anti-(T7 tag) Ig (Novagen) and 100 lL protein A±agarose beads (Pierce). The beads were washed three times with 1 mL immunoprecipitation buffer plus 350 m M NaCl, once with 1 mL 50 m M Tris/HCl (pH 7.4) and 150 m M NaCl, and eluted with 50 lL2´ sample buffer. The immunoprecipitates were resolved by SDS/PAGE, transferred to a poly(vinylidene di¯uoride) (PVDF) membrane, and subjected to Western blotting using either Ab ACVIII-A 1229±1248 antibody (as des- cribed previously [22]), or Living color peptide antibody (Clontech, 1 : 100 dilution; as described previously [20]). FRET measurements The manipulation of cells expressing YFP- and CFP-tagged proteins and imaging procedures were all t he same as those for GFP imaging. FRET between CFP and YFP was mea- sured and calculated for the entire image on a pixel-by-pixel 414 C. Gu et al. (Eur. J. Biochem. 269) Ó FEBS 2002 basis using a three-®lter ÔmicroFRETÕ method as described previously [20,25]. Brie¯y, to measure FRET, three images were acquired through YFP, CFP and FRET ®lter channels. The raw FRET images consist of both FRET and non-FRET components (the donor and acceptor ¯uorescence bleeding through the FRET ®lter). The extent of cross-bleeding is characteristic of the particu lar optical system and was determined using cells that express e ither CFP/8Tm2 or YFP/8Tm2. In several experiments we found that 55.3  0.8% of CFP and 1.28  0.06% of YFP ¯uorescence can bleed through the FRET channel. Therefore, to calculate the cross-over image, CFP and YFP images were multiplied by, respectively, 0.565 and 0.014. Finally, the corrected FRET (FRET C ) image was obtained by subtracting CFP and Y FP cross-over images from raw FRET images and is presented as a quantitative pseudocolor image. All manipulations with images were performed after subtraction of the background images. RESULTS Inactive mutant adenylate cyclases suppress the activity of wild-type adenylate cyclases in vivo In a multimeric assembly requiring the integrity of the whole complex for full function, it might be expected that one inactive subunit would exert a dominant-negative effect on activity. We evaluated this possibility with adenylate cyclases, focusing largely on AC8, which can be stimulated by Ca 2+ acting via calmodulin, b inding to the C -terminus [22]. Issues of speci®city of intermolecular interactions were add ressed w ith AC5 o r AC6, which are inhibitable by Ca 2+ , apparently independently of calmod- ulin [26]. Adenylate cyclases can be divided into ®ve major domains, the N-terminus, the ®rs t transmembrane cluster (Tm1), ®rst cytoplasmic loop (C1), second transmembrane cluster (Tm2) and second cytoplasmic loop (C2) (see later). The C1 and C2 regions are further subdivided into the highly conserved catalytic C1a and C2a regions and the less conserved C1b and C2b domains. In previous studies, by deleting part of the C1 region we generated an inactive mutant of AC8, termed AC8 D58 2)594 [22]. We wondered whether this mutant might suppress the activity of cotransfected wild-type AC8. Transfection of HEK 293 cells with wild-type AC8 resulted in a dramatic increase in cAMP accumulation in response to forskolin, t he en try of Ca 2+ triggered by store depletion (capacitative Ca 2+ -entry; CCE) or especially the combination of forskolin and CCE (Fig. 1A). Replacing half of the AC8 cDNA with empty vector caused no drop in cAMP accumulation. As expected, cAMP accumulation of HEK 293 cells transfected with the inactive AC8 mutant, A C8 D58 2)594 , was no different from that of cells transfected with empty vector, regardless of the stimuli (Fig. 1A). However, when cotransfected with wild-type AC8, AC8 D582)594 dramat- ically suppressed activity under a ll stimulation c onditions (Fig. 1A). These results are consistent with the formation of homomultimeric complexes of AC8 molecules. We wondered whether a similar approach might reveal that heteromultimeric complexes could form between different isoforms of adenylate cyclase. Consequently, cells were transfected with combinations of inactive or active AC8 and active or inactive AC5 and AC6 cDNAs. The inactive AC8, AC8 D582)594 , also suppressed the activity of AC5 and AC6 (Fig. 1B). Conversely, AC8 activity was suppressed by the corresponding, inactive mutant of AC6, AC6 D55 3)666 (Fig. 1 B). These results are consistent with the formation of heterodimers. Mutants do not misdirect wildtype adenylate cyclase The dominant negative effects of cotransfected adenylate cyclase mutants on adenylate cyclase activity could also arise from either a decreased expression or a misdirection of the wild-type adenylate cyclase. To test whether the location and/or the amount of wild-typ e A C8 expressed was altered Fig. 1. Suppression of adenylate cyclase activity by inactive mutants. (A) AC8 activity can be suppressed by the coexpression of AC8 D582)594 . HEK 293 cells were transfected with the same total amount of the indicated cDNAs. The cDNA ratio in the cotransfec- tions was 1±1. Transfected HEK 293 cells w ere pretreated with thapsigargin (1 00 n M for 4 min) to activate CCE. cAMP accumulation in the intact cells was measured for 1 min after adding, vehicle (Basal); 20 l M forskolin (Forsk); 20 l M forskolin and 4 m M Ca 2+ (Forsk/ Ca 2+ ); or 20 l M forskolin, 10 l M prostaglandin E 1 and 4 m M Ca 2+ (Forsk/PGE 1 /Ca 2+ ). (B) Suppression by an inactive mutant can occur with other adenylate cyclases. The cDNAs are shown under each bar. Assays were performed as in (A). cAMP a ccumulation in the trans- fected HEK 293 cells was measured for 1 min after adding, 20 l M forskolin and 10 l M prostaglandin E 1 for transfections involving AC5 and AC6; 20 l M forskolin, 10 l M prostaglandin E 1 and 4 m M Ca 2+ for transfections involving AC8. Ó FEBS 2002 Dimerization of adenylate cyclases (Eur. J. Biochem. 269) 415 when AC8 was coexpressed with AC8 D58 2)594 ,weemployed a G FP-tagged form of AC 8, GFP/AC8, which resembles the wild-type both in terms of catalytic activity and appropriate location in the plasma membrane [20]. Co-transfection with AC8 D58 2)594 , did not alter the plasma membrane localization of GFP/AC8 (Fig. 2A,B) and the expression level was also apparently quite similar. However, just as with the wild-type, the activity of GFP/AC8 was suppress ed by coe xpressi on with AC 8 D582)594 (Fig. 2C). Rescue of inactive mutants by half molecules of AC8 in vivo A corollary of the e xperiments described above involving dominant negative suppression of adenylate cyclase activity is the rescue of inactive, mutant mole cules by complement- ary, partial molecules. Tang and colleagues had shown that there was complementation of enzymatic activities between truncated AC1 and inactive point mutations [27]. Those experiments were performed with membranes prepared from Sf9 cells expressing various baculovirus-encoded constructs [27]. We wondered whether halves of AC8 could rescue t he activ ity of A C8 D582)594 expressed in HEK 293 cells. P reviously, we described a C -terminus deletion of AC8, AC8 D1126)1248 [22], which lacked part of the C2a domain and the entire C2b region. AC8 D11 26)1248 is completely inactive when expressed alone in HEK 293 cells, as are AC8 D582)594 , GFP/8Tm2C2 (the eGFP tagged second half of A C8) and 8NTm1C1/GFP (the eGFP tagged ®rst half of AC8; Fig. 3 [ 20]). The cAMP accumulation of HEK 293 cells transfected with these constructs alone was around 0.1% when the cells were stimulated by forskolin and CCE (Fig. 3). Cells cotransfected with the combination of either AC8 D582)594 and G FP/8Tm2C2 o r A C8 D11 26)1248 and 8NTm1C1/GFP also only had background adenylate cyclase activity (Fig. 3). In contrast, cAMP accumulation of cells cotransfected with either AC8 D582)594 and 8NTm1C1/ GFP or AC8 D1126)1248 and GFP/8Tm2C2, approached 0.5%, under the same assay conditions (Fig. 3). This result shows that complementation of activity can occur between separate molecules, presumably b y generating a complete catalytic core, which, of course, suggests intimate access between these constructs. However, rescued activity is only about one-tenth of the activity of the full length wild-type AC8. This inef®cient c oupling between molecules might suggest that the physical association between two catalytic domains from the same molecule is preferred. Such inef®ciency might also underlie the lack of detectable adenylate cyclase activities from cells cotransfected w ith AC8 D58 2)594 and AC8 D1126)1248 (Fig. 3). Fig. 2. The activity but not the expression of eGFP-tagged AC8 changed in the presence of AC8 D582)594 . HEK 293 cells were transfecte d with the same total amount of the indicated cDNAs. The cDNA ratio was 1±1 in the cotransfection. 24 h after transfection, half of the cells were plated on coated glass coverslips for eGFP-imaging; another half were plated in 24-well plates for in vivo assays. (A) Cells transfected with GFP/AC8 + vector. (B) Cells transfected with GFP/AC8 + AC8 D582)594 . (C) Assays were performed as in Fig. 1. The cAMP accumulation in these t ransfected cells was measured for 1 min after adding vehicle (Basal), 4 m M Ca 2+ (Ca 2+ ), 20 l M forskolin (Forsk), or 20 lm forskolin and 4 m M Ca 2+ (Forsk/Ca 2+ ). 416 C. Gu et al. (Eur. J. Biochem. 269) Ó FEBS 2002 Homo- and heteromeric interactions revealed by coimmunoprecipitation assays To test wh ether AC8 molecules actually bind to each other and to AC6 molecules, we performed coimmuno- precipitation assays with various epitope-tagged aden ylate cyclase constructs. We previously described the mutant AC8 D1)106, D1184)1248 or Ô8M13Õ, w hich lacks t he ®rst 106 residues at t he N-terminus and the last 65 residues at th e C-terminus [22]. It is constitutively active, i ndependent of Ca 2+ and is targeted correctly [20,22]. AC8 D1)106, D1184)1248 has a T7 epitope tag at its N-terminus, but it does not possess the epitope for the AC8 speci®c antibody, Ab VIII-A, which is directed against amino acids 1229±1248 [28]. We also constructed an AC6 with a T7 tag at its N-terminus. Wild-type AC8 was cotransfect- ed with either AC6 or AC8 D1)106, D1184)1248 into HEK 293 cells and 48 h later, coimmunoprecipitation assays were performed (see Materials and methods). The T7 tag antibody was used to pull down the AC6 or AC8 D1)106, D1184)1248 , r espectively, along with any associated proteins, which were then r un on SDS/PAGE and transferred t o a PVDF membrane. Ab VIII-A 1229±1248 antibody was used in the W estern blots to determine whether AC8 was present in association with either AC6 or AC8 D1)106, D1184)1248 . Indeed, AC8 did associate with AC8 D1)106, D1184)1248 , and to a lesser extent with AC6 (Fig. 4A). AC8 immunoreactivity was not detected in coimmunoprecipi- tations from any single transfection (Fig. 4A). Thus, these data are also consistent with the occurrence of both heteromeric and homomeric interactions between adeny- late cyc lase molecules. To narrow down the region of interaction between adenylate cyclase molecules, we cotransfected HEK 293 cells with AC8 D1)106, D1184)1248 and GFP-tagged parts of Fig. 3. Inactive mutants can be partially rescued by halves of AC8 in vivo. Top: diagram o f the constructs; the green box represen ts t he GFP molecule, the 12 small black bars represent the 12 putative transmembrane segments of AC8. The names of the domains of AC8 are above the G FP/AC8 in the corresponding region. A ssays were performed as in Fig. 1. cAMP accumulation was measured by adding 20 l M forskolin and 4 m M Ca 2+ . E xperiments were performed three times with similar results. The asterisks indicate value that are signi®cantly dierent from th e backgroun d (P < 0.05). The cDNAs transfected are indicated by the plus signs. Fig. 4. Homo- and heteromeric interactions between adenylate cyclases revealed by coimmunoprecipitation assays. Immunoprecipitations were performed wit h a nti-(T7 t ag)Ig. O nly A C8 D1)106, D1184 )1248 and AC6 have a T7 tag at their N-terminal. The immunoprecipitated proteins were run on SDS/PAGE and transferred onto PVDF m embranes. Cotransfectio n conditions are indicated on the top of each blot; molecular mass (kDa) is shown on the left of the blot. (A) The asso- ciation b etween AC8 and either AC8 D1)106, D1184)1248 or AC6 was tested. Ab ACVIII-A 1229±1248 antibody was used in the Western blotting. (B) T he assoc iatio n betwe en AC 8 D1)106, D1184)1248 and dierent parts of AC8 was tested. Living color peptide antibody recognizing the eGFP molecule was used in the Western blotting. The asterisk indicates the position of a no nspeci®c antibody band. The arrow heads sho w the positions of GF P/8Tm2C2 (c. 80 kDa) and GFP/8Tm2 (c. 55 kDa). The uppe r bands are p robably oligomeric forms. The amount of AC8 expressed in a ll of the transfections was very similar (data not shown). Ó FEBS 2002 Dimerization of adenylate cyclases (Eur. J. Biochem. 269) 417 AC8 [20]. The T7 tag antibody was used to pull down AC8 D1)106, D1184)1248 and any coimmunoprecipitating pro- teins, and anti-GFP living color peptide antibody was used in the subsequent Western blotting to identify the associated proteins. AC8 D1)106, D1184)1248 strongly interacted with the second half of AC8 (GFP/8Tm2C2, approximately 80 kDa) and the second transmembrane c luster (GFP/8Tm2; approximately 55 kDa), but not at all or only weakly with either the ®rst transmembrane cluster (8NTm1/GFP), the ®rst cytoplasmic domain (GFP/8C1) or the second cyto- plasmic domain ( GFP/8C2) (Fig. 4B). [The higher molec- ular mass bands in the GFP/8Tm2C2 and AC8 D1)106, D1184)1248 and G FP/8Tm2 and AC8 D1)106, D1184)1248 com- binations were likely multimeric forms (Fig. 4B)]. These coimmunoprecipitation experiments indicated that the second transmembrane domain was the major region responsible for bringing molecules together. However, coimmunoprecipitation requires the retention of inter- actions that will survive rather rigo rous detergent treatment and while positive r esults are informative, negative r esults do not necessarily prove that weaker interactions do not occur. One approach to overcoming this problem is FRET microscopy in living cells [25]. FRET r elies on sustained, intimate associations between proteins at distances on the order of 5 nm or less, although the c hemical n ature of the interaction is not a major consideration. Consequently we evaluated FRET analysis to probe the formation of adenylate cyclase oligomers in live cells. Higher order structures revealed by FRET microscopy Using FRET microscopy, we had previously shown that when tagged with CFP and YFP, the ®rst and second transmembrane clusters of AC8 interacted with each other, which resulted in the functional assembly of adenylate cyclase and traf®cking to the plasma membrane [20]. We also noted that the second transmembrane cluster of AC8 could form homooligomers, which were retained in the ER [20]. This latter homomeric interaction of the second transmembrane cluster reminded us of earlie r literature which suggested that adenylate cyclase could dimerize or oligomerize. To evaluate the possibility of dimerization using FRET analysis, CFP/8Tm2 and YFP/8Tm2 were cotransfected into HEK 293 cells with or without the untagged ®rst transmembrane cluster, 8NTm1. As expected from our previous studies, CFP/8Tm2 and YFP/8Tm2 associated with each other, yielding a strong FRET signal from the ER (Fig. 5A). Upon the inclusion of 8NTm1, both CFP/8Tm2 and Y FP/8Tm2 appeared at the plasma mem- brane yielding a strong FRET signal (Fig. 5B). This result indicated that more than one 8Tm2 molecule, one CFP- tagged and one YFP-tagged, was present in the tightly associated 8NTm1/8Tm2 complex at the plasma membrane. This result suggests that the trans membrane domains can mediate higher order assembly of adenylate cyclases. As a corollary, we cotransfected 8NTm1/CFP, 8NTm1/YFP and 8Tm2 in HEK 293 cells. In this case, although the presence of 8Tm2 ensured that appropriate intramolecular dimeriza- tion occurred resulting in traf®cking to the plasma mem- brane, only weak FRET was de tected between the 8NTm1/ CFP and 8N Tm1/YFP elements (data not shown). These data indicate that weaker associations occur between the ®rst transmembrane segments than between the second transmembrane segments. Quite curiously, when the anal- ogous experiment was performed with the full length CFP/ AC8 and YFP/AC8, even though they both located in the plasma membrane, no clear FRET signal was detected (Fig. 5C). This somewhat surprising result c ould be explained by the fact that the two AC8 molecules associate so that their N-termini are distant (> 5 nm) from each other or that the N-terminus of AC8 is too long and ¯exible to maintain a minimally effective distance for FRET to occur. Fig. 5. Homomeric interactions between the second transmem bran e cluster and full-length AC8. CFPandYFPtaggedconstructswere cotransfected into HEK 293 cells. Pictures in each row were captured from the same cell. The ®rst (CFP) and the second (YFP) columns show the CFP ¯uorescence and YFP ¯uores- cence, respectively. The third column (CFP/ YFP overlay) are the overlay of the CFP and YFP images of the cell, which shows colocal- ization. The FRET images are presented in the fourth column (FRET C ). FRET C is displayed as a quantitative pseudocolor image. ALUFI, arbitrary linear units of ¯uorescence intensity. (A) Cotransfection of CFP/8Tm2 and YFP/8Tm2. (B) Cotransfection of 8NTm1, CFP/8Tm2 and YFP/8Tm2. (C) Cotransfec- tion of CFP/AC8 and YFP/AC8. 418 C. Gu et al. (Eur. J. Biochem. 269) Ó FEBS 2002 To address the possibility that steric effects were precluding the detection of FRET between two full length adenylate cyclase molecules, we constructed a truncated AC8 in which the two transmembrane clusters were linked with either CFP or YFP (Fig. 6A). Remarkably, in this molecule, the conformations of the t wo transmembrane clusters and both the CFP and YFP molecules were correctly maintained, as the intact molecules could t raf®c separately to the plasma membrane (Fig. 6A,B). This observation extends our previous ®nding that the two transmembrane clusters, when coexpressed are necessary and suf®cient for the plasma membrane targeting of AC8 [20]. Strikingly, coexpression of 8Tm1/CFP/Tm2 and 8Tm1/YFP/Tm2 in HEK 293 cells yielded not only the expected colocalization, but also strong FRET signals in the plasma membrane, which establishes dimer formation (Fig. 6C). This is quite compelling evidence that the transmembrane domains of adenylate c yclase can mediate oligomerization. When cells w ere cotransfected with 8Tm1/CFP/Tm2 and YFP/AC8, a lthough they were colocalized in the plasma membrane, no FRET was detected (Fig. 6C). This again suggests that even though these m olecules could dimerize, inadequate access between the N-terminus of AC8 and the C1 region of different molecules precluded the detection of FRET. DISCUSSION The present group of studies have convinced us that adenylate cyclases dimerize and that functional conse- quences can accompany this d imerization. The dominant negative effects of inactive AC8 mutants on wild-type activity, coupled with the rescue of inactive mutants by complementary, but inactive, molecules led us to seek structural correlates to this apparent multimolecular inter- action, in which a catalytic center might be formed by the C1a and C2a domains from different molecules. These rescue experiments were reminiscent of earlier in vitro experiments using truncation mutants of AC1, which suggested that adenylate cyclase might dimerize [27]. In that study, when a nonepitope-expressing, C-terminally- truncated, active, AC1 was expressed along with a mutant AC1 that possessed no enzymatic activity but that did contain the C-terminal epitope, a signi®cant amount of the enzymatic activity could be immunoprecipitated [27]. This suggested that the functional C-terminal truncation m utant and the inactive (epitope-containing) mutant associated, or at least coimmunoprecipitated. Coimmunoprecipitation experiments, by themselves, can suggest interactions between molecules, although they do require persistent interactions that can withstand detergent. Thus, a balance must be established between the rigor that is required to avoid nonspeci®c interactions and the lowering of stringency that permits weak interactions to persist. Notwithstanding these limitation s, the coimmunoprecipita- tion experiments reported here, along with the functional interactions that we encountered, d id indicate that independent adenylate cyclase molecu les interacted and did so with speci®city. In this regard, the second transmem- brane cluster seemed to play a dominant role in the intermolecular interaction. The more discerning technique of FRET analysis in live cells showed that in addition to interacting with Tm1 and t raf®cking to t he plasma mem- brane [20], homomeric interactions could occur between two Tm2 domains in the plasma membrane, which meant that the transmembrane domains of adenylate cyclase could form higher order structures in the plasma membrane. This concept was proven with our construct that retained only the transmembrane domains with a CFP or YFP in the middle of the two clusters. This construct a lso traf®cked to the plasma membrane by itself, and formed multimers in the plasma membrane, as s een by FRET analysis. Therefore, the hydrophilic and hydrophobic portion of mammalian adenylate cyclases may be considered to have two Fig. 6. Tracking of the hydrophobic portion of AC8. (A) The left panel shows the structural diagram of 8Tm1/CFP/Tm2. The red and blue cylinders represent the transmembrane segments of the ®rst and second cluster, respectively. The blue ball in the middle rep- resents CFP. The right panel is the image of 8Tm1/CFP/Tm2 transfected cells. (B) The left panel shows the structural diagram of 8Tm1/YFP/Tm2, which is the same as 8Tm1/CFP/Tm2 except that CFP is replaced by a yellow ball, YFP. The right panel is the image of 8Tm1/YFP/Tm2 transfected cells. (C) and (D) are the FRET analyses arranged as in Fig. 5. 8Tm1/CFP/Tm2 and 8Tm1/YFP/ Tm2 were cotransfected in (C). 8Tm1/CFP/ Tm2 and YFP/AC8 were cotran sfected in (D). Ó FEBS 2002 Dimerization of adenylate cyclases (Eur. J. Biochem. 269) 419 roles. The hydrophilic portion is responsible for the adenylate cyclase activity and its regulation, while the hydrophobic portion governs the molecule's targeting and oligomerization. Whereas it seems reasonable to suggest that Tm2 domains bring molecules together, the f unctional rescue studies seem to suggest that interactions between the two catalytic domains are preferred within the same molecule rather than between two molecules. This suggestion comes from the f act that only half molecules can r escue inactive adenylate cyclase muta nts, and the re scued activit y is considerably less than that of the w ild-type, which suggests an inef®cient interaction. Moreover, two full-length inactive adenylate cyclase mutants, one mutated in the C1 loop and the other mutated in t he C2 loop, cannot complement each other's activity, which suggests that intermolecular interac- tions between C1 and C2 loops do not occur in the natural assembly of two adenylate cyclase molecules. Based on these various ®ndings, a model for the higher order assembly of adenylate c yclase can be proposed that minimally comprises two aden ylate cyclase molecules. The lack of FRET between two N-terminally-tagged molecules, coupled with an inef®cient rescue by partial molecules, along with the expectation that the N-terminus and C1 loop would be close to each other, based on intramolecu lar dimerization, makes it reasonable to speculate that the two adenylate cyclase molecules are arranged in a head-to-tail fashion when they dimerize. This arran gement is also consistent with previous data showing that the N-terminus and C-terminus of AC8 appeared to interact to permit regulation by Ca 2+ acting via calmodulin [22]. Although the reported studies establish that adenylate cyclase molecules dimerize, or form even higher order structures, it is premature to speculate on the precise advantages that this dimerization provides to t he cell. Nevertheless, one speculation that might be w orth raising is that adenylate cyclases could associate with other mem- brane proteins. It is well known that many heteromultimer- forming membrane proteins can homomultimerize in the absence of their normal partners, as is the case with voltage- gated Ca 2+ -channels, which can form functional assemblies of varying properties [29]. A similar situation may occur with adenylate cyclase. A substantial body of evidence already shows that Ca 2+ -sensitive adenylate cyclases and CCE channels are intimately colocalized, with the result that only Ca 2+ entering via CCE channels can regulate these cyclases (including AC8) while the release of Ca 2+ from internal stores or ionophore-mediated intracellular calcium ion concentration increases are quite ineffectual [30±32]. The mechanism for this association is quite unclear [33]. W hat if the multivalency of adenylate cyclase molecules provided the basis for the association of adenylate cyclases with either CCE channel proteins or scaffolding proteins, so that a complete adenylate cyclase complex was an association between adenylate cyclase molecules and CCE channel proteins? Premises for this type of behavior by other members of the ABC family of proteins include the ATP- activated K + -channel discussed earlier, which is comprised of a heterooctamer of four SUR protein subunits in associationwithfourK ir subunits [13,14]. Thus the data gathered presently, although initially appearing to introduce a layer of cumbersome complexity to the structure of adenylate cyclase, may actually be a step in resolving one of the more intrigu ing properties of Ca 2+ -sensitive adenylate cyclases, namely their essential colocalization with CCE channels. At the same time, these ®ndings render more prescient and add substance to a proposal ®rst raised over 20 years ago. ACKNOWLEDGEMENTS The authors thank M. Rodbell 2 for the original stimulus for t his study and Kent Fagan for useful comments on the manuscript. This work was supported by NIH grants GM 32483 a nd NS 28389 (to D. M. F. C.). REFERENCES 1. Rodbell, M., Lad, P.M., Nielsen, T.B., Cooper, D.M.F., Schlegel, W., Preston, M.S., Londos, C. & Kempner, E.S. (1981) The structure of adenylate cyclase systems. Adv. Cyclic. Nucleotide Res. 14, 3±14. 2. Schlegel, W., Kemp ner, E.S. & Rodbell, M . (1979) Activation o f adenylate cyclase in hepatic membranes involves interactions of the catalytic unit with multimeric complexes of regulatory pro- teins. J. Biol. Chem. 254, 5168±5176. 3. Schlegel, W., Cooper, D.M.F. & Rodbell, M. (1980) Inhibition and activation of fat c ell adenylate cyclase by GTP is me diated by structures of dierent size. Arch. Biochem. Biophys. 201, 678±682. 4. Haga, T., Haga, K. & Gilman, A.G. (1977) Hydrodynamic properties of the beta-adrenergic receptor and adenylate cyclase from wild typ e and variant S49 lymphoma cells. J. Biol. Chem. 252, 5776±5782. 5. Neer, E.J., Lok, J.M. & Wolf, L.G. (1984) Puri®cation and properties of the inhibitory guanine n ucleotide regulatory unit of brain adenylate cyclase. J. Biol. Chem. 259, 14222±14229. 6. Yeager, R.E., Heideman, W., Rosenberg, G.B. & Storm, D.R. (1985) Puri®cation of the calmodulin-sensitive adenylate cyclase from bovine cerebral cortex. Biochemistry 24, 3776±3783. 7. Smigel, M.D. (1986) Puri®cation of the catalyst of adenylate cyclase. J. Biol. Chem. 261, 1976±1982. 8. Krupinski, J., Coussen, F., Bakalyar, H.A., Tang, W.J., Feinstein, P.G., Orth, K., Slaughter, C., Reed, R.R. & Gilman, A.G. (1989) Adenylyl cy clase amino acid sequence: possible chann el- or transporter-like structure. Science 244, 1558±1564. 9. Haugeto, O., Ullensvang, K., Levy, L.M., Chaudhry, F.A., Honore,T.,Nielsen,M.,Lehre,K.P.&Danbolt,N.C.(1996) Brain glutamate transporter proteins form homomultimers. J. Biol. Chem. 271, 2 7715±27722. 10. Hebert, D.N. & Carruthers, A. (1991) C holate-solubilized ery- throcyte glu cose transporters exist as a mixture o f homodimers and homotetramers. Biochemistry 30, 4654±4658. 11. Kilic, F. & Rudnick, G. (2000) Oligomerization of serotonin transporter and its functional consequences. Proc. Natl Acad. Sci. USA 97, 3106±3111. 12. Wang, S., Yue, H., Derin, R.B., Guggino, W.B. & L i, M. (2000) Accessory protein facilitated CFTR±CFTR interaction, a molec- ular mechanism to p oten tiate the chloride channel activity. Cell 103, 169±179. 13. Bryan, J. & Aguilar-Bryan, L. (1999) Sulfonylurea receptors: ABC transporters that regulate ATP-sensitive K + channels. Biochim. Biophys. Acta. 1461, 285±303. 14. Schwappach, B., Zerangue, N., Jan, Y.N. & Jan, L.Y. (2000) Molecular basis for K(ATP) assembly: transmembrane interac- tions mediate a ssociation of a K + channelwithanABCtrans- porter. Neuron 26 , 155±167. 15. Tang, W.J. & Gilman, A .G. (1995) Construction of a soluble adenylyl cyclase activated by Gs alpha and forskolin. Science 268, 1769±1772. 420 C. Gu et al. (Eur. J. Biochem. 269) Ó FEBS 2002 16. Sunahara, R.K., Dessauer, C.W. & Gilman, A.G. (1996) Com- plexity and diversity of mammalian adenylyl cyclases. Annu. Rev. Pharmacol. Toxicol. 36, 461±480. 17. Zhang, G., Liu, Y., Ruoho, A.E. & Hurley, J.H. (1997) Structure of the adenylyl c yclase catalytic core. Nature 386, 247±253. 18. Tesmer, J.J., Sunahara, R.K., Gilman, A.G. & Sprang, S.R. (1997) Crystal structure of the catalytic domains of adenylyl cyclase in a complex with Gsa.GTPcS. Science 278, 1907±1916. 19. Bieger, B. & Essen, L.O. (2001) Structural analysis of adenylate cyclases from Trypanosoma brucei in their monomeric state. EMBO J. 20, 433±445. 20. Gu, C., S orkin, A. & Cooper, D.M.F. (2001) Persistent interac- tions between the two transmembrane clusters dictate the t argeting and functional assembly of adenylyl cyclase. Curr. Biol. 11, 185±190. 21. Seebacher, T., Linder, J.U. & Sch ultz, J.E. (2001) An iso form- speci®c interaction of the membrane anchors aects mammalian adenylyl cyclase type V activity. Eur. J. Biochem. 268, 105±110. 22. Gu, C. & Cooper, D.M.F. (1999) Calmodulin-binding site s on adenylyl cyclase type VIII. J. Biol. Chem. 27 4 , 8012±8021. 23. Evans, T., Smith, M.M., Tanner, L.I. & Harden, T.K. (1984) Muscarinic cholinergic receptors of two cell lines that regulate cyclic AMP metabolism by dierent molecular mechanisms. Mol. Pharmacol. 26, 395±404. 24. Putney, J.W. & Bird, G.S. (1993) The signal for capacitative cal- cium entry. Cell 75, 199±201. 25. Sorkin, A., McClure, M., Huang, F. & Carter, R. (2000) Inter- action of EGF receptor and grb2 in living cells visualized by ¯u- orescence r esonance energy transfer (FR ET) microscopy. Curr. Biol. 10, 1395±1398. 26. Cooper, D.M.F., Karpen, J.W., Fagan, K.A. & Mons, N.E. (1998) Ca 2+ -sensitive adenylyl cyclases. Adv. Second Messenger Phosphoprotein Res. 32, 23±51. 27. Tang, W.J., Stanzel, M. & Gilman, A.G. (1995) Truncation and alanine-scanning mutants of type I adenylyl cyclase. Biochemistry 34, 14563±14572. 28. Cali, J.J., Parekh, R.S. & Krupinski, J. (1996) Splice variants of type VIII adenylyl cyclase. Dierences in glycosylation and regulation by Ca 2+ /calmodulin. J. Biol. Chem. 271, 1089± 1095. 29. Singer, D ., Biel, M., Lotan, I ., Flockerzi, V., Hofmann, F. & Dascal, N. (1991) The roles of the subunits in the function of the calcium channel. Science 253, 1553±1557. 30. Chiono, M., Mahey, R., Tate, G. & Cooper, D.M.F. (1995) Capacitative Ca 2+ entry exclusively inhibits cAMP synthesis i n C6±2B glioma c ells. Evidence th at physiologically evok ed Ca 2+ entry regulates Ca 2+ -inhibitable adenylyl cyclase in non-excitable cells. J. Biol. Chem. 270, 1149±1155. 31. Fagan, K.A., Mahey, R. & Cooper, D.M.F. (1996) Functional co-localization of transfected Ca 2+ -stimulable adenylyl cyclases with capacitative Ca 2+ entry s ites. J. Biol. Chem. 27 1, 12438± 12444. 32. Fagan,K.A.,Mons,N.&Cooper,D.M.F.(1998)Dependence of the Ca 2+ -inhibitable adenylyl cyclase of C6±2B glioma cells on capacitative Ca 2+ entry. J. Biol. Chem. 273, 9297± 9305. 33. Fagan, K.A., Smith, K.E. & Cooper, D.M.F. (2000) Regulation of the Ca 2+ -inhibitable adenylyl cyclase type VI by capacitative Ca 2+ entry requires localization in cholesterol-rich domains. J. Biol. Chem. 275, 26530±265307. Ó FEBS 2002 Dimerization of adenylate cyclases (Eur. J. Biochem. 269) 421 . PRIORITY PAPER Dimerization of mammalian adenylate cyclases Functional, biochemical and ¯uorescence resonance energy transfer (FRET) studies Chen Gu 1 ,. molecules, coimmunoprecipitation and ¯uorescence resonance energy transfer (FRET) analysis between molecules in living cells. The results strongly suggest that adenylate cyclases