Site-directed Disulfide Mapping of Helices M4 and M6 in the Ca2+ Binding Domain of SERCA1a, the Ca2+ ATPase of Fast Twitch Skeletal Muscle Sarcoplasmic Reticulum*

In an attempt to define the spatial relationships among SERCA1a transmembrane helices M4, M5, M6, and M8, involved in Ca2+ binding, all six cysteine residues were removed from predicted transmembrane sequences by substitution with Ser or Ala. The cysteine-depleted protein retained 44% of wild type Ca2+ transport activity. Pairs of cysteine residues were then reintroduced to determine whether their juxtaposition would result in the formation of disulfide cross-links between transmembrane helices. In initial studies designed to map the juxtaposition of Ca2+ binding residues, Cys was substituted for Glu309 or Gly310 in transmembrane sequence M4, in combination with the substitution of Cys for Glu771 in M5; for Asn796, Thr799, or Asp800in M6; or for Glu908 in M8. These double mutants all retained the capacity to form a phosphoenzyme intermediate from Pi (but not from ATP in the presence of Ca2+), and in all but mutants E309C/N796C and G310C/N796C, phosphoenzyme formation was insensitive to 100 μm Ca2+. These results support the view that both Glu309 and Asn796 contribute to Ca2+ binding site II, which is not required for conversion of E 2, the substrate for Pi phosphorylation, toE 1. Cross-linking in mutants E309C/N796C and G310C/D800C established reference points for the orientation of M4 and M6 relative to each other and provided the basis for the prediction of potential additional cross-links. Strong links were formed with the pairs T317C/A804C and T317C/L807C near the cytoplasmic ends of the two helices and with A305C/L792C and A305C/L793C near the lumenal ends. These combined results support the conclusion that M4 and M6 form a right-handed coiled-coil structure that forms part of the pathway of Ca2+ translocation. In addition to providing a possible explanation for the mutation sensitivity of several pairs of residues in these helices, the proposed association of M4 and M6 supports a new model for the orientation of the two Ca2+ binding sites among transmembrane helices M4, M5, and M6.

The Ca 2ϩ ATPase of rabbit fast twitch skeletal muscle sarcoplasmic reticulum (SERCA1a) is a 110-kDa transmembrane protein, which transports Ca 2ϩ ions from the sarcoplasm to the lumen of the membrane system at the expense of ATP hydrolysis. We developed an expression/mutagenesis system for analysis of the functional consequences of site-specific mutagenesis of the Ca 2ϩ ATPase (1) and used it to identify six amino acids, located in transmembrane sequences M4, M5, M6, and M8, which appeared to provide Ca 2ϩ binding ligands (2,3). These were Glu 309 in M4; Glu 771 in M5; Asn 796 , Thr 799 , and Asp 800 in M6; and Glu 908 in M8. Since Ca 2ϩ binding to two sites in the ATPase is sequential and cooperative (4,5), it is of interest to assign Ca 2ϩ binding residues to each of the two sites. Extensive mutational analysis of the six Ca 2ϩ binding residues has provided criteria for the assignment of Glu 771 , Thr 799 , and Glu 908 to Ca 2ϩ binding site I (the more lumenal of the two stacked sites) and Glu 309 and Asn 796 to Ca 2ϩ binding site II (the more cytoplasmic of the two stacked sites) (6 -8). Asp 800 has been proposed to contribute to both sites (8). Mutants in five of the six Ca 2ϩ binding residues and in Gly 310 lost their capacity to occlude Ca 2ϩ , but E908D, which had lost Ca 2ϩ transport activity, retained the ability to occlude Ca 2ϩ (8,9), and E908A retained both Ca 2ϩ occlusion and transport (10).
Scanning mutagenesis of transmembrane segments M4, M5, M6, and M8 revealed that M4 and M6 contain patches of mutation-sensitive residues on one face of the central tiers of each helix (11,12). Helix M5, which runs antiparallel to M4 and M6, has a similar patch of mutation-sensitive residues. M4 and M6 share a sequence motif, (E/D)GLPA(V/T). The motif forms a prominent part of the mutation-sensitive patch, suggesting that it may be an important component of the Ca 2ϩ binding pore. A reverse sequence in the mutation-sensitive patch of M5 shares many characteristics of the M4/M6 motif. In particular, the sequences EGL (M4), EGV (reverse in M5), and DGL (M6) share a negative charge, a Gly linked to an acidic residue and a hydrophobic residue. In contrast, transmembrane helix M8 contained only one mutation-sensitive residue, Glu 908 , and only vestiges of the motif observed in M4 and M6. These results suggest that Glu 908 and, indeed, all of M8 is unlikely to play more than a peripheral role in Ca 2ϩ binding and translocation.
A limiting factor in our understanding of the mechanism of Ca 2ϩ transport is our lack of knowledge of the detailed structure of the Ca 2ϩ binding and translocation domain of SERCA1a. A method of site-directed disulfide mapping, in which cysteine residues are inserted in pairs into predicted transmembrane domains, followed by exposure to oxidant, has been used to map interactions between helices of several bacterial chemoreceptors (13)(14)(15). These proteins are homodimers, with each monomer consisting of two membrane-spanning helices. Cross-links between monomers were detected by testing for anomalous migration of monomers in nonreducing SDS-PAGE. 1 More recently, this method has been applied to eukaryotic multispanning membrane proteins such as rhodopsin (16) and P-glycoprotein (17). In the latter studies, the cDNA encoding a single large peptide chain was first divided into two fragments, and site-specific mutagenesis was used to introduce a single cysteine residue into each half. The two altered pieces were co-expressed and subjected to oxidizing conditions. Rejoining of the two expressed halves through disulfide bond formation was detected by SDS-PAGE.
Unlike P-glycoprotein and rhodopsin, expressed fragments of SERCA1 have not been shown to associate into an active enzyme complex. However, cleavage at the hypersensitive tryptic site, T1, located at Arg 505 in the large cytoplasmic domain between transmembrane helices M4 and M5, divides the protein into two segments of nearly equal length (18,19). Under conditions that stabilize the E2 conformation of the enzyme, this reaction is rapid, specific, and complete. Separation of the products by SDS-PAGE provides a convenient method of assay for cross-links formed between helix M4 and downstream helices. We have utilized this technique to explore the juxtaposition of the proposed Ca 2ϩ binding residues in the central tiers of helices M4, M5, M6, and M8, and we have extended the analysis to residues in the upper and lower tiers of the M4 and M6 helices, as illustrated in Fig. 1. The positions of cross-links demonstrate that M4 and M6 are likely to interact as a righthanded coiled-coil in the membrane. Insight into these structural arrangements has allowed us to propose a side-by-side arrangement of the Ca 2ϩ binding sites in the membrane (20).

EXPERIMENTAL PROCEDURES
Materials-Various enzymes for DNA manipulation were from New England Biolabs, Pharmacia Biotech Inc., Life Technologies, Inc., and Boehringer Mannheim Gmbh. Tissue culture reagents were from Life Technologies, Inc. Trypsin, soybean trypsin inhibitor, NEM, PIPES, Tris, MOPS, bovine serum albumin, Tween 20, aprotinin, and 1,10phenanthroline were from Sigma. 45 Ca 2ϩ and [␥-32 P]ATP were from Amersham Corp. 32 P i was from Mandel Scientific. Secondary antibodies were from Promega. Nicholson plastic models for helix modeling were from Labquip (Reading, UK).
Oligonucleotide-directed Mutagenesis, Expression of Mutant cDNA, and Preparation of Microsomes-Mutagenesis was carried out according to the method of Kunkel (21). Cell culture and transient transfections were performed as described previously (22). HEK-293 cells were harvested 40 h after transfection, and microsomes were prepared as described previously (1), with minor modifications made to accommodate the cross-linking assay. Briefly, cells were harvested 40 h after transfection, washed with phosphate-buffered saline, and swollen at 0°C for 10 min in hypotonic solution. Aprotinin was added, and the cells were homogenized with 40 strokes in a glass Dounce homogenizer. The homogenate was diluted with an equal volume of solution A (0.5 M sucrose, 300 mM KCl, 10 mM Tris-HCl, pH 7.5, 40 M CaCl 2 , and 5 mM freshly dissolved glutathione) and homogenized with an additional 20 strokes. Mitochondria and nuclei were removed from the suspension by centrifugation at 10,000 ϫ g for 30 min. The supernatant was made 0.6 M in KCl, and microsomes were pelleted by centrifugation at 180,000 ϫ g for 1 h. The microsomal pellet was suspended in solution B (0.5 ϫ solution A containing 10 mM Tris-HCl, pH 7.5). Microsomal protein was quantified (23) using bovine serum albumin as standard. Microsomes were flash-frozen in liquid N 2 and stored at Ϫ70°C. Expression of mutant SERCA1a was verified by immunoblotting with monoclonal antibody A52 (24) and quantitated by an enzyme-linked immunosorbent assay (25).
Ca 2ϩ Dependence of Ca 2ϩ Transport Activity and Analysis of Phosphoenzyme Intermediates-Ca 2ϩ uptake activity was assayed as described previously (11). Free Ca 2ϩ concentrations were calculated using the computer program of Fabiato (26). Assays of phosphoenzyme formation from ATP and stability of E 1 P or E 2 P in nontransporting mutants were carried out as described previously (2,27). Phosphorylation from P i was performed in a buffer containing 100 mM MOPS/Tris, pH 7.0, 10 mM MgCl 2 , 0.5 mM [ 32 P]KH 2 PO 4 , 20% (v/v) Me 2 SO, and either 100 M CaCl 2 or 2 mM EGTA. Approximately 10 g of microsomes were phosphorylated for 10 min at 22°C before quenching with 10 volumes of ice-cold 7% trichloroacetic acid. Sample preparation and polyacrylamide gel electrophoresis were performed as described previously (2,27).
Disulfide Cross-linking and Trypsin Digestion-Microsomes were thawed on ice, diluted to 1 mg of microsomal protein/ml, and reduced by the addition of 1 ⁄20 volume of 20 mM freshly dissolved glutathione. After a 5-min incubation at room temperature, samples were oxidized by the addition of an equal volume of a buffer containing copper phenanthroline as an oxidant and EGTA and vanadate, which are likely to promote the E 2 conformation. The final incubation conditions were 140 mM KCl, 5 mM MgCl 2 , 5 mM EGTA, 400 M sodium vanadate, 25 mM PIPES, pH 7, 150 mM sucrose, 1.75 mM glutathione, 0.3 mM CuSO 4 , and 0.9 mM 1,10phenanthroline. Following a 10-min incubation at room temperature, trypsin, at a concentration of 1 mg/ml, was added to a final ratio of 1:20 (trypsin/microsomal protein). After 4 min at room temperature, an equal volume of 2 ϫ loading buffer (28), containing 25 mM NEM and 245 g/ml soybean trypsin inhibitor in place of reducing agent, was added.
Glutaraldehyde Cross-linking-Glutaraldehyde cross-linking was performed by a modification of the method of Ross and McIntosh (29). Briefly, 40 l of SERCA1a⌬C microsomes in solution B, diluted to 1 mg of protein/ml, were warmed to 22°C. Cross-linking was initiated by the addition of 2 l of 125 mM glutaraldehyde, followed by incubation for 3 min at 22°C. Tryptic digestion was initiated by the addition of 1 l of 0.5 M EGTA, followed by 2 l of trypsin (1 mg/ml), and incubation for 3 min at 22°C. The reaction was stopped by the addition of 40 l of nonreducing loading buffer.
Electrophoresis and Immunoblotting of Proteins-For nonreducing gels, samples were loaded on 8% polyacrylamide gels (37.5:1 acrylamide:bisacrylamide) and run according to standard protocols (28). For reducing gels, DTT was added to 200 mM, and samples were incubated at 37°C for 5 min prior to loading on 10% acrylamide gels. Prestained standards (Bio-Rad) were used in both cases to visualize bands during the run. Nonstained standards (New England Biolabs) were loaded on nonreducing gels for more accurate determination of molecular weight. In both cases, loading volumes were adjusted so that approximately equal amounts of expressed SERCA1 were present in each lane. A semidry electroblot apparatus was used to transfer proteins to nitrocellulose membranes (30), and Ponceau red stain was used to visualize transferred protein. Membranes were blocked for 1 h in a solution of Tris-buffered saline, pH 7.5 (TBS), containing 0.5% (w/v) powdered milk and incubated with mouse monoclonal antibody A25 (N-terminal specific) or A52 (C-terminal specific) (24) at a concentration of 1 g/ml in TBS containing 0.1% (w/v) powdered milk for an additional 60 min at room temperature. Membranes were washed twice with TBS containing 0.5% Tween 20 for 15 min and incubated in TBS with horseradish peroxidase-conjugated anti-mouse secondary antibody (Promega) according to the manufacturer's instructions. Membranes were again washed in TBS/Tween, and expressed SERCA1 protein was detected with an enhanced chemiluminescence kit (Pierce SuperSignal). Bands were detected on Kodak Biomax x-ray film. Quantitation of band intensities was done by analysis of computer-scanned images of the autoradiographs using NIH Image version 1.59 software.

Construction and Analysis of Serca1⌬C-
The deduced transmembrane sequences of wild type SERCA1a contain six cysteine residues. Mutant Serca1a⌬C was constructed by mutating Cys 70 to Ser and Cys 268 , Cys 318 , Cys 774 , Cys 910 , and Cys 938 to Ala. The locations of Cys 318 , Cys 774 , and Cys 910 are indicated in Fig. 1. Cys 877 and Cys 890 , predicted to lie in a large lumenal loop and to form a disulfide cross-link (31), could not be mutated because substitution of either of these residues reduced expression dramatically. Attempts at simultaneous removal of cytoplasmic cysteine residues C-terminal to the T 1 site also decreased expression, although none of them appear, individ-ually, to be essential to function. 2 Serca1a⌬C, expressed in HEK-293 cells, transported Ca 2ϩ with activity equivalent to 44% of wild type (Table I). Apparent Ca 2ϩ affinity, measured as Ca 2ϩ dependence of Ca 2ϩ uptake, was increased slightly, from pCa 6.5 in wild type to pCa 6.74 in Serca1a⌬C (Table I). Since gross perturbations of the structure of the mutant protein would manifest as either loss of expression or loss of function, these data indicate that the overall structure of the Serca1a⌬C mutant was not significantly different from that of the wild type enzyme.
Replacement of Ca 2ϩ Binding Ligands with Cysteine-Our initial strategy was to introduce cysteine residues, one pair at a time, into different helices in the Ca 2ϩ binding domain. Cys was substituted for either Glu 309 or Gly 310 , in combination with a second substitution of Cys for either Glu 771 , Asn 796 , Thr 799 , Asp 800 , or Glu 908 . These residues, identified as probable Ca 2ϩ binding ligands (2), were expected to cluster close to each other in the transmembrane domain. In later mutants, a series of residues in M4 and in M6 were replaced with cysteine, on the basis of models that allowed us to predict which residues might lie close to each other. The locations of all of the residues that were changed to Cys in different experiments are shown in Fig. 1.
Enzymatic Activity of the Cysteine-substituted Serca1a⌬C-All of the doubly cysteine-substituted mutants were expressed in HEK-293 cells at levels similar to the wild type enzyme, as judged by Western blotting. None of the mutants in which Ca 2ϩ binding ligands were replaced had Ca 2ϩ uptake activities above that of untransfected cells, and none of them formed a high energy phosphoenzyme intermediate (E 1 P) by phosphorylation from ATP in the presence of Ca 2ϩ (data not shown). As shown in Fig. 2, all of these mutants formed a low energy phosphoenzyme intermediate (E 2 P) from P i in the absence of Ca 2ϩ and, in all but mutants E309C/N796C and G310C/ N796C, E 2 P formation was insensitive to 100 M Ca 2ϩ .
The observation that double mutant E309C/N796C showed no Ca 2ϩ activation of phosphoenzyme formation from ATP but normal Ca 2ϩ inhibition from P i at neutral pH is consistent with the hypothesis that Glu 309 and Asn 796 both contribute to Ca 2ϩ binding site II, occupation of which is not required for conversion of E 2 to E 1 conformations of the ATPase (32). The presence of 100 M Ca 2ϩ resulted in partial inhibition of P i phosphorylation of the double mutant G310C/N796C. This type of inhibition is consistent with an alteration of Ca 2ϩ affinity at Ca 2ϩ binding site I. Thus, an increase in the size of the residue (Gly to Cys) at position 310 appears to perturb site I, but not to disrupt it. Single mutations of Gly 310 to Val or Pro had a similar effect (11,33).
Detection of Disulfide Cross-links in the Central Tiers of M4 and M6 -Microsomal vesicles containing mutant ATPases were reduced with glutathione, a soluble, membrane-impermeant reducing agent, to prevent the formation of disulfide bonds between endogenous Cys residues located in the cytoplasmic domain. They were then incubated in a buffer containing EGTA and vanadate under conditions known to promote and maintain the E 2 conformation of the enzyme. Cu(II) (1,10phenanthroline) 3 was present in this buffer as a membranepermeant oxidant. After a 10-min incubation, trypsin was added for 4 min before loading buffer containing soybean trypsin inhibitor was added to stop the digestion. In this incubation buffer, trypsin cut predominantly and nearly completely at the T 1 site (Arg 505 ), as illustrated in Figs. 3-6, demonstrating that the enzyme was in the E 2 conformation. The loading buffer also contained NEM to block free cysteines and prevent spurious cross-linking in denatured SERCA1. Following electrophoresis and blotting onto nitrocellulose membranes, SERCA1 fragments were detected immunologically.
The results of experiments with cross-linking of Ca 2ϩ binding residues are presented in Fig. 3 and quantitated in Table II. The immunoblot in Fig. 3A was probed with monoclonal antibody A25, specific to the N-terminal fragment, and the immunoblot in Fig. 3B was probed with antibody A52, specific to the C-terminal fragment. All lanes in Fig. 3, A and B, contained heavy bands at 55 kDa, and double mutants E309C/N796, E309C/D800C, and G310C/D800C contained an additional band corresponding to a mass of 152 kDa. A very weak 152-kDa band was also observed with E309C/I799C. The 152-kDa band was detected with both antibodies and appeared with the same intensity relative to the 55-kDa band in all cases, demonstrating that both N-and C-terminal fragments were present in equal proportions (Table II). Although the anomalous mobility of the 152-kDa band suggested that its mass might approach that of three half molecules, antibody staining data are consistent only with a dimeric complex between the two halves of the molecule.
On the basis that the 152-kDa band represents a dimeric complex between the two halves of the cleaved SERCA1a molecule, clear cross-links formed in the double mutants E309C/ N796C, E309C/D800C and G310C/D800C, and a trace crosslink formed in E309C/T799C ( Fig. 3 and Table II). Cross-links were not formed in the double mutants E309C/E771C, G310C/ E771C, G310C/N796C, G310C/T799C, E309C/E908C, or G310C/E908C. These observations suggest that Ca 2ϩ binding residues in M4 are juxtaposed with Ca 2ϩ binding residues in M6 and possibly in M5, but not with Ca 2ϩ binding residues in M8 in the E 2 conformation. Fig. 4 shows a second gel electrophoretic pattern in which the  (12). Three lightly shaded residues, Cys 318 in M4, Cys 774 in M5, and Cys 910 in M8, were mutated to Ala. Additional mutations, not indicated, were Cys 70 to Ser in M1, Cys 268 to Ala in M3, and Cys 938 to Ala in M9. Darkly shaded residues were mutated from the residue indicated to cysteine. For cross-linking experiments, mutation to cysteine was carried out in pairs, as indicated under "Results."

TABLE I
Properties of mutants retaining Ca 2ϩ transport function Ca 2ϩ -dependent Ca 2ϩ transport activity was assayed as described under "Experimental Procedures." Maximal uptake activity was normalized to that of the wild type protein expressed in HEK-293 cells. K Ca , presented in pCa units, is the concentration at which half maximal Ca 2ϩ uptake rates were obtained. Data represent mean Ϯ S.E.; n ϭ 3.  Fig. 3 were separated under reducing conditions. The addition of DTT to the loading buffer resulted in the disappearance of both the 152-kDa band and the small amount of a 110-kDa residual band. This degree of sensitivity to reducing agents demonstrates that both the 152-and 110-kDa bands are formed by disulfide cross-links between the two halves of the cleaved SERCA1a molecule. We propose that the 110-kDa band represents cross-links very near to the T 1 cleavage site, leading to only minor changes in mobility, while the 152-kDa band represents cross-links in the transmembrane domain of the molecule, leading to more pronounced changes in mobility.

Cross-linking in Upper and Lower Tiers of M4 and M6
Helices-Since cross-links were abundant in the Ca 2ϩ binding tiers of the M4/M6 helices, it was of interest to determine whether cross-links might form in the tiers above and below the central tiers. Earlier analysis of mutation-sensitive residues in M4 and (v/v) Me 2 SO, 0.5 mM [ 32 P]P i , and either 100 M Ca 2ϩ (ϩ) or 2 mM EGTA (Ϫ), as described under "Experimental Procedures." Samples were quenched with ice-cold trichloroacetic acid, allowed to precipitate, washed with ice-cold trichloroacetic acid, and taken up in running buffer. Samples were separated by SDS-PAGE using the method of Weber and Osborn (52) with the running buffer adjusted to pH 6.3. Radioactivity was detected by autoradiography.

FIG. 3. Detection of disulfide cross-linked products.
Microsomes prepared from transfected HEK-293 cells were reduced with 1.5 mM glutathione and then incubated for 10 min in the presence of 0.3 mM Cu(II) (1,10-phenanthroline) 3 , as described under "Experimental Procedures." Trypsin was added in a ratio of 1:20 (trypsin/total protein), and digestion was carried out for 4 min before an equal volume of loading buffer containing soybean trypsin inhibitor and NEM was added. Glutaraldehyde cross-linking in Lane 13 was performed as described under "Experimental Procedures." Between 2 and 5 g of total protein was loaded in each lane, and samples were separated by SDS-PAGE (28) using a 2.5% stacking gel and an 8% resolving gel. Separated proteins were electroblotted onto nitrocellulose, and SERCA1a fragments were detected with monoclonal antibodies. A, detection with monoclonal antibody A25, specific for the N-terminal fragment of SERCA1a. B, detection with monoclonal antibody A52, specific for the C-terminal fragment of SERCA1a.  4. Separation of disulfide cross-linked products on reducing gels. Samples were prepared as described in the legend to Fig. 3, except that 200 mM DTT was added to the loading buffer as a reducing agent. Samples were separated on a 10% resolving gel. A, detection with A25 monoclonal antibody. B, detection with A52 monoclonal antibody. M6 revealed that T317D and A804V were conformational change mutants (12,34). On the basis that these residues, located at a distance from the Ca 2ϩ binding residues, might interact in a "knob-in-hole" fashion, we tested whether they might cross-link. As demonstrated in Fig. 3 Table II, the pair T317C/A804C formed a strong cross-link. Modeling of the two central cross-linking mutants, E309C/ N796C and G310C/D800C, together with the distal cross-linking mutant, T317C/A804C, using Nicholson plastic model sets, allowed us to predict that other cross-links might occur along a line that represented crossing of the M4/M6 helices at an angle of about 40°. On the basis of these predictions, a second round of cysteine substitutions was performed on residues in the upper and lower tiers of helices M4 and M6. Cross-linking results are presented as immunoblots in Figs. 5 and 6, and quantitation is presented in Table II.

and quantitated in
Double mutants T317C/L807C and L321C/G808C formed very strong cross-links. Double mutants A305C/L792C and A305C/L793C formed weaker cross-links. Double mutants I298C/I788C and A301C/I788C formed weak cross-links. Two of these, A301C/I788C and A305C/L792C, had a slightly higher mobility in the gel than the others (Figs. 3 and 5). Double mutants A303C/L792C, A303C/V795C, A306C/L792C, A306C/ V795C, and A313C/D800C did not form cross-links. Fig. 6 demonstrates that all proposed cross-links are sensitive to reducing conditions. These mutants illustrate that cross-links do occur along a line representing crossing of the M4/M6 helices at an angle of about 40°.
The group of double mutants outside of the proposed Ca 2ϩ binding cavity was assayed for Ca 2ϩ transport activity, and those that transported Ca 2ϩ were also assayed for apparent Ca 2ϩ affinity, measured as Ca 2ϩ dependence of Ca 2ϩ transport (25). Functional data, summarized in Table I, show that those mutants that retained low Ca 2ϩ transport activities had apparent Ca 2ϩ affinities similar to that of the wild type enzyme.
Transport-negative mutants were assayed for phosphoen-zyme formation from ATP and were tested for blocks in the Ca 2ϩ transport cycle. Those mutants in which Ca 2ϩ binding ligands were unaltered were phosphorylated by ATP in the presence of Ca 2ϩ (Fig. 7, lane 1). In Fig. 7, lanes 2 and 3, it is apparent that all of the Ca 2ϩ transport-deficient double mutants were blocked at the E 2 P dephosphorylation step, although the kinetics of the reaction differed among mutants. This phenotype has been identified in several single mutations involving residues in each of the helices M4, M5, and M6 (11,12,33). In these earlier studies, alteration of Ala 305 , Ala 306 , Gly 310 , or Ala 804 to Val resulted in a blockage of E 2 P dephosphorylation, while the mutant T317D was blocked in the E 1 P to E 2 P step (34).

DISCUSSION
In this study, disulfide cross-linking was used to gain insight into the structural relationships that exist among the three or four helices (M4, M5, M6, and possibly M8) that contribute residues to the two Ca 2ϩ binding and translocation sites in SERCA1a. Data obtained from the formation of site-directed cross-links demonstrate the juxtaposition of Ca 2ϩ binding ligands between M4 and M6, but not between M4 and M5 or between M4 and M8. Cross-links formed at both ends of transmembrane helices M4 and M6 provide evidence that the interaction between these helices occurs throughout their length. The association of Ca 2ϩ binding ligands in M4 and M6 provides valuable experimental evidence for modeling of the Ca 2ϩ binding sites.
Identification of Cross-linked Bands-Disulfide cross-linking of Serca1a⌬C, followed by tryptic cleavage at Arg 505 near the center of the 994-residue molecule, resulted in the formation of four types of bands. A major band at 55 kDa in all lanes undoubtedly represented the two cleaved fragments. In some lanes, an additional band appeared at 60 kDa, but, unlike the other bands, it was only detected with the N-terminal antibody. Since it was sensitive to reducing conditions, it may result from a cytoplasmic disulfide linkage formed within the N-terminal half of SERCA1a.
A weak band at 110 kDa appeared in all lanes. Since this band was sensitive to reducing conditions, it is most likely that it resulted from disulfide linkage and not from incomplete tryptic digestion. Only in lane 1 of Fig. 4 and lane 12 of Fig. 6 was a small amount of undigested protein observed. SERCA1a contains 18 cysteine residues in the cytoplasmic domain, and it is possible that some of these are sufficiently close to each other to form disulfide linkages under oxidizing conditions. Some cysteine residues lie near the tryptic cleavage site, and disulfide bonds there, unlike those formed in the transmembrane domain, could preserve the linear structure of the peptide, providing it with a mobility in SDS-PAGE similar to that of the native enzyme. In all cases, the intensity of this band accounted for less than 5% of the total protein (Table II).
The fourth band was a DTT-sensitive, variable band of 152 kDa. This band is likely to be formed from cross-linking of the two fragments in the transmembrane sequence, leading to their anomalous mobility in SDS-PAGE. When cross-linked soluble proteins are separated by SDS-PAGE under nonreducing conditions, they often migrate with higher mobility than the native protein, presumably due to a more compact structure. This is not necessarily true of membrane proteins. Although no unexpected mobility shift was found for cross-linked fragments of rhodopsin (16), disulfide cross-linking of chemoreceptor dimers decreased their mobility, giving them apparent masses 2-3 times greater than the monomer, depending on where the linkage occurred (14,15,35). When disulfide crosslinks were introduced into the multispanning transmembrane domain of the P-glycoprotein, its mobility corresponded to a higher apparent mass (17). For SERCA1a (29), a glutaraldehyde cross-link in the cytoplasmic domain between Lys 492 and Arg 678 , followed by tryptic cleavage at Arg 505 , decreased the mobility of the cross-linked protein so that its apparent mass corresponded to 135 kDa. In our hands, the apparent mass was 140 kDa. On the basis of these examples of anomalous mobility, it is reasonable to conclude that cross-linked, cleaved Serca1a⌬C could move in SDS-PAGE with a mobility corresponding to 152 kDa.
Further support for this view is provided by the observation that the ratio between the stain incorporated into the 152-and 55-kDa bands was the same when the samples were stained with either the N-terminal or C-terminal specific monoclonal antibodies, demonstrating that the 152-kDa bands contain both halves of the molecule in an equal ratio (Table II). These observations lead us to conclude that the 152-kDa bands result from disulfide cross-links formed between residues in M4 and M6, rejoining the two halves of the protein that were cleaved by tryptic digestion.
We have considered the possibility that cross-linking results from misfolding of the expressed protein, since many of the constructs were unable to transport Ca 2ϩ . An increased sensitivity to proteases is a clear sign of misfolding, and our tryptic digestions did not show this for any of the mutants in this study. Another indication of misfolding is extremely low expression levels, and, again, all of the mutants expressed at least 30% of wild type levels. In addition, all mutants showed phosphoenzyme formation, indicating that the cytoplasmic domains were intact in all cases. Accordingly, we conclude that misfolding did not contribute to the results obtained in this study.
Structural Predictions-The numerous cross-links formed between M4 and M6 establish that there is close contact between these helices under conditions that would normally favor the E 2 conformation. Maintenance of the protein in a single conformation would minimize helix movements, which might otherwise be an important factor, since each link was formed in a separate experiment. Movement of membrane-spanning helices during conformational changes has been described in the case of the aspartate receptor (36). In light of our proposed mechanism of Ca 2ϩ transport (20), it is likely that helix movements do occur as the enzyme moves among the different conformations that are critical to Ca 2ϩ transport, although these will be restrained by close packing (36).
The simplest way to evaluate possible helix packings is to use the method devised by Crick (37) of superimposing the helical nets of M4 and M6 to optimize residue intercalation and to mask in the cross-links as illustrated in Fig. 8. According to Chothia et al. (38,39) helix packing is facilitated by the ridges and grooves formed by rows of side chains that are separated by either three or four residues. These rows are identified as the i Ϯ 3 series and the i Ϯ 4 series. Packing of the i Ϯ 3 series of one helix into the i Ϯ 4 series of a second helix results in a left-handed crossing, with an average angle of interaction between 20 and 25°(3-4 interaction). Alternatively, if ridges formed by the i Ϯ 4 series of one helix pack into the grooves of the i Ϯ 4 series of a second helix, then the two helices cross with right-handed orientation, with an average angle of interaction between Ϫ40 and Ϫ50°(4 -4 interaction). The 4 -4 mode of packing is most common in interactions between soluble helices, whereas the 3-4 mode has been commonly seen in transmembrane helix packing (38,40).
The use of this method to model the interactions between M4 and M6 requires a defined starting point. The identification of cross-links between Ca 2ϩ binding ligands in M4 and M6, as well as sequence similarity around these residues, provides an obvious starting point for the superpositioning of M4 and M6. As illustrated in Fig. 8, Glu 309 and Asp 800 can be placed in the same relative position in both nets, forming the center of interaction. Fig. 8, A-C, shows 4 -4 packing, and Fig. 8, D-F, shows 3-4 packing of helices M4 and M6. In Fig. 8, A and D, residues that form strong cross-links are identified; in Fig. 8, B and E, weak links are identified; and in Fig. 8, C and F, substitutions that did not form linkages are identified. The weakly cross-linked residues in the centers of the helices are equally consistent with both models (compare length of arrows in panels B and E). However, the strongly cross-linked residues at either end of the helices are placed much closer together when the helices are drawn in a 4 -4 packing mode than when they are drawn in a 3-4 mode (compare length of arrows in FIG. 7. Phosphoenzyme formation and decay of inactive mutants. Microsomal fractions from transfected HEK-293 cells were phosphorylated with [␥-32 P]ATP in the presence of 100 M Ca 2ϩ for 7 s. All reactions were done on ice. Samples were then either exposed to 1 mM EGTA to block forward phosphorylation and permit forward dephosphorylation or exposed to 1 mM EGTA in the presence of 1 mM ADP to allow dephosphorylation through ATP formation. After the reaction, samples were quenched with ice-cold 7% trichloroacetic acid. Lane 1, samples were quenched immediately following phosphorylation. Lane 2, EGTA and ADP were added 10 s before quenching. Lane 3, EGTA was added 30 s before quenching. Following quenching, samples were treated as described in the legend to Fig. 2. panels A and D). Our linkage data are, therefore, most consistent with a 4 -4 crossing interaction. Since this interaction continues throughout the length of both helices, the two helices must form a right-handed coiled-coil. Maintenance of the large crossing angle required by 4 -4 packing would require significant bending of straight helices. This may be correlated with the presence of three conserved proline and two conserved glycine residues in M4 and M6. Recently published structures of the water channel aquaporin-1 have shown precisely this sort of interaction between transmembrane helices (41,42).
Implications of Cross-linking for Ca 2ϩ Binding Sites-The results of this study allow us to reconcile functional data obtained though mutagenesis with physical Ca 2ϩ binding data to form a consistent model of how the Ca 2ϩ binding pore is formed. SERCA1 binds two Ca 2ϩ ions, and the bound ions can be distinguished kinetically. Ca 2ϩ binding to the second site is cooperative (43), so that site I must be filled before site II. Binding of Ca 2ϩ to site II prevents the cytoplasmic exchange of Ca 2ϩ bound to site I, providing evidence that the two sites are "stacked" (4), with site I being more lumenal than site II. Although the two ions are randomized in the occluded state, there is evidence that stacking is regained once they face the lumen (5). Most models, therefore, illustrate the two sites with one stacked above the other.
Experiments involving ATP and P i phosphorylation of constructs containing mutations of the proposed Ca 2ϩ binding ligands have allowed their assignment to sites I and II (6 -8). The formation of the high energy phosphoenzyme intermediate, E 1 P, is dependent on occupation of both Ca 2ϩ binding sites (44). By contrast, formation of a low energy phosphoenzyme intermediate, E 2 P, from P i is inhibited by binding of Ca 2ϩ to a single site, because occupation of site I alone is sufficient to convert the P i -reactive E 2 P to the P i -nonreactive E 1 P (45). Since mutations of Glu 771 and Thr 799 prevent both E 1 P formation and inhibition of E 2 P formation, these residues were proposed to contribute to site I (8). On the basis that mutations of Glu 309 and Asn 796 have limited ability to prevent Ca 2ϩ inhibition of phosphoenzyme formation from P i , but profound effects on phosphorylation from ATP, they were assigned to site II (8). Direct measurement of Ca 2ϩ binding to the E309Q mutant expressed in Sf9 cells demonstrated the loss of a single Ca 2ϩ binding site. The remaining site was accessible from the lumenal side at pH 6.4, but not from the cytoplasmic side, supporting the view that Glu 309 contributes to site II (7). Intermediate effects on phosphorylation suggested that Asp 800 contributes to both sites (6,8,10).
In our experiments, the double mutant E309C/N796C was not phosphorylated from ATP, but was phosphorylated by P i (Fig. 2). At pH 7.0, phosphorylation from P i was inhibited by 100 M Ca 2ϩ . This observation is fully consistent with the contribution of both Glu 309 and Asn 796 to site II. The double mutant G310C/N796C was not phosphorylated from ATP but was phosphorylated from P i . In this mutant, 100 M Ca 2ϩ caused partial inhibition of P i phosphorylation. This result is consistent with loss of site II, together with a decrease in the Ca 2ϩ affinity of site I. Since mutation of Gly 310 to Val or Pro resulted in a block of E 2 P dephosphorylation, in a decrease in Ca 2ϩ affinity (as assayed by P i phosphorylation (11,33)), and even in loss of Ca 2ϩ occlusion for the G310P mutant (9), it now seems probable that Gly 310 is located in a position that impinges on site I, so that an increase in its side chain length will affect the Ca 2ϩ affinity of site I. The loss of Ca 2ϩ sensitivity of P i phosphorylation for double mutants E309C/E771C, E309C/ T799C, E309/D800C, and E309C/E908C is consistent with contributions of Glu 771 , Thr 799 , Asp 800 , and Glu 908 to site I, as proposed earlier (6,8).
A problem with the stacking of site II directly above site I is that Asn 796 in M6, assigned to site II, is closer to the lumen than Thr 799 , assigned to site I, the more lumenal of the sites. To resolve this problem, one model proposes that Asn 796 is not a Ca 2ϩ ligand at all (47,48). This proposal seems unlikely, since Asn 796 is required for Ca 2ϩ occlusion (8). Andersen and Vilsen (49) suggested that M6 is not fully helical but rather loops back on itself. However, there is both biochemical and structural evidence for 10 transmembrane helices (50,51). The crosslinking data presented in this paper provide the basis for a  (38). Helix M6 is shaded to help distinguish it from M4. Residues in helix M4 are drawn in large circles and labeled with capital letters using black text on a white background. Residues in helix M6 are drawn in small circles and labeled with small letters using white text on a black background. In all cases, residues for which cysteine was substituted are labeled according to their amino acid number in the SERCA1 sequence (46). A-C, a right-handed (4 -4) interaction is modeled by aligning the i Ϯ 4 series of both helices. D-F, a left-handed (3-4) interaction is modeled by aligning the i Ϯ 7 series of both helices. A and D, pairs of residues that formed strong links are indicated by double-headed arrows. B and E, pairs of residues that formed weak links are indicated by double-headed arrows. C and F, pairs of resides that did not form cross-links are indicated by doubleheaded arrows.
"side-by-side" model for the Ca 2ϩ binding site that accommodates the structural problems that arise with the "stacked sites" model (20). If M4 and M6 are oriented to optimize crosslinks between them, as indicated in Fig. 9A, then Glu 309 , Asn 796 , and Asp 800 would be positioned near the M4/M6 contact, while Thr 799 would lie to one side of this contact. If M5 were placed with Glu 771 apposed to Thr 799 in M6, then the ligands between M4 and M6 would be those assigned to site II (Glu 309 , Asn 796 ), while the ligands between M5 and M6 would be those assigned to site I (Glu 771 , Thr 799 ). Asp 800 in M6 would be in a position to contribute to both sites (Fig. 9B). This side-by-side sites model, more fully illustrated in Ref. 20, has the advantage that Asn 796 can contribute to site II without distortion of helix M6. This model also places Gly 310 in a position in which alteration in its size would impinge on Ca 2ϩ binding site I, rather than Ca 2ϩ binding site II. In a stacked model, Gly 310 would contribute only to site II, unless helix M4 were distorted.
Implications for Mutation Sensitivity outside of the Ca 2ϩ Binding Cavity-The strong cross-links found at the top and bottom of M4 and M6 provide possible explanations for previous mutagenesis data. Single mutants T317D and A804V are nonfunctional, conformation change mutants, while mutant L807A has a significantly lower Ca 2ϩ affinity than wild type (12,34). These functional consequences of mutagenesis might be explained in structural terms. The space between the hydrophobic Ala 804 and Leu 807 in M6 might better accommodate a polar Thr residue in M4 than a negatively charged Asp. Ala 804 might be small enough to provide space for the ␤branched Thr 317 , while Leu 807 might help lock it into place. At the bottom of M4, mutation of Ala 305 to Val was found to result in an E 2 P dephosphorylation block (11). Mutation of either Leu 792 or Leu 793 in M6 to Ala or Ser decreased Ca 2ϩ transport to less than 50% of wild type, although the slightly smaller Val was tolerated at either position (12,48). Ca 2ϩ affinity was not altered significantly by either of these two mutations in M6.
Positioning of the small Ala 305 in M4 between the bulky Leu 792 and Leu 793 residues in M6 provides a possible explanation for the sensitivity to residue size at position 305.