The Dimerization Domain of the b Subunit of theEscherichia coli F1F0-ATPase*

In this study a series of N- and/or C-terminal truncations of the cytoplasmic domain of the b subunit of the Escherichia coli F1F0 ATP synthase were tested for their ability to form dimers using sedimentation equilibrium ultracentrifugation. The deletion of residues between positions 53 and 122 resulted in a strongly decreased tendency to form dimers, whereas all the polypeptides that included that sequence exhibited high levels of dimer formation. b dimers existed in a reversible monomer-dimer equilibrium and when mixed with other b truncations formed heterodimers efficiently, provided both constructs included the 53–122 sequence. Sedimentation velocity and 15N NMR relaxation measurements indicated that the dimerization region is highly extended in solution, consistent with an elongated second stalk structure. A cysteine introduced at position 105 was found to readily form intersubunit disulfides, whereas other single cysteines at positions 103–110 failed to form disulfides either with the identical mutant or when mixed with the other 103–110 cysteine mutants. These studies establish that the bsubunit dimer depends on interactions that occur between residues in the 53–122 sequence and that the two subunits are oriented in a highly specific manner at the dimer interface.

The F 1 F 0 ATP synthase is a multisubunit enzyme complex that is responsible for the production of the bulk of intracellular ATP. This complex synthesizes ATP from ADP and inorganic phosphate by utilizing a transmembrane proton gradient as an energy source. Structurally and functionally the complex can be divided into two major domains: the membrane intrinsic F 0 domain and the peripheral F 1 domain. The F 0 domain of the prototypical Escherichia coli enzyme consists of three polypeptides in the stoichiometry of a 1 b 2 c 9 -12 that function in proton translocation across the inner membrane. The F 1 domain, which has the subunit composition of ␣ 3 ␤ 3 ␥ 1 ␦ 1 ⑀ 1 , performs the ATP catalytic functions. (For recent reviews of ATP synthase function and structure, see Refs. 1-3.) Structural studies have shown that the 3␣ and 3␤ polypeptides form a ring of alternating subunits surrounding a central region occupied by the N-and C-terminal helices of the ␥ subunit (4). The remainder of ␥ lies outside of the ␣ 3 ␤ 3 domain and is closely associated with the ⑀ subunit (5). This ␥⑀ subcomplex extends for 45 Å from F 1 to the membrane where interactions with the c subunits of the F 0 domain occur (6). Thus, ␥⑀ forms the central "stalk" often seen in electron micrographs (7). The binding change model (reviewed in Ref. 1) suggests that rotation of the ␥⑀ stalk relative to the ␣ 3 ␤ 3 domain results in sequential changes in the catalytic sites on the ␤ subunits, causing them to cycle through conformations favoring substrate binding, then catalysis, and finally product release. This mechanism requires a stator structure to hold the ␣ 3 ␤ 3 hexamer as the ␥⑀ stalk rotates within it. Recently, a structure that reaches from the membrane domain to the periphery of the catalytic domain and that is distinct from the central ␥⑀ stalk has been observed in electron micrographs (8). This "second stalk" may function as a stator formed by the b and ␦ subunits.
The b subunit is necessary for the stable binding of F 1 to the membrane and for correct assembly of the complex (9). b is a 156-residue polypeptide with a hydrophobic N-terminal membrane-spanning ␣ helix. The remainder of the protein is highly polar and extends into the cytoplasm where it interacts with the ␦ subunit and the ␣ 3 ␤ 3 domain. The cytoplasmic domain of b, consisting of residues 24 -156, has been expressed separately from the membrane spanning domain and forms a soluble, highly extended homodimer that can bind F 1 in a manner similar to the intact protein (10). Previous studies have shown that the two b monomers in the complex can be covalently cross-linked without abolishing activity, confirming that the dimeric state is the functional form of the protein (11). The affinity of b for F 1 or the isolated ␦ subunit is strongly correlated with its ability to form dimers (12). The only significant stretch of hydrophobic amino acids in the cytoplasmic domain occurs between Val 124 and Ala 132 . Mutation of Ala 128 to aspartate has been shown to disrupt dimerization of the cytoplasmic domain (13).
Cross-linking has demonstrated that the b-␦ interaction is between the C-terminal domains of each of the polypeptides (14). ␦ can also be cross-linked to the N-terminal region of the ␣ subunit located near the top of the of the catalytic hexamer (15). Therefore, the second stalk is believed to be formed by the b subunits extending upward from the membrane to interact with ␦, which binds on the top third of the ␣␤ domain. The second stalk could function as a relatively passive, rigid stator or play a more active role in catalysis by transiently storing energy in an elastic manner. Much of the cytoplasmic region of the b dimer is expected to exist as a pair of extended helical rods to span the 140 -150 Å from the membrane to the top of F 1 .
In this study, we have characterized a series of N-and C-terminal truncations of the soluble b domain to identify the minimal region necessary for formation of the homodimer. In addition we have made inferences about the structure of the b dimer based on sites of intersubunit disulfide formation and on changes in hydrodynamic behavior upon truncation of the sequence.

EXPERIMENTAL PROCEDURES
Plasmid Construction-Molecular biological procedures were carried out as described by Sambrook et al. (16). Constructions were confirmed by restriction endonuclease mapping and regions of DNA derived from PCR 1 products were sequenced to ensure that only the desired mutations had been introduced. Plasmid pDM3, which carries a synthetic sequence encoding b 24 -156 2 in pUC8, has been described previously (17). To stabilize mRNAs and enhance expression of various forms of b, a 202-base pair BfaI-HindIII fragment of pSD100 (17) carrying the unc transcriptional terminator was inserted downstream of the b 24 -156 synthetic gene, using the HindIII and NdeI sites. The resulting plasmid was called pDM3T.
A number of plasmids encoding N-or C-terminal truncations of the soluble form of b were used in this work; these plasmids and polypeptides are summarized in Fig. 1. Previously, work from this laboratory described three plasmids, pSD114, pSD111, and pKK1, which express soluble b polypeptides lacking 33, 52, or 66 residues, respectively, from the N-terminal of b (17). For the current studies a series of plasmids encoding C-terminal truncations was constructed by PCR mutagenesis. In these constructions, PCR using plasmid pDM3 as a template was carried out using the M13 reverse universal primer coupled with mutagenic primers containing a HindIII restriction site and a sequence complementary to a stop codon and the desired region of synthetic b sequence. PCR products were co-digested with HindIII and a second enzyme, both unique in plasmids pDM3 or pDM3T, and inserted into one of those plasmids using the same restriction sites. Sequences encoding some of the C-terminal truncations were pasted into pSD114 or pSD111 to produce forms of soluble b bearing deletions at both ends.
Another set of plasmids was constructed; each of these plasmids encoded a mutant form of b 24 -156 in which one of the amino acid residues between Ala 103 and Glu 110 of the normal b sequence was replaced by cysteine. The residue numbers cited refer to the position in the wild type b sequence. These plasmids were constructed by PCR mutagenesis and cloned into pDM3 using standard techniques as described previously (17).
Purification of Proteins-Induced cells expressing the polypeptide of interest were suspended in a volume of 50 mM Tris-HCl, pH 8.0, 10 mM MgCl 2 , 1 mM phenylmethylsulfonyl fluoride equal to 10 times their packed wet weight and disrupted by one passage through a French pressure cell at 20,000 p.s.i. Cell debris was removed by centrifugation for 10 min at 9,000 rpm in a Beckman JA-20 rotor. The supernatant solution was centrifuged for another 1.5 h at 38,000 rpm in a Beckman Ti-60 rotor.
The purification of polypeptides b 24 -156 , b 34 -156 , b 53-156 , b 24 -152 , and b 67-156 have been described previously (17). Other b truncated polypeptides were purified from the high speed supernatant solutions using generally similar techniques of ammonium sulfate precipitation, ion exchange chromatography, and size exclusion chromatography, with the modifications summarized below. During purification, fractions were analyzed by SDS-PAGE, and the final products were essentially pure, as judged by this technique.
b 24 -145 was precipitated with 40% saturated ammonium sulfate, redissolved and dialyzed against TE buffer (50 mM Tris-HCl, pH 8.0, 1 mM EDTA). The protein was loaded onto a DEAE-Sepharose column equilibrated with TE buffer and eluted with a linear gradient of 0 -250 mM NaCl in TE buffer. Fractions containing the protein were again dialyzed and loaded onto a second DEAE-Sepharose column; this second column was eluted with a gradient of 0 -200 mM NaCl in 25 mM imidazole-HCl buffer at pH 6.4. The protein was precipitated with 70% saturated ammonium sulfate and further purified by size exclusion chromatography on a column of Sephacryl S-300. b 24 -134 and b 24 -138 were both purified using the following procedure. The proteins were precipitated with 40% saturated ammonium sulfate, redissolved, and dialyzed against TE buffer. Upon application to a column of DEAE-Sepharose equilibrated with TE buffer, the proteins did not bind but upon elution with TE buffer were retarded slightly. This step was repeated on a second column of DEAE-Sepharose to remove yet more impurities. The protein was precipitated with ammonium sulfate and finally purified by size exclusion chromatography on a column of Sephacryl S-300.
b 24 -122 was precipitated with 35-55% saturated ammonium sulfate, dialyzed against TE buffer, and run through two columns of DEAE-Sepharose, similar to the procedure described for b 24 -134 and b 24 -138 , to remove other proteins. After precipitation with ammonium sulfate, final purification was by size exclusion chromatography on a column of Sephadex G-75.
b 24 -114 was unlike the other polypeptides described above in that a part of the protein was insoluble and precipitated during low speed centrifugation. The purification was continued with the portion of the protein that remained in the supernatant. Following high speed centrifugation, the b 24 -114 was precipitated from the supernatant by adding ammonium sulfate to a concentration of 55% saturation. The precipitate was redissolved in TE buffer and dialyzed, and most contaminating proteins were removed by passage through a column of DEAE-Sepharose, as for b 24 -134 and b 24 -138 . Appropriate fractions were pooled, the protein was precipitated with ammonium sulfate, and final purification was by size exclusion chromatography on a column of Sephacryl S-200.
b 34 -122 was precipitated with 45% saturated ammonium sulfate, dialyzed against TE buffer, and run through a column of DEAE-Sepharose, similar to the procedure described for b 24 -134 and b 24 -138 . Appropriate fractions were pooled and adjusted to pH 4.8 by addition of acetic acid. Final purification was achieved by loading the sample on a column of CM-Sepharose and elution with a linear gradient of 0 -1 M NaCl in 25 mM sodium acetate buffer, pH 4.8.
b 53-122 was precipitated with 45-70% saturated ammonium sulfate and run through two columns of DEAE-Sepharose, similar to the procedure described for b 24 -134 and b 24 -138 . Appropriate fractions from the second column were pooled, and the pH was adjusted to 5.0 by the addition of acetic acid. Final purification was achieved by loading the sample onto a column of CM-Sepharose equilibrated with 30 mM sodium acetate buffer, pH 5.0, and elution with a linear gradient of 100 -600 mM NaCl in 30 mM sodium acetate, pH 5.0.
Protein purity was assessed by 15% SDS-PAGE gels using standard glycine running buffers except for the constructs of molecular weight less than 10 kDa (b 34 -122 , b 53-122 ) where improved resolution was obtained using 15% Laemmli gels (18) but with a running buffer of 100 mM Tris, 100 mM Tricine, 0.1% SDS. All chemicals and solvents used were reagent grade. Protein concentrations were determined spectrophotometrically at 280 nm except b 67-156 , which lacked an aromatic chromophore and was observed at 240 nm. The extinction coefficients at 280 and 240 nm were based on concentrations determined by quantitative amino acid analysis.
Analytical Ultracentrifugation-Purified protein samples were dialyzed into 50 mM Tris-HCl, pH 7.5, 0.1 M NaCl, 1 mM EDTA, or 50 mM sodium acetate, pH 5.0 for ultracentrifugal analyses. Initial concentrations of the polypeptides analyzed were between 20 and 50 M of monomer. Unless otherwise noted all reported concentrations refer to monomer.
Sedimentation equilibrium data were collected in a Beckman Optima XL-A Analytical Ultracentrifuge with absorbance optics. A four-hole An-60Ti rotor and six-channel cells with Epon charcoal centerpieces were used. Absorbance measurements were taken in 0.002-cm radial steps and averaged over 10 observations. Samples were allowed to equilibrate for 24 h at the desired speed (30,000 or 36,000 rpm) and temperature (5, 20, or 40°C) before scanning. After the 40°C scan the samples were cooled to 20°C, allowed to equilibrate, and scanned again to ensure the reversibility of the monomer-dimer transition and to confirm that no degradation had occurred.
Software supplied by Beckman was used for data processing and curve fitting. The reported observed molecular weight, M obs , for a single species was calculated based on the average of three data sets. The monomer molecular weight, M c , was calculated from the amino acid sequence. The partial specific volumes of proteins were determined from the amino acid sequence by the method of Cohn and Edsall (19). The density of the solvents were measured in a pycnometer or, in some cases, calculated from published tables (20).
Sedimentation velocity samples were run at 60,000 rpm in doublesector cells at 5 and 20°C. Once full speed was attained the samples were scanned every 10 min for a total run time of up to 300 min. Each scanned point was the average of five observations, and all data reported were the averages of at least three runs. The data sets were processed using the SVEDBERG program (21) utilizing the modified Fujita-MacCosham model fitted for a single species. When necessary the observed values for sedimentation coefficients, s obs , were converted to s 20,w . Frictional coefficients (f) and frictional ratios (ƒ/ƒ min ) were calculated from M r and s 20,w by standard methods.
NMR Relaxation Measurements-NMR spectra were collected on a Varian Unity 500 MHz spectrometer equipped with a triple resonance probe and z axis pulsed field gradients. A sample of 15 N-labeled b 53-122 in 50 mM acetate buffer, pH 5.0, was equilibrated at 25°C. One-dimensional T1, T2, and 15 N NOE spectra were collected using the pulse sequences described in Farrow et al. (22). The T1 relaxation was measured in an array of 25 steps from 11.2 ms to 1.36 s in 56-ms increments. The T2 signal was measured from an array of 11 spectra with relaxation delays from 16.6 to 183.6 ms in 16.6-ms steps. For the T1 and T2 data the signal envelopes were integrated and fit to determine the first order exponential decay constant. The NOE experiments were run with either a 5-s relaxation delay for the no NOE base-line experiment or a relaxation delay of 2 s followed by a 3-s 1 H presaturation period to determine the magnitude of 15 N NOE. The ratios between the intensities of selected peaks in the NOE and no NOE experiments were calculated. Global correlation times, m , were calculated as described previously by Farrow et al. (22).
Chemical Cross-linking-The cross-linking of b 53-122, b 67-156 , b 24 -114 , and b 34 -156 was carried out for 10 min at room temperature using 1 mM bis(sulfosuccinimidyl)suberate (BS 3 ) (Pierce). The cross-linking reactions occurred in the presence of 50 mM triethanolamine buffer, 1 mM EDTA, pH 7.5, and were quenched by the addition of ethanolamine-HCl, pH 7.5, to a final concentration of 100 mM. Complete quenching was achieved by leaving the reactions standing for 10 min at room temperature followed by heating in SDS sample buffer. The products of the reactions were then run on 15% SDS-PAGE gels using the Tricine running buffers previously specified.
Disulfide Bond Formation-Forms of b 24 -156 bearing mutations incorporating cysteine into one of the positions between Ala 103 and Glu 110 were partially purified and tested for their ability to form disulfide bonds as described previously (17). Briefly, the partially purified proteins were reduced by dialysis into buffer containing 1 mM dithiothreitol and then dialyzed in buffer containing 0.1 M NaHCO 3 , 10 M CuCl 2 , and 10 mM cysteine at 4°C to induce disulfide bond formation. After 24 h, samples were treated with 15 mM N-ethyl maleimide in SDS-PAGE sample buffer to block unreacted thiol groups and analyzed by nonreducing SDS-PAGE.

RESULTS
Previous and current studies of the b subunit suggest a four-domain model, which is shown in Fig. 1A to assist in the presentation of data. The N-terminal 24-residue sequence is highly hydrophobic and is embedded in the membrane with the other F 0 subunits. The sequence from residues 25 to 52, which is not essential for dimer formation (17), is designated the tether domain, because it joins the membrane region to the beginning of the sequence essential for dimerization. The extent of the remaining sequence that is necessary for homodimer formation is defined in this report. Near the C terminus are sites required for binding to the ␦ subunit and the F 1 sector; we refer to this region as the ␦-binding domain.
Expression, Purification, and Properties of C-terminal Truncations of b Subunit-Previous work from this laboratory has characterized soluble forms of b lacking the N-terminal membrane spanning domain (10), three deeper N-terminal deletions (17), and minor C-terminal truncations (14). In the current work, a series of plasmids encoding deeper C-terminal truncations of b was constructed (Fig. 1), and the polypeptides were purified and characterized. All of the new forms were extracted as soluble polypeptides, except b 24 -114 , which was partly soluble and partly inclusion body. The purification of these proteins by conventional techniques of ammonium sulfate precipitation, ion exchange chromatography, and size exclusion chromatography is described under "Experimental Procedures." A number of properties of these polypeptides became apparent during this work. All C-terminal deletions redissolved easily following ammonium sulfate precipitation, unlike forms with an intact C terminus. Little difference in the percentage of ammonium sulfate required for precipitation was apparent, however, until the hydrophobic region between residues 124 and 132 was removed. Forms truncated at residue 138 or earlier in the sequence failed to bind to DEAE resin at pH 8.0, underscoring the acidic nature of the C-terminal region that had been removed. There are six acidic and two basic side chains between 134 and 156; deletion of this region changes the pI for the protein from about 6 to over 8. DEAE-Sepharose was nevertheless used during the purification of these polypeptides, because it absorbed most of the other proteins in the samples. The less acidic nature of these proteins resulted in a much enhanced solubility at pH 5.0 in comparison with forms like b 34 -156 , which precipitate as the pH is lowered toward 5.0 (17). This property was exploited through purification on CM-Sepharose.
Determination of Oligomerization State of the b Subunit Truncations-Sedimentation equilibrium centrifugation allows accurate determination of the molecular weight of soluble proteins under native conditions and hence the stoichiometry of multimeric complexes. Previous studies of the b 24 -156 construct have demonstrated that the isolated cytoplasmic domain exists primarily as a dimer (10).
In this study of the molecular weight of the b subunit truncations, sedimentation equilibrium data were fitted to a single component model to determine the average observed molecular weight, M obs , of the purified protein in solution. The data are presented in Table I (17). pDM3 and pDM3T encode the same polypeptide and differ only in the presence of a transcriptional terminator downstream of the expressed gene in pDM3T. pSD122 and pSD131 encode the same polypeptide and differ only in that pDM3 was the parental plasmid for pSD122, whereas pDM3T was the parental plasmid for pSD131. is reflected in the nonintegral values for these ratios.
In previous work it was demonstrated that b 24 -156 , b 34 -156 , and b 53-156 exist primarily as dimers, whereas the b 67-156 truncation was monomeric in solution (17). In the current work an analysis of the effect of temperature on the properties of these polypeptides was undertaken. The average oligomerization states of the solution species at 5, 20, and 40°C are listed in Table I. Comparison of the values at 5 and 20°C for the three dimeric constructs revealed that there was a trend for the M obs /M c ratio to increase from about 1.9 at 5°C to slightly more than 2 at 20°C. We have previously reported that b 24 -156 has a slight tendency to aggregate at 20°C when its concentration is higher than about 1 mg/ml (17). Because the values reported in Table I reflect average molecular weights of species present in the ultracentrifuge cell, this slight aggregation can raise the M obs /M c ratio to values above 2.0. At 40°C, the M obs /M c ratio dropped to less than 1.4 for b 24 -156 and b 34 -156 , indicating that the higher temperatures destabilized the dimer interaction, shifting the monomer-dimer equilibrium toward the monomeric state. The b 53-156 polypeptide gave higher values than the less truncated forms b 24 -156 and b 34 -156 at all temperatures examined; the M obs /M c value of 2.40 obtained at 20°C for this construct indicates that it is especially prone to aggregate to species higher than dimer. It is likely that the value obtained at 40°C, 1.71, was higher than those for b 24 -156 and b 34 -156 because of the effect of this aggregation on the average molecular weight, rather than because removing residues 34 -52 caused the remainder of the protein to form a tighter dimer. As expected, the b 67-156 construct exhibited M obs /M c values below 1.2 for the entire temperature range.
To determine the extent of sequence necessary for dimerization, a series of subtractive deletions from the C terminus were made and sedimentation equilibrium studies at 5, 20, and 40°C were carried out. Novel constructs studied here included b 24 -138 , b 24 -122 , and b 24 -114 . Deletion of all or part of the 123-156 sequence had little effect on dimerization. These constructs exhibited M obs /M c ratios ranging from 1.76 to 1.89 at 5 and 20°C, consistent with a majority of the protein existing as a dimer at those temperatures. At 40°C, values ranging from 1.09 to 1.29 were obtained, indicating a predominance of monomer at the higher temperature. The removal of the 115-122 sequence to produce b 24 -114 resulted in a dramatic decrease in the M obs /M c values measured at 5 and 20°C to 1.29 and 1.16, respectively. These results indicate that C-terminal truncations deeper than residue 122 resulted in a strong shift to monomer. None of the C-terminal deletion constructs exhibited the temperature-dependent increase in the M obs between 5 and 20°C that was observed in the polypeptides having an intact C-terminal sequence.
From these sedimentation equilibrium studies of the individual N-and C-terminal truncations, it appeared that only the region from residue 53 to residue 122 was necessary for dimerization. To test whether the 53-122 sequence was indeed sufficient for dimerization, a construct with both N-and C-terminal deletions was prepared. The M obs /M c for b 53-122 was close to 2 at both 5°C (1.96) and 20°C (1.89), whereas at 40°C this construct exhibited a thermal sensitivity of dimerization similar to that of the other dimeric b constructs. These results confirm that the sequence essential for dimerization of b is between residues 53 and 122.
Most of the experiments presented in Table I were carried out at pH 7.5, where all of the constructs are soluble. To gain insight into the interactions that stabilize the dimer, sedimentation equilibrium data were gathered on b 53-122 at pH values between 4 and 8, using a relatively low concentration of the protein at 20°C. The highest molecular weight was observed at pH 5.0, implying that dimer formation was strongest at this pH (data not shown). The effect of temperature at pH 5.0 was then examined using the same initial concentration as in the previous experiment carried out at pH 7.5 (Table I). A higher level of thermal stability was apparent with the M obs /M c values at 40°C increasing from 1.25 at pH 7.5 to 1.39 at pH 5.0. Other constructs that included the C-terminal sequence beyond residue 134 precipitated at pH values lower than 6.0.
To assess the reversibility of the monomer-dimer equilibrium, after each 40°C set of equilibrium data had been collected, the centrifuge was cooled to 20°C, and the protein was allowed to re-equilibrate. Subsequent data collection and analysis revealed that the M obs /M c ratios returned to values close to those observed at 20°C prior to raising the temperature to 40°C (data not shown). This result indicates that the observed changes were due to a reversible phenomenon rather than an irreversible process such as degradation. Further confirmation of reversibility was provided by a series of sedimentation equilibrium runs of the b 24 -152 construct at 35°C, a temperature at which average molecular weights indicate that both the monomeric and the dimeric states are well populated. Cells loaded with initial concentrations of 19 and 42 M polypeptide were equilibrated at 24,000, 30,000, and 36,000 rpm, and a cell loaded with 96 M polypeptide was equilibrated at the two lower rotor speeds. As expected for a reversible equilibrium, the average molecular weight obtained by fitting each data set to a model with a single species increased with the concentration (data not shown). Furthermore, the eight data sets could be simultaneously fitted to a monomer-dimer equilibrium model, yielding good fits and a K d for the equilibrium of 28 Ϯ 4 M at 35°C, pH 7.5.
Heterodimer Formation-Chemical cross-linking using bifunctional cross-linking agents has been used previously to demonstrate the ability of b 24 -156 (10) and single amino acid mutants (12) to form dimers. In the current work, cross-linking with BS 3 , which reacts with the amino groups on lysyl side chains to form links with an 11.4 Å spacer, was used to monitor the relative populations of homo-and heterodimers present in mixtures of different b constructs. As seen in Fig. 2A (lanes 2  and 6), treatment of isolated b 34 -156 or b 53-122 with BS 3 produced a strong dimer band on SDS-PAGE, as previously observed for b 24 -156 (10). The b 67-156 construct was cross-linked with a much lower efficiency, as seen in lane 2 of Fig. 2B, consistent with its existence as a monomer in solution.
From the reversibility of the monomer to dimer transition, as demonstrated by the sedimentation equilibrium data, one would expect that mixtures of two constructs would exchange and establish an equilibrium of monomers, homodimers, and heterodimers in solution. BS 3 cross-linking of b 53-122 in the presence of b 34 -156 ( Fig. 2A, lanes 3-5) resulted in the appearance of a heterodimer band on the gel midway between that of the b 53-122 homodimer and the b 34 -156 homodimer. The relative intensities of the bands indicate that the minimal dimer sequence efficiently forms heterodimers with b 34 -156 . In contrast, cross-linking b 53-122 in the presence of the monomeric b 67-156 (Fig. 2B) produced only a trace of heterodimers. A similar result was obtained when b 53-122 was cross-linked in the presence of monomeric b 24 -114 (data not shown). Although the levels observed on the cross-linking gels were not quantitative in regard to the relative populations of the monomer-dimer equilibrium, the relative heterodimer to homodimer populations should closely reflect those present in solution. These results reinforce the evidence that the 53-122 sequence contains all of the residues necessary for formation of dimers by the cytoplasmic domain of b.
Hydrodynamic Analyses of b Truncations-Sedimentation velocity analyses were used to study the hydrodynamic behavior of the b truncations to gain information on their shapes. Data were collected at 5 and 20°C and corrected for the effects of density and viscosity, and the resulting sedimentation coefficients (s 20,w ) and frictional ratios (ƒ/ƒ min ) are presented in Table II. Generally, soluble forms of the b subunit exhibited s 20,w values significantly lower than those reported for globular proteins of similar molecular weight and frictional ratios nearer 2 than 1. The dimer of b 24 -156 measured at 20°C can be compared (Table III) with a globular protein of nearly the same size, carbonic anhydrase, which sediments much more rapidly and has a frictional ratio just slightly higher than unity (23). These results indicate that the soluble domain of b deviates significantly from an ideal globular shape.
Sedimentation coefficients of some of the deletion constructs at 20°C have been reported previously (14); here those data sets have been reanalyzed using the SVEDBERG program (21) that allows more accurate determinations for small proteins. Starting with the data collected at 20°C, the N-terminal truncations from residues 24 to 53, which we call the tether domain, resulted in a modest decrease in the s 20,w at 20°C from 1.74 to 1.66 S, whereas the ƒ/ƒ min dropped from 1.93 to 1.71, implying that this region extends from the rest of the protein. In contrast, removal of the C-terminal four residues resulted in a large decrease in the s 20,w from 1.74 to 1.46 S and an increase in ƒ/ƒ min to 2.25, implying that the C-terminal deletion produced a more extended form of the molecule. Further deletion from residues 152 to 122 (compare b 24 -152 to b 24 -122 ) did not significantly alter the sedimentation coefficient, whereas the ƒ/ƒ min decreased from 2.25 to 1.95, implying deletion of an unfolded, highly extended portion of the protein. Removal of the entire C-terminal region, from residues 122 to 156 (compare b 24 -156 to b 24 -122 and b 53-156 to b 53-122 ) resulted in substantial decreases in s 20,w but very small changes in the ƒ/ƒ min values. Together, these results imply that the C-terminal domain is normally folded into a relatively compact structure at 20°C and that this folded structure is dependent on the last four residues of the protein but independent of the tether domain.
Comparison of the properties observed at 5°C with those at 20°C reveals that the constructs that included an intact C terminus exhibited significantly lower s 20,w values at 5°C, whereas the C-terminal truncations had very similar s 20,w values at the two temperatures. The similarity of the hydrodynamic parameters of b 24 -156 at 5°C to those of b 24 -152 at either 5 or 20°C suggests that the intact C-terminal domain is sensitive to cold-induced unfolding.
The shape of the dimerization domain by itself was investigated using the b 53-122 construct. The values obtained were not significantly dependent on temperature. A pH of 5.0, which would be expected to minimize the presence of monomers, gave a slightly higher sedimentation coefficient and a slightly lower frictional ratio, and these parameters probably provide the better view of the protein. In Table III we compare these values to those of myoglobin, a globular protein of nearly the same molecular weight, and to those of two coiled-coil sequences, one larger than b 53-122 , and the other smaller, which have been expressed from cortexillin (24). Myoglobin has a sedimentation coefficient of 2.04 S and ƒ/ƒ min of 1.105 (25). In comparison, b 53-122 sediments far more slowly (s 20,w ϭ 1.36 S), and its high frictional ratio of 1.66 indicates that this structure is elongated rather than globular. For coiled-coil polypeptides, the length of the structure will increase with chain length, but the average diameter will be unaffected, so the asymmetry of shape should be greater for longer coiled-coils, and they should have higher frictional ratios compared with shorter coiled-coils. This trend is apparent in the hydrodynamic parameters of the two cortexillin fragments, Ir-12hN and the larger Ir-4hC. The values for b 53-122 are intermediate between them, as would be expected if it were to adopt a similar structure of intermediate length. This suggests that a pair of helices that interact along their length may provide a good model for the dimerization domain.
NMR Relaxation Studies-The NMR relaxation rates of proteins in solution are dependent on overall hydrodynamic size and shape along with other effects such as chemical exchange. The global rotational correlation time, m , of b 53-122 was calculated to be 20.4 ns based on 15 N T1, T2, and NOE measurements, as described under "Experimental Procedures." This m was considerably longer than times measured for globular proteins in the same molecular weight range (see Table III). For example, the globular and relatively compact 16.8-kDa dimer of interleukin 8 has a measured m of 9.1 ns (26), less than half that of b 53-122 . The long m is consistent with sedimentation velocity data in indicating a highly elongated shape.
Disulfide Bond Formation-To look for sites of interaction between the two b subunits within the dimerization domain, each residue from Ala 103 to Glu 110 was individually mutated to a cysteine in b 24 -156 , and the tendency of the resultant proteins to form intersubunit disulfide bonds was examined. The partially purified proteins were reduced with dithiothreitol and then dialyzed against buffer containing 10 M CuCl 2 and 10 mM cysteine for 24 h, a technique we have used previously (17). The rationale behind this method is that Cu 2ϩ -catalyzed disulfide formation will occur between the polypeptides only if the cysteinyl residues are close together in the dimeric structure; otherwise mixed disulfides between the cysteinyl residue of the polypeptide and a free cysteine in solution will form. Upon analysis by nonreducing SDS-PAGE, the interpeptide disulfide will migrate as a dimer, whereas the mixed disulfide will migrate as the monomer. In our experience with soluble forms of b, inclusion of free cysteine in the buffer is required to ensure specificity of disulfide bond formation. In the absence of free cysteine, all cysteinyl residues tested thus far have formed disulfide bonds readily, probably because of conformational flexibility within the hydrophilic domain of the b subunit when it is not incorporated into F 1 F o . As seen in Fig. 3, treatment with Cu 2ϩ in the presence of cysteine led to formation of dimers to a major extent for the protein containing the A105C mutation, to a minor extent for the I109C mutation, and possible traces of dimer can be seen for Q106C and E108C, whereas other proteins remained entirely monomeric. In additional experiments (data not shown), Western blot analysis confirmed the presence of dimeric b species for the samples indicated, and the bands were missing in samples treated with dithiothreitol, confirming that they contained disulfide linkages. These results indicate that the residue at position 105 of one b subunit must be proximal to its counterpart in the other b subunit. The other residues in this region either are not close enough or have incorrect geometry to form disulfide bonds efficiently.
The cysteine-containing b 24 -156 proteins that did not form disulfides efficiently on their own were mixed in all possible combinations of pairs, and the experiment was repeated. No significant disulfide bond formation was observed in any case (data not shown). DISCUSSION In the work presented here, we have studied the structure of the b subunit of ATP synthase by the use of deletions, hydrodynamic analysis, and disulfide formation between introduced cysteines. These results have given a clearer picture of the b subunit architecture, providing some surprises.
Strength of b Dimerization-Sedimentation equilibrium  b Frictional ratio values were calculated from the data in the present work or taken from the reference except for the cortexillin fragments. The cortexillin f/f min values were calculated from the reported molecular weight, s 20,w value and an assumed partial specific volume of 0.73. c NA indicates that the value was not available. d Designations for coiled-coil fragments of cortexillin (24).
analysis of the polar domain of b at 20°C (10,12,13,17) has generally given molecular weight values consistent with near complete formation of dimer. Nevertheless, a careful analysis of the effect of concentration on the sedimentation coefficient (27) strongly suggested that a fraction of the subunit is monomeric at low concentrations, implying a relatively weak dimerization constant. The sedimentation equilibrium studies showed that b can be converted to a largely monomeric form by simply raising the temperature to 40°C, and the cross-linking studies confirmed that subunit exchange between different b constructs occurs readily at room temperature. The monomer/ dimer transition was readily reversible by a number of criteria, and the equilibrium displayed the expected concentration dependence.
The ease of thermally melting the cytoplasmic domain of b raises the question of the dimerization under growth temperatures, which are usually 37°C but may be as high as 42°C. It should be recognized that a concentration-dependent dynamic equilibrium exists between the monomer and dimer, and many factors outside the dimerization domain could influence this equilibrium. Among these are the membrane-embedded N-terminal domains of b (28), which would act to favor dimerization by holding the cytoplasmic domains in proximity, and interactions with ␦ (27) or possibly other parts of F 1 or F 0 . For these reasons, we doubt that the cytoplasmic domains would become monomeric at physiological temperatures. However, the weak nature of the b-b interaction implies that it may be readily influenced by stronger interactions in which the dimer may participate. Thus, it is highly possible that the b structure may change upon interaction with F 1 and that it may even vary during the course of catalysis.
It should also be noted that the weak dimerization of b complicates determination of its affinity for F 1 or for the ␦ subunit. Because these interactions depend on b in the dimeric form (12), there is no doubt that their apparent weakness (10,27) results in part from the monomerization of b at low concentrations.
Boundaries of the Dimerization Domain-When the current project was undertaken, we expected that the C-terminal boundary of the sequence required for dimerization would fall somewhere between the end of the hydrophobic sequence V 124 AILAVAGA 132 and the C-terminal Leu 156 . This expectation was based on evidence that mutations within the 124 -132 region cause defective assembly of ATP synthase (13), that substitution of aspartate for Ala 128 within the expressed polar domain of b produces an entirely monomeric protein (13), and that mutant forms of b with cysteines incorporated at positions 124, 128, or 132 have a strong tendency to form intersubunit disulfides (17). We were therefore surprised to discover that the protein remained dimeric following deletion of this region by truncation after residue Lys 122 . The studies presented here indicate that the thermal stability of the b 53-122 construct was similar to that of the complete polar domain, implying that residues outside the 53-122 region contribute little to the stability of the dimer.
Heterodimer formation between b 53-122 and b 34 -156 in the cross-linking studies confirmed the assignment of the minimal dimerization sequence by an orthogonal technique. If residues outside the b 53-122 sequence contributed significantly to interactions between the subunits or if the intersubunit interactions were significantly different in b 53-122 from those in the larger construct, one would expect that b 34 -156 would have preferentially formed homodimers. Because the observed likelihood of heterodimer formation was similar to that of homodimer formation, we conclude that the intersubunit interactions in b 53-122 faithfully reflect those in the entire polar domain. At present we cannot definitively reconcile the apparent close intersubunit association in the region containing residues 124 -132 with its lack of effect on dimerization.
Structure of the Dimerization Domain-Both the sedimentation coefficient and the global rotational correlation time of b 53-122 , determined at pH 5.0 where dimerization is nearly complete, revealed the dimerization domain to be highly extended. Under these conditions it is likely that the measured m (20.4 ns) for b 53-122 is a result of a distribution of correlation times for rotation about the long and short axes. In this case the increased magnitude of the measured m compared with that of globular proteins of the same molecular weight should be regarded as resulting from slow tumbling about at least one axis. The frictional ratio of b 53-122 shows the dimerization domain to have a degree of extension similar to that of a coiled-coil of the same molecular weight.
The existence of b 53-122 as a pair of parallel helices is also supported by the finding that cysteines introduced at position 105 had a strong tendency to form disulfides, whereas a weaker tendency was exhibited by positions 106 and 109. This periodicity is consistent with that of helices and is reminiscent of the pattern reported previously for cysteines introduced in positions 124 -132 (17). A significant difference, however, is that the disulfide formation at the nonoptimal positions was much lower in the 103-110 set compared with the 124 -132 set, suggesting that residues within the dimerization domain are more constrained than those in the 124 -132 region.
One of the cysteine mutations, Q104C, has been previously examined in the context of purified F 1 F 0 (29). Disulfide formation could be induced by treatment with Cu 2ϩ , so it was surprising that in the present examination of the cytoplasmic domain of b by itself, disulfides formed readily with cysteine at the 105 position but not significantly at the 104 position. It would be interesting to examine the tendencies of additional positions in the 103-110 region toward disulfide formation in ATP synthase, because the pattern obtained might provide direct evidence of conformational changes in b upon F 1 binding.
Domain Structure of the b Subunit-The present assignment of residues 53-122 as the dimerization domain suggests a division of the b sequence into four domains (Fig. 1A). The Nterminal 24 residues are largely hydrophobic and are essential for membrane anchoring (10). Recent studies have defined their helical structure in a solvent of chloroform/methanol/ water, and the tendencies toward disulfide formation between introduced cysteine residues have suggested a dimerization interface within the membrane (28). The sequence between this membrane-spanning domain and the dimerization domain is denoted here as the tether domain. The membrane-proximal segment of this domain is relatively hydrophobic and may interact with loop regions of the a or c subunits. The highly conserved residue Arg 36 in this region is essential for normal coupling between F 1 and F 0 (30). On the basis of our hydrodynamic analysis, the tether domain extends notably from the dimerization domain and may be flexible (14), but little else is known about its structure.
Following the dimerization domain is the enigmatic V 124 AILAVAGA 132 region, and residues nearer the C terminus, which are known to be involved in the interaction with ␦ (14); here we have denoted these regions together as the ␦-binding domain. Our evidence suggests that this region has a loosely folded structure not required for dimerization that may be disrupted either by deletion of four residues from the C terminus, as suggested previously (14), or else by cooling to 5°C. Interpretation of the sedimentation velocity results were slightly complicated by the weak dimerization of b, such that a small but significant proportion of the subunit is present as monomer under the conditions of the experiment. However, it appears that this proportion is similar at 5 and 20°C and so cannot account for the large effect. A second possible complicating factor, the higher order aggregation observed in the sedimentation equilibrium experiments at 20°C, does not appear to be a factor during the shorter sedimentation velocity runs. 3 Thus, the lower sedimentation coefficients measured at 5°C for b constructs containing the intact C terminus can be interpreted most reasonably as a cold-induced unfolding, suggesting the importance of hydrophobic interactions in the folded structure.
Functional Implications of the Current Findings-The role of the b 2 ␦ stalk in the mechanism of ATP synthesis is not yet fully understood. In the structures observed in the electron micrographs, the b subunits likely form the narrow linker from the membrane to F 1 (31). The highly elongated soluble constructs analyzed in this study are consistent with that role. We have shown that the b monomer-dimer transition is dynamic and reversible and depends only on interactions between residues 53 and 122. The potential functional importance of an elastic b structure as an energy storage device has been pointed out by Englebrecht and Junge (32); the reversibility of the dimerization interaction or conformational changes in the C-terminal domain may provide b 2 ␦ with elasticity.
It is also of interest to note the study of Sorgen et al. (33), who found that internal deletions of up to eleven amino acids (Ala 50 -Ser 60 or Leu 65 -Ile 75 ) could be made in the b sequence without completely eliminating F 1 F 0 function. All of the deletions that extended beyond residue 60 showed significant drops in activity that could only be overcome with overexpression of the mutant b subunits. On the basis of our studies, expressed cytoplasmic domains with such deletions would be expected to be monomeric. The fact that a functional F 1 F 0 complex could still be obtained provided the shortened b was overexpressed implies that the dimeric state can be stabilized by either the membrane anchor, as suggested above, or else by interactions with the F 1 sector.