Two classes of regulatory subunits coassemble in the same BK channel and independently regulate gating

High resolution proteomics increasingly reveals that most native ion channels are assembled in macromolecular complexes. However, whether different partners have additive or cooperative functional effects, or whether some combinations of proteins may preclude assembly of others are largely unexplored topics. The large conductance Ca2+-and-voltage activated potassium channel (BK) is well-suited to discern nuanced differences in regulation arising from combinations of subunits. Here we examine whether assembly of two different classes of regulatory proteins, β and γ, in BK channels is exclusive or independent. Our results show that both γ1 and up to four β2-subunits can coexist in the same functional BK complex, with the gating shift caused by β2-subunits largely additive with that produced by the γ1-subunit(s). The multiplicity of β:γ combinations that can participate in a BK complex therefore allow a range of BK channels with distinct functional properties tuned by the specific stoichiometry of the contributing subunits.

P roteomic 1,2 and functional 2-6 studies have revealed many partners that interact with BK channels, some of which are known to confer distinct tissue-specific functional properties on the BK pore-forming subunit. BK channels are dually activated by membrane depolarization and increases in intracellular [Ca 2 þ ] (refs [7][8][9]. The minimal functional unit of a BK channel is a homotetramer of pore-forming a-subunits (Fig. 1a), each containing intrinsic voltage-sensing and Ca 2 þsensing domains. However, BK channels may also contain any of two different families of regulatory proteins, b and g, which help define tissue-specific functional properties of a BK complex. The four members of the b family (b1-b4) and the four members of the recently discovered g family (g1-g4) differentially regulate BK function, influencing Ca 2 þ -dependence of activation 2,5,6,10,11 , current inactivation 4,5,12,13 and even pharmacology 5,14 .
energetically independent fashion to shift BK gating 20 . For g-subunits, both the position in the channel complex and the a:g stoichiometry remain unknown. In heterologously expressed a þ g1 channels, the g1-induced gating shift occurs in an all-or-none manner, consistent with an elementary functional unit of g1 (for example, monomer, dimer and tetramer) being sufficient to produce the full effect 21 . Here to determine whether different types of regulatory subunits coassemble in the same BK channels, we take advantage of the distinctive functional properties conferred on BK channels by b2and g1-subunits.
Using ensemble and single molecule approaches we report that g1 and 1-4 b2-subunits can be simultaneously present in the same BK channel and independently contribute to modulation of BK function.

Results
b2and c1-subunits coassemble in the same BK channel. The g1-subunit produces a remarkable negative shift of 120-140 mV in the voltage-range of BK channel activation either in the presence or absence of Ca 2 þ (2) (Fig. 1d,e,h,i). While BK channels composed of a alone require at least 10 mM Ca 2 þ to show appreciable open probability over physiologically relevant voltages, (Fig. 1h,i), a þ g1 channels show a similar probability in the total absence of intracellular Ca 2 þ (Fig. 1e,h). The ability of g1 to shift BK gating in 0 Ca 2 þ contrasts with the absence of a gating shift produced by b2 (Fig. 1h) under the same conditions 22 . The most readily identifiable effect of the b2-subunit is essentially complete inactivation that occurs following channel activation (Fig. 1f). Inactivation arises from the cytosolic N terminus of the b2-subunit 4,5,23 . Since BK channels can contain 1-4 b2-subunits, each acting independently, single a þ b2 channels exhibit one of four distinct inactivation rates 20 . The inactivation time constant (t inact ) therefore provides a measure of b2-subunit stoichioimetry within either a channel population or individual channels 20 . Although b2 produces little gating shift at 0 Ca 2 þ (23) (Fig. 1h), gating of a þ b2 channels at 10 mM Ca 2 þ is shifted about 40 mV leftward compared with a channels 3,5,20 which is more clearly observed using a non-inactivating b2 variant (b2DNt) (Fig. 1i). Thus, for a þ b2 channels, voltage steps up to þ 180 mV only weakly activate BK currents at 0 Ca 2 þ (Fig. 1f, left), while 10 mM Ca 2 þ produces robust activation of an inactivating current (Fig. 1f, right), whose t inact approaches a limiting value of B20 ms (Fig. 1j), indicating an average of 4 b2-subunits/channel in the population 20 . Another physiologically relevant property of inactivating channels, the voltage dependence of steady-state inactivation (SS-inactivation), is also an indicator of the presence of b2-subunits. During exposure of a þ b2 channels to 10 mM Ca 2 þ , the availability of noninactivated channels to open is very low at voltages above À 50 mV (Fig. 2a,c) with half-channel availability (V h ) at about À 110 mV. Together, the distinctive effects of b2and g1-subunits provide useful signatures to verify the presence of b2 and g1 in BK channels resulting from the coexpression of a þ b2 þ g1.
We therefore coexpressed g1 þ b2 subunits with a at relative mole fractions of message that would produce full effects of either g1 or b2 alone 20,21 . The coexpression of a þ b2 þ g1 subunits results in a prominent inactivating outward current even with 0 Ca 2 þ (Fig. 1g, left), similar to currents resulting from the expression of a þ b2 subunits when activated by 10 mM Ca 2 þ (Fig. 1f). The complete inactivation of currents obtained after coexpression of a þ b2 þ g1 subunits indicates that essentially all channels contain b2-subunits. Furthermore, that in such patches the voltage dependence of activation at 0 Ca 2 þ is shifted more than À 120 mV ( Fig. 1h) is diagnostic for the presence of g1. When the same patch is activated with 10 mM Ca 2 þ , essentially no current is observed (Fig. 1g, right) which suggests that a þ b2 þ g1 channels are constitutively inactivated with 10 mM Ca 2 þ at À 160 mV. Indeed, the fractional availability of a þ b2 þ g1 currents at 0 Ca 2 þ (Fig. 2b) exhibits a voltage dependence very similar to that observed with 10 mM Ca 2 þ for a þ b2 currents ( Fig. 2c-d). The markedly leftward-shifted steady-state inactivation curve of a þ b2 þ g1 channels also confirms that both b2and g1-subunits can coassemble in the same BK channels.
b2 and c1 occupy distinct positions in the BK channel complex. We next wondered whether channels containing both b2 þ g1 subunits can contain a full set of four b2-subunits. We imagined two kinds of assembly scenarios: (1) an independent model (Fig. 3a), where the presence of g1-subunits does not hinder the ability of four b2-subunits to fully populate a BK channel (b2 and g1 occupy different positions), or (2) an occlusive model (Fig. 3b), where the presence of g1 excludes the assembly of b2-subunits (b2 and g1 occupy overlapping positions). These two models can be tested by examination of the t inact arising from a set of single a þ b2 þ g1 channels obtained under conditions in which relative b2 subunit expression varies 20 . If the independent model is valid, g1-containing inactivating single channels should exhibit four distinct t inact (ref. 20), while the finding of less than four ARTICLE inactivation behaviours supports the occlusive model. By keeping both a and g1 constant but decreasing b2 in the RNA injection mix, we obtained a set of single channels that exhibit inactivation consistent with the presence of b2, but gating shifts consistent with the presence of g1 (Supplementary Fig. 1). The t inact observed from 31 channels appeared to group into four behaviours (Fig. 3c-f). For channels that might potentially contain either three or four b2-subunits, the expected differences in mean inactivation time constant are not great. We therefore examined in detail the ability of three-and fourcomponent Gaussian functions to fit the distribution. In evaluating fits of the distribution, we considered the impact of two criteria. First, since inactivation arises from independent inactivation domains 20 , the faster components should bear a simple arithmetic relationship to the slowest component. Second, the s.d. for any component, although perhaps not wellconstrained by the data, should be smaller than for any slower components, such that s 1 Zs 2 Zs 3 Zs 4 . Various considerations in the fitting process are given in the Methods and in Supplementary  Fig. 2. For the case that the s.d. are constrained as mentioned or where the mean values of each component are defined by the slowest component, the t inact distribution is better fit by a fourcomponent than by a three-component Gaussian distribution (Fig. 3g,h, Supplementary Fig. 2), indicating that channels containing the g1-induced effect can also contain one, two, three or four b2-subunits. b2 and c1 independently contribute to BK gating shifts. Both b2 and g1 produce leftward shifts in BK gating, but their effects are likely mediated by different mechanisms. Whereas the g1-effect can be explained by stronger coupling between the voltage-sensor movement and channel activation 2 , the effects of b2 appear more complex 22,[24][25][26] . We asked whether g1 and b2 effects might be additive or occlusive. For better elucidation of the b2 effects, we compared the gating shift resulting from the coexpression of g1 þ b2DNt versus that produced by each construct separately when coexpressed with BK-a-subunits (Fig. 4). The V h arising from the simultaneous presence of g1 and b2DNt approximately reflects the sum of the independent effects of g1 and b2DNt alone. These results indicate that, whatever the underlying molecular mechanism of the V h shift produced by the g1-subunit, it is predominantly energetically independent of that produced by the b2-subunit. Furthermore, there is no inhibitory allosteric coupling between the auxiliary subunits themselves.
Other b-subunits also coassemble with c1 in BK channels. Can other b-subunits also coassemble with g1 in BK channels?     An earlier report suggested that the presence of b1-subunits may occlude the ability of g1 to produce its gating shift 2 . Since the overexpression of one subunit might influence the successful expression of another, we used proportions of RNA for each subunit similar to those used in testing g1 þ b2 coassembly with a-subunits. Using a b1 construct in which its N terminus was replaced by the b2-N terminus (b1/b2Nt) so that inactivation reports the presence of b1, we found that all BK channels resulting from coexpression of a þ b1/b2Nt þ g1 simultaneously contain both types of regulatory subunits (Supplementary Fig. 3).

Discussion
The present work unambiguously shows that two different types of non-pore-forming regulatory subunits, b2 and g1, can coassemble in the same functional BK channel and independently regulate channel function. The a þ b2 þ g1 combination generates a BK channel with novel functional properties, which subtly change depending on the stoichiometry of b2 in the multimeric complex. In a normal cellular environment, the simultaneous presence of b2 þ g1 in BK channels might effectively remove them from availability for activation especially when intracellular calcium is increased. However, at 0 Ca 2 þ , such channels would not be fully inactivated (Fig. 2d): the window of overlap between the fractional availability curve and the activation curve spans approximately À 60 to À 30 mV, a range corresponding to normal resting potentials in many types of cells, including neurons. This essentially defines a range of voltage over which any contributions of a þ b2 þ g1 channels might be dynamically regulated by inactivation. Given the large conductance of the BK channel, only a small fraction of the total BK population would have to undergo cycles of inactivation, recovery from inactivation and activation to have some influence on cell excitability. Although it is unknown whether b2 þ g1 subunits are simultaneously present in any cell-type, reported message levels of both b2 and g1-subunits in some tissues 11,27 support the possibility that band g-subunits will copartner in at least some cells. For any cells which may express both a b2and a g1-subunit, our results establish that these two distinct regulatory partners of BK channels can simultaneously and independently contribute to modulation of BK function and do not appear to hinder the assemble of each other into a channel. The possibility of coassembly would also likely apply to other members of the b and g families, as supported by our results with b1 þ g1 subunits. Our findings highlight not only the critical importance of defining the identity of protein partners in native multimeric complexes within a cell or a specific cell location, but also the importance of understanding how individual contributions of distinct regulatory components and their stoichiometry can define fundamental properties of the complex. Data analysis. Analysis were made using Clampfit (Molecular Devices). For noninactivating currents, conductance values (G) were obtained from the tail currents, while for inactivating currents the peak current was used. G-V data sets were fitted with a single Boltzmann function: GðVÞ ¼ Gmax

Methods
, where V h represents the voltage of half activation and z is the valence of the voltage dependence. Notice that the a þ g1 injection ratio used (1:4 molar ratio) results in around 90% of BK channels expressed containing the full g1-induced effect 21 , in which case the fit to a single Boltzmann distribution provides a good estimation of V h and z for a þ g1 currents. Macroscopic inactivation time constants (t inact ) were obtained by fitting each current decay to a single exponential function. All error estimates are SEM. Single-channel traces were first processed using digital subtraction of leak and capacitive current defined from traces lacking any channel opening. t inact for each single channel was estimated from the ensemble current average of 50-100 identical sweeps recorded at 0 Ca 2 þ . The histogram of t inact was generated using a bin size of 2 ms and the distribution was fitted to the sum of three or four components Gaussian function: A n e x À tn ð Þ 2 2s 2 n where A n , t n and s n represent the amplitude, mean-t inact and s.d. of each component, respectively, with n ¼ 1 for the parameters of the slowest inactivating component, n ¼ 2 for the second slower and so on. Since each measured t inact represents the mean of an exponentially distributed population, the s.d. for the average of those grouped in the slowest component of the t inact -distribution should be larger than for faster components, such that s 1Z s 2Z s 3Z s 4 . Fitting to the sum of three or four Gaussian functions was first made allowing free variation of all fitting parameters. However, in both cases the rank of s.  Fig. 2a-b). Two factors may contribute to such deviations: first, that the numbers of entries in the histogram are insufficient to accurate define all aspects of each component and, second, that some components may actually arise from multiple components. We then constrained the s.d. in both cases to fulfil the expected criteria (s 1Z s 2Zy s n ) (see results in Fig. 3g-h). Furthermore, we took advantage of the idea that inactivation results from the independent movement of each inactivation domain to the central cavity of the pore 20 . Based on this latter consideration, in some cases we also constrained all mean values of each component to be dependent on the slowest component ( Supplementary Fig. 2c-d).
All cases yield better fits using four than three-component Gaussian distributions.
Chemicals. Salts and buffers were obtained from Sigma.