β2 and β3a regulatory subunits can coassemble in the same BK channels

Ca²⁺- and voltage-dependent large conductance Ca2⁺-activated K⁺ channel (BK)-type, large-conductance K⁺ channels exhibit extensive functional diversity that arises in part from tissue-specific expression of regulatory subunits (Gonzalez-Perez and Lingle, 2019) that coassemble with the pore-forming α subunit encoded by a single KCNMA1 gene (Butler et al., 1993; Salkoff et al., 2006). Three distinct families of regulatory subunits have been identified to date (β, γ, and LINGO) (Li and Yan, 2016; Gonzalez-Perez and Lingle, 2019; Dudem et al., 2020). Of these three families, the physiological roles of various β subunit isoforms have probably received the most attention, but many important questions remain unaddressed. The β subunit family (β1–β4) (Knaus et al., 1994a; Wallner et al., 1999; Xia et al., 1999; Brenner et al., 2000; Meera et al., 2000; Uebele et al., 2000; Weiger et al., 2000; Xia et al., 2000) includes four homologous mammalian genes (human: KCNMB1–KCNMB4; mouse: Kcnmb1–Kcnmb4), and the expressed subunits each uniquely influence functional properties of the resulting BK channels. This includes the range of voltages over which BK channels activate at a given Ca²⁺ (Nimigean and Magleby, 1999; Cox and Aldrich, 2000), the kinetics of activation and deactivation (Cox and Aldrich, 2000; Orio and Latorre, 2005), and the kinetics of BK channel inactivation onset and recovery (Wallner et al., 1999; Xia et al., 1999; Xia et al., 2000; Zeng et al., 2007). β subunits also impact such properties as single-channel current rectification (Zeng et al., 2003) and pharmacology (Xia et al., 1999; Meera et al., 2000; Zeng et al., 2003). Furthermore, the functional diversity arising from β subunits is influenced by the stoichiometry of assembly of a given β subunit (i.e., 0–4 β subunits per channel) in a BK complex (Ding et al., 1998; Wang et al., 2002); one unexplored question is whether two different β subunits can assemble in the same BK complex, in cells in which multiple types of β subunits may be expressed.

Despite the availability of screening techniques for RNA in different tissues and cells, there is little information regarding the extent to which coexpression of different β subunits may occur in the same cells. Although the Allen Brain Atlas qualitatively suggests regions of β subunit overlap (https://mouse.brain-map.org/experiment/show?id=81600586), one challenge is that message levels for some β subunits, as assessed by quantitative PCR in some tissues/cells, are remarkably low even in cases where BK currents may clearly exhibit the functional signature conferred by a given β subunit, e.g., β2 in mouse adrenal chromaffin cells (Martinez-Espinosa et al., 2014). To our knowledge, only one paper has proposed, based on functional criteria, the possible presence of two distinct β subunit isoforms. Specifically, based on functional and pharmacological criteria, cerebellar Purkinje cells have been proposed to express both β2-containing and β4-containing BK channels, perhaps segregated into different parts of the cell (Benton et al., 2013). Although it might be assumed that different BK β subunits should be able to coassemble in the same channels, this has not been tested, and the cerebellar results raise the possibility that segregation is possible. Recently resolved cryo-EM structures of the BKα/β4 complex (Tao and MacKinnon, 2019) have unveiled extensive inter-subunit contacts between the extracellular loops of adjacent β4 subunits, implying a potential symmetric requirement when multiple β subunits coassemble with BKα. As the extracellular loop structure may vary among different BK β subunits (Zeng et al., 2003), it is important to ascertain whether diverse β subunits can coassemble in a single BK channel and affect the function of the resulting channel. Here, we have tested the consequences of heterologous coexpression of β2 and β3a BK β subunits.

Taking advantage of distinct functional differences conferred by β2 and β3a subunits on BK channels, we use a combination of macroscopic current recordings, single-channel recordings, and quantitative biochemical evaluation of complexes to ask whether β2 and β3a subunits coassemble in single BK channel complexes. Both β2 and β3a subunits produce kinetically similar rates and extent of inactivation. However, they differ markedly in properties of tail currents during recovery from inactivation. For β2-mediated inactivation, there is almost no tail current over a time period that results in full recovery from inactivation (Solaro et al., 1997; Xia et al., 1999). In contrast, for β3a-mediated inactivation, a prolonged slow tail current is observed that results in a net current flux that exceeds by several fold that expected for a channel population recovering in accordance with a simple open channel block behavior (Zeng et al., 2007; Gonzalez-Perez et al., 2012). By determining the maximal BK current in a patch (which can be done by rapid removal of inactivation by trypsin), the slow tail currents provide a measure of the fraction of channels that have inactivated via a β3a N terminus. Using measurements of changes in tail current amplitude and the fraction of channels that are non-inactivating at different time points during trypsin digestion, we develop models for expectations based on either segregated assembly of β2 and β3a subunits vs. random mixing of subunits. The analysis depends on the application of a trinomial distribution model to evaluate predicted behaviors arising from the random mixing model, in comparison to the sum of two binomial distributions to describe the segregated assembly model. The results provide strong support for the idea that the distribution of β2 and β3a subunits among BK channels, when coexpressed in Xenopus oocytes, results from random assembly into the four β subunit–binding positions in each BK channel. Single-channel results, although not being sufficient to test for independence of subunit assembly, provide direct support that both β2 and β3a subunits can be present in the same BK channels, likely reflecting the full set of possible stoichiometries. Finally, quantitative biochemical separation following coexpression of tagged BK α, hβ2, and hβ3a subunits directly supports the idea that ternary (α:β2:β3a) complexes, of different stoichiometries, do form. Together, these results argue that different β subunits can coassemble in the same BK complexes, adding additional nuance to the potential physiological diversity that BK channels can adopt in native cells.

Stage IV Xenopus oocytes were obtained from Xenopus 1. BK channels were expressed by cRNA injection of the pore-forming mouse Slo1 (mSlo1) α-subunit alone or with human β2 (hβ2), mouse β3a (mβ3a), or mixtures of hβ2:mβ3a in defined ratios. Mouse β3a was preferred over human β3a because mβ3a produces faster and more complete inactivation and more prolonged tail currents compared with hβ3a, each feature being advantageous for the experiments described here (Zeng et al., 2008). cRNAs were injected at nominal ratios by weight. In preliminary tests with hβ2:mβ3a coexpression, we found that equimolar β2:β3a injection ratios yielded currents similar to those with β2 alone. Thus, β2:β3 injection ratios needed to be strongly biased toward higher mβ3a quantities to obtain clear indications of the presence of channels containing mβ3a. Yet, lower quantities of mβ3a cRNA coinjected with α alone successfully resulted in robust expression of β3a-containing channels when expressed in the absence of β2. Because of this, we tested changes in the amount of BK α expressed with different β2:β3a ratios and also remade cRNA multiple times without identifying any conditions that increased relative effectiveness of mβ3a assembly into channels when coexpressed with hβ2. Therefore, for experiments in which we have quantitatively evaluated the impact of different hβ2:mβ3a nominal expression ratios, both hβ2 and mβ3a cRNA were prepared in parallel with the same reagents, and tests of different ratios were all done from the same stock solutions and in the same batch of oocytes. For coexpression of Slo1 with hβ2, we used a fixed molar ratio of hβ2:mSlo1α of 4:1. For coexpression of Slo1:hβ2:mβ3a, the hβ2: mSlo1 ratio was fixed at 1:1, with mβ3a added to produce the hβ2:mβ3a ratios specified in the text. Under these conditions, this insured that there would always be a quantity of hβ2 sufficient to produce near full stoichiometric occupancy of β subunit sites of interaction.

During seal formation, oocytes were bathed in a standard frog ringer (in mM): 115 NaCl, 2.5 KCl, 1.8 CaCl₂, and 10 HEPES, at pH 7.4. Inside-out patches from Xenopus oocytes were used in all experiments. Currents were recorded using an Axopatch 200B amplifier (Molecular Devices), low pass-filtered at 10 kHz, and digitized at 100 kHz. Following excision, patches were quickly moved into a flowing low Ca²⁺ solution. The pipette/extracellular solution contained (in mM): 140 K-methanesulfonate, 20 KOH, 10 HEPES, and 2 MgCl₂, adjusted to pH 7 with methanesulfonic acid. The cytosolic face of the excised patches was perfused with intracellular solutions containing either 0 or 10 μM Ca²⁺ with the following (in mM): 140 K-methanesulfonate, 20 KOH, and 10 HEPES, pH adjusted to 7.0 with methanesulfonic acid, and one of the following: 5 EGTA (for nominally 0 Ca²⁺) or 5 HEDTA with added Ca²⁺ to make 10 μM Ca²⁺. A commercial set of Ca²⁺ standards (WPI) was used for calibration of Ca²⁺ concentrations with a Ca²⁺-sensitive electrode (Orion, Thermo-Fisher Scientific).

Experiments in which trypsin was used to remove BK inactivation were as previously reported (Zhang et al., 2006; Zhang et al., 2009). To summarize, a 0 Ca²⁺ solution containing 0.05 mg/ml trypsin (cat #T0303-1G; Sigma-Aldrich) was applied for timed intervals, ranging from 2.5 to 20 s. Both before and after each timed trypsin application, the patch was washed with enzyme-free 0 Ca²⁺ solution for 5 s. Between 0 Ca²⁺ applications, the patch was exposed to 10 μM Ca²⁺ solution, during which protocols to test for BK current properties were applied. Plots of trypsin removal of inactivation reflect the cumulative time a patch was exposed to a given trypsin concentration. An SF-77B fast perfusion stepper system (Warner Instruments) was used to switch among solutions and was controlled via pClamp software (Molecular Devices). Experiments were at room temperature (∼22–25°C). All chemicals were obtained from Sigma-Aldrich.

Data were analyzed using OriginPro 7.5 (OriginLab Corporation), Clampfit (version #9.2.1.9) (Molecular Devices), Microsoft EXCEL, or programs developed in this laboratory. Error bars in the figures represent SDs. Each measurement is averaged from at least four experiments.

Eq. 7a and Eq. 7b give, respectively, the changes in distributions of intact β2 subunits and IR-β2 subunits, and the same for the β3a-containing population, as trypsin digestion proceeds. Eq. 8 essentially defines the changes, based on two bionomial populations, in the expected fraction of channels of any starting composition to become non-inactivating (steady-state current). Similarly, changes in the fraction of channels that will have β3a-mediated slow tail current can be determined.

These equations are identical to the direct analytical descriptions given in Eqs. 4 and 5 to describe removal of inactivation and presumably can be used to fit predictions of digestion time course at different mole fractions of β2 and β3a subunits.

Changes in slow tail current behavior arising from inactivation mediated by the β3a N terminus depend not only on the number of β3a N termini in a given channel but also, for channels in which both β2 and β3a N termini are present, on the relative likelihoods (L_β2, L_β3a) that each type of N terminus may contribute to inactivation at the end of a depolarization. For any given channel with x β2 N termini and y β3a N termini (P(x,y), when L_β3a = L_β2, the likelihood that a channel with a given β2:β3a stoichiometry will be inactivated by β3a (pi_β3a) at the end of a depolarization is then given by pi_β3a = yL_β3a/(yL_β3a+xL_β2). For cases in which the relative likelihood is not identical and L_β3a = kL_β2, then pi_β3a = yL_β3a/(yL_β3a+xL_β3a/k). For channels of P(3,1), P(2,2), and P(1,3) with L_β3a = L_β2, pi_β3a = 0.25, 0.5, and 0.75, respectively, while for L_β3a = 3L_β2, pi_β3a = 0.5, 0.75, and 0.9.

In assessing the behavior of single channels arising from hβ2:mβ3a coexpression, similar considerations were used to make estimates of likely hβ2:mβ3a stoichiometry that might underlie the observed fractional occurrence of mβ3a-type tail current behavior (Fig. 8 C).

For protein expression, we used a modified human Slo1 construct, featuring a truncation in the RCK1–RCK2 linker and the C-terminal end, to improve biochemical behavior, used previously for single-particle cryo-EM reconstructions of hSlo1 (Tao and MacKinnon, 2019). This construct (hSlo1_EM) was tagged on the C terminus with a twin-strep tag with an intervening 3C-protease cleavage site. Full-length human β2 and human β3a genes were C-terminally tagged with mCherry-FLAG and eGFP-ρ1D4 tags, separated from the C terminus via a 3C-protease cleavage site. Plasmid DNA of these expression constructs, in pEG vectors, was isolated on a large scale using endotoxin-free Megaprep kits (Qiagen) and used for protein expression.

For the 1:1:1 (TR1) transfections, 300 ml suspension cultures of HEK293F cells, maintained in Gibco Freestyle 293 media, supplemented with 2 % Heat-Inactivated FBS (Gibco), were mixed with 225 μg each of Slo1-TS, β2-RF, and β3-Gρ, preincubated for 10 min at room temperature, with 2.025 mg of PEI (linear 25 kDa) in 30 ml of OptiMEM media. For 1:0.2:1.8 (TR2) transfections, the total weights of plasmids used for transfection were 225 μg of Slo1-TS, 45 μg of β2-RF, and 405 μg of β3-Gρ. After transfection, cells were grown at 37°C for 10–14 h, after which sodium butyrate was added to a final concentration of 10 mM, transferred to 30°C, and grown for another ∼55 h. Cells were pelleted, washed with PBS, and frozen until use. Each 300 ml culture yielded 4–5 g of wet pellets.

Frozen cell pellets were thawed on ice, resuspended in 20 ml of chilled buffer (1 M KCl, 100 mM Tris-Cl, 40 % vol/vol glycerol, 10 mM CaCl₂, and 10 mM MgCl₂,_, pH 7.8), and sonicated to generate a homogenous suspension. 20 ml of 2 % w/v digitonin was added to the homogenate, and the mixture was gently agitated at 4°C for 1.5 h. The lysate was spun at 100,000 g for 45 min 20–40 μl of the supernatant was preserved for fluorescence size exclusion chromatography (FSEC) analysis, and the remainder (∼40 ml) was incubated with 0.5-ml streptactin XT resin for 12–16 h. The resin was pelleted by spinning at 3,000 rpm for 10 min and subsequently washed five times, each time with 4 ml of wash buffer (WB: 500 mM KCl, 50 mM Tris-Cl, 20 % vol/vol glycerol, 0.1 % wt/vol digitonin, 10 mM CaCl₂, and 10 mM MgCl₂, pH 7.8). Total protein was eluted in 4 ml of WB supplemented with 50 mM biotin. 2 ml of the total Slo1 isolated was incubated with 50 μl of anti-FLAG M2 antibody resin (Sigma-Aldrich) or 50 μl anti-ρ1D4 resin (Cube Biotech) for 6 h at 4°C with gentle shaking. The resin was washed three times with 5× resin volume of WB and eluted in a total of 450 μl of WB supplemented with 0.2 mg/ml 3× FLAG peptide (anti-FLAG resin) or 450 μl of WB supplemented with 1 mM ρ1D4 peptide (3 steps, incubating the resin each time with 150 μl of elution buffer for ∼1 h). 400 μl of the FLAG-eluted protein was incubated with 50 μl equilibrated anti-ρ1D4 resin. Before incubation, the resin was thoroughly depleted of equilibration buffer (WB) by centrifuging the washed resin in spin columns (14,000 rpm for 1 min). After 4 h of incubation, the flow-through was collected by spin filtration, and the volume of the recovered flow-through was verified to ensure that it was within 95% of the input volume. Resin-bound protein was washed and eluted as described. Similarly, the ρ1D4-eluted protein (from the second step affinity stage) was purified via anti-FLAG affinity purification.

All results that report the mean and SD of some parameter are derived from some number (n) of patches collected from separate oocytes. When averaged data were fit with some function, parameter values are reported along with 90% confidence limits of the fit. However, statistical comparisons done on parameter values involve comparisons of individual parameter estimates, each from a distinct patch. All experiments that evaluate different ratios of injected β2 and β3a subunits utilized a single batch of β2 RNA and single batch of β3a RNA. When the ability of different models to fit data is compared, e.g., the time course of removal of inactivation for different β2:β3a injection ratios, we determined sum-of-squares not only for the entire data set but also for each β2:β3a ratio, allowing a better evaluation of whether a given model improves the quality of fit over all experimental conditions.

Fig. S1 demonstrates that slow β3a-mediated tail current time constants do not change as the average fraction of β3a subunits in a BK complex is reduced. Fig. S2 shows that differential likelihoods that a β3a vs. β2 N terminus will produce inactivation are unlikely to confound tests between the random mixing vs. segregated models. Fig. S3 show that partial occupancies of BK channels by β3a and/or β2 subunits are unlikely to confound the ability to distinguish between random mixing vs. segregated models. Fig. S4 compares global fits of the random mixing model vs. segregated model of the time course of trypsin-mediated removal of inactivation over 6 subunit injection conditions, utilized different constraints on model parameters. In all cases, the random mixing model yields better fits.

Biochemical (α+β1) (Knaus et al., 1994b) and structural (α+β4) (Tao and MacKinnon, 2019) work has indicated that BK pore-forming α subunits assemble in a complex of up to 4:4 with BK β subunits. A similar conclusion was reached based on functional properties of heterologously expressed inactivating α:β2 (and α:β1−β2ΝΤ) channels (Wang et al., 2002), with the nuance that functional BK complexes with less than a full complement of β subunits could also occur, a phenomenon that can also occur in native cells (Ding et al., 1998). Here, we address whether, when two β subunit isoforms are expressed in the same cell, they segregate (Segregated Model) into different BK channel complexes or can they coassemble (random mixing model) in the same complexes? We utilize three distinct approaches to address this question: first, the use of macroscopic current properties in patches; second, single-channel recordings; and, third, biochemical tests utilizing heterologous expression of tagged subunits. We begin with the macroscopic current approach, since this potentially permits a quantitative evaluation of the stoichiometry of assembly of different variants of β subunits within a large population of channels.

The macroscopic approach is predicated on the idea that, for two β subunit isoforms that inactivate with different properties, perhaps differences in inactivation behavior might serve as a reporter for the presence or absence of the fraction of channels that become inactivated by a given β subunit isoform in the channel population. Here, we summarize properties of inactivation mediated by the β2 and β3a isoforms when expressed alone that are critical for this approach. In each case, we presume that, of the four IDs per channel, inactivation arises from binding of one ID per channel, and only one ID can bind at a time (Wang et al., 2002). Thus, the rate of inactivation onset resulting from a channel with four intact IDs is four times that of a channel with only a single intact ID. Another important feature for the present analysis is that recovery from BK β-mediated inactivation is defined by dissociation of the single ID that has produced inactivation (Ding et al., 1998), irrespective of the number of β subunits in the channel. Both of these properties of BK β-mediated inactivation are shared with inactivation of the ShB Kv channel (MacKinnon et al., 1993). These characteristics of BK inactivation have been clearly established for β2-mediated inactivation but also appear to hold for β3a-mediated inactivation.

Macroscopically, the onset of mouse β3a-mediated inactivation is about two to three fold faster than β2-mediated inactivation (Fig. 1, A, B, and E). Despite the general similarity in the onset of β2 and β3a inactivation, they differ markedly in terms of tail currents following repolarizations that lead to recovery from inactivation. β3a-mediated inactivation is associated with a prolonged slow tail of current (Fig. 1 A) that opens essentially instantaneously upon repolarization (Zeng et al., 2007; Gonzalez-Perez et al., 2012; Zhou et al., 2025). In single-channel traces, repolarization of α+β3a channels results in immediate opening to an opening level of reduced conductance compared with normal BK opening (Fig. 1 C, bottom). This behavior has been explained by the idea that β3a inactivation involves entry first into a preinactivated open state (O*), which is in rapid voltage-dependent equilibrium with an inactivated state (I) (Zeng et al., 2007; Gonzalez-Perez et al., 2012; Zhou et al., 2025). This delays the return to a fully O state and delays recovery from inactivation. In contrast, no detectable tail current is observed during recovery from β2-mediated inactivation, either in macroscopic currents (Fig. 1 B) or in single-channel traces (Fig. 1 D). The differences in tail currents provide one of the key tools used here to evaluate the fractions of β2 and β3a N termini present in a channel population when the two variants are coexpressed.

The second critical feature necessary for our tests is that both β2- and β3a-mediated inactivation can be removed quantitatively by carefully timed applications of cytosolic trypsin (Fig. 2, A and B). During digestion by trypsin, the inactivation time constant slows approximately fourfold as N termini are removed, while the fraction of non-inactivating current increases. As described in the Materials and methods, channels only become fully non-inactivating when all N termini are digested. As a consequence, the digestion time course for both α+β2 and α+β3a current is described by I(t)=(1-exp(-t/τ))⁴ (Fig. 2 C). The ability of this function to describe the digestion time course in both cases supports the idea that each channel complex arises from the presence of close to 4 β subunits, each with an intact ID prior to trypsin application, and that all four N termini must be digested to remove inactivation. That a function raised to the fourth power fits adequately also supports the idea that most channels in these patches are fully occupied by 4 β subunits, whether β2 or β3a. Consistent with earlier results (Zhang et al., 2006; Zhang et al., 2009), the rate of removal of inactivation differs markedly between α+β3a and α +β2, with digestion about sevenfold slower for β2. This potentially provides a tool to differentially alter the relative fractions of β2 and β3a N termini in the channel population.

If the β3a-mediated slow tail directly reports on the fraction of channels that have inactivated via a β3a N terminus, trivially the amplitude of the slow tail during digestion by trypsin should be inversely related to the fraction of channels that become non-inactivating (f_ss) at different time points of digestion as observed in Fig. 2 D. In contrast, during digestion of α +β2 current, at the time point at which the β3a slow tail amplitude is readily measured, e.g., 3 ms, there is no change in tail current during α+β2 digestion (Fig. 2 D). For each patch, digestion by trypsin permits normalization of the slow tail amplitude to the total BK conductance in the patch. It should be noted (Fig. 2 B) that the initial amplitude of the slow tail at −160 mV reflects an appreciable fraction (∼0.4) of the total maximal BK conductance in the patch measured at +160 mV (Gonzalez-Perez et al., 2012; Zhou et al., 2025).

Might estimates of the slow tail amplitude be compromised by changes in the slow tail time constant? To evaluate this, we measured α+β3a slow tail currents under two conditions that alter the relative average fraction of β3a IDs in the channel population. First, we measured slow tail time constants and amplitudes (normalized to maximum outward current at +160 mV after trypsin) for α+β3a currents arising from different α:β3a injection ratios ranging from 1:1 to 40:1 α:β3a (Fig. S1, A and B). Although the normalized slow tail amplitude decreased by about 80% over this range of injection ratios, the slow time constants exhibited only modest changes (Fig. S1 B). Second, we measured slow time constants and normalized slow tail amplitude at different times of digestion by trypsin (Fig. S1, C and D). Again, over a large range of reductions in slow tail amplitude as the fraction of non-inactivating current increased, there was no appreciable change in the slow tail time constants (Fig. S1 D). This validates our use of slow tail amplitude as a measure of the fraction of channels inactivated by β3a at the end of depolarizations. It also supports the idea that the slow tails represent the behavior of a single β3a N terminus during recovery from inactivation, since reductions in the average number of N termini in a channel population do not alter the kinetics of that process.

We hypothesized that, by taking advantage of slow tail currents and the differential rates of digestion of β2 and β3a N termini, this may allow tests that distinguish between the Segregated and random mixing models of assembly. Before turning to experimental tests involving β2 and β3a coexpression, we first present the theoretical expectations based on the two models. Do the two models make predictions that may prove to be experimentally distinguishable?

The underpinnings for the two models are given in detail in the Materials and methods. In brief, both models assume four beta-occupancy sites in a channel. The time course of removal of inactivation predicted by the segregated model is given by Eq. 11, which is derived from the sum of two binomially distributed populations (Eq. 8). In contrast, the random mixing model requires a trinomial distribution (Eq. 12) to take into account the occupancy of each site in individual channels by either β2, β3a, or an N terminus–removed β, resulting in a digestion time course defined by Eq. 14.

We first evaluate expectations that arise from coexpression of different ratios of β2:β3a subunits in regard to how slow tail amplitude might be expected to vary simply as a function of mole fraction of β2 in the channel population. For the segregated model (Fig. 3 A), since we assume there is a sufficient quantity of β2 and β3a subunits to saturate available sites of β subunit occupancy, there are only the two stoichiometric possibilities, P(4β2,0β3a) and P(0β2,4β3a), such that, for the case illustrated (Fig. 3 A), the fraction of each combination in the population is simply the injected mole fractions, p_β2 = 0.5 and p_β3a = 0.5. Since slow tail amplitude reflects exclusively those channels that become inactivated by a β3a N terminus, while β2-inactivated channels do not contribute at all to slow tails, the slow tail amplitude is predicted to diminish linearly with the increase in fraction of channels that are inactivated by β2 (Fig. 3 C). For the mixed model (Fig. 3 B), at any given β2:β3a injection ratio, the fractional occurrence of each potential stoichiometry, P(4β2,0β3a), P(3β2,1β3a), P(2β2,2β3a), P(1β2,3β3a), and P(0β2,4β3a) (Fig. 3 B), can be determined from the binomial distribution, since there are no sites lacking a β subunit. The predicted slow tail amplitude is calculated from the likelihood that a channel of each stoichiometry will be inactivated by β3a at the end of a depolarization. For reasons of simplicity, we first assume that each individual N terminus, whether β2 or β3a, has an equal probability of producing inactivation, such that for a 3β2:1β3a channel, the likelihood that it ends up inactivated by β2 is threefold higher, and so on. With that assumption, the slow tail predictions for the mixed model result in a prediction for the relationship between normalized slow tail amplitude and mole fraction of β2 subunits that is identical to the segregated model (Fig. 3 C). This arises because, for any mole fraction injection ratio, irrespective of the model, there will always be the same fraction of β3a subunits in the channel population.

For a channel complex containing mixtures of β3a and β2 N termini, the assumption that any given N terminus has an equal likelihood of occupying the position of inactivation at the end of a depolarization is likely inaccurate. Based on the time constants of inactivation, the rate of inactivation for the β3a N terminus is at least threefold faster than for β2 (Fig. 1 C). However, differences in the lifetime of the inactivated states might also contribute to the likelihood of each type of subunit contributing to inactivation at the end of a depolarization, although we have no experimental evidence that addresses that possibility. Given the differences in rates of β2 and β3a-mediated inactivation, we have evaluated how such differences might impact on the expectations for the relationship between slow tail amplitude at a given β2/(β2+β3a) ratio. For two stoichiometries, P(4,0) and P(0,4), differences in inactivation likelihood can be ignored, while for P(3,1), P(2,2), and P(1,3), we adjusted the likelihoods for inactivation by β3a relative to β2 as described in the Materials and methods. Taking into account the different inactivation rates, a curvature in the expected relationships between slow tail amplitude as a function of the β2/(β2+β3a) ratio (Fig. 3 D) is introduced. The potential impact of the differential likelihoods of inactivation on our tests of models of assembly will be revisited later. Irrespective of any adjustments required to account for differential inactivation likelihoods, the main point to be made here is that simple variation of injection ratios of two different subunit isoforms is not useful in distinguishing between assembly models.

The differences in trypsin digestion rates of β2 and β3a N termini allow the stoichiometric composition of individual BK channels to be altered as a function of cumulative time of trypsin digestion, with uniquely different predictions in accordance with either the segregated model (Fig. 4 A) or the random mixing model (Fig. 4 B). For the segregated model (Fig. 4 A), the distribution of a population of BK channels containing either β2 or β3a subunits or inactivation-removed subunits, as N termini are cleaved, follows expectations given by the sum of two binomially distributed populations. Predicted digestion curves plotting the fraction of non-inactivating current (f_ss) from different starting β2/(β2+β3a) ratios are described by the double exponential given by Eq. 11 in the Materials and methods (Fig. 4 C). Notably, such curves exhibit a clear inflection point.

For the random mixing model (Fig. 4 B), the digestion results in 15 stoichiometric possibilities and requires application of the trinomial distribution analysis (see Materials and methods). We simulated the predicted time course of digestion in two ways: first, point-by-point calculation of the trinomial distribution expected based on changes in relative fractions of p_β2, p_β3a, and p_IR, i.e., the probability of occupancy by β2, β3a, or ID removed β’s (IR), as p_β2 and p_β3a are reduced in accordance with their separate trypsin digestion rates. From any initial starting distribution of occupancies by β2 and β3a subunits, P(β2,β3a), only when channels reach P(0,0) with inactivation fully removed do they contribute to f_ss. The predicted time course of changes in f_ss is also analytically described by Eq. 14. A distinguishing feature between the two models is the behavior at nominal 1:1 expression ratios, indicated by the green lines. Whereas for the segregated model, the curve is a simple double exponential function intermediate between the two extremes; for the mixed model, the curve is strongly shifted toward the β2 time course, despite the equal abundance of β3a in the population. This arises because of the pronounced slowing of digestion by the presence of even a single β2 subunit in a mixed channel.

The shapes of the predicted digestion curves (Fig. 4, C and D) appear sufficiently different that they may be amenable to experimental tests to distinguish between the models. These predicted changes in f_ss depend in no way on the differences in slow tail properties between β2- and β3a-mediated inactivation and are only defined by the relative digestion rates of the two different IDs. As such, the issue raised above regarding the differential likelihood that a channel may be more likely to be inactivated by a β3a ID than a β2 ID at the end of a depolarizing step does not apply.

Each model also makes different predictions in regard to how the slow tail current amplitude will vary as IDs are removed. As mentioned earlier, when only β3a is expressed, the two models are identical. However, as the fraction of β2 is increased, the initial normalized slow tail amplitude is reduced for both models because of the increase in the fraction of channels that are inactivated by β2 (Fig. 4, E and F). Moreover, the segregated model predicts that the changes in slow tail amplitude (A_ST) as a function of f_ss is largely linear as digestion proceeds, consistent with the behavior of BK channels containing only β3a. At the most elevated fractions of β2, there is a wide range of changes in f_ss for which there is no longer any slow tail current, reflecting the slow removal of inactivation of primarily β2-containing BK channels (Fig. 4 E). In contrast, for the random mixing model, at any ratio of both β2 and β3a coexpression, changes in A_ST exhibit marked curvature as IDs are removed. Since these plots are influenced by the properties of the slow tails, we also evaluated the impact of changes in the likelihood of inactivation (L_β3a vs. L_β2) at the end of a depolarization for individual β3a vs. β2 N termini. For the segregated model, since no channels contain both β2 and β3a subunits, any difference in relative β2 and β3a inactivation likelihoods is irrelevant. For the random mixing model, predicted A_ST vs. f_ss curves during digestion were generated for L_β3a/L_β2 ratios of 0.5, 1, 2, 3, and 4 (Fig. S2), with ratios of 3 or 4 being most consistent with the measured differences in inactivation time constants (Fig. 1 C). The differences in the idealized curves obtained for k_β3a/k_β2 of either 1 or 4, although clear (Fig. S2, E and F), are less than observed between the segregated and random mixing models.

For the considerations above, we have assumed that, irrespective of the model, all sites for β subunit occupancy are fully occupied, whether by β2 or β3a subunits. Earlier work, both from native cells (Ding et al., 1998) and with heterologous expression (Wang et al., 2002), indicates that BK channel complexes with <4 β2 subunits can occur. Although the experimentally determined trypsin digestion time courses for β2 and β3a removal of inactivation (Fig. 2) support the idea that almost complete occupancy by β subunits occurs in the present experiments, we were concerned that the occurrence of partial occupancies might compromise our ability to distinguish usefully between segregated and random mixing models. We therefore evaluated the predictions for a segregated model with the assumption that, once one type of β subunit associates in a complex, only other identical subunits can assemble. However, partial occupancy within the segregated populations can occur. Within this framework, we compared the impact of occupancies of 1.0, 0.8, and 0.6 within the segregated β2 and β3a channel populations (Fig. S3). With partial occupancy of 0.6, the fit of the predicted trypsin digestion time course for either the β2 alone population or the β3a alone population requires exponents of <2, clearly less than what is observed experimentally here (Fig. S3 C). However, even if partial occupancy were 0.6, the predicted dependence of slow tail amplitude as a function of the steady-state current at different time points of trypsin digestion would still mirror the linear relationships for the segregated model assuming full subunit occupancies (compare Fig. 4, E and F; and Fig. S3, D and E. Thus, partial occupancy in segregated populations does not predict behaviors that would be consistent with the random mixing model.

Guided by the above considerations, we expressed different ratios of β2 and β3a messages with mSlo1 and hβ2 at a fixed 1:1 weight ratio, with β3a added in to define the nominal β2:β3a ratios. Currents in excised patches were then recorded at different cumulative times of application of 0.05 mg/ml trypsin (Fig. 5, A–F). Patches were bathed with 0 Ca²⁺ during application of trypsin, but sample currents monitoring trypsin digestion were obtained with 10 μM Ca²⁺. In all cases, patches were only used if full removal of inactivation was achieved. As expected, steady-state current measured at the end of the depolarization to +160 mV increases more rapidly with trypsin digestion of patches with a larger fractional expression of β3a. In addition, it will be noted that all the experimentally injected ratios correspond to nominal mole fractions of β3a message much in excess of β2, all being at least 0.8 and higher. Even at the lowest nominal mole fraction of β3a, 0.8 (1β2:4β3a; Fig. 5 B), which would imply a fourfold excess of β3a relative to β2, the slow tail current amplitude remains only about half of that for β3a alone. Despite the large excess of β3a available, slightly less than half the channel population apparently has been inactivated by β3a. Based on the considerations above (Fig. 4 E), a slow tail amplitude that is half that of β3a alone would be expected to occur at 1:1 β2:β3a. This suggests that, under the conditions of our experiments, channel assembly, irrespective of the assembly model, is not occurring in accordance with the nominal mole fractions given by the injected subunit ratios.

To evaluate the adequacy of the two models, for the random mixing model, we fit the family of curves with Eq. 15, while for the segregated model we used Eq. 5. Irrespective of the fitting constraints (Fig. S4), the random mixing model yields a sum of the squares 2.5-fold lower than that for the segregated model. Constraining the digestion time constants to the values estimated from fitting the trypsin digestion time course for β3a alone and β2 alone and assuming all channels begin with n = 4 IDs, the full set of digestion time courses at different injection ratios can be reasonably well fit with a single free parameter (sf) (Fig. 6), yielding an SSQ of 0.0993, not too much different from the SSQ = 0.0929 obtained with 4 free parameters. That the scale factor constant, sf, works across all four injection ratios gives support to the idea that it reflects an intrinsic difference in efficacy of the β2 message or protein expression process in relation to β3a. Whatever the origins of this β2/β3 difference, the key point is that the mixed model better accounts for the observations.

The analysis above utilized only the time course of changes in f_ss as a function of trypsin digestion for different digestion ratios, whereas we showed above that the two models also exhibit different predictions for the behavior of slow tail amplitude as a function of changes in f_ss. Although we do not have an analytic equation that permits fitting of the relationship between A_ST and f_ss for the random mixing model, we compared the predictions for the A_ST and f_ss the relationship predicted based on the fitted values for the random mixing and segregated models (Fig. 7). For each set of patches evaluated with a given β2:β3a injection ratio, f_ss and A_ST values were separately averaged for each time point of trypsin application, resulting in mean ± SD values for both f_ss and the associated slow tail amplitude (A_ST, normalized to the current at +160 mV after trypsin digestion) (Fig. 7 A). From the fitted values obtained from the analysis above (Fig. 6), any given f_ss is predicted to define A_ST for each of the two models. However, rather than simply normalize to the maximum A_ST for the β3a alone data, we chose to step through a set of A_ST values (0.2 to 0.25) bracketing those obtained with β3a alone. This identified the A_ST value that yielded the minimum SSQ for the two cases: random mixing (Fig. 7 B) and segregated (Fig. 7 C). The SSQ for random mixing was almost half of that for the segregated model. It should also be noted that the random mixing model better recapitulates the curvature predicted in the A_ST vs. f_ss relationship than the segregated model (also consider Fig. S3).

We collected single-channel recordings from oocytes expressing BK α together with either β2 or β3a separately and then from patches injected with BK α with a 1:8 β2:β3a injection ratio. Patches were stimulated with a 100 ms depolarization to +160 mV, followed by a 20 ms repolarization to −160 mV and then followed by a second step to +160 mV (Fig. 8). Channel activity was then evaluated for whether the channel opened and inactivated during the initial depolarization, whether there was a tail opening following repolarization, and then whether the channel was again available for activation upon the second depolarization. For β2-containing channels, repolarization was invariably followed by an absence of openings, although more often than not the subsequent depolarization clearly revealed the channel had recovered from inactivation (Fig. 8 A). In contrast, for β3a-containing BK channels (Fig. 8 C), repolarization usually resulted into immediate opening in the O*-I subconductance behavior, which in some cases was terminated by an opening to a full BK tail opening level. Subsequent depolarization revealed that in many cases the β3a-containing channel may have recovered from inactivation, although in cases where a channel remained in the O*-I behavior at the end of the 20 ms repolarization, subsequent depolarization generally failed to produce opening. With oocytes in which β2 and β3a were coexpressed, we obtained 16 patches that exhibited β3a-type tail openings, but 14 of these patches also exhibited repolarization intervals where no opening was observed, but which were generally followed by a subsequent opening during the second depolarization, a characteristic of the β2-containing BK channels. However, such behavior can also be seen, albeit more rarely, for some patches expressing only β3a (trace 3 in Fig. 8 C), so it is not definitively predictive of either β2- or β3a-mediated inactivation. Yet, the presence of the instantaneous reduced amplitude tail openings is only associated with inactivation mediated by a β3a N terminus. As such, we determined the frequency of occurrence of such β3a-type tails for all sweeps for a set of 5 patches from β2 alone oocytes, 6 patches from β3a alone oocytes, and 16 patches with 1:8 β2:β3a coexpression (Fig. 8 D). β3a-type tails were never observed in β2-patches, at a frequency in excess of 0.9 for β3a-patches, and with a broad distribution of frequencies in patches with coexpressed β2:β3a subunits. In principle, the random assembly model predicts four potential stoichiometries containing at least one β3 a subunit (4:0, 3:1, 2:2, and 1:3 β3a:β2). 15 of 16 patches from coexpression had at least 40 trials, but one had only 14. Overall, the range of values is consistent with there being multiple combinations, but the number of patches and probably the number of sweeps for each patch are insufficient to look for groupings.

For the simplest case, where both β2 and β3a N termini may exhibit similar likelihoods of producing inactivation at the end of a depolarization (L_β2 = L_β3a), one would expect groups at 1.0, 0.75, 0.5, and 0.25 fractional likelihoods for 4:0, 3:1, 2:2, and 1:3 β3a:β2 assemblies. Qualitatively, the spread of observed values appears skewed to higher likelihoods of β3a tails (Fig. 8 D). As noted above, differences in inactivation rates between β2 and β3a (note ensemble average time constants; Fig. 8, A–C) suggest that the likelihood of being inactivated at the end of the 100 ms depolarization is higher for β3a than for β2. Although our results are insufficient to make any definitive interpretations regarding stoichiometry, we ranked our observations and placed these over the predictions for cases in which L_β3a/L_β2 varies from 1 to 4 (Fig. 8 E). The measured values are fully consistent with the existence of four possible β2/β3a stoichiometries, with L_β3a/L_β2 being about 2–3. It should be noted that no β2 alone-type channels were observed in the 16 single channels from the coinjection oocytes. Overall, we can conclude unambiguously that both β2 and β3a subunits can coassemble in single BK channels, and the basic inactivation behavior of each of the subunits does not seem altered by the presence of the other isoform.

We co-transfected suspension-adapted HEK293F cells with hSlo1_EM-TS, β2-RF and β3-Gρ (where TS, RF, and Gρ indicate twinstrep, mCherry-FLAG, and eGFP-ρ1D4 tags, respectively, appended onto the C terminus of the expression constructs), using plasmid weight ratios of 1:1:1 or 1:0.2:1.8. These two cases of transfection will be referred to as TR1 and TR2, respectively. Total protein was extracted in digitonin micelles, fractionated using a three-step purification scheme (Fig. 9 A), and each fraction was analyzed with dual-color fluorescence size exclusion chromatography, using eGFP and mCherry fluorescence (Fig. 9 B). In total cellular lysates, based on calibrated fluorescence intensities, the molar ratios of β2-RF:β3-Gρ were found to be ∼ 3.3 ± 0.21 (TR1) and ∼ 0.75 ± 0.18 (TR2) (Fig. 9 C). After the first affinity purification step, which isolates total Slo1, the ratio of eGFP to mCherry fluorescence intensities at the peak retention volume of the chromatogram (ρGC) indicated the presence of β2-RF and β3-Gρ, affiliated with Slo1, in a molar ratio of 7.32 ± 0.48 (TR1) and 1.08 ± 0.16 (TR2) (Fig. 9 C). These values are informed by the relative distributions of the binary (Slo1:β2 and Slo1:β3) and ternary (Slo1:β2:β3) complexes. Two additional sequential affinity purification steps capture Slo1 complexes containing both β2 and β3 subunits. For these triply purified proteins, the FSEC profiles showed robust eGFP and mCherry fluorescence, and ρGC indicated the presence of β2-RF and β3-Gρ in a molar ratio of 2.33 ± 0.11 (TR1) and 0.89 ± 0.09 (TR2) (Fig. 9 C). These reflect ternary complexes, Slo1:β2:β3, of mean stoichiometry 4:2.8:1.2 and 4:1.9:2.1, respectively. The nonintegral stoichiometry of the subunits indicates that the ternary complexes are most likely a heterogenous mix of complexes of different stoichiometries.

Comparison of the peak fluorescence intensities in the input and flow-through fractions of the last affinity capture step also allowed us to estimate the relative distribution of the binary and ternary complexes (Fig. 9 D). As would be expected, the FSEC profiles of the isolated binary complexes exhibited only a single fluorescence signal (Fig. 9 B), distinguishing them from the ternary or the heterogenous mix of complexes in other fractions. Ternary complexes were found to comprise 14 ± 4.1% or 58 ± 6.2% of the total Slo1–β2 complexes in the TR1 and TR2 cases, respectively, the remainder being binary Slo1–β2 complexes. Similarly, we estimated that ternary complexes comprised 49 ± 4.4% and 37 ± 6.6% of Slo1–β3 complexes in the instances of TR1 and TR2. The relative proportion of the ternary complexes inferred from our biochemistry experiments differs from that of the electrophysiological results. This perhaps originates from the use of different expression constructs or systems. Additionally, while electrophysiological methods reflect plasma membrane complexes, the biochemical methods inspect the pool of total (digitonin-extractable) protein. Nevertheless, consistent with electrophysiological results, the biochemistry experiments indicate that under both co-transfection conditions (TR1 and TR2), both binary and ternary complexes co-exist, that ternary complexes comprise a substantial proportion of the total pool of protein complexes, and that the relative proportion of the binary and ternary complexes and the stoichiometry of the ternary complexes vary with the expression levels of the auxiliary subunits.

Through a combination of analysis of macroscopic current, single channels, and biochemical tests, the results show unambiguously that β2 and β3a subunits of BK channels can coassemble in the same individual BK channel complexes. Taking advantage of differences in the time course of trypsin digestion of β2 and β3a N termini and also differences in β2- and β3a-mediated tail current properties, we measured tail current amplitudes and steady-state current amplitudes at different cumulative durations of trypsin application. Coupled with the application of a trinomial distribution analysis, we explicitly evaluated two models by which β subunits might be assembled in BK channel complexes: first, a random mixing model in which the two distinct β subunits can independently become associated with any one of the four total β subunit–binding sites in the BK channel tetramer, and second, a segregated model, in which BK channels can only contain up to four of a single species of BK β subunit. The random mixing model provided the best fit over four tested β2:β3a injection ratios. However, irrespective of the model, the optimal fit obtained by either model required introduction of a scaling factor (sf), which scaled the nominal injected mole fractions of β2 and β3a subunits to adjusted mole fractions. This presumably reflects some intrinsic difference between β2 and/or β3a message, protein synthesis, and/or capacity for assembly resulting in an effective β2 availability being about sixfold greater than that of β3a. The potential underpinnings of the requirement for this sf are addressed below.

The random mixing model provided not only the better fit to the overall data but also better captured the nature of the curvature in the both the digestion time course and also the relationship between tail current amplitude and f_ss. One assumption in our analysis is that all potential sites of β subunit occupancy are filled, whether by β2 or β3a. Although the molar amounts of β2 and/or β3a messages in our experiments are expected to be near saturation, there are suggestions of less than full stoichiometric assembly in some experiments, e.g., even with 4:1 β2:α. For example, the best fit of the trypsin digestion time course for either β2 or β3a can result in best fits with power terms somewhat <4.0. However, our evaluation of predicted power terms suggests that, under the conditions of our experiments, the observed power terms are consistent with both β2 and β3a fractional occupancies, in the worst case being in excess of 0.9 (power term of ∼3.4). We were concerned whether it was possible that segregated assembly might be occurring, but that less than full stoichiometric assembly might alter our digestion time course data to mask the predicted inflection points, making a segregated model appear like a random mixing model. However, our evaluations excluded that possibility.

Whereas the single-channel and biochemical tests provide direct support that coassembly does occur, such experiments are less useful for discerning particular models of assembly. For single channels, the challenge is that it is unrealistic to obtain enough separate patches under various coexpression conditions to determine likelihoods of various stoichiometries quantitatively, while for biochemistry the problem is that the extent of transfection may differ among cells, and furthermore, the purified protein does not exclusively reflect surface membrane channels. As such, although the macroscopic experiments are more indirect in regard to supporting coassembly, to the extent they do support coassembly, they suggest that, at least β2 and β3a, coassembly in BK complexes is guided by independent random assembly of available subunits. However, this may not be the case for other BK β subunits, and it should be noted that during the initial paper describing structures of α+β4 BK complexes (Tao and MacKinnon, 2019), the authors noted that they did not see any evidence for the presence of less than four β4 subunits in their complexes, suggesting that rules for assembly may differ among different β subunits. The approaches utilized in the present work should prove useful in addressing this question for other β subunit combinations for BK channels.

To confirm that both β2 and β3a subunits can be present in the same BK channel complex, single-channel recordings directly demonstrated that, with coexpression of β2 and β3a cRNA, recovery from inactivation of single channels exhibited features of β2-mediated inactivation in some sweeps while, in other sweeps, β3a-mediated inactivation. Although the specific stoichiometry of β2:β3a in the single channels was not determined, the distribution of likelihoods of observing β3a-type tails openings during recovery from inactivation was consistent with patches containing either 1:3, 2:2: 1:3, or 0:4 β2:β3a combinations. Furthermore, biochemical assessments of BK channel composition using differentially fluorescently tagged β2 and β3a subunits expressed in HEK cells directly revealed that ternary complexes of α:β2:β3 form, that that ternary complex comprises a substantial population of all complexes, and that relative amounts of ternary complexes as well as mean stoichiometry of the ternary complexes can be tuned by expression levels of the subunits.

The use of binomial distributions has been a powerful tool in the evaluation of models of protein subunit assembly (Fang et al., 2014) and has been applied to evaluation of a variety of ion channel complexes. For subunits of inwardly rectifying K⁺ channels (Glowatzki et al., 1995; Fakler et al., 1996) and glutamate-activated AMPA receptors (Mansour et al., 2001; Yu et al., 2021), evidence suggests that different isoforms of the pore-forming subunits assemble non-stochastically to form a functional channel. Moreover, in the pLGIC family of channels, it is well known that specific receptor subtypes adopt a preferential configuration, which is an important instance of non-stochasticity. However, in context of the channel-auxiliary subunit-dependent assemblies, non-stochastic assembly does not appear to be common. While cryo-EM reconstructions of native glutamate receptors identify receptors with distinct auxiliary subunits (TARP and CNIH) assembled into a single complex (Yu et al., 2021), the structures are able to reliably define the composition of perhaps only 5–10% of the pool of purified protein (corresponding to the number of particles that can be reliably aligned and refined to high resolution), which in itself might only represent a fraction of the total protein present in the cell. That specific assemblies exist in cells, where the expression of the individual subunits is genetically tuned, does not violate the prediction of binomial or stochastic assembly. Type 1 and 2A sulfonylurea receptors randomly assemble into KATP channel complexes to form heteromeric complexes in accordance with binomial statistics (Chan et al., 2008).

Despite the value of the binomial distribution for many biological questions, there are likely many situations in which other tools are required. Biochemical approaches increasingly have revealed examples of channel complexes with more than two components, e.g., ternary complexes of Kv4.2, with Kchip3 and DPP10 (Jerng et al., 2005; Jerng and Pfaffinger, 2012) and assembly of different KCNE subunits in complex with KCNMQ1 (Morin and Kobertz, 2007). However, as yet tools to address whether assembly is random or constrained do not seem to have been applied. To our knowledge, trinomial distribution analysis has not been used to evaluate questions pertinent to the assembly of molecules. Yet, it has been usefully applied to a variety of topics ranging from estimations of allele frequencies (Lynch, 2009) as an extension of the Hardy–Weinberg principle (Hardy, 1908; Edwards, 2008), financial engineering, stochastic modeling, and modeling outcomes of particle interactions or the particle decay processes in particle physics. In the present case, that a trinomial analysis could be tested in regard to β2 and β3a subunit assembly depended critically upon the unique properties conveyed on BK channels by the β2 and β3a subunits. However, the general approach could be potentially extensible to other cases if one can engineer functional reporters into the components being evaluated.

That equimolar expression of β2:β3a cRNA in oocytes or transfection with equimolar β2:β3a DNA in HEK cells results in a higher mole fraction of β2 subunits in the resulting ternary complexes was unexpected. However, even at the earliest biochemical measurements of molar ratios of β2 and β3a subunits in cell lysates, there is an excess of β2 subunits relative to β3a on the order of three to sixfold. Since the final ratios of β2:β3a in complexes tend to track roughly with the initial ratio in bulk lysates, for the present we prefer the view that something rate limiting in the translation process limits synthesis of β3a. In this regard, it is perhaps significant that when HEK cells are transfected with a ratio of 0.2:1.8 β2:β3a, with β3a message expected to be in ninefold excess, the relative abundance of β3a subunits in the final ternary complexes does not exhibit a similar excess, suggesting that too much β3a transcript may have a negative impact on the generation of the final protein product. Obviously, this work does not address whether there may be similar processes at play in native cells. However, that similar constraints on the eventual likelihood of β2 vs. β3a presence in a BK channel complex occur in both oocytes and HEK cells suggests it may represent differences in something intrinsic to the subunit sequences.

The present results provide clear proof-in-principle that different β subunit isoforms can coassemble in the same BK channels if they are both expressed in a cell. For an apparently well-studied channel, it is remarkable how poorly the subunit composition of BK channels in many tissues is understood. Despite great interest in the diversity of BK channel function that is generated by the β, γ, and LINGO families of regulatory subunits (Torres et al., 2007; Gonzalez-Perez and Lingle, 2019; Dudem et al., 2020) and quite detailed biophysical and mechanistic evaluation of the regulatory effects (Cox and Aldrich, 2000; Contreras et al., 2012; Castillo et al., 2015; Gonzalez-Perez et al., 2018), knowledge of where such subunits are found has not advanced to quite the same extent. There are multiple complicating factors that have hindered progress. For example, for β subunits, there is evidence that specific motifs may influence subunit trafficking (Toro et al., 2006), even to intracellular organelles (Zarei et al., 2007), such that the presence of a message and protein for a subunit in a specific tissue may not reflect cell surface expression. Furthermore, BK regulatory subunits are expressed at low message abundance, even in cases where functional hallmarks of subunit presence are unambiguous.

In some cases, unique functional properties conferred by regulatory subunits have facilitated identification of where they may be expressed. For example, for γ1-containing BK channels, the over −100 mV gating shifts produced by γ1 (Yan and Aldrich, 2010, 2012; Gonzalez-Perez et al., 2014) have enabled confirmation of cellular expression in a number of types of secretory epithelial cells (Yang et al., 2017; Gonzalez-Perez et al., 2021). The more modest gating shifts produced by β1 subunits match with the more negative gating range of BK channels in smooth muscle, while the inactivation properties of β2-containing BK channels (Wallner et al., 1999; Xia et al., 1999) match well with native BK channel inactivation in chromaffin cells (Solaro et al., 1995; Solaro et al., 1997; Martinez-Espinosa et al., 2014), in some pancreatic β cell–derived cell lines (Li et al., 1999), and also some DRG neurons (Li et al., 2007). Similarly, some features of β4-containing BK channels (Brenner et al., 2000), including resistance to block by iberiotoxin, has been associated with neuronal BK channels (Brenner et al., 2005). However, such studies largely reflect the very tip of the iceberg of where BK β subunits are expressed, whether the functional properties have been adequately defined to support identification of potential regulatory subunits, let alone address whether there is overlap in expression.

The challenges of using message as an indicator are apparent in work on mouse adrenal glands and adrenal medullary chromaffin cells. Quantitative PCR reported the presence of Kcnmb2 message in whole adrenal glands (Martinez-Espinosa et al., 2014), although at levels less than Kcnmb4 expression and two orders of magnitude less than the Kcnma1 message. Expression of a message of a regulatory subunit at such small fractions of the associated pore-forming subunit might raise questions about its relevance, but it is unambiguous that inactivation of BK channels in mouse chromaffin cells arises from expression of β2 subunits (Martinez-Espinosa et al., 2014).

Relevant to the present work, knowledge about loci of expression of Kcnmb3 subunits is exceedingly limited, with no reports of BK currents in native cells that exhibit any of the characteristic features of β3-containing BK channels. At this point, all we can say is that, should β2 and β3a subunits be expressed in the same cell, there should be ternary complexes. Ultimately, the net result of the formation of ternary complexes would a new nuance to observed functional diversity of BK channels available to modulate cellular excitability.

All data underlying main or supplementary figures are available from the corresponding authors upon reasonable request.

Olaf S. Andersen served as editor.

This work was supported by the National Institutes of Health (NIH) GM-118114 to Christopher J. Lingle and NIH GM-145719 and start-up funds from the Department of Molecular Physiology and Biophysics, The University of Iowa, to Sandipan Chowdhury.

Author contributions: Yu Zhou: conceptualization, data curation, formal analysis, investigation, methodology, software, validation, visualization, and writing—original draft, review, and editing. Vivian Gonzalez-Perez: investigation. Xiao-Ming Xia: conceptualization, data curation, investigation, and resources. Gopal S. Kallure: formal analysis, investigation, validation, and visualization. Sandipan Chowdhury: funding acquisition, investigation, methodology, project administration, resources, supervision, validation, visualization, and writing—original draft, review, and editing. Christopher J. Lingle: conceptualization, data curation, formal analysis, funding acquisition, methodology, project administration, resources, supervision, validation, visualization, and writing—original draft, review, and editing.