Activation mechanism of ATP-sensitive K+ channels explored with real-time nucleotide binding

The response of ATP-sensitive K+ channels (KATP) to cellular metabolism is coordinated by three classes of nucleotide binding site (NBS). We used a novel approach involving labeling of intact channels in a native, membrane environment with a non-canonical fluorescent amino acid and measurement (using FRET with fluorescent nucleotides) of steady-state and time-resolved nucleotide binding to dissect the role of NBS2 of the accessory SUR1 subunit of KATP in channel gating. Binding to NBS2 was Mg2+-independent, but Mg2+ was required to trigger a conformational change in SUR1. Mutation of a lysine (K1384A) in NBS2 that coordinates bound nucleotides increased the EC50 for trinitrophenyl-ADP binding to NBS2, but only in the presence of Mg2+, indicating that this mutation disrupts the ligand-induced conformational change. Comparison of nucleotide-binding with ionic currents suggests a model in which each nucleotide binding event to NBS2 of SUR1 is independent and promotes KATP activation by the same amount.


Introduction
ATP-sensitive K + channel (K ATP ) closure initiates the electrical response of pancreatic b-cells to metabolic changes induced by increased extracellular glucose (Ashcroft and Rorsman, 2013;Quan et al., 2011). K ATP 's metabolic sensitivity is accomplished through the coordinated activity of three classes of intracellular adenine nucleotide binding site (NBS), one inhibitory and two stimulatory. Despite the recent publication of several cryo-EM structures of K ATP showing these NBSs at near atomic resolution, the detailed mechanism by which energetic contributions from nucleotide binding to each site sum to affect channel gating remains obscure (Lee et al., 2017;Martin et al., 2017a;Puljung, 2018;Wu et al., 2018). K ATP is formed by four inward-rectifier K + channel subunits (Kir6.2 in b-cells), each associated with a modulatory sulfonylurea receptor (SUR1 in b-cells, Figure 1a) (Inagaki et al., 1995;Sakura et al., 1995;Aguilar-Bryan et al., 1995;Inagaki et al., 1997). The inhibitory NBS of K ATP is located on Kir6.2 and clearly resolved in several of the cryo-EM structures (Lee et al., 2017;Martin et al., 2017a;Puljung, 2018;Wu et al., 2018;Tucker et al., 1997). Like other ATP-binding cassette (ABC) proteins, SUR1 has two cytoplasmic nucleotide-binding domains (NBDs), which associate to form two NBSs at the dimer interface (Figure 1a,c) (Lee et al., 2017). NBS2, formed primarily by the nucleotide-binding Walker A and Walker B motifs of NBD2 and the ABC signature sequence of NBD1, is a consensus binding site, based on conservation of key catalytic residues with other ABC family members, and is competent to hydrolyze nucleotides (de Wet et al., 2007;Matsuo et al., 1999). In NBS1, formed by the Walker motifs of NBD1 and the signature sequence of NBD2, the catalytic Walker B glutamate is replaced with an aspartate (D854), rendering this site 'degenerate,' that is, incapable of nucleotide hydrolysis. NBS1 binds nucleotides in the absence of Mg 2+ , whereas it is thought that Mg 2+ is required for binding to NBS2 . The NBSs of SUR1  (Lee et al., 2017). TNP-ATP (from PDB accession # 3AR7) was positioned in both NBSs of SUR1 by alignment with MgATP (at NBS1) Figure 1 continued on next page mediate K ATP activation by Mg-nucleotides (Tucker et al., 1997;Nichols et al., 1996). SUR1 also confers inhibition by antidiabetic sulfonylureas (SUs) and activation by K ATP -specific K + channel openers (KCOs) (Tucker et al., 1997;Nichols et al., 1996;Gribble et al., 1997a;Gribble et al., 1997b).
Gating of K ATP is a complex function of the intrinsic opening and closing of the channel pore and the converging influences of the excitatory and inhibitory NBSs. Mutations that directly disrupt the NBSs or their ability to transduce nucleotide occupancy to the pore result in diseases of insulin secretion (Quan et al., 2011;Ashcroft et al., 2017). It is crucial to our understanding of K ATP to be able to independently measure (i) the occupancy of each NBS in real time, (ii) the effect each NBS has on K ATP conformation, and (iii) how binding to each NBS affects channel open probability (P open ). Electrophysiology, radioligand binding, and ATPase assays have provided rich mechanistic insight into K ATP gating. However, the ultimate readout of such studies is a function of nucleotide binding to all three NBSs as well as the conformational equilibria affected by binding. Photoaffinity labeling studies enable binding to each NBS to be separated, and some determination of their affinity and specificity, but require partially purified proteins, are performed over long incubation times, and involve irreversible covalent labeling .
An ideal method to separate the nucleotide interactions at each class of site on K ATP would have high spatial (specific for a single NBS) and temporal resolution, would require only small amounts of protein, could be used in a native environment (i.e. the cell or plasma membrane), and would operate under conditions compatible with electrophysiological experiments. We have developed a novel approach for studying ligand binding that fits these criteria. K ATP was labeled with the fluorescent non-canonical amino acid L-3-(6-acetylnaphthalen-2-ylamino)-2-aminopropionic acid (ANAP, Figure 1b) and binding of fluorescent, trinitrophenyl (TNP, Figure 1b) nucleotide derivatives was measured using FRET (Chatterjee et al., 2013). This enabled us to parse the individual contribution of NBS2 to nucleotide binding/channel gating and has provided insight into the mechanism by which Mg-nucleotides affect the conformation of SUR1, and the effect this conformational change has on P open .
In the nucleotide-bound structures of K ATP , the aromatic ring of Y1353 forms a p-stacking interaction with the adenine ring of ADP ( Figure 1c) (Lee et al., 2017;Wu et al., 2018). To probe nucleotide binding to K ATP , we measured FRET between SUR1-Y1353*/Kir6.2 and fluorescent TNPnucleotides (Figure 1d,e). The absorbance spectra of TNP nucleotides overlap with the emission spectrum of SUR1-Y1353* (Figure 1-figure supplement 2a), making them suitable FRET partners. Figure 1-figure supplement 2b shows the calculated distance dependence of FRET between SUR1-Y1353* and TNP-nucleotides. This steep distance dependence provides the spatial resolution necessary to discriminate between nucleotides bound directly at NBS2 (FRET efficiency close to 100%) or at other sites.
FRET between SUR1-Y1353*/Kir6.2 and bound TNP-ADP produced a concentration-dependent decrease (quenching) of the donor ANAP peak at 470 nm and a concomitant increase in the TNP-ADP fluorescence at~565 nm (Figure 1d,e). We quantified nucleotide binding using the decrease in ANAP fluorescence, as this signal was specific to K ATP (Figure 1-figure supplement 1b,c) and we observed non-specific association of TNP-nucleotides in sham-transfected membranes at high concentrations (Figure 1-figure supplement 2e). No change in fluorescence was observed when saturating concentrations of MgATP or MgADP were applied to SUR1-Y1353*/Kir6.2 channels (Figure 1-figure supplement 2f), indicating that quenching of the ANAP signal results from FRET and not an allosteric change in the environment surrounding Y1353* caused by nucleotide binding. Figure 1f shows concentration-response relationships for binding of MgTNP-ATP and MgTNP-ADP to SUR1-Y1353*/Kir6.2 channels. Both nucleotides quenched ANAP completely, consistent with direct binding of TNP-nucleotides at NBS2. MgTNP-ADP bound with~7 fold lower EC 50 than MgTNP-ATP ( Table 2), consistent with the lower EC 50 for activation of SUR1/Kir6.2-G334D channels by MgADP vs. MgATP (7.7 mM vs. 112 mM) (Proks et al., 2010).
Quite unexpectedly, we observed binding of both TNP-ATP and TNP-ADP to NBS2 in the absence of Mg 2+ , that is in 1 mM EDTA (Figure 1g). Both nucleotides bound with similar affinities in the absence of Mg 2+ , but neither fully quenched SUR1-Y1353*/Kir6.2 fluorescence ( Table 2). Although quenching was not 100%, the amount of FRET obtained in the absence of Mg 2+ is still consistent with nucleotides binding at NBS2. The difference in FRET efficiency at saturating nucleotide  (Lee et al., 2017;Martin et al., 2017a).
Despite the relatively close proximity of NBS1 to Y1353* (39 Å to TNP-ATP aligned at NBS1, Figure 1c), the FRET we observed was primarily the result of quenching by TNP-ATP/ADP bound at NBS2. When nucleotides are bound directly to NBS2, FRET efficiency is very high and any nucleotide binding at NBS1 would be expected to have little or no effect on our measurements (Corry et al., 2005). Nucleotides at NBS1 would only contribute to the observed FRET signal if NBS2 were unoccupied. However, with no nucleotide bound to NBS2, the NBDs would not dimerize and Y1353* would be~56 Å away from TNP-nucleotides bound to NBS1 (based on PDB 6BAA) (Martin et al., 2017a). At this distance, the predicted FRET efficiency is <20% (Figure 1-figure supplement 2b), much less that the FRET efficiencies we observed for TNP nucleotide quenching ±Mg 2+ (Figure 1f,g; Table 2).
In the Mg-nucleotide-bound 'quatrefoil' structure of K ATP , the inhibitory NBS of Kir6.2 is within~33 Å of Y1353 (Figure 1-figure supplement 3a) (Lee et al., 2017). To eliminate a possible contribution of nucleotide bound at Kir6.2 to our measured FRET signal, we measured binding of MgTNP-ADP to SUR1-Y1353* co-expressed with Kir6.2-G334D (Figure 1-figure supplement 3b). The EC 50 was very similar to that measured with SUR1-Y1353*/Kir6.2 ( Table 2), suggesting that either the NBS on Kir6.2 was too distant for FRET with Y1353*, or that appreciable binding to Kir6.2 did not occur at concentrations at which NBS2 was unoccupied.

Mg 2+ locks nucleotides at NBS2
If nucleotides can bind NBS2 in the absence of Mg 2+ , why is K ATP only activated in the presence of Mg 2+ ? The simplest explanation is that binding to NBS2 is Mg 2+ -independent, but Mg 2+ is required to support the conformational change that promotes channel activation. Presumably, this conformational change involves NBD dimerization, as observed in the cryo-EM structures of K ATP (Lee et al., 2017;Wu et al., 2018). Because nucleotide dissociation would require opening of the NBD dimer, dimerization is expected to slow the nucleotide off rate. We therefore measured the time course of nucleotide dissociation from SUR1-Y1353*/Kir6.2 in the presence and absence of Mg 2+ .
In the absence of Mg 2+ , TNP-ADP dissociated very rapidly from SUR1-Y1353*/Kir6.2 (Figure 2a, e; Table 3). Dissociation of TNP-ADP was greatly slowed by the presence of Mg 2+ (p=0.001). Similar, but less dramatic results were obtained with TNP-ATP (Figure 2b,e; p=0.0001; Table 3). Whereas the time courses of dissociation of TNP-ADP ±Mg 2+ and TNP-ATP in the presence of Mg 2+ were well described by single exponential decays, the dissociation of TNP-ATP in the absence of Mg 2+ appeared bi-exponential.  Our data are consistent with an activation model for K ATP in which the NBDs of SUR1 dimerize in the presence of Mg 2+ , preventing rapid nucleotide dissociation. However, it is formally possible that Mg 2+ could directly stabilize nucleotide binding without any conformational change in the NBDs. To test this hypothesis, we examined nucleotide dissociation rates in the presence of the sulfonylurea (SU) tolbutamide (Figure 2c,d,e, Table 3). The cryo-EM structure of Kir6.2/SUR1 in the presence of glibenclamide suggests that SUs prevent NBD dimerization, and electrophysiological studies further demonstrate that SUs prevent Mg-nucleotide induced current activation (Martin et al., 2017a;Proks et al., 2014). In the presence of tolbutamide, the dissociation rates for MgTNP-ADP and MgTNP-ATP were not significantly different from those measured for nucleotides in the absence of both tolbutamide and Mg 2+ (Figure 2e, Table 3; p=0.1, for TNP-ADP, p=1 for TNP-ATP). This argues that the slower off rate for TNP-ADP and TNP-ATP in the presence of Mg 2+ must result from NBD dimerization, not a direct stabilization of nucleotide binding by Mg 2+ .
As further support, we tested the effect of diazoxide (a KCO) on the nucleotide dissociation rate. It has been proposed that nucleotide binding at the NBDs of SUR2A stabilizes binding of the KCO pinacidil (Gribble et al., 2000). The time course of nucleotide dissociation in the presence of diazoxide did not closely follow a single exponential decay, but the drug greatly reduced the dissociation rates for both MgTNP-ADP and MgTNP-ATP (Figure 2c,d,e; Table 3). This suggests that KCOs function by stabilizing the activated (NBD-dimerized) conformation of SUR1.
In the absence of any conformational change (i.e. in Mg 2+ -free solution), nucleotide binding to NBS2 would be expected to follow a simple binding equilibrium (Scheme 1), with K d = k off /k on .
A simple activation model like the one proposed by Del Castillo and Katz for acetylcholine receptors (Scheme 2), in which nucleotides bind SUR1 and change its conformation can be invoked to explain the decrease in the apparent rate of nucleotide dissociation (Del Castillo and Katz, 1957). The nucleotide dissociation rate would depend on the kinetics of entry into (a) and exit from (b) the Figure 2 continued from SUR1-Y1353*/Kir6.2 in the presence of 500 mM tolbutamide (pink) or 340 mM diazoxide (teal). (c) The orange curve is the fit to the MgTNP-ADP data from (a). Tolbutamide: t = 6.4 s±0.2 s; diazoxide: t = 52.6 s±1.4 s. (d) The orange curve is the fit to MgTNP-ATP data from (b). Tolbutamide: t = 6.3 s±0.1 s; diazoxide: t = 60.5 s±1.4 s. (e) Time to 50% dissociation (t 0.5 ) for the individual fits to the data shown in (a)-(d) for better comparison between single-exponential and more complex time courses. Boxes represent mean ±SD. Individual data points are shown to the right of the boxes. Combined data from multiple experiments in (a)-(d) were fit to single exponential decays (Equation 2). Dissociation of TNP-ATP in the absence of Mg 2+ was better fit with a bi-exponential decay (Equation 3). DOI: https://doi.org/10.7554/eLife.41103.010 activated state (SUR1', in which the NBDs are dimerized), as well as the intrinsic k off . In Scheme 2, In our experiments, the off-rate for TNP-ADP was reduced~5 fold in the presence of Mg 2+ , whereas the EC 50 for TNP-ADP was largely unaffected (Figure 2e, Figure 1-figure supplement 4a, Table 2, Table 3). Mg 2+ reduced the off-rate for TNP-ATP by a factor of 2.2, but actually increased the EC 50 for binding by a factor of~3 (Figure 2e, Figure 1-figure supplement 4b, Table 2, Table 3). In the limiting case for Scheme 2, in which b>>a, the apparent EC 50 would be equal to the actual K d . (i.e. that measured in Mg 2+ -free solution). Therefore, the concentration-response curve for nucleotide binding would not be expected to shift to the right in the presence of Mg 2+ through a change in activation alone. As we observed such a shift for Mg-TNP-ATP, we conclude that in addition to stabilizing the NBDs in a dimerized state when nucleotides are present, Mg 2+ must also decrease the intrinsic on-rate for nucleotide binding (k on ).
The nucleotide dissociation rates measured in the presence of Mg 2+ are slower than the deactivation rates observed for SUR1/Kir6.2-G334D currents following nucleotide washout (Proks et al., 2010). However, it should be noted that current decay reflects a conformational change in the channel and not the rate of nucleotide dissociation, and may be affected by the presence of four SUR1 monomers per channel. Furthermore, the deactivation rates were measured for MgATP/ADP and not TNP-nucleotides. Ultimately, a direct comparison of TNP-nucleotide dissociation rates and deactivation rates will require simultaneous measurement of current and binding.

Mutation of the Walker A lysine of NBS2 affects the conformational change in SUR1
We next sought to explore the mechanistic consequences of mutating the Walker A nucleotide binding motif of NBS2. In other ABC proteins, and in the Mg-nucleotide-bound structure of K ATP , the Walker A lysine coordinates the b-and g-phosphates of bound nucleotides ( Figure 3a). Previous electrophysiological studies showed that mutation of the Walker A lysine in NBS2 to alanine (K1384A, K2A) reduced the maximal activation of SUR1/Kir6.2-G334D currents by MgATP and shifted the concentration dependence of MgADP activation >20 fold to higher concentrations, such that the maximal response could not be determined (Proks et al., 2014). This is consistent with the K2A mutation affecting channel activation.
As we observed no change in the TNP-ADP concentration-response in the absence of Mg 2+ , the K2A mutation may affect MgTNP-ADP binding indirectly, by reducing the ability of the nucleotide to cause NBD dimerization. To explore this idea further, we measured the rate of MgTNP-ADP dissociation from SUR1-Y1353*,K2A/Kir6.2 ( Figure 3f, Table 3). The K2A mutation increased the off-rate for MgTNP-ADP compared to SUR1-Y1353*/Kir6.2 channels (Table 3; p=0.04). There was no apparent change in the off-rate for TNP-ADP in the absence of Mg 2+ , compared to SUR1-Y1353*/Kir6.2 (p=0.8). Taken together, these data suggest that the K2A mutation destabilized NBD dimerization induced by MgTNP-ADP. However, these results do not rule out the possibility that the K2A mutation also affected the intrinsic on-rate for MgTNP-ADP.

Correlation of binding with gating
We next sought to correlate nucleotide binding with changes in channel activity. Despite our ability to demonstrate a Mg 2+ -and nucleotide-dependent conformational change in SUR1-Y1353*/Kir6.2, we were unable to measure activation of SUR1-Y1353*/Kir6.2 channels by Mg-nucleotides. There are two possible explanations for this observation. The first is that SUR1-Y1353* disrupts communication between the NBD-dimerized SUR1 and Kir6.2. Inspection of the Mg-nucleotide-bound 'quatrefoil' structure shows that position 1353 is adjacent to residues (e.g. R1352) at the interface between SUR1 and Kir6.2 (Lee et al., 2017). Mutation of R1352 disrupts K ATP activation (de Wet et al., 2012;Magge et al., 2004). Another possible explanation for the lack of activation of SUR1-Y1353*/Kir6.2 currents is that channels containing SUR1 subunits truncated at Y1352 (i.e. no ANAP incorporated) traffic to the plasma membrane. These channels would not be activated by nucleotides as they lack a complete NBS2. K ATP channels with SUR1 truncated at residue 1330 at the start of NBD2, traffic to the plasma membrane but are not activated by MgADP (Sakura et al., 1999).
In fluorescence measurements, we only observe full-length SUR1-Y1353*/Kir6.2 channels as only these are labeled with ANAP ( Figure 1-figure supplement 1). However, in electrophysiological experiments it may be possible to measure currents from channels with both full-length and truncated SUR1. Using a surface expression assay, we observed no difference in the ability of full-length SUR1-Y1353* and truncated SUR1-Y1353 stop (no ANAP included in the culture medium) to chaperone HA-tagged Kir6.2 to the plasma membrane (Figure 4-figure supplement 1a) (Zerangue et al., 1999). Truncated SUR1-Y1353 stop /Kir6.2 (expressed without pANAP) formed functional channels that were not activated by MgADP (Figure 4-figure supplement 1b). 100 mM MgADP increased full-length 3xFLAG_SUR1/Kir6.2 currents 3.6-fold over the current in nucleotidefree control solutions (Figure 4-figure supplement 1b, Table 4). In contrast, 100 mM MgADP inhibited current from truncated SUR1-Y1353 stop by 59% (Figure 4-figure supplement 1b, Table 4). This degree of inhibition is similar to the~60% reduction in wild-type SUR1/Kir6.2 current by 100 mM ADP in Mg 2+ -free solutions, in which only inhibition is expected (Proks et al., 2010), confirming that SUR1-Y1353 stop /Kir6.2 channels failed to activate. Currents from SUR1-Y1353 stop /Kir6.2 coexpressed with pANAP alone (66% full-length SUR1-Y1353* expected,  Table 4) (Schmied et al., 2014). This suggests that SUR-Y1353* fails to support nucleotide activation. To confirm this, we also looked at currents in patches excised from cells expressing  SUR1-Y1353*/Kir6.2-G334D (co-expressed with pANAP and pERF1-E55D). Wild-type SUR1 paired with Kir6.2-G334D produced channels that were activated by MgADP and inhibited by tolbutamide ( Figure 4-figure supplement 1c). When measured under the same conditions, SUR1-Y1353*/Kir6.2 was inhibited by tolbutamide, but we observed no activation by MgADP. We therefore sought to label an additional position near NBS2 of SUR1 that produces functional, nucleotide-activated channels when paired with Kir6.2. We mutated the codons for 34 different residues (amino acids 1386-1397, 1400-1408, and 1410-1423) to the amber stop codon and screened the constructs for their ability to promote surface expression of HA-Kir6.2 in the presence of ANAP. We focused on SUR1-T1397 stop , as this construct strongly increased the surface expression of HA-Kir6.2 in the presence of ANAP (Figure 4-figure supplement 1a; p=0.005). Further, in the absence of ANAP, this construct decreased the surface expression of HA-Kir6.2 (p=0.009) compared to no-SUR1 controls (Figure 4-figure supplement 1a), suggesting that channels with SUR truncated at this position may be selectively retained within the cell. Western blots of total protein from cells expressing SUR1-T1397 stop in the presence of ANAP and pANAP/peRF1-E55D confirmed that >95% of SUR1-T1397* was full length (Figure 4-figure supplement 1d). The expected distance between T1397* and TNP-nucleotides bound to NBS2 is~18.3 Å (Figure 4a), close enough for >99% FRET efficiency (Figure 1-figure supplement 2b).

Discussion
Taken together, our data suggest a modified model for SUR1/Kir6.2 K ATP activation by nucleotide binding to SUR1 (Figure 6). ADP and ATP bind to NBS2 in the absence of Mg 2+ . However, without Mg 2+ as a co-factor, adenine nucleotides cannot dimerize the NBDs and initiate the conformational change in SUR1 required to affect P open . As the NBDs fail to dimerize, nucleotides dissociate rapidly. If both Mg 2+ and nucleotides are present, the NBDs dimerize, resulting in slower nucleotide dissociation. NBD dimerization initiates a conformational change in SUR1 that ultimately communicates nucleotide occupancy to the pore, affecting P open . Inhibitory SUs prevent NBD dimerization. Thus, nucleotides dissociate rapidly even in the presence of Mg 2+ and P open is not affected by nucleotide occupancy at NBS2. KCOs have the opposite effect. They stabilize NBD dimerization, prolonging nucleotide occupancy at NBS2 and channel activation.
Although our data do not offer any direct structural insight into how NBD dimerization propagates to Kir6.2, our ability to measure MgTNP-ADP binding to NBS2 of SUR1 and channel activation under similar conditions furnishes some mechanistic insights into the relationship between nucleotide binding and channel opening. We simultaneously fit the combined data sets for MgTNP-ADP binding to SUR1-T1397*/Kir6. 2017a; Wu et al., 2018;Li et al., 2017;Martin et al., 2017b). show any direct interactions between neighboring SUR1 subunits. Therefore, we excluded models in which binding to one SUR1 directly affects binding to the other SUR1 subunits (i.e. with binding cooperativity). With these assumptions, the best fits to the data were obtained with an extended Monod-Wyman-Changeux model (Figure 5b; R 2 = 0.98, RSS = 0.15) (Monod et al., 1965). In this model,   binding of MgTNP-ADP to each NBS2 of SUR1 is independent and favors pore opening by the same amount. The data in Figure 5b were fit with the following expressions for binding and activation: Previous attempts to determine the stoichiometry of Mg-nucleotide activation of K ATP , that is how many SUR1 subunits must bind nucleotide in order to activate current, have produced varying results (Babenko, 2008;Hosy and Vivaudou, 2014;Tammaro et al., 2005). We also fit MgTNP-ADP binding to NBS2 and current activation for SUR1-T1397* with models in which fixed stoichiometries of 1-4 nucleotide-bound SUR1 subunits were necessary to increase P open ( Figure 5  the MWC model in Figure 5b, but none fit the data as well. An inspection of our model fits (Figure 5-figure supplement 2) shows that the residuals for the MWC model and the model in which two bound nucleotide molecules activate K ATP are small and centered around zero. The other three models show systematic errors. Whereas the model in which two binding events opens K ATP provides a reasonable fit to the data, this model also predicts E » 1, that is no activation. The MWC fits to our data predicts E = 2.2, which is similar to the value of 1.76 that we previously estimated from electrophysiology data (Proks et al., 2010;Vedovato et al., 2015). This would favor the MWC model. However, given the scatter in the data, further experimentation will be required to truly distinguish between these competing models (see below). Our binding and activation data were determined in separate experiments. Whereas the equations for binding and current activation both have a term (L) representing the intrinsic opening equilibrium, there is no way of knowing the P open of channels in an unroofed membrane. We expect P open to be quite low, as several minutes may elapse between cell unroofing and the completion of a given experiment, during which time channel rundown may occur (e.g. due to dissociation/degradation of PIP 2 ). Ideally, to derive more quantitative values, experiments should be performed using patch-clamp fluorometry in which binding and current are measured simultaneously (Biskup et al., 2007;Zheng and Zagotta, 2000). However, we were unable to obtain inside-out patches from cells expressing SUR1-T1397*/Kir6.2-G334D with sufficient fluorescence intensity to measure reliable spectra.
In our study, the concentration-response relationship for current activation was normalized to the minimum and maximum current values, as reflected in Equation 7. If the true maximum and minimum P open in a given experiment could be accurately determined, then the quantity E could be computed directly from the minimum P open in the absence of ligand, L/(L + 1), and the maximum P open at saturating nucleotide concentrations, E 4 L/(E 4 L + 1), and our models would be further constrained. In practice, however, these values are difficult to measure for K ATP in the presence of Mg 2+ as channel rundown due to loss of PIP 2 results in an intrinsic P open (L) that is constantly changing (Proks et al., 2016). Nevertheless, our analysis provides useful qualitative information regarding the mechanism by which nucleotide binding to NBS2 of SUR1 activates Kir6.2.
The affinities we measured for TNP-ADP and TNP-ATP binding are higher than those previously estimated for ATP and ADP from photoaffinity labeling studies on SUR1 in the absence of Kir6.2 or from ATPase assays preformed on isolated NBD2 fusion proteins (de Wet et al., 2007;Matsuo et al., 2000). This may reflect the absence of Kir6.2 or a higher affinity of TNP nucleotides for NBS2 of SUR1 compared to ATP and ADP, as observed for other ABC transporters (Oswald et al., 2008). Nevertheless, the EC 50 values obtained from MgTNP-ATP and MgTNP-ADP binding to SUR1-Y1353*/Kir6.2 were in reasonably good agreement with those measured for MgATP and MgADP activation of SUR1/Kir6.2-G334D currents (Proks et al., 2010). It is unlikely that the introduction of ANAP modified the binding affinity of NBS2 as the apparent affinity for TNP-nucleotides in the absence of Mg 2+ was unaffected by the site at which ANAP was inserted ( Table 2). The difference in apparent affinities between SUR1-Y1353* and SUR1-T1397* in the presence of Mg 2+ may indicate a difference in the ability of bound nucleotides to promote NBD dimerization.
MgADP can activate channels to an equal or greater extent than MgATP (Proks et al., 2010). Therefore, it is clear that the act of ATP hydrolysis per se is not required for channel activation, even though NBS2 is competent to hydrolyze ATP (de Wet et al., 2007;Matsuo et al., 1999). It is often stated that activation of K ATP channels by Mg 2+ -ATP proceeds through hydrolysis of MgATP to MgADP (Zingman et al., 2001;Zingman et al., 2002). However, there is no evidence that an irreversible step like ATP hydrolysis occurs during the gating cycle of the K ATP channel in the presence of MgATP, as is the case for the closely related ABC protein CFTR (Choi et al., 2008;Csanády et al., 2010). Furthermore, high concentrations of ATP have been shown, in radioligand binding assays, to cause SUR1 to change conformation in the absence of Mg 2+ , under which conditions ATP hydrolysis is not expected (Ortiz et al., 2013). Mg 2+ -free ATP, at high concentrations, was also recently demonstrated to activate K ATP formed by SUR1-Q1179R, a gain-of-function SUR1 mutation, paired with Kir6.2-G334D, as well as SUR constructs in which the Walker B glutamate was mutated to glutamine (Sikimic et al., 2018). As lack of Mg 2+ or mutation of the catalytic Walker B motif do not support MgATP hydrolysis, this suggests that ATP can activate K ATP directly. No Mg 2+free ATP activation was demonstrated when wild-type SUR1 was expressed with Kir6.2-G334D and we see no indication in our data that nucleotides caused SUR1 to change conformation in the absence of Mg 2+ (Figure 2) (Sikimic et al., 2018).
Our data suggest that ATP hydrolysis did not occur over the course of our experiments, as the off-rate for MgTNP-ATP was faster than that of MgTNP-ADP (t 0.5 = 10.5 s vs 14.7 s). If MgTNP-ATP were hydrolysed to MgTNP-ADP before being released (as is thought to be true of many ABC family members, including CFTR) the dissociation rates should have been similar. The observation that MgTNP-ATP dissociation was faster than MgTNP-ADP dissociation may reflect a lack of MgTNP-ATP hydrolysis during the course of our experiments. Alternatively, this may indicate a destabilizing effect of inorganic phosphate on MgTNP-ADP binding to NBS2 (which would not be present when binding MgTNP-ADP alone, but would be present as a product of MgTNP-ATP hydrolysis) on binding.
Our ability to directly measure nucleotide binding in real time will be of considerable value for future investigations. This will allow us to address the role of nucleotide binding at NBS1 and Kir6.2, and explore the mechanistic ramifications of disease-associated K ATP mutations. Furthermore, simultaneous binding and current measurements will yield precise information regarding the coupling of stimulatory and inhibitory nucleotide binding to the channel pore. This coupling of ligand binding to an effector domain (the channel pore) that lies tens of Å away provides an excellent model for studying long-range communication within protein complexes. It has also not escaped our notice that our method is readily adaptable to the study of other ABC proteins, ATP-gated channels like P2X receptors (for which different subtypes demonstrate distinct selectivity for MgATP vs. ATP) and any protein for which there is a suitable fluorescent ligand (Li et al., 2013).
Cell culture and expression of fluorescently tagged protein HEK-293T cells were obtained from and verified/tested for mycoplasma by LGC Standards (ATTC CRL-3216; Middlesex, UK). Frozen stocks were prepared directly from these cells and stored in liquid nitrogen. Working stocks were routinely replenished from frozen following 15-25 passages. Our working stock tested negative for mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit (Lonza Bioscience; Burton on Trent, UK). For experiments, cells were grown in 6-well plates (STARLAB; Milton Keynes, UK) or 30 mm culture dishes (STARLAB) on untreated 30 mm borosilicate coverslips (Thickness # 1, VWR International; Radnor, PA) or on poly-D-lysine coated FluoroDishes (FD35-PDL-100, World Precision Instruments; Hitchin, UK) in Dulbecco's Modified Eagle Medium (Sigma) supplemented with 10% fetal bovine serum, 100 U/mL penicillin, and 100 mg/mL streptomycin (Thermo Fisher Scientific; Waltham, MA) at 37˚C, 5%/95% CO 2 /air. SUR1 constructs were site-specifically tagged with the fluorescent amino acid ANAP using amber codon suppression as described by Chatterjee et al. (Chatterjee et al., 2013) 30 mm dishes of HEK-293T cells were co-transfected with pANAP (0.5-1 mg), Kir6.2 (0.5 mg), and SUR1 (1-1.5 mg) constructs containing an amber stop codon (TAG) at amino acid position 1353 or 1397 (1353 stop or 1397 stop ) using TransIT transfection reagent (Mirus Bio LLC; Madison, WI) in a ratio of 3 mL of TransIT per mg of total DNA. Following transfection the media was supplemented with 20 mM ANAP (free acid or methyl ester). pANAP encodes a tRNA/tRNA synthetase pair specific for ANAP. In the presence of ANAP, cells transfected with pANAP generate tRNAs charged with ANAP that recognize the amber stop codon. Thus, full-length ANAP-tagged SUR can be produced. Proper co-assembly of Kir6.2 subunits with SUR1 subunits is ensured by the fact that the two subunits rely on one another for proper exit from the ER and trafficking to the plasma membrane (Zerangue et al., 1999). An exception to this may be SUR1_GFP, which has been shown to traffic independently of Kir6.2 (Makhina and Nichols, 1998). After transfection, cells were cultured at 33˚C, 5%/95% CO 2 /air to slow growth and increase the per-cell protein yield (Lin et al., 2015). To enhance expression of fulllength, ANAP-labeled proteins, 1-2 mg of a plasmid containing a dominant negative eukaryotic release factor 1 (peRF1-E55D) was included in the transfection mix as indicated (Schmied et al., 2014). Experiments were performed 2-5 days post transfection.

Preparation of unroofed membrane fragments for imaging
Unroofed membranes were prepared using a modification of a protocol from the Heuser laboratory (Heuser, 2000;Usukura et al., 2012;Zagotta et al., 2016). Briefly, a coverslip containing adherent cells was removed from the cell culture medium and a small piece (0.5-1 cm) was broken off using jewelers forceps (#5) and washed in phosphate buffered saline (PBS, Thermo Fisher Scientific) diluted 1:3 in deionized water. Cells were exposed 3 Â 10 s to a solution of 0.01% poly-L-lysine (Sigma) with intervening washes in diluted PBS to adhere them more firmly to the coverslips. The coverslip fragment was then placed in a 30 mm culture dish filled with 2-3 mL of 1/3X PBS and sonicated very briefly (~100 ms) using a probe sonicator (Vibra-cell; Newtown, CT), leaving behind adherent plasma membrane and sometimes additional cellular material. The unroofed fragments in our experiments were sometimes visible under bright-field illumination, a potential result of our modified 'unroofing' procedure. Morphologically, such cell fragments resemble the 'partially unroofed' cells identified by Usukura et al. (Usukura et al., 2012) Cells cultured on poly-D-lysine coated FluoroDishes were unroofed in diluted PBS with no additional poly-L-lysine treatment. This procedure yielded a higher percentage of completely unroofed plasma membranes (i.e. nearly invisible with brightfield illumination).

Microscopy/spectroscopy
Unroofed membrane fragments were imaged directly on FluoroDishes (FD35-PDL-100, World Precision Instruments) or on broken coverslips placed in a FluoroDish (FD3510, World Precision instruments) using a Nikon TE2000-U microscope equipped with a 40x (S Fluor, 1.3 NA; Nikon, Kingston Upon Thames, UK) or 100x (Apo TIRF, 1.49 NA, Nikon) oil-immersion objective and low-fluorescence immersion oil (MOIL-30, Olympus; Southend-on-Sea, UK) or a 60x water immersion objective (Plan Apo VC, 1.20 NA, Nikon). ANAP was excited using a ThorLabs LED source (LED4D067) at 385 nm connected to the microscope using a liquid light guide in series with a 390/18 nm band-pass filter (MF390-18, ThorLabs; Newton, NJ) and MD416 dichroic (ThorLabs). For imaging, emitted light was filtered through a 479/40 nm band-pass filter (MF479-40, ThorLabs). GFP-tagged constructs were imaged using a similar setup with excitation at 490 nm, a 480/40 band-pass excitation filter (Chroma; Bellows Falls, VT), a DM505 dichroic mirror (Chroma), and 510 nm long-pass emission filter (Chroma). All images were acquired using a PIXIS 400B CCD camera (Princeton Instruments; Trenton, NJ). For spectroscopy, ANAP was excited as above, but emitted light was passed through a 400 nm longpass filter (FEL0400, ThorLabs) and directed through the slit of an Isoplane 160 spectrometer (300 g/ mm grating; Princeton Instruments) in series with the camera. The acquired images retained spatial information in the y-dimension, with the x-dimension replaced by wavelength (Figure 1-figure supplement 1a). Exposure times were typically 100 ms for brightfield images, 1-10 s for steady-state spectra and 1-5 s for time-course experiments. The emission spectrum of ANAP-labeled SUR1 showed a peak around 470 nm, which was used to differentiate labeled channels from autofluorescence (which typically had a broad emission spectrum), from cytoplasmic ANAP (or potentially ANAP-conjugated to tRNA), which peaks at~480 nm or from bright, fluorescent debris on the coverslip which was morphologically distinct from unroofed membranes and had spectra peaking at either~450 nm or 480-485 nm.
To verify that the fluorescence signal we observed in unroofed membrane fragments derived from ANAP-labeled SUR1, we co-transfected cells with Kir6.2 and GFP-tagged SUR1 with or without an amber stop codon at position Y1353. After unroofing, membranes from transfected cells could be identified by the green fluorescence of SUR1_GFP/Kir6.2. Figure 1-figure supplement 1b shows that membranes expressing SUR1_GFP-Y1353*/Kir6.2 were brightly fluorescent for both GFP and ANAP, whereas those from SUR1_GFP expressing cells had only GFP fluorescence, even though ANAP was included in the cell-culture medium. We also acquired emission spectra from membranes expressing SUR1_GFP-Y1353*/Kir6.2 (Figure 1-figure supplement 1c). The intensity of the ANAP peak in our fluorescence was linearly proportional to the peak GFP fluorescence, and the intensity of the ANAP peak extrapolated to zero GFP (i.e. no expression of SUR1_GFP-Y1353*) was zero. Taken together, these data suggest that non-specific ANAP background was very low. We also performed western blotting on FLAG-tagged SUR1-Y1353 stop constructs to verify that we produced full-length, ANAP-labeled SUR1, and that cells were unable to read through the amber stop codon to produce full-length protein in the absence of ANAP (Figure 1-figure supplement 1d).
To prevent accumulation of fluorescent nucleotides in the bath, FluoroDishes were perfused with either (in mM) 140 KCl, 20 NaCl, 10 MgCl 2 , 10 HEPES pH 7.2 with NaOH or 140 KCl, 20 NaCl, 1 EDTA, 10 HEPES pH 7.2 with NaOH using a peristaltic pump (Ismatec; Wertheim, Germany). Unroofed membranes were directly perfused with nucleotides using an 8-channel mFlow perfusion system (ALA Scientific Instruments; Farmingdale, NY). Solution change with the mFlow was rapid (t = 0.73 s±0.02 s, n=5) as measured from single exponential fits to wash-out of 50 mM tetramethylrhodamine-5-maleimide (TMRM) after switching to a solution with no TMRM. TMRM was excited with a 565 nm LED with a D540/25X band-pass filter (Chroma) and DM565 dichroic mirror (Chroma). Emitted light was collected through a D605/55M band-pass filter (Chroma). Solution exchange was also monitored during experiments with TNP-nucleotides by examining the emission spectrum of a background region in each image. However, the detection limit for TNP-nucleotides in solution was low as unbound nucleotides have a low quantum yield (enhanced by protein binding) and are not excited via FRET, as is the case for nucleotides bound to ANAP-labeled SUR1 (Broglie and Takahashi, 1983).
Images and spectra were acquired using LightField software (Princeton Instruments). The solution changer, camera, and light source were all controlled using pClamp 10.5 (Molecular Devices; San Jose, CA) and a DigiData 1440 A/D converter. All data were acquired at room temperature (18˚À22C ).

Analysis of images and spectra
Images were adjusted and displayed using Fiji (Schindelin et al., 2012). Spectra were analyzed using custom code written in Matlab (Mathworks; Natick, MA). The code was deposited on GitHub (https://github.com/mpuljung/spectra-analysis; copy archived at https://github.com/elifesciencespublications/spectra-analysis) (Puljung, 2019a;Puljung, 2019b). Regions of interest corresponding to fluorescent membrane fragments were manually selected from the spectral images and averaged for each wavelength. A background region of similar size was selected and the average background spectrum was subtracted from the spectrum of the region of interest. The peak fluorescence (~470 nm) was automatically selected by determining the maximum of a boxcar average of 31 points (corresponding to~10 nm) of the data in the absence of nucleotides. Data were corrected for photobleaching by fitting an exponential decay (F = Ae t/t + (1-A)) to the normalized peak of 5-6 images acquired prior to washing on TNP-nucleotides and dividing the raw spectra by the resulting fit at each cumulative exposure time (Figure 1-figure supplement 5). Photobleaching rates were similar regardless of the position of ANAP in SUR1. For concentration-response relationships, corrected peak data (at~470 nm) were normalized by the ANAP fluorescence at zero [TNPnucleotide] (F/F max ) and displayed as 1-F/F max . On occasion data were normalized instead to the fluorescence at 50 nM nucleotide, as we consistently saw no effect at this concentration. TNP-nucleotide concentration-response relationships were fit with the Hill equation where E max is the FRET efficiency at saturating concentrations, [TNP] is the concentration of TNPnucleotide, EC 50 is the half-maximal concentration and h is the Hill slope.
To generate the plot of GFP fluorescence vs. ANAP fluorescence in Figure 1-figure supplement 1c, spectra of ANAP-and GFP-tagged SUR1 subunits expressed in unroofed membrane fragments were acquired as above. We were able to resolve a GFP peak at 510 nm with 385 nm excitation, even though this wavelength is far from the l max for GFP. It is possible that GFP emission was enhanced via FRET between ANAP and GFP. The ANAP peak fluorescence at 470 nm did not overlap with the GFP emission peak. However, the peak GFP at 510 nm was contaminated by the shoulder of the ANAP peak. Therefore, the GFP fluorescence was corrected by subtracting the averaged (from 16 membranes) fluorescence of ANAP incorporated into SUR1, scaled to match the 470 nm peak in the ANAP-GFP spectrum.
To measure the time course of nucleotide unbinding, the fluorescence at the TNP-nucleotide peak (usually around 560 nm) was plotted as a function of time after exchanging to a zero-nucleotide solution, as following the time course of the rise in ANAP fluorescence was complicated by photobleaching. After complete wash out of nucleotides, the fluorescence at 560 nm was non-zero, as the shoulder of the ANAP emission spectrum extends into the range of wavelengths at which TNPnucleotides emit. Therefore, the time course data were corrected by subtracting a linear fit to the data following complete washout. This corrected for the shoulder of the ANAP spectrum as well as the continued photobleaching of ANAP. Data were normalized to the maximum fluorescence at t = 0 and fit with single-exponential decays or double-exponential decays Wash-out time courses following exchange to zero TNP-nucleotide solution should be independent of the nucleotide concentration before wash. The majority of our time-course experiments were conducted at the conclusion of our concentration-response experiments, and therefore reflect wash after applying a concentration of 1 mM. For experiments in diazoxide and tolbutamide, the nucleotide concentration prior to wash was 50 mM.
In some experiments, there was evidence of cross-contamination of our wash solution or lownucleotide solutions via back-flow of solutions with higher concentrations of nucleotide (as evident by accumulation of yellow TNP-nucleotides). We excluded data from such experiments and maintained/replaced valves on the mFlow as needed.

Electrophysiology
Currents were recorded from inside-out membrane patches excised from transfected HEK-293T cells. Pipettes were pulled to a resistance of 1-5 MW and filled with an extracellular solution containing (in mM) 140 KCl, 1.2 MgCl 2 , 2.6 CaCl 2 , and 10 HEPES, pH 7.4 or 140 KCl, 1 EGTA, 10 HEPES, pH 7.3. The intracellular (bath) solutions contained either (in mM) 107 KCl, 2 MgCl 2 , 1 CaCl 2 , 10 EGTA, and 10 HEPES, pH 7.2 for experiments showing nucleotide-dependent activation or 140 KCl, 1 EDTA, 1 EGTA, 10 HEPES pH 7.3 for experiments showing nucleotide-dependent inhibition. Nucleotides were added as indicated. Patches were perfused with an 8-channel mFlow perfusion system. Data were acquired at a holding potential of À60 mV using an Axopatch 200B amplifier and Digidata 1322A digitizer with pClamp 9.0 software (Molecular Devices). Currents were digitized at 10 kHz and low-pass filtered at 1 kHz. K ATP channels run down in excised patches (reviewed by Proks et al.) (Proks et al., 2016). For inhibitory concentration-response relationships, rundown was corrected by alternating test nucleotide solutions with nucleotide-free (control) solutions and expressing the test currents as a fraction of the average of the control currents before and after the test solution. Corrected data were fit to the following expression: with IC 50 representing the half-maximal inhibitory concentration. For activation concentrationresponses, the same protocol was used, but test concentrations were alternated with a saturating concentration (1 mM) of MgADP (Proks et al., 2010). Data were displayed at the current magnitude in the test concentration minus the current in zero nucleotide (I-I min ) divided by the current in saturating nucleotide (1 mM MgADP) minus the current in zero nucleotide (I max -I min ) and fit to the following expression: Currents were leak corrected by subtracting the remaining current after complete rundown.

Western blotting
In order to detect SUR1 via western blot, a triple FLAG-tagged construct (3xFLAG_SUR1, MDYKDHDGDYKDHDIDYKDDDDK; tag in an extracellular loop between amino acid positions T1042 and L1043) was used with or without amber stop codons at amino acid positions 1353 or 1397. HEK-293T cells were transfected in 6-well plates using the conditions described above, but with 1 mg of 3xFLAG_SUR1 DNA. After 48 hr, cells were washed and harvested with gentle pipetting in 1 mL of PBS (Sigma). Cells were pelleted (4 min at 200 x g yielding 20-50 mg pellets) and solubilized in 50-100 mL of 0.5% Triton X-100 in 110 mM potassium acetate, buffered at pH 7.4. 125 U of benzonase (Sigma) was added to each sample and samples were incubated at room temperature for 20 min. 7.5 mL of each sample was mixed with 2.5 mL of NuPAGE LDS sample buffer and 1 mL NuPAGE sample reducing agent (Invitrogen) and run on a NuPAGE 4-12% Bis-Tris gel (Invitrogen) at 200 V for 40 min in 1X MOPS-SDS buffer (Invitrogen). Protein was transferred overnight at 10 V to an Immobilon-P membrane (Merck Millipore; Burlington, MA) in 25 mM Tris, 192 mM glycine, and 20% methanol +0.1% SDS. Following transfer, the membrane was rinsed in TBS-Tw (150 mM NaCl, 25 mM Tris, pH 7.2, and 0.05% Tween 20) and incubated (with shaking) for 30 min in TBS-Tw +5% milk. The membrane was rinsed three times in TBS-Tw and incubated for 30 min with primary antibody (M2 anti-FLAG from Sigma, diluted 1:500 in TBS-Tw). The membrane was rinsed three more times in TBS-Tw and incubated for 30 min with secondary antibody (HRPconjugated sheep anti-mouse IgG, GE Healthcare Life Sciences; Freiburg, Germany) diluted 1:20,000 in TBS-Tw +1% milk. Finally, the membrane was rinsed 3 Â 10 min in TBS-Tw and developed using SuperSignal West Femto Max Sensitivity Substrate (Thermo Fisher Scientific). Images were collected using a C-DiGit scanner (Licor Biosciences; Lincoln NE) and band intensities were quantified using custom code written in MATLAB (https://github.com/mpuljung/spectra-analysis) (Puljung, 2019c).
Western blotting was performed ( Figure 1-figure supplement 1d) on the total protein from HEK-293T cells transfected with Kir6.2 and 3xFLAG-tagged SUR1-Y1353 stop DNA under various conditions. As the 3xFLAG tag was N-terminal to any introduced stop codons in SUR1, we used it to probe for both full-length and truncated SUR1. As a positive control, we transfected 3x-FLAGtagged SUR1 with no stop codon, which was only expected to produce full-length protein (lane 1). This protein ran above our 140 kDa marker as expected (predicted MW is~180 kDa). In the absence of ANAP or in cells transfected without the pANAP plasmid, only truncated 3xFLAG_SUR1 protein was evident (lanes 4 and 5), confirming that there was no full-length protein produced by accidental read-through of the amber (TAG) codon. In the presence of both pANAP and ANAP, we obtained a mixture of full-length (66%) and truncated (34%) SUR1. The majority of our experiments in unroofed membrane fragments were performed under such conditions, as the signals we measured (fluorescence) were derived solely from full-length (i.e. ANAP-labeled) channels. Finally, we were able to increase expression of full-length SUR1 as needed by transfecting an additional plasmid, peRF1-E55D, which encodes a dominant negative ribosomal release factor (Schmied et al., 2014). With the additional expression of ERF1-E55D, we obtained 92% full-length protein (Figure 1-figure supplement 1d). A similar approach was taken with SUR1-T1397 stop (Figure 4-figure supplement 1d).
In parallel experiments, we expressed SUR1_GFP-Y1353 stop in the absence of ANAP. Under such conditions, we observed a few cells with diffuse GFP fluorescence throughout the cytoplasm. We believe that the signal from such cells was derived from soluble GFP generated from an internal methionine after position 1353 and not read-through of the amber stop codon, as the distribution of this protein differed from that of SUR1_GFP (which was confined to the plasma membrane and presumably ER/Golgi). We never observed GFP fluorescence in unroofed membranes of such cells.

Surface expression assay
Surface expression of SUR1 constructs was assayed by their ability to chaperone HA-tagged Kir6.2 subunits to the plasma membrane using an adaptation of the method of Zerangue et al. (Zerangue et al., 1999) The HA tag plus linker (YAYMEKGITDLAYPYDVPDY) was inserted in the extracellular region following helix M1 of Kir6.2 between amino acids L100 and A101. HEK-293T cells were cultured in 12-well plates (STARLAB) on 19 mm cover slips treated with poly-L-lysine. Transfections were performed as described above, but DNA amounts were adjusted for the smaller scale (0.08 mg HA-tagged Kir6.2, 0.14 mg SUR1 constructs, 0.14 mg pANAP, 0.14 mg peRF1-E55D). Cells were incubated in DMEM (+10% FBS, 100 U/mL penicillin, and 100 mg/mL streptomycin) at 33˚C in a 95% air/5% CO 2 atmosphere for 48 hr in the presence or absence (as indicated) of 20 mM ANAP. Following incubation, cells were rinsed on ice with PBS and fixed for 30 min in PBS + 10% neutral buffered formalin. Cells were subsequently washed twice in PBS and blocked with 1% bovine serum albumin (BSA) in PBS for 30 min at 4˚C. Cells were then incubated for 1 hr at 4˚C in PBS + 1% BSA with a 1:1000 dilution of rat anti-HA monoclonal antibody (Roche; Basel, Switzerland) and washed 5 Â 10 min on ice with PBS + 1% BSA. HRP-conjugated goat anti-rat polyclonal antibody (diluted 1:2000, Jackson ImmunoResearch; Ely, UK) was applied in PBS + 1% BSA and cells were incubated an additional 30 min at 4˚C. Finally, cells were washed 4 Â 10 min in PBS + 1% BSA and 5 Â 10 min in PBS on ice. After the wash, coverslips were removed from PBS and placed in clean, untreated 35 mm culture dishes. 300 mL of SuperSignal ELISA Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific) were added and the luminescence was measured using a Glomax 20/20 Luminometer (Promega; Madison, WI) after a 10 s incubation. For display purposes, data were normalized to the mean of the luminescence with wild-type SUR1. Background was assessed with cells that were transfected with HA-tagged Kir6.2, but no SUR1.

Chemicals
ANAP trifluoroacetate salt and methyl ester were obtained from Asis Chemicals (Waltham, MA). A 1 mM stock of the salt was prepared in 30 mM NaOH and stored at À80˚C. The methyl ester was dissolved at 5 mM in DMSO and stored at À20˚C. The two forms were used interchangeably, as the final product (when ANAP is incorporated into a protein) is identical and in our experience both free acid and methyl ester were membrane permeant. Trinitrophenyl (TNP) nucleotide analogues were purchased from Jena Bioscience (Jena, Germany). 1 mM stocks were prepared in buffer and stored at À20˚C. Stocks were verified using UV-Vis spectroscopy (Beckman Coulter DU800 spectrophotometer; Pasadena, CA) as needed. Paper chromatography (mobile phase 40:10:25 n-butanol, glacial acetic acid: H 2 O) was used to verify that older TNP-ATP stocks had not undergone hydrolysis. All other chemicals were from Sigma. Diazoxide was prepared in a 34 mM stock in 100 mM KOH and stored at À20˚C before use. Tolbutamide was dissolved at 100 mM in DMSO and stored at À20˚C.

Data presentation and statistics
Error bars represent ±SEM. The boxes in Figure 2e and Figure 4-figure supplement 1a,b represent ±SD. The number of experiments (n) represents the number of patches or membranes used in a given experiment. These can be found in the tables for binding, electrophysiology, and time course data, and figure legends for other experiments. For surface expression assays, n represents the number of transfected coverslips. Plots and curve fitting were generated using Origin 9.1 (OriginLab Corporation; Northampton, MA). Curves in the figures (and parameters reported in the figure legends) represent fits to the combined data sets from multiple experiments, with errors representing the standard error of the fits. Shaded regions indicate 95% confidence intervals. Fit parameters reported in Tables are the mean ±SEM (Tables 1, 2 and 5) or ±SD (Tables 3 and 4) of the fits to individual experiments. Pairwise comparisons were performed using two-tailed Student's t-tests with the Welch correction (not assuming equal variance) and the Bonferroni correction for multiple comparisons, as needed. Comparisons of EC 50 values were performed on log-transformed data as the untransformed EC 50 is not normally distributed. Protein structures were displayed using PyMOL (Schrö dinger, LLC; New York, NY).