A mutually induced conformational fit underlies Ca2+-directed interactions between calmodulin and the proximal C terminus of KCNQ4 K+ channels

Calmodulin (CaM) conveys intracellular Ca2+ signals to KCNQ (Kv7, “M-type”) K+ channels and many other ion channels. Whether this “calmodulation” involves a dramatic structural rearrangement or only slight perturbations of the CaM/KCNQ complex is as yet unclear. A consensus structural model of conformational shifts occurring between low nanomolar and physiologically high intracellular [Ca2+] is still under debate. Here, we used various techniques of biophysical chemical analyses to investigate the interactions between CaM and synthetic peptides corresponding to the A and B domains of the KCNQ4 subtype. We found that in the absence of CaM, the peptides are disordered, whereas Ca2+/CaM imposed helical structure on both KCNQ A and B domains. Isothermal titration calorimetry revealed that Ca2+/CaM has higher affinity for the B domain than for the A domain of KCNQ2–4 and much higher affinity for the B domain when prebound with the A domain. X-ray crystallography confirmed that these discrete peptides spontaneously form a complex with Ca2+/CaM, similar to previous reports of CaM binding KCNQ-AB domains that are linked together. Microscale thermophoresis and heteronuclear single-quantum coherence NMR spectroscopy indicated the C-lobe of Ca2+-free CaM to interact with the KCNQ4 B domain (Kd ∼10–20 μm), with increasing Ca2+ molar ratios shifting the CaM-B domain interactions via only the CaM C-lobe to also include the N-lobe. Our findings suggest that in response to increased Ca2+, CaM undergoes lobe switching that imposes a dramatic mutually induced conformational fit to both the proximal C terminus of KCNQ4 channels and CaM, likely underlying Ca2+-dependent regulation of KCNQ gating.

Since the unexpected revelation that the well-known Ca 2ϩ sensitivity of voltage-gated Ca 2ϩ channels (VGCCs) and SKtype Ca 2ϩ -activated K ϩ channels are due to direct interactions with CaM, without the need for any kinases (12)(13)(14), a number of other channels have been shown to be similarly regulated by Ca 2ϩ ions (15). Particularly studied for CaM actions on VGCCs is the issue of apoCaM pre-association with the channels and the complex dynamic changes of the configuration of CaM between metal-free and Ca 2ϩ -loaded states. These changing configurations often involve "lobe switching" and have proved surprisingly distinct between the Ca V 1 (L-type) and Ca V 2 (Nand P/Q-type) channels (16 -22). For SK channels, similar issues are being studied, stemming from the surprising early conclusion that the high-affinity C-lobe of CaM is involved in pre-association with the channels (with neither EF-hand in the C-lobe occupied by Ca 2ϩ and the lower-affinity N-lobe acting as the Ca 2ϩ sensor, or "Ca 2ϩ switch," for gating). Consistent with that earlier structural hypothesis (12,(23)(24)(25) is a recent cryo-EM structure of a Ca 2ϩ /CaM-bound SK4 channel (26).
In neurons, heart, and smooth muscle, tetrameric voltagegated KCNQ (Kv7, "M-type") K ϩ channels play critical roles in regulating cellular excitability (27). CaM has been shown to regulate the trafficking and expression of KCNQ channels as well as their gating (28 -31). CaM acts as the Ca 2ϩ sensor for KCNQ channels via direct interactions with the proximal C terminus, thereby mediating the Ca 2ϩ -dependent modulatory action of several types of receptors linked to phospholipase C (7,30,32,33). It has been presumed for KCNQ channels that Ca 2ϩ loading of certain EF-hands of CaM induces a conformational change that inhibits channel opening. For KCNQ1-containing channels, however, Ca 2ϩ loading of CaM augments opening (34), perhaps in accord with their role in cardiomyocytes, inner ear, and epithelia in which KCNQ1 almost always is expressed together with KCNE ␤-subunits (35,36). In all cases, the extent and manner of pre-association of apoCaM with the channels and the nature of that conformational change are vigorously under debate.
It has been argued that apoCaM is required for KCNQ channels to properly function (28,31,34,(37)(38)(39), but it is still unclear whether truly metal-free CaM pre-associates with the channels. All studies have shown two highly conserved domains in the proximal C terminus, the A and B domains, as the loci of CaM actions (Fig. 1, A and B). This proximal half of the C terminus, which we call the regulatory domain (RD), besides containing highly conserved A and B domains involved in CaM interactions (39,40), also contains sites of regulation of opening by phosphatidylinositol 4,5-bisphosphate (PIP 2 ) (41) and protein kinase C, the regulatory site of the latter being just after the B domain (42)(43)(44). Rich interplay between these molecules at the RD is proposed to exquisitely regulate KCNQ channels.
Recent structural investigations suggest either that Ca 2ϩ / CaM embraces both the A and B domains of KCNQ1, -4, and -5 and KCNQ2/3 hybrids (45)(46)(47) or that Ca 2ϩ -loading of CaM induces the A domain to be released from the trimeric complex, leaving Ca 2ϩ -loaded CaM to wrap tightly around the B domain alone (48,49). In contrast, a recent solution NMR study of a similar complex of the A and B domains of KCNQ2 and CaM suggested only minor changes in the structure of the complex between low and high [Ca 2ϩ ], arguing against a dramatic structural change in KCNQ channels in response to intracellular rises in [Ca 2ϩ ] (50).
We investigated this issue for KCNQ4, as this isoform is expressed in cells and tissues mainly as homomeric channels, simplifying our interpretations. We used a gamut of biophysical chemical and structural analyses, such as heteronuclear singlequantum coherence NMR spectroscopy (HSQC-NMR) and X-ray crystallography. In our investigations, we used separate synthetic peptides corresponding to the A and B domains of KCNQ2-4 to gain better insight into CaM interactions with the domains independent from each other. We were scrupulously careful to know both the free [Ca 2ϩ ] and the stoichiometric ratios of [Ca 2ϩ ], CaM, and A and/or B domains in all of our experiments. The goal of this inquiry was to build a stepwise model of the mechanism of CaM binding to the KCNQ4 C terminus from Ͻ10 nM to physiologically high [Ca 2ϩ ].

The A and B domains appear intrinsically disordered in the absence of CaM
The amino acid sequences of the A and B domains of KCNQ channels are conserved within the KCNQ family (Fig. 1B). The A domain, also known as the "A helix," contains a highly conserved IQXXR amino acid sequence characteristic of IQ CaMbinding motifs found in a multitude of other CaM target proteins (51). The A domain also contains 1-12 and 1-16 motifs. The B domain, also called the "B helix," contains multiple canonical and noncanonical CaM-binding motifs, including 1-5-10 and 1-14 hydrophobic anchoring residues (52). Having multiple CaM motifs within the RD suggests that CaM may adopt several different orientations when binding to KCNQ channels, as recently suggested (50).
The existing co-crystal and cryo-EM structures of CaM with purified full channel and purified fragments containing the A and/or B domains show the A and B domains adopting an ␣-helix secondary structure or assembled in coiled-coil arrangements when embraced by CaM (45)(46)(47)49). However, it is not known whether the free A and B domains of KCNQ channels adopt helical structure in the absence of CaM, despite their common reference as the "A helix" or "B helix." We used CD spectroscopy to analyze the secondary structure of synthesized peptides corresponding to the A or B domains of KCNQ2-4 subunits. Although peptides typically appear disordered in the absence of a binding partner, CD can indicate whether they adopt secondary structure on their own, such as that corresponding to the NSCate domain of L-type Ca 2ϩ channels, which has ␣-helical structure in the absence of CaM (53). Each of the KCNQ peptides displayed single peaks at 205 or 220 nm (Fig. 1C), suggesting that the A and B domains may be intrinsically disordered in the absence of CaM and require CaM to adopt their coiled-coil nature in reported structures. In contrast, CaM displayed dual peaks at 208 and 222 nm, which is the signature for proteins with ␣-helices. Because the CD buffer contained Ca 2ϩ in equimolar ratio to protein, the CaM in this experiment is expected to only be partially Ca 2ϩ -bound. This Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels experiment also suggests that Ca 2ϩ does not directly induce the secondary structure of the A and B domains. To highlight the malleable nature of these CaM-binding domains, we generally call these regions the "A and B domains" in this study. Under "Discussion," we comment on the importance of these observations in the role of CaM in proper channel expression.

The independent Q4A and Q4B peptides adopt an anti-parallel ␣-helical conformation enveloped by CaM in the presence of Ca 2؉
To determine with high resolution how CaM and the independent A and B domains of KCNQ4 spontaneously assemble under conditions of high [Ca 2ϩ ], compared with existing structures of CaM that have been co-expressed with proteins of the connected KCNQ-AB fragments, we obtained the X-ray crystal structure of Ca 2ϩ /CaM in complex with KCNQ4 A domain (Q4A) and KCNQ4 B domain (Q4B) peptides. We refer to this structure as Ca 2ϩ /CaM:Q4A:Q4B (PDB entry 6N5W) ( Fig. 2A). Elongated, hexameric crystals grew to full size after 8 days at room temperature in the presence of 2 mM free [Ca 2ϩ ]. No crystals were observed in Ca 2ϩ -free conditions (in a formation buffer of HBS plus 2 mM EGTA) in our extensive array of crystallization screens. The X-ray structure of Ca 2ϩ /CaM:Q4A: Q4B was determined by molecular replacement using the structure of Ca 2ϩ /CaM:KCNQ1-AB (PDB entry 4UMO) as a model and refined to a resolution of 2.15 Å (crystallographic statistics found in Table S1). The trimeric structure includes residues 4 -147 of CaM, Arg 338 -Asp 356 of Q4A, and Asp 525 -Phe 549 of Q4B. The schematic in Fig. 2B illustrates our understanding of this complex as it exists within the entire KCNQ4 subunit. A common observation of existing structures of Ca 2ϩ / CaM with the A and B domains of KCNQ1 and KCNQ4 and of the frog oocyte KCNQXem channel is that Ca 2ϩ is shown coordinated by both EF-hands in the N-lobe but is only occasionally present in EF-hands of the C-lobe. This has been a surprising finding, because, as mentioned above, for free CaM in isolation, the EF-hands in the C-lobe have a higher affinity for Ca 2ϩ than does the N-lobe, although this discrepancy was also seen previously for VGCCs and SK channels (54). Our Ca 2ϩ /CaM:Q4A: Q4B crystal structure also shows EF-hands I and II of the CaM N-lobe to each coordinate a Ca 2ϩ ion. EF-III and -IV are empty, despite this complex being formed in a buffer containing an excess of Ca 2ϩ (albeit modest) relative to the number of EFhands (222 M CaM and 2 mM Ca 2ϩ ). The crystallization buffer contained citrate, which also binds Ca 2ϩ but with much lower affinity than CaM (K a ϭ 10 3.5 M Ϫ1 ) (55). Citrate has also been shown to directly interact with the C-lobe of CaM, but we do not observe such an interaction in this structure (56). We also point out that the Ca 2ϩ /CaM:Q1AB structure was not obtained in a high-citrate buffer (48) but is very similar to this structure (RMSD ϭ 1.68 Å) and was also found to lack Ca 2ϩ ion density in A, representative schematic of a KCNQ protein subunit depicting the S1-S6 transmembrane helices, the intracellular N and C termini, and the A domain (teal) and B domain (gold) within the proximal half of the C terminus, referred to as the RD. The schematic does not show the likely intimate proximity of the RD to the plasma membrane. For reference, the PIP 2 interaction sites are shown as gray circles, and the protein kinase C phosphorylation site conserved between KCNQ2-5 is represented by a black circle. B, sequence alignments of the A domain (left) and B domain (right) of human KCNQ1-5 subunits taken from Uniprot. All sequences represent "isoform 1" of each subunit, with the exception of the KCNQ2 sequences that represent the universally used isoform 4 and those of KCNQ4a. Alignments were performed using the PRALINE alignment tool, and the colors were adjusted using Photoshop Elements. The dots above the alignments indicate the interactions with Ca 2ϩ /CaM shown in Fig. S3 and Fig. 2. Black dots indicate the residues with strong interactions between the KCNQ4 peptides and CaM that are different from those of KCNQ1, and gray dots indicate strong interactions for which the interacting residues are conserved. Open circles indicate those KCNQ4 residues having nonbonded contacts with CaM, and the star indicates the residues of KCNQ1 that may cause changes of the CaM backbone compared with KCNQ4. C, CD spectra of the peptides used in this study and of CaM. Two dashed vertical lines at 208 and 222 nm indicate the points of deflection typical of proteins such as CaM (black) with high helical content. All of the Q2-4 A and B peptides appear to lack helical content in the absence of CaM.

Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels
the C-lobe EF-hands, suggesting this to be more related to [Ca 2ϩ ] or that Ca 2ϩ ions dissociate from the C-lobe upon complex formation. Contrary to these findings, the Ca 2ϩ /CaM: Q3AQ2B structure (PDB entry 5J03) found Ca 2ϩ coordinated in all four EF-hands formed in a crystallization buffer containing a stoichiometric ratio of [Ca 2ϩ ] to [CaM] of 125:1. Thus, we must admit the likely possibility that a greater stoichiometric excess of Ca 2ϩ to CaM than what we and others have used is required to fully load all "loadable" C-lobe EF-hands under crystallization conditions (45). A color plot of crystallographic B-factors of the crystal structure shows the relative level of disorder, and we observe that the EF-hands III and IV map the highest disorder (Fig. S1). This higher potential for disorder in the C-lobe may arise from this trimeric configuration, a result of the C-lobe releasing previously bound Ca 2ϩ ions and rendering EF-hands III and IV less important for stabilizing the structure. The preference of Ca 2ϩ for the N-lobe in this CaM-ion channel co-crystal structure is consistent with the N-lobe, not the C-lobe, being obligatory for CaM function as the Ca 2ϩ sensor that regulates channel opening (57), but as our model in Fig.  10 illustrates, does not mean that the C-lobe is not the first lobe to load Ca 2ϩ ions.
Similar to the other published Ca 2ϩ /CaM:KCNQ structures, our crystal structure shows the A and B domains adopting helical conformations in the presence of Ca 2ϩ -loaded CaM. Taking the CD spectra into account, we suggest that CaM imposes ␣-helical structure on the A and B domains of KCNQ channels. The structure shows CaM wrapped around the A and B domains, assembled into a coiled-coil configuration in antiparallel orientation, with the N-lobe bending around the B domain and the C-lobe folded around the A domain (Fig. 2  Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels atoms included in the calculation using PyMOL. An interaction plot summarizes multiple van der Waals (nonbonded contacts) in addition to several hydrogen bonds and salt bridges that hold CaM together with the A and B domains (Fig. S2). The residues involved in these interactions are indicated by circles above the residues in the conserved alignment panel in Fig. 1B. This interaction profile is almost identical, as shown, with that reported previously (48) and suggests a high degree of similarity with most of the other structures of Ca 2ϩ /CaM interacting with KCNQ1-5. A notable exception is the reported structure of Ca 2ϩ /CaM interacting only with the KCNQ4 B domain (48,49). Expanded representations of the interior of the Ca 2ϩ / CaM:Q4A:Q4B co-crystal structure obtained here reveal that these interactions of the A and B domains occur with the two loops within the N-and C-lobes and the N-lobe-C-lobe linker region of CaM (Fig. 2, C and D). We also observed that the more highly conserved residues of A and B domains (pink and orange residues, colored according to the scale in Fig. 1B) tend to face the interior of CaM rather than facing toward each other.
An overlay of this structure with the KCNQ1-AB domainswapped complex (PDB entry 4UMO) (Fig. 2E) (RMSD ϭ 1.68 Å) indicates that the only noticeable difference in these interactions localizes to the N-lobe-C-lobe linker of CaM. The expanded view reveals that Ile 539 of Q4B allows the CaM linker to have closer contact with the B domain (Fig. 2F). The bulker side chain of the KCNQ1 homolog, Arg 519 , appears to push the linker further away from the B domain. Whether this small structural difference translates to explaining the functional differences between CaM actions on KCNQ1 and KCNQ4 channels remains to be seen. Overall, these data, featuring the assembly of independent proteins, confirm that the peptides interact with Ca 2ϩ /CaM similarly to the co-expressed pre-assembled AB proteins with CaM.

Ca 2؉ /CaM binds the B domain with higher affinity than the A domain
We wondered whether there might be an obligate "stepwise" mechanism of complex formation between CaM and the A and B domains. The direct biochemical binding affinity of fulllength CaM for each KCNQ domain individually has not been reported, although indirect measurements of apparent affinities indicate that Ca 2ϩ /CaM binds the A domain of KCNQ2 in the nanomolar range (39,58). We used isothermal titration calorimetry (ITC) to assess the thermodynamic parameters of Ca 2ϩ /CaM interactions with the A and B domains of KCNQ2-4. Peptides corresponding to the A domains each displayed a moderate binding affinity for Ca 2ϩ /CaM, with K d values of ϳ0.44 -1.2 M (Fig. 3 (top) and Table 1). The B domain peptide of KCNQ2 also displayed a moderate affinity for Ca 2ϩ / CaM (K d ϳ0.24 M). The B domains of KCNQ3 and KCNQ4, however, bound to Ca 2ϩ /CaM very tightly, with K d values between 4 and 13 nM (Fig. 3, bottom). The stoichiometry for Ca 2ϩ /CaM binding to each peptide was near 1:1, with any differences explainable by remaining uncertainty in the precise peptide concentrations. Although we find here that fully Ca 2ϩloaded CaM has far greater affinity for the B domain versus the A domain when present alone, when together, Ca 2ϩ -loaded CaM must strongly interact with both the A and B domains. Because the crystal structures show that a single CaM embraces the A and B domains together, the very different affinities of the two domains are consistent with there being a specific order to

Table 1 Summary of ITC results of KCNQ A and B domain peptides binding Ca 2؉ /CaM
Values are mean Ϯ S.D.

Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels
how CaM induces formation of this trimeric complex. This is one of the central advances of this work.

Formation of the Ca 2؉ /CaM:Q4A:Q4B complex necessitates a highly ordered mechanism
Additional ITC experiments were performed to further characterize the interactions between Ca 2ϩ /CaM and peptides of the KCNQ4 A and B domains (Q4A and Q4B peptides) simultaneously. First, we tested whether the A and B domains have the ability to interact in the absence of CaM. We did not observe any evidence of binding between the A and B domains using ITC (Fig. 4A), and we also did not observe any crystals formed of only Q4A and Q4B peptides. Both suggest that CaM is necessary to bring these two domains together. Next, we tested whether Ca 2ϩ /CaM must bind the A domain or the B domain first to form a stable trimeric complex. We found that Q4A did not display a measurable interaction with a preformed complex of Ca 2ϩ /CaMϩQ4B (Fig. 4B). Reversing this order, however, showed Q4B to bind to the preformed complex of Ca 2ϩ /CaMϩQ4A with very high affinity (K d ϭ 0.5 Ϯ 0.2 nM) ( Fig. 4C), even higher than of Q4B binding to Ca 2ϩ /CaM alone. This is congruent with the data of Xu et al. (49), who suggested that CaM may be required to interact with the A domain first to form the trimeric complex with the B domain and who reported a putative structure of Ca 2ϩ /CaM wrapped around the B domain only. Although we conclude that the ultimate configuration of Ca 2ϩ -loaded CaM is wrapped around both domains, we cannot rule out the possibility that Ca 2ϩ /CaM is wrapped around solely the B domain, under certain conditions that we do not yet know. Combining these data with our above ITC experiments, we suggest that an "energy barrier" must be overcome to form the trimeric complex: Ca 2ϩ /CaM first binds the "lower affinity" A domain before binding the "higher affinity" B domain to form the trimeric complex. This would imply that for Ca 2ϩ /CaM to form a fully functional complex with both the A and B domains of KCNQ4 channels, a mechanism must be in place to ensure the proper order of binding.

ApoCaM binds the B domain but not the A domain of KCNQ4
We were not able to obtain a crystal of apoCaM with the separate KCNQ4 A and B domains in our screens, as mentioned above. Additionally, our ITC data showed the Q4A to display negligible interaction with apoCaM or a preformed complex of apoCaMϩQ4B in buffer containing 1 mM EGTA (Fig. S3, A and  B). However, we noticed weak interactions between Q4B and apoCaM, characterized by an increase in endothermic heat exchange followed by a decrease in the endothermic profile ( Fig. S3C). A similar isotherm was observed for Q4B binding to apoCaMϩQ4A (Fig. S3D), indicating that apoCaM might weakly interact only with the B domain and not the A domain. Because we were unable to accurately determine a binding constant from these ITC data due to limited amounts of these proteins, we turned to microscale thermophoresis (MST), which requires a much smaller quantity of protein. We examined the interaction of fluorescently labeled apoCaM (200 nM) with Q4A or Q4B in chelexed buffer (ChHBS) supplemented with 0.5 mM EGTA, ensuring that this low concentration of CaM was wholly free of Ca 2ϩ . These experiments revealed that apoCaM interacts with the B domain of KCNQ4 with a K d of ϳ10 M (confidence interval 6 -17 M; Fig. 5A), which is within the range of free [CaM] in cytoplasm under very low [Ca 2ϩ ] conditions. Similar to our ITC results, the MST experiments suggest that the A domain does not interact with apoCaM (Fig. 5B).

The C-lobe of the apoCaM is loosely associated with the KCNQ4 B domain
We used solution HSQC-NMR to track interactions of apoCaM with Q4A and Q4B. Spectroscopy was performed using [ 15 N]apoCaM and Q4A and Q4B in 1 mM EGTA. Peptides were added to a slight excess of 150 M CaM based on the 1:1 stoichiometry we had observed by ITC under high-[Ca 2ϩ ] conditions. Fig. 6A shows the NMR spectrum of isolated apoCaM, with expanded regions (Fig. 6, B-D) comparing critical spectral peaks of apoCaM before and after the addition of Q4A and Q4B. This apoCaM spectrum closely matches the assignments of the vertebrate apoCaM spectrum shared by John Putkey (UT Health, Houston TX), further confirming that the starting CaM was indeed in the Ca 2ϩ -free state. The addition of Q4A did not cause any chemical shifts or peak alterations to those of isolated apoCaM (Fig. 6B). In contrast, the addition of Q4B induced obvious changes to the apoCaM spectrum, revealing the apoCaM residues affected by interaction with the B domain. We observed significant changes in peaks corresponding to res-

Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels
idues Ala 88 , Asn 97 , Ile 100 , and Ser 101 and between Asp 131 and Glu 139 , as shown in the expanded region in Fig. 6C and the full spectrum in Fig. S4. The addition of Q4B to the solution containing apoCaMϩQ4A (Fig. 6D) caused nearly identical chemical shifts or alterations to the spectral peaks of apoCaMϩQ4B, further suggesting that the B domain alone is responsible for interactions with apoCaM. To determine the loci of apoCaM interactions with the B domain, we mapped the residues displaying significant chemical peak alterations observed from the above experiment onto an existing solution NMR structure of apoCaM (PDB entry 1DMO) (5). That analysis unequivocally revealed the C-lobe of apoCaM to interact with the B domain of KCNQ4 (Fig. 6E). Specifically, the residues that were most affected by the Q4B titrations localized to EF-hands III and IV. We plotted those significant changes in peak height, or line broadening, to Q4B titrated over a range of concentrations, which yielded K d ϭ 16 Ϯ 5 M (Fig. S5). Although peak height is not a widely accepted method to determine equilibrium constants, as line broadening is not a direct report of binding, this result is very similar to that obtained using our MST experiments in Fig. 5, thus confirming by two distinct approaches that apoCaM binds the KCNQ4 B domain with a moderate affinity consistent with known cellular physiology. Based on these results, the schematic in Fig. 6F represents our interpretation of the structural interaction between apoCaM and a single KCNQ4 subunit at resting cytoplasmic [Ca 2ϩ ]. It suggests a conformation of the apoCaM:KCNQ4 complex quite distinct from our Ca 2ϩloaded crystal structure (Fig. 2) and others (45,46,50), of Ca 2ϩ / CaM in complex with the A and B domains of KCNQ1-4, and the cryo-EM structure of Ca 2ϩ /CaM:KCNQXem (47), which all show the B domain interacting with the N-lobe of CaM and the A domain embraced by the C-lobe. These results challenge our ini-tial supposition that apoCaM would be constitutively bound to the A domain and suggest a much more dynamic mechanism of Ca 2ϩ directing CaM interactions with KCNQ4 channels. The implications of these results are discussed below.

Lobe switching of CaM occurs with half-loaded EF-hands
The above findings imply that a rise in [Ca 2ϩ ] causes a dramatic lobe switching between CaM and the KCNQ4 A and B domains. To test this hypothesis, we performed HSQC-TROSY experiments over a range of [Ca 2ϩ ] designed to cover Ͻ1% loading of the four EF-hands of CaM to that maximal. To avoid competition between EGTA and CaM for Ca 2ϩ , we carefully controlled for [Ca 2ϩ ], using ChHBS in the absence of EGTA as our starting "zero-Ca 2ϩ " buffer. This allows a molar stoichiometry of [Ca 2ϩ ]/[CaM] of no more than 1:100, resulting in an apoCaM spectrum almost identical to the spectrum of apoCaM in HBS buffer ϩ 1 mM EGTA (Fig. S6). We monitored changes of the emission spectrum of double-labeled, deuterated CaM ([ 2 H-15 N]apoCaM) (50 M), which allows better resolution of the spectrum, requiring lower protein concentration than the single-labeled [ 15 N]apoCaM used in the earlier experiments. The better resolution was necessary to track changes in the CaM and CaMϩQ4B spectra over increasing Ca 2ϩ titrations.
With the first addition of Ca 2ϩ at 1:4 stoichiometry (1 Ca 2ϩ ion/4 EF-hands), we observed obvious changes in the peaks of isolated CaM as the spectra displayed little overlap (Fig. 7 (A-C), green and purple peaks). At higher ratios of Ca 2ϩ /EFhands, more peaks began to overlap due to fewer peak shifts or changes, indicating less response to increased [Ca 2ϩ ] as the EF-hands became more loaded (Fig. 7 (A-C), purple and orange peaks). This suggests that CaM on its own is very responsive to small rises in [Ca 2ϩ ]. In contrast, CaMϩQ4B displayed much Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels spectral overlap at a 1:4 ratio of Ca 2ϩ to EF-hands, observed as very few changes in the peaks (Fig. 7 (D-F), green and purple peaks). At the point of half-loading of Ca 2ϩ to the EF-hands (2:4 stoichiometry), the CaMϩQ4B spectrum displayed a sudden shift, observed by the separation of orange peaks from the purple peaks. A graphical plot is shown summarizing the significant peak changes of CaM and CaMϩQ4B with each addition of Ca 2ϩ (Fig. 7G). Overall, we suggest that Q4B changes the relationship between CaM and [Ca 2ϩ ]. When apoCaM is bound to Q4B, more Ca 2ϩ is needed to cause a change in the configuration of CaM.
We were able to track many of the amino acid residues from "zero" Ca 2ϩ (0:4) up to a ratio of 2:4 or 3:4 Ca 2ϩ /CaM EF-hands. Although more shifts were observed in the spectra, as previously noted in Fig. 7, only those residues that changed in position Ͼ55% from the previous titration, and that we could track from the initial zero [Ca 2ϩ ] spectrum, are plotted. Fig. 8A shows that those residues localize to both lobes of apoCaM changed over the range of 0:4 to 1:4 Ca 2ϩ /CaM EF-hands, and even more residues localized throughout the protein changed at a 2:4 stoichiometry (Fig. 8B).
Interestingly, in the presence of Q4B, only the C-lobe residues of CaMϩQ4B displayed peak alterations over a range from 0:4 to 1:4 Ca 2ϩ /CaM EF-hands (Fig. 8C). This suggests that Ca 2ϩ binds CaM first to the C-lobe of Q4B-bound CaM. As shown in Fig. 6, the C-lobe interacts with Q4B in the absence of Ca 2ϩ , so this finding also suggests that Ca 2ϩ may displace Q4B from the C-lobe. At a 2:4 stoichiometry, residue emissions shifted throughout the CaM protein (Fig. 8D), followed by changes mostly within the N-lobe upon further increases of Ca 2ϩ molar ratio (Fig. 8E). Fig. 8 illustrates our interpretation of the changes of CaM structure in relation to the B domain, based on these NMR data. The coordinated response of CaM to Ca 2ϩ , in which responding residues shift from the C-lobe  ]. We conclude that 1) Ca 2ϩ ions interfere with Q4B interactions with apoCaM; 2) Ca 2ϩ ions initially load the C-lobe of the apoCaM:Q4B complex, instead of the N-lobe; and 3) a highly coordinated sequence of binding-unbinding-rebinding steps occurs over a physiological range of very low to high [Ca 2ϩ ], which we here refer to as a "lobe-switching mechanism." We elaborate on this in Fig. 10.

Ca 2؉ /CaM-mediated inhibition of KCNQ4 channels is not associated with shifts in voltage dependence
Regarding the CaM-mediated inhibition of KCNQ channel gating by rises in [Ca 2ϩ ] i , there have been conflicting reports of this action involving shifts of the voltage dependence of activation or changes in gating kinetics. Studying endogenous M current in sympathetic neurons or cloned KCNQ2/3 heteromers or KCNQ2-5 homomers, our group has not ever observed any changes in activation or deactivation kinetics induced by any form of CaM over the full range of physiological [Ca 2ϩ ] i (7,57). There are two alternatively spliced isoforms of human KCNQ4, called KCNQ4a and KCNQ4b, displaying very distinct activation kinetics. Most laboratories, including ours, have used KCNQ4a, which we and others call "KCNQ4." Xu et al. (59) found overexpression with KCNQ4a of a "dominant-negative"

Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels
CaM, in which all four EF-hands have been mutated to be incapable of loading Ca 2ϩ ions, shifted the voltage dependence of activation by ϳϪ36 mV. Other laboratories have reported much more modest changes of voltage dependence of ϳ10 mV for KCNQ2 or KCNQ3 homomers or KCNQ2/3 heteromers, with only minor changes in kinetics (30, 60 -62). The case of KCNQ1-containing channels is unique in that 1) Ca 2ϩ -loaded CaM augments currents from those channels, instead of depressing them (34), and 2) the S2-S3 linker of KCNQ1 is wholly nonconserved with that of KCNQ2-5. Indeed, a cysteine triplet conserved among the latter is the site of reactive oxygen species, whose action dramatically increases the opening of KCNQ2, KCNQ4, and KCNQ2/3, but not KCNQ1-containing channels (63)(64)(65), whereas the S2-S3 linker of KCNQ1 has been recently suggested to gently engage bound CaM, resulting in shifts in voltage dependence of those channels also by ϳ10 mV (47). A recent crystal structure of apoCaM with the joined A and B domains of KCNQ4 suggested a very large lobespecific action of CaM on the voltage dependence of activation of KCNQ4, by Ͼ35 mV, and a strong effect on activation kinetics ϳ3.4-fold in magnitude (48). Because that group coupled such dramatic effects on voltage dependence and kinetics with their suggested structural model and were so divergent from our previous studies (7, 57), we decided to revisit this issue.
Chinese hamster ovary (CHO) cells were co-transfected with KCNQ4 and either WT CaM, CaM(1,2), CaM(3,4), or CaM(1,2,3,4), where CaM(1,2) and CaM (3,4) refer to N-or C-lobe mutants that cannot bind Ca 2ϩ ions in that lobe, respectively, and CaM(1,2,3,4) (i.e. D20A/D56A/D93A/D129A) cannot bind Ca 2ϩ in either lobe. All of the constructs, which were the same cDNA as used previously by us (57), were resequenced in their entirety and found to be correct. As before, we performed experiments in the perforated patch variant of wholecell voltage clamp and, moreover, ensured that the resulting

Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels
currents were not so large as to either induce series resistance errors or "soak up" all of the PIP 2 molecules in the plasma membrane, both of which might cause artifacts in our data. Cell lysates from each group were also immunoblotted with anti-KCNQ4 or anti-CaM antibodies to ensure a reasonable balance of expression of the two proteins. We found that such cells expressed very similar currents from each group, with very similar properties (Fig. 9). For cells co-transfected with KCNQ4 and WT CaM, CaM(1,2), CaM(3,4), or CaM(1,2,3,4), the midpoint voltage of activation (V1 ⁄ 2 ) values were Ϫ20.3 Ϯ 0.7, Ϫ18.1 Ϯ 5.0, Ϫ18.2 Ϯ 5.3, and Ϫ19.1 Ϯ 1.1 mV, respectively (n ϭ 6, 5, 4, and 5). The activation kinetics at 10 mV were likewise not significantly divergent for cells in each group, having the weighted time constants (see "Experimental procedures") of 420 Ϯ 145, 395 Ϯ 96, 355 Ϯ 68, and 291 Ϯ 37 ms, respectively. Likewise, there were no significant differences in the deactivation kinetics at Ϫ60 mV, which were 70 Ϯ 5, 50 Ϯ 6, 90 Ϯ 15, and 85 Ϯ 6 ms, respectively. Whereas the modest number of cells studied does not rule out minor changes in activation kinetics in the presence of different forms of CaM, any such differences could only be small.

Discussion
The results from this study highlight several key elements of the dynamics of CaM interaction with the A and B domains of KCNQ2-4. First, we showed that the A and B domains of the channels are likely to be intrinsically disordered in the absence of CaM and that interactions with CaM Ca 2ϩ -loaded in at least its N-lobe imposes ␣-helical secondary structure on those domains. Next, we demonstrated that apoCaM does indeed bind KCNQ4 subunits with moderate affinity with only the C-lobe of apoCaM interacting with the B domain of KCNQ4. Finally, the major advance of this work is our finding that a dramatic lobe switching of CaM interaction with the KCNQ4 A and B domain occurs from low to high [Ca 2ϩ ].
Despite our findings that the A and B domains are likely to be disordered in the absence of CaM, the above experiments do not allow us to conclude whether apoCaM induces the helical structure on the B domain, a question planned for future studies. However, the co-crystal structures of KCNQ A-B with CaM all clearly show the A and B domains helical, whether in the presence of divalent cations, such as Mg 2ϩ or Ca 2ϩ , or after removal of either divalent from the preformed structure (48). This induced conformational change in CaM target proteins has been widely observed, such as for Ca 2ϩ /CaM-dependent kinases (66) and other ion channels (53, 66 -68). Our finding that apoCaM binds the B domain might account for the obligatory need for the presence of CaM reported by several laboratories either for functional expression of KCNQ channels or

Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels
their assembly in the plasma membrane of either tissue culture cells or neurons (28,34,37,39,60). The reason likely has to do with the obligatory interactions of all KCNQ channels for interactions with PIP 2 in the plasma membrane for function (69, 70). Given that the two most important domains of the channels for PIP 2 interactions are in the proximal C terminus (Fig. 10) (41,46,71), such a change in structure may prevent the RD from stable interactions with PIP 2 (30), thus hindering opening. Another group has also shown an intricate relationship between channel interactions with PIP 2 and CaM (44). However, as opposed to test tube experiments, all cells, especially excitable cells, express CaM at high levels in cytoplasm. As CaM is increasingly Ca 2ϩ -loaded in response to increasing [Ca 2ϩ ] i , we propose that the tight embrace of the A and B domains (now certainly helices) condenses the RD to a springlike structure, pulling it away from the plasma membrane and interfering with those KCNQ-PIP 2 interactions, thus causing suppression of M current. We present our model in more detail in Fig. 10.
To properly interpret our data, we must consider the physiology of CaM molecules and Ca 2ϩ ions in the cytoplasm. Because the affinity of CaM for almost all of its cellular targets greatly increases upon Ca 2ϩ loading (neurogranin being a notable converse example (6,(72)(73)(74) (75)(76)(77), such as in response to strong neuronal stimulation and rapid firing. It is important to note that, unlike the case of CaM actions on VGCCs, for which highly local free [Ca 2ϩ ] at the inner mouth of the pore may approach millimolar concentrations (16,78,79), no Ca 2ϩ ions are flowing through open K ϩ channels, and so it is likely that, as for the analysis of CaM actions on SK K ϩ channels, it is global [Ca 2ϩ ] that should be most relevant for our thinking. This supposition is tempered, however, by the discovery of KCNQ channels and Ca 2ϩ -permeable channels clustered together in microdomains in sensory neurons (80), and similar multichannel complexes likely exist in brain as well (81)(82)(83). Thus, we cannot say with certainty the precise [Ca 2ϩ ] in the local micro-environment of KCNQ channels in nerve, heart, and muscle that corresponds to CaM being maximally "switched on." Studied in isolation via ITC, we found Ca 2ϩ /CaM to have a much higher affinity for the B domains than for the A domains of KCNQ2-4, with KCNQ3 and KCNQ4 most notably so. However, the affinity increased by an order of magnitude when the A domain was present as well. Consistent with those measurements is the high-resolution crystal structure we obtained of Ca 2ϩ -loaded CaM wrapped around the A and B domains, very similar to that reported earlier for a variety of KCNQ subtypes (45)(46)(47)50). None of that work suggests a role of CaM in cross-linking the C termini of the subunits in the tetramer, but rather suggests that CaM interacts within individual subunits. A caveat to our conclusions is our ITC results, in which the addition of the A domain to a preformed Ca 2ϩ /CaM-B helix complex did not induce a thermodynamic signal; thus, we cannot rule out a configuration of Ca 2ϩ /CaM wrapped around the B domain alone under certain cytoplasmic conditions, as suggested by another group (48,49). It is important to remember that absolute affinities assayed in test tube experiments with isolated A and B domains and CaM molecules are likely much different from those when the RD is attached to the rest of the channel, namely coupled to the gating machinery, as well as the changes in affinities of the EF-hands of CaM for Ca 2ϩ when pre-associated with the relevant domains of the channels (18,22,54). However, we assert that the relative affinities and thermodynamic parameters are likely to parallel our findings in intact cell experiments, giving us insight into conformational changes over ranges of physiological [Ca 2ϩ ] and [CaM].
Our observations of the Ca 2ϩ titrations shifting from emission changes in C-lobe residues to those including the N-lobe of CaM prebound to Q4B lead us to propose a lobe-switching  shown as a helix). This twisting motion may impose torque on the PIP 2 interaction sites in the proximal C terminus, partially weakening their interactions with PIP 2 . 4, in the final step, under a strong [Ca 2ϩ ] signal (such as strong stimulation of certain G q/11 -coupled receptors), the EF-hands of the N-lobe become occupied by Ca 2ϩ ions, enhancing its affinity for the B domain, inducing it into a helical formation, retaining C-lobe binding to the A domain (still a helix). This final twisting motion may completely twist or pull away the PIP 2 interaction sites from the inner leaflet of the membrane, severely hindering the ability of the C terminus to bind PIP 2 , resulting in maximal Ca 2ϩ /CaM-mediated inhibition of neuronal M channels. The subnanomolar affinity of the CaM:AϩB trimeric complex may allow Ca 2ϩ ions to rapidly move in and out of EF-III and -IV while maintaining a stable complex during such elevated [Ca 2ϩ ] conditions; thus, accounting for crystals variably observed to contain Ca 2ϩ ions in the C-lobe EF-hands.

Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels
model. The issue of lobe dependence and/or specificity for CaM actions (both CDI and CDF) on VGCCs is represented by a vast literature that reveals surprisingly stark differences between L-type (Ca V 1) and N and P/Q-type (Ca V 2) Ca 2ϩ channels. For the former, the C-lobe is recognized as the "Ca 2ϩ sensor" provoking CDI, whereas for the latter, the N-lobe has been proposed to fulfill that role (16). Moreover, for both Ca V 1 and Ca V 2 channels, lobe switching has been proposed as a key structural mechanism, involving an N-terminal region of the channels (84,85), and it is tempting to think that this CaM-mediated regulatory mechanism is conserved among channel types (86). Notably, there have not been any structural analyses of CaM/ KCNQ channel interactions that include the N terminus of KCNQ channels, which have been suggested to play key roles in gating by interacting with the C terminus, involving syntaxin and CaM (62,87). Our thinking also takes into account our earlier work indicating Ca 2ϩ loading of the N-lobe of CaM to be critical for Ca 2ϩ /CaM-mediated suppression of KCNQ2, KCNQ4, and KCNQ2/3 channels, with Ca 2ϩ loading of the C-lobe unimportant. That work was performed on living cells expressing full-length, functional channels, studied under perforated patch whole-cell recording (57) rather than only analyses of C-terminal fragments.
We were unable to obtain a crystal of this or any other "apo" state under metal-free conditions using our crystallization screen of 1,728 conditions by assembling the discrete proteins in the absence of Ca 2ϩ . Our apoCaM data contrast with a recent report showing co-expressed (pre-assembled) apoCaM co-crystallized with the AB fragment of KCNQ4 (PDB entry 6B8L) in which Ca 2ϩ had been removed from the preformed complex (48) and another showing the solution NMR structure of apoCaM:KCNQ2-AB (PDB code 6FEG) that had been similarly co-expressed and purified (50). Aside from their absence of Ca 2ϩ ions, these "apo" structures are still quite similar to that found for the Ca 2ϩ /CaM:KCNQ-AB complexes (45)(46)(47). We note that the primary difference between these apo structures and the apo model we propose is in how the complex was assembled; in this study, we combined each Ca 2ϩfree protein/peptide separately with the crystallization buffer, whereas all others had been co-expressed and preassembled in a divalent cation-rich medium (LB/2YT) prior to removal of the Ca 2ϩ ions with EGTA. We think this suggests that the complex itself is more stable than the transient Ca 2ϩ state of the CaM-EF-hands.
Millet and colleagues (50) called their structure "intermediate," with regard to metal, consistent with this complex not being wholly Ca 2ϩ -free, and reported that complete unloading of all Ca 2ϩ from the preformed complexes was difficult. Under those circumstances, any further loading of Ca 2ϩ ions to the complex displayed a K d value using FRET assays of ϳ1 M (50), a value similar to that globally in the cytoplasm of neurons in response to a stimulus. Our solution NMR and MST experiments suggest that the C-lobe of apoCaM weakly interacts with the KCNQ4 B domain under Ca 2ϩ conditions at which at most 1% of the CaM EF-hands could be Ca 2ϩ -loaded, and at extremely low [Ca 2ϩ ] i , free CaM may be within that range in the cytoplasm (76,77). It is unclear what physiological condition would correspond to such a scarcity of Ca 2ϩ ions relative to CaM molecules, but under those conditions, we found the C-lobe of CaM to interact solely with the B domain of KCNQ4. Titration of Ca 2ϩ into the apoCaMϩQ4B complex measured by NMR suggests that Ca 2ϩ must reach a stoichiometry of 1:4 (EF-hands) to displace the B domain from the C-lobe. Taken together, our results can only be reconciled by a profound change in the configuration of the KCNQ4 RD when CaM is half-loaded by Ca 2ϩ (two of the four EF-hands bound by Ca 2ϩ ), under which the Ca 2ϩ -loaded N-lobe now strongly interacts with the B domain with very high affinity. This configuration seems to be particularly stable, as consistently found by all investigators.
Like many studies, we here ignore the significant concentration of free Mg 2ϩ ions, estimated to be ϳ0.5-1 mM in neurons, and given that the known affinities of Mg 2ϩ ions for the CaM EF-hands are within that range, predict significant occupancy of CaM EF-hands at tonic [Ca 2ϩ ] i (88 -90). Thus, it is likely that in resting cells at which free [Ca 2ϩ ] is low, some or all of the EF-hands of CaM are not empty but rather are occupied by Mg 2ϩ ions. The high-resolution crystal structure of the Mg 2ϩloaded N-lobe of CaM reveals Mg 2ϩ ions in both EF-hands, but in contrast with Ca 2ϩ occupancy, 2-4 H 2 O molecules are also included in each EF-hand (89), which could electrostatically shield much of the charge of divalent ions and perhaps alter interactions of the RD with the membrane. The occupancy by CaM by Mg 2ϩ ions when CaM is "off" was explicitly assumed for the solved CaM/Na V 1.2/FGFHF complex (91) and recently examined for CaM/KCNQ4 RD interactions (48). This raises the likelihood of "alkali earth-metal exchange" being part of the switching of CaM "on" or "off" in its functional interactions with the channels. This topic has also been explored for the case of synaptotagmin, whose structure when loaded with Mg 2ϩ (under which exocytosis is inhibited) has also been determined (92,93). In all of these cases, the structure of the CaM/target complex was suggested to be only subtly distinct between Mg 2ϩ -loaded and Ca 2ϩ -loaded forms. How these results can be in accord with the indisputable role of Ca 2ϩ as the "switch" for CaM actions on these proteins remains to be elucidated.
Based on our results, we propose a "lobe-switching model" in which Ca 2ϩ ions compete with the B domain for binding the C-lobe under low [Ca 2ϩ ], causing the B domain to dissociate from CaM in a stepwise CaM action on KCNQ4 channels (Fig.  10). The true "apo" configuration ( Fig. 10, 1) can only occur during extremely low (Ͻ10 nM) [Ca 2ϩ ], and we do not know whether and when such a low value occurs in the cytoplasm of living cells, which contain up to 1 mM free [Mg 2ϩ ]. Nonetheless, in this state, the C-lobe of apoCaM binds the B domain with modest affinity, and the N-lobe of CaM and the A domain of the channel are not involved. Because the RD is in a flexible, "open" configuration, we suppose it could be anchored to the plasma membrane via the two PIP 2 -binding sites located on both ends of the A domain (the "S6Jx" or "pre-A helix" site and the linker between the A and B domain). In this state, both the A and B domains are likely still disordered and nonhelical. It is unclear whether the channel can function in this fully "apoCaM" state. At local free [Ca 2ϩ ] near resting levels in excitable cells (50 -100 nM), the complex rapidly adopts a CaM displacement configuration, in which the B domain dissociates from the C-lobe and

Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels
CaM is reconfigured to a "target-ready" state that can recognize the A domain (Fig. 10, 2). This brief state then likely quickly transitions to the CaM rebinding configuration (Fig. 10, 3), as a result of Ca 2ϩ -loaded C-lobe (half-loaded CaM) binding to the A domain, imposing an ␣-helical structure on the A domain and condensing the proximal C terminus. In the first three of these configurations, the RD is relatively relaxed, allowing the C terminus to extend the two proposed PIP 2 -binding sites, the "S6Jx" or "pre-A helix," and the "A-B linker," to interact with PIP 2 at the plasma membrane-cytoplasm interface.
We take note, however, that our earlier work found exogenous expression of mutant C-lobe CaM that cannot bind Ca 2ϩ ions to not perturb Ca 2ϩ -mediated suppression of KCNQ2, KCNQ4, and KCNQ2/3, channels expressed in CHO cells (57). How can that be reconciled with our data and model here, which require Ca 2ϩ ions in the C-lobe in steps 1 and 3 in Fig. 10? We think the answer lies in the extremely high affinity of Ca 2ϩ ions for the C-lobe, and the presence of endogenous CaM in nearly all mammalian cells, with even more so in neurons. Thus, KCNQ channels already assembled in the Golgi (with the C termini in the cytoplasm) are prebound by endogenous CaM with Ca 2ϩ ions in the C-lobe and do not lose those ions. Overexpression of C-lobe mutant CaM is very much higher than expression of endogenous CaM. However, no inhibition of neuronal KCNQ channels can occur without the Ca 2ϩ /N lobe-dependent embrace of the A and B helices, which is what pulls the RD away from the plasma membrane, disrupting the obligatory interactions between PIP 2 and the two PIP 2 interaction sites in the proximal C terminus of the channels that are required for channel opening.
We hypothesize that upon a physiological signal that causes a substantial rise in free [Ca 2ϩ ] i in the vicinity of the "primed" CaM/RD complex, the triplex configuration occurs (Fig. 10, 4), as Ca 2ϩ binds the N-lobe of CaM, causing it to latch around the B domain. This induces a pulling/torsional Ca 2ϩ -induced motion, as suggested recently (50), likely disrupting the critical interactions between the RD and PIP 2 , causing inhibition of channel gating for neuronal M channels. We note that this "condensed" trimolecular configuration, as opposed to the "open" structure of free Ca 2ϩ -loaded CaM, is due to its intimate interaction with the A and B helices, with which its many bonds and sites of interaction provide the energetics for the resultant tight configuration of CaM. Thus, we suggest here that both the A and B helices and CaM rearrange their secondary and tertiary structures in response to each other, similar to the "mutually induced fit" already noted for CaM interactions with Ca 2ϩ / CaM kinases (66). This insight we believe to represent another advance of this paper.
Finally, with a rise in [Ca 2ϩ ] that is more than transient, this configuration is "locked," and the double anti-parallel helical conformation of the complex is stabilized, creating a compact geometry that severely disrupts PIP 2 interactions with KCNQ C termini that likely takes some minutes to reverse, as seen physiologically. Although Ca 2ϩ must be bound to the C-lobe first to displace CaM from the B domain, it is Ca 2ϩ loading of the N-lobe that creates this highly stable trimeric complex, which allows freedom within EF-hands III and IV of the C-lobe, rendering their ligation of Ca 2ϩ optional at this final step. Our lobe-switching model could provide the explanation for the dis-crepancies reported in the literature of how Ca 2ϩ directs CaM in its binding to, and regulation of, KCNQ channels, as it does incorporate some role of metal loading of the C-lobe as (an early) part of the mechanism (48) but retains Ca 2ϩ binding to the N-lobe as the modulatory switch, in accord with our physiological experiments in cells (57). It also is in accord with the need for CaM for functional expression (28,39,58). However, we do not believe that CaM interacts with the voltage-sensor domain of KCNQ2-5 channels and that as for PIP 2 actions on those channels (94), effects on voltage dependence are minimal. Future studies to test the affinities of the peptides with inactive N-or C-lobe CaM mutants and these domains will further probe whether this lobe-switching mechanism is indeed correct.

Buffers and protein preparation
Peptides corresponding to the A and B domains of KCNQ2-4 were synthesized to 95% purity (Peptide 2.0, Chantilly, VA). The lyophilized peptides were reconstituted in HBS buffer, which consists of 20 mM HEPES and 150 mM NaCl at pH 7.4, made with deionized water. We name each peptide according to the subunit isoform and the CaM binding domain: Q2A, Q2B, Q3A, Q3B, Q4A and Q4B, referring to the A or B domains of KCNQ2-4, respectively. A plasmid containing WT, untagged vertebrate CaM (plasmid pETGQ.HCaM) was a gift from William N. Zagotta (University of Washington). Vertebrate CaM was expressed in BL21 competent cells at 37°C for 6 h in LB or minimal essential medium containing [ 15 N]ammonium chloride, depending on the experimental design, and then purified using phenyl-Sepharose matrix (GE Healthcare). For experiments requiring higher resolution of NMR spectra we expressed in D 2 O (Millipore-Sigma). The eluent was further purified through a Superdex 75 column on an AKTA FPLC system (GE Healthcare). For divalent metal-free CaM (apoCaM) studies, the protein was exchanged to HBS buffer that had been soaked with Chelex reagent (Bio-Rad). We refer to this chelexed buffer as "ChHBS." Glassware and other containers used for apoCaM measurements were prerinsed with 10 mM EGTA, followed by 2ϫ ChHBS rinses, prior to the addition of proteins. The total [Ca 2ϩ ] in the ChHBS buffer was determined by inductively coupled plasma MS to be ϳ500 nM (Northwestern University, Evanston, IL). The peptides used for apoCaM interactions were reconstituted in ChHBS. The concentrations of the peptides and CaM were determined by amino acid analysis at the Texas A&M University Protein Chemistry Laboratory core (College Station, TX).

CD
Proteins were diluted to ϳ30 M in potassium phosphate buffer (KH 2 PO 4 ), pH 7.4, containing 5 mM NaCl and ϳ5-30 M Ca 2ϩ (determined by inductively coupled plasma MS and fluorescence spectroscopy using Ca 2ϩ indicator dyes) and placed in a 0.5-cm path length quartz cuvette. The molar ellipticity of each peptide was reported using a Jasco J-810 CD spectrometer at 4°C. CD spectra were recorded from 270 to 190 nm in 0.1-nm steps. The CD signals were corrected by subtracting the spectra from buffer only. The CD analysis plotting tool, CAPITO, was used to determine the predicted content of ␣-heli-Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels ces, ␤-sheets, and random coils of the indicated peptides and proteins (95).

ITC
ITC titrations were performed at 25 or 37°C using a VP-ITC microcalorimeter (MicroCal/Malvern Instruments). Titrations were conducted in HBS buffer supplemented with 1 mM EGTA or 0.5 mM CaCl 2 . Samples were degassed for at least 15 min. 5-10 M CaM was placed in the ITC cell, and 50 -200 M peptide was added to the titration syringe. Each ITC experiment consisted of at least 24 injections of 10 l of titrant, preceded by one 2-l injection, which is traditionally used to "prime" the system for ligand diffusion during the temperature equilibration between the syringe and cell contents (96). Data were analyzed with MicroCal Origin version 7.0, using the built-in curve-fitting models.

Crystallization, structure determination, and refinement
WT CaM purified in-house (see above) was mixed with Q4A in an equimolar ratio at room temperature, in HBS buffer supplemented with 2 mM CaCl 2 or 2 mM EGTA, and Q4B was added at an equimolar ratio 30 min later. The final concentration of each protein was 222 M. Automated screening for crystallization was carried out using the sitting drop vapor-diffusion method with an Art Robbins Instruments Phoenix system in the X-ray Crystallography Core Laboratory at UT Health San Antonio. Crystals of Ca 2ϩ -loaded complexes were initially obtained from Microlytic MCSG-III screen condition 60; optimized in 1.3 M sodium citrate, 0.1 M HEPES, pH 7.0; and flashcooled in liquid nitrogen prior to data collection. Data for two crystals were collected at the Advanced Photon Source NE-CAT beamline 24-ID-E and integrated and scaled together using XDS (97). The structure of Ca 2ϩ /CaM:Q4A:Q4B was determined by the molecular replacement method implemented in PHASER (98) using a truncated version of PDB entry 4UMO as the search model. Coordinates were refined using PHENIX (99), including simulated annealing with torsion angle dynamics and TLS refinement, alternated with manual rebuilding using COOT (100). The model was verified using composite omit map analysis (101,102) to minimize model bias. Data collection and refinement statistics are shown in Table S1. Renderings of the structures were performed using PyMOL software (PyMOL Molecular Graphics System, version 2.0, Schrödinger, LLC).

MST
apoCaM in ChHBS was conjugated with Alexa Fluor 594 (Alexa-CaM) using a protein-labeling kit (Invitrogen). Q4B or Q4A peptides were serially diluted 1:2 in ChHBS containing an additional 0.5 mM EGTA. The final concentration of Alexa-CaM in each well of the dilution series was 200 nM, and each sample was centrifuged prior to loading to standard capillaries. After capturing the proteins in 16 different capillaries, fluorescence was measured in a Nanotemper Monolith 1.115, using an excitation power of 60 and MST power of 80. Microscale thermophoresis was recorded using the MO control software, and the binding affinities were analyzed using PALMIST and GUSSI (Chad Brautigam, UT Southwestern, Dallas, TX) (103, 104).

HSQC-NMR
NMR titrations of [ 15 N]CaM with unlabeled Q4A and Q4B peptides were performed using HSQC (non-TROSY) experiments. All non-TROSY experiments were conducted in HBS, 1 mM EGTA, and 10% (v/v) D 2 O at 298 K on a Bruker Avance 700 NMR spectrometer. NMR titrations of deuterated [ 2 H-15 N]CaM with Ca 2ϩ were performed in ChHBS using TROSY-HSQC to reduce line broadening by cancelling dipole-dipole coupling and chemical shift anisotropy, thus producing wellresolved spectra. The methods were performed similarly to those described elsewhere (105,106).
The assignments for mammalian apoCaM were shared with us by John Putkey (UT Health, Houston, TX), and the assignments for Ca 2ϩ /CaM were kindly provided by Walter Chazin (Vanderbilt University, Nashville, TN) and Adriaan Bax (National Institutes of Health, Bethesdsa, MD). The raw spectrometer format data were processed using nmrPipe and nmrDraw (107). The peaks were calculated and visualized using SPARKY version 3.115 software (T. D. Goddard and D. G. Kneller, University of California, San Francisco), and in some cases, the overlays were formatted using Adobe Illustrator.
Estimation of equilibrium constants of apoCaM for the KCNQ4 B domain were evaluated by NMR titration data as follows. The peak heights of the titration spectra were normalized against the reference spectrum without titrant, followed by calculation of the mean and S.D. of the normalized peak heights after titration. Only those residues showing normalized peak heights at least 2 ϫ S.D. above the mean were plotted. Using these peak data, the K d values were determined using GraphPad Prism version 7, by fitting those data to the binding equation, y ϭ B max ⅐x/(K d ϩ x) ϩ NS⅐x ϩ background, in the one-sitetotal nonlinear model. Although plotting peak height (line broadening) is not a widely accepted or accurate method for determining affinities, we used this method as an approximation to guide us in further experimental approaches for testing binding affinities.

Perforated patch-clamp electrophysiology
CHO cells were grown in 100-mm tissue culture dishes (Falcon, Franklin Lakes, NJ) in Dulbecco's modified Eagle's medium with 10% heat-inactivated fetal bovine serum plus 0.1% penicillin/streptomycin in a humidified incubator at 37°C (5% CO 2 ) and passaged every 4 days. Cells were discarded after ϳ30 passages. CHO cells were first passaged onto 35-mm plastic tissue culture dishes and transfected 24 h later with FuGENE HD reagent (Promega), according to the manufacturer's instructions. The total amount of cDNA used was 0.55 g, which is less than that typical but required for such high-expressing channels like KCNQ4 (41), which otherwise might "soak-up" all of the PIP 2 in the membrane, allowing artifactual interactions between CaM and the channels that are unphysiological. The next day, cells were plated onto coverglass chips, and experiments were performed over the following 1-2 days. Pipettes were pulled from borosilicate glass capillaries (BF150-86-10HP; Sutter Instruments) using a Flaming/Brown micropipette puller P-97 (Sutter Instruments) and had resistances of 2-3 M⍀ when filled with internal solution and measured in Mutually induced fit of Ca 2؉ /CaM action on KCNQ4 K ؉ channels standard bath solution. The external Ringer's solution contained 160 mM NaCl, 5 mM KCl, 2 mM CaCl 2 , 1 mM MgCl 2 , and 10 mM HEPES, pH 7.4, with NaOH. The pipette solution contained 160 mM KCl, 5 mM MgCl 2 , and 10 mM HEPES, pH 7.4, with KOH with added amphotericin B.
Membrane current was measured with pipette and membrane capacitance cancellation, sampled at 5 ms, and filtered at 1 kHz using an EPC10 amplifier and Patchmaster software (HEKA). In all experiments, the perforated patch method of recording was used with amphotericin B (600 ng/ml) in the pipette solution (108). Amphotericin was prepared as a stock solution as 60 mg/ml in DMSO. In these experiments, the access resistance was typically 7-15 M⍀, 5-10 min after seal formation. Series resistance compensation was routinely compensated ϳ60%, and liquid junction potential corrections (Ͻ2 mV in this case) were not applied. Cells were placed in a 500-l perfusion chamber through which solution flowed at 1-2 ml/min. Inflow to the chamber was by gravity from several reservoirs, selectable by activation of solenoid valves (Warner Scientific). Bath solution exchange was essentially complete by Ͻ30 s. Experiments were performed at room temperature. Cells that displayed KCNQ4 currents over 1 nA at 0 mV were not studied, nor were experiments accepted if the uncorrected series resistance was Ͼ5 M⍀, due to undue sequestering of free [PIP 2 ] or to residual series resistance voltage errors. Currents were studied by holding the membrane potential at Ϫ80 mV and applying 500-ms depolarizing pulses from Ϫ80 to 40 mV, followed by a 400-ms step to Ϫ60 mV, every 3 s. To estimate voltage dependence, tail current amplitudes at Ϫ60 mV were fit to a single exponential starting at a time ϳ5-10 ms after the repolarization (when the residual capacity transient has subsided), and the amplitudes were normalized and plotted as a function of test potential. The data were fit with Boltzmann relations of the form, I/I max ϭ I max / (1 ϩ exp((V1 ⁄ 2 Ϫ V)/k)), where I max is the maximum tail current, V1 ⁄ 2 is the voltage that produces half-maximal activation of the conductance, and k is the slope factor. Values from cell populations were compared using a two-tailed t test. In such group comparisons, the use of S.E. values is most statistically correct. The activation kinetics were estimated using a weighted double-exponential fit (A⅐exp(Ϫ/t 1 ) ϩ B⅐exp(Ϫ/t 2 )) of the first 400 ms of the currents, where 1 and 2 are the fast and slow time constants and A and B are the "weights." Thus, the weighted deactivation time constants reported are given by (A⅐ 1 ϩ B⅐ 2 )/(A ϩ B). The deactivation time constants were measured by fitting the deactivating current to a single exponential at Ϫ60 mV, as described above.