The arrhythmogenic N53I variant subtly changes the structure and dynamics in the calmodulin N-terminal domain, altering its interaction with the cardiac ryanodine receptor

Mutations in the genes encoding the highly conserved Ca2+-sensing protein calmodulin (CaM) cause severe cardiac arrhythmias, including catecholaminergic polymorphic ventricular tachycardia or long QT syndrome and sudden cardiac death. Most of the identified arrhythmogenic mutations reside in the C-terminal domain of CaM and mostly affect Ca2+-coordinating residues. One exception is the catecholaminergic polymorphic ventricular tachycardia–causing N53I substitution, which resides in the N-terminal domain (N-domain). It does not affect Ca2+ coordination and has only a minor impact on binding affinity toward Ca2+ and on other biophysical properties. Nevertheless, the N53I substitution dramatically affects CaM's ability to reduce the open probability of the cardiac ryanodine receptor (RyR2) while having no effect on the regulation of the plasmalemmal voltage-gated Ca2+ channel, Cav1.2. To gain more insight into the molecular disease mechanism of this mutant, we used NMR to investigate the structures and dynamics of both apo- and Ca2+-bound CaM-N53I in solution. We also solved the crystal structures of WT and N53I CaM in complex with the primary calmodulin-binding domain (CaMBD2) from RyR2 at 1.84–2.13 Å resolutions. We found that all structures of the arrhythmogenic CaM-N53I variant are highly similar to those of WT CaM. However, we noted that the N53I substitution exposes an additional hydrophobic surface and that the intramolecular dynamics of the protein are significantly altered such that they destabilize the CaM N-domain. We conclude that the N53I-induced changes alter the interaction of the CaM N-domain with RyR2 and thereby likely cause the arrhythmogenic phenotype of this mutation.

Calmodulin (CaM) 6 is a ubiquitous protein that serves as an intracellular Ca 2ϩ sensor with more than 300 interaction partners (1). In its mature processed form, CaM consists of 148 amino acid residues and is composed of two globular domains (N-and C-domain) connected by a flexible linker region (2). Each domain contains two EF-hand motifs that can bind one Ca 2ϩ each, and the Ca 2ϩ affinity is higher for the C-domain than the N-domain, giving CaM a wide Ca 2ϩ -sensing range. This wide range is used to conduct the Ca 2ϩ signal in a variety of pathways because Ca 2ϩ -binding induces large conformational changes in CaM that alter its affinity for other protein targets (3,4). CaM is an extremely conserved protein: all vertebrate species have the exact same protein sequence (2,5). Selection pressure against CaM mutations is further underlined by the fact that humans have three different genes (CALM1-3) all coding the exact same CaM sequence (6).
In the cardiac context, CaM interacts with several ion channels (4,(7)(8)(9)(10). This includes the L-type voltage-gated calcium channel (Ca V 1.2), for which CaM serves as the Ca 2ϩ sensor mediating the Ca 2ϩ signal that accelerates the inactivation of the channel in a process termed Ca 2ϩ -dependent inactivation (CDI) (8,11). CaM also interacts with the cardiac ryanodine receptor (RyR2), which governs the release of Ca 2ϩ stored in the sarcoplasmic reticulum in cardiomyocytes. RyR2 is activated by Ca 2ϩ entering the myocyte through Ca V 1.2 or by a high sarcoplasmic reticulum luminal Ca 2ϩ concentration, while it is inhibited by CaM at both diastolic and systolic Ca 2ϩ concentrations (12,13).
Mutations in RyR2 and some of its auxiliary proteins have been linked to catecholaminergic polymorphic ventricular tachycardia (CPVT), a condition that can lead to ventricular arrhythmias and sudden cardiac death (14 -17). Recently, two individual CALM1 mutations were found in patients suffering from CPVT (N53I and N97S; CaM numbering throughout this article is according to the mature protein sequence without the initial Met). Subsequently, additional mutations in CaM-encoding genes have been reported. Some of these are linked to CPVT, some cause long QT syndrome, and some cause both. Others cause idiopathic ventricular fibrillation. Several other CaM variants have been identified through genome database searches but without being linked to any cardiac phenotype (18 -23).
Many CaM mutations reduce CDI of Ca V 1.2, inhibition of RyR2, or both. Several mutations within the C-domain lead to long QT syndrome and have been shown to either obliterate or decrease the extent of CDI. In stark contrast, CaM-N53I shows a negligible effect on CDI (22,24,25), but a strong effect on RyR2, where different studies show either no inhibitory effect or an activating effect on RyR2 at different cellular Ca 2ϩ concentrations. This is in sharp contrast to the RyR2 inhibition observed by CaM-WT at the same Ca 2ϩ concentration (26 -30).
However, the mechanistic details that explain how CaM mutations lead to cardiac arrhythmia are not always understood. It is puzzling that one mutated allele can cause a dominant phenotype in the presence of five healthy alleles. All of the reported arrhythmogenic mutations are observed in the CaM C-domain except for E45K, which was discovered very recently, and N53I (23). The Asn 53 residue is involved neither in Ca 2ϩ coordination nor in the packing of the hydrophobic core of the N-domain. It is situated in one of the helices, protruding out into the solvent (18 -20, 22). Previous structural investigations of CaM mutants have often shown large structural changes, but the structure of CaM-N53I in complex with the Ca V 1.2 IQ domain did not significantly differ from CaM-WT in complex with the Ca V 1.2 IQ domain (31,32). Although this result is expected, because the function of Ca V 1.2 is not perturbed by CaM-N53I (24), it does not explain the strong effect of N53I on RyR2 function, and the mechanism of action thus remains elusive.
Studies on CaM-N53I have shown how this variant causes increased RyR2 activity in murine ventricular myocytes, HEK293 cells, and single-channel RyR2 recordings (24,26,28,29). The mutation also increases the heart rate in ␤-adrenergic stimulated zebrafish, similar to the tachycardia phenotype observed in the human index case discovered 8 years ago (5,30). Although the effect of CaM-N53I in a cellular setting is evident, only small differences have been observed when looking at CaM-N53I biophysically. The Ca 2ϩ affinity of both CaM-N53I domains, along with the Ca 2ϩ affinity in the presence of a peptide representing the CaMbinding domain of RyR2 (CaMBD2), shows only small differences compared with CaM-WT (27,29,30). Based on the cellular effects described above, it seems likely that the N53I mutation interferes with the interaction between CaM and RyR2. However, a recent high-resolution cryo-EM structure of the CaM-WT in complex with full-length RyR2 does not support this hypothesis because amino acid Asn 53 seems not to interact with RyR2 under any of the investigated conditions (33).
To investigate the structural effect of the N53I mutation, we determined the solution structures and dynamics of CaM-N53I both in the apo-and Ca 2ϩ -loaded state by NMR, along with solving the crystal structures of both Ca 2ϩ /CaM-WT and -N53I in complex with CaMBD2 of RyR2. The affinity of CaM-N53I for RyR2-CaMBD2 was also determined. The overall goal of this work is to find the differences that can explain the arrhythmogenic nature of CaM-N53I.

Results
Throughout this paper, the CaM numbering is for mature CaM, meaning that the initial methionine is excluded. Both apoCaM and Ca 2ϩ /CaM possess eight ␣-helices, which will be named A-H, of which A-D are in the N-domain, whereas E-H are in the C-domain. Structures of other CaM variants have already been solved in which it has been shown that a mutation can give rise to a structural change (31). We therefore started out by solving the structures of both apoCaM-N53I and Ca 2ϩ / CaM-N53I by NMR to investigate for structural differences compared with CaM-WT.

NMR structure of apoCaM-N53I
The two CaM domains are connected by a flexible linker that causes them to sample a lot of different spatial conformations, making a simultaneous alignment of both domains in an ensemble impossible. Therefore, an ensemble of each domain of apoCaM-N53I was compared with its corresponding domain from apoCaM-WT (PDB code 1CFD) (34) (Fig. 1, A and B). In the N-domain (Fig. 1A), differences are observed in the Ca 2ϩbinding loops in both EF-hands but also in the flexible loop connecting the two EF-hands between helix B and C. No structural changes are observed in helix C harboring Ile 53 , which protrudes into the solvent like Asn 53 in the WT. The variation in the 20 apoCaM-N53I structures is largest in the aforementioned loop regions. Comparing the apoCaM-N53I and apoCaM-WT C-domains (Fig. 1B), similar loop flexibility is observed with both interhelical loops and the EF-handconnecting loop being flexible. Overall, the apoCaM-N53I C-domain is less well-defined than the N-domain because of a lower density of NOEs in the C-domain, but the same observation holds true for apoCaM-WT (35).
Because of the structural similarity between apoCaM-WT and apoCaM-N53I, we analyzed chemical shift differences in the [ 1 H-15 N]-HSQCs to see which other residues were affected by the mutation (Fig. S1). As expected, the residues near the mutated Ile 53 residue are affected because the chemical environment changes by substituting an amino acid. Notably, the two Ca 2ϩ -coordinating residues Asn 60 and Thr 62 are also affected, even though they are 10 -12 Å away from Ile 53 .

Effect of calmodulin mutation N53I NMR structure of Ca 2؉ /CaM-N53I
The ensemble of Ca 2ϩ /CaM-N53I structures is compared with the first structure of the Ca 2ϩ /CaM-WT ensemble (PDB code 1X02) (36) (Fig. 2, A and B). Compared with apoCaM, a higher degree of similarity is observed in the Ca 2ϩ /CaM comparison because the difference in the flexible loop regions are smaller. Ile 53 is located right at the end of helix C in Ca 2ϩ /CaM-N53I, and no major changes are present compared with Ca 2ϩ / CaM-WT ( Fig. 2A). The side chain of residue 53 is not within the hydrophobic binding pocket that is exposed when CaM binds Ca 2ϩ , so it is not at the typical interface whereby CaM binds amphipathic helices.
Because of the high degree of similarity between the two Ca 2ϩ -bound structures, we investigated chemical shift differences in the [ 1 H-15 N]HSQC spectra of Ca 2ϩ /CaM-N53I and Ca 2ϩ /CaM-WT (Fig. S2). Not surprisingly, the differences are largest in residues 49 -57 on ␣-helix C near the mutation site. However, as opposed to the apo structure, minor changes are also observed in residues 28 -33 on ␣-helix B, which is part of the first EF-hand in the N-domain). In summary, small structural changes are observed both for the apoCaM-and Ca 2ϩ / CaM-N53I structures compared with their respective WT structures, along with some chemical shift changes propagating to residues not in the immediate vicinity of Ile 53 .

Crystal structure of Ca 2؉ /CaM-WT bound to CaMBD2 of RyR2
Because the CaM-N53I variant is believed to cause dysregulation of RyR2 in arrhythmia patients, we investigated the CaM-RyR2 interaction. CaM interacts with RyR2 through CaMBD2 encompassing residues 3581-3611, and a deletion of this domain abolishes CaM binding to RyR2 (37). Because the resolution of Ca 2ϩ /CaM bound to full-length RyR2 is limited, with many side chains in CaM unresolved, we first solved the crystal structure CaMBD2 in complex with Ca 2ϩ /CaM-WT at 1.8 Å resolution. Three residues differ between CaMBD2 from RyR1 and RyR2 (RyR1/RyR2: Lys 3614 /Arg 3581 , Arg 3629 /Lys 3596 , and Thr 3639 /Ala 3606 ). CaM exerts different actions on these two RyR isoforms, where RyR1 is activated and inhibited in the presence of low and high Ca 2ϩ concentration, respectively, whereas RyR2 is inhibited during both low and high Ca 2ϩ concentration (12,37). A comparison of Ca 2ϩ /CaM-WT bound to the RyR1 and RyR2 CaMBD2 indicates that CaM-WT binds similarly to both RyR CaMBD2 isoforms with the characteristic 1-17 spacing between the Trp and Phe hydrophobic anchor residues in the CaMBD2s, shown by a root-mean-square deviation (RMSD) of 0.259 Å (Fig. 3A). Of the three residues that differ between the two CaMBD2s, only Lys 3596 is resolved in the complex with CaMBD2 from RyR2.
Arg 3629 in RyR1 interacts with both Leu 18 and Ser 38 from CaM through two hydrogen bonds and through a water-mediated hydrogen bond, respectively. The Lys 3596 lysyl group in RyR2 is smaller than the guanidino group in Arg 3629 , thereby increasing the distance to Leu 18 and Ser 38 from CaM. Lys 3596 is also angled slightly more toward the peptide backbone than Arg 3629 , increasing the distance to CaM residues Leu 18 and Ser 38 even further (Fig. 3B). The two unresolved residues Arg 3581 and Ala 3606 in the RyR2 complex correspond to Lys 3614 and Thr 3639 in the RyR1 complex, respectively. In the RyR1 complex, Lys 3614 does not make any interactions, but Thr 3639 forms hydrogen bonds from its side chain to the carbonyl groups of Cys 3635 and Phe 3636 . This interaction is not possible for the Ala 3606 side chain in the RyR2 complex and could lead to instability and an altered structure. However, the structures are still highly similar in this C-terminal part of the CaMBD2s. Despite these differences in binding, we previously found the affinities of Ca 2ϩ /CaM for the RyR1 and RyR2 CaMBD2 segments to be very similar, with a K d ϳ45 nM (38).

Crystal structure of Ca 2؉ /CaM-N53I bound to CaMBD2 of RyR2
We next solved the structure of the Ca 2ϩ /CaM-N53I in complex with CaMBD2 from RyR2 to a resolution of 1.9 Å. When the structure of Ca 2ϩ /CaM-WT and Ca 2ϩ /CaM-N53I bound to CaMBD2 from RyR2 are superposed, subtle conformational changes are observed in the vicinity of residue 53 (Fig. 4A). When zooming in on the area harboring EF-hand 2, subtle changes in both the position of the Ca 2ϩ and the Ca 2ϩ -coordi-

Effect of calmodulin mutation N53I
nating residues are seen (Fig. 4B). The conformation of the two CaM-N53I domains is also highly similar to the CaM-WT domains, with an RMSD of 0.384 Å. For the individual domains, RMSD values of 0.221 Å for the N-domain (residue 4 -73) and 0.190 Å for the C-domain (residue 85-145) are obtained. In the crystal structure of Ca 2ϩ /CaM-WT bound to CaMBD2 from RyR1 (PDB code 2BCX), Asn 53 forms hydrogen bonds with CaM residues Gln 49 and Asp 56 , the latter through a water molecule. We therefore investigated the hydrogen bond network in helix C in both the Ca 2ϩ /CaM-WT and -N53I:RyR2-CaMBD2 crystal structures (Fig. 4, C and D). In the Ca 2ϩ /CaM-WT: RyR2-CaMBD2 complex, Asn 53 O ␦1 forms a hydrogen bond with Gln 49 H ⑀2 , but the water-mediated hydrogen bond to Asp 56 is absent (Fig. 4C). For the Ca 2ϩ /CaM-N53I:RyR2-CaMBD2 complex, the side chain of Gln 49 is not resolved, probably because of the lost hydrogen bond because the side chain of Ile 53 cannot participate in a hydrogen bond. The chemical shift of Gln 49 also changes strongly upon mutation of Asn 53 in the  Labelled are the two aromatic anchor residues Trp 3620 /Phe 3636 and Trp 3587 /Phe 3603 in CaMBD2 for RyR1 and RyR2, respectively. The green circle highlights the only amino acid substitution (Arg 3629 /Lys 3596 ) that is resolved in the RyR2 CaMBD2 structure. B, close-up of the area containing the Arg 3629 /Leu 3629 difference between the two CaMBD2s. Distances mark bond length where it is seen how Arg 3629 from CaMBD2 in RyR1 is capable of making hydrogen bonds with both Leu 18 and through water Ser 38 in CaM, whereas the distance is too long for hydrogen bonds from Lys 3629 from RyR2 CaMBD2.

Effect of calmodulin mutation N53I
absence of CaMBD2 (Fig. S9). It is, however, not possible to conclude whether this is due to a disruption of its H-bond with Asn 53 or merely due to a change in the chemical environment. Another hydrogen bond, formed between the backbone atoms of Asp 50 and Glu 54 in the WT structure, is absent in the mutant Ile 53 structure because the distance between the two atoms has increased from 2.19 to 2.48 Å (Fig. 4D).
Asn 53 is found on the solvent-exposed side of CaM, where it interacts neither with Ca 2ϩ nor with CaMBD2. When looking at a surface hydrophobicity plot of the two structures, it is clear that the N53I substitution turns the surface from polar to hydrophobic (Fig. 5). The solvent-accessible area for residues Asn 53 and Ile 53 is 89 and 190 Å 2 , respectively. This exposed hydrophobic area may promote additional interactions with RyR2 that are not observed for WT CaM.

Ensemble refinement of CaM-WT and -N53I bound to CaMBD2 of RyR2
We performed an ensemble refinement (40, 41) on the two crystal structures of the Ca 2ϩ /CaM-WT:RyR2-CaMBD2 and Ca 2ϩ /CaM-N53I:RyR2-CaMBD2 complexes to investigate whether the N53I mutation could lead to instability of helix C in CaM (Fig. 6). The ensemble shows that the linker connecting the two domains is highly flexible together with the loop regions between helix B-C and F-G and the N and C termini in both CaM-WT and CaM-N53I (Fig. 6, A-C). This is in agreement with { 1 H}-15 N-NOE values recorded by NMR on Ca 2ϩ / CaM-WT and -N53I, although these data are recorded in the absence of CaMBD2 (Fig. S6). Interestingly, a detailed investigation of the stability of helix C in CaM-N53I reveals an extended distance between Asp 50 O and Glu 54 N on the backbone, compared with CaM-WT (Fig. 6D). This result correlates with the missing hydrogen bond illustrated in Fig. 4 (C and D).

Effect of CaM-N53I on binding to CaMBD2 of RyR2
We have previously reported the affinity of CaM-WT and several other CaM mutants toward the RyR2 CaMBD2 under both apo and Ca 2ϩ -saturating conditions, using isothermal titration calorimetry (ITC) (28). These data show significant changes in CaMBD2 affinity for some of the mutants. By the same procedure, we determined the affinity of both apo-and
The binding of apoCaM to CaMBD2 is entropy-driven, whereas the binding of Ca 2ϩ /CaM to CaMBD2 is enthalpydriven for both CaM-WT and -N53I, in agreement with previous results for CaM-WT (28,38). For CaM-N53I, the affinities of the apo-and Ca 2ϩ interaction with the RyR2 CaMBD2 are low (28.7 M) and high (0.015 M), respectively, showing only minor differences when compared with CaM-WT binding to the RyR2 CaMBD2 (apo 27.8 M, Ca 2ϩ 0.012 M) ( Table S4).

Effect of the N53I mutation on CaM dynamics
In addition to the missing hydrogen bonds in the Ca 2ϩ /CaM-N53I:RyR2-CaMBD2 crystal structure, another significant difference caused by the N53I mutation is an increased Ca 2ϩ dissociation rate from the N-domain in CaM-N53I, reported previously both in the absence and presence of RyR2-CaMBD2 (30,42). We therefore investigated the dynamics of both CaM-WT and CaM-N53I in the absence and presence of Ca 2ϩ by NMR. In the apo form, no significant differences in the { 1 H}-15 N-NOE spectra can be observed between CaM-WT and -N53I (Fig. S3). Comparing the { 1 H}-15 N-NOE spectra for Ca 2ϩ /CaM-WT and Ca 2ϩ /CaM-N53I, a mobility difference is observed in amino acids 29 -31 and 47-51 situated on ␣-helices B and C, respectively ( Fig. 8 and Fig. S6). This result correlates well with the chemical shift perturbations observed between the Ca 2ϩ -bound forms (Fig. S2).
We then recorded 15 N-T 1 and 15 N-T 2 relaxation for both CaM-WT and -N53I in the absence and presence of Ca 2ϩ , but there were no significant differences in those parameters between CaM-WT and -N53I (Fig. S4, S5, S7, and S8). Because these experiments only probe the motions on the picosecond to nanosecond time scale, hydrogen deuterium exchange (HDX) experiments were performed to examine the slow hydrogen backbone amide exchange, which is on the time scale of 10 3 -10 7 s. For both apoCaMs, the exchange took place too fast to be observable, but for Ca 2ϩ /CaM-WT and Ca 2ϩ /CaM-N53I, HDX was observable for residues mainly found in the core of each CaM domain (Fig. 9). HDX could be observed for many of the same residues in the two proteins, but the exchange rates of To investigate intramolecular mobility on the millisecond time scale, T 2 -relaxation dispersion was measured for both CaM-WT and -N53I. A significant difference was observed here. In apoCaM-WT, conformational exchange rates (k ex ) for backbone nitrogen atoms in the N-domain are much lower than those in the C-domain (Fig. 10A), indicating a less stable fold of the C-domain, a well-known characteristic of CaM (44). Strikingly, in apoCaM-N53I, k ex of backbone nitrogen atoms in the N-domain are similar to the residues in the C-domain (Fig.  10B). This points to a destabilization of the N-domain of apoCaM-N53I. The relaxation dispersion of both Ile 52 and Asn 53 /Ile 53 is shown in Fig. 10 (C and D). For Ca 2ϩ /CaM-WT and Ca 2ϩ /CaM-N53I, a completely different result is obtained because nearly all residues in both variants have either a much lower or a nonmeasurable k ex (Fig. 10, E and F) indicating more  conditions. Isotherms were fit to a one-site binding model. DP is the measured power difference between the reference and the sample cell to maintain a zero temperature between the cells.

Effect of calmodulin mutation N53I
stable domains. This higher stability of Ca 2ϩ /CaM compared with apoCaM has already previously been reported (45) and is corroborated by the results presented here.

Discussion
Mutations in the highly conserved Ca 2ϩ -sensing protein CaM can lead to cardiac arrhythmia and cardiac death. The first human CaM mutation, CaM-N53I, was identified by genetic linkage in a large family with dominantly inherited CPVT. Importantly, the large sarcoplasmic Ca 2ϩ -release channel RyR2 is the only other protein in which mutations are dominantly linked to the same phenotypical expression, and CaM is an important regulator of RyR2 gating properties. Studies have demonstrated an effect of CaM-N53I on zebrafish heart rhythm and intracellular Ca 2ϩ -release events in ventricular myocytes (25,30). Also, the regulation of RyR2 by CaM-N53I was shown to be aberrant because CaM-N53I displays an activating effect on the channel compared with CaM-WT, which shows an inhibitory effect (26,27). There are four putative regions of CaM interaction with RyR2, of which CaMBD2 is required for CaM-regulation (37,38,(46)(47)(48). Even though cryo-EM structures of CaM-WT bound to full-length RyR2 are now available, these structures suffer from poor local resolution around CaM, and a mechanistic explanation for the disease mechanism underlying the N53I variant is still lacking.
So how can our results help to understand the arrhythmogenic phenotype? First of all, our data show that the overall structure of CaM-N53I remains unaltered in the apo, the Ca 2ϩ , and the Ca 2ϩ ϩ CaMBD2-bound state. However, we discern two major differences that likely underlie the disease mechanism. First, the N53I mutation leads to an exposed large hydrophobic residue, which directly could drive interactions between the CaM N-domain and RyR2 that otherwise do not occur. Second, the mutation leads to a destabilization of the N-domain, resulting in an increase in the internal dynamics of the protein. In the apo-state, 15 N-T 2 relaxation occurs faster around the mutation site and around amino acids 27-31 (Fig.  S5). Also, the heteronuclear { 1 H}-15 N-NOE, a parameter sensitive to intramolecular mobility on the picosecond to nanosecond time scale, of amino acids in this region of the protein drops  . Hydrogen deuterium exchange data for Ca 2؉ /CaM-WT (black) and Ca 2؉ /CaM-N53I (red). Protection factors p ϭ k rc /k ex are determined by comparing measured hydrogen-deuterium exchange rates, k ex , with the rate in a random coil model in which the residues are completely unprotected, k rc (43). A larger value indicates a slower exchange. Some amino acids are missing a value because of the exchange rate being too fast to measure with HDX.

Effect of calmodulin mutation N53I
slightly (Fig. S3), hinting at slightly increased mobility. Most significantly, T 2 -relaxation dispersion of the CaM-N53I N-domain changes dramatically: although the CaM-WT N-domain displays almost no relaxation dispersion, the N53I mutation introduces T 2 -relaxation dispersion throughout the whole N-domain (Fig. 10, A and B). This means that the N-domain of apoCaM-N53I undergoes conformational exchange on the millisecond time scale contrary to the N-domain of apoCaM-WT, in which no conformational exchange could be detected. Taken together, the data suggest that there is a higher degree of relatively slow internal motions in the N-domain of CaM-N53I than in CaM-WT. This can be viewed as a constantly ongoing, slow conformational exchange between two or more states. There could be a single excited state, which we do not know much about, populated for a fraction of time, or there could be a more general destabilization of the fold of the N-domain. This is also reflected in the previously reported lower melting temperature for the N-domain of CaM-N53I compared with CaM-WT (30).
In the Ca 2ϩ -bound form, mobility changes are not as pronounced: the heteronuclear { 1 H}-15 N-NOE values around

Effect of calmodulin mutation N53I
amino acid 30 and throughout helix C are slightly decreased in CaM-N53I, but also other regions of the mutant display small changes ( Fig. 8 and Fig. S6). The 15 N-T 1 values (Fig. S7) of Ca 2ϩ /CaM-N53I are generally lower and 15 N-T 2 values (Fig.  S8) are generally higher than in the WT. This points to an increased overall mobility of the mutant. Native PAGE experiments reported earlier (30), however, do not support a significant change in hydrodynamic radius of CaM-N53I in solution. Neither Ca 2ϩ /CaM-N53I nor Ca 2ϩ /CaM-WT display significant T 2 -relaxation dispersion, probably because of Ca 2ϩ binding which yields a strong contribution to the stability of the fold in both the N-and C-domains. Only few hydrogen-deuterium exchange rates could be obtained. They are generally slightly faster for Ca 2ϩ /CaM-N53I than for Ca 2ϩ /CaM-WT, also in helices B and C, with the notable exception of Ile 52 , whose H N exchanges much faster in the N53I mutant, and the mutated amino acid 53, whose H N exchanges much slower in the mutated protein (Fig. 9), suggesting an altered H-bonding network in the vicinity of the mutation site. Despite the absence of larger, detectable mobility changes, the N53I mutation was shown to also destabilize the N-domain of Ca 2ϩ /CaM-N53I by 4.7 kJ/mol, relative to Ca 2ϩ /CaM-WT (30). Chemical shift perturbation ( Fig. S1 and Fig. S2) occurs not only around the mutation site, but also throughout helix C and also in parts of helix B, pointing at a change in the time-averaged chemical environment of the affected amino acid backbone atoms H N and N.
What might be the reason for the drastic destabilization of the N-domain in N53I? The X-ray structure of the Ca 2ϩ /CaM-N53I:RyR2-CaMBD2 complex provides a clue (Fig. 4, C and D): the side chains of Gln 49 (Fig. S9).
It is difficult to explain how a mutation on amino acid Asn 53 , which does not directly take part in Ca 2ϩ binding and is remote to the CaMBD2-binding site of CaM (Figs. 4A and 5), could possibly influence the interaction of CaM with RyR2. Therefore, additional interaction sites in RyR2 apart from CaMBD2 were suggested (37,38,(46)(47)(48) to hold the key to the dysregulation of RyR2 by CaM-N53I. Indeed, recent cryo-EM structures of the RyR2-CaM complex (33) revealed that the N-domain of CaM has additional interaction sites with RyR2, both in the apo and in the Ca 2ϩ -bound forms. In the apo-form (PDB code 6JI8), helix C of CaM is close to amino acid Lys 2558 of RyR2, but because of unresolved amino acids in this region of RyR2, the exact interaction site with CaM helix C cannot be determined. Also, other parts of the N-domain show interactions with RyR2. Importantly, CaM EF-hand 2, a region where we can see chemical shift changes introduced by the N53I mutation, is not resolved in the cryo-EM structure but appears close enough to the RyR2 surface to make an interaction with this region possible. In the Ca 2ϩ -bound form of the RyR2-CaM complex (PDB code 6JV2), the side-chain carboxylate of Asp 50 , which is located very close to amino acid Asn 53 , is only 2.4 Å away from the guanidine group of Arg 2209 in RyR2 (Fig. 11B). Superposition of our crystal structure of Ca 2ϩ /CaM-N53I:RyR2-CaMBD2 onto the cryo-EM structure (PDB code 6JV2) reveals slight differences especially in the Ca 2ϩ -binding EF-hands of CaM, and an overall RMSD of 1.83 Å is obtained for superposition on N, C ␣ , and CЈ over all helix residues (Fig. 11A).
These secondary interaction sites revealed by the cryo-EM structures allow us to propose that the destabilization of helix C in CaM-N53I disturbs the interaction of CaM with RyR2 not by altering the interaction with CaMBD2 but by altering the secondary interactions made by the N-domain of CaM. We show that the N53I mutation does not alter the affinity of CaM for CaMBD2, neither in the apo nor in the Ca 2ϩ -bound state (Fig. 7 and Table S4). However, even though the N53I mutation has only a small impact on the Ca 2ϩ affinity of CaM alone (30), a reduced Ca 2ϩ -affinity for the CaM-N53I:CaMBD2 complex was reported (29), because of an increase in the dissociation rate of Ca 2ϩ from the complex (42). Single-channel ion conductivity experiments demonstrated that CaM-N53I is not as efficient in closing RyR2 as CaM-WT (28). However, our results

Effect of calmodulin mutation N53I
only show minute changes in protein structure and moderate changes in its intramolecular dynamics.
This demonstrates how finely tuned the regulation of the heartbeat is. Its correct regulation depends critically on very precise affinity and kinetics of the tripartite RyR2-CaM-Ca 2ϩ interaction. It has already been established that RyR2 mutations can lead to spontaneous Ca 2ϩ release and CPVT and that mutations in some RyR2-interacting proteins can lead to CPVT (49). Ca 2ϩ binding to the N-domain of CaM is vital for RyR2 regulation, and it is believed that N-domain binding to CaMBD2, which only occurs at systolic Ca 2ϩ concentrations, is at the heart of the feedback loop that closes RyR2 at high cytosolic Ca 2ϩ . CaM-N53I, with its higher Ca 2ϩ dissociation rate, is not be able to keep RyR2 closed tightly enough (28), leading to a leaky RyR2 and consequently CPVT.
Ca 2ϩ binding to the N-domain of the apoCaM:RyR2-CaMBD2 complex is accompanied by a large conformational rearrangement, both of the N-domain and of the interaction epitopes between CaM and RyR2. Any stabilization or destabilization of those interaction epitopes, both in the apo and in the Ca 2ϩ -bound states, will have an influence on the apparent Ca 2ϩ affinity. We here demonstrate that the interaction epitopes of the CaM N-domain are affected by the N53I mutation in both states. It is worth bearing in mind that the CaM-N53I variant represents the largest known cohort of people living with an arrhythmogenic CaM variant because the gene has been carried over for generations, whereas most of the other arrhythmogenic CaM variants are de novo mutations. This marks the CaM-N53I variant as less severe compared with the other CaM variants. In light of that, it is not surprising that we observe only small changes in the protein structure. In conclusion, we have shown that the N53I mutation does not alter the overall structure of CaM, but the mutation does change the internal dynamic behavior and stability of the N-domain, and this disturbance in the N-domain is likely to cause the observed phenotype.

Protein expression and purification for NMR
For uniformly 13 C-and 15 N-labeled CaM-N53I, Escherichia coli Rosetta-2 (DE3) cells (Novagen) were transformed with a plasmid from a previous study (30) containing the gene for the CaM-N53I variant. Cells were first grown in LB medium containing no isotope labeled compounds. At A 600 ϳ0.5, the medium was centrifuged, the supernatant was discarded, and a portion of the pelleted cells was resuspended in standard M9 minimal medium containing 15 N-ammonium sulfate and [U-13 C] D-glucose to an A 600 of ϳ0.05. At an A 600 of ϳ0.5, the expression was induced by 1 mM isopropyl ␤-D-1-thiogalactopyranoside and incubated for 18 h at 25°C. The cells were harvested by centrifugation and resuspended in 5 ml of lysis buffer (20 mM Tris-HCl, 50 mM NaCl, 2 mM MgSO 4 , and 1 mM ␤-mercaptoethanol (BME) at pH 7.5) per g of cell mass. The cells were lysed by three cycles of freeze-thawing followed by sonication. The lysate was centrifuged at 30,000 ϫ g at 4°C for 45 min, and the supernatant was applied to a 120-ml amylose column (New England Biolabs), washed with 5 column volumes (CV) of buffer A (20 mM Tris-HCl, 200 mM NaCl, pH 7.4), and eluted with 100% (v/v) buffer B (buffer A and 10 mM maltose). The fusion protein was cleaved with recombinant tobacco etch virus (TEV) protease overnight at 4°C. The sample was then applied to a 50-ml Q-Sepharose anion-exchange column equilibrated in buffer C (20 mM Tris-HCl, 50 mM NaCl, pH 7.5), washed with 3 CV of 30% (v/v) buffer D (20 mM Tris-HCl, 500 mM NaCl, pH 7.5), and then eluted with a linear gradient from 30 to 100% (v/v) buffer D. The eluted protein was concentrated to 2 ml. The sample was applied to a HiLoad 16/60 Superdex200 column (GE Healthcare) using buffer E (2 mM HEPES, 100 mM KCl, pH 6.5). The molecular weight of the purified protein was verified by MALDI-TOF MS, and concentration was determined by absorption at 280 nm.
The chemical shift assignments for 1 H, 13 C, and 15 N of the stereo array isotope-labeled Ca 2ϩ /CaM -WT (52) were used as a template for the assignment of Ca 2ϩ /CaM-N53I, whereas the chemical shift assignment for 1 H, 13 C, and 15 N of apoCaM-WT 7 was used as template for apoCaM-N53I. Both assignments were performed manually using the software CARA 1.8.4.2 (53) with their respective templates used. The NEASY module was used to collect and integrate NOESY cross-peaks. Based on the chemical shift assignments from H N , H ␣ , C ␣ , CO, and N, TALOSϩ (54) was used for a preliminary empirically prediction of the protein backbone dihedral angles and . The backbone dihedral angles from TALOSϩ and the integrated and assigned NOESY peak lists were used to calculate an ensemble of 20 structures with CYANA 3.0 (55-58). YASARA 14.12.2 (59) was used to perform an energy refinement both in vacuo and water on the ensemble of 20 model structures, with the NOVA (59) and YASARA (60) force fields, respectively. Finally, the two ensembles were validated using the protein structure validation software suite (61) to determine the quality of the structures. A summary of distance restraints and the details from the CYANA calculations can be seen in Table S1.

Cloning, expression, and purification for crystallography
The pEGST_CaM was coexpressed with CaMBD2 in a modified pET28a vector containing in tandem, a His tag, maltosebinding protein, and a cleavage site for the TEV protease in front of CaMBD2 (62) in E. coli Rosetta (DE3) pLacI (Novagen) at 37°C. The expression was induced at A 600 ϳ0.6 with 0.4 mM isopropyl ␤-D-1-thiogalactopyranoside for 4 h. The cells were lysed by sonication in buffer A (250 mM KCl, 10 mM HEPES, pH 7.4, 10 mM CaCl 2 ) and supplemented with 2 ml of glycerol, 25 mg ml Ϫ1 lysozyme, 25 mg ml Ϫ1 DNase, and 1 mM phenylmethanesulfonyl fluoride. The lysate was applied to a 25-ml PorosMC (Tosoh Biosep) column, washed with 5 CV of buffer A followed by 5 CV of 30% buffer B (250 mM KCl, 500 mM imidazole). Protein was eluted with 5 CV of 100% buffer B. The eluate was 2-fold diluted with buffer A supplemented with 10 mM BME, applied to a 40-ml amylose column, and washed with 3.5 CV buffer A containing 10 mM BME. The protein was eluted with 3.5 CV of buffer C (buffer A with 10 mM maltose and 10 mM BME). The sample was dialyzed against buffer A overnight at 4°C and simultaneously cleaved with recombinant TEV protease.
The cleaved sample was 2-fold diluted in buffer A and applied to a PorosMC column collecting the flow-through along with 2 CV of buffer A wash. The protein was concentrated and supplemented with 10 mM BME and applied to a Superdex200 column (GE Healthcare) in buffer A containing 10 mM BME. To remove remaining fusion protein, the sample was 2-fold diluted in buffer A and applied onto a 5-ml gravity Co 2ϩ -affinity column, collecting flow-through and 1 CV of buffer A wash. The sample was then concentrated in a spin filter, while the buffer was changed into a crystallization buffer containing (25 mM KCl, 10 mM HEPES, pH 7.4, 2 mM CaCl 2 , 5 mM DTT). Protein crystals were grown by the hanging-drop method at room temperature. The Ca 2ϩ /CaM-WT:RyR2-CaMBD2 and Ca 2ϩ /CaM-N53I:RyR2-CaMBD2 complexes were crystallized in 0.1 M sodium acetate, pH 4.70, and 23% PEG 550 monomethyl ether, with Ca 2ϩ /CaM-N53I:RyR2-CaMBD2 crystals used for seeding of Ca 2ϩ /CaM-WT:RyR2-CaMBD2.

X-ray data collection and structure determination
Crystals were harvested and flash-frozen in solutions containing the original growth conditions supplemented with 20% PEG 400. Diffraction data were collected at the Canadian Macromolecular Crystallography Facility 08B1-1 and the Advanced Photon Source at Argonne National Laboratory. The data were processed with XDS (63) and scaled with Aimless (CCP4 package). The structures of both RyR2 CaMBD2 complexes were solved by molecular replacement with Phaser (Phenix package) (64) using the 2.0 Å crystal structure of the Ca 2ϩ /CaM-WT: RyR1-CaMBD2 complex as a model (Lys 3614 -Asn 3643 , RyR1 CaMBD2) (PDB code 2BCX) (65). The model was subsequently manually built with COOT (66) and refined with phenix.refine (67). The final maps contain electron density for residues 3582-3605 of CaMBD2 from RyR2, residues 4 -76 and 81-145 for CaM-WT, and residues 4 -77 and 82-146 for CaM-N53I, as well as four well-ordered Ca 2ϩ . Because the linker region of CaM and CaMBD2 N and C termini are poorly represented in the electron density maps, these residues are omitted from the final model of both complexes. The statistics are shown in Table S2. Structures were deposited in the PDB with codes 6Y4O and 6Y4P.

Time-averaged X-ray restrained ensemble refinement
The two crystal structures solved here for Ca 2ϩ /CaM-WT: RyR2-CaMBD2 (PDB code 6Y4O) and Ca 2ϩ /CaM-N53I: RyR2-CaMBD2 (PDB code 6Y4P) were used for Ensemble refinement. A time-averaged electron density restrained molecular dynamics simulation was applied to generate a grouping or ensemble of structures that describe the structural dynamics inherent in the electron density in a quantifiable manner. Manual addition of residues not found in the density, addition of poorly resolved side chains, and removal of conformers were performed using COOT, and the final PDB code files were subjected to Phenix.ready_set to generate ligand restraints and add explicit hydrogen atoms. Ensemble refinement was performed using the implementation in Phenix version 1.10.1-2155 (67)

Effect of calmodulin mutation N53I
was performed for optimal values of pTLS (0.6 -0.9), Tbath (2.5, 5, and 10), and Tx (0.3, 0.6, and 1.2ϫ the resolution-dependent value) scoring for the lowest final R free . Selected optimal parameters were pTLS of 0.8, T_bath of 2.5, and Tx of 0.6 for both structures with average R free /R work of 0.1639/0.2234 for the WT and 0.1684/0.2509 for the N53I structure. These parameters were used to perform five random seed repeats for each structure. Root-mean-square fluctuation values for the five repeated ensembles were calculated using PyMOL (PyMOL Molecular Graphics System, version 1.8, Schrödinger) for all residues together using a custom Python script ens_tool.py (40). N-O distances for residues in ␣-helix C, harboring the N53I mutation, were calculated using residues and frequency histograms created using GraphPad Prism v7.0. Statistics are shown in Table S3.

Isothermal titration calorimetry
The ITC experiments were performed as described previously (28).

Data availability
The atomic coordinates and structure factors or NMR restraints, respectively, have been deposited in the Protein Data Bank (PDB codes 6Y4O for Ca 2ϩ /CaM-WT bound to RyR2-CAMBD2, 6Y4P for Ca 2ϩ /CaM-N53I bound to RyR2-CAMBD, 6Y94 for Ca 2ϩ /CaM-N53I, and 6Y95 for apoCaM-N53I). The NMR chemical shifts have been deposited in the BioMagRes-Bank (accession nos. 34497 for apoCaM-N53I and 34496 for Ca 2ϩ /CaM-N53I). All other data are contained within the manuscript.