Predicting the Damaging Potential of Uncharacterized KCNQ1 and KCNE1 Variants

Abstract

Voltage-gated potassium channels Kv7.1, encoded by the gene KCNQ1, play critical roles in various physiological processes. In cardiomyocytes, the complex Kv7.1-KCNE1 mediates the slow component of the delayed rectifier potassium current that is essential for the action potential repolarization. Over 1000 KCNQ1 missense variants, many of which are associated with long QT syndrome, are reported in ClinVar and other databases. However, over 600 variants are of uncertain clinical significance (VUS), have conflicting interpretations of pathogenicity, or lack germline information. Computational prediction of the damaging potential of such variants is important for the diagnostics and treatment of cardiac disease. Here, we collected 1750 benign and pathogenic missense variants of Kv channels from databases ClinVar, Humsavar, and Ensembl Variation and tested 26 bioinformatics tools in their ability to identify known pathogenic or likely pathogenic (P/LP) variants. The best-performing tool, AlphaMissense, predicted the pathogenicity of 195 VUSs in Kv7.1. Among these, 79 variants of 66 wildtype residues (WTRs) are also reported as P/LP variants in sequentially matching positions of at least one hKv7.1 paralogue. In available cryoEM structures of Kv7.1 with activated and deactivated voltage-sensing domains, 52 WTRs form intersegmental contacts with WTRs of ClinVar-listed variants, including 21 WTRs with P/LP variants. ClinPred and paralogue annotation methods consistently predicted that 21 WTRs of KCNE1 have 34 VUSs with damaging potential. Among these, 8 WTRs are contacting 23 Kv7.1 WTRs with 13 ClinVar-listed variants in the AlphaFold3 model. Analysis of intersegmental contacts in CryoEM and AlphaFold3 structures suggests atomic mechanisms of dysfunction for some VUSs.

1. Introduction

Voltage-gated potassium channels Kv7.1, encoded by the gene KCNQ1, play important roles in the physiology and pathophysiology of the heart [1,2]. Homotetrameric Kv7.1 associated with different beta subunits, which are encoded by KCNE genes, are expressed in various organs including the colon, kidney, stomach, inner ear, and testis; for reviews, see [3,4]. In the heart, hKv7.1 channels mediate the slow-delayed rectifier potassium current I_Ks, which is essential for the normal cardiac rhythm [5]. Mutations in the KCNQ1 gene are associated with several heart diseases [3], including long QT syndrome [6,7], short QT syndrome [8,9], atrial fibrillation [10,11], and sudden cardiac death [12,13]. Recently, Kv7.1 was found to mediate sex-dependent cold sensation [14].

Each Kv7.1 subunit has six transmembrane (TM) segments. Segments S1 to S4 form the voltage-sensing domain (VSD). Segments S5 and S6, with a membrane-reentrant P-loop between them, contribute a quarter to the pore domain. The C-terminal domain, which is critical for the channel tetramerization, binds calmodulin, which regulates gating [15]. KCNE1 is the primary accessory subunit for the cardiac channel Kv7.1 [16]. Kv7.1-mediated slow potassium current I_Ks plays a relatively minor role under normal heart conditions, but becomes crucial under strong adrenergic stimulation, such as during stress or exercise [17]. Other voltage-gated potassium channels involved in repolarization include Kv1.4, Kv1.5, Kv1.11, Kv4.2, and Kv4.3; see ref. [18] for a review. Thousands of disease-associated mutations in potassium channels are reported in ClinVar [19] and other databases. Here, we focus on missense disease mutations in KCNQ1 and KCNE1.

As of 6 March 2025, the ClinVar database lists about 1000 disease-associated missense variants of KCNQ1. Among these, 519 variants are of unknown clinical significance (VUSs), 131 variants are reported with conflicting interpretations of pathogenicity (CIP), and germline classification is not provided (NP) for 70 variants. The most common diseases associated with Kv7.1 variants are long QT syndrome [20] and short QT interval syndrome [9]. The KCNE1 β subunit has only 103 residues, but as many as 624 missense variants are reported in ClinVar. Among these, 100 variants are VUSs, 21 variants are CIPs, and 24 are NP variants. Artificial neuronal networks have been used to predict some biophysical characteristics of KCNE1 variants [21]. Functional studies, including high-throughput electrophysiology, were employed to reclassify many Kv7.1 VUSs as P/LP variants [22] and reveal molecular mechanisms underlying pathogenicity [23]. The large number of yet uncharacterized VUSs motivates the application of computational methods to predict their damaging potential.

The American College of Medical Genetics and Genomics and the Association for Molecular Pathology recommend using in silico predictive algorithms for the interpretation of variants [24]. Many computational tools based on different principles have been developed to predict the likelihood that a mutation would damage the protein function [25,26]. The performance of in silico tools ranges between 60 and 80% and may depend on the disease phenotype [27] and the protein type [28]. For instance, MetaLR, MetaSVM, and MCap have shown top performance in predicting pathogenicity for variants associated with cardiovascular abnormalities [25]. However, some methods yielded many false-positive and false-negative predictions of pathogenicity in individual protein families [25,29]. Thus, MetaSVM predicted a pathogenic effect for 75% of benign variants of the cardiac sodium channel Nav1.5 [30]. Selecting a tool with a high success rate of correct predictions for specific protein families and adjusting the pathogenicity threshold can improve predictions.

Previously, we used various bioinformatics tools combined with the paralogue annotation method [31] to predict the damaging potential for numerous VUSs of the cardiac sodium channel Nav1.5 [30], calcium channel Cav1.2 [32], and the TRPM4 channel [33]. The paralogue annotation method employs multiple sequence alignment of functionally and structurally related proteins, focusing on residues in sequentially matching positions where a disease mutation is known for at least one family member. A VUS in the matching position of the channel under investigation is then predicted as a likely damaging (LD) variant.

In this study, we collected 1750 benign and pathogenic missense variants of Kv channels from several databases and tested 26 bioinformatics tools in identifying damaging variants. The best-performing tool, AlphaMissense, in combination with the paralogues annotation method, predicted damaging potential for 79 VUSs. We further used another tool, ClinPred, in combination with paralogue annotations to predict the damaging potential for 34 VUSs of 21 WTRs in KCNE1. In available cryoEM structures of Kv7.1 with activated and deactivated VSDs and in the AlphaFold model of KCNE1-bound Kv7.1, 52 WTRs of likely damaging VUSs form intersegmental contacts with ClinVar-listed WTRs, including 21 WTRs with P/LP variants. Analysis of state-dependent contacts revealed in 3D-aligned structures suggests an atomic mechanism of dysfunction for some VUSs.

2. Results and Discussions

2.1. Universal Residue Labels of Kv7.1 and Its Paralogues

In this study, we show UniProt residue numbers of Kv7.1 and its paralogues, as well as PLIC labels, which are universal for P-loop ion channels [34]. A PLIC label refers to the segment, the channel subunit (when necessary), and residue position relative to the reference residue in the segment, which is most conserved in the multiple sequence alignment of P-loop channels. The reference residues have numbers 550 in TM helices or 850 in P-loops. Subunits are designated “A”, “B”, “C”, and “D”. When viewed from the extracellular side, subunits A to D are arranged clockwise [35]. PLIC labels facilitate recognition of residue locations in different P-loop channels. Several 3D-aligned structures of Kv7.1 and Kv7.2 channels with PLIC labels are available in the database https://plic3da.com. Supplemental file PLIC_Uniprot_Kv.xlsx provides relations between PLIC labels and UniProt numbers for potassium channels mentioned in this study.

2.2. Composing a Broad Dataset of Missense Variants for Channel hKv7.1 and Its Paralogues

According to the Ensemble database, there are 31 paralogues of KCNQ1 (ENSG00000053918), but only 14 of them have P/LP variants that are used in the paralogue annotation method. For KCNQ1 and its 14 paralogues (

Table 1. Known missense variants of hKv7.1 and its paralogues.

Gene a	UniProt ID b	P/LP c	VUS d	Common Neutral e
KCNA1	Q09470	38	245	14
KCNA2	P16389	44	207	10
KCNA5	P22460	4	276	56
KCNB1	Q14721	87	212	62
KCNC1	P48547	11	149	11
KCNC2	Q96PR1	12	44	37
KCNC3	Q14003	10	173	82
KCND2	Q9NZV8	7	176	22
KCND3	Q9UK17	21	207	17
KCNQ1	P51787	299	519	43
KCNQ2	O43526	394	474	57
KCNQ3	O43525	39	458	49
KCNQ4	P56696	28	153	63
KCNQ5	Q9NR82	20	228	65
KCNV2	Q8TDN2	45	361	103

^a Genes of voltage-gated potassium ion channels, which have P/LP variants in public databases; ^b Accession number of a protein in the UniprotKB database; ^c Number of P/LP variants; ^d Number of VUSs; ^e Number of variants from the gnomAD database, which occur in a population with allele frequency > 0.00001 and are absent in the ClinVar database.

), we collected a total of 5632 missense variants from the gnomAD, ClinVar, Ensembl, and Humsavar databases (Table S1). These include 1059 P/LP variants, 691 common neutral variants (with AF > 0.00001), and 3882 uncharacterized variants or VUSs. We refer to this dataset as the “broad dataset.” The largest number of P/LP variants (394) was identified in KCNQ2. For the hKv7.1 channel, we found 299 P/LP variants, 43 benign variants (with AF > 0.00001), and 519 VUSs (Table 1 and Table S1).

2.3. Distribution of Missense Variants in Topological Regions of hKv7.1

Many pathogenic variants are localized in the C-terminal region and P-loop of the channel. Most of the P/LP variants are associated with the long QT syndrome 1 (LQT1).

2.4. Comparing Performance of Bioinformatics Tools

We have chosen only those tools that predicted pathogenicity for >70% variants in our dataset. Thus, we excluded from our analysis algorithms EVE, MutPred, and MutationTaster. To compare the performance of 26 prediction tools (

Table 2. Performance of variant interpretation tools.

Tool	Deleterious Threshold a	Sensitivity b	Specificity c	MCC d	ACC e	AUC f
AlphaMissense	>0.5	0.95	0.86	0.81	0.9	0.96
ClinPred	>0.8	0.97	0.79	0.77	0.87	0.95
DEOGEN2	>0.5	0.96	0.67	0.65	0.81	0.95
MetaRNN	>0.6	0.97	0.82	0.79	0.89	0.95
VARITYR	>0.5	0.93	0.86	0.78	0.89	0.95
REVEL	>0.45	0.99	0.56	0.59	0.76	0.94
VARITYER	>0.5	0.9	0.85	0.75	0.87	0.93
VEST4	>0.5	0.97	0.73	0.72	0.85	0.93
LISTS2	>0.85	0.97	0.54	0.55	0.73	0.91
PROVEAN	<−1.5	0.96	0.62	0.61	0.78	0.91
SIFT4G	<0.05	0.9	0.75	0.65	0.82	0.9
CADD	>3	0.93	0.68	0.63	0.8	0.89
MetaLR	>0.4	0.99	0.15	0.25	0.54	0.88
MPC	>2	0.57	0.92	0.51	0.74	0.88
MutationAssessor	>1.7	0.89	0.68	0.58	0.78	0.88
MVP	>0.75	0.99	0.43	0.49	0.68	0.88
PPH_HVAR	>0.45	0.91	0.69	0.61	0.79	0.88
SIFT	<0.0045	0.82	0.8	0.62	0.81	0.88
MetaSVM	>0	0.99	0.28	0.37	0.61	0.86
PPH_HDIV	>0.45	0.93	0.56	0.52	0.73	0.86
MCap	>0.05	1	0.19	0.31	0.57	0.84
PrimateAI	>0.6	0.98	0.46	0.5	0.7	0.84
DANN	>0.5	1	0.04	0.14	0.52	0.76
GenoCanyon	>0.7	0.89	0.35	0.29	0.62	0.68
FATHMM	<−1	0.99	0.08	0.18	0.5	0.64
ESM1b	<−3	0.69	0.35	0.05	0.51	0.52

^a Deleterious threshold is the custom pathogenicity threshold that divides variants into two categories: pathogenic or benign. The larger or smaller the score is than the threshold, the more likely the variant is damaging. ^b Sensitivity characterizes the number of P/LP variants, which were predicted as P/LP by the tool. ^c Specificity characterizes the number of benign variants, which were predicted as benign by the tool. ^d MCC, Matthews Correlation Coefficient. ^e ACC, Accuracy indicates the predictive accuracy of the tool. ^f AUC, Area under the ROC curve.

), we compiled a test set with 691 true-positive (TP) and 1059 true-negative (TN) observations obtained from our broad dataset (Table S1). Pre-computed algorithm scores were retrieved from the database dbNSFP v4.5. For each tool, we determined the optimal pathogenicity threshold and calculated sensitivity, specificity, and accuracy (Table 2). We used the area under the Receiver Operating Characteristic (ROC) curve as the performance measure (Figure 1A).

AlphaMissense and ClinPred demonstrated the best performance (AUC = 0.96) in the broad dataset (Figure 1), followed by VARITYR (AUC = 0.95), MetaRNN (AUC = 0.95), and DEOGEN2 (AUC = 0.95). AlphaMissense correctly classified 96% of the P/LP variants as pathogenic variants and 85% of the common neutral variants as tolerated variants (Table 2).

DEOGEN2, MetaRNN, and VARITYR also have high predictive parameters with AUC = 0.95. The lowest accuracy across all methods was found for ESM1b (AUC = 0.52), FATHMM (0.64), and GenoCanyon (0.68). The results indicate that AlphaMissense and ClinPred are the best-performing pathogenicity predictors for variants in the Kv family.

2.5. Paralogue Annotation of Kv7.1 Variants

Using multiple sequence alignments of hKv7.1 and its paralogues, we mapped each residue with a known P/LP variant from a paralogue protein onto the corresponding amino acid position of hKv7.1. A total of 555 known P/LP variants in paralogues were mapped to 195 amino acid positions in the hKv7.1 channel (Table S2). In these positions, we found 410 variants of hKv7.1, including 174 VUSs, 226 P/LP variants, and 10 common neutral variants. In some cases, multiple variants were mapped to the same sequence position. Forty-three P/LP variants of paralogues were mapped to the P-loop region of hKv7.1, forty-seven P/LP paralogue variants to the C-terminal region, and twenty-four P/LP variants to the S4 segment (Table S2).

2.6. AlphaMissense and Paralogue Annotations Consensually Predicted Damaging Potential for 79 VUSs

As many as 519 variants of Kv7.1 in our dataset are currently classified as VUSs (Table 1). We used the best-performing tool, AlphaMissense, to predict the damaging potential of these VUSs. AlphaMissense identified 195 VUSs with a pathogenicity threshold > 0.5 as P/LP variants. Among these, we further selected those variants that are annotated as P/LP in at least one of the fourteen paralogues of hKv7.1 (Table 1) with a conservation score across paralogues Cs > 0.3. Both methods consensually predicted damaging potential for 79 VUSs (

Table 3. Likely damaging VUSs of hKv7.1 ^a.

PLIC Label	hKv7.1 Variant	Paralogue	ClinPred	α Missense	Current Change b
1.532	E115D	KCNQ2-E86K	0.878	0.997
1.545	A128P	KCNQ2-Y98X	0.185	0.861
1.548	L131P	KCNQ2-L101H	0.995	0.958	↓↓
1.550	V133A	KCNA1-I177N, KCNB1-I199F, KCNQ2-V103D	0.984	0.964	↓
1.553	C136F	KCNQ2-C106G	1	0.954	↓↓
1.554	L137P	KCNQ2-L107F	0.986	0.999
1.557	S140R	KCNA1-F184C	0.996	0.999	↓↓
1.560	S143F	KCNB1-N209K	0.997	0.953
2.544	V164A	KCNQ2-I134N	0.998	0.797
2.550	E170G	KCNQ2-E140A	1	0.987
2.551	Y171H	KCNV2-Y317X	0.998	0.998
2.556	W176S	KCNQ2-W146X	1	0.926
2.609	G189A	KCNQ2-G159E/R/V	0.998	0.978	↓↓
3.549	I201V	KCNA2-I258N	0.944	0.507
3.550	D202V	KCNQ2-D172G	1	0.998
3.560	V212A	KCNQ2-V182M	0.994	0.958
3.560	V212F	KCNQ2-V182M	0.997	0.917
3.600	G219E	KCNA1-E283K, KCNQ2-G189D	0.989	0.817
4.542	A223T	KCNQ2-A193D/V	0.994	0.772
4.547	R228W	KCNA2-R294H, KCNQ2-R198W	1	0.966
4.553	Q234L	KCNA2-R300S, KCNB1-R303Q, KCNC3-R420H, KCNQ2-Q204H	0.998	0.965
4.553	Q234R	KCNA2-R300S, KCNB1-R303Q, KCNC3-R420H, KCNQ2-Q204H	0.97	0.992
4.558	L239V	KCNQ2-I209S/T	0.829	0.86
4.559	H240R	KCNQ2-R210C/H/P	0.331	0.949
4.559	H240Q	KCNQ2-R210C/H/P	0.993	0.991
5.518	G245R	KCND3-S304F	0.998	0.998	↓↓
5.523	L250P	KCNA1-I314T, KCNB1-S319F/Y	0.995	1
5.525	G252R	KCNB1-G321S	1	0.996
5.525	G252S	KCNB1-G321S	0.998	0.885
5.526	S253A	KCNQ2-S223F/P	0.972	0.792
5.530	I257S	KCNQ2-A227V	0.993	0.947
5.533	Q260H	KCNQ2-K230M	0.991	0.99
5.535	L262V	KCNC3-F448L	0.996	0.952
5.536	I263K	KCNB1-G332V	0.997	0.997
5.537	T264S	KCNQ2-T234A/P	0.992	0.895
5.540	Y267F	KCNQ2-Y237C	0.996	0.644
5.541	I268V	KCNC2-F388S	0.878	0.674
5.548	F275L	KCNQ2-L245P	0.996	0.986
5.552	F279C	KCND3-V338E, KCNQ2-L249P	0.999	0.788
5.556	A283T	KCNQ2-A253S/T	0.996	0.504
5.557	E284G	KCNQ2-E254D	1	0.972
5.613	S298R	KCNQ2-T263A/I	0.997	0.977
5.837	A300G	KCNQ2-A265P/T/V	0.986	0.747
5.843	G306E	KCNQ2-G271D/R/S/V, KCNQ3-G310D/V	0.999	0.999
5.844	V307M	KCNB1-T372N/I	0.991	0.79
5.844	V307L	KCNB1-T372N/I	0.99	0.895
5.844	V307E	KCNB1-T372M/I	0.993	0.989
5.849	T312S	KCNA2-T374A, KCNC2-T437A, KCNQ2-T277N/P/S	0.995	0.966
5.850	I313F	KCNQ2-I278F/M/T, KCNQ3-I317M/T	0.993	0.991
5.855	K318N	KCNQ2-K283E	0.958	0.93
5.856	V319M	KCNQ2-Y284C/D	0.995	0.616
6.543	A329V	KCND3-G384S, KCNQ2-A294G/S	0.999	0.979
6.543	A329T	KCND3-G384S, KCNQ2-A294G/S	0.987	0.935
6.544	S330Y	KCND3-S385P	0.998	0.993
6.545	C331Y	KCNQ2-T296P	0.998	0.976
6.549	F335C	KCNA1-A395S	0.997	0.856
6.551	I337M	KCNA2-V399M	0.962	0.775
6.553	F339V	KCNQ2-F304C, KCNQ2-F304S	0.999	0.972
6.555	A341T	KCNA1-A401V, KCNB1-A406V, KCNQ2-A306P/T/V/E	0.996	0.976
6.563	S349A	KCNB1-N414D	0.995	0.92
6.563	S349L	KCNB1-N414D	0.999	0.995
6.566	A352P	KCNB1-S417P, KCNQ2-A317T, KCNQ3-A356T	0.999	0.997
6.566	A352D	KCNB1-S417P, KCNQ2-A317T, KCNQ3-A356T	0.998	1
6.569	K354R	KCNQ2-K319E	0.992	0.768
6.571	Q357R	KCNA1-R417X, KCNB1-E422A	0.996	0.984
6.571	Q357E	KCNA1-R417X, KCNB1-E422A	0.939	0.566
6.574	R360T	KCNQ2-R325G, KCNQ3-R364C/H	0.999	0.999
6.574	R360K	KCNQ2-R325G, KCNQ3-R364C/H	0.992	0.976
7.013	L374V	KCNQ2-L339Q, KCNQ2-L339R	0.995	0.85
7.030	T391P	KCNQ2-T359K	0.998	0.898
7.165	K526E	KCNQ2-K552N	0.932	0.967
7.165	K526Q	KCNQ2-K552N	0.838	0.717
7.165	K526N	KCNQ2-K552N	0.984	0.995
7.180	V541I	KCNQ2-V567D	0.903	0.814
7.187	G548D	KCNQ2-G574D/S	0.999	0.999
7.194	R555L	KCNQ2-R581G	0.998	0.994
7.196	K557R	KCNQ2-K583N	0.982	0.63
7.222	R583S	KCNQ2-K606X	0.885	0.902
7.222	R583G	KCNQ2-K606X	0.79	0.606
7.230	R591P	KCNQ2-R622P	0.997	0.997

^a Shown are PLIC labels that are universal for P-loop channels [34] and UniProt residue numbers of specific channels. ^b Current density decreases strongly (↓↓) or moderately (↓) according to [22].

).

Alfred George and co-authors [22] functionally assessed 78 Kv7.1 variants using a repurposed automated electrophysiology platform originally designed for drug discovery. The authors reclassified over 65% of the examined VUSs and found strong concordance with conventional electrophysiology results. Among the 79 likely damaging VUSs predicted in our study (Table 3), 6 were functionally characterized [22]; 5 of these exhibited a strong reduction in current density, and 1 showed a moderate decrease (Table 3). These experimental data support the predictive power of our approach. In the next section, we describe intersegmental contacts of WTRs in likely damaging VUSs, which may help guide their selection for functional studies.

2.7. Intersegmental Contacts Involving WTRs with Likely Damaging VUSs

The above bioinformatics approaches strongly suggest the damaging potential of 79 VUSs in Kv7.1, but the molecular mechanisms of the variants’ dysfunction are unclear. Analysis of intersegmental contacts and their state dependency may suggest mechanisms of stabilization or destabilization of specific channel states that underlie the molecular mechanisms. We this goal in mind, we 3D-aligned cryoEM structures of Kv7.1 with VSDs in the activated (8sik) and deactivated (8sin) states [36], as described in the Methods. Upon a VSD deactivation, helix S4 underwent a substantial downshift (Figure 2A) and rotation (Figure 2B), while rearrangements of helices S5 and S6 were rather small. In both 8sik and 8sin structures, the pore at the level of serines S^6.563 is narrow (Figure 3A), implying a closed channel [36]. Surprisingly, leucines L^6.567 in the 8sik structure form a much tighter constriction than in 8sin (Figure 3A). However, CA atoms of leucines L^6.567 in both structures are close to each other, and respective CA-CB bonds are collinear (Figure 3A), indicating that the tighter constriction in 8sik is due to the sidechain rotations rather than rearrangement of the S6 bundle.

In epithelial cells, KCNE3 associates with KCNQ1 to form apparently voltage-independent channels. By using voltage clamp fluorimetry, Larsson and co-authors [37] found that KCNE3 shifts the voltage dependence of S4 movement to extreme hyperpolarized potentials, making the channel constitutively open. These results suggest that KCNE3 primarily affects the voltage-sensing domain and only indirectly affects the gate. In a later study, Kasuya and Nakajo used the KCNQ1–KCNE3–calmodulin complex structure to examine amino acid residues in KCNE3 and the S1 segment of the KCNQ1 [38]. By changing the side-chain bulkiness of these interacting residues, the authors found that the distance between the S1 segment and KCNE3 is optimized to achieve constitutive activity, suggesting that the tight binding of the S1 segment and KCNE3 is crucial for controlling the VSDs.

While VSDs in the cryoEM structure (6v01) of KCNE3-bound Kv7.1 are deactivated [6], the S6 bundle is much wider than that in 8sin (Figure 3B), indicating that KCNE3 binding is sufficient to widen the pore even when VSDs are deactivated. Thus, although cryoEM structures of Kv7.1 with activated VSDs and open pore domain are lacking, available cryoEM structures of Kv7.1 allow analyzing intersegmental contacts of WTRs of likely damaging VUSs and, to some extent, their state-dependency.

We visualized intersegmental contacts in VSDs involving WTRs of likely damaging VUSs and WTRs of other ClinVar-reported variants in structures with deactivated (Figure 2C) and activated (Figure 2D) VSDs. Since the pore domain and especially its extracellular half are more 3D-conserved than VSDs, we analyzed intersegmental contacts in the pore domain in the structure with deactivated VSDs (Figure 4, Figure 5 and Figure 6), and then in the model with the KCNE1-bound channel. These contacts are listed in Table 4, Table 5, Table 6 and Table 7.

The vast majority of Kv7.1 P/LP variants, which are reported in ClinVar, as well as likely damaging VUSs (Table 3), are associated with long QT syndrome, indicating that respective mutations destabilize the open channel, decrease the I_Ks current, and thus prolong the action potential. In the case of the cardiac channel Nav1.5, for many intersegmental state-dependent contacts involving WTRs with ClinVar-reported variants, substitution of either contact partner causes the channel dysfunction associated with the same syndrome [39]. By analogy, we propose that if two WTRs of Kv7.1 form an intersegmental contact with one partner having a ClinVar-reported P/LP variant and another partner is a likely damaging VUS, it strongly increases the reliability of our prediction on the likely damaging VUS.

Table 4. Likely damaging (LD) VUSs whose WTRs contact WTRs of ClinVar-reported P/LP variants ^a.

LD VUS	P/LP	State b			LD VUS	P/LP	State
V133/1.550A	R231/4.550L		↓		L262/5.535V	P343/6.557A		↓
	Q234/4.553P	↑		o		P343/6.557A/S	↑
C136/1.553F	S225/4.544L/W		↓		T264/5.537S	L233/4.552M		↓
	L156/2.536P	↑		o		L251/5.524Q/P	↑
	Q234/4.553P	↑				G269/5.542R/S/V/D	↑
L137/1.554P	S225/4.544L/W		↓		Y267/5.540F	G229/4.548D		↓
	G229/4.548D		↓			L233/4.552M		↓
	Q234/4.553P	↑			E284/5.557G	T322/5.859P/A/R		↓
	I235/4.554N	↑				P320/5.857S		↓
	I274/5.547D	↑				G325/6.539W
	R231/4.550S/C/L/H			o	G306/5.843E	L273/5.546I/V/P/R/F		↓
	Q234/4.553P			o	V307/5.844M/L/E	S330/\6.544Y		↓
	I235/4.554L/N			o	T312/5.849S	T312/5.849I/S		↓
S140/1.557R	S225/4.544L/W		↓		I313/5.850F	G314/5.851R/D/S		↓
	L156/2.536P	↑				T312/5.849S/I		↓
	R231/4.550S/C/L/H	↑		o		T309/5.846I		↓
	Q234/4.553P	↑		o	V319/5.856M	Y315/5.852S		↓
	L156/2.536P			o		W304/5.841R/L/S		↓
S143/1.560F	R231/4.550S/C/L/H			o		W304/5.841G		↓
V164/2.544A	S209/3.557P			o		Y315/5.852D		↓
E170/2.550G	R231/4.550S/H/C/L		↓		I337/6.551M	F340/6.554L		↓
D202/3.550V	Q234/4.553P		↓		F339/6.553V	L251/5.524Q		↓
	R231/4.550S/C/L/H		↓			L251/5.524P		↓
	R234/4.562S/C/L/H/P	↑			A341/6.555T	A344/6.558E/V		↓
	R243/4.562S/C/L/H/P			o	S349/6.563A/L	G345/6.559R/V/E		↓
Q260/A5.533H	L251/D5.524Q/P		↓		R360/6.574K/T	R539/7.178W/Q		↓
L262/5.535V	P343/6.557A/S			o	K557/7.196R	R555/7.194S/C		↓

^a ClinVar of 4 March 2025; ^b ↓, contacts in the hKv7.1 cryoEM structure with “down” VSDs (PDB ID: 8sin) [36]; ↑, contacts in the hKv7.1 cryoEM structure with “up” VSDs (PDB ID: 8sik) [36]; o, Contacts in hKv7.1-KCNE3-PIP3 cryoEM structure with open PD and “down” VSDs (PD ID: 6v01) [40].

Table 4. Likely damaging (LD) VUSs whose WTRs contact WTRs of ClinVar-reported P/LP variants ^a.

LD VUS	P/LP	State b			LD VUS	P/LP	State
V133/1.550A	R231/4.550L		↓		L262/5.535V	P343/6.557A		↓
	Q234/4.553P	↑		o		P343/6.557A/S	↑
C136/1.553F	S225/4.544L/W		↓		T264/5.537S	L233/4.552M		↓
	L156/2.536P	↑		o		L251/5.524Q/P	↑
	Q234/4.553P	↑				G269/5.542R/S/V/D	↑
L137/1.554P	S225/4.544L/W		↓		Y267/5.540F	G229/4.548D		↓
	G229/4.548D		↓			L233/4.552M		↓
	Q234/4.553P	↑			E284/5.557G	T322/5.859P/A/R		↓
	I235/4.554N	↑				P320/5.857S		↓
	I274/5.547D	↑				G325/6.539W
	R231/4.550S/C/L/H			o	G306/5.843E	L273/5.546I/V/P/R/F		↓
	Q234/4.553P			o	V307/5.844M/L/E	S330/\6.544Y		↓
	I235/4.554L/N			o	T312/5.849S	T312/5.849I/S		↓
S140/1.557R	S225/4.544L/W		↓		I313/5.850F	G314/5.851R/D/S		↓
	L156/2.536P	↑				T312/5.849S/I		↓
	R231/4.550S/C/L/H	↑		o		T309/5.846I		↓
	Q234/4.553P	↑		o	V319/5.856M	Y315/5.852S		↓
	L156/2.536P			o		W304/5.841R/L/S		↓
S143/1.560F	R231/4.550S/C/L/H			o		W304/5.841G		↓
V164/2.544A	S209/3.557P			o		Y315/5.852D		↓
E170/2.550G	R231/4.550S/H/C/L		↓		I337/6.551M	F340/6.554L		↓
D202/3.550V	Q234/4.553P		↓		F339/6.553V	L251/5.524Q		↓
	R231/4.550S/C/L/H		↓			L251/5.524P		↓
	R234/4.562S/C/L/H/P	↑			A341/6.555T	A344/6.558E/V		↓
	R243/4.562S/C/L/H/P			o	S349/6.563A/L	G345/6.559R/V/E		↓
Q260/A5.533H	L251/D5.524Q/P		↓		R360/6.574K/T	R539/7.178W/Q		↓
L262/5.535V	P343/6.557A/S			o	K557/7.196R	R555/7.194S/C		↓

^a ClinVar of 4 March 2025; ^b ↓, contacts in the hKv7.1 cryoEM structure with “down” VSDs (PDB ID: 8sin) [36]; ↑, contacts in the hKv7.1 cryoEM structure with “up” VSDs (PDB ID: 8sik) [36]; o, Contacts in hKv7.1-KCNE3-PIP3 cryoEM structure with open PD and “down” VSDs (PD ID: 6v01) [40].

Table 5. Likely damaging (LD) VUSs whose WTRs form intersegmental contacts with WTRs of other ClinVar-reported VUSs ^a.

LD VUS	Contact b		LD VUS	Contact b
V133/A1.550I/A	R228/A4.547W	VUS → LD	K318/A5.855N	D301/A5.838Y	VUS
	G229/A4.548S/V	VUS	S349/A6.563A/L	S349/B,D6.563A/L	VUS → LD
	Y267/B5.540F	VUS → LD		A352/B,D6.566P/D	VUS → LD
C136/A1.553F	R228/A4.547W	VUS→ LD	A352/A6.566P/D	L353/B6.567P	VUS
L137/A1.554P	G229/A4.548S/V	VUS		S349/B6.563A/L	VUS → LD
	R228/A4.547W	VUS → LD		G350/B6.564R	VUS
D202/A3.550V	Q234/A4.553L/R	VUS → LD	R360/A6.574K/T	P535/A7.174T	VUS
S253/A5.526A	K354/A6.568R	VUS → LD	T391/A7.030P	R518/A7.157Q/P	VUS
Y267/A5.540F	G229/D4.548S/V	VUS	V541/A7.180I	V541/B,D7.180I	VUS → LD
E284/A5.557G	V324/6.538L/I/F	VUS		Y545/B7.184F	VUS
A300/A5.837S/G	K326/B6.540E	VUS	G548/A7.187D	Y545/B7.184F	VUS
T312/A5.849S	I313/5.850F	VUS → LD
	I337/A6.551M	VUS → LD

^a ClinVar Data on 4 March 2025; ^b In the cryoEM structure of hKv7.1 (PDB ID: 8sin) with voltage sensor in the down state [36].

Table 5. Likely damaging (LD) VUSs whose WTRs form intersegmental contacts with WTRs of other ClinVar-reported VUSs ^a.

LD VUS	Contact b		LD VUS	Contact b
V133/A1.550I/A	R228/A4.547W	VUS → LD	K318/A5.855N	D301/A5.838Y	VUS
	G229/A4.548S/V	VUS	S349/A6.563A/L	S349/B,D6.563A/L	VUS → LD
	Y267/B5.540F	VUS → LD		A352/B,D6.566P/D	VUS → LD
C136/A1.553F	R228/A4.547W	VUS→ LD	A352/A6.566P/D	L353/B6.567P	VUS
L137/A1.554P	G229/A4.548S/V	VUS		S349/B6.563A/L	VUS → LD
	R228/A4.547W	VUS → LD		G350/B6.564R	VUS
D202/A3.550V	Q234/A4.553L/R	VUS → LD	R360/A6.574K/T	P535/A7.174T	VUS
S253/A5.526A	K354/A6.568R	VUS → LD	T391/A7.030P	R518/A7.157Q/P	VUS
Y267/A5.540F	G229/D4.548S/V	VUS	V541/A7.180I	V541/B,D7.180I	VUS → LD
E284/A5.557G	V324/6.538L/I/F	VUS		Y545/B7.184F	VUS
A300/A5.837S/G	K326/B6.540E	VUS	G548/A7.187D	Y545/B7.184F	VUS
T312/A5.849S	I313/5.850F	VUS → LD
	I337/A6.551M	VUS → LD

^a ClinVar Data on 4 March 2025; ^b In the cryoEM structure of hKv7.1 (PDB ID: 8sin) with voltage sensor in the down state [36].

Table 6. Intersegmental contacts ^a between WTRs of likely damaging (LD) VUSs and WTRs of CIP or NP variants ^b.

LD VUS	Contact		LD VUS	Contact
V164/A2.544A	S209/A3.557P	NP	I313/A5.850F	G314/B5.851C/A	NP
A223/A4.542T	Y278/B5.551H	NP		T309/B5.846S/R	NP
Q260/A5.533H	L236/D4.555R/P	CIP		V308/B5.845D	NP
T264/A5.537S	L236/D4.555R/P	CIP	K318/A5.855N	D301/A5.838V	CIP
	L233/D4.552P	CIP	I337/A6.551M	T311/D5.848A	NP
Y267/A5.540F	L233/4.552P	CIP		T311/D5.848I	CIP
F275/A5.548L	A226/D4.545V	CIP	A341/A6.555T	A344/D6.558T/G	CIP
E284/A5.557G	T322/5.859K	CIP	S349/A6.563A/L	S349/B,D6.563P	CIP
	F296/5.611S	CIP	A352/A6.566P/D	S349/B6.563P	CIP
T312/A5.849S	I337/A6.551F	CIP		G350/B6.564V	CIP
			T391/A7.030P	R518/A7.157G	CIP

^a In the cryoEM structure of hKv7.1 (PDB ID: 8sin) with the voltage sensor in the down state [36]; ^b ClinVar Data of 4 March 2025.

Table 6. Intersegmental contacts ^a between WTRs of likely damaging (LD) VUSs and WTRs of CIP or NP variants ^b.

LD VUS	Contact		LD VUS	Contact
V164/A2.544A	S209/A3.557P	NP	I313/A5.850F	G314/B5.851C/A	NP
A223/A4.542T	Y278/B5.551H	NP		T309/B5.846S/R	NP
Q260/A5.533H	L236/D4.555R/P	CIP		V308/B5.845D	NP
T264/A5.537S	L236/D4.555R/P	CIP	K318/A5.855N	D301/A5.838V	CIP
	L233/D4.552P	CIP	I337/A6.551M	T311/D5.848A	NP
Y267/A5.540F	L233/4.552P	CIP		T311/D5.848I	CIP
F275/A5.548L	A226/D4.545V	CIP	A341/A6.555T	A344/D6.558T/G	CIP
E284/A5.557G	T322/5.859K	CIP	S349/A6.563A/L	S349/B,D6.563P	CIP
	F296/5.611S	CIP	A352/A6.566P/D	S349/B6.563P	CIP
T312/A5.849S	I337/A6.551F	CIP		G350/B6.564V	CIP
			T391/A7.030P	R518/A7.157G	CIP

^a In the cryoEM structure of hKv7.1 (PDB ID: 8sin) with the voltage sensor in the down state [36]; ^b ClinVar Data of 4 March 2025.

Table 7. CryoEM structure ^a with activated VSDs: intersegmental contacts of WTRs in VSD and S5 ^b.

LD VUS	Contact
C136/A1.553F	M159/2539L	VUS
S140/A1.557R	Q260/A5.533H	LD VUS
V164/A2.544A	M210/3558T/I	VUS
E170/A2.550G	H240/4.559Q	LD VUS
	V129/1546I/G/A	VUS
D202/A3.550V	H240/4.559Q	LD VUS
L239/4.558V	Y267/A5.540F	LD VUS
	F275/A5.548S/L	LD VUS
H240/4.559Q/R	E170/A2.550G	LD VUS
	D202/A3.550H	LD VUS
V241/4.560I	Y267/5.540F	VUS
Y267/A5.540F	V241/4.560I	LD VUS
	L239/4.558V	LD VUS
F275/A5.548L	L239/4.558V	LD VUS

^a In the cryoEM structure of hKv7.1 (PDB ID: 8sik) with the voltage sensor in the up state [36]; ^b ClinVar Data of 4 March 2025.

Table 7. CryoEM structure ^a with activated VSDs: intersegmental contacts of WTRs in VSD and S5 ^b.

LD VUS	Contact
C136/A1.553F	M159/2539L	VUS
S140/A1.557R	Q260/A5.533H	LD VUS
V164/A2.544A	M210/3558T/I	VUS
E170/A2.550G	H240/4.559Q	LD VUS
	V129/1546I/G/A	VUS
D202/A3.550V	H240/4.559Q	LD VUS
L239/4.558V	Y267/A5.540F	LD VUS
	F275/A5.548S/L	LD VUS
H240/4.559Q/R	E170/A2.550G	LD VUS
	D202/A3.550H	LD VUS
V241/4.560I	Y267/5.540F	VUS
Y267/A5.540F	V241/4.560I	LD VUS
	L239/4.558V	LD VUS
F275/A5.548L	L239/4.558V	LD VUS

^a In the cryoEM structure of hKv7.1 (PDB ID: 8sik) with the voltage sensor in the up state [36]; ^b ClinVar Data of 4 March 2025.

Such contacts with known P/LP residues involve 31 likely damaging VUSs of 21 WTRs, including eight variants in VSD, seven variants in S5, eight variants in P-loop, and eight variants in S6 (Table 4). Contacts within a VSD and between VSD and S5 are clearly state-dependent (cf. Figure 2C,D). In particular, H-bonds R^4.550---S^1.560, Q^4.553---E^2.540_, salt bridges H^4.559: E^2.550 and H^4.559: D^3.550, and hydrophobic interactions L^4.558: F^5.548 stabilize the “up” state of S4 and thus activated state of VSD (Figure 2D). Mutations R^4.550L, D^3.550H, Q^4.553P, E^2.550G (Table 4), L^4.558V, and F^275/A5.548S/L (Table 7) would weaken or eliminate these contacts and destabilize the activated VSD. This would decrease the probability of the channel open state, reduce I_Ks, and prolong the action potential in cardiomyocytes, explaining why respective variants are associated with the gain-of-function long QT syndrome. However, some of the above mutations (e.g., R^4.550L and E^2.550G) would also destabilize deactivated VSD (Figure 2D). Therefore, the long QT syndrome associated with these variants indicates that respective mutations destabilize the activated state of VSD more strongly than its deactivated state.

Table 4 also shows seven likely damaging VUSs in the cytoplasmic part of S6 (positions 6.553–6.574) and six likely damaging VUSs in the cytoplasmic part of S5 (positions 5.533–5.540). These residues undergo substantial movements upon activation gating, implying that respective mutations destabilize the channel open state. Importantly, WTRs of the respective variants contact WTRs with ClinVar-reported P/LP variants (Table 4). These residues are marked with red labels in Figure 4.

The second category of contacts includes 16 WTRs of 26 likely damaging VUSs and 25 partner WTRs of 35 ClinVar-reported VUSs (Table 5). Among the contact partners, 13 WTRs have 18 VUSs that we also predicted as likely damaging variants (Table 4). The fact that VUSs of both partners in such contacts were independently selected by AlphaMissense and paralogue annotation methods further justifies our predictions that respective VUSs are likely damaging, making them promising objects for future functional analyses.

The third category of contacts includes 15 WTRs of 20 likely damaging VUSs and 16 contact-partner WTRs for which 26 variants are reported in ClinVar as either CIP variants or their germline classification is not provided (Table 6). We suggest a damaging potential of the CIP variants because at least one submitter reported likely pathogenicity of the variant, and the respective WTR is contacting a WTR with a likely damaging VUS.

We found four WTRs of ClinVar-reported benign/likely benign (B/LB) variants that form intersegmental contacts with WTRs of likely damaging VUSs (Supplementary Figure S1). Among these contacts, only B/LB mutation S^1.560F would affect the polar contact with T^2.553 and thus the stability of the VSD state.

2.8. Predicting Damaging Potential for KCNE1 VUSs with High ClinPred Score and ClinVar-Reported Variants in Sequentially Matching Positions of Paralogues

KCNE1 association with Kv7.1 is important for normal heart rhythm; for reviews, see [5,16]. As of March 2025, over 600 variants of KCNE1 are reported in ClinVar, but only three missense variants (T⁷I, M¹L, and G⁵²R) are classified as pathogenic. Among four known KCNE1 paralogs (KCNE2, KCNE3, KCNE4, and KCNE5), KCNE2 has five P/LP variants, while for the other paralogues, only VUSs are reported. Therefore, the paralogues annotation method for KCNE subunits is less reliable than for Kv channels. Since AlphaMissense correctly predicted the damaging effect of fewer than 60% of KCNE variants, we used the second top-performing tool (Section 2.4), ClinPred, with the standard cutoff score (deleterious threshold) of 0.5. ClinPred and the paralogue annotations consensually predicted the damaging potential of 34 VUSs for 21 WTRs in KCNE1 (

Table 8. KCNE1 VUSs reclassified as LP variants.

VUS	Cs	Functional Study a	ClinPred	Paralogue
S28L	0.6	Smaller current	0.818	KCNE5-VUS:D44H
R32C	0.8	faster activation	0.622	KCNE3-VUS:R47G,KCNE3-VUS:R47Q, KCNE3-VUS:R47W
P35S	0.4	faster activation	0.957	KCNE2-VUS:V41A
L48I	0.9	~ current	0.963	KCNE2-Disease:M54T, KCNE2-VUS:M54V
L48F	0.9	faster activation	0.808	KCNE2-Disease:M54T, KCNE2-VUS:M54V
L48P	0.9		0.999	KCNE2-Disease:M54T, KCNE2-VUS:M54V
M49T	0.7		0.958	KCNE5-VUS:L65F
F53L	0.8	~ current	0.756	KCNE2-VUS:M59I, KCNE5-VUS:F69V
F53C	0.8	Small current	0.996	KCNE2-VUS:M59I,KCNE5-VUS:F69V
G55R	0.8		0.994	KCNE2-VUS:S61P
T58P	0.6	Slower activation	0.982	NA-Disease:T58P,KCNE3-VUS:V72G
T58A	0.6		0.987	NA-Disease:T58P,KCNE3-VUS:V72G
L59P	0.7	LoF	0.978	KCNE2-VUS:V65L,KCNE2-VUS:V65M, KCNE5-VUS:G75R
G60D	0.8	Small current	0.994	KCNE2-VUS:A66V,KCNE3-VUS:S74R, KCNE5-VUS:G76D
G60V	0.8		0.995	KCNE2-VUS:A66V,KCNE3-VUS:S74R, KCNE5-VUS:G76D
I61F	1		0.983	KCNE2-VUS:I67M
I66L	0.7		0.905	KCNE3-VUS:T80I
R67G	0.9	Small current	0.996	KCNE3-VUS:R81C
R67S	0.9	Small current	0.987	KCNE3-VUS:R81C
R67L	0.9	Small current	0.99	KCNE3-VUS:R81C
R67H	0.9	Small current	0.873	KCNE3-VUS:R81C
K69E	0.9		0.953	KCNE3-VUS:R83C,KCNE3-VUS:R83P, KCNE5-VUS:R85H
K70Q	0.9		0.977	KCNE3-VUS:K84E,KCNE5-VUS:K86E
K70M	0.9	Small current	0.994	KCNE3-VUS:K84E,KCNE5-VUS:K86E
K70E	0.9		0.983	KCNE3-VUS:K84E,KCNE5-VUS:K86E
L71V	0.4		0.956	KCNE2-Disease:R77W, KCNE3-VUS:V85A
E72K	0.4		0.984	KCNE5-VUS:V88D,KCNE5-VUS:V88I
H73R	0.6		0.931	KCNE2-VUS:H79R,KCNE5-VUS:E89K
S74W	0.4	Small current	0.998	KCNE2-VUS:S80P,KCNE3-VUS:R88C, KCNE3-VUS:R88H
S74P	0.4	Smaller current	0.629	KCNE2-VUS:S80P,KCNE3-VUS:R88C, KCNE3-VUS:R88H
Y81C	0.7	Smaller current	0.998	KCNE2-VUS:Y87C
I82M	0.4	Smaller current	0.885	KCNE3-VUS:I96S
I82V	0.4	Smaller current	0.978	KCNE3-VUS:I96S
I82F	0.4		0.993	KCNE3-VUS:I96S

^a ref. [41]; ~ current means “approximately the same current as in the wildtype channel”.

). Among the 34 VUSs, 20 variants were explored in a recent high-throughput functional study of 68 KCNE1 variants [41], and for most of these, the pick current decrease in the KCNQ1-KCNE1 assembly is reported (Table 8). These experimental data confirm the reliability of our approach to predict the damaging potential of KNCE1 variants.

2.9. AlphaFold 3 (AF3) Model of Kv7.1 with KCNE1

In the lack of cryoEM structures of Kv7.1-KCNE1, we created an AF3 multimer model [42,43] of complex Kv7.1-KCNE1 and Monte Carlo (MC) minimized its energy with rigid backbones (Figure 7). In this model, the large N-terminal portion of each KCNE1 folded up to fill the space between two VSDs, while the smaller C-terminal part extended to the cytoplasm. WTRs of pathogenic variants M¹L, T⁷I, and G⁵²R are shown as blue spheres. Likely damaging VUSs of KCNE1 (Table 8) are shown as sticks in close-up views of Figure 7. Among multiple Kv7.1 WTRs that form contacts with KCNE1, six have known P/LP variants, and several others have likely damaging VUSs (Table 3). Since these data alone are insufficient to prove our AF3 model of Kv7.1-KCNE1, we compared the model with the cryoEM structure of KCNE3-bound Kv7.1 as described in the next section.

2.10. CryoEM Structure of KCNE3-Bound Kv7.1

Kv7.1 co-assembles with KCNE3 in non-excitable cells [40,44], but there is evidence that such complexes are also expressed in the human heart, especially in diseased hearts [45]. KCNE1 and KCNE3 sequences are very different (Figure 8A), but there is sequence similarity in the middle part where KCNE3, which is resolved in the cryoEM structure (6v01) of KCNE3-bound Kv7.1 [40], where the TM helix is bound between S1 of VSD and the extracellular parts of helices S5 and S6 (Figure 8B). In the 3D-aligned cryoEM structure of Kv7.1-KCNE3 and the AF3 model of Kv7.1-KCNE1, the TM part of the KCNE3 helix (M⁶⁰ to T⁸⁰) overlaps with the KCNE1 helix from L⁴⁵ to I⁶⁶ (Figure 8C). It should be noted that the full-fledged cryoEM structure and the AF3 model are 3D-aligned by minimizing RMS deviations of sequentially matching CA atoms in P-loop helices P1 (see Methods) rather than sequentially matching CA atoms in the KCNE TM helices (Figure 8A). The perfect 3D match of the TM helices validates the position of the respective KCNE1 TM helix in our AF3 model (Figure 7).

3. Methods and Materials

3.1. Sequence Data of Human Channels and Collection of Variants

The sequence of hKv7.1 was obtained from the UniProt database [46] (accession number P51787). Paralogues of the hKv7.1 channel were identified using Ensembl [47]. Missense mutations for Kv7.1 and its paralogues were collected from three databases: Humsavar, Ensembl Variation [48], and ClinVar [49]. Only P or LP variants were extracted from Ensembl Variation, ClinVar, and Humsavar. VUSs were also obtained from ClinVar. Common benign (neutral) variants, along with their minor allele frequencies (AF), were sourced from the population database gnomAD [50]. Variants with AF > 0.00001 that are not present in ClinVar were considered benign [51,52]. The total number of collected P/LP, VUS, and common neutral variants is given in Table 1. All variants were compiled into a comprehensive dataset (Table S1).

3.2. Topology of the Kv7.1 Channel

The hKv7.1 regions were defined in accordance with the UniProt entry P51787. The pore-forming α1 subunit of Kv7.1 assembles from four subunits. A large cytoplasmic domain plays critical roles in subunit assembly, interactions with regulatory proteins, and modulation of channel function [53].

3.3. Multiple Sequence Alignment and Paralogue Annotation

The paralogue annotation method identifies P/LP missense variants by transferring annotations across families of related proteins [31]. Previously, we applied a modified version of this method to predict likely damaging variants for channels hNav1.5 [30], hCav1.2 [32], and TRPM4 [33]. Here, we used the same approach to predict likely damaging variants for VUSs of hKv7.1.

For each paralogue channel, P/LP variants were collected. The amino acid sequences of hKv7.1 and its paralogue channels were aligned using the multiple sequence alignment program Tcoffee [54]. Proteins lacking P/LP variants were excluded from the alignment [26]. Since disease-causing mutations tend to occur at evolutionarily conserved positions, we computed position-specific conservation scores (Cs), which range from 0 (no conservation) to 1 (identical), with Cs = 0.8 indicating high conservation. These scores reflect the conservation of physicochemical properties (small, polar, hydrophobic, tiny, charged, negative, positive, aromatic, aliphatic, and proline) in the alignment [55]. Cs values were calculated using the Zvelebil method [56], as implemented in the Amino Acid Conservation Calculation Service [57]. Variants occurring at positions with Cs > 0.3 were classified as P/LP variants.

3.4. Sequence-Based Prediction of Likely Damaging Variants

Missense variants were annotated with scores from 29 algorithms (REVEL, VEST4, MVP, CADD, LIST.S2, VARITY_R, VARITY_ER, AlphaMissense, EVE, MPC, MVP, DANN, CenoCanyon, PrimateAI, DEOGEN2, M-CAP, MetaLR, MetaSVM, MetaRNN, FATHMM, PROVEAN, MutationAssessor, MutPred, PolyPhen2-HVAR, PolyPhen2-HDIV, SIFT, SIFT4G, LRT, MutationTaster), which were obtained from the dbNSFPv4.5 database [58]. To generate binary predictions (Damaging/Tolerated), we used thresholds determined as the optimal pathogenicity threshold from the AUC-ROC curve (Table 2).

The ‘probably damaging’ and ‘possibly damaging’ classes predicted by Polyphen were merged into a single ‘damaging’ class. For results from the MutationAssessor server, which subdivides mutants into four categories, we treated categories high (‘H’) or medium (‘M’) as ‘Damaging’, and categories low (‘L’) or neutral (‘N’) as ‘Tolerated’.

The overall prediction performance of the 29 methods was assessed by calculating sensitivity, specificity, Matthews Correlation Coefficient (MCC), and accuracy (ACC) as follows:

$$\mathrm{S}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{v}\mathrm{i}\mathrm{t}\mathrm{y}={\displaystyle \frac{TP}{TP+FN}};$$

(1)

$$\mathrm{S}\mathrm{p}\mathrm{e}\mathrm{c}\mathrm{i}\mathrm{f}\mathrm{i}\mathrm{c}\mathrm{i}\mathrm{t}\mathrm{y}={\displaystyle \frac{TN}{TN+FP}};$$

(2)

$$\mathrm{M}\mathrm{C}\mathrm{C}={\displaystyle \frac{TP\times TN-FP\times FN}{\sqrt{(TP+FP)(TP+FN)(TN+FP)(TN+FN)}}};$$

(3)

$$\mathrm{A}\mathrm{C}\mathrm{C}={\displaystyle \frac{TP+TN}{TP+FP+TN+FN}}.$$

(4)

The following abbreviations are used in these equations:

• TP (true-positive) is the number of disease-causing variants correctly predicted to be pathogenic;
• FN (false-negative) is the number of disease-causing variants incorrectly predicted as tolerated;
• TN (true-negative) is the number of neutral variants correctly predicted as tolerated;
• FP (false-positive) is the number of neutral variants incorrectly predicted as pathogenic;
• MCC is a correlation coefficient between the observed and predicted binary classification, ranging from −1 (total disagreement between prediction and observation) to 1 (perfect prediction).

For the test dataset, we selected common neutral and P/LP variants from our comprehensive dataset (Supplementary Table S1). We also calculated the area under the ROC (Receiver Operating Characteristic) curve (AUC) using the pROC library in the R programming language. ROC curves were generated by plotting sensitivity against (1—specificity) at each threshold for each algorithm. The AUC can range from 0 (completely random) to 1 (perfectly correct prediction). The absence of a variant annotation negatively impacts prediction accuracy. Therefore, we included only those algorithms that predicted the pathogenicity of over 30% of variants in our dataset (Table S1).

3.5. Molecular Modeling

All computations were performed with the freely available ZMM program (www.zmmsift.ca). Energy was calculated using the AMBER force field [59] with environment- and distance-dependent dielectric function [60]. Energy was optimized with Monte Carlo (MC) minimizations [61] in the space of generalized coordinates [62]. We used the AlphaFold3 (AF3) server (https://alphafold.ebi.ac.uk) to predict multimeric complexes of hKv7.1 with four full-fledged KCNE1 subunits. CryoEM structures of Kv7.1 with the activated and deactivated VSDs, as well as AF3 models, were imported by the ZMM program, and side chain conformations were MC-minimized with rigid backbones. MC-minimizing trajectories were terminated when the last 100th energy minimization did not improve the protein energy. All structures were 3D-aligned by minimizing the root mean square deviations of C^α atoms in P1 helices against a reference crystal structure of the chimeric potassium channels Kv1.2-Kv2.1 channel (PDB ID: 2R9R), the first eukaryotic P-loop potassium channels whose crystal structure was obtained with a resolution below 2.5 Å [63].

Intersegmental contacts are defined as those where two sidechains are within 5Å from each other. Such contacts were automatically identified by ZMM in MC-minimized structures. We used PyMol v099 (Schrödinger, New York, NY, USA) to visualize cryoEM structures and AF3 models of the Kv7.1 channel and its complex with KCNE1. Other details of computations may be found elsewhere [64].

4. Conclusions

In this study, we compiled a comprehensive dataset encompassing known pathogenic/likely pathogenic (P/LP) variants of the hKv7.1 channel and its 14 paralogues. Our analysis identified AlphaMissense as the top-performing bioinformatics tool for predicting (P/LP) variants in Kv channels. AlphaMissense and the paralogue annotations method consensually predicted 79 VUSs of Kv7.1 as likely damaging (LD) variants. We further predicted that 34 KCNE1 VUSs are LD variants. Many wildtype residues (WTRs) of LD variants make state-dependent intersegmental contacts with WTRs of known P/LP variants or LD variants in cryo-EM structures or AlphaFold3 models, suggesting atomic mechanisms of the variants’ dysfunction. Respective variants of Kv7.1 and KCNE1 are promising objects for future functional analyses.

Study Limitations

Our computations predicted the damaging potential of many reported disease-associated mutations in KCNQ1 and KCNE1, but functional studies are necessary to reclassify respective VUSs as P/LP variants. Since the cryoEM structures of Kv7.1 considered here do not provide relations between VSD activation and the pore opening, mutations with likely damaging potential within the VSD do not necessarily affect the I_Ks current; functional studies are necessary to explore these variants. We analyzed contacts of WTRs but not respective VUSs and suggested that mutations would modify these contacts. Despite these limitations, we hope that our study, which incorporates paralogue annotations and analysis of experimental and AF3-predicted 3D structures, provides more reliable predictions of the damaging potential of VUSs than more traditional bioinformatics approaches.