Likely Pathogenicity of Uncharacterized KCNQ1 and KCNE1 Variants

1. Introduction

Voltage-gated potassium channels Kv7.1 encoded by gene KCNQ1 play important roles in physiology and pathophysiology of the heart (Wu and Larsson, 2020; Kekenes-Huskey et al., 2022). Homotetrameric Kv7.1 associated with different beta subunits, which are encoded by KCNE genes, is expressed in various organs including colon, kidney, stomach, inner ear and testis, for reviews see (Jespersen et al., 2005; Abbott, 2014). In the heart, hKv7.1 permeates slow-delayed rectifier potassium current I_Ks, which is essential the normal cardiac rhythm (Sanguinetti and Seebohm, 2021). Mutations in the KCNQ1 gene are associated with several heart diseases (Abbott, 2014), including long QT syndrome (Sun and MacKinnon, 2017; Bains et al., 2024), short QT syndrome (Bellocq et al., 2004; Rothenberg et al., 2016), atrial fibrillation (Chen et al., 2003; Hateley et al., 2021) and sudden cardiac death (Albert et al., 2010; Campuzano et al., 2015). Recently, Kv7.1 was found to mediate sex-dependent cold sensation (Kiper et al., 2024).

Each Kv7.1 subunit has six transmembrane (TM) segments. Segments S1 to S4 form the voltage-sensing domain (VSD). Segments S5 and S6 with 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 that regulates gating (Sachyani et al., 2014). KCNE1 is the primary accessory subunit for the cardiac channel Kv7.1 (Wrobel et al., 2012).

As of March 6, 2025, the ClinVar database lists about 1,000 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 (Huang et al., 2018) and short QT interval syndrome (Bellocq et al., 2004). 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 (Phul et al., 2022). Functional studies, including high throughput electrophysiology were employed to reclassify many Kv7.1 VUSs as P/LP variants (Vanoye et al., 2018) and reveal molecular mechanisms underlying pathogenicity (Brewer et al., 2025). Nevertheless, large number of still uncharacterized VUSs motivates application of computational methods to predict their pathogenicity.

The American College of Medical Genetics and Genomics and the Association for Molecular Pathology recommend using in silico predictive algorithms for interpretation of variants. Various bioinformatics tools based on different principles have been developed to predict pathogenicity and tolerance of genetic variants, with success rates ranging between 60 and 80%. The performance of in silico tools may depend on the disease phenotype. For instance, MetaLR, MetaSVM, and MCap have shown top performance in predicting pathogenicity for variants associated with cardiovascular abnormalities. However, some methods yielded many false-positive and false-negative predictions of pathogenicity in individual protein families. Thus, MetaSVM predicted a pathogenic effect for 75% of benign variants of the cardiac sodium channel Nav1.5. 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 (Walsh et al., 2014) to reclassify as P/LP variants numerous VUSs of cardiac sodium channel Nav1.5 (Tarnovskaya et al., 2020), calcium channel Cav1.2 (Tarnovskaya et al., 2021) and TRPM4 channel (Tarnovskaya et al., 2023). The paralogue annotation method employs a 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 P/LP variant.

In this study, we collected 1,750 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 pathogenicity of 79 variants. We further used another tool, ClinPred in combination with paralog annotations to predict 34 VUSs of 21 WTRs in KCNE1. In available cryoEM structures of Kv7.1 with activated and deactivated voltage sensing domains and in the AlphaFold model of KCNE1-bound Kv7.1, 52 WTRs of LP-reclassified variants form intersegment contacts with ClinVar-listed WTRs, including 21 WTRs with P/LP variants. Analysis of state-dependent contacts in the cryoEM structures and AlphaFold model suggests atomic mechanism of dysfunction for some variants.

2. Results and Discussions

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

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 (Tikhonov et al., 2025). 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 (Gigolaev et al., 2025). 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 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 contain P/LP variants that are used in the paralog annotation method. For KCNQ1 and its 14 paralogues (Table 1), we collected a total of 5,632 missense variants from the gnomAD, ClinVar, Ensembl, and Humsavar databases (Table S1). These include 1,059 P/LP variants, 691 common neutral variants (with AF > 0.00001), and 3,882 uncharacterized variants or VUSs. We refer to this dataset as the “broad dataset.” The largest numbers of P/LP variants (394) were 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 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, which predicted pathogenicity for > 70% variants in our dataset. Thus, we excluded from our analysis algorithms EVE, MutPred, and MutationTaster. To compare performance of 26 deleteriousness prediction scores (Table 2), we compiled a test set with 691 true positive (TP) and 1,059 true negative (TN) observations obtained from our broad dataset (Table S1). Pre-computed algorithm scores were retrieved from 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 Paralogues Annotations Consensually Predicted Pathogenicity of 79 VUS

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 pathogenicity 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 paralogs of hKv7.1 (Table 1) with a conservation score across paralogs Cs > 0.3. Both methods consensually predicted 79 P/LP variants (Table 3). Among these only 6 variants are functionally characterized by high-throughput functional analysis (Vanoye et al., 2018). Five of these have a very strong decrease of the current density and one shows a moderate decrease (Table 3). These experimental data confirm the predictive power of our approach. Below we describe intersegment contacts of WTRs for some LP-reclassified VUSs that may guide their selection for functional studies.

2.7. Intersegment Contacts Involving WTRs with LP-Reclassified VUSs

Above bioinformatics approaches strongly suggest likely pathogenicity of 79 VUSs in Kv7.1, but molecular mechanisms of the variants' dysfunction are unclear. Analysis of intersegments 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 (Mandala and MacKinnon, 2023) as described in Methods. Upon a VSD deactivation, helix S4 undergoes substantial downshift (Figure 2A) and rotation (Figure 2B), while rearrangements of helices S5 and S6 are 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 (Mandala and MacKinnon, 2023). 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.

Co-expression of Kv7.1 with KCNE3 yields constitutively open channels by promoting VSD activation (Barro-Soria et al., 2015; Kasuya and Nakajo, 2022). While VSDs in cryoEM structure (6v01) of KCNE3 bound Kv7.1 are deactivated (Sun and MacKinnon, 2017), 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 intersegment contacts of WTRs with LP-reclassified VUSs and, to some extent, their state-dependency.

We visualized intersegment contacts in VSDs involving WTRs of LP-reclassified 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 is more 3D conserved than VSDs, we analyzed intersegment contacts in the pore domain in structure with deactivated VSDs (Figure 4, Figure 5 and Figure 6), and then in structure with 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 LP-reclassified 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 case of cardiac channel Nav1.5, for many intersegment state-dependent contacts involving WTRs with ClinVar reported variants, substitution of either contact partner causes the channel dysfunction associated with the same syndrome (Korkosh et al., 2021). By analogy, we propose that if two WTRs of Kv7.1 form an intersegment contact with one partner reported in ClinVar with P/LP variant and another partner has a LP-reclassified VUS, it strongly increases the reliability of our VUSs reclassification as LP variants.

Such contacts with known P/LP residues involve 31 LP-reclassified 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. Figs. 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 probability of the channel open state, shorten 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 activated state of VSD stronger than its deactivated state.

Table 4 also shows seven LP-reclassified VUSs in the cytoplasmic part of S6 (positions 6.553 - 6.574) and six LP-reclassified VUSs in the cytoplasmic part of S5 (positions 5.533 -5.540). These residues undergo substantial movements upon the activation gating implying that respective disease mutations destabilize the channel open state. Importantly, WTRs of 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 LP-reclassified VUSs and 25 partner WTRs of 35 ClinVar reported VUSs (Table 5). Among the contact partners, 13 WTRs have 18 VUSs that we also reclassified as LP variants (Table 4). The fact that VUSs of both partners in such contacts were independently reclassified as LP variants by AlphaMissense and paralogue annotation methods further justifies LP reclassifications for respective VUSs making them promising objects for future functional analyses.

The third category of contacts includes 15 WTRs of 20 LP-reclassified 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 respective WTR is contacting WTR with LP-reclassified VUS.

2.8. Re-Classifying 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 (Wrobel et al., 2012; Sanguinetti and Seebohm, 2021). 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), only KCNE2 contains 5 P/LP variants, while for other paralogues only VUSs are reported. Therefore, the paralogues annotation method for KCNE subunits is less reliable than for Kv channels. Nevertheless, ClinPred and the paralogues annotations consensually predicted likely pathogenicity of 34 VUSs for 21 WTRs in KCNE1 (Table 8). Among the 34 VUSs, 20 variants were explored in a recent high-throughput functional study of 68 KCNE1 variants (Vanoye et al., 2025), and for most of these the pick current decrease in the KCNQ1-KCNE1 assembly is reported (Table 8). These experimental data confirm reliability of our bioinformatics approach to predict likely pathogenicity of KNCE1 variant.

2.9. AF3 Model of Kv7.1 with KCNE1

In lack of cryoEM structures of Kv7.1-KCNE1, we created an AF3 model of complex Kv7.1-KCNE1 and MC-minimized its energy with rigid backbones (Figure 7). In this model, the large N-terminal portion of a single 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. KCNE1 VUSs reclassifies as LP variants (Table 8) are shown as sticks in close-up views of Figure 7. Among multiple Kv7.1 WTRs that form contacts with KCNE1, six are known P/LP variants and several others are LP-reclassified 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 next section.

2.10. CryoEM Structure of KCNE3-Bound Kv7.1

Kv7.1 co-assembles with KCNE3 in non-excitable cells (Abbott, 2016; Sun and MacKinnon, 2020), but there are evidence that such complexes are also expressed in human heart, especially in disease heart (Lundquist et al., 2005). 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 (Sun and MacKinnon, 2020) where the TM helix of KCNE3 is bound between S1 of VSD and extracellular parts of helices S5 and S6 (Figure 8B). In 3D-aligned cryoEM structure of Kv7.1-KCNE3 and AF3 model of Kv7.1-KCNE1, the TM part of KCNE3 helix (M⁶⁰ to T⁸⁰) overlaps with 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 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 position of respective KCNE1 TM helix in our AF3 model (Figure 7).

3. Methods

3.1. Sequence Data of Human Channels and Collection of Variants

The sequence of hKv7.1 was obtained from the UniProt database (UniProt, 2015) (accession number P51787). Paralogues of the hKv7.1 channel were identified using Ensembl (Dyer et al., 2025). Missense mutations for Kv7.1 and its paralogues were collected from three databases: Humsavar, Ensembl Variation (Boutet et al., 2016), and ClinVar (Landrum et al., 2016). 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 (Karczewski et al., 2020). Variants with AF > 0.00001 that are not present in ClinVar were considered benign (Walsh et al., 2017; Kaltman et al., 2018). 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 (Haitin et al., 2009).

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 (Walsh et al., 2014). Previously, we applied a modified version of this method to predict P/LP variants for channels hNav1.5 (Tarnovskaya et al., 2020), hCav1.2 (Tarnovskaya et al., 2021), and TRPM4 (Tarnovskaya et al., 2023). Here we used the same approach to predict P/LP 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 (Sievers et al., 2011). Proteins lacking P/LP variants were excluded from the alignment (Niroula and Vihinen, 2016). 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 (Livingstone and Barton, 1993). Cs values were calculated using the Zvelebil method (Zvelebil et al., 1987), as implemented in the Amino Acid Conservation Calculation Service (Golicz et al., 2018). Variants occurring at positions with Cs > 0.3 were classified as P/LP variants.

3.4. Sequence-Based Prediction of Pathogenicity

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 (Liu et al., 2020). 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:

$$\mathrm{Sensitivity}={\displaystyle \frac{TP}{TP+FN}};$$

(1)

$$\mathrm{Specificity}={\displaystyle \frac{TN}{TN+FP}};$$

(2)

$$\mathrm{MCC}={\displaystyle \frac{TP\times TN-FP\times FN}{\sqrt{\left(TP+FP\right)\left(TP+FN\right)\left(TN+FP\right)\left(TN+FN\right)}}};$$

(3)

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

(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 (Weiner et al., 1986) with environment- and distance-dependent dielectric function (Garden and Zhorov, 2010). Energy was optimized with Monte Carlo (MC) energy minimizations (Li and Scheraga, 1987) in the space of generalized coordinates (Zhorov, 1981). We used AlphaFold3 (AF3) server (https://alphafold.ebi.ac.uk) to predict complexes of hKv7.1 with the full-fledged KCNE1 subunit. 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. MCM trajectories were terminated when the last 100^th energy minimization did not improve the protein energy. All structures were 3D-aligned by minimizing the root mean square deviations of С^α atoms in P1 helices against a reference crystal structures 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 the resolution below 2.5 Å (Long et al., 2007).

Intersegment contracts 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 (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 (Zhorov, 2021).

4. Conclusions

In this study, we compiled a comprehensive dataset encompassing known pathogenic and likely pathogenic variants in the hKv7.1 channel and its 14 paralogues. Our analysis identified AlphaMissense as the top-performing bioinformatics tool for predicting likely pathogenic/pathogenic (LP/P) variants in Kv channels. Using AlphaMissense and the paralogue annotation method, we consensually predicted 79 VUSs of Kv7.1 as LP variants. We also used bioinformatics to predict LP status for 34 KCNE1 VUSs. Many wild type residues (WTRs) of LP-reclassified variants make state-dependent intersegments contacts with WTRs of known P/LP variants or reclassified LP variants in cryo-EM structures or AlphaFold-3 models, suggesting atomic mechanisms of the variants' dysfunction. Respective variants of Kv7.1 and KCNR1 are promising objects for future functional analyses.