High-throughput discovery of trafficking-deficient variants in the cardiac potassium channel K_V11.1

Abstract

Background:

KCHN2 encodes the K_V11.1 potassium channel responsible for I_Kr, a major repolarization current during the cardiomyocyte action potential. Variants in KCNH2 that lead to decreased I_Kr have been associated with Type 2 Long QT syndrome (LQT2). The mechanism of LQT2 is most often induced loss of K_V11.1 trafficking to the cell surface. Accurately discriminating between variants with normal and abnormal trafficking would aid in understanding the deleterious nature of these variants; however, the volume of reported nonsynonymous KCNH2 variants precludes the use of conventional methods for functional study.

Objective:

We report a high-throughput, multiplexed screening method for KCNH2 genetic variants capable of measuring the cell surface abundance of hundreds of missense variants in the resulting K_V11.1 channel.

Methods:

We developed a method to quantitate K_V11.1 variant trafficking on a pilot region of 11 residues in the S5 helix.

Results:

We generated trafficking scores for 220/231 missense variants in the pilot region. For 5/5 variants, high-throughput trafficking scores validated when tested in single variant flow cytometry and confocal microscopy experiments. We further explored these results with planar patch electrophysiology and found that loss-of-trafficking variants indeed do not produce I_Kr. Conversely, but expectedly, some variants that traffic normally were still functionally compromised.

Conclusions:

Here, we described a new method for detecting K_V11.1 trafficking-deficient variants in a multiplexed assay. This new method accurately generated trafficking data for variants in K_V11.1 and is extendable both to all residues in Kv11.1 and to other cell surface proteins.

INTRODUCTION

KCNH2, also known as the human Ether-à-go-go-Related Gene (hERG), encodes the rapid component of the delayed inward-rectifying, voltage-gated, cardiac potassium channel K_V11.1. This channel produces a major repolarizing current, I_Kr, in cardiomyocytes. K_V11.1 loss-of-function variants are the second most common cause of congenital long QT syndrome (LQTS2; MIM: #613688, Figure 1A)¹ whilst K_V11.1 gain-of-function variants are the most common cause of short QT syndrome (SQTS; MIM: #609620).² LQTS2 is characterized by a prolonged QT interval on the electrocardiogram, and SQTS, a rarer disease, is characterized by a short QT interval. Both LQTS2 and SQTS predispose individuals to syncope and sudden death.^{2, 3} To date, more than 1,000 unique non-synonymous KCNH2 variants have been discovered in patients presenting with arrhythmias and in public databases of mostly unaffected individuals.^{1, 4} With the onset of large population sequencing projects and increase in clinical genetic testing, the number of observed KCNH2 variants is rapidly growing, much faster than the detailed characterization of these variants.⁹ An important challenge for KCNH2 variant annotation is the identification and characterization of potentially disease-causing KCNH2 variants found in individuals with or without a clinical phenotype.

Figure 1: K_V11.1 trafficking deficiency and Long QT Syndrome.

A) KCNH2 variants can cause Long QT Syndrome Type 2 (LQTS2) by reducing I_Kr inward potassium current. LQTS2 manifests as a prolonged action potential in cardiomyocytes and a prolonged QT interval on the electrocardiogram. B) Schematic of K_V11.1 trafficking to the plasma membrane. A common mechanism of LQTS2 is reduced trafficking of K_V11.1 to the cell surface. C) For missense variants, trafficking-deficiency is the dominant mechanism of LQTS2 (178/210 variants; 85%)^8–10. LQTS2 variants that have been assessed for cell surface abundance are labeled in red (trafficking-deficient; <50% of wildtype) or white (trafficking-normal; 50–150% of wildtype). A Short QT Syndrome-associated variant, N588K, is labeled in blue; this variant has increased cell surface abundance (>150%). PAS indicates the approximate location of the PAS (Per-Arnt-Sim) domain thought to play a role in cell surface expression, cNBD indicates the Cyclic Nucleotide Binding Domain.

The most common mechanism by which pathogenic variants cause protein dysfunction is destabilization-induced misfolding.^5–7 K_V11.1 is a 1,159 residue protein with six membrane-spanning helices that forms a homotetrameric voltage-gated potassium channel at the plasma membrane surface in cardiomyocytes. For K_V11.1, misfolding most commonly causes loss of trafficking to the plasma membrane, resulting in fewer functional channels at the cell surface (Figure 1B).^8–10 178/211 (84%) K_V11.1 variants associated with LQT2 whose cell surface abundance has been assayed have defective trafficking to the cell surface (Figure 1C).^{8, 11}

Here, we develop an assay which combines the emerging technology of saturation mutagenesis with a high-throughput flow-cytometry-based trafficking assay. Our approach builds on previously developed methods of deep mutational scanning (DMS)^{12, 13} and cell surface abundance assays^{8, 11, 14, 15} to create a high-throughput cell surface trafficking assay. Although our method measures the same cell surface abundance of K_V11.1 protein as previously established Western blot-based and flow cytometry methods,¹⁵ the major advancement here is the ability to readily scale to all 22,021 (1159 × 19) missense substitutions in K_V11.1. We can now collect the necessary data to evaluate trafficking which would take years with traditional western blot assays. This provides complementary information to in silico predictions^{12, 16} at a scale currently inaccessible to other functional characterization techniques, such as automated patch clamp electrophysiology¹⁷ and CRISPR-generated, or naturally-occurring, variants in induced pluripotent stem cell-derived cardiomyocytes.¹⁸

METHODS

Methods for KCNH2 mutagenesis, generation of stable cell lines, sequencing of libraries, flow cytometry, confocal microscopy, patch clamping, and corresponding analyses are described in detail in the Supplemental Material.

RESULTS

We modified a flow cytometry-based cell-surface trafficking assay¹⁵ for compatibility with a DMS of KCNH2 (Figure S1). To accomplish this, we inserted an hemagglutinin (HA) tag¹⁹ into the first extracellular loop of K_V11.1, overexpressed this plasmid in HEK293T cells, and isolated cells using flow cytometry.²⁰ Control variants of KCNH2 were generated within the HA-KCNH2 construct, and trafficking characteristics were validated with flow cytometry and confocal microscopy (Figure 2A). Wildtype HA-KCNH2 yielded a high K_V11.1 surface abundance, detected by both flow cytometry and confocal microscopy (Figure 2B). In contrast, G601S and A561V variants, which have previously been described to have mistrafficking-related loss-of-function defects,^{21, 22} had reduced cell surface abundance using both methods (Figure 2B). Additionally, the G601S variant, but not A561V, could be rescued by incubation with E-4031, as previously observed (Figure S2).¹⁰

Figure 2: K_V11.1 cell-surface trafficking assay.

A) Diagram of antibody labeling of surface or both surface and internal K_V11.1. Live cells are stained with Alexa 647-labeled anti-HA antibody, which only labels surface K_V11.1. Cells are assayed by flow cytometry. For confocal imaging, cells are fixed, permeabilized, and stained with an Alexa 488-labeled anti-HA antibody, which labels internal K_V11.1. B) Flow cytometry assay of HEK293T cells expressing wildtype K_V11.1, G601S (partial loss-of-function), or A561V (near total loss-of-function). The x-axis indicates mCherry level, which is a marker of protein expression. C) Confocal imaging of HEK293T cells expressing wildtype, G601S, or A561V K_V11.1.

Deep mutational scan of an 11-amino-acid region of K_V11.1

We next generated a barcoded plasmid library of nearly every possible variant in an 11-amino-acid region (residues 545–555) of KCNH2 using PCR with degenerate primers²³ in a cDNA construct broken into tiles (Table S1, Figure S3). The 11-amino-acid region covers a segment stretching from outside the membrane on the S4-S5 linker to within the membrane on the S5 helix. This segment was chosen for its location, which changes from adjacent to within the membrane, and structural diversity, which varies between flexible and helical. After subassembly²⁴, 14,466 unique barcodes were detected, which included 220/231 possible missense variants (Figure S4). The 11-amino-acid experiment successfully distinguished between two control classes of variants: nonsense variants that were predicted to be undetectable at the cell surface and synonymous variants that were predicted to have wildtype channel abundance at the cell surface (Figure 3A). Initial trafficking scores were calculated from a weighted average of each variant’s abundance in the four sorted pools (see Supplemental Materials for details). These scores were averaged across two independently transfected and sorted replicate experiments. Nonsense variants had an average score of 5 (0 was the lowest possible score) and synonymous variants had an average score of 101 (100 is the score for wildtype; Figure 3). The missense variant scores ranged from loss-of-function to wildtype-like; the mean score for missense variants was 55 (Figure 3A and B). Trafficking scores were highly consistent across the two replicate experiments (Spearman’s ρ = 0.77). Seven variants were unable to be analyzed due to lack of barcoding during the generation of the KCNH2 plasmid library (grey boxes in Figure 3B). Of note, methionine is encoded by only one codon and therefore lack synonymous variants.

Figure 3: Deep mutational scan identifies trafficking-defective variants.

A) Trafficking scores for nonsense, synonymous, and missense K_V11.1 variants in the pilot region from Y545 to C555. Synonymous and nonsense variants were well-separated, with a mean score of 5 and 101, respectively. Missense variants had a range of scores from near-wildtype to loss-of-function, with a mean score of 55. B) A heatmap of the trafficking scores from the pilot region also displayed in A). K_V11.1 residue number and identity are indicated in the leftmost column and amino-acid substitution is indicated on the top row. The black dot indicates synonymous variants. Color of the box indicates trafficking score compared to wildtype with blue as low expression to the plasma membrane and red as increased expression at the plasma membrane. Gray boxes indicate variants were absent in either the barcode libraries or trafficking results and therefore a trafficking score could not be determined. Out of the 220 variants assayed, 86 were WT-like (score > 75), 57 were mild loss-of-trafficking (75 > score > 50), 48 were loss-of-trafficking (50 > score > 25), and 73 were severe loss-of-trafficking (score < 25). Note the low trafficking scores of charged or polar amino-acids closer to the middle of the membrane (residues 550–555) (see Figure 4).

Trafficking scores and K_V11.1 topology

We explored the structural basis of the trafficking scores with a three-dimensional structure of K_V11.1 (Figure 4, PDB: 5VA1).²⁵ Residues Y545-V549 tolerated substitutions more than residues L550-C555, especially secondary structure-disruptive substitutions (glycine and proline) and hydrophilic substitutions (Figure 3B). These observations are consistent with the topology and position within the membrane of the pilot region: residues 545–549 join the S4 and S5 helices on the intracellular membrane leaflet, a segment transitioning from random coil to α-helical secondary structure (Figure 4). Missense substitutions at residues 545–549 were less likely on average to induce mistrafficking compared to residues 550–555, a mean trafficking score of 70.2 versus 42.2, respectively. In contrast, residues 550–555, closer to the membrane interior, were less tolerant to any substitution, especially polar and charged residues. The trafficking scores correlated modestly with computational predictions, with a Spearman’s ρ of 0.31, 0.21, and 0.42 with PROVEAN,²⁶ PolyPhen-2,²⁷ and REVEL,²⁸ respectively (Table S3).

Figure 4. Polar and charged residues close to the middle of the membrane cause trafficking-deficiency.

CryoEM structure of K_V11.1 (PDB: 5VA1)⁴¹ with residues mutagenized in a pilot study of a high-throughput trafficking assay highlighted as red or blue spheres. These residues are on the intracellular half of the S5 transmembrane helix. The red to blue transition from V549 to L550 reflects the transition from residues tolerant of hydrophilic substitution to residues intolerant of hydrophilic substitution, respectively, as observed in the high-throughput trafficking assay (Figure 3B).

Validation of trafficking assay scores

To validate our trafficking scores, we selected five variants for further characterization. These variants were selected based on their unintuitive DMS trafficking results: L553N and A547D are in regions intolerant of hydrophilic substitution yet had a trafficking score like wildtype (scores of 108 and 99, respectively); M554A is in a region tolerant of hydrophobic substitution yet does not traffic (score of 4) yet M554W is a disruptive substitution and traffics like wildtype (score of 99); F551Y is a relatively conservative substitution yet is not tolerated (score of 2). In single variant validation experiments, these five variants had confocal and flow cytometry cell surface abundance concordant with their DMS scores (Figure 5 and Table 1).

Figure 5: Validation of trafficking scores.

Validation of selected variants by analytical flow cytometry (left) and confocal microscopy (right). In the leftmost column, DMS score is indicated directly below variant name. X-axis is mCherry fluorescence (indicating K_V11.1 expression), the y-axis is surface stained K_V11.1. The remaining columns are extracellularly stained K_V11.1 (AF647, purple) intracellularly stained K_V11.1 (AF488, green), and both merged with DAPI stained nucleus (blue).

Table 1:

Trafficking score validation

Variant	DMS score	Flow/Confocal validation	Patch clamp
A547D	99	WT-like	LOF
L553N	108	WT-like	Fast deactivation
M554A	4	No trafficking	LOF
M554W	99	WT-like	LOF
F551Y	2	No trafficking	--

PROVEAN²⁶ and PolyPhen-2²⁷ predict all variants listed are deleterious (Table S3). REVEL²⁸ score is only available for F551Y (only variant possible with single nucleotide variation), which is predicted to be neutral.

In addition, we examined four variants in the studied region that were previously observed in literature reports or in gnomAD.⁴ L552S was previously reported as trafficking deficient variant⁸ and has been observed in 90 cases of Long QT syndrome in the literature, 28 unaffected individuals, and 20 alleles in the gnomAD database. Our DMS score for L552S was 25, consistent with the strong association with LQTS2 and loss of trafficking phenotype as previously reported.⁸ Three other variants, A547T, V549M, and F551L, were not observed in any literature cases of Long QT Syndrome and have been observed in 1–2 cases in gnomAD, too few to assess relevance from patient data alone. All of these variants had normal trafficking scores (Table S2). Thus, the DMS trafficking scores are consistent with the available data for variants in this region.

To further assess the trafficking results, we collected planar patch clamp data on four of the five validation variants. M554A also had a total loss of I_Kr, consistent with its loss of trafficking DMS score (score of 4) and its loss of trafficking in validation experiments. Surprisingly, the other three validation variants, which had detectable trafficking by DMS and in validation experiments, also had loss-of-function defects by patch clamp (Figure 6). Despite being present at the cell surface (Figure 5), M554W and A547D did not produce measurable I_Kr, while L553N had detectable I_Kr, but with lowered peak tail current and faster deactivation rate, essentially producing a loss-of-function phenotype (Figure 6). Therefore, the high-throughput trafficking scores were successful at identifying loss-of-function variants that acted through a trafficking mechanism, but not by other mechanisms, unlike PROVEAN, PolyPhen-2, and REVEL, which predicted all variants in the validation set to be deleterious.

Figure 6: Functional characterization of L553N.

Ai) All channels were fully activated at +40mV for 1s before stepping to −120mV for 3s to allow for recovery from inactivation and measure current density. The peak tail current traces for WT (black) and L553N (grey) are shown for the region highlighted in the voltage protocol. Aii) Violin plot of peak tail current amplitudes normalized to cell capacitance. Bi) Typical families of current traces recorded during Isochronal activation protocols for WT (black) and L553N (grey). Bii) Violin plot of V_0.5 values derived by fitting Boltzmann function to the tail current amplitudes (see Supplementary Methods for details). Ci) The current traces for WT (black) and L553N (grey) were shown for the region highlighted in the voltage protocol. Cii) Violin plot of the weighted time constant (T_weighted) for channel deactivation at −50mV, (see methods for detail of calculation of T_weighted). In all violin plots, median and quartiles are shown as solid and dotted lines, respectively. The reported two-tailed P-values were derived from Mann-Whitney tests. D) Structural context of L553 (on helix S5), which points between residues V657 and N658 on the S6 helix within the same subunit.

DISCUSSION

A high-throughput assay for identifying KCNH2 variant trafficking

We developed a scalable, multiplexed flow cytometry and high-throughput sequencing assay to measure the cell surface abundance of hundreds (and potentially thousands) of variants in KCNH2. We generated trafficking scores for 220/231 possible missense variants in an 11-amino-acid region of KCNH2. We validated these scores in several ways. First, the scores correlated very well (Spearman ρ = 0.77) across the two replicate experiments. Second, the scores distinguished nonsense from synonymous variants (Figure 3). Finally, 5/5 missense variants with a range of surprising, unintuitive trafficking scores were validated by flow cytometry and confocal immunocytochemistry on single variants (Figure 5). Therefore, we believe these scores accurately capture cell surface abundance, which is the mechanism for ~80% of all long QT variants studied to date.^8–10

Improved understanding of KCNH2 trafficking deficiency

From our trafficking data, we could observe two segments in the 11-residue pilot with distinct substitution tolerances. In the segment comprised of residues 545–549, hydrophilic and secondary structure-disrupting substitutions were tolerated. However, in the segment comprised of residues 550–555, hydrophilic substitutions were more likely to lead to compromised trafficking. The tolerance of hydrophobic or hydrophilic substitution reflect the observed topology of the K_V11.1 S5 helix⁴¹ (Figures 3B and 4). Our trafficking data have a Spearman’s ρ of 0.31, 0.21, and 0.42 with PROVEAN, PolyPhen-2, and REVEL respectively, indicating our DMS scores provide distinct information to that available in in silico these predictive models.

In addition, we unexpectedly found L553N trafficked like wildtype and produced measurable peak tail current. However, this variant had a dramatically faster deactivation time (the time it takes the channel to stop conducting after the cell is repolarized; Figure 6). L553 points in between residues N658 and V659 (both known to accelerate deactivation) in the S6 helix within the same subunit.²⁹ Given our result, we speculate the asparagine (N) substitution at residue 553 stabilizes an interaction between N658 and L553N resulting in a more stable deactivated state which allows this variant to escape trafficking quality control but does not allow the channel to function normally (Figure 6D).

Approaches to studying KCNH2 variants

This work represents a new tool for studying KCNH2 variants in high-throughput. Several studies have previously quantified trafficking of K_V11.1, typically using assays that rely on Western blotting or confocal imaging, which are not scalable to the thousands of possible missense variants within the protein.^{10, 15, 30} Recent advances in automated patch clamp electrophysiology have enabled the rapid characterization of dozens of voltage-gated potassium channel variants,^{17, 31} however, the throughput of these methods does not allow for the rapid screening of the thousands of possible variants in KCNH2. In contrast, computational predictions like Polyphen2, PROVEAN, or REVEL can be easily generated for every variant in KCNH2. However, while these computational predictions can partially predict disease risk, they are imperfect and often cannot capture the variant-induced defect (Table S3). Lastly, this method may also be used to investigate variant-specific and therapy-specific trafficking rescue for a multitude of variants per experiment (Figure S3).

Improving the prediction of KCNH2 variant pathogenicity

Our assay has the potential to sample all possible amino acid substitutions in KCNH2 in an unbiased manner. This will provide a database for patients and clinicians to identify the effect a previously uncharacterized mutation may have on patient disease propensity. Future work will expand this method to much larger regions of KCNH2, with a focus on regions enriched for known long QT variants (Figure 1). However, our electrophysiological results highlight the need to approach trafficking data with caution. A547D and M554W trafficked normally in the high-throughput assay and in single variant validation experiments, yet did not produce I_Kr currents in patch clamp experiments. This result indicates all variants with normal trafficking scores need to be interpreted with caution, as they may still cause loss of function through other mechanisms.

Limitations

Here, we described a method to investigate whether a missense variant in KCNH2 affects the steady-state concentration of mutant K_V11.1 channels at the plasma membrane. The trafficking score derived from this method will indicate if the variant channel traffics normally; however, we cannot assess whether a low-trafficking variant will have a dominant negative effect or whether wildtype-trafficking variants will produce wildtype I_Kr. The assay described was developed for the dominant isoform in HEK293T cells and therefore does not account for the expression of additional KCNH2 splice variants, common variants (e.g. T897 and L1047), and other subunits known to contribute to physiologically relevant I_Kr in cardiomyocytes.^32–34 To better understand the implications of the DMS trafficking scores in the clinical context, future work is needed to investigate K_V11.1 variant trafficking in these contexts.

Conclusion

We have developed a high-throughput trafficking assay to characterize cell surface expression of KCNH2 variants in a massively parallel fashion. The vast majority of KCNH2 variants only have limited available patient data and most are currently annotated as variants of uncertain significance. The method we put forward here enables the collection of in vitro experimental variant defect data at a scale commensurate with the scale of new KCNH2 variants being discovered. Lastly, this method can be expanded to other ion channels and transmembrane proteins where altered cell surface abundance is a major mechanism of disease pathogenesis.