Prognostic Value of Multiplexed Assays of Variant Effect and Automated Patch-clamping for KCNH2-LQTS Risk Stratification (2024)

The complete version history of this preprint is available at medRxiv.

Associated Data

Supplementary Materials

Data Availability Statement

Abstract

Graphical Abstract.

Congenital Long QT Syndrome (LQTS) is a frequently fatal inherited arrhythmia syndrome with strong monogenic contributions¹. Rare, loss-of-function variants in KCNH2, which encodes for the cardiac repolarizing potassium ion channel current I_Kr, are implicated in approximately 30% of Long QT syndrome cases (LQT2)². Integrating genetic data into clinical practice is hindered by complex variant interpretation, incomplete penetrance, and environmental interactions. Variant-specific information, including functional data, can assist with variant classification and diagnosis^3,4; however, whether variant-specific information can assist with prognosis is unknown. Here, we implemented two high-throughput functional assays to study the KCNH2-LQT2 relationship: 1) Multiplexed Assays of Variant Effect (MAVE)^5,6, and 2) Automated Patch Clamping (APC)⁷. The MAVE probed the effect on protein trafficking of nearly all possible KCNH2 missense variants, providing prospective evidence for unascertained variant future classification⁸. APC recorded detailed effects on variant protein electrophysiological function after clinical ascertainment. To evaluate the significance of these and other variant-specific attributes in stratifying the risk of LQTS events, we recorded QTc, survival data, and cardiac events in 1150 patients heterozygous for KCNH2 missense or in-frame insertion/deletion variants.

We quantified the trafficking effect of 18,323 total variants by MAVE (Figure 1A; Supplemental Table 1)^5,6,9. We observed excellent stratification of WT normalized variant trafficking scores across synonymous (mean 100.4±23.9%), missense (mean 71.4±46.4%), and nonsense (mean 17.9±36.0%) variant classes (Figure 1B/​/C;C; Supplemental Table 2). Consistent with a previous study, nonsense variation had minimal effect on trafficking function downstream of residue 863 (Supplemental Figure 1)¹⁰. In parallel, we used the SyncroPatch 384 PE platform to comprehensively interrogate potassium peak tail currents among KCNH2 variants observed in our clinical cohorts, the literature, and gnomAD¹¹ (Figure 1D; Supplemental Methods). 266/506 variants had severe loss of function (Z < −4, Figure 1E, Supplemental Table 3) with good concordance between MAVE and APC datasets, including across “hot spot domains” (Figure 1F–G; Supplemental Figure 2). We derived ClinGen-recommended assay calibration and Z-score thresholds using the same set of variant controls for the MAVE dataset as our previously described KCNH2 APC assay^12,13 (Supplemental Figure 3 and Supplemental Table 4). This calibration showed that the MAVE data may prospectively apply strong pathogenic functional criteria and moderate benign functional criteria. Most variants studied in both assays were concordantly annotated with an abnormal threshold of an assay Z-score less than −2 (365/420; Figures 1F and Supplemental Figure 4). 23 variants had normal trafficking but abnormal current density, suggesting abnormalities in gating or ion permeability.

We next collected clinical data on 1150 patients recruited from five tertiary arrhythmia clinics in Japan (N=289), Italy (N=275), the USA (N=261), France (N=259), and New Zealand (N=66) heterozygous for 317 unique KCNH2 missense or in-frame insertion/deletion variants. LQTS was diagnosed based on Schwartz score >3.5 or repeated Corrected QT intervals (QTc) >480 ms without secondary causes (see methods for full details)¹⁴. Events were either ventricular tachycardia, ventricular fibrillation, sudden cardiac death, appropriate ICD shocks, and/or syncope; severe events did not include syncope unless while on beta blockers. QTc intervals were higher among probands than family members (N = 503/1150; mean 504±55 vs 472±69 ms, respectively; Figure 2A), females than males (N = 606/1150; mean 494±52 vs 477±77 ms; Figure 2B), and individuals experiencing cardiac events (N = 323/1150, mean 515±59 vs 474±64 ms, respectively; Figure 2C).

Clinical risk stratification of LQTS-cardiac events currently relies on patient-specific features such as sex and QTc¹⁵. We hypothesized that including variant-specific data (e.g., variant functional, in silico, and structural properties) alongside patient-specific data could improve prognosis of adverse events. In a risk model using patient-specific QTc and sex, both functional datasets were significantly associated with event risk (Figure 2D–E). We integrated prospectively available MAVE data from the current study into our previously described Bayesian prospective LQTS penetrance tool (referenced here as ‘Penetrance’; data hosted at variantbrowser.org for community use)¹⁶. Among 1117 patients with sex, QTc, and a LQTS penetrance estimate, we observed 296 cardiac events through the age of 40. To build a more comprehensive model, we tested the inclusion of additional variant predictive features. We reduced model complexity to the most significant covariates using multiple approaches detailed in the Supplemental Methods and Supplemental Figures 5–7. Using a Royston-Parmar time-to-event model, we observed optimal prognostic value incorporating MAVE functional into our LQTS penetrance prediction feature (Figure 2F)^17,18. Interestingly, we observed in Figure 2F that MAVE data not integrated into this model became significantly protective when both APC and MAVE were included in the same model. This may be explained by greater phenotype resiliency against disrupted trafficking versus disrupted gating (consistent with possible dominant negative vs haploinsufficiency mechanisms)¹⁹. These findings show that prospectively available MAVE data provide near equivalent functional evidence strength for variant classification (diagnosis); however, APC data collected after variant ascertainment can provide additional data for event risk stratification (prognosis).

We next built a risk stratification model combining the significant predictors and demonstrated excellent performance in stratifying event risk (Figure 2G). We then compared model accuracy using an ROC/AUC approach (Figure 2H and ​and2I).2I). We found that prospective (patient-specific features + LQTS penetrance estimate conditioned by MAVE data) and a model including prospective and APC data both significantly improved upon prognosis from patient-specific features alone. Lastly, we provide sex-stratified nomograms of our risk prediction model of events at 20 years for variants with prospectively available LQTS penetrance estimate (Supplemental Figure 8).

We present the first comprehensive high-throughput functional studies of a ClinGen definitive evidence LQTS-associated gene². We demonstrate that integrating continuous and quantitative, variant-specific features with detailed clinical phenotyping improves risk assessment for a ‘monogenic’ genotype-phenotype relationship. We propose that similar approaches involving continuous variant-specific features, such as high-throughput functional studies and quantitative penetrance estimates, will further improve risk stratification in other gene-disease pairs.

Online Methods

Full methods are described in the online appendix. All clinical data used in this analysis received IRB/REC approval. The research reported in this paper adhered to guidelines included in the Helsinki Declaration as revised in 2013. All code used to analyze these data may be found on GitHub at https://github.com/kroncke-lab.

Supplementary Material

Acknowledgements

Data and Code Availability

Code used to analyze raw data and generate figures and tables are available at the Kroncke Lab GitHub site. Bayesian variant functional scores for MAVE and APC, in silico predictions, structural features, and clinical data are available at variantbrowser.org. Participant-level data are not included for concern of reidentification and privacy protection.

Literature Cited