Transcriptional dysregulation underlies both a monogenic arrhythmia syndrome and common modifier of cardiac repolarization

Abstract

Background:

The Brugada syndrome (BrS) is an inherited arrhythmia syndrome caused by loss-of-function variants in the cardiac sodium channel gene SCN5A in ~20% of subjects. We identified a family with four individuals diagnosed with BrS harboring the rare G145R missense variant in the cardiac transcription factor TBX5 and no SCN5A variant.

Methods:

We generated induced pluripotent stem cells (iPSCs) from two members of a family carrying TBX5-G145R and diagnosed with Brugada syndrome. After differentiation to cardiomyocytes (iPSC-CMs), electrophysiologic characteristics were assessed by voltage- and current-clamp experiments (n=9–21 cells per group) and transcriptional differences by RNA sequencing (n=3 samples per group), and compared to iPSC-CMs in which G145R was corrected by CRISPR/Cas9 approaches. The role of platelet derived growth factor (PDGF)/PI3K pathway was elucidated by small molecule perturbation. QTc association with serum PDGF was tested in the Framingham Heart Study Cohort (n=1,893 individuals).

Results:

TBX5-G145R reduced transcriptional activity and caused multiple electrophysiologic abnormalities, including decreased peak and enhanced “late” cardiac sodium current (I_Na), which were entirely corrected by editing G145R to wild-type. . Transcriptional profiling and functional assays in genome unedited and edited iPSC-CMs showed direct SCN5A down-regulation caused decreased peak I_Na, and that reduced PDGF receptor (PDGFRA) expression and blunted signal transduction to phosphoinositide 3-kinase (PI3K) was implicated in enhanced late I_Na. Both Tbx5 regulation of the PDGF axis and disruption of PDGF signaling causing arrhythmia risk were conserved in murine model systems. PDGF receptor blockade markedly prolonged normal iPSC-CM action potentials and plasma levels of PDGF in the Framingham Heart Study were inversely correlated with QTc (P<0.001).

Conclusions:

These results not only establish decreased SCN5A transcription by the TBX5 variant as a cause of Brugada syndrome, but also reveal a new general transcriptional mechanism of arrhythmogenesis of enhanced late sodium current caused by reduced PDGF receptor-mediated PI3K signaling.

Introduction

Studies of familial cardiac arrhythmia syndromes have used heterologous expression of ion channels and their ancillary subunits to validate causative genomic variants by demonstrating that they alter protein function in ways consistent with known physiology. More recently, cardiomyocytes from induced pluripotent stem cells (iPSC-CMs) derived from patients with these diseases have been used to confirm conventionally accepted mechanisms. For example, heterologous expression has shown that long QT syndrome-associated variants in cardiac potassium channels decrease K⁺ current, and studies in iPSC-CMs demonstrate the expected action potential prolongation. In addition, CRISPR/Cas9 gene editing in iPSC-CMs has been used to further validate variants in known Mendelian disease genes, such as SCN5A in the Brugada Syndrome (BrS)¹ and PLN in dilated cardiomyopathy.²

BrS is an inherited arrhythmia disease with a characteristic electrocardiogram and an increased risk of sudden cardiac death due to ventricular fibrillation.^{3, 4} Genotype-phenotype studies have identified decreased peak sodium current due to abnormal gating or reduced cell surface expression as a common cellular electrophysiologic hallmark of BrS.^{5, 6} Loss-of-function variants in the voltage-gated cardiac sodium channel gene SCN5A are the most common genetic cause of BrS,⁷ but are identified in only ~20% of cases.^{8, 9}

In a family with BrS and no SCN5A mutation, we previously reported the identification of a novel missense variant of unknown significance in the gene encoding the cardiac transcription factor TBX5 (c.G433A, p.G145R; Fig 1a and 1b).¹⁰ TBX5 is known to modulate SCN5A expression,¹¹ and common variants in this gene have been implicated by genome-wide association studies as modulators of QRS duration, an electrocardiographic correlate of cardiac conduction, slowed in BrS.¹² TBX5 variants resulting in TBX5 haploinsufficiency or severe alterations in DNA binding (e.g. TBX5-G80R) cause the Holt-Oram syndrome (HOS),¹³ a rare disease with abnormal development of the upper limbs and the heart. In mice, Tbx5 haploinsufficiency phenocopies HOS-related impaired septation and abnormal limb development, and decreases cardiac expression of gap junction proteins, atrial natriuretic peptide, and the cardiac sodium channel.^{14, 15} However, the functional effects of TBX5 variants associated with less severe alterations in DNA binding have not been established and no causal relationship with BrS has previously been reported. Here we use gene editing in patient-specific iPSC-CMs to prove that TBX5-G145R causes familial BrS. Transcriptional profiling and follow-up functional studies identify a TBX5-regulated arrhythmogenic network centered on phosphoinositide 3-kinase (PI3K)-mediated signaling.

Figure 1. Clinical and biochemical characterization of a familial Brugada syndrome-associated TBX5 variant.

(a) Representative Brugada syndrome ECG pattern of affected family member with QT dispersion of >100ms. Red arrows highlight ST segment elevation in right precordial leads. (b) Sanger sequencing confirmation of hypothesized risk allele in TBX5 c.G433A. (c) Representative positive procainamide challenge testing with J-point elevation elicited in V2. (d) Electrophoretic mobility shift assay demonstrates TBX5-G145R has reduced DNA binding capacity by 97% compared to wild-type protein. TBX5-G80R included as loss of function control variant that has no DNA binding capacity. (e) TBX5-G145R has reduced luciferase activity compared to WT which was performed in triplicate.

Methods:

Patient enrollment and study approval:

This study was reviewed and approved by the Institutional Review Board at Vanderbilt University Medical Center (Nashville, TN). Detailed protocols are available in Supplemental Material. iPSC-derived cardiomyocyte RNA-sequencing reads are available in the NCBI Short Read Archive with accession number GSE221945. Murine cardiomyocyte RNA-sequencing raw data was deposited in GEO with an accession number GSE125823.

Statistical analyses:

Electrophysiology and Biochemistry Statistics: Data are mean ± SEM unless otherwise indicated. Statistical differences among two groups were tested with two-tailed Student’s t test or Mann-Whitney test for nonparametric comparisons. Statistical differences among more than two groups were assessed using one-way ANOVA followed by post-hoc Bonferroni correction. Results were considered statistically significant if the p-value was less than 0.05. Statistical analyses were performed using GraphPad Prism 6.

iPSC-CM RNA-sequencing statistical analyses: The gene expression differential analysis was performed using DESeq2 (v1.12.4). The genes with absolute fold change larger than 1.5 and false discovery rate less than 0.05 were designated as significantly differentially expressed between groups.

Murine RNA-sequencing statistical analyses: Differential expression testing was performed using edgeR v3.16.5¹⁶ and limma v3.30.13.¹⁷ Low level genes were removed within each condition using median log-transformed counts per gene per million mapped reads (cpm) cutoff of 0. Pearson correlation coefficients were calculated using the base R stats function cor between samples using the log counts per million (CPM) reads.

Proteomic Statistics:

Protein measures were log-transformed and standardized to mean=0, standard deviation=1 within respective batches. Age- and sex-adjusted regression models were performed for quartile of protein (independent variable) versus the corrected QT (QTc) interval (dependent variable). QTc was calculated using the Bazett formula.

Results

Clinical characteristics of study family

The proband of the family originally presented as a 14 year-old male with chest pain most likely due to costochondritis (proband noted as IV.9 in Family pedigree provided in Figure S1a). He had a history of palpitations but no syncopal episodes. His presenting ECG demonstrated a spontaneous type I Brugada pattern (Fig 1a). No SCN5A variant was identified on commercial sequencing. He was referred for an electrophysiology (EP) study for risk stratification which was a standard of care at the time and this revealed inducible polymorphic ventricular tachycardia for which he was treated with primary prevention placement of an implantable cardioverter defibrillator (ICD). A procainamide challenge to elicit the BrS pattern was not performed given his spontaneous type 1 BrS pattern. He declined participating in future research endeavors beyond genetic evaluation. Subsequent screening identified three additional family members with Brugada Syndrome, his mother (family member III.17), sister (family member IV.7), and brother (family member IV.10). The mother also displayed a spontaneous type 1 BrS pattern. There was one family member, a maternal great aunt to proband, who died suddenly at age 59 years but additional clinical history is not available (family member II.5; Figure S1a)

Targeted Sanger sequencing was performed in the parents and children identifying the previously reported novel, missense variant in TBX5 that is absent in gnomAD 2.1 and gnomAD 3 (c.G433A; p.G145R, Fig 1b, Figure S1a).¹⁰ The elder sister (noted as IV.7 in pedigree in Figure S1a) was also screened based upon the family history and symptoms of palpitations without syncopal events. Procainamide challenge demonstrated a positive test with >2mm elevation of J point in V1 and downsloping ST segment elevation and milder changes in V2; nonsustained polymorphic ventricular tachycardia was induced during the EP study. The youngest sibling (member IV.10 in pedigree) had many ECGs that were negative for spontaneous BrS pattern. A procainamide challenge test was negative at age 17 but a re-challenge at age 22 produced a positive test with >2mm elevation of J point in V2, consistent with the known age-related penetrance of this condition (Fig 1c and Figure S1a).^{18, 19}

Review of all available ECGs from variant carriers demonstrated rate corrected QT intervals (QTc) were normal (422–447 msec in the member IV.7 (sister) and 375–422 msec in member III.17 (mother)). However, we did observe evidence of abnormally long repolarization, localized to specific leads: QT inter-lead dispersion, assessed as the difference between longest and shortest QT measured in all 12 ECG leads (Fig 1a), was strikingly prolonged (104 and 78 msec [normal <60 msec;²⁰ pathologic in long QT syndrome >100 msec²¹]), and QTc in individual right precordial leads was as long as 521 msec. There is one family member (II.5) with sudden death at age 59 but no family history of syncope (See extended pedigree in Figure S1a).

TBX5-G145R causes BrS

Electrophoretic mobility shift assays showed a severe but incomplete loss of DNA binding capacity by TBX5-G145R relative to TBX5-WT and reduced transcriptional regulation in a TBX5-responsive enhancer luciferase reporter assay (Fig 1d and 1e). This markedly decreased function is consistent with a potential causative role in BrS in the index family. As the number of individuals within the family was insufficient to generate statistical evidence of linkage, we generated and validated three iPSC clones from the elder sister (family member IV.7) in the kindred diagnosed with BrS (TBX5^G145R/WT) and four clones from healthy population controls (Figure S2). To directly test if the missense TBX5 variant caused the BrS phenotype, we edited the TBX5 c.G433A missense variant back to the reference sequence in two BrS clones using RNA-guided Cas9 nuclease to generate isogenic controls (Figure S3). Sequencing of predicted off-target exonic regions did not identify off-target activity of Cas9 (Table S1).

Expression of TBX5 in TBX5^G145R/WT iPSC-CMs at 35 days post-cardiac induction was no different than population or isogenic controls by qPCR and RNA-sequencing (Fig 2a). However, total SCN5A transcripts, inclusive of both fetal and adult isoforms, were reduced by 79% compared to each control group by qPCR (P<0.05; Fig 2a); this result was also seen in the RNA sequencing experiments described below. In whole cell voltage-clamp experiments TBX5^G145R/WT iPSC-CMs displayed ~50% reduced peak sodium current (−58±4.9 pA/pF) compared to the control groups (population: −124±8.0; isogenic: −133±6.5) at a test voltage of −30 mV (P<0.05; Summary data pooled by genotype presented in Fig 2b and 2d; individual clone data presented in Figure S4–7 and Table S4). L-type calcium and potassium currents, inconsistently implicated in BrS pathogenesis,⁷ were not different between groups (Figure S8–10). Thus, TBX5^G145R/WT cells demonstrate reduced sodium current, the electrophysiologic hallmark of BrS, attributable to reduced SCN5A expression. The findings of phenotype reversal in the isogenic control cells establish TBX5-G145R as the underlying cause.

Figure 2. Abnormal sodium channel expression and electrophysiology in TBX5^G145R/WT iPSC-CMs.

(a) RNA levels of TBX5 and SCN5A normalized to RPL19 transcript in iPSC-CMs. In this and subsequent figures, data from TBX5^G145R/WT cells are indicated in red and from TBX5^G145R/WT cells with G145R edited back to G in blue. Summary data of 4–6 cardiomyocyte differentiations for four independent clones per group. (b) Representative whole cell voltage clamp recordings at voltage steps from −100 to 20 mV in 10 mV increments. Inset voltage clamp protocol. Summary data presented in panel e. (c) Representative sodium current recordings highlighting late current. Inset voltage clamp protocol. (d) Peak sodium current density plotted against test voltages (Population control (ctrl): n=14; TBX5^G145R/WT: n=16; Isogenic ctrl: n=9). (e) Voltage-dependence of steady-state inactivation and activation. (f) Quantification of late sodium current. Ranolazine treatment: 3μM. (Population control (ctrl): n=10; TBX5^G145R/WT: n=10; Isogenic ctrl: n=7). All summarized data displayed as mean±S.E.M. Means were compared using one-way ANOVA followed by post-hoc Bonferroni correction.

Further arrhythmogenic behaviors in TBX5^G145R/WT iPSC-CMs

The TBX5^G145R/WT cells displayed abnormalities beyond reduced peak sodium current and, as we describe here, these were also restored to wild-type behavior in genome edited cells. There was an 8.7 mV shift in the voltage (V_1/2) at which half-maximal activation (Isogenic control −42.9±0.5 mV; TBX5^G145R/WT −34.2±0.6 mV) and a 10 mV shift for inactivation (Isogenic control −88.9±0.3; TBX5^G145R/WT −78.9±0.3; Summary data pooled by genotype presented in Fig 2e and individual clone data presented in Figure S7 and summarized in Table S4). In addition to these shifts in activation and inactivation of NaV1.5 (the channel encoded by SCN5A), we also observed a marked increase in a non-inactivating (“late”) component of the sodium current, I_Na-L, at 2.2±1% of peak sodium current in TBX5^G145R/WT iPSC-CMs but negligible I_Na-L in the population and isogenic controls accounting for 0.25±0.03% and 0.26±0.04% of peak I_Na, respectively (Fig 2c and 2f). Treatment of TBX5^G145R/WT iPSC-CMs with the late sodium current blocker ranolazine (3 μM) completely eliminated I_Na-L (Fig 2c and 2f).²² Enhanced I_Na-L has been reported in a number of settings, notably as a contributor to heart failure and increased action potential duration (APD) in the congenital or drug-induced forms of the long QT syndrome.^23–25 TBX5^G145R/WT iPSC-CMs showed normal resting membrane potential and action potential amplitude but did display increased APD at 90% repolarization (APD₉₀) (Summary data pooled by genotype presented in Fig 3a–b and individual clone data presented in Figure S11 and summarized in Table S2). We generated an additional iPSC line from the affected mother to confirm the key, unanticipated electrophysiologic abnormalities. This clone also demonstrated increased I_Na-L and APD prolongation (Figure S11). Arrhythmogenic early afterdepolarizations, delayed afterdepolarizations, and triggered action potentials were observed in 4/18 of TBX5^G145R/WT iPSC-CMs with pacing at 1Hz but not in any cell (n=20) studied from either control group (Fig 3b–3d). Afterdepolarizations with occasional triggered action potentials were also observed in the rare spontaneously beating individual TBX5^G145R/WT iPSC-CMs (representative tracing in Figure S12). These electrophysiologic abnormalities were also abolished by 3 μM ranolazine, further implicating enhanced I_Na-L as the cause of the increased APD (Fig 3b–3d).

Figure 3. Arrhythmogenic changes in cardiac action potential in TBX5^G145R/WT iPSC-CMs.

(a) Representative action potential traces recorded at 1 Hz. (b) Averaged resting membrane potential (RMP), action potential amplitude (APA), action potential duration at 50% (APD₅₀) and 90% repolarization (APD₉₀). (Population control (ctrl): n=13; TBX5^G145R/WT: n=21; Isogenic ctrl: n=16). * indicates P<0.05 compared to TBX5^G145R/WT iPSC-CMs. (c) Early afterdepolarizations and delayed afterdepolarizations after a triggered beat (TB) (d) were observed in TBX5^G145R/WT iPSC-CMs but not in population or isogenic control cells (arrows). (e) Superimposed action potentials from single iPSC-CMs demonstrate APD variability in the TBX5-G145R cell. (f) Poincaré plots comparing the APD of each beat (N APD₉₀) to the APD (N+1 APD₉₀) of the subsequent beat. Dashed ellipse represents scatter area. All summarized data displayed as mean±S.E.M. Ranolazine treatment: 3μM in TBX5^G145R/WT: n=21. Means were compared using one-way ANOVA followed by post-hoc Bonferroni correction. See Table S2 for summary data including number of cells studied per group.

Beat-to-beat variability in APD is another arrhythmogenic feature linked to I_Na-L.²⁶ In healthy control iPSC-CMs, Poincaré plots (APD_n vs APD_n+1) revealed very little beat-to-beat APD variability, whereas there was substantial variability, abolished by ranolazine, in TBX5^G145R/WT iPSC-CMs (Fig 3e and 3f).²⁷ Metrics of variability in Poincaré plots include scatter plot area, short term variability (STV), and long term variability (LTV) in APD,²⁸ and all were significantly increased in TBX5^G145R/WT iPSC-CMs relative to population and isogenic controls (Table S2). Ranolazine exposure returned the TBX5^G145R/WT iPSC-CM Poincaré plot parameters to control values. The APD variability did not result from alterations in calcium control as parameters of calcium handling were similar in disease and control groups (Table S3).

We considered the possibility that the increased I_Na-L and afterdepolarizations might arise as a direct result of decreased NaV1.5 abundance (e.g. by altered interactions with function-modifying subunits). To test this idea, we generated SCN5A heterozygous knockout iPSC-CMs from the isogenic control cells (TBX5^WT/WT; SCN5A^+/−), and found NaV1.5 levels reduced to a similar degree as in TBX5^G145R/WT myocytes but no increased APD variability or afterdepolarizations (Fig 4 and Table S2).

Figure 4. Heterozygous SCN5A knockout iPSC-CM arrhythmia models generation and characterization.

(a) Schematic of genome editing result using CRISPR/Cas9 in patient-derived iPSCs to generation loss of function SCN5A Brugada syndrome model. Modified nucleotide variants and amino acids indicated as red text. (b) Representative superimposed action potentials recorded from isogenic TBX5^WT/WT; SCN5A^-/WT. Summary data provided in Table S2, n=4.

We further probed the impact of the single-cell arrhythmogenic phenotype in a tissue-like context²⁹ by assaying the electrical properties of spontaneously beating monolayers of iPSC-CMs using extracellular field potentials (EFPs) and the contractile properties using impedance measurements.³⁰ TBX5^G145R/WT and control monolayers displayed regular beating rates at 50–60 beats/min with minimal variability in the beat frequency, while EFP duration (FPD) was significantly longer in TBX5^G145R/WT iPSC-CMs (Figure S13a and S13b). Poincaré plots were similar to those seen in the single cell patch clamp experiments, with TBX5^G145R/WT iPSC-CM monolayers showing afterdepolarizations as well as larger STV and FPD variability, both of which were reduced by 3 μM ranolazine (Figure S13c–13e and Table S5). In contrast to the dramatic differences in EFP characteristics, we observed only modest differences in impedance parameters between lines. Impedance pulse width at 30% peak height was minimally prolonged in TBX5^G145R/WT monolayers, consistent with the prolonged action potentials in these cells, and there was no difference in 50% peak height or 90% peak height between lines (Figure S13f). These data indicate that contractile function was minimally affected in TBX5^G145R/WT cells, in contrast to the significant changes in these parameters observed in models of cardiomyopathy³¹ and consistent with normal contractile function in individuals with BrS.

Changes in gene expression with TBX5-G145R

Holt-Oram Syndrome (HOS) is defined by septal defects and upper limb abnormalities and is caused by severe loss-of-function (LOF) variants in TBX5. Our family members do not have overt HOS presentation recognized prior to the identification of the TBX5-G145R variant. Additional clinical phenotyping and family history found subtle and variably expressive thumb hypoplasia, and a single member with a ventricular septal defect (see family pedigree in Figure S1a and S1b). Biochemical assays demonstrated that TBX5-G145R is not a complete LOF variant, since it maintained ~3% DNA binding capacity and ~38% transcriptional activity compared to wild-type TBX5 (Fig 1d and 1e). In order to understand the differences in gene regulation resulting from the BrS-identified G145R variant compared to canonical HOS TBX5 haploinsufficiency, we generated isogenic iPSC lines carrying heterozygous TBX5 frameshifting insertion or deletion (indel) variants, thereby creating an HOS iPSC-CM model (TBX5^-/WT; Figure S14 and S15). We performed RNA-sequencing (at least three clones each) from population controls as well as the following isogenic groups: TBX5^G145R/WT, the CRISPR-corrected TBX5^G145R/WT group (TBX5^WT/WT), and TBX5^-/WT. Consistent with the qPCR data, the TBX5-G145R samples showed a 53±13% reduction in SCN5A expression in the RNA-sequencing dataset (Fig 5a). Human iPSC-CMs express the fetal and adult isoforms of SCN5A which are both reduced in TBX5-G145R containing cardiomyocytes. The fetal isoform was reduced by 50% and the adult form by 70% compared to control cardiomyocytes. Sixteen other genes rarely associated with BrS including SCN10A and the function-modifying subunits SCN1B, and SCN3B were not differentially expressed (Table S6).⁵

Figure 5. Reduced phosphoinositol-3 kinase activity in TBX5^G145R/WT iPSC-CMs.

(a) SCN5A expression measured by RNA-sequencing for each study group normalized to population control. See methods for experimental details. (b) Volcano plot of differentially expressed genes in TBX5^-/WT compared to TBX5^G145R/WT iPSC-CMs. Red dots highlight genes implicated in heart development biologic pathway. (c) Volcano plot of differentially expressed genes in TBX5^G145R/WT compared to population control iPSC-CMs. Red dots highlight genes implicated in regulation of phosphoinositol-3 kinase signaling biologic pathway. (d) Quantification of Western blot analysis using phospho-specific antibodies of known phosphoinositol-3 kinase targets S6 and Akt normalized to total protein of each target. n = 3 biologic replicates per group (e) Western blot analysis for phosphorylation of Akt after 2-hour incubation with vehicle or ligands in combination or individually. Ligands are those identified by gene ontology with reduced expression by RNA-sequencing. (f) Representative HPLC traces and summary data of inositol species quantification. *indicates P<0.05

Unsupervised hierarchical clustering of adult human ventricle-enriched genes identified a distinct gene expression signature in TBX5^-/WT iPSC-CMs relative to TBX5^G145R/WT iPSC-CMs, which clustered more closely with the control lines (Figure S16a and S16b). There was a greater reduction in SCN5A expression in TBX5^-/WT iPSC-CMs compared to TBX5^G145R/WT iPSC-CMs and a significant reduction in expression of NPPA, a previously characterized direct TBX5 target (Fig 5a–b).¹⁵ Functional pathway analysis of the 234 genes differentially expressed in haploinsufficient TBX5^-/WT iPSC-CMs compared to the TBX5^G145R/WT revealed enrichment for genes governing heart morphogenesis and cardiovascular development including reduced levels of cardiac transcription factors HEY2^{32, 33} and ETV1³⁴ and structural genes MYL2 and MB, supporting TBX5-G145R as a hypomorphic loss-of-function variant and the known role of TBX5 in cardiogenesis (Fig 5b, Figure S16a–c, and Table S7).³⁵ Interestingly, HEY2 has also been implicated as a regulator of SCN5A function and of the BrS phenotype.^{32, 36, 37}

The distinct effect of TBX5^G145R/WT on I_Na-L phenotypes prompted us to perform additional gene ontology analyses in order to identify other arrhythmogenic pathways beyond reduced SCN5A expression. Due to the hypomorphic impact of TBX5-G145R on gene expression, we expanded our analysis to include all genes with significant (adjusted p-value <0.05) and sub-threshold (>1.5-fold change in expression with an uncorrected p value < 0.05) differential expression compared to control lines as part of hypothesis generating evaluation (Fig 5c and Table S8–S9). This approach identified functional enrichment in genes involved with regulation of the phosphoinositol-3 kinase (PI3K) signaling pathway in TBX5^G145R/WT iPSC-CMs compared to control cardiomyocytes. TBX5^G145R/WT iPSC-CMs have reduced gene expression of multiple upstream transmembrane receptors (PGFRA, TEK, FLT1) or their ligands (IGF1 and HB-EGF) that are activators of the PI3K cascade.

Reduced PI3K activity with resultant enhanced late sodium current has been implicated as a mechanism of arrhythmia risk in diabetes ²³ or caused by exposure to QT prolonging drugs,^{25, 38} so we hypothesized that our observed cellular arrhythmia phenotypes were due to altered PI3K signaling. Phospho-specific antibodies against the PI3K targets Akt and S6 and semi-quantitative analysis by Western blot revealed a non-statistically significant trend for reduced basal phosphorylation in TBX5^G145R/WT iPSC-CMs compared to isogenic myocytes of 31% and 45% reduction, respectively (Fig 5d). To more accurately quantify PI3K activity, we utilized a more sensitive method to assess phosphorylation state by quantifying radiolabeled phosphoinositides by HPLC after tritiated-inositol labelling,³⁹ and found TBX5^G145R/WT iPSC-CMs had a 61% reduction in basal levels of PIP₃ (the downstream effector of PI3K signaling) compared to isogenic control myocytes (P<0.05; Fig 5f). We used Western blot analysis of Akt phosphorylation to study the effects of PI3K pathway ligands identified by pathway analysis (IGF1, VEGF, PDGF, HB-EGF, and Angiopoietin-1). There was 3–4 fold increase in Akt phosphorylation with pooled exposure to these ligands. Follow-up experiments demonstrated that this effect was attributable to PDGF alone, with minimal effects of the other ligands (Fig 5e). These results were confirmed by HPLC quantification of radiolabeled phosphoinositides which showed that one-hour stimulation with PDGF increased PIP₃ levels 3.4 and 3.9-fold in TBX5^G145R/WT and isogenic control iPSC-CMs, respectively (Fig 5f and Figure S17), while treatment with IGF-1 and VEGF had no effect on Akt phosphorylation or PIP₃ levels.

Electrophysiologic effects of PDGF and of the PDGF receptor antagonist crenolanib

We tested the potential of PDGF to blunt the arrhythmogenic phenotype we observed in TBX5^G145R/WT iPSC-CMs. As shown in Fig 6a and 6b, treatment with PGDF in these cells significantly reduced I_Na-L (P=0.003, Figure S18), shortened APD and reduced APD variability in individual iPSC-CMs; the effects were similar to those observed with intracellular perfusion with PIP3 (Fig 6c and Table S2).^{23, 25} The rightward voltage shifts in sodium channel activation and inactivation were also restored to the V_1/2 recorded in control myocytes at −44.3±1.2 and −86.9±1.4 mV (n=9), respectively. By contrast, treatment with IGF1 did not impact the described arrhythmia phenotypes (Figure S19). To test if PDGF directly blocks I_Na-L, we studied I_Na-L in HEK cells transfected with wild-type NaV1.5 or mutant F1486L that causes SCN5A-associated type 3 congenital long QT syndrome⁴⁰ because of enhanced I_Na-L: PDGF caused no change in the magnitude of I_Na-L in wild-type or F1486L cells (Figure S20). Treatment with PDGF also completely abolished afterdepolarization-triggered action potentials and resolved APD variability in isolated TBX5^G145R/WT iPSC-CMs (Fig 6b and Table S2). Additionally, PDGF exposure also reduced the EFP duration variability in TBX5^G145R/WT iPSC-CM monolayers (Table S5). The extent of this reduction was similar to that seen with the I_Na-L blocker ranolazine, supporting a common mechanism of increased I_Na-L (Table S5).

Figure 6. Reduced phosphoinositol-3 kinase activity promotes arrhythmias iPSC-CMs.

(a) Representative I_Na traces with repetitive stimulation with acute treatment with PDGF in TBX5^G145R/WT iPSC-CM. (b) Representative superimposed action potentials recorded from TBX5^G145R/WT iPSC-CM treated with vehicle or PDGF for 2-hours. Summary data provided in Table S2. (c) Representative action potential recording over time with intracellular PIP₃ treatment in TBX5^G145R/WT iPSC-CMs. (d) Representative APs recorded from control iPSC-CMs at baseline, after 2-hour crenolanib treatment, and crenolanib + ranolazine treatment. All summarized data displayed as mean±S.E.M. Means were compared using one-way ANOVA followed by post-hoc Bonferroni correction.

In heterologous expression systems, we have previously reported that the effect of PI3K on late sodium current is mediated by the α-isoform.⁴¹ We also reported that the PI3Kα inhibitor A66 phenocopied the anticipated electrophysiologic phenotypes of increased late sodium current and APD prolongation in control iPSC-CMs⁴² and here we tested if chemical blockade of the PDGF pathway had a similar result. We found that the PDGF receptor inhibitor crenolanib strikingly prolonged action potential duration in control iPSC-CMs, to >1000 msec, and this was reversed to normal duration with ranolazine exposure (Fig 6d). The findings were also seen in adult mouse left ventricular myocytes, where crenolanib treatment caused marked APD prolongation and afterdepolarization-triggered action potentials (Fig 7a).

Figure 7. TBX PDGF phenotypes conserved in mouse cardiomyocytes.

(a) Representative mouse ventricular cardiomyocyte (CM) action potentials recorded at pacing rate of 5 Hz and (e) 0.1Hz after 2 hour treatment with vehicle or 100 nM crenolanib. Triggered action potentials (TA) were observed in Crenolanib treated cells paced at 0.1Hz (b) Pearson correlation and hierarchical clustering of biological replicates for the transcriptional profiling of mouse embryonic left ventricle 24-hours after conditional knockout in Tbx5^flox/flox and controls (left). Differential expression testing of transcriptional profiles with Pdgfra and Pdgfrl (bold) and members of the PDGF-BB signaling axis that predict inhibition (blue), activation (red), or affected (black) by conditional knockout of Tbx5 (right).

PDGF dysregulation in developing mouse and in a human population

We addressed whether TBX5 regulation of PDGF signaling occurs in biological contexts other than TBX5^G145R/WT iPSC-CMs by studying established Tbx5 null murine models. We found that in multiple developmental settings, removal of Tbx5 caused diminished expression of genes in the PDGF axis, including Pdgfb, Pdgfd, and Pdgfra. We analyzed RNA-Seq data from of heart tubes collected from embryonic day 9.5 (E9.5) embryos with germline deletion of Tbx5.^{15, 43} Tbx5 deleted heart tubes displayed a significant 1.65-fold decrease in Pdgfra expression compared to littermate controls. To avoid the confounding requirement of Tbx5 in early heart organogenesis, we also analyzed gene expression from the E12.5 left ventricle 24-hours following inducible deletion of Tbx5 using R26R-CreERT2. We observed reduced expression of two PDGF receptors, Pdgfra and Pdgfrl (Fig 7b). Furthermore, upstream ingenuity pathway analysis predicted significant reduction of the PDGF-BB signaling axis (z-score −2.91, p-value=1.15E-10, rank 18/928 pathways by p-value). It has previously been shown that Tbx5 null cardiomyocytes have APD prolongation compared to wild-type littermate controls (51–68 vs 144–188 ms).^{44, 45} These results support the role of TBX5 in regulating the PDGF/PI3K signaling cascade in both in vitro human iPSC-CMs and in vivo murine cardiac developmental contexts.

To investigate if PDGF signaling broadly influences electrical signaling in the general human population, we directly measured plasma growth factor levels in healthy individuals in the Framingham Heart Study Offspring Cohort with ECG data available (n=1,893 individuals). There was no correlation between QT measurement and insulin and IGF-1 but there was an inverse correlation between higher plasma levels of PDGF-B and shorter QTc intervals (Fig 8, P=0.0001). These data support a common mechanism of PDGF being a modifier of cardiac repolarization as evaluated by QTc in the general population.

Figure 8. Altered levels of plasma PDGF is associated with QT interval in the general population.

(a) Mean QTc +/− SE plotted for quartiles of log-transformed measured plasma PDGF beta in healthy individuals.

Discussion

Correct assignment of causation to a rare genetic variant is critical to the care of affected individuals and their families and is becoming a growing problem as the use of panel-based sequencing is increasingly identifying rare variants of unknown significance. To date, the strongest evidence that a rare genetic variant causes a familial arrhythmia syndrome has come from linkage analysis, which requires large kindreds with multiple affected family members. In the absence of large kindreds, causation has sometimes been inferred by the discovery of a rare variant in a logical candidate gene and functional studies that support causation. However, this logic can be flawed, as evidence reviews have, for example, downgraded assertions of pathogenicity in some putative BrS and LQTS genes.^{7, 46, 47} Thus, stringent experimental evidence is required for variants discovered in novel genes, not established for a particular disease, as in our index case reported here. This provides an example of how the development of patient-specific iPSC-CMs represents an important new approach to address this problem. The reduced sodium current in our patient-specific iPSC-CMs compared to population controls is consistent with the known pathophysiology of BrS, but does not address whether TBX5-G145R causes or even contributes to the cellular or clinical phenotype. Another TBX5 variant (F206L) has been described in familial BrS further supporting TBX5 as potential BrS gene but our work establishes genetic causation.⁴⁸ It is the correction by gene editing of this electrophysiologic abnormality that firmly establishes G145R as the cause of the abnormality in this family and that TBX5 is a novel BrS causative gene. However, further work is required to determine if all or only a subset of TBX5 variants increases an individual’s risk of BrS or additional arrhythmias.

Transcriptional profiling in patient-derived and genetically-edited cells implicated decreased expression of a set of genes converging on PI3K signaling, and a single ligand, PDGF, rescued the defect in PI3K signaling and reversed the arrhythmogenic phenotype. These mechanisms are conserved across species as genetic deletion of Tbx5 in mice reduced expression of multiple genes in the PDGF axis, including Pdgfra, and chemical inhibition of the PDGF receptor resulted in arrhythmogenic APD prolongation in both human iPSC-CMs and murine ventricular cardiomyocytes. The results of these functional studies were supported using the completely orthogonal population proteomic analysis, which identified serum PDGF levels as a modifier of human QTc, a known proxy of arrhythmia risk which is all the more remarkable since initial studies implicating perturbed PI3K signaling as an arrhythmogenic pathway studied arrhythmias due to exposure to QT-prolonging drugs.^{25, 38} These results thus define a novel genetically-defined transcriptional mechanism, culminating in decreased PI3K activity, of arrhythmia susceptibility.

Post-translation modifications of NaV1.5 by phosphorylation and nitrosylation are known modifiers of I_Na-L.⁴⁹ We have shown that chronic exposure to the QT prolonging agent dofetilide (previously classified as a “pure” blocker of the potassium current I_Kr) enhanced I_Na-L in Chinese hamster ovary cells expressing the cardiac sodium channel and in human iPSC-CMs through reduced PI3K activity that was rescued by intracellular delivery of PIP₃.²⁵ In the present report we identified a novel transcriptional mechanism of disrupted regulation of upstream PI3K activators causing increased I_Na-L. QT prolongation is a recognized potentially fatal side effect of tyrosine kinase inhibitors (TKIs), particularly those that have off-target blockade of PDGF α-receptor.^{23, 50} Others have shown that exposure of human iPSC-CMs to the TKIs nilotinib and vandetanib prolonged the duration of spontaneous action potentials and decreased the spontaneous beat rate,⁵⁰ and the data reported here also raise the possibility that PDGF receptor targeting therapies may carry arrhythmogenic potential. The role of PDGF in the pathogenesis of arrhythmia and contractile phenotypes continues to evolve. Phase II clinical trial safety data are unavailable for crenolanib but other PDGF inhibitors, such as midostaurin and quizartinib, are associated with QT prolongation in 11% and 5–13% of patients, respectively.⁵¹ PDGF signaling is critical in the electromechanical remodeling in sheep atrial myocyte and myofibroblast coupling.⁵² Further, the PDGF pathway has also been implicated in FLNC and LMNA-associated cardiomyopathies pathogenesis through enhanced expression of PDGFRA or PDGFRB.^{53, 54} A more complete understanding of the PDGF axis on both arrhythmic and contractile phenotypes is needed.

An additional variant in TBX5 (p.D111Y) has been identified in family with LQTS with normal sodium current density and presence of increased late sodium current, prolonged action potential, and after depolarizations attributed to dysregulation of Ca/calmodulin kinase IIδ signaling cascade.⁴⁸ There is growing evidence that demonstrates that disrupted TBX5-mediated transcription can result in a spectrum of cardiac manifestations to ranging from abnormal cardiac development to electrophysiologic manifestations. Our data presented in this report extend our understanding of TBX5 biology and support the idea that the mechanism of the action potential prolongation due to TBX5-G145R or reduced PDGF signaling is increased I_Na-L secondary to reduced PI3K activity.

A limitation of this study is that we have not identified a translational direction to directly modify growth factor signaling as a therapeutic. Growth factor signaling is delicately balanced and utilizing exogenous PDGF therapeutically thus may not be feasible; for example, mouse models of exogenous PDGF-AA delivery led to atrial fibrosis and atrial fibrillation.⁵⁵ Therefore, we propose an alternative and safer approach to treating the reduced receptor tyrosine kinase signaling contributions to arrhythmia may be to target the downstream mechanism, such as increased I_Na-L, a known effect of ranolazine. We measured growth factor levels in plasma but these data will not precisely define the local signaling context within the functional myocardium in the healthy donor studied. Our study established an association between plasma PDGF-B and QTc and causation is inferred based upon data collected in our cell and murine model systems.

The phenotypes of increased I_Na-L and APD prolongation that we observed are not typical features of BrS; rather they are a recognized cause of QT interval prolongation in a range of settings, including the congenital long QT syndrome (cLQTS). As described in the “Clinical characteristics of the study family” section, the rate corrected QT interval (QTc) in subjects from whom we generated and studied iPSC-CMs was normal on multiple ECGs but did have evidence of abnormally long repolarization, localized to specific leads: QT inter-lead dispersion and QTc in individual right precordial leads as long as 521 msec. QT inter-lead dispersion has been proposed as a better predictor of arrhythmia risk than QTc itself and remains debated in the electrophysiology community.^{56, 57} Further, global electrical heterogeneity is associated with increased risk of drug induced-torsades de pointes independent of QTc interval.⁵⁸ The observation of action potential prolongation at the single cell level and QTc that is only mildly abnormal is also compatible with the idea raised by others that iPSCs may reveal cellular phenotypes that are not uniformly translated to the whole heart, may require environmental cues, or are influenced by basal or fluctuating endocrine/paracrine signaling.⁵⁹ For example, in the case of some cardiomyopathies, cellular phenotypes are easily appreciated in iPSC-CMs while no phenotype occurs in mouse models or humans until adulthood or after stress.^{31, 60}

This report establishes a transcriptional mechanism of reduced PI3K signaling and arrhythmia risk driven by increased late sodium current. There are downstream questions the data raise that would merit further study. While we demonstrate a transcriptional relationship between TBX5 activity and PDGF activity, the presented experiments cannot differentiate from a primary (PDGFRA as a direct target of TBX) or secondary regulatory role. Human ChIP-sequencing data for TBX5 in cardiomyocytes are not currently achievable with commercially available antibodies which would provide insight into the localization of the transcription factor (mutant and wild-type) within the genome. Additionally, we observe significant prolongation of the action potential in TBX5-G145R containing iPSC-CMs while there is no phenotype on calcium handling properties. We observed normal decay of the calcium transient in our cells and this may reflect the fact that iPSC-CMs are immature relative to the cardiomyocytes in the adult human heart with calcium handling characteristics more similar to neonatal or fetal cardiomyocytes. Future studies are warranted to investigate the relationship between late sodium current and calcium handling.

Patient-directed investigation of rare genetic variation in combination with population-level validation has the potential to provide insight into underlying disease mechanisms and risk. Even though the index family does not display clinical long QT syndrome, this family may possess increased susceptibility of drug-induced QT prolongation and the population study supports the results of the observed cellular phenotypes. In combination, this report raises the strong possibility that extracellular growth factors that modify PI3K activity may alter susceptibility to arrhythmias in diverse settings, such as heart failure or drug exposure as discussed above. The balance of PI3K and Akt in the failing heart modifies protective processes such as influencing chronic hypertrophic response,⁶¹ cardiomyocyte apoptosis,⁶² and, as demonstrated in this report, the magnitude of arrhythmogenic late sodium current, and thus represents a novel pathway for arrhythmia susceptibility and management.

Clinical Perspective.

What is new?

• Using CRISPR/Cas9 editing in patient-derived induced pluripotent stem cells, we establish a missense variant in the transcription factor gene TBX5 as a cause of Brugada syndrome.

• We also show that these cells demonstrate reduced platelet derived growth factor (PDGF) receptor-mediated signaling resulting in enhanced late sodium current and prolonged action potentials; this effect is phenocopied by PDGF receptor inhibition in wild-type cells.

What are the clinical implications?

• Tyrosine kinase inhibitors with activity against the PDGF receptor may result in QT prolongation and increased arrhythmia risk.

• Data from the Framingham Heart study support the idea that serum PDGF is a modifier of human QTc, and thus may be a biomarker for arrhythmia risk.

Abbreviations

APA

action potential amplitude

APD

action potential duration

APD₅₀

action potential duration at 50% repolarization

APD₉₀

action potential duration at 90% repolarization

BrS

Brugada syndrome

cLQTS

congenital long QT syndrome

EB

embryoid bodies

EFP

extracellular field potentials

EP

electrophysiology

FDP

field potential duration

FHS

Framingham Heart Study

gRNA

guide RNA

HEK

human embryonic kidney

HOS

Holt-Oram syndrome

ICD

implantable cardioverter defibrillator

Indel

insertion or deletion

iPSC

induced pluripotent stem cell

iPSC-CM

induced pluripotent stem cell-derived cardiomyocytes

I_Kr

delayed rectifying potassium currents

I_Ks

slow delayed rectifying potassium currents

I_Na

sodium current

I_Na-L

late component of sodium current

LOF

loss-of-function

LTV

long term variability

PA

posteroanterior

PI3K

phosphoinositide 3-kinase

PDGF

platelet derived growth factor

RMP

resting membrane potential

RNZ

ranolazine

SR

sarcoplasmic reticulum

STV

short term variability

TA

triggered action potentials

TKI

tyrosine kinase inhibitor

QTc

corrected QT interval