Pathogenicity Prediction of GABAA Receptor Missense Variants

Introduction

Epilepsy is one of the most common neurological diseases in the world with a broad phenotypic spectrum.¹ Recent advances in genome sequencing identified an increasing number of genes that are associated with epilepsy.² According to protein functions, epilepsy-associated genes can be grouped to ion channels, enzymes and enzyme modulators, transports and receptors, and others.³ Genetic epilepsy is often linked to developmental delay, movement disorder, and other comorbidities.⁴ Due to the important role of neurotransmitter-gated ion channels in controlling the excitation-inhibition balance in the central nervous system, genes encoding these ion channels, including excitatory N-methyl-D-aspartate (NMDA) receptors and inhibitory γ-aminobutyric acid type A (GABA_A) receptors, are recognized as prominent epilepsy-causing genes.⁵ Here, we focus on GABA_A receptors, the primary inhibitory neurotransmitter-gated ion channels in the human brain.⁶ They mediate the fast inhibitory GABA-induced chloride currents and hyperpolarize the postsynaptic membranes to reduce neuronal firing.

Proteostasis maintenance of GABA_A receptors is essential for their function in the central nervous system.⁷ GABA_A receptors are assembled as pentamers from a specific combination of 19 subunits, including α1-α6 (GABRA1-A6), β1-β3 (GABRB1-B3), γ1-γ3 (GABRG1-G3), δ (GABRD), ϵ (GABRE), θ (GABRQ), π (GABRP), and ρ1-ρ3 (GABRR1-R3). The distribution of GABA_A receptors is throughout the brain regions, and the most abundant subtype is composed of two α1 subunits, two β2 subunits, and one γ2 subunit.⁸ To function, GABA_A receptor subunits need to fold in the endoplasmic reticulum (ER) with the assistance of molecular chaperones and subsequently assemble with other subunits to form heteropentamers. The properly assembled receptors exit the ER and traffic to the plasma membrane to act as chloride channels. Unassembled and misfolded subunits are retained in the ER, which could be routed to the degradation pathway by the ER-associated degradation.⁹ Recent quantitative proteomics analysis identified the proteostasis network that regulates the folding, assembly, trafficking, and degradation of GABA_A receptors.¹⁰

Recent cryo-electron microscopy (cryo-EM) studies solved the high-resolution structures of pentameric GABA_A receptors, including α1β2γ2 receptors¹¹ and α1β3γ2 receptors.¹² The pentameric receptors are arranged as β-α1-β-α1-γ2 counterclockwise when viewed from the synaptic cleft (Figure 1A). Each pentamer has two binding sites for the neurotransmitter, GABA, at the interfaces between β subunits and α1 subunits. Residues from β subunits constitute the principal binding site, denoted as “positive” (+) side, whereas residues from α1 subunits constitute the complementary binding site, denoted as “negative” (−) side. Each subunit shares a common structural scaffold, including a large extracellular N-terminal domain (NTD), four transmembrane helices (TM1-TM4), and loops connecting transmembrane helices (a short intracellular TM1–2 loop, a short extracellular TM2–3 loop, and a long intracellular TM3–4 loop), and a short extracellular C-terminus (Figure 1B, 1C). The secondary structures of the NTD contain two α-helices, ten β-sheets (β1-β10), and connecting loops (Figure 1C, 1D). GABA_A receptors belong to the Cys-loop receptor superfamily.⁷ The signature Cys-loop in GABA_A receptor subunits is designated as loop 7. Biochemical studies revealed that several segments in GABA_A receptor subunits play an important role in binding the ligand: the binding loops in the principal side are called loop A–C, whereas the binding loops in the complementary side are called loop D–F (Figure 1C, 1D).

Figure 1

Open in figure viewerPowerPoint

Structures and sequence alignment of GABA_A receptors. (A) Cartoon representation of pentameric α1βγ2 receptors, built from 6X3S.pdb. The principal side of one subunit is denoted as “+”, whereas the complementary side of one subunit is denoted as “−”. (B) The schematic of the primary protein sequence of a GABA_A receptor subunit. NTD, N-terminal domain; M1-M4, transmembrane helices 1 to 4. (C) The secondary structures of a GABA_A receptor subunit. The two cysteines in the signature Cys-loop are colored in yellow. (D) The sequence alignment of major human GABA_A receptor subunits, including α1, β2, β3, and γ2. The residue positions that harbor clinical missense variants are highlighted. According to ClinVar annotation, pathogenic variants are colored in red, uncertain in yellow, and benign in green.

To date, over 1000 clinical variants in genes encoding GABA_A receptor subunits have been recorded in ClinVar (www.clinvar.com), including missense, nonsense, and frameshift variants. However, the clinical significance of these variants is not adequately addressed since most of them lack functional characterization and many of them are classified as uncertain or conflicting interpretations. For the limited number of characterized GABA_A receptor variants, accumulating evidence indicated that proteostasis deficiency that resulted from misfolding and excessive degradation of the variants is a major disease-causing mechanism.¹³ Adapting the ER proteostasis network pharmacologically corrected the misfolding and restored the surface trafficking and thus ion channel function for a variety of pathogenic GABA_A receptor variants.¹⁴ Another important disease-causing mechanism is that the missense variants result in the channel gating defects and altered electrophysiological properties, such as current kinetics, current amplitude, and ligand potency.

Here, we applied two state-of-the-art modeling tools, namely AlphaMissense¹⁵ and Rhapsody,¹⁶ to comprehensively predict the pathogenicity of saturating missense variants in the major subunits of GABA_A receptors (α1, β2, β3, and γ2). AlphaMissense represents a major technological advance in the field of the prediction of missense variants by integrating structural context and evolutionary conservation.¹⁵ Other machine learning-based prediction approaches have limitations on the training databases, which are prone to human bias,¹⁷ lack of precise structural information,¹⁸ or inadequate genetic evolutionary constraint.¹⁹ AlphaMissense overcomes these limitations by combining the following strengths. First, AlphaMissense utilizes the training datasets with weak labels from population frequency data; second, AlphaMissense fine-tunes highly accurate protein structures afforded by AlphaFold;²⁰ third, AlphaMissense is capable of learning evolutionary constraints according to amino acid sequences. Consequently, AlphaMissense outperforms other prediction models across multiple clinical benchmarks.¹⁵ Recently, AlphaMissense has been applied to predict pathogenicity of cystic fibrosis transmembrane conductance regulator (CFTR) variants and the results correlated well with certain clinical benchmarks.²¹ Here, we also compared the AlphaMissense and Rhapsody predictions with the ClinVar clinical benchmarks, aiming to provide insights for clinical interpretation and guidance for future experimental investigation of missense variants in GABA_A receptors.