Deciphering the impact of ABCA4 genetic variants of unknown significance in inherited retinal disease through computational and functional approaches

Abstract

Variants in the ABCA4 gene are a fundamental cause of several inherited retinal degenerations (IRDs), including Stargardt macular dystrophy, retinitis pigmentosa, and cone-rod dystrophy. These three ABCA4-driven diseases are estimated to cause blindness in 1.4 million people worldwide. As a result, genetic testing of ABCA4 is increasingly common in clinical settings. Of the 4111 identified variants in ABCA4, 1668 are missense, of which 47 % are of unknown pathogenicity (variants of unknown significance, VUS). This genetic uncertainty leads to three fundamental problems: (i) for IRD patients with multiple unclassified ABCA4 mutations, it is impossible to predict which variant will cause disease in relatives who have not yet developed it; (ii) development of variant-specific therapies remains limited; and (iii) these variants cannot be used to predict disease prospectively, which is essential for life-planning decisions and for directing patients to new clinical trials. This chapter describes approaches to deciphering the impact of ABCA4 genetic variants of unknown significance (VUS) using a combination of in silico and in vitro analyses. By leveraging complementary fields—protein biochemistry and computational biology—to create a “sequence-structure-function” workflow, where in silico 3D protein structural analysis of ABCA4 sequence variants serves as a tool to predict disease severity and clinical pathogenicity in conjunction with first-line bioinformatic tools and functional analysis. This approach represents a helpful step forward in understanding how ABCA4 variants affect structure and function and in evaluating their potential to cause inherited retinal diseases.

1. ABC transporters and the ABCA subfamily

ATP-binding cassette (ABC) transporters represent one of the largest and most diverse families of membrane proteins, playing critical roles in various physiological processes across species (Biswas-Fiss et al., 2018; Dean, Moitra, & Allikmets, 2022; Juan-Carlos, Perla-Lidia, Stephanie-Talia, Mónica-Griselda, & Luz-María, 2021). These transporters are characterized by their ability to harness the energy derived from ATP hydrolysis to translocate a wide range of substrates across cellular membranes. The substrates transported by ABC proteins include ions, lipids, peptides, and metabolic products, making them integral to numerous cellular functions, including lipid metabolism, antigen presentation, and drug resistance (Juan-Carlos et al., 2021; Kaminski, Piehler, & Wenzel, 2006).

The ABC transporter superfamily in humans can be categorized into seven subfamilies, designated ABCA through ABCG, based on sequence homology, structural organization and phylogenetic analysis (Dean et al., 2022; Juan-Carlos et al., 2021). Among these, the ABCA subfamily is particularly noteworthy for its involvement in lipid transport and homeostasis. Members of the ABCA subfamily are primarily expressed in tissues with high lipid metabolic activity, such as the liver, lungs, and eyes (Kaminski et al., 2006). These transporters are integral to the maintenance of cellular lipid composition and the removal of toxic lipid metabolites, emphasizing their importance in both normal physiology and disease states. A number of inherited diseases are linked to genetic variants in the ABCA subfamily, including ABCA1 in familial high-density lipoprotein (HDL) deficiency and Tangier Disease, ABCA3 in pediatric interstitial lung disease and fatal surfactant deficiency, ABCA12 in lamellar and harlequin ichthyosis, and ABCA7 in Alzheimer’s disease (Biswas-Fiss et al., 2018; Dean et al., 2022; Juan-Carlos et al., 2021; Kaminski et al., 2006).

2. ABCA4 transporter protein

The ABCA4 transporter protein is a member of the ABCA subfamily, playing a key role in maintaining health of the retina and normal vision. It is encoded by the ABCA4 gene (OMIM #601691), located on chromosome 1p22, which spans approximately 128 kilobases of genomic DNA and consists of 50 exons. ABCA4 is predominantly expressed in the photoreceptor cells of the retina (Papermaster, Reilly, & Schneider, 1982), where it is responsible for transporting retinal byproducts generated during phototransduction from the photoreceptor outer segment discs into the cytoplasm for further processing (Fig. 1A) (Biswas-Fiss & Kurpad, 2010; Biswas-Fiss, Affet, Ha, & Biswas, 2012; Quazi & Molday, 2014; Quazi, Lenevich, & Molday, 2012; Tsybovsky & Molday, 2010). A recent study also suggests that ABCA4 is expressed in low amounts in the Retinal Pigment Epithelium (RPE), where it may have a functional role beyond merely being present due to photoreceptor outer segment shedding and phagocytosis (Lenis et al., 2018). ABCA4 function is critical for the visual cycle and the preservation of photoreceptor cell integrity, and defects in this process lead to retinal degenerative diseases (Allikmets et al., 1997; Allikmets, 2000b; Sun & Nathans, 2000, 2001; Sun, Smallwood, & Nathans, 2000; Zhang et al., 1999). Understanding the structure of ABCA4 is crucial for elucidating its functional mechanism and the pathogenic effects of various mutations.

Fig. 1. Localization and topological organization of the ABCA4 protein.

(A) ABCA4 is a crucial component of the visual cycle, localized in the outer segment discs of cone and rod photoreceptor cells in the retina. The protein plays a key role in transporting retinal byproducts of phototransduction from the photoreceptor outer segments into the cytoplasm. (B) The topological organization of ABCA4 within the lipid bilayer is depicted, showing the detailed domain structure. The protein is organized into two homologous halves, each containing an extracytoplasmic domain (ECD), a transmembrane domain (TMD), a nucleotide-binding domain (NBD), and a regulatory subdomain (RD). The figure illustrates the orientation of these domains with the ECDs protruding into the lumen and the NBDs positioned on the cytoplasmic side. Created in BioRender. Cevik et al. (2024). https://BioRender.com/a45u169.

2.1. Structural insights into ABCA4

ABCA4 is a full-length transporter composed of 2273 amino acids, organized into two homologous halves. Each half contains a transmembrane domain (TMD), an extracyoplasmic domain (ECD), a nucleotide-binding domain (NBD), and a regulatory domain (RD) (Fig. 1B). Within the outer segment discs of cone and rod photoreceptor cells, ABCA4 is positioned with the ECDs facing into the disc lumen and the NBDs oriented on the cytoplasmic side (Fig. 1A).

Recent cryo-electron microscopy (cryo-EM) studies have provided detailed insights into the molecular architecture of ABCA4 (Liu, 2021a; Liu, 2021b; Scortecci, 2023; Scortecci, Van, & Molday, 2021a; Scortecci, Van, & Molday, 2021b; Xie, Zhang, & Gong, 2021a; Xie, Zhang, & Gong, 2021b; Xie, Zhang, & Gong, 2021c). These studies have resolved the structure of ABCA4 in multiple conformational states, including the apo state, ATP-bound state, and substrate-bound forms, at resolutions ranging from 2.9 to 3.6 Å. These experimental structures reveal that ABCA4 adopts an elongated shape, extending approximately 230–240 Å across the lipid bilayer (Liu, Lee, & Chen (2021a); Scortecci et al., 2021; Xie, Zhang, Fang, Du, & Gong, 2021).

2.1.1. Transmembrane domains (TMDs)

The TMDs of ABCA4 consist of 12 transmembrane helices (six in each half), which form the core of the transporter and are responsible for substrate translocation (Liu et al. (2021a); Scortecci, Van Petegem, & Molday, 2023; Xie, Zhang, Fang, Du, & Gong, 2021). In the ATP-free state, the TMDs adopt an outward-facing conformation, creating a hydrophobic cavity that extends into both the lumen and the lipid bilayer, which is important for substrate recruitment. The substrate-binding pocket, located at the interface between the TMDs, is accessible from the lumen leaflet of the membrane (Fig. 2A) (Liu et al., 2021a; Scortecci et al., 2021; Xie et al., 2021). In the substrate-bound state, the TMDs maintain an outward-facing conformation, with the substrate occupying the large hydrophobic cavity, stabilized by hydrophobic and ionic interactions and capped by a loop from ECD1 (Scortecci et al., 2021; Xie et al., 2021). Upon ATP binding, the TMDs undergo a significant conformational shift to an inward-facing state, where the two TMDs come into close contact, collapsing the substrate-binding cavity and forming a tightly sealed structure (Fig. 2A) (Liu et al., 2021a; Scortecci et al., 2021; Xie et al., 2021).

Fig. 2. Structural highlights from the cryo-EM studies of ABCA4.

(A) Surface representation of the ABC transporter in the ATP-free state (left panels) and the superimposed ATP-free (cyan) and ATP-bound (gray) states (right panels), demonstrating the substrate-binding pocket (red), which is located at the interface between the TMDs and ECDs and is accessible from the luminal leaflet of the membrane. Upon ATP binding (right), the TMDs undergo a conformational shift to an inward-facing state, collapsing the substrate-binding cavity and forming a tightly sealed structure. The previously exposed hydrophobic cavity is blocked, and the two TMDs come into close contact. (Demonstrated on NRPE-bound and ATP-bound cryo-EM structures, 7e7o, 7e7q (Xie, 2021a, 2021b)). (B) The domain-swapping arrangement of the nucleotide-binding domains (NBDs) and regulatory subdomains (RDs). The NBD1 and NBD2 (blue and green) are linked to regulatory domains RD1 and RD2, forming a cross-domain arrangement, where RD1 (an extension of NBD1) is positioned beneath NBD2, and RD2 (an extension of NBD2) beneath NBD1. The close-up views on the right depict the intricate interdomain interactions stabilized by hydrogen bonds (red lines) and salt bridges (blue lines), which are almost exclusively formed upon ATP binding, reinforcing the cross-domain interaction network. Demonstrated on ATP-bound cryo-EM structure, 7lkz (Liu, 2021c). Created in BioRender. Cevik et al. (2024). https://BioRender.com/g84x856.

2.1.2. Extracytoplasmic domains (ECDs)

The ECDs of ABCA4 are large, glycosylated regions located above the TMDs. These domains exhibit a complex structure with multiple regions, including a base, a tunnel, and a lid (Liu, Lee, & Chen, 2021b; Scortecci et al., 2021; Xie et al., 2021). The tunnel region, which is accessible from the lumen side, is lined with hydrophobic residues and may play a role in substrate passage (Scortecci et al., 2021; Xie et al., 2021). However, due to their high flexibility, the ECDs are the least well-resolved structures. Despite this, ECDs exhibit clear conformational changes between the ATP-bound and NRPE-bound states (Xie et al., 2021). The ECDs are stabilized by several polar interactions and disulfide bridges, crucial for maintaining the structural integrity of the transporter (Liu, Lee, & Chen, 2021c; Scortecci et al., 2021). The presence of glycosylation sites further suggests that the ECDs may play a role in the proper folding and trafficking of ABCA4.

2.1.3. Nucleotide-binding domains (NBDs)

The NBDs of ABCA4 are highly conserved among ABC transporters, containing key motifs required for ATP binding and hydrolysis, such as the Walker A and B motifs, the ABC signature motif, and various loops involved in ATP coordination (Tsybovsky & Molday, 2010). In the absence of ATP, the NBDs remain separated, allowing the TMDs to adopt an outward-facing conformation. ATP binding, however, induces a large tweezer-like conformational change, causing the NBDs to dimerize and bringing the TMDs into close proximity, facilitating substrate translocation across the membrane (Scortecci et al., 2021; Xie et al., 2021). The NBDs are also linked to regulatory domains (RDs), which could potentially modulate transporter activity. Structurally, RD1 and RD2 display a domain swap, where RD1 (an extension of NBD1) is positioned beneath NBD2, and RD2 beneath NBD1. This cross-domain arrangement is further stabilized by ATP binding, reinforcing an interdomain interaction network (Fig. 2B). Additionally, the RDs exhibit an “ACT-like fold” with the characteristic βαββαβ topology (Liu, Lee, & Chen, 2021d), which may play a crucial role in regulating the protein’s functional activity.

While cryo-EM studies have yielded a wealth of structural information, their static nature limits the ability to capture dynamic events involved in the entire transport cycle. One of the unresolved questions is how the regulatory domains, RD1 and RD2, modulate ABCA4’s activity. The current understanding of these domains as “regulatory” is largely inferred from their sequence homology and structural classification of domain organization, yet their precise mechanistic role remains elusive. Our hypothesis is that the unresolved sections of both RD1 and RD2, which seem to extend toward TMD1 and the ATP/Mg binding sites, respectively, may undergo conformational changes upon substrate binding or passage, and this conformational shift could be the mechanism behind ATPase activity stimulation in the presence of the retinal substrate (Fig. 3).

Fig. 3. Proposed regulatory mechanism of ABCA4 functional activity highlighting the RD1 and RD2 domains and their unresolved regions as observed in all available cryo-EM studies.

(A) The unresolve part of the RD1 (in red in apo state (PDB: 7e7i), and blue in NRPE-bound state (PDB:7e7o)) extend between the interaction network of the NBDs-RDs and the intracellular helix 3 (IH3), which connects to transmembrane domain 1 (TMD1). (B) The unresolved portion of the RD2 (in yellow in apo state (PDB: 7e7i), and purple in NRPE-bound state (PDB: 7e7o)) appears to extend towards the ATP and Mg binding site. The unresolved segments may undergo conformational shifts upon retinal substrate binding or passage, potentially facilitating ATPase activity stimulation. ATP (in brown) and Mg (in green) are shown docked into the structure by superimposing the ATP-bound structure (PDB: 7e7q) onto others, with the cartoon representation hidden for visual clarity. Created in BioRender. Cevik et al. (2024). https://BioRender.com/l17h721.

Furthermore, several regions remain unresolved in the cryo-EM structures and, when modeled, show random coil structures that likely represent intrinsically disordered regions (IDRs). IDRs often provide the necessary flexibility for protein dynamics and are crucial for facilitating interactions with small molecules (Dunker et al., 2008). ABCA4 appears to have several of these disordered regions, notably in the N-terminal cytoplasmic side at amino acid locations 862–914, 1162–1203, and 1280–1340. These regions correspond to the TMD1-NBD1, NBD1-RD1, and RD1-IH3 interfaces, respectively (Fig. 4). Interestingly, the individually expressed and purified N-terminal cytoplasmic region of ABCA4 has been shown to interact exclusively with 11-cis-retinal (Biswas-Fiss et al., 2012). In addition, ABCA4 has also been shown to clear excess 11-cis-retinal by facilitating its transport in the form of 11-cis-N-ret-PE from the lumen to the cytoplasmic side of the membrane (Quazi & Molday, 2014). These additional IDRs on the N-terminal cytoplasmic side—the only significant difference from its otherwise symmetric counterpart, the NBD2 side—may provide insights into the exclusive 11-cis-retinal interaction that has been reported, or possibly suggest other functional roles for these regions. Investigating the role of these IDRs, particularly in the context of retinal molecule interactions, could reveal novel aspects of ABCA4’s molecular mechanisms.

Fig. 4. Unresolved random coils in the cytoplasmic side of the ABCA4 structure, likely representing intrinsically disordered regions (IDRs).

(A) Random coil regions are shown in blue on the NBD1 side and in green on the NBD2 side of the cytoplasm. The model was generated by separately modeling the first (N-terminus) and second halves (C-terminus) of the protein using AlphaFold2 and then superimposing them onto the cryo-EM structure (PDB: 7e7i). (B) Superimposition of the N-terminal and C-terminal cytoplasmic regions demonstrates that the N-terminal side contains longer disordered regions than the otherwise structurally identical C-terminal side. These disordered regions are located at amino acids 862-914, 1162-1203, and 1280-1340, corresponding to the TMD1-NBD1, NBD1-RD1, and RD1-IH3 interfaces, respectively, and may play a crucial role in ABCA4's function. Created in BioRender. Cevik et al. (2024). https://BioRender.com/z81n658.

2.2. Functional role of ABCA4 in the visual cycle and retinal health

Phototransduction is a process by which light is converted into electrical signals, enabling vision. This cascade begins when the G-protein-coupled receptor rhodopsin binds to 11-cis-retinal, a derivative of vitamin A, and triggers the conversion of light into an electrical signal (Lamb & Pugh, 2006).

Following the activation of rhodopsin, 11-cis-retinal is isomerized into all-trans-retinal (Palczewski, 2006; Sparrow, Wu, Kim, & Zhou, 2010). To reset the photoreceptor to its light-sensitive state, all-trans-retinal must be removed and recycled. However, all-trans-retinal is highly reactive and can form toxic compounds when it interacts with membrane lipids (Sparrow et al., 2010).

Once all-trans-retinal is released into the lipid bilayer of the photoreceptor outer segment disks, it forms a reversible covalent bond with phosphatidylethanolamine (PE), creating N-retinylidene-phosphatidylethanolamine (N-ret-PE) (Sparrow et al., 2010; Xu, Molday, & Molday, 2023). ABCA4 primarily functions by flipping N-ret-PE from the inner to the outer leaflet of the photoreceptor disk membrane, initiating the first step in the visual cycle—the regeneration of 11-cis-retinal— which is completed through subsequent enzymatic reactions in the retinal pigment epithelium (RPE) (Palczewski, 2012; Xu et al., 2023).

Dysfunction in ABCA4 disrupts the normal flipping of N-ret-PE across the photoreceptor disk membrane (Sun & Nathans, 2000, 2001; Sun et al., 2000). When genetic variants in ABCA4 result in a deficient or defective protein, or when the protein fails to be properly targeted to the membrane, N-ret-PE accumulates on the inner leaflet of the disk membrane, and in the RPE cells. This accumulation increases the likelihood of a second retinaldehyde molecule reacting with N-ret-PE, leading to the formation of an irreversible and insoluble bis-retinoid compound known as A2-phosphatidylethanolamine (A2PE) (Crouch, Koutalos, Kono, Schey, & Ablonczy, 2015; Sparrow et al., 2010).

A2PE is a precursor to di-retinoid-pyridinium-ethanolamine (A2E), a toxic bis-retinoid that forms after the hydrolysis of A2PE, which is highly detrimental to photoreceptor cells and RPE (Sparrow et al., 2010). It contributes to the buildup of lipofuscin, a fluorescent pigment that accumulates within and beneath the RPE, leading to cellular dysfunction and death (Crouch et al., 2015). The presence of A2E and other bis-retinoids is directly linked to the formation of characteristic pisciform flecks observed in patients with ABCA4-related retinal diseases (Al-Khuzaei et al., 2021).

Moreover, A2E and related bis-retinoids are phototoxic and can generate reactive oxygen species (ROS) when exposed to light, exacerbating oxidative stress and further damaging retinal cells (Marie et al., 2018). Therefore, ABCA4 plays a critical role in efficiently clearing these reactive molecules and recycling 11-cis-retinal for the phototransduction process, ultimately ensuring retinal health (Fig. 5).

Fig. 5. Functional role of ABCA4 in retinal health and pathology.

The left panel represents the healthy retina, where ABCA4 mediates the efficient translocation of N-retinylidene-phosphatidylethanolamine (N-ret-PE) across the photoreceptor disk membrane, enabling the recycling of all-trans-retinal into 11-cis-retinal as part of the visual cycle. This process ensures that toxic compounds are efficiently cleared from the photoreceptor outer segment, maintaining retinal health. The right panel illustrates retinopathy, where dysfunction of ABCA4 due to genetic variants leads to the accumulation of N-ret-PE on the inner leaflet of the disk membrane, resulting in the formation of A2E, a toxic bis-retinoid. A2E contributes to lipofuscin buildup and photoreceptor damage, exacerbated by oxidative stress. This accumulation is linked to the characteristic features of ABCA4-related retinal diseases. Created in BioRender. Cevik et al. (2024). https://BioRender.com/e66h359.

2.3. ABCA4 genetic variants and disease association

Pathogenic variants in ABCA4 have been linked to a series of inherited retinal diseases (IRDs), including Stargardt macular dystrophy (STGD1, OMIM #248200), cone-rod dystrophy (CRD, OMIM #604116), autosomal recessive retinitis pigmentosa (arRP, OMIM #601718), and are also believed to contribute to the complex disease age-related macular degeneration (AMD) (Al-Khuzaei et al., 2021; Allikmets & Dean, 2008; Allikmets & Shroyer, 1997; Allikmets et al., 1997; Allikmets, 2000a; Azarian & Travis, 1997; Cremers et al., 1998; Klevering et al., 2004; Koenekoop, 2003; Lewis et al., 1999; Mullins et al., 2012; Weng et al., 1999; Zhang et al., 1999).

Stargardt disease is the most common ABCA4-related retinal disorder, characterized by the early onset of central vision loss, accumulation of lipofuscin in the retinal pigment epithelium (RPE), and the presence of pisciform flecks around the macula (Al-Khuzaei et al., 2021). It is considered a relatively common “rare genetic disease,” affecting roughly 1 in 8000 to 10,000 individuals globally (Raimondi et al., 2023; Runhart et al., 2022). CRD is marked by the degeneration of cone photoreceptors, which leads to an initial loss of color and central vision, eventually followed by rod photoreceptor involvement. In contrast, RP typically begins with the degeneration of rod photoreceptors, causing night blindness and peripheral vision loss before affecting central vision later in the disease progression (Gill, Georgiou, Kalitzeos, Moore, & Michaelides, 2019; Van Huet et al., 2013).

To date, over 4000 variants in the ABCA4 gene have been identified (https://www.ncbi.nlm.nih.gov/clinvar/, accessed on 31 October 2024), contributing to the genotypic and phenotypic heterogeneity of ABCA4-related IRDs. This wide variability in clinical presentations can be attributed in part to the severity of individual mutations and the combined effect of unique combination of alleles inherited by each patient (Cornelis et al., 2017; Khan et al., 2020; Shroyer, Lewis, Yatsenko, Wensel, & Lupski, 2001; Zhang et al., 2015). Some ABCA4 variants are associated with milder phenotypes or later disease onset, which may be due to the retention of partial protein function. Conversely, more severe mutations that result in a complete loss of ABCA4 function tend to cause earlier onset and more aggressive progression of retinal degeneration (Curtis, Molday, Garces, & Molday, 2020; Garces et al., 2018; Klevering et al., 1999; Klevering et al., 2002; Klevering, Deutman, Maugeri, Cremers, & Hoyng, 2005; Lee et al., 2017; Lewis et al., 1999; Xiao, Ye, Chen, Zheng, & Yuan, 2022; Zernant et al., 2018; Zhu et al., 2021). The complex relationship between genotype and phenotype in ABCA4-related diseases is also believed to be influenced by factors such as modifier genes, and environmental conditions (Al-Khuzaei et al., 2021). This complexity creates significant challenges in the diagnosis, prognosis, and management of these disorders, highlighting the importance of developing a comprehensive understanding of the underlying molecular mechanisms.

Various types of mutations have been reported in the ABCA4 gene, including missense, nonsense, frameshift, and splice-site changes, as well as larger deletions and duplications. Missense variants are particularly noteworthy due to their prevalence and the challenges in determining their functional impact compared to other mutation types. Consequently, a significant number of ABCA4 missense variants remain classified as variants of uncertain significance (VUS). These missense variants are dispersed throughout the entire open reading frame (ORF) (Fig. 6), and can affect multiple aspects of ABCA4 protein function, such as substrate binding, ATP hydrolysis, membrane targeting, and protein stability.

Fig. 6. Structural distribution of pathogenic missense variants in ABCA4.

The pathogenic missense variants (blue spheres) are dispersed throughout the ABCA4 protein. These variants can be associated with a range of deleterious effects, such as reduced protein stability, conformational changes, impaired ATP binding and hydrolysis, as well as defects in substrate recognition and transport, contributing to the pathogenicity of ABCA4-related retinal diseases. Created in BioRender. Cevik et al. (2024). https://BioRender.com/m61u337.

3. Computational tools and in silico methods in predicting variants’ pathogenicity

The rapid expansion of genomic data has uncovered a vast array of genetic variants, many of which remain classified as variants of uncertain significance (VUS), posing a challenge to clinical interpretation. In genes like ABCA4, where more than 800 VUS have been identified, addressing this uncertainty is critical for advancing precision medicine. Computational tools have become indispensable in predicting the pathogenicity of these variants, providing valuable insights when clinical data is unavailable or experimental validation is impractical.

Pathogenicity prediction tools such as PolyPhen-2 (Polymorphism Phenotyping v2) (Adzhubei et al., 2010), SIFT (Sorting Intolerant From Tolerant) (Sim et al., 2012), and MutationTaster (Schwarz, Cooper, Schuelke, & Seelow, 2014) serve as initial screening methods for variant assessment. These tools primarily rely on evolutionary conservation, sequence homology, and biochemical properties to predict whether a variant is likely to be pathogenic. Comprehensive tools like CADD (Combined Annotation Dependent Depletion) (Rentzsch, Witten, Cooper, Shendure, & Kircher, 2019) enhance this assessment by integrating a wide range of annotations into a single metric, offering a robust measure of deleteriousness. More recent machine learning-based ensemble methods, such as REVEL (Rare Exome Variant Ensemble Learner) (Ioannidis et al., 2016), have further refined predictive accuracy by combining multiple algorithms into a cohesive model.

Despite these advancements, these tools still have inherent limitations. Notably, they do not explain the underlying mechanisms of pathogenicity and are often constrained by the available annotations and training data, which may not fully account for the complexity of genetic variants or their interactions within diverse biological contexts (Ittisoponpisan et al., 2019). This is especially relevant for transporter membrane proteins like ABCA4, where the structural impacts of variants can deviate significantly from those in soluble proteins (Iqbal et al., 2020).

To overcome these limitations, molecular modeling and structural prediction techniques, in conjunction with using available experimental structure models to infer the impacts of variants have become increasingly important. Especially for missense variants, these methods offer a biological context, and more nuanced understanding of how genetic variants alter protein architecture and function. This approach allows researchers to visualize conformational changes, predict molecular interactions, and assess protein stability, providing a complementary layer of evidence to traditional prediction tools.

4. Approaches to studying ABCA4 membrane protein in vitro

Initial studies focused on expressing individual ABCA4 domains in bacterial systems to biochemically characterize the wild-type (WT) protein and its variants (Biswas & Biswas, 2000; Biswas-Fiss & Kurpad, 2010; Biswas-Fiss et al., 2012; Biswas-Fiss, 2003, 2006). While these efforts provided valuable insights into the roles of specific domains, motifs and residues, they were limited in scope, as they did not fully capture the functional complexity and interactions of the entire protein.

A major challenge in studying full-length ABCA4 in vitro is the difficulty of solubilizing and reconstituting this large membrane protein into artificial membranes. With 12 transmembrane helices, ABCA4 requires a lipid environment that preserves its native conformation and functionality. Solubilization and reconstitution processes can disrupt crucial lipid-protein interactions, which may lead to alterations in protein conformation and function (Hardy, Desuzinges Mandon, Rothnie, & Jawhari, 2018). Furthermore, reproducing the balanced lipid composition and conditions of native cellular membranes is challenging, and deviations from the natural context may result in misleading conclusions regarding the protein’s behavior and the characterization of its variants (Hardy et al., 2018; Jahn & Radford, 2005).

Due to these limitations, less than 10 % of ABCA4 variants have been functionally characterized at the protein level using in vitro assays, including enzymatic assays, expression, and localization studies (Ahn, Beharry, Molday, & Molday, 2003; Curtis et al., 2020; Garces et al., 2018; Garces, Scortecci, & Molday, 2020; Kim et al., 2022; Molday, Wahl, Sarunic, & Molday, 2018; Sun et al., 2000; Xu et al., 2023) (https://www.hgmd.cf.ac.uk/ac/index.php). This leaves a significant gap in understanding how specific variants impact the protein’s function and their contribution to disease. Therefore, alternative strategies are necessary to advance the functional characterization of the many uncharacterized ABCA4 variants.

4.1. Domain-specific approaches for functional assessment of ABCA4

Studies using purified and reconstituted bovine or human ABCA4 expressed in mammalian cell lines have provided experimental systems for probing ABCA4 function (Ahn et al., 2003; Beharry, Zhong, & Molday, 2004; Sun et al., 2000). However, these whole-molecule studies have limitations due to ABCA4’s large size, low abundance, and rapid loss of stability and activity upon removal from native ROS disk membranes (Tsybovsky, Wang, Quazi, Molday, & Palczewski, 2011). In a multifunctional protein, such as ABCA4, it is difficult to assign the true raison d’être of each domain using this approach. Expressing and characterizing individual functional domains has been shown to be an effective method in studies of other ABC proteins, such as the MDR1 and CFTR transporters (Chen & Bahl, 1993; Duffieux et al., 2000; Ling, 1997; Zhang et al., 2008). ABCA4, a particularly large membrane protein (>220 kDa), presents unique experimental challenges due to its bipartite structure. Consequently, systematic analysis of each individual domain’s structure and function in recombinant form is a valuable and precise approach, as demonstrated by previous studies on the NBD1/NBD2 and ECD1/ECD2 domains of ABCA4 and by similar work in vision research (Murakami, Frey, Lin, & Antonetti, 2012).

These approaches have been used to provide mechanistic details on the distinct nucleotide binding and hydrolysis aspects of the NBDs as well as to demonstrate the retinoid binding properties of ECD2 and NBD1 (Biswas-Fiss, Kurpad, Joshi, & Biswas, 2010; Biswas-Fiss et al., 2012).

4.2. Virus-like particles to express and functionally characterize ABCA4 and its variants

Given the technical challenges in studying full-length ABCA4, such as solubilization and reconstitution, alternative approaches are being explored to facilitate the functional characterization of ABCA4 VUS. One such approach is the use of virus-like particles (VLPs), a versatile platform for expressing membrane proteins in a controlled and near-native environment. VLPs represent a robust tool for studying ABCA4 because they offer several advantages over traditional in vitro systems.

VLPs are non-infectious particles that resemble the structure of viruses but lack viral genetic material, making them a safe and effective system for protein expression (Jeong & Seong, 2017; Sari-Ak et al., 2019; Sari-Ak et al., 2021). Structurally, VLPs can be classified as either enveloped or non-enveloped (Jeong & Seong, 2017). Non-enveloped VLPs consist of viral capsid proteins that spontaneously assemble into particles, whereas enveloped VLPs are enclosed by a lipid bilayer originating from the host cell membrane. This lipid envelope makes enveloped VLPs (eVLPs) particularly suitable for studying membrane proteins, as they closely mimic the native lipid bilayer found in cellular membranes, allowing for critical lipid-protein interactions for proper enzymatic function that are often disrupted in other artificial membrane systems (Cevik, Biswas, Ghosh, & Biswas-Fiss, 2024).

Another major advantage of VLPs is their ability to provide a consistent and uniform orientation of membrane proteins due to the nature of the budding process from the host cell plasma membrane (Cevik et al., 2024; Sari-Ak et al., 2021). This orientation uniformity is critical for functional assays, as incorrect orientation can lead to misleading functional data or a complete lack of activity.

This approach simplifies the expression and characterization of membrane proteins, enabling studies to be conducted under conditions that more accurately reflect the native environment. As a result, VLPs offer a practical and scalable solution to overcoming the challenges of solubilization and reconstitution, which can significantly accelerate the pace of in vitro studies to characterize ABCA4 variants and advance our understanding of their role in retinal diseases.

5. Comprehensive strategies for assessing ABCA4 variant pathogenicity using multi-modal approaches

5.1. Bioinformatic exploration of ECD2 domain specific variants

ABCA4’s extracytoplasmic domain 2, ECD2 extends from the transmembrane domain of the ABCA4 protein in a loop-like structure and has been shown to bind the retinoid all-trans-retinal (ATR), a byproduct of the phototransduction cycle (Biswas-Fiss et al., 2010). Recent cryo-EM studies have reported that ECD2 forms a lid-tunnel-base structure with ECD1, suggesting the domain’s role in retinoid interaction and translocation (Liu et al., 2021; Scortecci et al., 2021; Xie et al., 2021). Of the more than 1600 missense genetic variants in ABCA4, over 200 are located within the ECD2 domain, underscoring its critical role in protein function (Borras et al., 2017; Nicora et al., 2018; Richards et al., 2015; Starita et al., 2017). The 270-amino-acid ECD2 domain lacks enzymatic activity and does not contain any known functional motifs, making the analysis of single nucleotide variants (SNVs) challenging. Extensive studies are required to better understand the ECD2 domain, its role in ECD2-retinoid binding, and its impact on ABCA4 function and disease.

A total of 207 ECD2 variants were identified on NIH’s ClinVar database by querying “ABCA4.” ECD2 variants were then identified based on their amino acid locations within the domain (Landrum et al., 2018). Due to the high number of ECD2 missense variants, prioritizing these variants for further analysis is essential. This prioritization was achieved using first-line, protein-based informatics tools.

5.1.1. Consistency of tools predicting the impact of single nucleotide variants (SNVs) on ABCA4 functionality in the ECD2 domain

Single nucleotide variants (SNVs) significantly impact protein function, and accurately predicting the extent of this impact remains a critical challenge in bioinformatics. Consistency across bioinformatic tools reduces the likelihood of errors and improves confidence in the results. To examine congruence across pathogenicity prediction tools, ABCA4-ECD2 variants were entered into the following predictive software: Polyphen2, SIFT, Mutation Taster (Adzhubei, Jordan, & Sunyaev, 2013; Ng & Henikoff, 2001; Steinhaus et al., 2021). Pathogenicity results for each variant from each database were gathered in tabular form. Variants were further classified according to the number of pathogenic scores received from the corresponding databases. For instance, zero pathogenic results classified a variant as pathogenic across zero tools (PAZT) while one pathogenic result classified a variant as pathogenic across one tool (PAOT), and two pathogenic results classified a variant as pathogenic across two tools (PATT). Lastly variants with three pathogenic results were labeled as pathogenic across all tools (PAAT) which signified the most potential impact on the function of the protein (Fig. 7). In silico predictors identified 75 ECD2 PAAT variants in total. These PAAT variants span the entirety of the ECD2 domain and represent a pool of potential “high-impact” targets to be functionally studied.

Fig. 7. Prioritization of ECD2 Variants using in silico tools (workflow).

ECD2 missense variants were first collected from NIH’s Clinvar database and then evaluated by the following predictive tools: Polyphen2, Mutation Taster, and SIFT. Prediction scores from all three databases were recorded for each variant. The variants were then categorized according to the number of pathogenic scores received. PAOT (pathogenic across one tool) indicated that the variant received only one pathogenic score and PATT (pathogenic across two tools). Lastly PAAT (pathogenic across all tools) represented a distinct group which received three pathogenic scores. Created in BioRender. Jones, (2024). https://BioRender.com/d79c319.

5.1.2. Predicting functionally significant residues based on sequence conservation: identification of critical conserved motifs in ECD2

Considering that the ABCA subfamily members have similar domains and topology, it follows that many of these domains may contain conserved sequences. The ECD2 domain was thus studied for the presence of conserved regions that may potentially contribute to the overall function of the domain.

ABCA family members were selected and aligned using the multiple sequence alignment tool on the Uniprot database (UniProt, 2023). The alignment was isolated to the ECD2 domain and critical conserved motifs (CCMs) were defined as sequence blocks that were present (conserved) in all ABCA family members (Fig. 8). The CCMs (7 identified) are also potential targets as they have remained in the ECD2 sequence throughout the evolution of the ABCA subfamily. This strongly implicates the CCMs in the functioning of the ECD2 domain as well as the ABCA4 protein. In conjunction, PAAT ECD2 variants were also mapped to the CCMs using a bar graph-like diagram (Fig. 9). The CCMs were found to contain over 52 % of PAAT variants, which further implicated the importance of the conserved sequences to the ECD2 domain.

Fig. 8. Multiple sequence alignment of human ABCA4 and ABCA4 homologues.

ECD2 in human ABCA4 was compared to the ECD2 domain in ABCA4 of other species. The same ECD2 domain is also shown in other ABCA subfamily members (ABCA1, ABCA7, and ABCA8). Conserved regions (7 observed) were labeled as critical conserved motifs (CCMs). CCMs and conserved amino acids are highlighted in blue and green respectively. The ECD2 sequence is indicated by the red arrow directly under the alignment. Created in BioRender. Jones, (2024). https://BioRender.com/n47n168.

Fig. 9. PAAT variants mapped to the ECD2 domain.

PAAT variants are shown alongside the ECD2 domain (in blue). Variants within their perspective CCM (underlined) are shown on the right while those variants outside of CCMs are shown on the left. 51 % of PAAT variants were found to be present in a CCM versus another region. Created in BioRender. Jones, (2024). https://BioRender.com/r69n879.

The number of ECD2 missense variants continues to increase, therefore it is important to establish methods to prioritize them for future studies in exploring their impact on ABCA4 protein structure and function as well as predicting their contribution to IRD-pathogenicity.

5.2. Integrated in silico and in vitro approaches to assess ABCA4 variants at protein-level

The disproportionate prevalence of missense VUS among all ABCA4 variants emphasizes the need for a focused study of these specific variants.

This section expands on the previously developed in silico pipeline (Cevik, Biswas, & Biswas-Fiss, 2023; Cevik, Wangtiraumnuay, et al., 2023) and virus-like particle (VLP)-based functional assays (Cevik et al., 2024), applying these methodologies to a selected subset of ABCA4 missense VUS. Specifically, we assess the pathogenicity of seven VUS (p.Y345D, p.Y603F, p.L844R, p.G1559E, p.Y1889N, p.E2031K, p.Y2165C). Three of these VUS (p.Y603F, p.G1559E, p.Y2165C) were structurally analyzed previously (Cevik et al. 2023), and selected here for in vitro validation due to their structural significance, are now evaluated functionally. Additionally, a benign variant (p.T1428M), which was structurally assessed previously, serves as a control (Cevik et al. 2023). The remaining VUS were selected based on initial structural assessment, indicating potential deleterious effects.

Of particular interest are the four VUS involving tyrosine (Tyr) substitutions. Tyrosine, an aromatic amino acid, plays a crucial role in protein stability and function due to its unique phenolic hydroxyl group, which enables various interactions. Its amphipathic nature facilitates membrane protein orientation, anchoring the protein within the lipid bilayer. In ABCA4, Tyr residues are typically located at the interface between the lipid bilayer and the aqueous environment, where they may contribute to the protein’s structural integrity (Fig. 10). Many Tyr side chains in ABCA4 form hydrogen bonds, likely stabilizing interactions between key domains (Fig. 10). In addition to hydrogen bonding, the aromatic ring of tyrosine is involved in non-covalent interactions, such as π-π and cation-π interactions. These interactions are thought to play a significant role in substrate binding, as observed in cryo-EM structures (Fig. 10) (Xie et al., 2021). Substituting Tyr with a non-aromatic residue in these regions could disrupt these stabilizing interactions, potentially impairing the protein’s function. Moreover, the extracytoplasmic domain 1 (ECD1) of ABCA4 contains a disproportionate number of tyrosine residues—23 out of a total of 65—suggesting that Tyr in this domain may be of particular interest for functional studies (Fig. 10).

Fig. 10. Tyrosine (Tyr) residues of ABCA4.

Full-length structure of ABCA4 depicting Tyr residues (pink spheres), highlighting their unique localization pattern. The extracytoplasmic domain 1 (ECD1), shown in the top-right rectangle, contains the highest density of Tyr residues, particularly at the base region of the base-tunnel-lid structure formed by the ECDs. The bottom-right rectangle presents a close-up view of the transmembrane region, where Tyr residues form hydrogen bonds (dashed lines) and other non-covalent interactions (navy blue sticks) with NRPE and PE, contributing to the stability of transmembrane domains and the structural integrity of ABCA4. Created in BioRender. Cevik et al. (2024). https://BioRender.com/a94l055.

Overall, missense mutations of Tyr residues could have significant implications for ABCA4’s structural and functional integrity, disrupting hydrophobic and hydrophilic interactions, destabilizing domain-domain communication, and impairing membrane orientation, ultimately leading to protein dysfunction.

In sections 5.2.1 and 5.2.2 of this chapter, we report on the use of a multi-modal approach to elucidate the functional consequences of these specific missense VUS in ABCA4, providing critical insights that may inform their reclassification under the ACMG/AMP guidelines for pathogenicity.

5.2.1. Computational protein structure analysis and pathogenicity prediction

The structural analysis of the selected ABCA4 variants was conducted using available cryo-EM structures (Liu, 2021a; Liu, 2021b; Scortecci, 2023; Scortecci et al. 2021a; Scortecci et al. 2021b; Xie et al. 2021a; Xie et al. 2021b; Xie et al. 2021c), and AlphaFold2 models (Jumper et al., 2021; Mirdita et al., 2022) as detailed previously (Cevik et al. 2023; Cevik, Wangtiraumnuay, et al., 2023). Structural changes were visualized using PyMOL2 software (Schrodinger, 2015), while pathogenicity predictions were performed using PolyPhen-2 (Adzhubei et al., 2010), REVEL (Ioannidis et al., 2016), CADD (Rentzsch et al., 2019), AlphaMissense (Cheng et al., 2023), and MutPred2 (Pejaver et al., 2020). As part of the pathogenicity prediction pipeline, allele frequencies were sourced from the GnomAD database (Karczewski et al., 2020), and evolutionary conservation scores were obtained from ConSurf (Ashkenazy et al., 2016).

The computational structural analysis provided key insights into the potential impacts of ABCA4 variants in this study. The p.T1428M variant was previously evaluated, revealing no structural alterations, consistent with its benign classification (Fig. 11A) (Cevik et al. 2023). In contrast, all other variants displayed distinct structural defects, indicative of potential pathogenicity (Fig. 11).

Fig. 11. In silico structure analysis of variant proteins.

(A) The p.Y345D variant, located near the retinal substrate binding site, likely disrupts the hydrophobic pocket and impairs the interaction with the ATR moiety. (B) The p.L844R variant, found in transmembrane domain 1, replaces a lipid exposed leucine with a positively charged arginine, either introducing a hydrophilic side chain to the hydrophobic environment (relative solvent accessibility is 60 %) or leading to steric clashes in the all internally positioned rotamers. (C) The p.Y1889N variant, located in transmembrane domain 2, loses a possible hydrogen bond with the “EH3-turn-EH4 insertion, which could compromise protein stability. (D) The p.E2031K variant in nucleotide-binding domain 2 causes the loss of a possible salt bridge with intracellular helix 3, a bond present only in the ATP-bound state. Wild-type residues are shown as blue sticks, and substitutions are highlighted in red. Created in BioRender. Cevik et al. (2024). https://BioRender.com/t87b994.

The p.Y345D variant, located near the retinal substrate binding site, demonstrated a disrupted interaction with the ATR moiety, potentially impairing retinal transport (Fig. 11B).

For p.Y603F, the substitution of tyrosine with phenylalanine resulted in the loss of a key interdomain hydrogen bond between ECD1 and ECD2, likely compromising protein stability and domain interactions (Fig. 11C) (Cevik et al. 2023).

The p.L844R variant, located in transmembrane domain 1 (TMD1), replaces a hydrophobic leucine residue that is exposed to the lipid bilayer, with a positively charged arginine. Structure analysis evaluated the potential for the arginine side chain to reposition inward to avoid unfavorable contact with the lipid environment. However, the arginine could not be repositioned without inducing steric clashes or significant conformational changes. Therefore, this substitution introduced a charged residue into a hydrophobic environment, potentially leading to disruption of the membrane interaction (Fig. 11D).

The p.G1559E showed substantial steric clashes that were unavoidable across multiple rotamer positions (Fig. 11E) (Cevik et al. 2023).

Two variants exhibited potential pathogenic effects by disrupting critical domain-domain interactions. The p.Y1889N variant, located in transmembrane domain 2 (TMD2), resulted in the loss of a stabilizing hydrogen bond with an “EH-turn-EH insertion” (Fangyu Liu et al., 2021), reducing the stability (∆∆G = +3.58 kcal/mol) (Fig. 11F). Similarly, the p.E2031K variant in nucleotide-binding domain 2 (NBD2), led to the loss of a salt bridge with intracellular helix 3 (IH3), which is only present in the ATP-bound conformation (Fig. 11G).

Lastly, the p.Y2165C variant, located in regulatory domain 2 (RD2), disrupted both intra- and interdomain hydrogen bonds between RD1 and RD2 (Fig. 11H) (Cevik et al. 2023).

All seven VUS involved substitutions of highly evolutionarily conserved residues, as indicated by Consurf analysis (Ashkenazy et al., 2016), and showed consensus predicted pathogenicity across multiple computational tools, including PolyPhen-2, REVEL, AlphaMissense, CADD, and MutPred2.

5.2.2. In vitro expression and functional assessment of ABCA4 variants using the VLP platform

The in vitro expression of ABCA4 variants (c .1033 T > G (p.Tyr345Asp), c.1808A > T (p.Tyr603Phe), c .2531 T > G (p.Leu844Arg), c .4283 C > T (p.Thr1428Met), c .4676 G > A (p.Gly1559Glu), c .5665 T > A (p.Tyr1889Asn), c .6091 G > A (p.Glu2031Lys), and c.6494 A > G (p.Tyr2165Cys)) was conducted using the virus-like particle (VLP) platform with a baculovirus-mediated expression vector system in insect cell culture, as previously described (Cevik et al., 2024). Briefly, variant constructs were generated through site-directed mutagenesis and confirmed by sequencing. The recombinant proteins were expressed in High Five^™ insect cells, which simultaneously produced VLPs containing the ABCA4 variants. The ABCA4-VLPs were harvested from the culture medium and purified using differential centrifugation followed by ultracentrifugation. The expression levels, VLP targeting, and functional activity of the variants were assessed using Western blot analysis and enzymatic assays, following established protocols (Cevik et al., 2024). Data were reported as mean ATP hydrolysis rates ± standard deviation (SD) based on biological replicates. Statistical significance was assessed using one-way ANOVA separately for basal and retinal-stimulated activity, with significance set at p < 0.05.

5.2.2.1. Expression of ABCA4 variants

The uniform baculovirus infection and expression of all eight variants was successfully achieved, as confirmed by fluorescence signals and WBA of the High Five^™ insect cell lysates (Fig. 12).

Fig. 12. Expression and VLP-incorporation analysis of the variants.

(A) Fluorescence microscopy images of High5 insect cells showing the mCherry reporter expression, confirming successful and uniform baculovirus infection. (B) Western blot analysis of benign variant p.T1428M and (C) seven variants of uncertain significance (VUS), demonstrating the expression and VLP-targeting levels of ABCA4 variants compared to the wild-type. Created in BioRender. Cevik et al. (2024). https://BioRender.com/t49b834.

The WBA revealed comparable expression levels between the wild-type ABCA4 and variant proteins with the exception of the p.E2031K, and p.Y2165C variants, which displayed a noticeably fainter ABCA4 band compared to the wild-type and other variants (Fig. 12C, left). The use of β-tubulin as a loading control confirmed equal protein loading across all samples, suggesting that the reduced ABCA4 signal is specific to those variants.

5.2.2.2. Assessment of ABCA4 variants’ membrane targeting

The ability of the ABCA4 variants to be targeted to the membrane and incorporated into virus-like particles (VLPs) was assessed through WBA of the purified VLP samples. Successful membrane targeting was indicated by the presence of an ABCA4 band in the WBA for the variants: p.Y345D, p.Y603F, p.T1428M, p.Y1889N, and p.E2031K, showing comparable levels of ABCA4 to the wild-type control (Fig. 12C, right).

However, for the p.L844R and p.G1559E variants, no ABCA4 bands were detected in the purified VLP samples (Fig. 12C, right). The absence of detectable ABCA4 in the VLP preparations of these two variants suggests potential defects in folding or trafficking to the host cell plasma membrane, resulting in their failure to integrate into the VLP membrane environment.

5.2.2.3. Assessment of ABCA4 variants’ enzymatic activity

We performed the basal and retinal-stimulated ATPase activity assay with the 6 variants (p.Y345D, p.Y603F, p.T1428M, p.Y1889N, p.E2031K, and p.Y2165C) that were targeted to the VLP membrane.

The benign variant p.T1428M exhibited basal and retinal-stimulated ATPase activities similar to those of the wild-type (Fig. 13).

Fig. 13. ATPase Activity of VLP-ABCA4 Variants.

The bar graph shows the basal and retinal-stimulated ATPase activities of ABCA4 incorporated into virus-like particles (VLPs) for WT and seven variants (T1428M, Y345D, Y603F, E2031K, Y1889N, Y2165C), alongside a negative control (-Control). The data represent the mean ± SD of independent biological replicates (n = 3). Statistical analysis was conducted using one-way ANOVA followed by Tukey's HSD post hoc test.

The p.Y345D variant showed a complete loss of retinal-stimulated ATPase activity (p < 0.001), while retaining most of the basal activity (p = 0.002) (Fig. 13), suggesting that this substitution disrupts interactions critical for effective substrate binding and catalytic function. According to cryo-EM structures, Tyr-345 is a key residue within the NRPE binding pocket, interacting with the ATR moiety (Fig. 11), and its substitution likely compromises these crucial interactions.

The p.Y603F variant exhibited a partial reduction in retinal-stimulated ATPase activity (p < 0.001) (Fig. 13). Although the aromatic nature of tyrosine is retained with phenylalanine, the absence of the hydroxyl group may impair specific hydrogen bonds essential for proper enzyme activation (Fig. 11).

The p.Y1889N variant also showed a complete loss of retinal-stimulated ATPase activity (p < 0.001), while maintaining basal activity (Fig. 13). This suggests that the substitution of tyrosine with asparagine disrupts crucial interactions necessary for substrate-induced activation, likely due to the loss of stabilizing hydrogen bonds in the transmembrane domain (Fig. 11).

The p.E2031K variant also showed a significant reduction in retinal-stimulated ATPase activity (p < 0.001) (Fig. 13). The replacement of a negatively charged glutamate with a positively charged lysine likely disrupts a critical salt bridge, leading to instability in the ATP-bound conformation (Fig. 11).

The p.Y2165C variant exhibited a complete loss of ATPase activity, both in basal and retinal-stimulated conditions (p < 0.001) (Fig. 13). The substitution of tyrosine with cysteine disrupts several hydrogen bonds and alters the conformational dynamics necessary for ATP hydrolysis, indicating severe functional impairment (Fig. 11).

5.2.2.4. Reclassification of the variants using ACMG/AMP guidelines

Section 5.2 exemplifies the elucidation of the functional consequences of selected ABCA4 variants of uncertain significance (VUS) using combined in silico and in vitro approaches. By focusing on a subset of these missense VUS, we sought to provide a deeper understanding of their pathogenic potential and facilitate their reclassification under the ACMG/AMP (American College of Medical Genetics and Genomics) guidelines (Richards et al., 2015). The ACMG guidelines provide standardized criteria for classifying genetic variants based on their clinical significance. They consider factors such as variant type, population frequency, computational predictions, functional evidence, segregation data, and phenotypes of affected individuals.

The benign variant p.T1428M expressed and targeted at a similar rate with WT, and retained 100 % of ATPase activity, with no functional impairment observed in vitro. On the other hand, all analyzed VUS showed varying degrees of impairment in the membrane localization, basal or retina-stimulated ATPase activity.

The variants p.L844R and p.G1559E displayed a failure to localize to the VLP membrane, indicative of protein misfolding or trafficking defects. These findings align with the in silico structural predictions, where both variants showed severe stereochemical issues, suggesting a destabilized protein structure. The resulting steric clashes and electrostatic repulsion likely prevent proper folding and membrane insertion. Additionally, multiple lines of pathogenicity evidence and clinical data support these observations, as both variants are associated with retinal disease phenotypes (Stone et al., 2017). Additionally, reported patients carrying these variants had a pathogenic trans allele, while the cis allele was free from known pathogenic variants (PM3), further support their classification as likely pathogenic (LP) under the ACMG/AMP guidelines (Fig. 14).

Fig. 14. Reclassification of ABCA4 variants of uncertain significance (VUS) following ACMG/AMP guidelines.

Variants were evaluated using integrated in vitro and in silico analyses. PP2 could also be applied to all missense variants in the ABCA4, as the guidelines suggest that missense variants in gene with low rate of benign missense variants suggest supporting level of pathogenicity evidence. These findings justify reclassification of the VUS to likely pathogenic (LP) or pathogenic, aligning with the ACMG/AMP guidelines (Richards et al., 2015). NF: not found. PS, PM, PP are strong, moderate, and supporting pathogenicity evidence, respectively. Created in BioRender. Cevik et al. (2024). https://BioRender.com/y95k636.

Tyrosine residues play a pivotal role in the structural and functional stability of ABCA4, particularly due to their amphipathic nature and their involvement in critical hydrogen bonding and aromatic interactions. The four variants involving tyrosine substitutions—p.Y345D, p.Y603F, p.Y1889N, and p.Y2165C—exhibited varying degrees of functional impairment, highlighting the importance of these residues.

The p.Y345D and p.Y1889N variants, both of which replace tyrosine with non-aromatic residues, resulted in a complete loss of retinal-stimulated ATPase activity. This suggests a significant disruption in protein-substrate interactions, as tyrosine residues are often involved in critical interactions with the retinal substrate or in maintaining the structural integrity of substrate-binding sites. In silico analysis supports these findings, with p.Y345D disrupting the retinal binding site, and p.Y1889N compromising the structural stability of the transmembrane domain through the loss of stabilizing hydrogen bonds. The computational pathogenicity predictions (PP3) and absence in population databases (PM2) justify their reclassification as LP (Fig. 14).

The p.Y2165C variant, located in regulatory domain 2, resulted in a complete loss of ATPase activity. This tyrosine-to-cysteine substitution disrupts several intradomain and interdomain hydrogen bonds, likely altering the conformational flexibility necessary for ATP hydrolysis. The computational predictions (PP3) further support its classification as LP (Fig. 14).

The p.Y603F variant, which substitutes tyrosine with phenylalanine, retains the aromatic nature but loses the hydroxyl group essential for hydrogen bonding. This loss is reflected in the significantly reduced retinal-stimulated ATPase activity, indicating impaired protein-substrate interactions. Notably, substitutions of Tyr-603 to Cys and His have been reported as pathogenic in ClinVar, providing strong evidence of pathogenicity (PS1) according to the ACMG/AMP guidelines. Combined with computational evidence (PP3), this variant can be classified as pathogenic (Fig. 14). Similarly, the p.E2031K variant in NBD2 results in the loss of a salt bridge critical for stabilizing the ATP-bound conformation, leading to reduced retinal-stimulated ATPase activity. The pathogenicity of this variant is further supported by computational evidence (PP3) and population data (PM2), suggesting its likely pathogenic status (Fig. 14).

Our findings offer substantial evidence for the reclassification of the studied VUS, emphasizing the utility of the combined in silico and VLP-based platform in the functional analysis of ABCA4 variants. This integrative approach could be invaluable in advancing our understanding of ABCA4-related retinal diseases and improving the clinical interpretation of genetic variants.

6. Conclusions and outlook

The retina-specific ATP-binding cassette transporter protein, ABCA4, plays a critical role in the visual cycle by translocating retinoid byproducts of phototransduction, preventing their toxic accumulation. Dysfunction caused by genetic variants of ABCA4 is one of the primary causes of a spectrum of inherited retinal degenerations (IRDs). Although over 4000 ABCA4 variants have been identified, many remain unclassified in terms of their pathogenic impact. Accurate classification is critical not only for understanding disease progression but also for the personalization of treatment plans, recruitment into clinical trials, and the development of targeted gene therapies.

The growing number of variants of uncertain clinical significance, coupled with the experimental challenges of studying the large transmembrane protein ABCA4, highlights the urgent need for high-throughput and feasible methods. The comprehensive understanding of ABCA4 variants and their pathogenicity can be achieved with an integrative approach that combines in silico predictions with experimental validation. Such a strategy is crucial for both broadening our molecular understanding of ABCA4 functional mechanisms and providing more precise clinical insights for patient management.

In silico analyses offer a valuable first step, enabling the efficient scanning and prioritization of a vast number of variants. These computational approaches can predict pathogenicity and molecular consequences of a variation. However, to obtain a complete picture of variant pathogenicity, these predictions must be complemented by in vitro experiments, which allow for detailed molecular and biochemical characterization. Effective in vitro studies can provide direct evidence of functional deficits caused by specific missense ABCA4 variants, ranging from altered substrate binding to ATP hydrolysis or improper membrane localization.

This integrated strategy aligns with the American College of Medical Genetics and Genomics and the Association for Molecular Pathology (ACMG-AMP) guidelines (Richards et al., 2015), which emphasize the importance of using multiple lines of evidence when interpreting the pathogenicity of genetic variants. By incorporating both predictive and experimental data, this approach ensures a more accurate classification of variants.

Future directions for this work can build upon both the in silico and in vitro findings, with several promising avenues to explore. On the in silico front, developing an automated, deep learning-based prediction tool specifically tailored for ABCA4 could greatly benefit the research community by rapidly predicting the effects of ever emerging novel variants. This tool would incorporate a structure-focused approach while maintaining multiple lines of computational evidence, which has proven effective in predicting pathogenicity. Furthermore, expanding this tool to cover other members of the ABC transporter superfamily with known disease associations could offer broader applications, improving diagnostic precision across related genetic disorders.

On the in vitro side, the VLP system demonstrates immense potential as a versatile platform for functional assays. Future efforts could focus on designing a high-throughput transporter assay leveraging this system. While previous assays, such as the proteoliposome-based ATP-dependent transfer assay (Quazi et al., 2012), have significantly advanced our understanding of N-retinylidene-PE transport, they remain limited in scalability and throughput. The VLP platform, with its ability to maintain native topology and functional activity, offers a compelling alternative that could accommodate a broader range of variants.

By combining computational tools with advanced in vitro techniques, this integrated approach promises to accelerate the functional characterization of ABCA4 variants, contributing to the molecular understanding of retinal diseases. Ultimately, these advancements will support the translation of genetic findings into meaningful clinical applications, improving outcomes for individuals affected by ABCA4-related IRDs.