Development of a novel prediction model based on protein structure for identifying RPE65-associated inherited retinal disease (IRDs) of missense variants

Introduction

The retinal pigment epithelium-specific 65 kD protein (RPE65) is a retinoid isomerohydrolase that plays a critical role in the regeneration of 11-cis retinol in the visual cycle. Encoded by the RPE65 gene (OMIM 180069), this protein is expressed exclusively in the retinal pigment epithelium (Moiseyev et al., 2005). Autosomal recessive mutations in RPE65, often involving bi- or multi-allelic mutations, can lead to photoreceptor degeneration in humans (Gu et al., 1997). Clinically, a majority of RPE65 variants (approximately 67%) are commonly associated with Leber congenital amaurosis (LCA) or retinitis pigmentosa (RP) phenotypes (approximately 16%) (Stenson et al., 2017), which have similar fundus manifestations (Aoun et al., 2021). In 2017, the US Food and Drug Administration approved voretigene neparvovec (Luxturna®; Spark Therapeutics, Philadelphia, PA, USA) gene therapy for the treatment of patients with viable retinal cells and confirmed biallelic RPE65 mutation-associated retinal dystrophy (Russell et al., 2017). However, while timing the initiation of gene therapy is an important consideration, we still need more information about the natural history of the disease to better guide clinical applications (Botto et al., 2022; Sodi et al., 2021). Therefore, understanding the correlation between mutation and phenotype is critical.

RPE65 is a beta-propeller fold protein comprised of seven blades. Splicing and frameshift mutations in this protein can result in a truncated and non-functional protein product that is presumed to be null and irrelevant (Gu et al., 1997). However, predicting the significance of missense variations associated with RPE65-mediated inherited retinal diseases (IRDs) is challenging as both benign and pathogenic variations coexist in almost every disease-associated gene (Lek et al., 2016). In RPE65-associated IRDs, the analysis of missense mutants is particularly challenging as it is difficult to predict the significance of variants of uncertain significance (VUS). Mutations causing disease often occur in regions with secondary protein structures, which are crucial for protein stability and function (Khan & Vihinen, 2007; Yue, Li & Moult, 2005). In 2009, the crystal structure of RPE65 was resolved, revealing details of its active site architecture and oligomeric state (Kiser et al., 2009). Recent research has identified notable features of the RPE65 protein structure, including an iron center coordinated by a 4-His/3-Glu motif, a hydrophobic/cationic patch on the protein’s exterior, an active site cavity, and a dimeric form (Kiser, 2022). Despite the knowledge that protein secondary structure influences protein function, few studies have investigated the correlations between secondary structure and missense/phenotypes in RPE65-mediated IRDs (Lorenz et al., 2008; Thompson et al., 2000).

In this study, we aimed to investigate the relationship between missense variations in RPE65 and their impact on protein features. We analyzed missense variants in both pathogenic and normal populations and developed a new prediction model based on protein structure to calculate the risk of VUS occurring in RPE65 missense variants. We also examined the correlations between secondary structure and phenotypes in probands with RPE65 variants to gain new insights into the role of RPE65 missense variants in the pathogenic mechanisms of RPE65-associated IRDs. Our study provides important guidance for future gene therapy strategies and could ultimately lead to more effective treatment options for RPE65-associated IRDs.

Materials & Methods

Results

Characteristic 3D features of pathogenic and population missense variants of RPE65

According to the ClinVar and HGMD databases, missense mutations account for 67.58% and 54.89%, respectively, of all mutations in pathogenic (P and LP) RPE65 variants. This indicates that missense mutations are the most significant type of mutation in RPE65 genes (Fig. 1A). After deduplication, we marked 153 pathogenic (P and LP), and 300 population amino acid variations (as reported in the UniProt database) in the RPE65 monomer (Fig. 1B). We then calculated the substitution matrix score (BLOSUM 62) between the two groups and found that the score of population variants was higher than that of RPE65-IRDs (P < 0.0001) (Fig. 1C). To systematically identify the 3D features associated with “pathogenic” and “population” variants, we analyzed the 3D sites affected in 153 pathogenic variants (ClinVar and HGMD databases) and 300 general population variants (UniProt database) from RPE65 genes (Table S2). Among the sixteen features, sheet (OR = 2.55, P = 0.000), β-sheet (OR = 2.64, P = 0.000), strands (OR = 2.57, P = 0.000), β-hairpins (OR = 1.78, P = 0.004), Fe²⁺ (iron center) (OR = 5.73, P = 0.001), and active site cavity (OR = 2.76, P = 0.006) were significantly correlated with RPE65-associated IRDs missense variants. Additonally, helix (OR = 0.37, P = 0.009), loop (OR = 0.55, P = 0.003), helices (OR = 0.42, P = 0.11), and helix-helix interactions (OR = 0.43, P = 0.046) were also correlated (Fig. 1D).

Development of a prediction model based on the RPE65 protein structure

LASSO binary logistic regression was utilized to select the top fourteen 3D variables from the sixteen structures analyzed (Figs. 2A and 2B). A nomogram was subsequently created from these variables (Fig. 2C). The receiver operating characteristic (ROC) curves generated demonstrated strong predictive capability with an area under curve (AUC) value of 0.75131 (Fig. 3A). Calibration curves were generated to evaluate the calibration of the RPE65-associated IRDs nomogram (Fig. 3B). Next, the risk of RPE65 missense variants of uncertain significance (VUS) occurring was calculated using a nomogram (Fig. 4A). The highest risk of VUS occurrence was observed in H241R and R44L, where amino acids at both mutation sites were altered. H241R influenced the Fe²⁺ and active site cavity of functional features. Both missense mutations were located in the β-sheet and in strands of 8-class-secondary structure(Fig. 4B), which were identified as risk factors for pathogenicity (Fig. 1C). Furthermore, changes in residues were observed. The amino acid site 241 changed from Histidine (H) to Arginine (R), decreasing its connection to Y239 (Fig. 4C). Similarly, the residue at position 44R interacted with H68, F469, T525, and F526. In contrast, when mutated to 44L, the residue was only connected to H68 (Fig. 4D). In summary, these findings may indicate that both missense mutations are pathogenic or likely pathogenic.

Genotype analyses of patients with RPE65 variants

Twenty-one probands with RPE65 variants (13 males and eight females) were enrolled in this study (Fig. S2). Missense mutations accounted for 72.4% of the total mutation types with some overlaps, which were identified in the RPE65 monomer (Fig. 5A). A total of 26 different mutations were observed, including 15 missense, five frameshifts, one nonsense, and five splicing mutations. Thirteen patients had biallelic mutations, and one patient (F13) had four allelic mutations, including one missense (Asp482Asn) and three frameshift mutations (Leu270Hisfs11, Trp271Lysfs11, and Ser269Metfs13) (Table 1). For the purpose of analysis, patient F8 was selected at random (Fig. 5B). The pathogenic missense mutation H68Y (His68Tyr) was located in the sheet, β-sheet, strands, and β-hairpins, all of which are risk factors. In contrast, P419S (Pro419Ser), which is a variant of uncertain significance (VUS), with a BLOSUM 62 score of −1, was located in the sheet, loop, strands, β-hairpins, and β-bulges, and its risk score ranged from 0.3 to 0.4, with no change in residue contact (Fig. 5C). Thus, it can be concluded that P419S has two risk factors and has a high possibility of being pathogenic.

Table 1:

RPE65 variants identified in this cohort of patients.

Nucleotide change	Amino acid change	Exon/intron	ACMG category	Patients	Mutation type	Reference
c.1399C >G	Pro467Ala	E13	VUS	F1, F3	Missense	Reported
c.272G >A	Arg91Gln	E4	P	F2, F15, F21	Missense	Reported
c.271C >T	Arg91Trp	E4	P	F2, F12	Missense	Reported
c.1338G >T	Arg446Ser	E12	P	F5, F7	Missense	Reported
c.1543C >T	Arg515Trp	E14	P	F6, F14, F20	Missense	Reported
c.1444G >A	Asp482Asn	E13	LP	F6, F13	Missense	Reported
c.1255C >T	Pro419Ser	E12	VUS	F8	Missense	Reported
c.202C >T	His68Tyr	E3	P	F8	Missense	Reported
c.1590C >A	Phe530Leu	E14	LP	F9, F16, F17	Missense	Reported
c.997G >C	Gly333Arg	E9	VUS	F10	Missense	Reported
c.334T >A	Cys112Ser	E4	VUS	F10	Missense	Reported
c.93A >G	Thr31Thr	E2	VUS	F15, F21	Synonymous	Reported
c.335G >A	Cys112Tyr	E4	VUS	F16	Missense	Reported
c.1520C >T	Ala507Val	E14	VUS	F17	Missense	Novel
c.1051G >A	Glu351Lys	E10	VUS	F19	Missense	Novel
c.493C >T	Gln165*	E5	P	F4	Nonsense	Reported
c.837del	Phe279Leufs46	E8	LP	F11	Frameshift	Novel
c.376del	Val126*fs1	E5	LP	F12	Frameshift	Novel
c.808_809insA	Leu270Hisfs11	E8	LP	F13	Frameshift	Novel
c.809_810insGAAG	Trp271Lysfs11	E8	LP	F13	Frameshift	Novel
c.805_806insTGGA	Ser269Metfs13	E8	LP	F13	Frameshift	Novel
c.94+2T >A	_	I2	LP	F1	Splicing	Novel
c.245+4A >G	_	I3	LP	F18	Splicing	Novel
c.354-2A >G	_	I5	LP	F7	Splicing	Novel
c.998+1G >A	_	I10	LP	F3	Splicing	Reported
c.858+1delG	_	I9	LP	F9	Splicing	Novel

DOI: 10.7717/peerj.15702/table-1

Notes:

E

Exon

I

Intron

P

Pathogenic

LP

Likely pathogenic

VUS

variants of uncertain significance

ACMG

The American College of Medical Genetics and Genomics

Discussion

Predicting the 3D structure that a protein will assume based solely on its amino acid sequence has been a significant challenge in research for over 50 years (Anfinsen, 1973; Dill et al., 2008). A recent study has revealed the molecular effect of missense variants by accomplishing a comprehensive characterization of amino acid positions in protein structures, providing reference for the clinical interpretation of pathogenic and benign missense variants (Iqbal et al., 2020). Current studies show that most of the known mutations in the functional region of the RPE65 protein can cause retinal disease (Kiser, 2022). However, the way in which the protein structure influences disease remains unclear. In this study, we explored the correlations between the pathogenicity and population of RPE65 missense variants. Using a relatively large patient sample, we analyzed the association between pathogenic and population missense variations to RPE65. We found that the missense variants of RPE65-associated IRDs are related to the sheet and β-sheet, strands, β-hairpins, iron center, and active site cavity, which may indicate that the missense variants located in these sites are prone to be pathogenic.

In previous studies, researchers focused on clarifying the frequency and phenotypes characteristic of different races or distributions (Gao et al., 2021; Li et al., 2020; Lopez-Rodriguez et al., 2021). However, to date, genotype-phenotype correlations in patients with RPE65 variants have not been well established (Gao et al., 2021). Recently, a study analyzed the molecular characterization of the RPE65 cohort and genotype-phenotype correlation according to the number of RPE65 loss-of-function (LoF) alleles in the Italian population (Testa et al., 2022). Yet there are no studies analyzing the correlation between phenotype and missense variants. In this study, we indicated that the clinical characteristic was correlated to changes in amino acid residues, helix, helices, and β-hairpins in the missense mutation of RPE65. An AA change may be one of the factors influencing the phenotype of RPE65-associated IRDs with missense variants.

Determining whether a genetic variation in a patient is responsible for their disease can be challenging. Previously, researchers attempted to predict the pathogenicity of RPE65 mutations using an empirical algorithm to estimate pathogenic probability (EPP), which was validated for certain RPE65 variants (Philp et al., 2009; Stone, 2003). However, this method may have limited application in many cases. Recently, Cho et al. (2022) analyzed carrier frequency and expected incidence of RPE65-associated IRDs in East Asians and Koreans using exome data from the Genome Aggregation Database (gnomAD) and the Korean Reference Genome Database (KRGDB). In our study, we investigated the association between populations and the pathogenicity of RPE65 missense variants and developed a prediction model specifically tailored for missense variants of RPE65 based on its secondary structure and functional features, without considering other missing factors. We also discussed how to apply this model in clinical practice. For patients whose RPE65 variants are classified as variants of uncertain significance (VUS), we recommend locating the missense variant and referring to the nomogram to calculate the risk score. If the variant affects structural features strongly correlated with pathogenicity and receives a high-risk score, it is reasonable to classify it as pathogenic. But for variants with limited risk factors or low-risk scores, caution is necessary when estimating their pathogenicity using this model.

This study has certain limitations that should be considered. In cases where patients exhibit multi-allelic mutations, it is important to ascertain the presence and abundance of a mutation in the gene locus. Precise quantification of EGFR mutation abundance has been reported to not only enable better patient selection for EGFR-TKI treatment but also to facilitate the development of more effective treatment strategies for patients with a low abundance of EGFR mutations (Zhou et al., 2011). In our current model, the focus is limited to the protein structure at the mutation site, as information regarding mutation abundance is unavailable. This limitation may be a contributing factor to the suboptimal predictive accuracy observed in our study.

Conclusion

In this study, we investigated the relationship between pathogenic and population missense variations of RPE65 and protein 3D features and developed a novel prediction model (AUC = 0.7531). Furthermore, we analyzed the correlation between phenotype and protein structure in a Chinese cohort of patients with RPE65 missense variants. We developed a complementary method presenting a novel approach to predicting the potential pathogenicity of RPE65 missense variants based on protein structure. Our findings may provide valuable insights for the accurate diagnosis of RPE65-mutated inherited retinal diseases.

Supplemental Information