Determination of the Population Frequency of Monoallelic and Biallelic Predicted Pathogenic RPE65 Variants in a Normal Database

Abstract

Purpose

The RPE65-associated retinopathies include biallelic Leber congenital amaurosis 2 and severe early childhood-onset RP, and monoallelic RP with choroidal involvement. The population frequencies of these diseases have previously only been estimated from epidemiological studies, but are deduced here from the number of predicted pathogenic heterozygous or carrier variants in a normal cohort using bioinformatic analyses.

Methods

RPE65 variants were downloaded from gnomAD v4.1 (n = 807,162) and annotated with ANNOVAR. The population frequency of heterozygous predicted pathogenic variants was calculated from the numbers of pathogenic structural and copy number variants, null variants, rare missense changes classified disease-causing computationally, and from founder variants not already included. The computed population frequency for biallelic disease was compared with frequencies deduced from pathogenic variants reported in ClinVar.

Results

The population frequencies of heterozygous predicted pathogenic RPE65 variants were 1 in 408 using our strategy and 1 in 415 using ClinVar. Variants were structural in 50 people (3%), null in 793 (41%), and missense in 1,115 (57%). Predicted pathogenic variants were commonest in Admixed American (1 in 257) and African American (1 in 268) and least common in Middle Eastern (1 in 3,031) and Ashkenazi (1 in 1,851) people. A possible founder variant (p.Tyr431His) was present in 1 in 21 Amish. The variant associated with monoallelic RPE65-retinopathy (p.Asp477Gly) was present in 1 in 294,930 Europeans.

Conclusions

Biallelic RPE65-associated retinopathy affects at least 1 in 665,856 of the population, but is more common in Admixed Americans (1 in 264,196) and African Americans (1 in 287,296).

The inherited retinal diseases (IRD) are a group of genetic eye diseases characterized by progressive degeneration of the retinal photoreceptor cells and visual loss. The rod–cone phenotype (RP) presents with night blindness and a constricted visual field. Cone-mediated central and color vision are affected later. IRD is now the leading cause of blindness in working-age adults in developed countries and, altogether, is estimated to affect 1 in 3,000 to 4,000 individuals.¹ More than 300 genes have been implicated,² most of which involve structural or functional proteins in the RPE or photoreceptors. At least 20 of these diseases now have gene-based therapies available or in development.³

One such gene is RPE65. IRD from biallelic RPE65 pathogenic variants (RPE65-IRD or RPE65-associated retinopathy) represents a form of retinal degeneration where the photoreceptor cells may be relatively preserved but dysfunctional^4,5 and sometimes amenable to treatment with gene augmentation early in the disease course. Clinical and genetic diagnosis is now possible, and people with biallelic pathogenic or likely pathogenic RPE65 variants are referred for consideration of treatment with voretigene neparvovec (Luxturna).⁶ The beneficial effect is mainly on rod function, and outcomes are modest for traditional measures such as visual acuity, which mainly reflects cone abnormalities. Nevertheless, the novel multiluminance mobility test has demonstrated a marked and sustained improvement.⁷

Clinical manifestations of RPE65-associated retinopathy include the biallelic Leber congenital amaurosis 2 (MIM204100) and RP20 (MIM613794),⁸ monoallelic RP87 with choroidal involvement (RP87, MIM618697),⁹ and variations of these.

The spectrum of clinical manifestations for RPE65-associated retinopathy varies from mild to severe.^10,11 Leber congenital amaurosis 2 is diagnosed in children who are born blind or diagnosed within a few months of life.¹² Nystagmus is typical. Fundoscopy demonstrates attenuated retinal vessels and abnormal retinal pigmentation.¹² There is widespread hypoautofluorescence^13,14 and an unrecordable ERG response due to impaired vitamin A metabolism.^4,15,16 Older patients have chorioretinal atrophy with no recordable rod or cone response on ERG.¹⁶

RP87 with choroidal involvement has a later onset and more slowly progressive loss of central and peripheral vision, including night blindness with choroidal and retinal atrophy.⁹ This variant is incompletely penetrant and some people with a pathogenic variant are not affected.¹¹ Retinal features range from nearly normal to bone spicules and focal pigment deposits in the midperiphery, to a diffuse chorioretinal atrophy. ERG demonstrates rod function that also varies from mildly affected to unrecordable.⁹ No treatment is available.

RPE65 is expressed in the RPE and encodes retinoid isomerohydrolase, which regenerates the chromophore 11-cis retinol that is critical in phototransduction in the photoreceptor cells. Pathogenic variants associated with biallelic RPE65-retinopathy occur throughout the gene and generally include both null and missense changes (https://simple-clinvar.broadinstitute.org/) (Fig. 1). The RPE65 variant type contributes to disease severity. Biallelic loss-of-function variants result in absent or nonfunctional protein, and early-onset disease. Hypomorphic typically missense variants result in later onset, milder disease.

Figure 1.

Distribution of pathogenic and likely pathogenic missense and nonsense variants in RPE65. From Simple ClinVar https://simple-clinvar.broadinstitute.org/ This is a diagram of the RPE65 protein that demonstrates that pathogenic changes are located throughout the protein, and comprise at least nonsense and missense changes. This observation was used to justify searching for variants of all types in the gene.

The biallelic RPE65-associated retinopathies are rare and have been reported in 2 to 3 per 100,000 individuals.^17,18 However, epidemiological studies suggest that they are less common. One study found a prevalence for Leber congenital amaurosis 2 of 1 to 2 per 100,000, and RPE65 variants were responsible for 2% to 16%. Likewise, the prevalence of RP was estimated to be 11 to 26 per 100,000, and an RPE65 pathogenic variant was found in 0.2% to 2.0% of these.¹⁹ These estimates are based on the number of affected individuals in ophthalmology clinics rather than from genomic data of unbiased populations. When estimates were made directly from genetic data in the East Asian and Korean populations, an RPE65 variant carrier rate was found of one of 992, corresponding with a biallelic disease frequency of 1 in 3,941,308.²⁰ An analysis of the genetics from six world populations suggests that biallelic variants are over-represented in people with African ancestry.²¹ The risk is also higher in consanguineous populations, and the RPE65- associated retinopathies are more common in the Netherlands and Israel because of founder variants (p.Tyr368His, c.95-2A>T, respectively).^22,23 Founder missense changes in people of Irish or Scottish backgrounds (p.Asp477Gly) and in Flemish or Belgian populations (p.Glu519Lys) account for all known cases of the monoallelic disease.^9,11,24,25

Making the diagnosis of biallelic RPE65-associated retinopathy is important because of the availability of treatment that may preserve vision. Knowing the population frequency helps to alert clinicians to the likelihood of encountering an affected person, aids in health service planning including genetic testing and paying for treatment, and encourages pharmaceutical companies to develop further therapies.

The aim of the present study was to determine the population frequency of heterozygous/carrier and biallelic disease-associated RPE65 variants using bioinformatics and a large normal variant database and to compare these estimates with the results from pathogenic variants in ClinVar. The population frequency is then calculated from the carrier frequency. Our strategy used many of the American College of Medical Genetics/Association for Molecular Pathology (ACMG/AMP) criteria,²⁶ but did not have access to clinical data and relied principally on computational tools.

A similar strategy has helped to determine the population frequencies for Alport syndrome, polycystic kidney disease, Gitelman syndrome, mucopolysaccharidoses, and Fabry disease.^27–31 Furthermore, in at least two of these diseases the population frequencies determined computationally were confirmed by an independent method such as biochemical testing or histopathology.^27,31 The adult nature of our cohort was not a weakness of the study because we were not looking to identify individuals with biallelic childhood-onset disease, but rather calculating the likelihood of manifesting disease from the offspring of heterozygotes.

Our data are different from those derived from ophthalmology clinics. They demonstrate how often predicted pathogenic variants occur in the population, but the number of people with the disease depends on the penetrance of individual variants. People with less penetrant changes may have milder disease, no disease, or different manifestations. Penetrance may vary even in family members with the same genetic change. We have examined both monoalleic and biallelic forms of IRD-RPE65, even though treatment is not currently available for monoallelic disease.

Methods

Population Database

The population frequency of IRDs owing to predicted pathogenic variants in RPE65 (RPE65-IRD) was estimated using genetic variants (GRCh38) from the Genome Aggregation Database (gnomAD v4.1.0 www.gnomAD.broadinstitute.org, with a mean of 797,922 samples examined for RPE65) (canonical transcript: ENST00000262340.6/NM_000329.3). We have described our method previously.²⁹ gnomAD is a database with aggregated information from many different sequencing projects including UK Biobank where the majority have undergone whole exome sequencing and 63,046 have undergone whole genome sequencing. In general, these were adults in whom severe childhood-onset diseases had been excluded. Participants were unrelated, and sex, age, and ancestry, but not clinical data, were provided. The database was examined from August to September 2024.

Written, informed consent was obtained from all gnomAD participants at recruitment into the original studies for the use of their anonymized data for further research, so that this project did not require separate ethical approval.

Structural and Copy Number Variants

Sequencing for structural variants was available for 63,046 samples and for copy number variants in 464,297 samples. Structural and copy number variants that affected exons and were deletions were considered predicted pathogenic and the number corrected for the smaller sample sizes.

Annotation of Null and Missense Variants

Variants in the RPE65 gene were downloaded from gnomAD, annotated in ANNOVAR (https://annovar.openbioinformatics.org/) and evaluated for pathogenicity based on their null type, and for missense changes, based on whether they were rare, pathogenic in computational tools and affected a conserved amino acid (Fig. 2). This population frequency was then compared with that calculated from the number of gnomAD variants assessed as pathogenic or likely pathogenic in the ClinVar database (https://www.ncbi.nlm.nih.gov/clinvar/).

Figure 2.

Bioinformatic strategy for estimating population frequency of biallelic RPE retinopathy. The filtering process used to determine the population frequency of monoallelic predicted pathogenic variants in RPE65.

Filtering Processes

Variants located in the 5′ or 3′ untranslated region, in the intronic or noncanonical splice regions, or that were synonymous were excluded.

Null Variants

Null variants including nonsense, frameshift, and canonical splice site changes were classified as predicted pathogenic, except null variants in the last exon or the last 50 nucleotides of the penultimate exon that were presumed to escape nonsense-mediated decay.²⁶ Null variants that were predicted pathogenic were included in the total cohort no matter how many times they occurred in gnomAD.

Missense Variants

Missense variants were classified predicted pathogenic if they were rare (occurring in <25 people in gnomAD for monoallelic disease) and were pathogenic in all bioinformatic prediction tools: PolyPhen-2 (PP2, score of ≥0.95, http://genetics.bwh.harvard.edu/pph2/), Sorting Intolerant From Tolerant (SIFT4G, score of ≤0.05 from Varsome, https://varsome.com/), MutationTaster (prediction: disease causing, D or A https://www.mutationtaster.org/); and GERP (http://mendel.stanford.edu/sidowlab/downloads/gerp/index.html³²; score of >3.00).

Our assessment strategy for predicted pathogenic missense variants was validated in the following way. Up to 20 pathogenic or likely pathogenic and 20 benign or likely benign missense variants were chosen from the ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/) and the Leiden Open Variation Database (LOVD; https://www.lovd.nl/) and assessed using our strategy for predicted pathogenicity. Sensitivity, specificity, and positive and negative predictive values were then calculated (Table 1).

Table 1.

Population Frequency of Predicted Pathogenic RPE65 Variants in gnomAD V4 Using Our Strategy or REVEL for Missense Variant

Pathogenic Structural and Copy Number Variants (Corrected)	Pathogenic Null Variants	Predicted Pathogenic Missense Variants by Our Method	Predicted Pathogenic Missense Variants by REVEL Score ≥0.932	Founder Variants Not Counted Already	Population Frequency of Predicted P Variants Using Structural, Null, Missense by Our Method Plus Founder Variant	Population Frequency Using Structural, Null Variants and REVEL >0.932 for Missense Changes Plus Founder Variant
50 people	542 people	1,115 people	2,837 people (includes 2115 copies of c.881A>C, p.Lys294Thr)	251 people	50 + 542 + 1,115 + 251 = 1,958/797,922 or 1 in 408 people are heterozygotes or carriers and 1 in 665,856 has biallelic disease	50 + 542 + 2837 + 251 = 3,680 or 1 in 217 people are heterozygotes or carriers, and 1 in 188,356 has biallelic disease

Missense variants were also assessed by REVEL scores, which have been reported to have the best performance for distinguishing pathogenic from rare neutral variants with allele frequencies of less than 0.5%.^33,34 REVEL scores combine pathogenicity predictions from 13 individual computational scores with 8 based on conservation. Here we took a score of 0.932 or greater as pathogenic.

Founder Variants

A number of variants assessed predicted pathogenic in our analysis were found more than 25 times in the gnomAD 4.1 cohort. Those that were reported as pathogenic or likely pathogenic in ClinVar were considered founder variants and the counts added to our total of predicted pathogenic changes. In addition, other founder variants that affected a noncanonical splicing site were identified and counted.

Calculation of Population Frequencies

The population frequency for heterozygotes with predicted pathogenic variants was then calculated based on the assumption that each individual had only one variant. The population frequency for biallelic variants was calculated from the formula: ¼ × (population frequency for total heterozygous predicted pathogenic variants)².

Population Frequencies of Predicted Pathogenic Variants in Different Ancestries

Population frequencies of predicted pathogenic variants were calculated in people from all 10 ancestries included in gnomAD (African/African American, admixed American, Ashkenazi, East Asian, Finnish, European [non-Finnish], Amish, Middle Eastern, South Asian, and Remaining). Calculations did not take into account overall consanguinity.

Population Frequencies of RPE65-IRD Using Other Variant Databases

In addition, the population frequencies of RPE65-IRD pathogenic variants were calculated based on how often variants were assessed as pathogenic or likely pathogenic in ClinVar (ClinVar https://www.ncbi.nlm.nih.gov/clinvar/), LOVD (https://www.lovd.nl/), or the Human Gene Mutation Database (HGMD Professional 2024.2 https://hgmd.cf.ac.uk/ac/index.php) were also found in gnomAD. Variants in ClinVar classified as conflicting where there were both a variant of unknown significance and pathogenic or likely pathogenic assessment were classified as predicted pathogenic for the calculation. Variants were included even if allele counts were greater than 25 because the initial assessments were presumed to include clinical data and use the ACMG/AMP criteria that increased accuracy.

Statistical Analysis

Results for differences between ancestries were compared with the χ² test with Yates correction (Graphpad, La Jolla, CA, USA; https://www.graphpad.com/).

Results

Predicted Pathogenic RPE65 Variants in the Overall gnomAD Cohort

The RPE65 gnomAD dataset included a total of 1,958 heterozygous predicted pathogenic variants in 797,922 people (0.24%) or 1 predicted pathogenic variant in 408 people (Table 1).

Predicted pathogenic variants included no structural variants but copy number variants in 25 people in a cohort of 464,297, which was corrected to 50 people for the whole cohort, as well as 542 people with a null variant and 1,115 people with a predicted pathogenic missense change (Table 1). There were also 251 copies of c.11+5G>A, a confirmed founder variant which was not included in our analysis but was considered in ClinVar. The heterozygous predicted pathogenic carrier rare of 1 in 408 corresponded with an overall biallelic population frequency of 1 in 665,856 based on our assessment of missense variants.

The validation studies of our assessment method demonstrated a sensitivity of 65%, specificity of 100%, positive predictive value of 100% and negative predicted value of 59% (Table 2), (Supplementary Table S1) consistent with our population frequency underestimating the number of heterozygous pathogenic variants.

Table 2.

Validation of Filtering Strategy for Predicted Pathogenic RPE65 Missense Variants

	Database Contributions	Sensitivity (%)	Specificity (%)	Positive Predictive Value (%)	Negative Predictive Value (%)
RPE65	LOVD (B/LB n = 6, P/LP n = 20) ClinVar (B/LB n = 4)	65%	100%	100%	59%

When a REVEL score of greater than or equal to 0.932 was used instead of our missense assessment strategy, there were 2,837people with a predicted pathogenic missense variant. When the structural and copy number and null and founder variants were added, there was a total of 3,680 people or 1 in 217 with a predicted pathogenic variant using REVEL for the missense changes. This number was high because of c.881A>C corresponding with the p.Lys294Thr variant, which was found in 2,115 people and was not positive in our assessment and assessed as likely benign in ClinVar. The population frequency for predicted pathogenic variants using REVEL after this variant was excluded was 1 in 689.

Predicted Pathogenic RPE65 Variants According to Different Pathogenic Variant Databases

ClinVar

One hundred thirty-nine pathogenic or likely pathogenic variants in ClinVar were found in 1,925 people in gnomAD corresponding to a population frequency of 1,925/797,922 or 1 in 415 (Table 3).

Table 3.

Population Frequencies of RPE65 Retinopathy Using Variant Databases (ClinVar, LOVD, and HGMD)

ClinVar	Predicted Pathogenic Variants	139 Variants in 1925 People
	Frequency	1,925/797,922 (0.24%) or 1 in 415 people are heterozygotes or carriers and 1 in 688,900 has biallelic disease
LOVD	Predicted pathogenic variants	15 variants in 119 people
	Frequency	119/797,922 (0.01%) or 1 in 6,705 people are heterozygotes or carriers and 1 in 179,828,100 has biallelic disease
HGMD	Predicted pathogenic variants	45 variants in 178 people
	Frequency	178/797,922 (0.02%) or 1 in 4483 people are heterozygotes or carriers and 1 in 80,389,156 has biallelic disease

These population frequencies are for carriers or heterozygotes of the pathogenic variants and we have also calculated the population frequency of autosomal recessive or biallelic disease. ClinVar assessments are considered the most accurate because variants have been submitted from accredited testing laboratories on samples from patients with suspected IRD and assessed according to the ACMG criteria over the past 5 years.

LOVD

Fifteen pathogenic or likely pathogenic variants in LOVD were found in 119 people in gnomAD corresponding to a population frequency of 119 in 797,922 or 1 in 6,705 for predicted pathogenic variant carriers.

HGMD

Forty-five pathogenic variants were found in 178 people in gnomAD corresponding with a population frequency of 1 in 4,483 for predicted pathogenic variant carriers.

Predicted Pathogenic RPE65 Variants in People of Different Ancestries in gnomAD

The population frequencies of predicted pathogenic variants differed in people of different ancestries (Table 4). Heterozygous predicted pathogenic variants were more common in African/American (1 in 268; P < 0.0001 compared with European) and admixed American people (1 in 257; P < 0.0001). One variant was common in the Amish and meant that the population frequency in these people was 1 in 21, which was also more common than in the European population (P < 0.0001). Predicted pathogenic variants were least common in people of Middle Eastern (one in 3,031; P = 0.02) and Ashkenazim backgrounds (1 in 1,851; P < 0.0001). However, both the Amish and Middle Eastern cohorts were small.

Table 4.

Confirmed Predicted Pathogenic Founder Variants in Different Ancestries

Transcript	Protein	Additional Evidence of Pathogenicity	Total No. of People With This Variant in gnomAD v4.1	Ancestries (No. of People With the Variant)
c.11+5G>A	–	Pathogenic (ClinVar)	251	Afr/Am (7), Eur (240), remaining (4)
c.118G>A	p.Gly40Ser	Pathogenic (ClinVar)	97	Afr/Am (4), Ashk (3), Eur (83), remaining (7)
c.271C>T	p.Arg91Trp	Pathogenic (ClinVar)	118	Afr/Am (16), Adm/Am (1), E Asian (5), Eur (86), Ashk (1), S Asian (2), remaining (7)
c.292_311del	Frameshift	Pathogenic (ClinVar)	32	Ad/Am (30), Eur (1), remaining (1)
c.304G>T	Stop-gained	Pathogenic (ClinVar)	35	Ashk (1), Finn (5), Eur (29)
c.370C>T	p.Arg124Ter	Pathogenic (ClinVar)	44	Afr/Am (23), Ad/Am (1), Eur (17), remaining (3)
c.746A>G	p.Tyr249Cys	Pathogenic (ClinVar)	19	Eur (2), S Asian (17)
c.893del	p.Lys298SerfsTer27	Pathogenic (ClinVar)	101	Eur (101)
c.1102T>C	p.Tyr368His	Pathogenic (ClinVar)	228	Eur (223), remaining (5), confirmed founder variant
c.1291T>C	p.Tyr431His	Variant of unknown significance (ClinVar); PM1, PM2, PP3, score = 5 (Alamut)	22	Amish (22)
c.1543C>T	p.Arg515Trp	Pathogenic (ClinVar)	38	Afr/Am (4), Eur 95), S Asian (1), East Asian (28)

These variants fulfilled all our criteria for pathogenicity except for rarity but were Pathogenic in ClinVar except for p.Tyr431His which is a known founder mutation.

Founder Variants for Leber Congenital Amaurosis 2 and RP 20

A number of variants that were assessed as predicted pathogenic using our algorithm but were present in more than 25 people were found in gnomAD4.1 (Table 5). Where these variants were considered pathogenic or likely pathogenic in ClinVar, they were added to the missense counts. Two founder variants (c.11+5G>A [n = 231] and c.246-11A>G [n = 3]) affected noncanonical splicing sites.²⁰

Table 5.

Population Frequencies of Carriers and Biallelic Disease Based on Predicted Pathogenic RPE Variants in People of Different Ancestries

Ancestry	Total Predicted Pathogenic SV and CNV Variants (Corrected)	Pathogenic Null Variants	Predicted Pathogenic Missense Variants	Founder Variants Not Counted Previously*	Total Predicted pathogenic Variants and Heterozygotes	Population Frequency of Biallelic Disease
African/African American (n = 37,545)	0	67	66	9	142/37,545 or one in 264 P < 0.0001	1 in 278,784
Latino/Admixed American (n = 30,019)	0	58	59	–	117/30,019 or 1 in 257 P < 0.0001	1 in 264,196
Ashkenazi (n = 14,804)	0	2	6	–	8/14,804 or 1 in 185 P < 0.0001	1 in 13,704,804
East Asian (n = 22,448)	0	7	55	–	62/22,448 or 1 in 362 P = 0.23	1 in 524,176
Finnish (n = 32,026)	16	25	17	–	58/32,026 1 in 552; P = 0.08	1 in 1,218,816
Middle Eastern (n = 3,031)	0	1	0	–	1/3,03 1 in 3031; P = 0.02	1 in 36,747,844
European (n = 590,031)	22	345	767	241	1,375/590,031 or 1 in 429 (reference)	1 in 736,164
Amish (n = 456)	0	0	22	–	22/456, 1 in 21; P < 0.0001	1 in 1764
South Asian (n = 45,546)	2	19	79	–	100/45,546 or 1 in 455; P = 0.61	1 in 828,100
Other (n = 31,256)	10	18	44	4	76/31,256 or 1 in 411; P = 0.74	1 in 675,684
Total	50 (3%)	542 (28%)	1115 (57%)	251 (13%)	1,958/797,922 or 1 in 408	1 in 665,856

^*Founder variants: c.11+5G>A; c.246-11A>G.

Results in bold correspond to P < 0.05.

Founder Variant for RP 87 With Choroidal Involvement

Only two pathogenic RPE65 variants have been identified in RP 87 with choroidal involvement (p.Asp477Gly and p.Glu519Lys). gnomAD4.1 has two copies of p. Asp477Gly, corresponding with a population frequency of 2 in 797,922 or 1 in 398,961 people overall, and 1 in 294,930 Europeans. This may be higher in a Scottish or Irish population. There are five copies of p.Glu519Lys, corresponding with a population frequency of 5 in 797,922 or 1 in 159,584 overall, and 1 in 147,465 in Europeans. One person in the South Asian cohort had this variant as well (1/91,072).

Discussion

Predicted pathogenic heterozygous RPE65 variants were present in 1 in 408 normal people using our strategy and 1 in 415 using pathogenic variants from ClinVar. These corresponded with population frequencies for biallelic RPE65-retinopathy of 1 in 665,856 and 1 in 688,900, respectively, which are much more common than the previously reported and clinically ascertained 1 in 849 for carriers and 1 in 3 million for biallelic disease. The population frequencies varied in people of different ancestries, with the commonest being 1 in 264,196 in admixed Americans and least common with 1 in 13,704,804 in Ashkenazim. There were many founder variants, including a likely founder change in the Amish. The founder variants associated with RP87 with choroidal involvement were present in 1 in 294,930 people overall and 1 in 147,465 Europeans. The p.Glu519Lys was also found in South Asians.

The strengths of this study were the use of a large database of genetic variants, from a normal unbiased adult population, the rigorous bioinformatics analysis, and the multiple approaches taken, as well as the examination of different ancestries. Previous studies in other rare diseases have not required all computational results (PolyPhen-2, SIFT, Mutation Taster, and GERP) to be positive for a variant to be classified pathogenic. In addition, we evaluated our assessment strategy, compared our results with REVEL, reviewed the database for possible founder variants, and compared our overall results with variants assessed pathogenic and likely pathogenic in ClinVar, LOVD and HGMD.

However, our approach also had limitations, which mean our population frequencies may be underestimates. The gnomAD variant database used mainly whole exome sequencing for variant detection, but this does not identify large deletions, duplications, or deep intronic splicing changes, which anyway are difficult to interpret. However most disease-causing variants are located in the exons. Assessment of our testing algorithm indicated a sensitivity of only 65% for predicted pathogenic missense changes. In addition, many clinically identified people with a biallelic RPE65-retinopathy are from consanguineous parents and have homozygous variants; nevertheless, rates of consanguinity are difficult to predict. We also detected a number of common missense variants that fulfilled all our criteria for pathogenicity, but had insufficient evidence in ClinVar (Supplementary Table S2). Those that are proven to be pathogenic will further increase the population frequency.

The greatest risk of inaccuracy in this study was in interpreting the missense variants. Our assessment required a predicted pathogenic missense variant to be rare, pathogenic in all three computational tools, and to affect a conserved amino acid. Diagnostic laboratories have clinical and possibly family segregation data that help with variant interpretation. gnomAD does not include such data, which meant that not all the ACMG/AMP criteria could be applied.²⁶ However, we were still able to use some: structural variants (deletions, PVS1), null variants (PVS1), and, for missense changes, known disease-causing gene (PM1), rare variants (PM2), and computational (in silico) data (PP3). We were not able to use the classifications of de novo variants (PM6), functional studies (PS3), cis-trans testing (PM3), segregation analysis (PP1), or using the phenotype to support variant claims (PP4).

The population frequency derived from ClinVar is also likely to be an underestimate, but for different reasons. It relies on variant assessments submitted by diagnostic laboratories rather than the systematic study of all variants included in gnomAD, so that many gnomAD variants have no classification. ClinVar also includes many variants of uncertain significance. Nevertheless, ClinVar has numerous RPE65 variants because many diagnostic laboratories automatically transfer their variants and assessments. In addition, variants considered pathogenic or likely pathogenic by ClinVar are more likely to be accurate because testing laboratories use both clinical data and the ACMG/AMP criteria in their assessments. In contrast, LOVD and HGMD have fewer pathogenic RPE65 variants and their assessments may be less accurate because many variants were reported before the adoption of the ACMG/AMP criteria and the use of gnomAD to assess rarity.

Our results are different from the study performed in East Asian and Korean people that found a heterozygote or carrier frequency of 1 in 836 and 1 in 1,816, corresponding with population frequencies of 1 in 3,941,308 and 1 in 13,184,161, respectively.²⁰ The difference is probably because we included structural and founder changes. Our result for the population frequency for people of East Asian ancestry was not different from that for Europeans.

Although our estimated population frequency of biallelic predicted pathogenic variants in the population was 1 in 665,856 or 1 in 264,196 in admixed Americans or 1 in 1,764 for Amish, these results do not mean that everyone with 2 predicted pathogenic variants develops biallelic disease. This is because of incomplete penetrance or the presence of two pathogenic variants on the same chromosome (‘in cis’). Some people develop milder or later onset disease. Penetrance varies even in different family members with the same variant. Disease severity also depends on the effect of modifying genes and interacting alleles.

Overall, our assessment found that 435 people (35%) in gnomADv4.1 with a predicted pathogenic variant had a null variant and 818 (65%) had a missense change. Structural RPE65 variants were uncommon (50 in 797,922 or 1 in 15,958), but few people had undergone whole genome sequencing to detect such changes. Missense variants were much more common than reported previously, probably because null variants were easier to identify as pathogenic in routine laboratories. This relative increase in missense changes and the number of missense founder variants may predispose to milder disease.

Possible founder variants present in more than 15 people and classified predicted pathogenic in our assessment and pathogenic in ClinVar were counted in our estimate of population frequency, to reach as accurate an estimate as possible.

In conclusion, heterozygous predicted pathogenic RPE65 variants occur in at least 1 in 408 people, corresponding with a population frequency of biallelic disease of 1 in 665,856. Treatment is only available for biallelic disease. People of admixed American, African-American, and possibly Amish ancestries appear more likely to develop biallelic disease. Monoallelic RPE65-retinopathy disease owing to founder variants is present in 1 of 294,930 or 1 in 147,465 Europeans. Unknown founder variants may occur in other ancestries. Most people with disease-causing variants in developed countries have already been diagnosed, but those from developing countries or societies where consanguinity³⁵ is common are still at risk. Finally, rare RPE65 variants may be associated with novel phenotypes.

These results are based on current computational tools; however, more accurate estimates of the population frequency of RPE65-associated retinopathy will be available with future refinements and a better understanding of population structure and ancestry data.