Genetic Correlates of Phenotypic Variability in c.5882G>A p.(Gly1961Glu)-Associated Stargardt Disease

Abstract

Purpose

To quantitatively investigate disease progression and genotype–phenotype correlations in Stargardt disease patients harboring the common ABCA4 variant c.5882G>A p.(Gly1961Glu), aiming to evaluate the contribution of second ABCA4 variants to disease severity.

Methods

This multicenter study included 77 Stargardt disease patients uniformly carrying the ABCA4 c.5882G>A p.(Gly1961Glu) variant from Nijmegen, Bonn, and Basel. Ellipsoid zone (EZ) loss was measured using automated segmentation of optical coherence tomography images. Progression was assessed with linear mixed-effects models. Disease severity was quantified by the age-at-criterion EZ loss, defined as the estimated age when EZ atrophy reached 6.25 mm², allowing a comparison of the effects of different second alleles.

Results

A total of 52 patients were included in the study. The median age at first observation was 41 years (interquartile range, 27–52 years). The median age at criterion EZ loss across all patients was 43 years (interquartile range, 31–51 years). Substantial variability in the progression of EZ loss was observed among patients despite the shared c.5882G>A p.(Gly1961Glu) genotype. Differences in specific second ABCA4 variants were strongly associated with variation in disease progression, with some variants linked to markedly earlier progression (e.g., c.1957C>T, −50.7 years) and others associated with delayed progression (e.g., c.1648G>A, +19.8 years).

Conclusions

The ABCA4 c.5882G>A p.(Gly1961Glu) variant is associated with substantial interpatient variability in disease severity, heavily influenced by the nature of the second ABCA4 allele. These findings highlight the importance of precise genotypic characterization in prognostic counseling, clinical trial stratification, and the design of targeted gene therapies.

Stargardt disease (STGD1) is one of the most common inherited retinal dystrophies, resulting from biallelic pathogenic variants in the ABCA4 gene.¹ The disease is characterized by progressive central vision loss owing to photoreceptor degeneration and RPE atrophy.^2,3 Considerable phenotypic variability exists among patients, with differences in age of onset, progression rate, and severity, presenting a major challenge for both predicting disease progression and therapeutic development.^4–6

Among the numerous ABCA4 variants identified, the c.5882G>A p.(Gly1961Glu) variant is one of the most frequently observed, typically associated with a bull's eye maculopathy.^7–9 STGD1 associated with c.5882G>A p.(Gly1961Glu) is typically characterized by early loss of foveal vision, but a relatively mild disease and little to no extramacular involvement.^7–9 This mutation has been particularly interesting in clinical and genetic studies, given its potential implications for disease progression and therapeutic intervention. Recently, it was demonstrated that base editing of the c.5882G>A p.(Gly1961Glu) variant is feasible in primates and human retinal organoids, creating opportunities for potential gene-editing therapies targeting this mutation.¹⁰

Before such a therapy can be tested, a clear genotype–phenotype correlation must first be established to map the natural history of the disease accurately. The earliest studies predominantly used qualitative or semiquantitative approaches to phenotype classification. Notably, Fishman et al.¹¹ (1999) highlighted distinct phenotypic characteristics associated with specific ABCA4 mutations, laying the groundwork for subsequent natural history studies using more quantitative methodologies.

Previous investigations, based on linear models, aimed to assess the variant-specific effect on disease severity.¹² Key studies have used such models to identify the genetic correlates of disease severity in terms of retina-wide visual field changes,¹³ macular ellipsoid zone (EZ) loss,¹⁴ or ERG-based assessments.^15,16 Despite these advances, a more precise approach to genotype–phenotype correlation is needed. In particular, studies with one allele being held constant are lacking to date.

We hypothesized that, by maintaining one allele constant—specifically c.5882G>A p.(Gly1961Glu)—it is possible to achieve a more refined analysis of the second ABCA4 variant's contribution to disease severity. To this end, this study aimed to leverage a fully quantitative approach by using optical coherence tomography (OCT)-based detection of EZ loss as an objective metric for genotype–phenotype correlation in c.5882G>A p.(Gly1961Glu)–associated STGD1. This refined approach may enhance the precision of phenotypic classification and improve our understanding of the genetic determinants underlying STGD1 progression.

Methods

Patient Recruitment and Ethical Approval

Data for this study were collected from three tertiary referral centers: Nijmegen (the Netherlands), Bonn (Germany), and Basel (Switzerland). Ethical approval was obtained/waived at each site, ensuring compliance with institutional and national guidelines for human research (METC Oost-Nederland: waived, 2017-3535; Ethikkommission Nordwest- und Zentralschweiz: 2022-01242; the data from Bonn was retrospectively analyzed and anonymized in accordance with German data protection regulations and federal laws; therefore, approval from an ethics committee was not required). Informed consent was obtained from all participants, including permission for the secondary use of coded patient data.

Imaging and Data Extraction

Patient demographics, ABCA4 mutations, and visual acuity data were derived from medical records. Patients were included when they carried at least two ABCA4 mutations and had a clinical phenotype of STGD1.¹⁷ At least one of the ABCA4 mutations had to be c.5882G>A p.(Gly1961Glu). Spectral-domain OCT images (Heidelberg Spectralis, Heidelberg, Germany) were retrospectively collected from Bonn and Nijmegen, and prospectively in Basel (NCT05674058). EZ loss segmentation was performed using a validated convolutional neural network for automated layer extraction at the GRADE reading center.¹⁶ En face images of EZ thickness were generated, and the area of EZ loss was semiautomatically quantified using ImageJ 1.53t (National Institutes of Health, Bethesda, MD, USA) based on the layer absence in the photoreceptor outer segment thickness map. In eyes with a central bull's eye phenotype, an ellipse was fitted to the atrophic area using the create ellipse function. The extreme values of this ellipse were used for modeling the spatial growth (e.g., superior, inferior, nasal, and temporal) of the atrophic area over time. Images with the area of EZ loss exceeding the image frame of the en face thickness map were excluded. EZ loss area measurements underwent a square root transformation to account for two-dimensional growth to ensure linearity in subsequent statistical modeling.

Statistical Analysis

A linear model was applied to analyze EZ loss progression, accounting for patient-specific intercepts and slopes. This model enabled individualized assessment of disease progression while controlling for interpatient variability.

To quantify disease severity, we calculated the age-at-criterion EZ loss, defined as the age at which a (square root–transformed) EZ loss of 2.5 mm was expected (equivalent to 6.25 mm²). We selected this threshold because lesion growth is approximately linear within this range, allowing the age-at-criterion to serve as a time-invariant measure of disease severity.¹⁴ Although the 2.5-mm cutoff was to some extent arbitrary, the linearity of the lesion growth ensured that using other thresholds within this range would yield the same rank order of severity across patients.¹⁴ This age-at-criterion metric provided a time-invariant estimate of disease severity, ensuring that assessments remained consistent regardless of when a patient first presented to the clinic. The calculation incorporated both the lesion size at the initial visit and its estimated progression rate.

In cases where the lesion size at the first visit exceeded the OCT image frame—indicative of a severe phenotype—the EZ loss area was conservatively set to 36 mm², based on the dimensions of the imaging field. Consequently, severity estimates for these cases are slightly conservative, yielding marginally lower estimates of variant-associated disease burden.

Genotype–Phenotype Correlation

Subsequently, we estimated the effect of the second ABCA4 variant on the age-at-criterion EZ loss based on linear models. For patients with three ABCA4 variants, where the DNA report did not specify which mutations were in cis, gnomAD was used to assess if variants were most likely in cis (e.g., c.[1622T>C;3113C>T]). If this information was unavailable, the patient was excluded from this analysis. The age-at-criterion for each ABCA4 variant was compared with previously reported age-at-criterion values or the variant's documented effect on disease progression in the literature. Of the 10 articles that reported mutation-specific disease progression, we were able to compare with 4 articles.^13–16 The other studies were not suitable for comparison owing to pooled severity estimates without mutation specification, ordinally scaled severity estimates, or only one shared variant.^18–23 Figure 1 visualizes the methods used in this study.

Figure 1.

Visualization of the methods used for this study, according to a previously specified approach.¹⁴

Results

Patient Demographics and Data Quality

A total of 52 patients were included from the three centers (24 from Radboud University Medical Center Nijmegen, 20 from the Eye Clinic of the University Hospital Basel, and 8 from University Hospital Bonn) (Table 1). The convolutional neural network was able to segment the OCT scans of 50 patients reliably (as checked by grader JP). In nine patients, atrophy at the first measurement exceeded the image frame of the OCT image. An overview of these patients can be found in Supplementary Table S1. The flow chart inclusion can be found in Supplementary Figure S1. Of the remaining patients, the size of EZ loss and the spatial growth of the loss were measured.

Table 1.

Overview of the Cohort

Center	Patients Included	Age at First Visit, Years	VA OD*	VA OS*
1	24	48 (35–57)	0.20 (0.10–0.45)	0.20 (0.10–0.40)
2	20	29 (25–45)	0.17 (0.13–0.20)	0.19 (0.13–0.33)
3	8	41 (23–51)	0.10 (0.09–0.10)	0.10 (0.10–0.20)
Total	52	41 (27–52)	0.13 (0.10–0.21)	0.16 (0.10–0.38)

VA, visual acuity.

Values are number or median (interquartile range).

^*At first included visit.

Spatial Pattern of Bull's Eye Phenotype

The median area of EZ atrophy at the first available measurement was 2.82 mm² (interquartile range, 1.06–4.94 mm²). The mean growth rate of the square root–transformed atrophy in the whole cohort was 0.085 ± 0.0039 mm/year. The atrophic area's boundary was most distant from the fovea in the temporal direction (1.26 mm), followed by the nasal, inferior, and superior directions (1.23, 0.96, and 0.90 mm, respectively). The unidirectional growth rate was also highest in the temporal direction (0.054 mm/year), followed by the inferior, nasal, and superior directions (0.050 mm/year, 0.049 mm/year, and 0.048 mm/year, respectively) (Fig. 2).

Figure 2.

Spatial distribution of the atrophic area (a) and directional progression of the outer retinal atrophy (b).

Age-at-Criterion EZ Loss Across Variants

The median age at which the atrophic area size was estimated to reach 6.25 mm² (age-at-criterion size) was 43 years (interquartile range, 31–51 years). The intercept of the linear model analysis in which we assessed the effect of the second mutation on the age-at-criterion size was 51.7. Eight patients had a third variant in ABCA4. Of those, four patients had confirmed that the third variant was not in cis with c.5882G>A p.(Gly1961Glu); in two patients, phase was imputed based on gnomAD data; and in one case, one variant was considered as benign and thereby the other variant was included in the analysis. One patient was excluded from the genotype–phenotype analysis because it was unknown whether the third variant of this patient was in cis or in trans with c.5882G>A p.(Gly1961Glu). An overview of the cases with three mutations in ABCA4 can be found in Supplementary Table S2. According to our analysis, c.1957C>T had the most severe effect on the age-at-criterion size, decreasing it by −50.7 years. The mildest second mutation in our cohort was c.1648G>A, delaying the age-at-criterion size by 19.8 years. The most prevalent second mutation in our cohort, c.768G>T, which occurred in 11 patients, only decreased the age-at-criterion size by 7.6 years. Table 2 shows the effect of all mutations on the age-at-criterion size. Figure 3 contains the clinical phenotype of a severe, a common, and a mild patient to illustrate the phenotypic variability between patients carrying different mutations.

Table 2.

The Effect on the Age-at-criterion Size of Each Individual Second Variant in Our Cohort

Variant	Effect on Age-at-Criterion Size (Intercept = 51.7)	No. of Patients
c.1957C>T	−50.7	1
c.1834C>T	−28.0	1
c.2626C>T	−26.9	1
c.5882G>A	−23.9	1
c.[1622T>C;3113C>T]	−22.6	2
c.2894A>G	−22.6	1
c.4793C>A	−22.4	1
c.[5603A>T;6282+1G>C]	−21.3	1
c.3970del	−20.5	1
c.2768del	−18.4	2
c.3056C>T	−15.8	1
c.1853G>A	−14.2	1
c.5161_5162del	−12.7	1
c.3322C>T	−11.8	1
c.3819duo	−11.8	1
c.2041C>T	−10.7	1
c.6284A>T	−8.0	1
c.768G>T	−7.6	11
c.[3210_3211dup;5603A>T]	−6.2	1
c.3259G>A	−2.7	1
c.3874C>T	−2.2	1
c.4734del	−0.2	1
c.716G>A	0.0	1
c.5917delG	0.2	1
c.859-506G>C	1.8	2
c.5116del	4.6	1
c.5461-6T>C	7.8	1
c.3819dupT	9.1	1
c.5318C>T	13.7	1
deletion exons 44 to 50	15.7	1
c.[5461-10T>C;5603A>T]	16.9	3
c.4539+2001G>A	17.5	2
c.1648G>A	19.8	1

Figure 3.

Patient examples including a case with c.1957C>T, a case with c.4540-2A>G, and a case with c.768G>T. Note the phenotypic variability between these patients, ranging from early-onset disease to a milder late-onset phenotype. (Of note, the mild case with ABCA4 c.4540-2A>G as the second variant was not included in the quantitative analysis owing to an erroneous segmentation as it displayed an optical gap phenotype on OCT.)

Comparison With Prior Genotype–Phenotype Studies

Additionally, we compared our findings with previous genotype–phenotype correlation studies. As in Glinton et al.,¹⁶ c.[1622T>C;3113C>T] was associated with a more severe/earlier onset phenotype, whereas c.[5461-10T>C;5603A>T] was associated with a milder/later-onset phenotype. Although c.5882G>A p.(Gly1961Glu) was estimated as a mild variant in all previous studies, in our study, where c.5882G>A p.(Gly1961Glu) represented the estimates for c.5882G>A p.(Gly1961Glu) as the second variant (i.e., homozygous), it was not associated with an additional delay of the age-at-criterion size. Regarding the other studies exploring quantitative genotype–phenotype correlations, our study provided a limited number of shared variants. Figure 4 provides an overview of the comparisons between our findings and those of other studies. No patients overlapped among these cohorts.

Figure 4.

Comparison of the estimated effect of allele severity between previous studies and the current study.

Discussion

In this study, we demonstrated that disease progression in STGD1 patients harboring the c.5882G>A p.(Gly1961Glu) variant can be assessed quantitatively through automated OCT segmentation of the EZ. Contrary to the anticipated consistency of the bull's eye phenotype, our results uncovered significant interpatient variability associated with this genotype. This observed variability may be partly attributable to the nature of the second ABCA4 variant. Importantly, our patient cohort uniformly carried the c.5882G>A p.(Gly1961Glu) variant on one allele, whereas previous studies have included a broader range of genotypic combinations, resulting in an improved ability to assess the second variant's effect.^13–16

Currently, there is no standardized methodology for classifying the severity of ABCA4 variants. Beyond clinical genotype–phenotype correlations, several additional strategies have been used. These include examining statistical enrichment of variants among STGD1-affected populations compared with the general population, analyzing deviations in the frequency of homozygous variants relative to expected occurrences, and evaluating the severity of variants through their interaction with known mild alleles in compound heterozygous presentations. However, despite these varied approaches, the classification of variants continues to face considerable complexities and inconsistencies. For instance, the c.5882G>A p.(Gly1961Glu) variant examined herein demonstrates a severe phenotype with early-onset foveal atrophy and rapid loss of central visual acuity, yet simultaneously presents a relatively mild phenotype when assessed regarding retina-wide involvement. Genetically, c.5882G>A p.(Gly1961Glu) is typically considered mild, given the high minor allele frequency but paucity of affected homozygotes.²⁴

Not surprisingly, the complex c.[1622T>C;3113C>T] allele was associated with a particularly severe disease manifestation, consistent with previous reports showing that this allele accelerates disease onset relative to early truncating variants, likely owing to a combined loss-of-function and toxic gain-of-function mechanism.²⁵ However, we did not observe previous genotype–phenotype association for most other variants. For example, a missense variant not considered to be severe (c.1957G>T)¹⁶ was among the most severe phenotypes in our cohort, whereas a deletion of six exons was associated with a surprisingly mild phenotype.

This apparent inconsistency is possibly explained by the composition of our cohort, with all alleles occurring in trans with the common c.5882G>A variant by definition. Based on the results and discrepancy with previous results, the overall severity is likely predominantly driven by the c.5882G>A allele. Consequently, the contribution of the second allele becomes difficult to disentangle except for very severe alleles such as c.[1622T>C;3113C>T].

These findings have important clinical implications, particularly for prognostic counseling, because they highlight the effect of specific ABCA4 variants on the disease progression—enabling the clinician to more accurately estimate the expected disease trajectory in STGD1 patients carrying these mutations. However, especially in the context of patients with ABCA4 c.5882G>A as the first variant, usually observed genotype–phenotype correlations are less distinct.

Estimating disease progression based on specific mutations is not only useful in clinical practice. In clinical trials, clustering patients based on the expected disease progression (e.g., only fast-progressing patients) may help in attaining statistical power more easily, requiring a smaller sample size in the already limited patient population with STGD1. Participant stratification may help in creating more homogeneous groups for the clinical trials. Current clinical trials do not differentiate between ABCA4 variants, potentially unintentionally leading to heterogeneous treatment responses.

In addition to our genotype–phenotype correlation analysis, we investigated the spatial pattern of EZ atrophy expansion in patients carrying the c.5882G>A p.(Gly1961Glu) variant. Our results showed that EZ atrophy progressed fastest in the temporal and nasal directions, followed by inferior and superior expansion. To our knowledge, only one previous study has examined the directional growth of atrophy in STGD1. In contrast with our findings, that study did not report any significant directional differences in atrophy progression. However, their cohort included all Stargardt cases, whereas we focused specifically on individuals with the c.5882G>A p.(Gly1961Glu) variant, which is known to display a distinct pattern of disease progression.²⁶ Interestingly, the preferred retinal locus in STGD1 patients is reported in having a tendency to adopt a fixation point superior to the fovea. Our finding of the least EZ atrophy progression in the superior direction may help to explain or complement this observation.²⁷

A major limitation of this study is the small cohort size, which substantially constrains the strength and generalizability of the interpretations that can be drawn from our findings. In particular, the limited number of patients restricts the robustness of variant-specific severity estimates. Most second variants were observed only once in the cohort, meaning that their associated phenotypes are based on single-patient observations rather than reproducible patterns. Of the 33 mutations described, only 6 occurred in more than 1 individual, and several of these recurrent variants were identified in sibling pairs. Because of the small cohort size, we could not statistically correct for relatedness, and this factor may have affected the results for those particular mutations. Another limitation of our study is the absence of segregation analysis, which prevented confirmation that the identified variants were present in trans configuration. Larger multicenter studies incorporating comprehensive OCT biomarkers are warranted for improved accuracy. Such expanded analyses would facilitate deep learning–based modeling of disease progression, enabling individualized predictions with enhanced clinical utility.

Additionally, the recently identified deep intronic variant c.769-784C>T, known as a significant cis-acting modifier of c.5882G>A p.(Gly1961Glu), was not evaluated owing to the reliance on routine diagnostic protocols. As it is known that c.769-784C>T is present in approximately 15% of the c.5882G>A p.(Gly1961Glu) cases, it can be expected that eight cases in our study carry c.769-784C>T in trans with c.5882G>A. This factor may influence our results, because the complex c.[769-784C>T;5882G>A] allele is associated with a more severe phenotype.²⁸ Moreover, recent findings suggest that rare and common variants in the ROM1 and PRPH2 genes may act as trans-modifiers of STGD1, potentially contributing further to between-patient variability.²⁹

Conclusions

STGD1 patients harboring the c.5882G>A p.(Gly1961Glu) variant commonly exhibit the distinctive bull's eye phenotype. However, considerable variability in disease severity remains evident, even within a cohort uniformly carrying c.5882G>A p.(Gly1961Glu) on one allele. Notably, specific second alleles, such as the complex c.[1622T>C;3113C>T] allele, were associated with greater disease severity. This observation suggests additional pathogenic mechanisms in STGD1 beyond simple loss of function, which may complicate translational efforts.