Very Large Cystoid Macular Lesions Identified Using Outlier Analysis of Genetically Confirmed Inherited Retinal Disease Cases

ABSTRACT Background Cystoid macular lesions (CML) in inherited retinal diseases (IRDs) can contribute to vision impairment. Studying the morphologic range and outlier presentations of CML may inform clinical associations, mechanistic research, and trial design. Thus, we aim to describe the distribution of optical coherence tomography (OCT) parameters in IRD cases with CML and identify phenotype–genotype associations in very large cystoid macular lesions (VLCML). Materials and methods This cross-sectional study retrieved clinical information from electronic records from January 2020 to December 2021. VLCML cases were identified using the robust distance (Mahalanobis) of the correlation between central foveal thickness (CFT) and total macular volume (TMV) and a 99.9% probability ellipse. The distribution of OCT parameters was calculated by genotype and phenotype. Results We included 173 eyes of 103 subjects. The median age was 55.9 (interquartile range [IQR], 37.9, 63.7) and 47.6% (49/103) were females. Patients had disease-causing mutations in 30 genes. The most common genes included USH2A (n = 18), RP1 (n = 12), and ABCA4 (n = 11). Robust distance analysis showed that the prevalence of VLCML was 1.94% (n = 2 patients, 4 eyes). VLCML was seen in cases of NR2E3 (119-2A>C) and BEST1 (1120_1121insG) mutations. The median CFT in cases without VLCML was 269 µm (IQR 209, 318.50) while the median for VLCML cases was 1,490 µm (IQR 1,445.50, 1,548.00) (P < .001). Conclusions Subjects with different IRD genotypes may develop VLCMLs. Future studies could consider the range and outlier values of CML foveal thickness when determining inclusion criteria and biostatistical plans for observational and interventional studies.


Introduction
Cystoid macular lesions (CML) in subjects with inherited retinal diseases (IRD) are a biologically complex trait with a broad phenotypic range.These characteristics make CML a particularly compelling target for extreme phenotype analysis.Studying extreme phenotypes is an approach that increases the power to detect molecular determinants of human diseases and illuminate genotype-phenotype correlations (1,2).In this approach, phenotype severity is classified as extreme if it falls beyond a statistically defined threshold, and then molecular and genetic assays are targeted to the extreme phenotype subgroup.This workflow ensures that those resourceintensive assays are deployed in a rationally selected sample and is particularly relevant to multifactorial or complex traits.
CML in subjects with IRD are a particularly complex and phenotypically diverse trait that might benefit from this approach.Among IRD subjects, more than 50% develop CML (3)(4)(5)(6).Multiple mechanisms, including vitreous traction, inflammation, and loss of cell-to-cell adhesion, have been proposed as mechanisms of CML in IRD, yet the molecular pathogenesis and definitive treatment remain unknown (7,8).Acknowledging that resource limitations preclude molecular studies in all IRD patients with CML, we implemented an approach to define a phenotypic subgroup for this purpose.Thus, this study aimed to describe a very large cystoid macular lesion (VLCML) phenotype in IRD, estimate its frequency, and determine associated genotypes.In addition, we aim to describe the distribution of optical coherence tomography (OCT) structural parameters underlying the non-VLCML cases (i.e.99.9% of cases with CML) to potentially aid in the design of future prospective studies and therapeutic trials.

Methods
In this cross-sectional study, we retrieved records of IRD patients at the Johns Hopkins Wilmer Eye Institute from January 2020 to December 2021.The Johns Hopkins University Institutional Review Board approved this study.Informed consent was waived due to the retrospective design.We included subjects with confirmed IRD genotypes and evidence of CML on optical coherence tomography (OCT, Heidelberg Spectralis).Demographic and clinical characteristics were obtained from electronic records.For this study, genetic testing results were considered confirmatory if mutations were classified as either likely pathogenic or pathogenic according to the American College of Medical Genetics and Genomics guidelines (9).In subjects with autosomal recessive alleles, homozygosity or compound heterozygosity was required.Segregation analyses were not routinely performed.Subjects without OCT images were excluded from the analysis.In all images, the internal limiting membrane and retinal pigmented epithelium (RPE) were traced to ensure the repeatability and accuracy of the measurements.We used the 6 mm scan with 49 horizontal rasters per scan.Following manual correction of foveal centration and retinal layer segmentation, we extracted the following measurements from the manufacturer's software: (1) central foveal thickness (CFT, [µm]), (2) total macular volume (TMV, [mm 3 ]), and (3) maximum thickness (Tmax [µm]).

Statistical analysis
Baseline characteristics were calculated using descriptive statistics, and the Wilcoxon rank-sum test tested the difference in medians.First, we evaluated the correlation between CFT and TMV.Then, we created a 99.9% (%) probability prediction ellipse for the bivariate distribution.To assess VLCML phenotypes, we identified values with high leverage on both axes of the distribution.Values with high leverage were defined as points with a greater robust distance (Mahalanobis) from the whole sample (10,11).The Mahalanobis distance (also referred to as the greatest robust distance) is used to describe, statistically, how 'far' the value of any single value is from the population mean.A larger value indicates a greater 'distance' from the mean, and large values are typically associated with statistical outliers or unusually extreme sample values.The statistical methodology underlying the greatest robust distance estimation was described by Pasantra C. Mahalanobis in 1936.To obtain this value, we compute the dataset's mean vector, covariance matrix, and inverse of the covariance matrix.The following formula is used to calculate the Mahalanobis distance for sample x: where 'inv' denotes matrix inversion and '′' denotes the transpose operation (12).The distribution of structural parameters in the sample was evaluated separately by including or excluding VLCML cases and for VLCML cases alone.All analyses performed were conducted in R version 4.0 (13).

Results
We identified 106 records of IRD cases with CML using OCT imaging.Of these, three patients with missing imaging were excluded.Thus, the final sample included 173 eyes of 103 subjects.The median age at the time of the OCT evaluation was 55.9 years (interquartile range [IQR], 37.9, 63.7), and 47.6% (49/103) were female (Table 1).The prevalence of VLCML in the study sample was 1.94% (n = 2 patients, 4 eyes) (Figure 1a), and all cases were under 35 years at the time of OCT evaluation.

Very large cystoid macular lesions
Case 1 was a male subject with compound heterozygous mutations in NR2E3 (119-2A>C and 259dup (Ala87Glyfs*54)), resulting in enhanced S-cone syndrome (ESCS).In this subject, VLCML in the right eye (OD) was characterized by CFT 1490 µm, TMV 25.7 mm 3 , and Tmax 1535 µm.In the left eye (OS), it was characterized by CFT 1312 µm, TVM 30.7 mm 3 , and Tmax 1768 µm in the left eye (OS) (Figure 4a).A summary table of the cases is available in Table 2.
Case 2 was a female subject with homozygous mutations in BEST1 (1120_1121insG) resulting in autosomal recessive bestrophinopathy.The fundoscopic evaluation showed CML in both eyes and fleck-like deposits throughout the macula at baseline.In this subject, VLCML in the OD was characterized by CFT 1490 µm, TMV 22.8 mm 3 , and Tmax 1511 µm and CFT 1722 µm, TMV 18.1 mm3, and Tmax 1368 µm in the OS (Figure 4b).There were no clinical signs of choroidal neovascularization, such as hemorrhage, RPE detachment, or exudates in either case.Fluorescein angiography (FA) demonstrated pooling without vascular leakage or occlusion in both cases.

Discussion
The results showed that a small group of IRD patients may present extreme CML parameters by OCT above the 99.9% distribution threshold.VLCML is genetically heterogeneous: in this sample, VLCML cases had mutations in NR2E3 and BEST1.The OCT thickness and volume values in VLCML cases were statistically significantly larger compared to non-VLCML cases and had significant leverage in the parameter's distribution.This series also shows that VLCML can arise in IRD with mutations primarily affecting photoreceptor cells (NR2E3) and RPE cells (BEST1) (14,15).Some shared characteristics between VLCML and non-VLCML cases included the presence of cystoid spaces in the outer nuclear layer, inner nuclear layer, and inner plexiform layer.The most notable difference was the size of the lesions.Both cases with VLCML had pockets measuring more than 400 microns in the greatest linear dimension and were located closer to the fovea.Potentially, the data from our study may be leveraged to enable further research on CML in IRD.Firstly, regarding research on molecular pathogenesis, VLCML cases selected from a cohort of CML cases may be prioritized for aqueous humor, serum biomarker studies, or genomic sequencing.For this purpose, the statistical threshold for VLCML may be modified depending on the resources available for biochemical, metabolic, or sequencing assays.Secondly, the distribution of CML structural parameters in IRD cases may aid the design of prospective natural history studies for this condition.This approach is usually referred to as extreme phenotype study and is sometimes used in genetic research to help identify the molecular basis of complex diseases or traits.VLCML cases are at one extreme end of a phenotype (observable characteristics), and these individuals have the most severe presentations.For instance, VLCML cases may be analysed separately to determine if there were unique structural and functional progression characteristics in VLCML versus non-VLCML cases.In addition, VCLML cases may lead to identifying novel cellular and molecular mechanisms that influence CML formation, potentially leading to the development of new and rationally targeted treatments.Thirdly, for future treatment studies on CML in the context of IRD, VLCML cases may skew baseline structural or outcome parameters and may need to be explicitly addressed in terms of inclusion/exclusion criteria.If numerous outliers are identified in a certain group, it may suggest that the group was not well defined or that the study population was highly heterogeneous.Outliers can have a significant impact on statistical analyses, such as regression or mean calculation which rely on the assumption of normal distribution.Formulating a biostatistical plan to remove or adjust for outliers, in this particular case based on the foreknowledge of VLCML as an entity, can help to ensure the accuracy and generalizability of sample analyses.The robust method and inclusion of cases below the 98th percentile has previously been recognized as an optimal method to create reference limits of outlier analysis in randomized clinical trials (16).Lastly, the response to the treatment of VLCML cases deserves further study.It is plausible that extreme structural characteristics are associated with an unfavourable response to treatment and that further phenotype risk stratification is needed.In addition, there is increasing interest in understanding the molecular mechanisms and clinical implications of these particular cystoid retinal lesions that do not show leakage on FA (17).
Previous publications have reported a high frequency of CML in cases with NR2E3 and BEST1 mutations (18)(19)(20).Khojasteh et al. reported that macular retinoschisis was observed in 95% of patients in a 25-patient cohort with BEST1 mutations (18).Moreover, it is estimated that 50% to 70% of patients with ESCS present cystic macular changes (20).Those reports did not specifically address VLCML but are in broad agreement with a possible phenotype-genotype association between these mutations and CML and/or VLCML.
A report of ocular findings in Zellweger spectrum disorders included images of CML, including at least one case that appeared similar to VLCML in terms of its severity (21).The Zellweger spectrum disorders are associated with abnormalities in peroxisome biogenesis due to mutations in the PEX gene family.These data suggest that VLCML may occur in syndromic cases of hereditary retinopathy, and, moreover, that the perturbation of specific molecular pathways may be associated with VLCML.
We acknowledge several limitations in the study.The main limitation is the lack of visual acuity and other functional data to illuminate visual function deficits associated with VLCML.Other limitations include the lack of FA in all CML cases and of grading center evaluation of CML thickness in this dataset.We also did not systematically exclude other causes of CML, such as diabetic macular oedema, relying instead on a presumed link with IRD based on clinical documentation.In addition, we acknowledge the limitations of using crosssectional analysis and the inherent lack of time-resolved data over the patient's clinical course.This study is not designed to delineate the natural history of CML or VLCML in patients with IRD.It is widely recognized that CML can be dynamic phenomena with sometimes widely fluctuating structural severity.The temporal characteristics of VLCML are unknown.Future studies should consider these methodological improvements.Regardless, we believe this data will be important to determine in future research given the association between macular thickness and VA in patients with CML   (22,23).To our knowledge, there is a paucity of published data detailing the normative range and outlier values of cystoid macular lesions among IRD cases, and thus this information may be relevant for future clinical trial design, particularly in terms of inclusion and exclusion criteria and statistical analysis.
In summary, we applied robust distance analysis to CML OCT measurements data derived from a cohort of IRD cases with genetic confirmation, to identify cases that exceeded a biostatistical threshold definition of normality.Thus, we identified two cases of VLCML, occurring in distinct genetic types of IRD.These data indicate that certain IRD patients with CML may present extreme structural changes and suggest that the extreme phenotype may occur more broadly in multiple IRD genotypes.The presence of extreme or outlier values of CML thickness could be important to consider in the context of clinical research design.Future studies could consider the typical range, and extreme outlier values, of CML thickness when determining inclusion criteria and biostatistical plans for observational and interventional studies.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Figure 1 .
Figure 1.Very large cystoid macular lesions identification and central foveal thickness distribution stratified by genotype.(a) Very large cystoid macular lesions (VLML) were defined as points with the greatest leverage in both axes of the distribution using the robust distance (Mahalanobis) and 99.9% probability ellipse.Red dots represent the outliers based on the traditional statistical definition (less than 2.5th or greater than 97.5th percentiles).(b) Central tendency and distribution of central foveal thickness by genotype.

Table 2 .
Summary of extreme cystoid macular lesion cases.