Deep Learning Enables Genome-Wide Association Studies of Microvascular Features

Deep Learning Enables Genome-Wide Association Studies of Microvascular Features | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Deep Learning Enables Genome-Wide Association Studies of Microvascular Features Adham Elwakil, Sophie Strebel, Mathieu Charbonnier, José Vargas Quiros, and 10 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9039865/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Microvascular dysfunction underlies a wide range of cardiovascular, neurodegenerative and metabolic diseases, yet its genetic architecture remains largely unexplored owing to the lack of scalable in vivo phenotyping. Optical coherence tomography angiography (OCTA) enables direct visualization of human capillary networks and provides a unique window into microvascular biology. Here we perform, to our knowledge, the first genome-wide association study of capillary-scale microvasculature directly measured in vivo, using OCTA-derived foveal avascular zone (FAZ) area as a quantitative trait. We analysed two population-based European cohorts with automated deep learning–based image quality control and FAZ segmentation, followed by cohort-specific genome-wide association analyses and fixed-effects meta-analysis. Genomic inflation was minimal across analyses. Meta-analysis identified genome-wide significant loci on chromosomes 17 and 9 associated with FAZ area. These findings establish OCTA-derived FAZ metrics as genetically tractable microvascular phenotypes and provide a foundation for large-scale genetic studies of human microvasculature and its role in systemic disease. Computational Biology GWAS Microvasculature Deep Learning Retina FAZ Figures Figure 1 Figure 2 1. Introduction Microvascular dysfunction is a unifying mechanism across cardiovascular, neurodegenerative, renal and metabolic diseases, yet direct in‑vivo assessment of human microvasculature at scale remains challenging. The retinal circulation offers a uniquely accessible microvascular bed that shares developmental origin, calibre and autoregulatory properties with cerebral and coronary vessels, providing a non‑invasive window into systemic microvascular health 1 , 2 . Consistent with this homology, qualitative and quantitative abnormalities of retinal microvessels have been associated with incident ischemic heart disease, stroke and cardiovascular mortality in population‑based cohorts 3 . Optical coherence tomography (OCT) has transformed ophthalmic imaging by enabling rapid, high‑resolution, three‑dimensional visualization of retinal structure in clinical and population settings. Optical coherence tomography angiography (OCTA) extends this capability by exploiting motion contrast to reconstruct depth‑resolved images of the retinal and choroidal microvasculature without exogenous dye injection, allowing repeated quantification of vessel density, capillary morphology and flow voids in routine practice and research 4 . OCTA thus provides, for the first time, scalable access to the human capillary network in vivo, including the perifoveal capillary plexuses and the foveal avascular zone (FAZ), an anatomically well‑defined capillary‑free region at the center of the macula. The FAZ is central to foveal architecture and visual function. Its size and shape correlate with foveal pit morphology and visual acuity in healthy eyes, supporting its role as a structural determinant of high‑acuity vision 5 . Normative OCTA studies have shown that FAZ area exhibits substantial inter‑individual variation even among healthy adults, and is associated with central retinal thickness, macular vessel density and other local anatomical parameters, while being only weakly influenced by age and refractive error 6,7 . However, OCTA-derived FAZ measurements can be influenced by acquisition artefacts, segmentation definitions, and device-specific reconstruction, motivating robust automated phenotyping pipelines and harmonized analysis to enable genetic discovery 8 . Beyond normal variation, OCTA studies have linked FAZ enlargement and capillary rarefaction to a broad spectrum of ocular and systemic conditions. In diabetes, FAZ area and contour irregularity increase with severity of diabetic retinopathy and are detectable even in eyes without clinically apparent retinopathy, consistent with early macular ischemia 9 . FAZ enlargement and reduced parafoveal vessel density have also been reported in hypertensive disease, chronic kidney disease, neurodegenerative disorders, and systemic inflammatory or infectious conditions, suggesting that FAZ morphology integrates cumulative microvascular injury across multiple disease processes 4,9 . In Alzheimer’s disease and its preclinical stages, OCTA studies and systematic reviews have described reduced macular vessel density and, in several cohorts, a trend towards larger FAZ, raising the possibility that retinal microvascular changes could serve as non‑invasive biomarkers of early neurodegeneration 10 , 11 . In parallel, genome‑wide association studies (GWAS) have begun to dissect the genetic architecture of retinal traits derived from conventional imaging. Fundus photography–based analyses have identified numerous loci for retinal arteriolar and venular calibre and tortuosity, and demonstrated genetic links between retinal vasculometry and cardiometabolic traits, including blood pressure 12 , 13 . Structural OCT studies in large biobanks have mapped hundreds of loci for retinal layer thicknesses and other macular morphology traits, and shown that these endophenotypes capture shared genetic risk with glaucoma, age‑related macular degeneration and systemic cardiometabolic conditions 14 , 15 . More recently, deep‑learning based phenotyping of foveal pit depth has enabled GWAS of foveal morphology, implicating pathways involved in retinal development, pigment biology and extracellular matrix organization 16 . Despite this rapid progress, current retinal GWAS have almost exclusively focused on structural traits from color fundus photography or cross‑sectional OCT—such as vessel calibre, tortuosity and retinal layer thickness—rather than microvascular features measured directly from OCTA. As a result, the inherited determinants of capillary‑scale metrics, including FAZ area, perimeter, and shape descriptors, remain largely unexplored, even though these traits are strongly implicated as sensitive markers of local and systemic microvascular health. To our knowledge, no prior large‑scale GWAS has specifically targeted OCTA‑derived FAZ phenotypes as quantitative traits. In this study, we leverage OCTA imaging in two population-based European cohorts to perform genome-wide association analyses of FAZ morphology. Using fully automated deep learning–based quality control and segmentation, we derive quantitative FAZ phenotypes capturing size and shape, conduct GWAS within each cohort, and combine results through meta-analysis. We further apply gene-based and pathway-level aggregation as exploratory analyses and provide empirical benchmarks on signal strength and discoverability of OCTA-derived microvascular traits. By establishing a scalable framework for OCTA genetics, our work lays the groundwork for consortium-scale studies aimed at elucidating the inherited architecture of human microvasculature in vivo. 2. Methods 2.1 Study population The discovery analyses were conducted in OphtalmoLaus (OL), a population-based ophthalmic cohort from the Lausanne region, and replication was performed in part of the Rotterdam Study (RS-II, RS-III, RS-IV), a population-based cohort from Rotterdam, The Netherlands. Detailed cohort recruitment and examination protocols have been described previously 17,18 . Participants included in the present analyses had available macular OCTA imaging and genome-wide genotype data and were of predominantly European genetic ancestry, defined using principal component analysis (PCA). After applying automated OCTA image quality control and standard genotype quality filters, the final sample sizes were 1263 in OL and 1889 in RS. The study adhered to the tenets of the Declaration of Helsinki and was approved by the Commission cantonale d’éthique de la recherche sur l’être humain (CER-VD), Canton of Vaud, Switzerland. All participants provided written informed consent. 2.2 OCTA image acquisition Macular OCTA imaging in both OphtalmoLaus (OL) and the Rotterdam Study (RS) was performed using the Topcon Triton swept-source OCT system (Topcon Corporation, Tokyo, Japan), following the manufacturer’s standardized acquisition and motion-correction protocol. For each eye, an en-face OCTA scan centered on the fovea was acquired over a 4.5× 4.5 mm field in OL and 3 × 3 mm in RS. Scans were included only when the fovea was within the central region of the acquisition and suitable for FAZ delineation (see Image quality control). Analyses were restricted to the superficial capillary plexus (SCP). The SCP slab was defined identically across cohorts using the device layer segmentation boundaries, extending from the inner limiting membrane (ILM) to the inner plexiform layer/inner nuclear layer (IPL/INL) interface (or, equivalently, the lower boundary of the IPL as provided by the device segmentation). En-face OCTA images were generated from this slab using the manufacturer’s projection and decorrelation algorithm without the use of exogenous dye. To ensure comparability across different scan sizes, all FAZ phenotypes were computed in physical units (µm²) using the device-reported pixel spacing and scan dimensions embedded in the OCTA metadata. Because the FAZ is centrally located at the fovea, it is fully contained within both acquisition fields of view, enabling direct comparison of FAZ area across scan protocols. Figure 1 shows a comparison of the FAZ area distribution across the two cohorts. 2.3 Image quality control Quality control (QC) of OCTA images was performed using a fully automated, in-house deep learning–based QC model, specifically trained to assess FAZ-relevant image quality. The model produced a binary accept/reject decision for each scan based on image signal quality, motion artefacts, foveal centration, and segmentation suitability. No manual review or post-hoc correction was performed. If one eye failed QC, the contralateral eye was retained for analysis provided it passed QC. 2.4 FAZ segmentation and phenotype definition FAZ segmentation was performed using CapillaryX 19 , an automated deep learning pipeline for OCTA vascular segmentation and quantitative phenotyping. CapillaryX is based on the OCT2Former transformer architecture 20 and was initially trained on the publicly annotated OCTA-500 dataset 21 . To account for device and protocol specific variability, the pretrained FAZ model was fine-tuned using a subset of manually annotated superficial capillary plexus (SCP) en-face OCTA images from the OphtalmoLaus (OL) cohort. FAZ segmentation was performed directly on superficial capillary plexus (SCP) en-face OCTA projections using the dedicated FAZ model within CapillaryX. As part of the automated CapillaryX pipeline, predicted FAZ masks underwent built-in post-processing to identify the anatomically relevant central connected component using a center-weighted scoring approach, followed by hole filling to ensure a contiguous and topologically valid avascular region. FAZ boundaries were delineated fully automatically without user intervention during phenotype extraction. The primary quantitative phenotype derived in this study was FAZ area, expressed in square micrometers (µm²), computed using the physical pixel spacing of each scan. No explicit outlier exclusion was applied. For participants with high-quality scans in both eyes, the mean FAZ area across eyes was used as the final phenotype. If only one eye passed QC, the available eye was retained. 2.5 Covariates and phenotype preprocessing FAZ area values were adjusted for covariates using linear regression. The following covariates were included: Age, Sex, Age², Sex × Age, Sex × Age², Spherical refractive power, Cylindrical refractive power, Spherical power², Cylindrical power², Genetic principal components PCA₀–PCA₉. Residuals from the covariate-adjusted model were then transformed using a rank-based inverse normal transformation to ensure approximate normality prior to genome-wide association analysis. 2.6 Genotyping, quality control, and imputation Genotyping in OphtalmoLaus was performed using genome-wide SNP arrays following standard manufacturer protocols. Sample-level quality control excluded individuals with low genotype call rate, sex discordance, excessive heterozygosity, or cryptic relatedness beyond predefined thresholds. Variant-level QC excluded SNPs with low call rate, deviation from Hardy–Weinberg equilibrium, or minor allele frequency (MAF) below 1%. Genotypes were phased and imputed to a reference panel using established pipelines. Post-imputation, variants passing standard imputation quality filters and with MAF ≥ 0.01 were retained for association testing. All genomic coordinates refer to GRCh37. 2.7 Genome-wide association analysis Genome-wide association analyses were performed using BGENIE under an additive genetic model. The inverse-normal–transformed FAZ area residual served as the outcome variable, and imputed allele dosages were used as predictors. Association models included the full set of covariates described above. Genome-wide significance was defined as P < 5 × 10⁻⁸, with suggestive associations reported at P < 1 × 10⁻⁶. For each genome-wide significant locus, the lead SNP was defined as the variant with the smallest P-value within a ±250 kb window. 2.9 Gene‑ and pathway‑level analyses (PascalX) Gene- and pathway-level association analyses were performed using PascalX, a summary-statistics–based framework that aggregates SNP-level association signals into gene scores while accounting for local linkage disequilibrium (LD). GWAS summary statistics derived from the FAZ area association analyses were used as input. SNPs were mapped to genes based on GRCh37 gene annotations, with gene boundaries defined as the transcribed region extended by ±50 kb upstream and downstream to capture proximal regulatory variation. LD structure was modeled using genotype data from the 1000 Genomes Project Phase 3 European reference panel, consistent with the predominantly European ancestry of the study population. For each gene, PascalX computed a gene-level association statistic by integrating SNP-level P-values within the defined gene window while correcting for LD-induced correlation between variants. Pathway enrichment analyses were subsequently performed using curated gene sets from Gene Ontology (GO), KEGG, and Reactome databases. For each pathway, PascalX tested whether the observed distribution of gene-level association scores deviated from the null expectation, indicating enrichment of genetic association signal within biologically defined pathways. Multiple testing correction was applied using the default PascalX framework, and pathways surpassing the corrected significance threshold were considered statistically significant. All gene and pathway analyses reported in this study were conducted exclusively using PascalX; no additional functional mapping, expression quantitative trait locus (eQTL) prioritization, or fine-mapping tools were applied. 2.10 Ethical approval This study was conducted in accordance with the tenets of the Declaration of Helsinki. The OphtalmoLaus study was approved by the Commission cantonale d’éthique de la recherche sur l’être humain (CER-VD) (PB_2019-00168, Canton of Vaud, Switzerland), and all participants provided written informed consent. The Rotterdam Study was approved by the Medical Ethics Committee of the Erasmus MC (registration number MEC 02.1015) and by the Dutch Ministry of Health, Welfare and Sport (Population Screening Act WBO, license number 1071272-159521-PG), and all participants provided written informed consent for participation and use of their data for genetic and ophthalmic imaging research 3. Results 3.1 OCTA-derived FAZ phenotyping Using automated deep learning–based segmentation of macular OCTA images, we quantified the area of the foveal avascular zone (FAZ) in participants from the OphtalmoLaus (OL) cohort and the Rotterdam Study (RS). FAZ area was derived from superficial capillary plexus en-face OCTA images using the CapillaryX pipeline, following automated image quality control. When high-quality scans were available for both eyes, FAZ area was averaged across eyes; otherwise, the available eye was retained. After quality control and preprocessing, FAZ area residuals were adjusted for demographic, ocular, and genetic covariates and transformed using a rank-based inverse normal transformation prior to genetic analysis. Descriptive statistics of FAZ area in each cohort are provided in Table 2. 3.2 Genome-wide association analysis of FAZ area Genome-wide association analyses of FAZ area were performed separately in OL and RS, followed by inverse-variance weighted fixed-effects meta-analysis. After quality control, 7,109,791 variants were tested in OL, 7,483,593 variants in RS, and 6,983,731 variants shared across cohorts were included in the meta-analysis. Quantile–quantile plots showed minimal deviation from the null expectation in both cohorts and in the meta-analysis (genomic inflation factor λGC = 1.001 in OL, 1.024 in RS, and 1.014 in the meta-analysis), indicating no evidence of substantial residual population stratification (Supplementary Fig. S1). 3.3 Association signals in individual cohorts In the OL cohort, genome-wide association analysis identified several loci reaching suggestive significance (P < 1 × 10⁻⁶), with the strongest signal observed at rs8070929 on chromosome 17 (chr17:79,530,993; GRCh37), which reached near-genome-wide significance (P = 2.61 × 10⁻⁷, β = −0.212, SE = 0.041) (Fig. 2a; Table 1). An additional suggestive locus was observed at rs4352797 on chromosome 7 (chr7:52,302,196; P = 4.12 × 10⁻⁷, β = −0.288, SE = 0.056). In the RS cohort, no locus reached genome-wide significance (Fig. 2b). At the chromosome 17 lead variant rs8070929, nominal evidence of association was observed (P = 4.00 × 10⁻⁶, β = −0.171, SE = 0.037), whereas the chromosome 7 signal did not replicate (P = 0.812) (Table 1). 3.4 Meta-analysis across cohorts Meta-analysis of OL and RS strengthened association evidence at two independent loci. The lead variant rs8070929 on chromosome 17 reached genome-wide significance in the combined analysis (P = 5.04 × 10⁻¹², β = −0.189, SE = 0.027) (Fig. 2c; Table 1). A second locus on chromosome 9, represented by rs7872217, also surpassed the genome-wide significance threshold in the meta-analysis (P = 5.03 × 10⁻¹⁰) (Table 1; Supplementary Figs. S2–S3). A suggestive locus on chromosome 7 identified in OL did not reach genome-wide significance in the meta-analysis (P = 7.15 × 10⁻⁴). Because allele coding differed across cohorts for some variants, effect estimates are reported as provided by each cohort without allele harmonization. Consequently, effect directions should be interpreted cautiously, while statistical evidence for association is reflected by cohort-specific and meta-analytic P-values. 3.5 Summary of lead loci Table 1 summarizes lead variants for independent loci identified through meta-analysis using ±250 kb clumping, together with corresponding association statistics in the OphtalmoLaus (OL) cohort and the Rotterdam Study (RS). Manhattan plots for OL, RS, and the meta-analysis are shown in Fig. 2, and regional association plots for the genome-wide significant loci are provided in Supplementary Fig. S3. 3.6 Gene- and pathway-level association analyses Gene-based and pathway-level association analyses were performed using PascalX to aggregate SNP-level association statistics while accounting for local linkage disequilibrium. No gene reached the significance threshold after correction for multiple testing in the discovery, replication, or meta-analysis. Similarly, pathway enrichment analyses using curated gene sets did not identify any pathways significantly enriched for association signal. These results indicate that, at the current sample size, FAZ area does not appear to be driven by single genes or pathways with large aggregate effects detectable by gene-based testing. 4. Discussion This study represents, to our knowledge, the first genome-wide association analysis of human microvasculature directly measured in vivo, leveraging OCTA to quantify capillary-scale architecture rather than relying on indirect or macrovascular proxies. By focusing on the foveal avascular zone, an anatomically well-defined capillary-free region, our work establishes a new class of genetically tractable microvascular phenotypes. The retinal circulation offers a unique opportunity to study human microvasculature non-invasively at scale, sharing developmental origin, calibre range, and autoregulatory properties with cerebral and coronary microvessels. OCTA enables direct visualization of capillary networks that are otherwise inaccessible in living humans, positioning retinal OCTA as a gateway for genetic studies of systemic microvascular biology. Despite modest cohort sizes, meta-analysis identified identified genome-wide significant associations at two independent loci on chromosomes 17 and 9, supported by suggestive evidence in individual cohorts and minimal genomic inflation. This demonstrates that capillary-scale microvascular traits can yield detectable common-variant associations, while also highlighting the power limitations inherent to current OCTA datasets. Compared with GWAS of macrovascular retinal traits or structural OCT phenotypes, OCTA-derived microvascular traits are more sensitive to acquisition artefacts, segmentation uncertainty, and protocol heterogeneity. These factors likely attenuate effect sizes and reduce discovery yield, underscoring the importance of automated quality control, standardized phenotyping, and harmonized acquisition for future microvascular genetics efforts. FAZ area is tightly coupled to foveal specialization and retinal development, making it plausible that genetic determinants of foveal morphology also influence capillary organization. Large GWAS of foveal pit depth and macular structure have revealed extensive genetic regulation of foveal anatomy, providing a natural framework for interpreting FAZ genetics and motivating future integrative analyses across vascular and structural retinal traits. Because microvascular dysfunction is a shared mechanism across cardiovascular, neurodegenerative, renal, and metabolic diseases, establishing genetically informed microvascular endophenotypes opens new avenues for linking imaging-derived traits to systemic disease risk. OCTA-based GWAS may thus complement existing vascular genetics by capturing biological variation at the level where many diseases originate. Scaling OCTA GWAS through multi-center consortia, expanding beyond FAZ area to richer capillary descriptors (shape, irregularity, annular density), and integrating allele-harmonized meta-analysis and colocalization approaches will be essential to fully resolve the genetic architecture of human microvasculature. Conclusion This study demonstrates that capillary-scale microvascular traits directly measured in vivo using OCTA are amenable to genome-wide association analysis, establishing a new class of genetically tractable phenotypes for human microvascular biology. By performing, to our knowledge, the first GWAS of human microvasculature based on direct imaging rather than macrovascular proxies, we position OCTA-derived FAZ metrics as a unique bridge between retinal imaging, vascular genetics, and systemic microvascular health. Meta-analysis across two population-based cohorts identified genome-wide significant loci associated with FAZ area, while maintaining minimal genomic inflation, supporting the biological plausibility of the observed association and the feasibility of microvascular GWAS despite current sample-size limitations. At the same time, the modest number of detected loci highlights the technical and statistical challenges inherent to OCTA-based phenotyping, underscoring the need for standardized acquisition, robust automated quality control, and large-scale collaborative efforts. Together, these findings lay the groundwork for consortium-scale OCTA genetics and motivate future studies expanding beyond FAZ area to richer microvascular descriptors, with the potential to link genetically informed microvascular endophenotypes to systemic vascular and neurodegenerative disease risk. Declarations Acknowledgements This work was supported by the Swiss National Science Foundation (grant no. CRSII5 209510 to Sven Bergmann). We thank the authors of the OCTA-500 dataset (Li et al.) for collecting and publicly releasing this dataset on IEEE DataPort, which made this research possible. References Dorr, A. et al. Amyloid-β-dependent compromise of microvascular structure and function in a model of Alzheimer’s disease. Brain 135 , 3039–3050 (2012). Patton, N. et al. Retinal vascular image analysis as a potential screening tool for cerebrovascular disease: a rationale based on homology between cerebral and retinal microvasculatures. J Anat 206 , 319–348 (2005). Witt, N. et al. Abnormalities of retinal microvascular structure and risk of mortality from ischemic heart disease and stroke. Hypertension 47 , 975–981 (2006). Merriott, D. J. et al. Optical coherence tomography and optical coherence tomography angiography in systemic disease. Taiwan J Ophthalmol 15 , 364–377 (2025). Samara, W. A. et al. CORRELATION OF FOVEAL AVASCULAR ZONE SIZE WITH FOVEAL MORPHOLOGY IN NORMAL EYES USING OPTICAL COHERENCE TOMOGRAPHY ANGIOGRAPHY. Retina 35 , 2188–2195 (2015). Fujiwara, A. et al. Factors affecting foveal avascular zone in healthy eyes: An examination using swept-source optical coherence tomography angiography. PLoS One 12 , e0188572 (2017). Ghassemi, F. et al. The quantitative measurements of foveal avascular zone using optical coherence tomography angiography in normal volunteers. J Curr Ophthalmol 29 , 293–299 (2017). Linderman, R. E. et al. Variability of Foveal Avascular Zone Metrics Derived From Optical Coherence Tomography Angiography Images. Transl Vis Sci Technol 7 , 20 (2018). Freiberg, F. J. et al. Optical coherence tomography angiography of the foveal avascular zone in diabetic retinopathy. Graefes Arch Clin Exp Ophthalmol 254 , 1051–1058 (2016). O’Bryhim, B. E., Apte, R. S., Kung, N., Coble, D. & Van Stavern, G. P. Association of Preclinical Alzheimer Disease With Optical Coherence Tomographic Angiography Findings. JAMA Ophthalmol 136 , 1242–1248 (2018). Rifai, O. M. et al. The application of optical coherence tomography angiography in Alzheimer’s disease: A systematic review. Alzheimers Dement (Amst) 13 , e12149 (2021). Jiang, X. et al. GWAS on retinal vasculometry phenotypes. PLoS Genet 19 , e1010583 (2023). Tomasoni, M. et al. Genome-wide Association Studies of Retinal Vessel Tortuosity Identify Numerous Novel Loci Revealing Genes and Pathways Associated With Ocular and Cardiometabolic Diseases. Ophthalmol Sci 3 , 100288 (2023). Currant, H. et al. Genetic variation affects morphological retinal phenotypes extracted from UK Biobank optical coherence tomography images. PLoS Genet 17 , e1009497 (2021). Zekavat, S. M. et al. Phenome- and genome-wide analyses of retinal optical coherence tomography images identify links between ocular and systemic health. Sci Transl Med 16 , eadg4517 (2024). Hunt, C. et al. Genome-Wide Insights Into the Genes and Pathways Shaping Human Foveal Development: Redefining the Genetic Landscape of Foveal Hypoplasia. Invest Ophthalmol Vis Sci 66 , 22 (2025). Meloni, I. et al. Cohort profile: OphtalmoLaus, an extension of the CoLaus|PsyCoLaus cohort to investigate the relationships between ocular, cardiovascular, and cognitive parameters. (2025) doi:10.21203/rs.3.rs-7660406/v1. Hofman, A. et al. The Rotterdam Study: 2014 objectives and design update. Eur J Epidemiol 28 , 889–926 (2013). Elwakil, A. et al. CapillaryX: A Fine-Tunable Pipeline for OCTA Segmentation and Feature Extraction. (2025) doi:10.21203/rs.3.rs-8176394/v1. Tan, X. et al. OCT2Former: A retinal OCT-angiography vessel segmentation transformer. Comput. Methods Programs Biomed. 233 , 107454 (2023). Li, M. et al. OCTA-500: A retinal dataset for optical coherence tomography angiography study. Med. Image Anal. 93 , 103092 (2024). Tables Table 1. Lead variants for independent loci associated with FAZ area identified through meta-analysis, with corresponding association statistics in the OphtalmoLaus (OL) cohort and the Rotterdam Study (RS). Loci were defined using ±250 kb clumping around lead SNPs. Effect estimates (β) and alleles (A1/A2) are reported as provided by each cohort without allele harmonization. Chr Lead SNP Position (GRCh37) Cohort A1 A2 β SE P 17 rs8070929 79,530,993 Meta G T -0.189 0.027 5.04×10⁻¹² 17 rs8070929 79,530,993 OL G T -0.212 0.041 2.61×10⁻⁷ 17 rs8070929 79,530,993 RS T G -0.171 0.037 3.65×10⁻⁶ 9 rs7872217 21,573,443 Meta G A -0.170 0.027 5.03×10⁻¹⁰ 9 rs7872217 21,573,443 OL G A -0.102 0.044 1.98×10⁻² 9 rs7872217 21,573,443 RS A G -0.215 0.035 9.8×10⁻¹⁰ Table 2. Descriptive statistics of FAZ area in the OphtalmoLaus (OL) cohort and the Rotterdam Study (RS), including sample size (N), mean, standard deviation, minimum, maximum, range, 25th percentile, median, 75th percentile, and interquartile range (IQR). Area distribution characterisation OL RS N 1288 1899 Mean 260,261.87 315,748.97 Standard deviation 118,522.84 122,109.15 Minimum 3,856.20 1,767.20 Maximum 1157552.49 762,370.08 Range 1153696.29 760,602.88 25th Percentile 181735.84 230,950.95 Median 250208.13 314,362.79 75th Percentile 325255.74 396,162.06 IQR 143519.90 165,211.11 Additional Declarations The authors declare no competing interests. Supplementary Files SINERGIAOCTAGWASFAZpreprintV1SI.docx Deep Learning Enables Genome-Wide Association Studies of Microvascular Features Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9039865","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":601248638,"identity":"ceda2223-5482-4711-8a6b-99b26c46a7f7","order_by":0,"name":"Adham Elwakil","email":"","orcid":"https://orcid.org/0009-0009-4881-6643","institution":"Platform for Research in Ocular Imaging, Fondation Asile des Aveugles","correspondingAuthor":false,"prefix":"","firstName":"Adham","middleName":"","lastName":"Elwakil","suffix":""},{"id":601248639,"identity":"6886f097-0764-4d2b-b049-de7b6ad2a154","order_by":1,"name":"Sophie Strebel","email":"","orcid":"","institution":"Platform for Research in Ocular Imaging, Fondation Asile des Aveugles","correspondingAuthor":false,"prefix":"","firstName":"Sophie","middleName":"","lastName":"Strebel","suffix":""},{"id":601248640,"identity":"7762cc17-4777-4481-81e1-f199081e351f","order_by":2,"name":"Mathieu Charbonnier","email":"","orcid":"","institution":"Platform for Research in Ocular Imaging, Fondation Asile des Aveugles","correspondingAuthor":false,"prefix":"","firstName":"Mathieu","middleName":"","lastName":"Charbonnier","suffix":""},{"id":601248641,"identity":"33ccdcf8-cc04-43f3-a6a5-3eaa5eefb677","order_by":3,"name":"José Vargas Quiros","email":"","orcid":"https://orcid.org/0000-0002-9018-4928","institution":"Department of Ophthalmology, Erasmus University Medical Center, Rotterdam, The Netherlands","correspondingAuthor":false,"prefix":"","firstName":"José","middleName":"Vargas","lastName":"Quiros","suffix":""},{"id":601248642,"identity":"fa475a9f-6efe-4457-8e67-ac402edceb13","order_by":4,"name":"Bart Liefers","email":"","orcid":"","institution":"Department of Ophthalmology, Erasmus University Medical Center, Rotterdam, The Netherlands","correspondingAuthor":false,"prefix":"","firstName":"Bart","middleName":"","lastName":"Liefers","suffix":""},{"id":601248643,"identity":"30291354-27dd-40af-9ea9-be6ec740ef89","order_by":5,"name":"Lahcen Belatchane","email":"","orcid":"","institution":"Faculty of Medicine and Maieutics, Claude Bernard University Lyon 1, Lyon, France","correspondingAuthor":false,"prefix":"","firstName":"Lahcen","middleName":"","lastName":"Belatchane","suffix":""},{"id":601248644,"identity":"6f47ebcc-3e6e-4383-a604-7da100d47dea","order_by":6,"name":"Nastasia Foa","email":"","orcid":"","institution":"Department of Ophthalmology, University of Lausanne, Fondation Asile des Aveugles, Jules Gonin Eye Hospital, Lausanne, Switzerland.","correspondingAuthor":false,"prefix":"","firstName":"Nastasia","middleName":"","lastName":"Foa","suffix":""},{"id":601248645,"identity":"9d8a86be-555d-41f9-953c-7223a148f7f2","order_by":7,"name":"Sven Bergmann","email":"","orcid":"","institution":"Department of Computational Biology, University of Lausanne, Lausanne, Switzerland.","correspondingAuthor":false,"prefix":"","firstName":"Sven","middleName":"","lastName":"Bergmann","suffix":""},{"id":601248646,"identity":"d283c9f4-1d97-4e26-b458-e88f80c3836e","order_by":8,"name":"Caroline Klaver","email":"","orcid":"","institution":"Department of Ophthalmology, Erasmus University Medical Center, Rotterdam, The Netherlands","correspondingAuthor":false,"prefix":"","firstName":"Caroline","middleName":"","lastName":"Klaver","suffix":""},{"id":601248647,"identity":"3d004093-ca74-49d4-8fa1-b5160f168254","order_by":9,"name":"Ciara Bergin","email":"","orcid":"","institution":"Department of Ophthalmology, University of Lausanne, Fondation Asile des Aveugles, Jules Gonin Eye Hospital, Lausanne, Switzerland","correspondingAuthor":false,"prefix":"","firstName":"Ciara","middleName":"","lastName":"Bergin","suffix":""},{"id":601248648,"identity":"8af333c2-07a8-42f0-84d1-0aad9bf22824","order_by":10,"name":"Reinier Schlingemann","email":"","orcid":"","institution":"Department of Ophthalmology, University of Lausanne, Fondation Asile des Aveugles, Jules Gonin Eye Hospital, Lausanne, Switzerland","correspondingAuthor":false,"prefix":"","firstName":"Reinier","middleName":"","lastName":"Schlingemann","suffix":""},{"id":601248649,"identity":"f8bb1754-9ec7-4b7e-a8b1-573d32ef5d49","order_by":11,"name":"VascX Research Consortium","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"VascX","middleName":"Research","lastName":"Consortium","suffix":""},{"id":601248650,"identity":"89bbb58c-0a1c-45f8-8339-544a4fe8cb0b","order_by":12,"name":"Mattia Tomasoni","email":"","orcid":"","institution":"Platform for Research in Ocular Imaging, Fondation Asile des Aveugles, Jules Gonin Eye Hospital, Lausanne, Switzerland.","correspondingAuthor":false,"prefix":"","firstName":"Mattia","middleName":"","lastName":"Tomasoni","suffix":""},{"id":601248651,"identity":"98af4c4b-cb9f-4c96-af05-69ac6553af7a","order_by":13,"name":"Ilenia Meloni","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3ElEQVRIiWNgGAWjYBACNnYgkQBigRkVcgz8QEH8WphhWsCMM8YMkg0EtIBVwhmMbcYMBgcIaOFjZj724EENQzQ/iPFwnoE8wwHmZw/wO4wt3SDhGEPuzGYgI3GbgWFjA5u5AX4tPGYSCWwMuRsOAxmJ2/4wNjPwsEng18L/TSLhH0PufrCWOQb2bYS1ABUktgFtAVmX2GCQ2ENYCxtQZZ9E7ozDbGkSCccMkmeARPBpkW9vfib545tNbn978zHJHzUGtvuPNz/DqwUKkNUw41Q1CkbBKBgFo4BYAABDzTn2K8DgHgAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0001-9078-0538","institution":"Platform for Research in Ocular Imaging, Fondation Asile des Aveugles, Jules Gonin Eye Hospital, Lausanne, Switzerland.","correspondingAuthor":true,"prefix":"","firstName":"Ilenia","middleName":"","lastName":"Meloni","suffix":""}],"badges":[],"createdAt":"2026-03-05 12:11:42","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-9039865/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9039865/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":104180591,"identity":"506a7d3f-e6bc-4d85-9253-905920b64c0b","added_by":"auto","created_at":"2026-03-08 17:16:22","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":411658,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDistribution of FAZ area in the OL and RS cohorts. \u003c/strong\u003eKernel density estimates of the foveal avascular zone (FAZ) area are shown for the OL (blue, \u003cem\u003en\u003c/em\u003e= 1288) and RS (orange, \u003cem\u003en\u003c/em\u003e = 1889) cohorts. Shaded curves represent the estimated probability density of FAZ area within each cohort, while dashed vertical lines indicate the cohort medians. The RS cohort exhibits a right-shifted distribution compared with OL, suggesting larger FAZ areas on average.\u003c/p\u003e","description":"","filename":"Fig1.png","url":"https://assets-eu.researchsquare.com/files/rs-9039865/v1/631005e05732640902e8a18c.png"},{"id":104180601,"identity":"e9c95fe7-eb0b-4660-b250-6549d3a90e60","added_by":"auto","created_at":"2026-03-08 17:16:33","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":314540,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGenome-wide association analysis of FAZ area. \u003c/strong\u003eManhattan plots showing genome-wide association results for foveal avascular zone (FAZ) area in (a) the OphtalmoLaus (OL) cohort, (b) the Rotterdam Study (RS), and (c) the fixed-effects meta-analysis across cohorts. Each point represents a single nucleotide polymorphism plotted by chromosomal position (x-axis) and −log₁₀(P value) (y-axis), with alternating shading indicating chromosomes. The horizontal dashed line denotes the conventional genome-wide significance threshold (P = 5 × 10⁻⁸). While individual cohorts show suggestive association signals, the meta-analysis reveals multiple loci surpassing genome-wide significance.\u003c/p\u003e","description":"","filename":"Fig2.png","url":"https://assets-eu.researchsquare.com/files/rs-9039865/v1/47c7687aa3eb48fb9ab47dbf.png"},{"id":104409217,"identity":"fa8e2f72-fc3c-430c-8f22-6c438568c9c7","added_by":"auto","created_at":"2026-03-11 12:44:22","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1550723,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9039865/v1/5ea70317-eab3-418c-9a40-c2e7fcde44bf.pdf"},{"id":104404052,"identity":"74ca01b6-e560-4ac7-9f9d-e2c4ce1148df","added_by":"auto","created_at":"2026-03-11 12:19:39","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2992216,"visible":true,"origin":"","legend":"\u003cp\u003eDeep Learning Enables Genome-Wide Association Studies of Microvascular Features\u003c/p\u003e","description":"","filename":"SINERGIAOCTAGWASFAZpreprintV1SI.docx","url":"https://assets-eu.researchsquare.com/files/rs-9039865/v1/c650f317130bcbd8a547292b.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003e\u003cstrong\u003eDeep Learning Enables Genome-Wide Association Studies of Microvascular Features\u003c/strong\u003e\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eMicrovascular dysfunction is a unifying mechanism across cardiovascular, neurodegenerative, renal and metabolic diseases, yet direct in‑vivo assessment of human microvasculature at scale remains challenging. The retinal circulation offers a uniquely accessible microvascular bed that shares developmental origin, calibre and autoregulatory properties with cerebral and coronary vessels, providing a non‑invasive window into systemic microvascular health\u003csup\u003e1\u003c/sup\u003e\u003csup\u003e,\u003c/sup\u003e\u003csup\u003e2\u003c/sup\u003e. Consistent with this homology, qualitative and quantitative abnormalities of retinal microvessels have been associated with incident ischemic heart disease, stroke and cardiovascular mortality in population‑based cohorts\u003csup\u003e3\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eOptical coherence tomography (OCT) has transformed ophthalmic imaging by enabling rapid, high‑resolution, three‑dimensional visualization of retinal structure in clinical and population settings. Optical coherence tomography angiography (OCTA) extends this capability by exploiting motion contrast to reconstruct depth‑resolved images of the retinal and choroidal microvasculature without exogenous dye injection, allowing repeated quantification of vessel density, capillary morphology and flow voids in routine practice and research\u003csup\u003e4\u003c/sup\u003e. OCTA thus provides, for the first time, scalable access to the human capillary network in vivo, including the perifoveal capillary plexuses and the foveal avascular zone (FAZ), an anatomically well‑defined capillary‑free region at the center of the macula.\u003c/p\u003e\n\u003cp\u003eThe FAZ is central to foveal architecture and visual function. Its size and shape correlate with foveal pit morphology and visual acuity in healthy eyes, supporting its role as a structural determinant of high‑acuity vision\u003csup\u003e5\u003c/sup\u003e . Normative OCTA studies have shown that FAZ area exhibits substantial inter‑individual variation even among healthy adults, and is associated with central retinal thickness, macular vessel density and other local anatomical parameters, while being only weakly influenced by age and refractive error\u003csup\u003e6,7\u003c/sup\u003e. However, OCTA-derived FAZ measurements can be influenced by acquisition artefacts, segmentation definitions, and device-specific reconstruction, motivating robust automated phenotyping pipelines and harmonized analysis to enable genetic discovery\u003csup\u003e8\u003c/sup\u003e. \u003c/p\u003e\n\u003cp\u003eBeyond normal variation, OCTA studies have linked FAZ enlargement and capillary rarefaction to a broad spectrum of ocular and systemic conditions. In diabetes, FAZ area and contour irregularity increase with severity of diabetic retinopathy and are detectable even in eyes without clinically apparent retinopathy, consistent with early macular ischemia\u003csup\u003e9\u003c/sup\u003e. FAZ enlargement and reduced parafoveal vessel density have also been reported in hypertensive disease, chronic kidney disease, neurodegenerative disorders, and systemic inflammatory or infectious conditions, suggesting that FAZ morphology integrates cumulative microvascular injury across multiple disease processes\u003csup\u003e4,9\u003c/sup\u003e. In Alzheimer’s disease and its preclinical stages, OCTA studies and systematic reviews have described reduced macular vessel density and, in several cohorts, a trend towards larger FAZ, raising the possibility that retinal microvascular changes could serve as non‑invasive biomarkers of early neurodegeneration\u003csup\u003e10\u003c/sup\u003e\u003csup\u003e,\u003c/sup\u003e\u003csup\u003e11\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eIn parallel, genome‑wide association studies (GWAS) have begun to dissect the genetic architecture of retinal traits derived from conventional imaging. Fundus photography–based analyses have identified numerous loci for retinal arteriolar and venular calibre and tortuosity, and demonstrated genetic links between retinal vasculometry and cardiometabolic traits, including blood pressure\u003csup\u003e12\u003c/sup\u003e\u003csup\u003e,\u003c/sup\u003e\u003csup\u003e13\u003c/sup\u003e. Structural OCT studies in large biobanks have mapped hundreds of loci for retinal layer thicknesses and other macular morphology traits, and shown that these endophenotypes capture shared genetic risk with glaucoma, age‑related macular degeneration and systemic cardiometabolic conditions\u003csup\u003e14\u003c/sup\u003e\u003csup\u003e,\u003c/sup\u003e\u003csup\u003e15\u003c/sup\u003e. More recently, deep‑learning based phenotyping of foveal pit depth has enabled GWAS of foveal morphology, implicating pathways involved in retinal development, pigment biology and extracellular matrix organization\u003csup\u003e16\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eDespite this rapid progress, current retinal GWAS have almost exclusively focused on structural traits from color fundus photography or cross‑sectional OCT—such as vessel calibre, tortuosity and retinal layer thickness—rather than microvascular features measured directly from OCTA. As a result, the inherited determinants of capillary‑scale metrics, including FAZ area, perimeter, and shape descriptors, remain largely unexplored, even though these traits are strongly implicated as sensitive markers of local and systemic microvascular health. To our knowledge, no prior large‑scale GWAS has specifically targeted OCTA‑derived FAZ phenotypes as quantitative traits.\u003c/p\u003e\n\u003cp\u003eIn this study, we leverage OCTA imaging in two population-based European cohorts to perform genome-wide association analyses of FAZ morphology. Using fully automated deep learning–based quality control and segmentation, we derive quantitative FAZ phenotypes capturing size and shape, conduct GWAS within each cohort, and combine results through meta-analysis. We further apply gene-based and pathway-level aggregation as exploratory analyses and provide empirical benchmarks on signal strength and discoverability of OCTA-derived microvascular traits. By establishing a scalable framework for OCTA genetics, our work lays the groundwork for consortium-scale studies aimed at elucidating the inherited architecture of human microvasculature in vivo.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003ch2\u003e2.1 Study population\u003c/h2\u003e\n\u003cp\u003eThe discovery analyses were conducted in OphtalmoLaus (OL), a population-based ophthalmic cohort from the Lausanne region, and replication was performed in part of the Rotterdam Study (RS-II, RS-III, RS-IV), a population-based cohort from Rotterdam, The Netherlands. Detailed cohort recruitment and examination protocols have been described previously\u003csup\u003e17,18\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eParticipants included in the present analyses had available macular OCTA imaging and genome-wide genotype data and were of predominantly European genetic ancestry, defined using principal component analysis (PCA).\u003c/p\u003e\n\u003cp\u003eAfter applying automated OCTA image quality control and standard genotype quality filters, the final sample sizes were 1263 in OL and 1889 in RS.\u003c/p\u003e\n\u003cp\u003eThe study adhered to the tenets of the Declaration of Helsinki and was approved by the Commission cantonale d’éthique de la recherche sur l’être humain (CER-VD), Canton of Vaud, Switzerland. All participants provided written informed consent.\u003c/p\u003e\n\u003ch2\u003e2.2 OCTA image acquisition\u003c/h2\u003e\n\u003cp\u003eMacular OCTA imaging in both OphtalmoLaus (OL) and the Rotterdam Study (RS) was performed using the Topcon Triton swept-source OCT system (Topcon Corporation, Tokyo, Japan), following the manufacturer’s standardized acquisition and motion-correction protocol. For each eye, an en-face OCTA scan centered on the fovea was acquired over a 4.5× 4.5 mm field in OL and 3 × 3 mm in RS. Scans were included only when the fovea was within the central region of the acquisition and suitable for FAZ delineation (see Image quality control).\u003c/p\u003e\n\u003cp\u003eAnalyses were restricted to the superficial capillary plexus (SCP). The SCP slab was defined identically across cohorts using the device layer segmentation boundaries, extending from the inner limiting membrane (ILM) to the inner plexiform layer/inner nuclear layer (IPL/INL) interface (or, equivalently, the lower boundary of the IPL as provided by the device segmentation). En-face OCTA images were generated from this slab using the manufacturer’s projection and decorrelation algorithm without the use of exogenous dye.\u003c/p\u003e\n\u003cp\u003eTo ensure comparability across different scan sizes, all FAZ phenotypes were computed in physical units (µm²) using the device-reported pixel spacing and scan dimensions embedded in the OCTA metadata. Because the FAZ is centrally located at the fovea, it is fully contained within both acquisition fields of view, enabling direct comparison of FAZ area across scan protocols. Figure 1 shows a comparison of the FAZ area distribution across the two cohorts.\u003c/p\u003e\n\u003ch2\u003e2.3 Image quality control\u003c/h2\u003e\n\u003cp\u003eQuality control (QC) of OCTA images was performed using a fully automated, in-house deep learning–based QC model, specifically trained to assess FAZ-relevant image quality. The model produced a binary accept/reject decision for each scan based on image signal quality, motion artefacts, foveal centration, and segmentation suitability.\u003c/p\u003e\n\u003cp\u003eNo manual review or post-hoc correction was performed. If one eye failed QC, the contralateral eye was retained for analysis provided it passed QC.\u003c/p\u003e\n\u003ch2\u003e2.4 FAZ segmentation and phenotype definition\u003c/h2\u003e\n\u003cp\u003eFAZ segmentation was performed using CapillaryX\u003csup\u003e19\u003c/sup\u003e, an automated deep learning pipeline for OCTA vascular segmentation and quantitative phenotyping. CapillaryX is based on the OCT2Former transformer architecture\u003csup\u003e20\u003c/sup\u003e and was initially trained on the publicly annotated OCTA-500 dataset\u003csup\u003e21\u003c/sup\u003e. To account for device and protocol specific variability, the pretrained FAZ model was fine-tuned using a subset of manually annotated superficial capillary plexus (SCP) en-face OCTA images from the OphtalmoLaus (OL) cohort.\u003c/p\u003e\n\u003cp\u003eFAZ segmentation was performed directly on superficial capillary plexus (SCP) en-face OCTA projections using the dedicated FAZ model within CapillaryX. As part of the automated CapillaryX pipeline, predicted FAZ masks underwent built-in post-processing to identify the anatomically relevant central connected component using a center-weighted scoring approach, followed by hole filling to ensure a contiguous and topologically valid avascular region. FAZ boundaries were delineated fully automatically without user intervention during phenotype extraction.\u003c/p\u003e\n\u003cp\u003eThe primary quantitative phenotype derived in this study was FAZ area, expressed in square micrometers (µm²), computed using the physical pixel spacing of each scan. No explicit outlier exclusion was applied.\u003c/p\u003e\n\u003cp\u003eFor participants with high-quality scans in both eyes, the mean FAZ area across eyes was used as the final phenotype. If only one eye passed QC, the available eye was retained.\u003c/p\u003e\n\u003ch2\u003e2.5 Covariates and phenotype preprocessing\u003c/h2\u003e\n\u003cp\u003eFAZ area values were adjusted for covariates using linear regression. The following covariates were included: Age, Sex, Age², Sex × Age, Sex × Age², Spherical refractive power, Cylindrical refractive power, Spherical power², Cylindrical power², Genetic principal components PCA₀–PCA₉.\u003c/p\u003e\n\u003cp\u003eResiduals from the covariate-adjusted model were then transformed using a rank-based inverse normal transformation to ensure approximate normality prior to genome-wide association analysis.\u003c/p\u003e\n\u003ch2\u003e2.6 Genotyping, quality control, and imputation\u003c/h2\u003e\n\u003cp\u003eGenotyping in OphtalmoLaus was performed using genome-wide SNP arrays following standard manufacturer protocols. Sample-level quality control excluded individuals with low genotype call rate, sex discordance, excessive heterozygosity, or cryptic relatedness beyond predefined thresholds. Variant-level QC excluded SNPs with low call rate, deviation from Hardy–Weinberg equilibrium, or minor allele frequency (MAF) below 1%.\u003c/p\u003e\n\u003cp\u003eGenotypes were phased and imputed to a reference panel using established pipelines. Post-imputation, variants passing standard imputation quality filters and with MAF ≥ 0.01 were retained for association testing. All genomic coordinates refer to GRCh37.\u003c/p\u003e\n\u003ch2\u003e2.7 Genome-wide association analysis\u003c/h2\u003e\n\u003cp\u003eGenome-wide association analyses were performed using BGENIE under an additive genetic model. The inverse-normal–transformed FAZ area residual served as the outcome variable, and imputed allele dosages were used as predictors.\u003c/p\u003e\n\u003cp\u003eAssociation models included the full set of covariates described above. Genome-wide significance was defined as P \u0026lt; 5 × 10⁻⁸, with suggestive associations reported at P \u0026lt; 1 × 10⁻⁶. For each genome-wide significant locus, the lead SNP was defined as the variant with the smallest P-value within a ±250 kb window.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e2.9 Gene‑ and pathway‑level analyses (PascalX)\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eGene- and pathway-level association analyses were performed using PascalX, a summary-statistics–based framework that aggregates SNP-level association signals into gene scores while accounting for local linkage disequilibrium (LD). GWAS summary statistics derived from the FAZ area association analyses were used as input. SNPs were mapped to genes based on GRCh37 gene annotations, with gene boundaries defined as the transcribed region extended by ±50 kb upstream and downstream to capture proximal regulatory variation.\u003c/p\u003e\n\u003cp\u003eLD structure was modeled using genotype data from the 1000 Genomes Project Phase 3 European reference panel, consistent with the predominantly European ancestry of the study population. For each gene, PascalX computed a gene-level association statistic by integrating SNP-level P-values within the defined gene window while correcting for LD-induced correlation between variants.\u003c/p\u003e\n\u003cp\u003ePathway enrichment analyses were subsequently performed using curated gene sets from Gene Ontology (GO), KEGG, and Reactome databases. For each pathway, PascalX tested whether the observed distribution of gene-level association scores deviated from the null expectation, indicating enrichment of genetic association signal within biologically defined pathways. Multiple testing correction was applied using the default PascalX framework, and pathways surpassing the corrected significance threshold were considered statistically significant.\u003c/p\u003e\n\u003cp\u003eAll gene and pathway analyses reported in this study were conducted exclusively using PascalX; no additional functional mapping, expression quantitative trait locus (eQTL) prioritization, or fine-mapping tools were applied.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e2.10 Ethical approval\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eThis study was conducted in accordance with the tenets of the Declaration of Helsinki. The OphtalmoLaus study was approved by the \u003cem\u003eCommission cantonale d’éthique de la recherche sur l’être humain (CER-VD)\u003c/em\u003e (PB_2019-00168, Canton of Vaud, Switzerland), and all participants provided written informed consent. The Rotterdam Study was approved by the Medical Ethics Committee of the Erasmus MC (registration number MEC 02.1015) and by the Dutch Ministry of Health, Welfare and Sport (Population Screening Act WBO, license number 1071272-159521-PG), and all participants provided written informed consent for participation and use of their data for genetic and ophthalmic imaging research\u003c/p\u003e"},{"header":"3. Results","content":"\u003ch2\u003e3.1 OCTA-derived FAZ phenotyping\u003c/h2\u003e\n\u003cp\u003eUsing automated deep learning–based segmentation of macular OCTA images, we quantified the area of the foveal avascular zone (FAZ) in participants from the OphtalmoLaus (OL) cohort and the Rotterdam Study (RS). FAZ area was derived from superficial capillary plexus en-face OCTA images using the CapillaryX pipeline, following automated image quality control. When high-quality scans were available for both eyes, FAZ area was averaged across eyes; otherwise, the available eye was retained.\u003c/p\u003e\n\u003cp\u003eAfter quality control and preprocessing, FAZ area residuals were adjusted for demographic, ocular, and genetic covariates and transformed using a rank-based inverse normal transformation prior to genetic analysis. Descriptive statistics of FAZ area in each cohort are provided in Table 2.\u003c/p\u003e\n\u003ch2\u003e3.2 Genome-wide association analysis of FAZ area\u003c/h2\u003e\n\u003cp\u003eGenome-wide association analyses of FAZ area were performed separately in OL and RS, followed by inverse-variance weighted fixed-effects meta-analysis. After quality control, 7,109,791 variants were tested in OL, 7,483,593 variants in RS, and 6,983,731 variants shared across cohorts were included in the meta-analysis.\u003c/p\u003e\n\u003cp\u003eQuantile–quantile plots showed minimal deviation from the null expectation in both cohorts and in the meta-analysis (genomic inflation factor λGC = 1.001 in OL, 1.024 in RS, and 1.014 in the meta-analysis), indicating no evidence of substantial residual population stratification (Supplementary Fig. S1).\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e3.3 Association signals in individual cohorts\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eIn the OL cohort, genome-wide association analysis identified several loci reaching suggestive significance (P \u0026lt; 1 × 10⁻⁶), with the strongest signal observed at rs8070929 on chromosome 17 (chr17:79,530,993; GRCh37), which reached near-genome-wide significance (P = 2.61 × 10⁻⁷, β = −0.212, SE = 0.041) (Fig. 2a; Table 1). An additional suggestive locus was observed at rs4352797 on chromosome 7 (chr7:52,302,196; P = 4.12 × 10⁻⁷, β = −0.288, SE = 0.056).\u003c/p\u003e\n\u003cp\u003eIn the RS cohort, no locus reached genome-wide significance (Fig. 2b). At the chromosome 17 lead variant rs8070929, nominal evidence of association was observed (P = 4.00 × 10⁻⁶, β = −0.171, SE = 0.037), whereas the chromosome 7 signal did not replicate (P = 0.812) (Table 1).\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e3.4 Meta-analysis across cohorts\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eMeta-analysis of OL and RS strengthened association evidence at two independent loci. The lead variant rs8070929 on chromosome 17 reached genome-wide significance in the combined analysis (P = 5.04 × 10⁻¹², β = −0.189, SE = 0.027) (Fig. 2c; Table 1). A second locus on chromosome 9, represented by rs7872217, also surpassed the genome-wide significance threshold in the meta-analysis (P = 5.03 × 10⁻¹⁰) (Table 1; Supplementary Figs. S2–S3).\u003c/p\u003e\n\u003cp\u003eA suggestive locus on chromosome 7 identified in OL did not reach genome-wide significance in the meta-analysis (P = 7.15 × 10⁻⁴).\u003c/p\u003e\n\u003cp\u003eBecause allele coding differed across cohorts for some variants, effect estimates are reported as provided by each cohort without allele harmonization. Consequently, effect directions should be interpreted cautiously, while statistical evidence for association is reflected by cohort-specific and meta-analytic P-values.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e3.5 Summary of lead loci\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eTable 1 summarizes lead variants for independent loci identified through meta-analysis using ±250 kb clumping, together with corresponding association statistics in the OphtalmoLaus (OL) cohort and the Rotterdam Study (RS). Manhattan plots for OL, RS, and the meta-analysis are shown in Fig. 2, and regional association plots for the genome-wide significant loci are provided in Supplementary Fig. S3.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003e3.6 Gene- and pathway-level association analyses\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eGene-based and pathway-level association analyses were performed using PascalX to aggregate SNP-level association statistics while accounting for local linkage disequilibrium. No gene reached the significance threshold after correction for multiple testing in the discovery, replication, or meta-analysis. Similarly, pathway enrichment analyses using curated gene sets did not identify any pathways significantly enriched for association signal.\u003c/p\u003e\n\u003cp\u003eThese results indicate that, at the current sample size, FAZ area does not appear to be driven by single genes or pathways with large aggregate effects detectable by gene-based testing.\u003c/p\u003e"},{"header":"4. Discussion","content":"\u003cp\u003eThis study represents, to our knowledge, the first genome-wide association analysis of human microvasculature directly measured in vivo, leveraging OCTA to quantify capillary-scale architecture rather than relying on indirect or macrovascular proxies. By focusing on the foveal avascular zone, an anatomically well-defined capillary-free region, our work establishes a new class of genetically tractable microvascular phenotypes.\u003c/p\u003e\n\u003cp\u003eThe retinal circulation offers a unique opportunity to study human microvasculature non-invasively at scale, sharing developmental origin, calibre range, and autoregulatory properties with cerebral and coronary microvessels. OCTA enables direct visualization of capillary networks that are otherwise inaccessible in living humans, positioning retinal OCTA as a gateway for genetic studies of systemic microvascular biology.\u003c/p\u003e\n\u003cp\u003eDespite modest cohort sizes, meta-analysis identified identified genome-wide significant associations at two independent loci on chromosomes 17 and 9, supported by suggestive evidence in individual cohorts and minimal genomic inflation. This demonstrates that capillary-scale microvascular traits can yield detectable common-variant associations, while also highlighting the power limitations inherent to current OCTA datasets.\u003c/p\u003e\n\u003cp\u003eCompared with GWAS of macrovascular retinal traits or structural OCT phenotypes, OCTA-derived microvascular traits are more sensitive to acquisition artefacts, segmentation uncertainty, and protocol heterogeneity. These factors likely attenuate effect sizes and reduce discovery yield, underscoring the importance of automated quality control, standardized phenotyping, and harmonized acquisition for future microvascular genetics efforts.\u003c/p\u003e\n\u003cp\u003eFAZ area is tightly coupled to foveal specialization and retinal development, making it plausible that genetic determinants of foveal morphology also influence capillary organization. Large GWAS of foveal pit depth and macular structure have revealed extensive genetic regulation of foveal anatomy, providing a natural framework for interpreting FAZ genetics and motivating future integrative analyses across vascular and structural retinal traits.\u003c/p\u003e\n\u003cp\u003eBecause microvascular dysfunction is a shared mechanism across cardiovascular, neurodegenerative, renal, and metabolic diseases, establishing genetically informed microvascular endophenotypes opens new avenues for linking imaging-derived traits to systemic disease risk. OCTA-based GWAS may thus complement existing vascular genetics by capturing biological variation at the level where many diseases originate.\u003c/p\u003e\n\u003cp\u003eScaling OCTA GWAS through multi-center consortia, expanding beyond FAZ area to richer capillary descriptors (shape, irregularity, annular density), and integrating allele-harmonized meta-analysis and colocalization approaches will be essential to fully resolve the genetic architecture of human microvasculature.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study demonstrates that capillary-scale microvascular traits directly measured in vivo using OCTA are amenable to genome-wide association analysis, establishing a new class of genetically tractable phenotypes for human microvascular biology.\u003c/p\u003e\n\u003cp\u003eBy performing, to our knowledge, the first GWAS of human microvasculature based on direct imaging rather than macrovascular proxies, we position OCTA-derived FAZ metrics as a unique bridge between retinal imaging, vascular genetics, and systemic microvascular health.\u003c/p\u003e\n\u003cp\u003eMeta-analysis across two population-based cohorts identified genome-wide significant loci associated with FAZ area, while maintaining minimal genomic inflation, supporting the biological plausibility of the observed association and the feasibility of microvascular GWAS despite current sample-size limitations.\u003c/p\u003e\n\u003cp\u003eAt the same time, the modest number of detected loci highlights the technical and statistical challenges inherent to OCTA-based phenotyping, underscoring the need for standardized acquisition, robust automated quality control, and large-scale collaborative efforts.\u003c/p\u003e\n\u003cp\u003eTogether, these findings lay the groundwork for consortium-scale OCTA genetics and motivate future studies expanding beyond FAZ area to richer microvascular descriptors, with the potential to link genetically informed microvascular endophenotypes to systemic vascular and neurodegenerative disease risk.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch1\u003eAcknowledgements\u003c/h1\u003e\n\u003cp\u003eThis work was supported by the Swiss National Science Foundation (grant no. CRSII5 209510 to Sven Bergmann).\u003c/p\u003e\n\u003cp\u003eWe thank the authors of the OCTA-500 dataset (Li et al.) for collecting and publicly releasing this dataset on IEEE DataPort, which made this research possible.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDorr, A. \u003cem\u003eet al.\u003c/em\u003e Amyloid-\u0026beta;-dependent compromise of microvascular structure and function in a model of Alzheimer\u0026rsquo;s disease. \u003cem\u003eBrain\u003c/em\u003e \u003cstrong\u003e135\u003c/strong\u003e, 3039\u0026ndash;3050 (2012).\u003c/li\u003e\n\u003cli\u003ePatton, N. \u003cem\u003eet al.\u003c/em\u003e Retinal vascular image analysis as a potential screening tool for cerebrovascular disease: a rationale based on homology between cerebral and retinal microvasculatures. \u003cem\u003eJ Anat\u003c/em\u003e \u003cstrong\u003e206\u003c/strong\u003e, 319\u0026ndash;348 (2005).\u003c/li\u003e\n\u003cli\u003eWitt, N. \u003cem\u003eet al.\u003c/em\u003e Abnormalities of retinal microvascular structure and risk of mortality from ischemic heart disease and stroke. \u003cem\u003eHypertension\u003c/em\u003e \u003cstrong\u003e47\u003c/strong\u003e, 975\u0026ndash;981 (2006).\u003c/li\u003e\n\u003cli\u003eMerriott, D. J. \u003cem\u003eet al.\u003c/em\u003e Optical coherence tomography and optical coherence tomography angiography in systemic disease. \u003cem\u003eTaiwan J Ophthalmol\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 364\u0026ndash;377 (2025).\u003c/li\u003e\n\u003cli\u003eSamara, W. A. \u003cem\u003eet al.\u003c/em\u003e CORRELATION OF FOVEAL AVASCULAR ZONE SIZE WITH FOVEAL MORPHOLOGY IN NORMAL EYES USING OPTICAL COHERENCE TOMOGRAPHY ANGIOGRAPHY. \u003cem\u003eRetina\u003c/em\u003e \u003cstrong\u003e35\u003c/strong\u003e, 2188\u0026ndash;2195 (2015).\u003c/li\u003e\n\u003cli\u003eFujiwara, A. \u003cem\u003eet al.\u003c/em\u003e Factors affecting foveal avascular zone in healthy eyes: An examination using swept-source optical coherence tomography angiography. \u003cem\u003ePLoS One\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, e0188572 (2017).\u003c/li\u003e\n\u003cli\u003eGhassemi, F. \u003cem\u003eet al.\u003c/em\u003e The quantitative measurements of foveal avascular zone using optical coherence tomography angiography in normal volunteers. \u003cem\u003eJ Curr Ophthalmol\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 293\u0026ndash;299 (2017).\u003c/li\u003e\n\u003cli\u003eLinderman, R. E. \u003cem\u003eet al.\u003c/em\u003e Variability of Foveal Avascular Zone Metrics Derived From Optical Coherence Tomography Angiography Images. \u003cem\u003eTransl Vis Sci Technol\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 20 (2018).\u003c/li\u003e\n\u003cli\u003eFreiberg, F. J. \u003cem\u003eet al.\u003c/em\u003e Optical coherence tomography angiography of the foveal avascular zone in diabetic retinopathy. \u003cem\u003eGraefes Arch Clin Exp Ophthalmol\u003c/em\u003e \u003cstrong\u003e254\u003c/strong\u003e, 1051\u0026ndash;1058 (2016).\u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Bryhim, B. E., Apte, R. S., Kung, N., Coble, D. \u0026amp; Van Stavern, G. P. Association of Preclinical Alzheimer Disease With Optical Coherence Tomographic Angiography Findings. \u003cem\u003eJAMA Ophthalmol\u003c/em\u003e \u003cstrong\u003e136\u003c/strong\u003e, 1242\u0026ndash;1248 (2018).\u003c/li\u003e\n\u003cli\u003eRifai, O. M. \u003cem\u003eet al.\u003c/em\u003e The application of optical coherence tomography angiography in Alzheimer\u0026rsquo;s disease: A systematic review. \u003cem\u003eAlzheimers Dement (Amst)\u003c/em\u003e \u003cstrong\u003e13\u003c/strong\u003e, e12149 (2021).\u003c/li\u003e\n\u003cli\u003eJiang, X. \u003cem\u003eet al.\u003c/em\u003e GWAS on retinal vasculometry phenotypes. \u003cem\u003ePLoS Genet\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, e1010583 (2023).\u003c/li\u003e\n\u003cli\u003eTomasoni, M. \u003cem\u003eet al.\u003c/em\u003e Genome-wide Association Studies of Retinal Vessel Tortuosity Identify Numerous Novel Loci Revealing Genes and Pathways Associated With Ocular and Cardiometabolic Diseases. \u003cem\u003eOphthalmol Sci\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, 100288 (2023).\u003c/li\u003e\n\u003cli\u003eCurrant, H. \u003cem\u003eet al.\u003c/em\u003e Genetic variation affects morphological retinal phenotypes extracted from UK Biobank optical coherence tomography images. \u003cem\u003ePLoS Genet\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, e1009497 (2021).\u003c/li\u003e\n\u003cli\u003eZekavat, S. M. \u003cem\u003eet al.\u003c/em\u003e Phenome- and genome-wide analyses of retinal optical coherence tomography images identify links between ocular and systemic health. \u003cem\u003eSci Transl Med\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, eadg4517 (2024).\u003c/li\u003e\n\u003cli\u003eHunt, C. \u003cem\u003eet al.\u003c/em\u003e Genome-Wide Insights Into the Genes and Pathways Shaping Human Foveal Development: Redefining the Genetic Landscape of Foveal Hypoplasia. \u003cem\u003eInvest Ophthalmol Vis Sci\u003c/em\u003e \u003cstrong\u003e66\u003c/strong\u003e, 22 (2025).\u003c/li\u003e\n\u003cli\u003eMeloni, I. \u003cem\u003eet al.\u003c/em\u003e Cohort profile: OphtalmoLaus, an extension of the CoLaus|PsyCoLaus cohort to investigate the relationships between ocular, cardiovascular, and cognitive parameters. (2025) doi:10.21203/rs.3.rs-7660406/v1.\u003c/li\u003e\n\u003cli\u003eHofman, A. \u003cem\u003eet al.\u003c/em\u003e The Rotterdam Study: 2014 objectives and design update. \u003cem\u003eEur J Epidemiol\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e, 889\u0026ndash;926 (2013).\u003c/li\u003e\n\u003cli\u003eElwakil, A. \u003cem\u003eet al.\u003c/em\u003e CapillaryX: A Fine-Tunable Pipeline for OCTA Segmentation and Feature Extraction. (2025) doi:10.21203/rs.3.rs-8176394/v1.\u003c/li\u003e\n\u003cli\u003eTan, X. \u003cem\u003eet al.\u003c/em\u003e OCT2Former: A retinal OCT-angiography vessel segmentation transformer. \u003cem\u003eComput. Methods Programs Biomed.\u003c/em\u003e \u003cstrong\u003e233\u003c/strong\u003e, 107454 (2023).\u003c/li\u003e\n\u003cli\u003eLi, M. \u003cem\u003eet al.\u003c/em\u003e OCTA-500: A retinal dataset for optical coherence tomography angiography study. \u003cem\u003eMed. Image Anal.\u003c/em\u003e \u003cstrong\u003e93\u003c/strong\u003e, 103092 (2024).\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1. Lead variants for independent loci associated with FAZ area identified through meta-analysis, with corresponding association statistics in the OphtalmoLaus (OL) cohort and the Rotterdam Study (RS). Loci were defined using \u0026plusmn;250 kb clumping around lead SNPs. Effect estimates (\u0026beta;) and alleles (A1/A2) are reported as provided by each cohort without allele harmonization.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"602\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 7.16667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eChr\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLead SNP\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePosition (GRCh37)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.6667%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCohort\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eA1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eA2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10%;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026beta;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.5%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8333%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eP\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 7.16667%;\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14%;\"\u003e\n \u003cp\u003ers8070929\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1667%;\"\u003e\n \u003cp\u003e79,530,993\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.6667%;\"\u003e\n \u003cp\u003eMeta\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10%;\"\u003e\n \u003cp\u003e-0.189\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.5%;\"\u003e\n \u003cp\u003e0.027\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8333%;\"\u003e\n \u003cp\u003e5.04\u0026times;10⁻\u0026sup1;\u0026sup2;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 7.16667%;\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14%;\"\u003e\n \u003cp\u003ers8070929\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1667%;\"\u003e\n \u003cp\u003e79,530,993\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.6667%;\"\u003e\n \u003cp\u003eOL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10%;\"\u003e\n \u003cp\u003e-0.212\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.5%;\"\u003e\n \u003cp\u003e0.041\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8333%;\"\u003e\n \u003cp\u003e2.61\u0026times;10⁻⁷\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 7.16667%;\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14%;\"\u003e\n \u003cp\u003ers8070929\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1667%;\"\u003e\n \u003cp\u003e79,530,993\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.6667%;\"\u003e\n \u003cp\u003eRS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10%;\"\u003e\n \u003cp\u003e-0.171\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.5%;\"\u003e\n \u003cp\u003e0.037\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8333%;\"\u003e\n \u003cp\u003e3.65\u0026times;10⁻⁶\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 7.16667%;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14%;\"\u003e\n \u003cp\u003ers7872217\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1667%;\"\u003e\n \u003cp\u003e21,573,443\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.6667%;\"\u003e\n \u003cp\u003eMeta\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10%;\"\u003e\n \u003cp\u003e-0.170\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.5%;\"\u003e\n \u003cp\u003e0.027\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8333%;\"\u003e\n \u003cp\u003e5.03\u0026times;10⁻\u0026sup1;⁰\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 7.16667%;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14%;\"\u003e\n \u003cp\u003ers7872217\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1667%;\"\u003e\n \u003cp\u003e21,573,443\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.6667%;\"\u003e\n \u003cp\u003eOL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10%;\"\u003e\n \u003cp\u003e-0.102\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.5%;\"\u003e\n \u003cp\u003e0.044\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8333%;\"\u003e\n \u003cp\u003e1.98\u0026times;10⁻\u0026sup2;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 7.16667%;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 14%;\"\u003e\n \u003cp\u003ers7872217\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 24.1667%;\"\u003e\n \u003cp\u003e21,573,443\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10.6667%;\"\u003e\n \u003cp\u003eRS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 5.83333%;\"\u003e\n \u003cp\u003eG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 10%;\"\u003e\n \u003cp\u003e-0.215\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 8.5%;\"\u003e\n \u003cp\u003e0.035\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 13.8333%;\"\u003e\n \u003cp\u003e9.8\u0026times;10⁻\u0026sup1;⁰\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u003c/strong\u003e Descriptive statistics of FAZ area in the OphtalmoLaus (OL) cohort and the Rotterdam Study (RS), including sample size (N), mean, standard deviation, minimum, maximum, range, 25th percentile, median, 75th percentile, and interquartile range (IQR).\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"374\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003eArea distribution characterisation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOL\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRS\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003eN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e1288\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e1899\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003eMean\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e260,261.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e315,748.97\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003eStandard deviation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e118,522.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e122,109.15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003eMinimum\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e3,856.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e1,767.20\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003eMaximum\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e1157552.49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e762,370.08\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003eRange\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e1153696.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e760,602.88\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003e25th Percentile\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e181735.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e230,950.95\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e250208.13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e314,362.79\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003e75th Percentile\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e325255.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e396,162.06\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 45.4545%;\"\u003e\n \u003cp\u003eIQR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.0053%;\"\u003e\n \u003cp\u003e143519.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 27.5401%;\"\u003e\n \u003cp\u003e165,211.11\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"This work was supported by the Swiss National Science Foundation (grant no. CRSII5 209510 to Sven Bergmann).","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"GWAS, Microvasculature, Deep Learning, Retina, FAZ","lastPublishedDoi":"10.21203/rs.3.rs-9039865/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9039865/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMicrovascular dysfunction underlies a wide range of cardiovascular, neurodegenerative and metabolic diseases, yet its genetic architecture remains largely unexplored owing to the lack of scalable in vivo phenotyping. Optical coherence tomography angiography (OCTA) enables direct visualization of human capillary networks and provides a unique window into microvascular biology. Here we perform, to our knowledge, the first genome-wide association study of capillary-scale microvasculature directly measured in vivo, using OCTA-derived foveal avascular zone (FAZ) area as a quantitative trait. We analysed two population-based European cohorts with automated deep learning–based image quality control and FAZ segmentation, followed by cohort-specific genome-wide association analyses and fixed-effects meta-analysis. Genomic inflation was minimal across analyses. Meta-analysis identified genome-wide significant loci on chromosomes 17 and 9 associated with FAZ area. These findings establish OCTA-derived FAZ metrics as genetically tractable microvascular phenotypes and provide a foundation for large-scale genetic studies of human microvasculature and its role in systemic disease.\u003c/p\u003e","manuscriptTitle":"Deep Learning Enables Genome-Wide Association Studies of Microvascular Features","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-08 17:16:12","doi":"10.21203/rs.3.rs-9039865/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4709c083-7157-43d8-aeac-90c51264fe31","owner":[],"postedDate":"March 8th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":63989020,"name":"Computational Biology"}],"tags":[],"updatedAt":"2026-03-08T17:16:12+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-08 17:16:12","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9039865","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9039865","identity":"rs-9039865","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}