Association between polygenic risk for schizophrenia and retinal morphology: A cross-sectional analysis of the United Kingdom Biobank

We examined the relationship between genetic risk for schizophrenia (SZ), using polygenic risk scores (PRSs), and retinal morphological alterations. Retinal structural and vascular indices derived from optical coherence tomography (OCT) and color fundus photography (CFP) and PRSs for SZ were analyzed in N = 35,024 individuals from the prospective cohort study, United Kingdom Biobank (UKB). Results indicated that macular ganglion cell-inner plexiform layer (mGC-IPL) thickness was significantly inversely related to PRS for SZ


Introduction
Schizophrenia (SZ) is a severe psychiatric disorder characterized by disturbances in perception, cognition, thought, affect, and behavior (Bleuler, 1911;Strauss et al., 2014;Uhlhaas and Mishara, 2007), with high rates of suicide (Hor and Taylor, 2010) and functional disability (Harvey et al., 2012).Additionally, SZ is associated with elevated rates of cardiovascular and metabolic disease, premature mortality, and substantial economic and healthcare burden (Chong et al., 2016;Correll et al., 2022).To address these issues, biomarkers that indicate risk for SZ, likely response to treatment, and long-term prognosis are needed.Multiple potential biomarkers of SZ have been identified (Kraguljac et al., 2021;Nguyen et al., 2018), but many of these are not feasible for acquisition in everyday clinical practice.For example, neuroimaging biomarkers and cerebrospinal fluid analysis involve invasive, expensive, and/or time-consuming procedures.Retinal biomarkers, on the other hand, can serve as a window to the brain (see below) and can address several of the shortcomings inherent in other CNS measures.Importantly, they can also further our understanding of the etiology and pathophysiology of SZ, as retinal neural tissue loss and microvasculature abnormalities have been widely demonstrated in this disorder (Komatsu et al., 2022;Silverstein et al., 2020).
There are several reasons why studying the retina in individuals with SZ can advance our understanding of the disorder's biological mechanisms and related pathologies.First, the retina is part of the central nervous system (CNS) that grows from the same tissue as the brain during early development and shares many of the same features as the brain (e.g., neurons, glial cells, neurotransmitters and receptor types, blood-brain/retina barriers, layered architecture) (Snyder et al., 2021).As such, retinal morphology can be and often is used as a proxy for CNS functioning.Retinal neural layer thickness is related to level of cognitive function, gray matter volume, white matter volume, integrity of white matter microstructure, hippocampal volume, and cerebral blood flow in healthy adults (Garzone et al., 2023;Mauschitz et al., 2022;Moran et al., 2022;Mutlu et al., 2017).Retinal structural abnormalities are also associated with disease diagnosis, severity, progression, and cognitive decline in neurodegenerative disorders, most notably, multiple sclerosis, Parkinson's disease (PD), and Alzheimer's disease (AD) (Asanad et al., 2020;Ge et al., 2021;Satue et al., 2016) and retinal structural findings can predict cognitive decline over a 3-year period (Ko et al., 2018;Sekimitsu et al., 2023) in general population samples.Second, the retina can be easily examined in vivo using optical coherence tomography (OCT) and its extension, OCT angiography (OCTA), and color fundus photography (CFP).OCT is a highly reliable (intraclass correlation > 0.99) (Wadhwani et al., 2015) non-invasive retinal imaging technique that captures high-resolution images of the retina in a matter of seconds and generates indices of the thickness and volume of retinal structures (e.g., macula, optic disc) and specific retinal layers.It is well-tolerated by patients with serious mental illness (Green et al., 2022), implying that clinical integration of this methodology within psychiatry may be feasible.Third, retinal layer thickness reduction findings are now well-established in SZ (Komatsu et al., 2022;Silverstein et al., 2020;Lizano et al., 2020).For example, compared to healthy controls, individuals with SZ tend to have thinner retinal layers (e.g., macula ganglion cell layer-inner plexiform layer [GC-IPL] and retinal nerve fiber layer [RNFL]), thinner overall maculae, and smaller macular volume (Komatsu et al., 2022;Wagner et al., 2023a).These retinal abnormalities may be indicative of accelerated aging, as a recent study (Blose et al., 2023) revealed that the relationships between age and retinal neural layer thickness and volume in individuals with SZ were significantly stronger when compared to controls, especially for the GC-IPL, which is particularly vulnerable to neurodegeneration (Goldberg & Corredor, 2009) and more strongly associated with cognition than other retinal indices (Moran et al., 2022).In another recent study, retinal structural differences between SZ patients and controls were shown to be more prominent with advanced age (Wagner et al., 2023a).Note that while a significant proportion of these retinal neural changes may be secondary to diabetes and hypertension (Wagner et al., 2023a), which are present at elevated rates in SZ (Liao et al., 2011) and are known to cause retinal damage (Lee et al., 2019;van Dijk et al., 2012), macula findings in SZ have generally remained significant even when controlling for medical disease and have been found in studies that excluded patients with diseases that can affect the retina (Komatsu et al., 2022;Wagner et al., 2023a;Boudriot et al., 2023).Correlations of retinal alterations with brain and cognitive findings in SZ have also begun to appear (Bannai et al., 2020;Liu et al., 2020;Remy et al., 2023;Silverstein et al., 2018), suggesting the potential utility of incorporating OCT and fundus indices into clinical screening and monitoring efforts.
With OCTA and CFP, retinal microvasculature metrics can be computed from the images.These include vessel density, width of the blood vessels, fractal dimension, extent of vessel branching, and tortuosity.Because retinal blood vessels are structurally, physiologically, and embryologically similar to cerebral microvessels, they can provide a window into cerebrovascular and cardiovascular functioning (Kennedy et al., 2023), and retinal and cerebral cortex microcirculation are significantly associated (van Dinther et al., 2022).Retinal microvascular measures can also predict cardiometabolic and cerebrovascular disease (Chua et al., 2021;Wagner et al., 2020), which have high prevalence rates in SZ (Correll et al., 2022;Li et al., 2014).For example, changes in retinal vascular indices have been found to predict systemic diseases such as hypertension, diabetes, coronary artery disease, and stroke (McGeechan et al., 2009a(McGeechan et al., , 2009b;;Wong et al., 2006;Wong and Mitchell, 2004).In addition, abnormal OCTA findings of retinal microvasculature are characteristic of neurodegenerative disorders (Tsokolas et al., 2020), further driving the point that retinal morphological indices are related to brain health.Differences in retinal microvasculature have also been found in individuals with SZ.Specifically, SZ is associated with wider retinal venules and narrower retinal arterioles (Kennedy et al., 2023), as well as abnormalities in retinal perfusion density, overall vessel length, and fractal dimension (all reduced), and corresponding increased foveal avascular zone size (Appaji et al., 2019a(Appaji et al., , 2019b;;Bannai et al., 2022;Silverstein et al., 2021).There is some evidence, however, that most of these vascular changes are due to the presence of metabolic and cardiovascular disease comorbid with SZ, at least in older patients (Wagner et al., 2023a).However, retinal microvascular changes have been observed in first-episode psychosis patients without diabetes or hypertension (Silverstein et al., 2021).
As noted above, while retinal morphological alterations in SZ have been established, the extent to which they reflect neurodevelopmental and neurodegenerative processes inherent to SZ versus the effects of lifestyle, comorbid disease, and/or medications is still unclear.One way to investigate this issue is to examine the association between retinal alterations and genetic risk for SZ using polygenic risk scores (PRSs).PRSs are "single value estimate[s] of an individual's genetic liability to a phenotype" [Choi et al., 2020[Choi et al., , p. 2760] that represent the sum of risk alleles weighted by their corresponding estimated effect sizes derived from genome-wide association studies (GWAS) [Choi et al., 2020[Choi et al., , p. 2760]].They are generated based on the number and strength of genetic risk variants for SZ (Choi et al., 2020).The advantages of using PRSs are that they are relatively simple to calculate, inexpensive, non-invasive, require only a single measurement, and can be used to determine risk for disorders, such as SZ, before their onset, making them clinically useful (Choi et al., 2020;Zheutlin and Ross, 2018).SZ may be an ideal candidate for clinical integration of PRSs (Zheutlin et al., 2019) given its strong genetic basis (with heritability estimates between 64 % and 81 %) (Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014; Sullivan et al., 2003) and because its PRS explains a larger proportion of phenotypic variance for SZ than it does for any other psychiatric illness (7 %) (Schizophrenia Working Group of the Psychiatric Genomics Consortium, 2014).Additionally, genetic risk for SZ is associated with alterations in brain structure (Neilson et al., 2019) and with elevated risk for many different health conditions, including respiratory diseases, digestive diseases, neurological disorders, urinary syndromes, pituitary gland dysfunction, heart failure, and other mental disorders (Pillinger et al. 2023;Veeneman et al., 2022;Zhang et al., 2022;Zheutlin et al., 2019), which is important for risk stratification.To date, no published studies have examined the relationship between polygenic risk for SZ and retinal morphological alterations.However, macular thickness single-nucleotide polymorphisms (SNPs) have been identified that map onto SZ, neurodegeneration, body mass index (BMI), and type-2 diabetes, suggesting that genetics may contribute to overlap of these features in people with SZ (Gao et al., 2019).In addition, a recent study (Sekimitsu et al., 2023) examining polygenic risk for AD and PD and their relationship with retinal indices provided insight into the pathogenesis of neurodegenerative disorders and, importantly, demonstrated the potential of PRSs and retinal imaging to be used as screening tools/markers for cognitive decline, signs of early neurodegenerative disease, and risk stratification, generating the possibility for this methodology to be used in psychiatry for SZ.All of this evidence is consistent with accumulating evidence for a shared genetic basis between brain and retinal structure (Zhao et al., 2023).Because retinal and brain changes in SZ are thought to reflect a neurodegenerative process and the effect sizes of retinal findings in SZ (e.g., Wagner et al., 2023a) are similar to those observed in the neurodegenerative disease literature (e. g., Chan et al., 2019;Cheung et al., 2015), significant associations between polygenic risk for SZ and retinal morphology alterations are plausible.
The aim of the current study was to examine the relationship between polygenic risk for SZ and individual differences in retinal structure and microvasculature in a large general population cohort (the United Kingdom Biobank [UKB]) (Chua et al., 2019).We hypothesized that PRS for SZ would be associated with inner retinal morphological alterations, such that greater risk for SZ would predict thinner RNFL, mGC-IPL, and inner nuclear layer (INL) and larger optic cup-to-disc ratio.INL was examined because it is the most active site for dopamine in the retina through A2 amacrine cells (Roy and Field, 2019;Witkovsky et al., 2004), and dysfunctional presynaptic dopaminergic activity is thought to be a significant factor in the pathology of SZ (Howes et al., 2012).Additionally, INL thinning (among other structural changes) has been observed in Parkinson's disease (Alves et al., 2023;Wagner et al., 2023b), which is known to involve reduced dopaminergic activity (Emamzadeh and Surguchov, 2018).Second, we hypothesized that PRS for SZ would be associated with alterations in retinal microvascular indices, as evidenced by reduced retinal microvascular caliber, increased retinal microvascular tortuosity, and increased retinal microvascular fractal dimension.Lastly, we explored whether the relationship between retinal layer thinning and age would be more pronounced in those with higher PRS for SZ.

Design, participants, and setting
This cross-sectional analysis used data from the UKB, a large-scale biomedical database and research resource of over 500,000 participants recruited between 2006 and 2010.Participants aged 37-73, registered with the UK National Health Service (NHS), and living within 25 miles of one of 22 assessment centers were invited to participate.Participants gave informed consent to undergo verbal interviews and complete touchscreen questionnaires, physical measurements, and biological investigations.Further information can be found in the study protocols (http://www.ukbiobank.ac.uk/resources/) (Bycroft et al., 2018;Sudlow et al., 2015).As part of an enhanced ophthalmic assessment, a subset of 67,321 participants received retinal imaging with color fundus photography (CFP) and optical coherence tomography (OCT) at the baseline visit (Chua et al., 2019).For this report, we included participants with European genetic ancestry, based on principal components analysis (to mitigate population stratification) and complete schizophrenia PRS data who had undergone retinal imaging.We did not exclude individuals with schizophrenia in the analyses examining the relationships between PRS and retinal morphology indices.Our primary pre-planned analysis was to investigate the association between schizophrenia PRS and thickness of the RNFL and mGC-IPL, the retinal structures for which there is the greatest amount of evidence of SZ-related decline in the published literature.
UKB was conducted in accordance with the principles of the Declaration of Helsinki; ethical approval was obtained from the North-West Research Ethics Committee (reference: 06/MRE08/65).This analysis was conducted under UKB application ID: 36,741.

Polygenic risk score
We used the standard PRS set for schizophrenia constructed by (Thompson et al., 2022) and provided through the UKB showcase (Field ID: 26,275).In brief, this PRS was generated using three external genome-wide association study (GWAS) data on 98,456 individuals with schizophrenia and 334,331 controls.Raw PRS values were centered and variance-standardized as per methods described previously (Thompson et al., 2022).The UKB was not included in the standard PRS construction.Description of the GWAS data, details of the meta-analysis, and PRS calculation are described in (Thompson et al., 2022).
To investigate associations between the schizophrenia PRS and secondary exposure variables, we used the entire UKB cohort of European genetic ancestry (n = 407,943).We first examined the relationship between the PRS and schizophrenia diagnosis, the latter defined through both verbal interview and hospital admissions data (coded through a preceding International Classification of Diseases 10th revision, schizophrenia code: F20) (Davis et al., 2018).We also investigated the association between the PRS and presumed confounders of the relationship between schizophrenia and retinal morphologyage, sex, socioeconomic deprivation, alcohol consumption, smoking status, hypertension, diabetes mellitus, and BMI.Socioeconomic deprivation was defined using quartiles of the Townsend deprivation index, an area-based measure of material deprivation derived from four themesunemployment, non-car ownership, non-home ownership, and household overcrowding (Townsend et al., 1988).Alcohol consumption, smoking status, hypertension, and diabetes mellitus were defined through self-report touchscreen-administered questionnaires.BMI was calculated by dividing weight (kg) by height (m 2 ), measured at a UKB assessment center.

Retinal imaging
The outcome measures were retinal morphological features derived from imaging.Non-mydriatic macula-centered CFP and OCT were acquired from UKB participants using the Topcon 3D-OCT 1000 Mark II (Fig. 1, Topcon Corporation, Tokyo, Japan).OCT images were acquired over a 6.0 mm × 6.0 mm 2 field and had 128 horizontal B scans and 512 A scans per B scan (Patel et al., 2016).All imaging was conducted at the baseline visit.
CFPs were processed using the open-source deep learning algorithm, AutoMorph (Zhou et al., 2022).AutoMorph crops the CFP, assesses image quality, segments the optic nerve and retinal vessels, and provides a tabular export of quantitative clinically interpretable indices of retinal anatomy.This includes retinal vascular characteristics, such as tortuosity, caliber, and fractal dimension, as well as optic nerve morphology, such as optic cup and disc height.Optic cup-disc ratio (CDR) was calculated as the ratio between cup and disc sizes with an average for both horizontal and vertical dimensions.
OCT images were segmented for sublayer thicknesses of the macular retinal nerve fiber layer (mRNFL), ganglion cell-inner plexiform layer (GC-IPL), and inner nuclear layer (INL) using the Topcon Advanced Boundary Segmentation Tool (TABS), a software providing automated segmentation of retinal sublayers using dual-scale gradients (Keane et al., 2016).We averaged thicknesses from the four parafoveal subfields, as defined by the Early Treatment Diabetic Retinopathy Study Research Group (1985).Sublayer boundaries were defined according to the lexicon proposed by the International Nomenclature for OCT panel (Staurenghi et al., 2014).
For image quality control, we used image quality metrics from both AutoMorph and TABS.AutoMorph outputs a categorical score of 'good', 'ok' and 'poor'; we only included images labelled as 'good' or 'ok'.TABS provides additional metadata for each image to establish scan quality based on segmentation error, movement artefact, and poor quality.For consistency with other reports, we excluded the poorest 20 % of images based on these specific image quality control metadata.

Statistical analysis
Distribution of data was visualized using quantile-quantile plots and assessed statistically.Continuous variables were summarized using mean ± standard deviation and categorical variables through percentages.We used linear regression to estimate the association between the schizophrenia PRS and our secondary exposure variables.To investigate the association between schizophrenia PRS (primary exposure) and retinal morphological features (outcome), we fitted linear mixed effects regression models using maximum likelihood estimation with a random effect on the intercept to account for the multilevel data structure of eyes nested within participants (Cruz-Herranz et al., 2016).We adjusted for any confounders defined a priori that had a significant bivariate B.A. Blose et al. association with the schizophrenia PRS (defined as p < 0.05).We investigated schizophrenia PRS as both a continuous variable and in quintile groups, suspecting that some associations may only be present for those most at risk of schizophrenia.We additionally stratified the cohort into ten-year age groups to investigate the effect of age on the association between schizophrenia PRS and retinal morphology, to enhance interpretation and examine any differences across age groups.Degrees of freedom for multilevel modelling were estimated using Satterthwaite's approximation (Satterthwaite, 1946).Retinovascular indices were standardized (mean-centered, divided by standard deviation) for adjusted analyses (raw values are less interpretable for these features).All analyses were conducted in R version 4.1.0(R Core Team, 2016) and used the lme4 and lmertest packages.

Results
From an original cohort of 67,321, there were 35,024 participants with PRS who had sufficient quality retinal image data to include in the analysis (Fig. 2).Of these participants, n = 40 (0.1 %) had a SZ diagnosis.Baseline characteristics of the imaging cohort in the UKB and demographics are displayed in Table 1.Table 2 shows the association between SZ PRS (hereafter just indicated as PRS) and the secondary exposure variables across all eligible UKB participants who had PRS assessment.Higher PRS was significantly associated with SZ diagnosis, female sex, and greater Townsend deprivation index (3rd and 4th quartiles).Additionally, higher PRS was significantly associated with a lower likelihood of hypertension and diabetes, higher cholesterol, and lower BMI.Further, higher PRS corresponded to a greater likelihood of previous and current smoking and alcohol use.No significant relationships were observed between PRS and age or refractive error.
The relationships between PRS and the OCT variables are displayed in Table 3.Both unadjusted and adjusted analyses revealed no significant associations between PRS and mRNFL or cup-to-disk ratio.There was an inconsistent association between PRS and INL, with a significant relationship between the second and third quintiles of PRS and INL thickness but no significant relationship for the fourth or fifth quintile in the unadjusted analyses (and only significant for the third quintile in adjusted analyses).In both the unadjusted and adjusted analyses, greater PRS was significantly associated with thinner mGC-IPL for each PRS quintile, demonstrating a dose-response pattern with each higher quintile showing an incrementally greater effect estimate.The relationship with mGC-IPL thickness was also significant when PRS was used as a continuous variable.
Table 4 shows the association between PRS and mGC-IPL stratified by ten-year age groups.For both the 40-49 and 50-59 age groups, greater PRS (4th and 5th PRS quintiles for the 40-49 group and 5th quintile for the 50-59 group) significantly predicted thinner mGC-IPL.However, no significant relationship was observed between the PRS  quintiles and mGC-IPL thickness in the 60-69 age group.These findings indicate that higher genetic risk for SZ was related to thinner mGC-IPL in the younger, but not older, age groups.
Lastly, Table 5 shows the relationships between PRS and the retinovascular indices.Both unadjusted and adjusted analyses revealed no significant associations between any of the arteriolar or venular indices and PRS with the exception of venular tortuosity.In both the adjusted and unadjusted analyses, PRS had a negative association with venular tortuosity for the 2nd, 4th, and 5th PRS quintile, suggesting that greater polygenic risk for schizophrenia was associated with reduced venular tortuosity.

Discussion
We examined the association between SZ PRS and retinal morphological phenotypes in large cohort totaling 35,024 participants from the UKB.Several notable findings emerged: (1) Greater PRS was associated with thinner mGC-IPL, a pattern that persisted after adjusting for sex, Townsend deprivation index, hypertension, diabetes, alcohol use, smoking, cholesterol levels, and BMI, while no significant relationships were found for the other OCT metrics; (2) An association between greater PRS and thinner mGC-IPL was observed in the younger age groups (40-49 and 50-59 years) but not in oldest age group (60-69 years); and (3) retinal microvasculature indices were largely unrelated to PRS, with the exception of retinal venular tortuosity.Each of these will be discussed in turn below.
The finding that greater genetic risk for SZ was associated with thinner mGC-IPL even after adjusting for covariates is further evidence that retinal thinning in SZ may reflect a neurodegenerative process that is not solely a result of lifestyle and other secondary consequences of the disease and builds upon several studies finding reduced mGC-IPL thickness in patients with SZ while controlling for many of these variables (e.g., diabetes, hypertension) (Boudriot et al., 2023;Wagner et al., 2023a).Additionally, the mGC-IPL findings suggest that the multiple genetic factors involved in SZ have neurodegenerative effects even when observed at moderate levels of aggregation and that, beyond this level, increased genetic risk is not associated with significantly greater retinal atrophy.In contrast, an association with greater polygenic risk was not found for RNFL thickness, INL thickness, or cup-to-disc ratio.This is consistent with reviews and meta-analyses indicating that macular retinal findings are the most robust retinal findings in SZ (Komatsu et al., 2022;Silverstein et al., 2020), a pattern also evident in neurodegenerative disorders (Britze et al., 2017;Cheung et al., 2015;González-López et al., 2014).This may be the case because the macula contains about half of all the retinal ganglion cells, and these are particularly sensitive to insults (Corredor and Goldberg, 2009), and the mGC-IPL contains the cell bodies and dendrites of those neurons (Curcio and Allen, 1990).Thus, the macular region appears to be more sensitive to retinal pathology, and, as such, thinning in the mGC-IPL tends to precede RNFL thinning in neurodegenerative disease (Liao et al., 2021;Komatsu et al., 2022).It has also been proposed that astrogliosis, which primarily occurs in the RNFL, may mask RNFL thinning in neurodegenerative disease measured via OCT and, therefore, RNFL measurements may not be as sensitive as mGC-IPL measurements in tracking neurodegeneration (Green et al., 2010;Saidha et al., 2011).
Interestingly, we observed differences across age groups, such that PRS was associated with mGC-IPL thinning in the younger age groups (40-49 and 50-59 years) but not the oldest age group (60-69 years).We suspect this to be the case because there may have been additional factors significantly affecting retinal neural tissue integrity in older subjects, thereby reducing the impact of PRS.While such factors still  need to be determined in future studies, they likely include effects of normal aging on the retina, undiagnosed early eye disease (e.g., glaucoma, age-related macular degeneration), which is associated with ganglion cell loss even before it can be detected clinically (Kerrigan-Baumrind et al., 2000;Makous, 2004), effects of other medical diseases (especially those that involve neuroinflammation), and/or effects of sedentary behavior (Loprinzi et al., 2015).The lack of a consistent association between PRS and the retinal vascular variables was unexpected, as retinal neural and vascular findings are significantly related (Silverstein et al., 2021;Wei et al., 2017).This was also surprising given an influential theory of SZ that posits a primary neuroinflammatory process, followed by vascular changes, eventually leading to neural changes (Hanson and Gottesman, 2005), and there is preliminary evidence from investigations with first-episode psychosis patients that retinovascular changes may occur prior to retinal neural changes (Green et al., 2022;Lai et al., 2020;Silverstein et al., 2021).In addition, these results contradict a prior study on retinal microvasculature in participants from the Brisbane Longitudinal Twin Study (Gillespie et al., 2013;Wright and Martin, 2004), which found significantly wider retinal venules in unaffected co-twins of probands with psychosis symptoms (Meier et al., 2015).It is possible that vascular changes are less related to the genetic variables that currently comprise the PRS than are neural thickness variables and/or that a more positive relationship could emerge as additional genetic variables related to schizophrenia are identified.Another possibility is that vascular changes in SZ, along with other aspects of retinal structure that were not related to PRS (e.g., optic disc parameters such as RNFL thickness and cup-to-disc ratio), may be more related to the consequences of the disorder known to affect retinal health (e.g., hypertension, diabetes, smoking, alcohol use, etc.) than to genetic factors.This may also explain why a significant association with polygenic risk for SZ was only found for retinal venular tortuosity and none of the other retinovascular indices.Retinal vascular tortuosity has been shown to be a more stable index, as it is not affected by pulse variations (Hao et al., 2012;Kalitzeos et al., 2013) and is less influenced by systemic cardiovascular risk factors relative to retinal vascular caliber (Appaji et al., 2019b;Kirin et al., 2017;Taarnhøj et al., 2008).A further explanation for the lack of consistent findings with the retinal microvascular indices is that a PRS is not deterministic, and even those in the highest PRS quintile are unlikely to have SZ.We quantified a genetic predisposition to SZ rather than schizophrenia itself and are assuming a linear relationship across the spectrum of genetic risk.This is very different from examining associations with a well-defined binary clinical phenotype.

Limitations
These findings should be interpreted in light of several limitations.First, it is known that the UKB is not representative of the general Table 3 Associations between schizophrenia polygenic risk score (primary exposure) and inner retinal morphology and optic cup-disc ratio (outcomes). 2 For cup-to-disc ratio analysis, there was data available for 24,364 participants.

Table 4
Association of polygenic risk of schizophrenia and macular ganglion cell-inner plexiform layer thickness stratified by ten-year age groups.population in the UK and is affected by healthy volunteer bias (Tyrrell et al., 2021), limiting the external validity of the findings.In addition, the analyses were restricted to individuals of European ancestry, so we cannot generalize the results to those of non-European ancestry.Therefore, future studies using more diverse samples of other ancestries are warranted.Third, the cross-sectional design of the study limits the ability to make conclusions regarding the relationship between genetic risk for SZ and the rate of retinal thinning over time.Another limitation is that the SZ PRS was associated with several other variables, including BMI, smoking status, cholesterol, etc., and so inferring any direct causality (from PRS to retinal findings, in this case) as in a Mendelian randomization is limited.In addition, we were limited to testing the PRS as a single instrument variable, as the component SNPs underlying the PRS are not in the public domain.If the component SNPs were known, it would have been possible to use each SNP as an instrument variable to carry out further Mendelian randomization experiments and sensitivity analyses to explore potential pleiotropy.Lastly, the amount of phenotypic variance of SZ explained by PRS scores is small, so small effects may have gone undetected (Pillinger et al., 2023;Trubetskoy et al., 2022).

Conclusion and future directions
In conclusion, we found evidence that mGC-IPL thinning is related to the genetic basis of SZ.By contrast, this study did not find strong evidence of associations between genetic liability for SZ and retinal microvasculature and other retinal neural metrics (mRNFL, INL, and cup-to-disc ratio), suggesting that factors other than genetic variants associated with SZ contribute to these retinal phenotypes.Our results build upon previous studies of retinal morphological differences in individuals with SZ by demonstrating that some of these alterations, especially macular findings, may reflect a neurodegenerative process that is inherent to the disorder, while others, particularly in retinal microvasculature, may be secondary consequences.Future studies are needed to clarify the clinical characteristics of SZ patients who have stronger vs weaker relationships between SZ PRS and macular findings, as well as the meaning of macula thinning for outcomes such as transition to a psychotic disorder, good vs. poor treatment response, structural and functional changes in the brain, and long-term cognitive and functional outcomes.Lastly, more research is needed to clarify the extent to which neurodevelopmental factors lead to retinal structural features, such as thinner macula in children and adolescents prior to a first psychotic episode.Given the overlap between neurodevelopmental and neurodegenerative processes (Schor et al., 2021), it is also possible that this distinction will be superseded by a more unified, life-course approach.Nevertheless, there is a need to better understand the retinal characteristics of people who eventually develop a schizophrenia spectrum disorder, as well as the extent to which such findings are predictive of brain structural and functional changes and functional decline.

Fig. 1 .
Fig. 1.Left panel: An example color retinal photograph with overlaying outputs from AutoMorph segmentation of the optic nerve cup, optic nerve disc and retinal vasculature.Right panel: A macular optical coherence tomography slice through the fovea with the location of the retinal nerve fiber layer, ganglion cell-inner plexiform layer and inner nuclear layer highlighted.GC-IPL: ganglion cell-inner plexiform layer, INL: inner nuclear layer, RNFL: retinal nerve fiber layer.
This work was supported by a New York Fund for Innovation in Research and Scientific Talent (NYFIRST) from the Empire State Development Fund.

Table 1
Baseline table for all eligible participants.

Table 2
Association between schizophrenia PRS and secondary exposure variables for entire cohort (N = 407,943).
Model adjusted for sex, Townsend deprivation index, hypertension, diabetes, alcohol, smoking, cholesterol and body mass index.

Table 5
Associations between schizophrenia polygenic risk score and retinovascular indices.