Ocular and systemic associations and heritability of retinal arterial wall‐to‐lumen ratios in a twin cohort

To investigate ocular and systemic factors associated with the retinal arterial wall‐to‐lumen ratio (WLR) and to determine the relative contribution of genetic and environmental variation to WLR in healthy adults.

Vessel wall characteristics can be inferred, nevertheless, from the displacement of veins at arteriovenous crossings (Christoffersen & Larsen, 1999).Furthermore, a difference in refractive indexes enables neuronal stroma and vessels to be distinguished using fundus photography with adaptive optics (Bakker et al., 2022;Karst et al., 2019).This practice follows a principle developed in larger organs on the basis of ultrasonography, where a higher relative thickness of the vessel wall has been found to be related to end-organ damage, ageing and arterial hypertension.
The present study explored the range, variation and heritability of retinal artery wall and lumen characteristics in a population-based cohort of twins using adaptive optics fundus photography.

| M AT ER I A L S A N D M ET HOD S
Inclusion was between March 2019 and June 2020 in the Copenhagen Twin Cohort Eye Study, a prospective observational study based on a zygosity-and sexbalanced cohort of twin pairs.One-half of the cohort was recruited from the prospective GEMINAKAR cohort of twins without diabetes or cardiovascular disease at inclusion (Benyamin et al., 2007).These subjects had previously participated in a comparable examination in 1999.The other half was recruited from the Danish Twin Registry, and they were matched on zygosity and sex without consideration of glycaemic and cardiovascular characteristics.A preceding statistical power calculation found that 40 monozygotic (MZ) and 34 dizygotic (DZ) twin pairs would be needed to detect a broad-sense heritability of at least 50% (at 80% power, alpha 0.05), assuming a 3:2 ratio of common environmental variance to non-shared environmental variance (Visscher, 2004;Visscher et al., 2008).This ratio was chosen on the assumption that relatively high homogeneity would characterise the study population due to (1) high cultural uniformity in Danish lifestyle and (2) known volunteer bias in twin research, which results in smaller between-pair variance (Lykken et al., 1978), which could have been exacerbated further due to the recruitment of only healthy individuals in the GEMI-NIKAR study (Benyamin et al., 2007).
Zygosity was verified using forensic genetic testing.Oral and written informed consent was obtained from all participants.The protocol was in accordance with the Declaration of Helsinki.
The study was approved by the regional Health Research Ethics Committee (registration number H-18052822) and the Danish Data Protection Agency (registration number VD-2018-434).
Adaptive optics fundus images were recorded using a commercially available adaptive optics floodillumination ophthalmoscope (rtx1-e, Imagine-Eyes).Forty single fundus images were recorded in all right eyes during a 2-s period, and an averaged image was generated automatically.A major superotemporal branch retinal artery devoid of crossings and bifurcations, with clearly detectable borders on both sides of the vessel and a lumen diameter (LD) of at least 50 μm, was targeted at a distance of 1-1.5 optic disc diameters superior to the optic disc.A single grader (CVC) masked from all participant information conducted three measurements of the retinal arterial wall thickness (WT) using semi-automated software (AOimage, Imagine-Eyes).
The axial length (AL) of the eye affects the magnification of the fundus, making biomarkers that rely on absolute values, such as LD and outer diameter (OD), difficult to compare between studies, as many use a standard value for AL.The fact that WLR is a dimensionless ratio and hence not affected by this issue makes it a more reliable biomarker.For the sake of comparability, we did, however, convert all measurements from minutes of arc to μm using the Littmann-Bennett formula (Bennett et al., 1994; given below), which is dependent on AL: To test the intra-rater variability of vessel measurements, 50 randomly selected participants were tested twice on two separate occasions.The mean difference (bias) between first and second measurements was assessed using a paired sample t-test.
Current height and weight were self-reported on the day of the study visit.Body mass index (BMI) was defined as the body weight in kilograms divided by the squared height in metres.Tobacco use was self-reported.Active or previous tobacco users were registered as smokers had they recorded ≥1 pack-year (≥20 cigarettes daily for 365 consecutive days).Blood pressure was measured at the start of the study visit while sitting.Only participants receiving antihypertensive treatment at the time of the study visit were categorised as hypertensive.
Each participant had a blood sample taken for assessment of high density (HDL), low density (LDL) and very low density (VLDL) lipoprotein, total cholesterol and triglycerides.Blood samples were stored in a tube with lithium heparin and analysed using a colorimetric test (VITROS 5600, Ortho Clinical Diagnostics).
Besides adaptive optics imaging, the participants underwent ophthalmic examinations, including measurement of AL using the IOL Master 700 (Carl Zeiss Meditec AG) and measurement of intraocular pressure (IOP) using a rebound tonometer (iCare TA01i, Icare Oy).
The study investigated WLR in relation to age, sex, blood pressure, smoking, BMI, a standard blood lipid profile, AL and zygosity.Univariate quantitative genetic models were used to estimate the relative contribution of genetic and environmental factors to retinal artery wall thickness variations.

| Statistics
RStudio statistical software (RStudio: Integrated Development Environment for R, Inc., Boston, MA) was used for statistical Arterial wall-lumen ratio was calculated as: We measured each artery three times and used the mean for analysis.Since the geometrical effect of WLR is closely related to outer vessel diameter, we adjusted for the outer diameter in all analyses of WLR.
The descriptive comparison of MZ and DZ twins lists variables as means ± standard deviation, unless otherwise specified.The normality of the distributions of differences was tested by histogram inspection and using quantile-quantile plots.The Wald test was used for continuous variables and the Chi-square test for categorical variables.Likelihood ratio testing showed a borderline significant difference (p = 0.0514) between a saturated model (assuming equal inter-twin variance and correlation) and an unsaturated model (assuming unequal intertwin variance and correlation), and hence we used the unsaturated model.
A linear mixed model with fixed effects adjusted for data clustering due to inter-twin dependability was used to analyse associations between WLR and ocular or systemic descriptive variables.WLR was transformed logarithmically to achieve optimal model fitting, variance homogeneity and distribution of residuals.Estimates were calculated with 95% confidence intervals and presented as % change in WLR per 1 (10 for blood pressure) unit change in exposure by subtracting the exponential function of the logarithmic estimate by 1 and multiplying this value by 100.All tests were two-sided, and the level of statistical significance was set at p < 0.05.
The foundation of twin studies stipulates that MZ twins are genetically identical, whereas DZ twins share approximately half of their segregation genes.A descriptive estimate of the genetic and environmental influences on WLR was therefore calculated using the bivariate cotwin correlation in MZ and DZ pairs, respectively.The model (Scheike et al., 2014) assumes greater genetic influence if the within-pair similarity for a phenotype is greater in MZ pairs than in DZ pairs.
A univariate quantitative genetic model (ACDE) was used to decompose the phenotypic variance of the considered parameters into four components: additive genetic effects (A), dominant (non-additive) genetic effects (D), common environmental effects representing the contribution of the shared family environment by both twins (C) and unique environmental effects that apply only to each individual twin (E).
The best-fitting model was selected by comparing the Akaike information criterion (AIC) of the possible models and choosing the one with the lowest AIC score (highlighted in bold in Table 3).

| R E SU LT S
A total of 429 same-sex twin pairs were invited to participate, and 176 twin pairs completed the Copenhagen Twin Cohort Eye Study protocol.Of the 176 pairs, 31 were excluded due to insufficient quality of adaptive optics imaging with unidentifiable artery boundaries in one or both twins.Such pairs were flagged by a primary investigator (CVC), and exclusion was decided together with a retinal specialist and clinical professor (ML).
The analysis thus included 145 twin pairs, of which 78 were MZ and 67 DZ, comprising 158 women and 132 men aged 58.4 ± 9.8 years (range 30-81 years).Arterial hypertension was present in 19% of subjects.The outer diameter of the artery segment of interest was 116.3 ± 13.0 μm and the lumen diameter was 91.2 ± 11.8 μm.The mean wall-lumen ratio was 0.279 ± 0.04 and was comparable for the MZ and DZ groups (Table 1).
Intra-rater variability, defined as the mean difference between the first and second measurements, was 0.35 ± 2.47 μm for outer diameter and 0.35 ± 2.21 μm for the lumen diameter, respectively.The mean difference (bias) for WLR was 0.00093 ± 0.018 with a statistically non-significant difference between the first and second measurements (p = 0.71) and with limits of agreement spanning from −0.0337 to 0.0356.
There was a statistically significant difference between the MZ and DZ groups regarding two parameters: DZ twins were older than the group of MZ twins, and although the number of participants who had registered as smokers did not differ between the groups, members of the DZ group had consumed more pack years than those of the MZ group.Lastly, although borderline significant, OD was not significantly larger in the DZ group compared to the MZ group (Table 1).
In both crude and adjusted analyses, increasing age was associated with a higher WLR, as were increasing mean arterial blood pressure and mean diastolic blood pressure.Systolic blood pressure was associated with a higher WLR in the crude model, but the significance did not persist in the adjusted model.In both crude and adjusted analyses, increasing OD was associated with a lower WLR pr.μm increase in OD (Table 2).
Increasing AL was associated with a lower WLR.There was no statistically significant relationship between WLR and sex, height, weight, BMI, or smoking.Models including age, sex and OD showed no significant association between WLR and HDL, LDL, VLDL, total cholesterol or triglycerides.
A polygenetic model for quantitative genetic analysis of WLR found that the model that explained most variation with the smallest number of variables was an AE model adjusted for age, sex and outer diameter.In this model, additive genetic factors (A; i.e. broad-sense heritability) explained 21% (95% CI: 1-41%; Table 3) of the total phenotypic variance in WLR, with non-shared environmental factors accounting for the remaining 79% (95% CI: 59-99%; Table 3).
As broad-sense heritability was found to be lower than what the study was designed to detect, a reiterative statistical power calculation was conducted using the actual number of included MZ (n = 78) and DZ (n = 67) twins.The analysis found that the actual inclusion allowed a broad-sense heritability of 38% or more to be detected at 80% statistical power (alpha 0.05).Accordingly, statistical power was 35% for the detected broad-sense heritability value of 21%.

WLR = OD − LD LD
This study of the heritability of retinal artery wall thickness in adult twins found a dominating influence of environmental factors, explaining most of the variation in the retinal artery wall-to-lumen ratio (WLR).
The design of this twin cohort study was optimised towards finding higher levels of heritability, and larger sample sizes are required to accurately quantify small genetic effects (Martin et al., 1978;Sham et al., 2020;Verhulst, 2017;Visscher et al., 2008).However, it can reasonably be inferred from the study that the heritability of WLR less than 38%, which supports that WLR is predominantly influenced by environmental factors over heredity.
The aim of the study was not to reliably separate interindividual variance attributable to environmental factors into shared and non-shared constituents, as such studies require significantly more twin pairs (Visscher, 2004;Visscher et al., 2008).Hence, we cannot conclude whether the high degree of environmental influence found on WLR is mostly attributable to factors that can be modified on individual level.However, a previous study found that anti-hypertensive medication reduced the WLR of participants (Evangelou et al., 2018), suggesting that medical blood pressure modulation can modify WLRs.
The mean WLR in the study cohort was 0.28 ± 0.04, a value close to what has been found in comparable study populations (Bakker et al., 2022).We also confirmed the previous finding that the retinal artery wall-to-lumen ratios decrease with increasing outer diameter of the artery (Bakker et al., 2022).
In our study, the relative thickness of the vessel wall increased with age, mean arterial blood pressure and diastolic blood pressure.In a risk factor analysis, only increasing age and higher blood pressure were associated with a higher WLR.These observations add further weight to the existing evidence that ageing and arterial hypertension are the primary factors related to retinal artery wall thickening.This is in agreement with branch retinal vein occlusion being interpreted as an effect of arterial hypertension that is attributable to retinal artery wall thickening, albeit with some effect of random variations in the layout of the retinal vascular network (Bertelsen et al., 2012;Christoffersen & Larsen, 1999).
The proportion of subjects who had arterial hypertension was representative of the prevalence of this condition in the adult population of Denmark (Kronborg et al., 2009).The study analysed blood pressure as a continuous variable and found a 1.29% increase in WLR for each 10 mmHg increase in blood pressure.Some studies operate with an either/or definition of hypertension (De Ciuceis et al., 2017, 2018;Gallo et al., 2016;Koch et al., 2014;Mehta et al., 2019;Meixner & Michelson, 2015), grouping participants with manifest hypertension together with participants who receive antihypertensive treatment.A vast majority of these studies, summarised in a recent systematic review of studies of vascular biomarkers using adaptive optics ophthalmoscopy, found that ageing and blood pressure elevation are major independent determinants of retinal artery wall thickening (Gallo et al., 2016;Koch et al., 2014;Mehta et al., 2019;Meixner & Michelson, 2015).
The twins in this study of adults demonstrated a decrease in WLR with increasing ocular axial length of the eye.This may reflect the stretching of the posterior pole that takes place with axial elongation during childhood and adulthood.Such an effect was not found in the population-based CCC2000 cohort study of 16-17-year-olds, suggesting a modulating effect of ageing (Laigaard et al., 2020).
We did not find any association between WLR and body mass index in our study population.Uncertainty remains regarding a potential relationship, as some studies have found an association (Koch et al., 2014;Laigaard et al., 2020) and others have not (Arichika et al., 2015;Meixner & Michelson, 2015).We also found no association between WLR and the components of the classic lipid profile (HDL-, LDL-, VLDL-, total cholesterol and triglycerides).The absence of a correlation between wallto-lumen ratio and smoking is in agreement with previous studies (Gopinath et al., 2011;Salvetti et al., 2014).
Limitations of the study include the risk of selection bias, which was minimised by recruiting randomly and population-based.Although our sample size was sufficient to conclude that WLR is predominantly influenced by environmental factors, a four-fold higher sample size would have been needed to accurately and more reliably detect the genetic variance effect size found (Visscher et al., 2008).
The study included a low number of participants with diabetes (3%) and hence limited power to discern a role of hyperglycaemia on WLR and a low risk of bias from this variable.Tobacco use and body weight were selfreported, potentially causing information bias, which could have led to an underestimation of their effects.Participants were categorised as hypertensive based on current antihypertensive therapy, possibly underestimating the real proportion of hypertensive subjects.Blood pressure was only measured once, leaving a potential effect of white coat hypertension uncompensated for.The avoidance of bias from white-coat hypertension was also the rationale for not basing the diagnosis of arterial hypertension on current blood pressure.
While the study confirmed previous findings relating to WLR (Bakker et al., 2022), future studies should T A B L E 3 Polygenetic modelling for retinal artery wall-to-lumen ratios.

Genetic components
Environmental components examine a larger number of retinal to enable better assessment of the robustness of findings and the potential for the use of arterial wall-to-lumen ratios in clinical studies.Finally, we had to resort to statistical compensation for the limitation that WLR increases with decreasing artery diameter.
The strengths of our study include the highly accurate quantification and repeatable measurements of WLR.Additionally, our twin study design with a homogenous study population minimised the risk of overestimating heritability, as a near-identical environmental influence among twins during childhood and early adulthood can be assumed.
In summary, this twin cohort study found that thicker retinal artery walls were closely linked to increasing age and higher arterial blood pressure, the latter being mediated by environment over genes.

AC K NO W L E DGE M E N T S
Funding of the study was provided by VELUX FONDEN (grant agreement no.00055617), Rigshospitalet (grant E-23334-02), P. Carl Petersens Fond (grant 19102), Helsefonden (grant 19-B-0063), Aase og Ejnar Danielsens Fond (grant 18-10-0698), Beckett Fonden (grant 19-2-3490), Einar Willumsen Fonden (grant 500028) and Horizon 2020, the European Union's Framework Programme for Research and Innovation, under grant agreement no.780989 (MERLIN).JB was supported by the VELUX FODEN (grant 00028975).SPR was supported by the OPOS Foundation and the Alfred-Vogt Foundation.The funding organisations had no role in the design or conduct of this research.

a estimate: % increase in wall-to- lumen ratio (95% CI) p-value Adjusted b estimate: % increase in wall-to-lumen ratio (95% CI) p-value
Characteristics of study population.Retinal artery wall-to-lumen ratio in relation to ocular and systemic variables.Adjusted for BMI, age, sex, the axial length of the eye and outer diameter of the artery used for analysis.
T A B L E 1 b Estimates are proportions attributable to A, C, D and E factors [95% CI].p-values are based on a likelihood ratio test (ACE-model used as reference).The best fitting model according to AIC, was the AE-model (in bold).Confidence intervals are adjusted for age, sex and the outer diameter of the artery used for measurement.Abbeviations: A, additive genetic effects; AIC, Akaike's information criterion; C, common environment effects; CI95, 95% confidence interval, D, dominant genetic effects; E, unique environmental effects. Note: