Retinal and Cerebral Microvasculopathy: Relationships and Their Genetic Contributions

Abstract

Purpose: Retinal microvasculopathy may reflect small vessel disease in the brain. Here we test the relationships between retinal vascular parameters and small vessel disease, the influence of cardiovascular risk factors on these relationships, and their common genetic background in a monozygotic twin cohort.

Methods: We selected 134 cognitively healthy individuals (67 monozygotic twin pairs) aged ≥60 years from the Netherlands Twin Register for the EMIF-AD PreclinAD study. We measured seven retinal vascular parameters averaged over both eyes using fundus images analyzed with Singapore I Vessel Assessment. Small vessel disease was assessed on MRI by a volumetric measurement of periventricular and deep white matter hyperintensities. We calculated associations between RVPs and WMH, estimated intratwin pair correlations, and performed twin-specific analyses on relationships of interest.

Results: Deep white matter hyperintensities volume was positively associated with retinal tortuosity in veins (P = 0.004) and fractal dimension in arteries (P = 0.001) and veins (P = 0.032), periventricular white matter hyperintensities volume was positively associated with retinal venous width (P = 0.028). Intratwin pair correlations were moderate to high for all small vessel disease/retinal vascular parameter variables (r = 0.49–0.87, P < 0.001). Cross-twin cross-trait analyses showed that retinal venous tortuosity of twin 1 could predict deep white matter hyperintensities volume of the co-twin (r = 0.23, P = 0.030). Within twin-pair differences for retinal venous tortuosity were associated with within twin-pair differences in deep white matter hyperintensities volume (r = 0.39, P = 0.001).

Conclusions: Retinal arterial fractal dimension and venous tortuosity have associations with deep white matter hyperintensities volume. Twin-specific analyses suggest that retinal venous tortuosity and deep white matter hyperintensities volume have a common etiology driven by both shared genetic factors and unique environmental factors, supporting the robustness of this relationship.

The eye, and especially the retina, may reflect vascular pathology elsewhere in the body, and is easily assessable through direct imaging like fundus photography. Subtle changes of the microvasculature of the retina, such as retinal vessel diameter changes or increased tortuosity, can be measured using automated detection software on fundus photographs.¹ 

The retina shares many similarities with the brain, being embryologically derived from the same tissue and possessing similar structural and functional features such as a blood–retina barrier and glial cell connections.² This makes the eye an ideal potential biomarker for vascular and degenerative processes occurring in the brain.^2–6 Many degenerative diseases of the brain also contain a vascular component. Vascular dementia is an obvious example, but also in diseases such as Alzheimer's disease, vascular changes play an important role.^7,8 These pathologic vascular changes of the brain are summarized under the term cerebral small vessel disease, such as white matter hyperintensities on MRI.^9–11 White matter hyperintensities are commonly observed white spots in the cerebral white matter of especially elderly people on MRI and are thought to reflect (micro)vascular damage.¹² Several groups have found relationships between retinal vascular parameters and cerebral small vessel disease, especially in people predisposed to cerebral small vessel disease.^13–18 

Vascular changes in the eye and the brain show a moderate to high level of genetic influence.¹⁹ Heritability (variation explained by the genome) estimates lie between 0.49 and 0.83 for ophthalmologic vascular parameters^20–23 and between 0.64 and 0.77 for small vessel disease.^24–27 By performing a study on monozygotic twins, we aimed to elucidate the genetic and environmental influence on possible relationships between these two vascular systems. For this purpose, two sets of analyses will be performed; a cross-twin cross trait and a twin-difference analysis. A cross-twin cross-trait analysis takes the trait of twin 1 to predict a different trait of the co-twin. If a relationship is found, this suggests that the relationship between said traits is at least partially explained by shared factors, either genetic or environmental, between the twins.²⁸ A twin-difference analysis correlates the within twin-pair differences of a continuous trait to the within twin-pair differences in a second trait. If a relationship between these differences is found, this suggests a unique environmental factor mediating both traits, as differences within monozygotic twins can only be due to environmental exposures that are not shared by twins. If there is a causal relationship between the two traits and the causal trait is partly genetic, and partly environmental, difference scores will also be correlated.²⁹ 

A confounding factor in the relationship between vasculature in the eye and the brain is systemic disease, as systemic vascular disease shows relationships to both retinal vessel parameters as well as brain small vessel disease. For example, hypertension has been shown to be associated with decreased arteriolar diameter,^30–34 and a smaller arteriovenous ratio was found to be related to atherosclerotic findings^35,36 and an increased chance of developing diabetes mellitus.^37,38 Additionally, it is known that cardiovascular risk factors such as hypertension play an important role in the development of cerebral small vessel disease.^39,40 

In this study, we aimed to (1) test the associations between retinal vascular parameters and small vessel disease in the context of cardiovascular risk factors, (2) estimate the upper limit of the genetic contribution to retinal vascular parameters and small vessel disease in a population of healthy, elderly, monozygotic twins, and (3) investigate the underlying mechanism of the associations between retinal vascular parameters and small vessel disease with cross-twin cross-trait and within-pair difference analyses. 

Methods

Participants

We invited 217 participants aged ≥60 years from the Netherlands Twin Register^27,41 who take part in the Amsterdam substudy of the European Medical Information Framework for Alzheimer's Disease (EMIF-AD) PreclinAD cohort. The study followed the Tenets of the Declaration of Helsinki, and written informed consent was obtained from all participants. The study was approved by the Medical Ethics Committee of the VU University Medical Center in Amsterdam. 

Inclusion criteria were as follows: monozygosity, age ≥ 60 years; cognitively healthy as defined by Telephone Interview for Cognitive Status modified (TICS-m) score >22⁴²; Geriatric Depression Scale (GDS) score <11⁴³; Consortium to Establish a Registry for Alzheimer's Disease (CERAD) 10-word list immediate and delayed recall >−1.5 SD of age-adjusted normative data⁴⁴; and Clinical Dementia Rating (CDR) scale of 0 with a score on the memory sub domain of 0.⁴⁵ 

Exclusion criteria were as follows: stroke resulting in physical impairment, neurodegenerative disorders, cancer with terminal life expectancy, uncontrolled diabetes mellitus, and alcohol consumption >35 units (1 unit = 10 mL or 8 g pure alcohol) per week. 

Cardiovascular Risk Factors

Clinical data were collected during a face-to-face interview for medical history, medication intake, smoking habits, and educational attainment. All participants also underwent physical examinations. Blood pressure was measured three times in a lying position with a 5-minute interval between measurements, and the mean of these three measurements was used for the analysis. After a minimum 2-hour fasting period, participants underwent a blood draw to determine lipid profile and glycated hemoglobin in the morning. The cardiovascular risk profile for each participant was summarized using the Framingham Risk Score, which includes the following factors: age, sex, total cholesterol, high-density lipoprotein (HDL), systolic blood pressure, antihypertensive medication use, diabetes mellitus, and smoking.⁴⁶ This risk index represents the 10-year risk of a major cardiovascular event. 

Ophthalmologic Examination

All participants underwent the following ophthalmologic examinations: best-corrected visual acuity, IOP, refraction data, slit-lamp examination, indirect fundoscopy, and fundus photography. Tropicamide 0.5% was used for pupil dilation to enable these examinations. All photographs were assessed by an experienced ophthalmologist (HTN) for unexpected pathology. Participants suffering from ophthalmologic conditions interfering with retinal vasculature or image quality were excluded from analysis (cataract, macular degeneration, glaucoma, diabetic retinopathy, vascular occlusions), as well as ungradable images. 

Fundus Photography and Quantitative Assessment of Retinal Vasculature

Digital fundus images of 50° field of view were obtained from both eyes of most participants (Topcon TRC 50DX type IA; Topcon Medical Systems, Inc., Oakland, CA, USA), centered on the optic nerve head. These images were graded by a trained grader (JAvdK) using the Singapore I Vessel Assessment (SIVA) software (version 3.0; National University of Singapore, Singapore).^47–49 Values from both eyes of every participant were averaged; if only one suitable image was present, only this eye was included. Intrarater scores were calculated in 20 patients. Based on this, we selected seven mostly dimensionless retinal vascular parameters to account for refraction error with an intraobserver intraclass correlation, absolute agreement, of >0.80 for the analyses. These are central retinal artery equivalent, central retinal vein equivalent, arteriole–venular ratio, fractal dimension of the arteriolar network, fractal dimension of the venular network, curvature tortuosity of the arterioles, and curvature tortuosity of the venules. All values for retinal vascular parameters were obtained within zone C (i.e., 0.5 to 2 disc diameters around the optic nerve head). 

Magnetic Resonance Imaging

Whole brain scans were obtained using a 3T scanner using an eight-channel head coil (Philips Ingenuity Time-of-Flight PET/MRI-scanner; Philips Medical Systems, Best, Netherlands). Isotropic structural three-dimensional (3D) T1-weighted images (1.00-mm³ isotropic voxels, repetition time [TR] = 7.9 ms, echo time [TE] = 4.5 ms, flip angle = 8°) and 3D sagittal Fluid Attenuated Inversion Recovery sequences (1.12-mm³ isotropic voxels, TR = 4800 ms, TE = 279 ms, inversion time = 1650 ms) were acquired. An experienced neuroradiologist visually assessed all scans for incidental findings. 

The scans were visually rated by a single experienced rater (MtK), who was blinded to twin pairing. 

Cerebral small vessel disease was assessed by white matter hyperintensities, using 3D sagittal FLAIR on magnetic resonance imaging (MRI). White matter hyperintensities were automatically segmented using a previously described algorithm.^27,50 In short, the algorithm can make a distinction between different types of abnormal image patterns without pathologic a priori knowledge, enabling detection of abnormal intensity clusters, which is particularly useful in segmenting white matter hyperintensities. The white matter hyperintensities map was then dichotomized into periventricular and deep regions using the relative distance between the ventricular surface and the cortical sheet at a ratio of 50%/50%. 

Total intracranial volume measurements were obtained from the application of a label fusion algorithm.⁵¹ 

Statistical Analysis

To normalize the distribution for white matter hyperintensities volume and arterial and venular tortuosity, a log transformation was applied. White matter hyperintensities were further corrected for head size using the total intracranial volume. Analyses with these as dependent variables are thus reported in ratios instead of regression coefficients. A ratio can take any value from 0 onward, where 1 means there is no relationship, <1 means there is a negative relationship (i.e., with an increase in the independent variable, the dependent variable will decrease), and >1 means there is a positive relationship (i.e., with an increase in the independent variable, the dependent variable will also increase). Relationships between retinal vascular parameters, white matter hyperintensities, and cardiovascular risk factors were investigated with generalized estimating equations (GEE) using SPSS (version 22; IBM, Chicago, IL, USA), to correct for clustering in the data from twin pairs, as well as confounders such as age and sex. Monozygotic siblings are very much alike and should thus not be treated as separate individuals, GEE is a type of regression analysis which can take such dependencies into account. Intratwin pair correlations were calculated using Pearson/Spearman correlation coefficients. An intratwin pair correlation is the correlation between the two siblings within a twin pair. One would thus expect this correlation to be high when a trait is mainly determined by shared factors, of either genetic or environmental etiology. Cross-twin cross-trait analyses, which use the trait of twin 1 to predict a different trait of the co-twin, were performed using structural equation modeling implemented in OpenMx running in RStudio (version 1.1.383; RStudio, Inc., Boston, MA, USA). Twin difference analyses were performed by using SPSS. Cross-twin cross-trait and twin difference analyses were performed on significant relationships between retinal vascular parameters and white matter hyperintensities only. Analyses were corrected for age and sex, and standardized values for retinal fractal dimension of the veins and retinal tortuosity of both veins and arteries were used for the cross-twin cross-trait analyses. Standardizing values means making a Z-score out of a parameter (with a mean of 0 and SD of 1.96). We applied a standardization to some parameters for this particular analysis, because the script used for this analysis has trouble in dealing with values of a very different size (in this case comparing very small values to values around 3). 

Results

Of the 217 approached subjects, 15 were excluded due to: epileptic seizures (N = 1), heart problems (N = 1), not meeting neuropsychologic inclusion criteria (N = 3), neurodegenerative disease (N = 2), conditions disabling a hospital study visit (N = 4), transient ischemic attacks (N = 1), or being a sibling to one of the excluded cases (N = 3), leaving a total of 202 participants (100 monozygotic twin pairs and 2 singletons). Of these, 12 were excluded from analyses due to ophthalmologic disease. Of the remaining 190 participants, only 134 participants (67 twin pairs) were complete pairs who had useable data for most included parameters (i.e., sufficient quality fundus pictures and MRI data, complete Framingham Risk Score). This subset was demographically very similar to the whole population. Table 1 shows the demographics of the population in this study. 

Table 1

 

Demographics From Study Population

 

After correction for clustering, age, sex, and total intracranial volume, only increased central retinal vein equivalent was related to a higher volume of periventricular white matter hyperintensities (Table 2). 

Table 2

 

Associations Between Retinal Vascular Parameters and WMH, Given in Ratios Due to a Log Transformation Applied to the Dependent Variables

 

An increased fractal dimension of arteries and veins and increased tortuosity of veins were related to deep white matter hyperintensities volume. Table 2 shows all the associations between retinal vascular parameters and white matter hyperintensities volumes. 

To look for possible mediating cardiovascular risk factors in the above described relationships, cardiovascular risk factors were analyzed in relation to both retinal vascular parameters and small vessel disease (Supplementary Table S1). None of the cardiovascular risk factors were significantly related to both the retinal vascular parameters and small vessel disease in any of the relationships described above. When we repeated the analyses of the associations between retinal vascular parameters and small vessel disease with correction for Framingham risk score, the associations remained similar (Supplementary Table S2). 

The intratwin pair correlations were moderate to high for all small vessel disease/retinal vascular parameters variables (Table 3). 

Table 3

 

Intratwin Pair Correlation Coefficients

 

For the associations found between the retinal vascular parameters and small vessel disease (Table 2), we performed cross-twin cross-trait and twin difference analysis. Cross-twin cross-trait analysis showed a relationship between retinal tortuosity of the veins of twin 1 with deep white matter hyperintensities volume of the co-twin (r = 0.23, P = 0.030; Table 4), whereas other markers showed a trend (P ≈ 0.1; Table 5). 

Table 4

 

Cross-Twin Cross-Trait Correlations for Significant Relationships Between RVPs and SVD

 

Table 5

 

Twin Difference Correlations for Significant Relationships Between RVPs and SVD

 

Within twin-pair differences for retinal venous tortuosity were related to within twin-pair differences in deep white matter hyperintensities volume (r = 0.39, P = 0.001; Table 5). 

Discussion

Increased tortuosity of retinal veins and increased fractal dimension of retinal arteries and veins were related to a higher volume of deep white matter hyperintensities. An increase in the width of retinal veins was associated with a higher volume of periventricular white matter hyperintensities. Retinal venous tortuosity was also associated to deep white matter hyperintensities in cross-twin cross-trait and twin difference analysis, suggesting that these vascular changes have a shared underlying factor related to genetic and nonshared environmental factors. 

We found that retinal venous tortuosity was positively associated with deep white matter hyperintensities volume. In the retina, increased vascular tortuosity is thought to be indicative of vessel wall dysfunction, tissue hypoxia, disturbed blood–retina barrier, and disturbed blood flow.⁵² Similar mechanisms have been described to lead to white matter hyperintensities in the brain,⁵³ suggesting similar processes on a microvascular level are occurring in synchrony in both the eye and the brain. This was supported by cross-twin cross-trait and twin-difference analyses. Cross-twin cross-trait analyses showed that the retinal venous tortuosity of twin 1 was associated with deep white matter hyperintensities volume of the co-twin. This means that the relationship between these traits is partly explained by shared factors between the twins.²⁸ It is likely that in our elderly cohort, age 60 years and older, the association is mainly explained by genetic factors, as discussed below. Twin difference analysis showed that within twin-pair differences for retinal venous tortuosity were related to within twin-pair differences in deep white matter hyperintensities volume. As any difference between monozygotic twins is due to environmental factors unique to each twin; this means that the association is also driven by nonshared environmental factors, possibly supporting a causal relationship between these traits.²⁹ As a direct causal influence of one trait on the other is unlikely in our case, our results suggest there is a common unique environmental factor influencing both retinal venous tortuosity and deep white matter hyperintensities volume, in addition to shared genetic factors. 

An increase in retinal venous tortuosity has been found to be associated with hypertension.⁴⁹ Hypertension also has a relationship to white matter hyperintensities volume,⁵⁴ so it would seem likely that the relationship between retinal venous tortuosity and white matter hyperintensities volume was driven through hypertension. Yet, a diagnosis of hypertension did not show a relationship to venous tortuosity in our study (Supplementary Table S1), and adding Framingham risk score to our model did not change the significance of the relationship found between venous tortuosity and deep white matter hyperintensities volume (Supplementary Table S2). This suggests there is a relationship between the tortuosity of veins in the retina and deep white matter hyperintensities volume regardless of cardiovascular risk factors such as hypertension. 

Our finding that increased fractal dimension in the retina was associated with higher volume of deep white matter hyperintensities was unexpected. Reduced fractal dimension has been associated with stroke, cognitive dysfunction, and hypertension^3,48,55–59 and is thought to represent microvascular damage, due to destruction and collapse of small vessels, thus creating a “simpler' network.”⁶⁰ As white matter hyperintensities also represents vascular damage,⁵³ one would expect that decreased retinal fractal dimension is associated with higher white matter hyperintensities volume, rather than the opposite. An explanation could be a physiologic variation in the fractal dimension of the retinal vasculature. It may be that those who are born with a more complex network of retinal vessels may have a different makeup of vessels in the brain as well, which makes individuals more vulnerable to the development of white matter hyperintensities. 

Cross-twin cross-trait and twin difference analysis did not show any associations, suggesting that the association between fractal dimension and deep white matter hyperintensities does not have a shared etiology. 

We found that increased venous width (central retinal vein equivalent) was associated with increased periventricular white matter hyperintensities volume. Other studies have shown that increased central retinal vein equivalent was related to cardiovascular risk factors such as hypertension, diabetes, and smoking behavior.^31,61,62 It is postulated that widening of the venules may reflect endothelial dysfunction, and it is hypothesized to be a general marker of retinal ischemia and hypoperfusion.⁶² Our results are in line with this; white matter hyperintensities represent (micro)vascular damage in the brain, and an increase in central retinal vein equivalent in the eye could suggest a similar process occurring in the eye. 

Cross-twin cross-trait and twin difference analysis did not show any associations, suggesting that the association between retinal venous width and deep white matter hyperintensities does not have a shared etiology. 

Interestingly, most relationships between the eye and brain in this study were found with deep white matter hyperintensities, rather than periventricular white matter hyperintensities. Periventricular white matter hyperintensities has been shown to have a stronger association with cardiovascular risk factors, as it is supposed to be particularly vulnerable to decreases in blood flow due to it being anatomically located at the arterial border zone.⁶³ This was confirmed by our study; after correction for multiple confounders, only periventricular white matter hyperintensities showed significant relationships with smoking behavior and Framingham risk score (Supplementary Table S1). As both periventricular white matter hyperintensities and microvascular changes of the eye are associated with cardiovascular risk factors, one would expect the relationship between microvascular changes in the eye to be more strongly related to periventricular white matter hyperintensities rather than deep white matter hyperintensities. This suggests that the underlying pathophysiology of retinal vascular parameters is likely to be heterogeneous. 

For small vessel disease, we found high intratwin pair correlations ranging from 0.84 to 0.87. Other studies have found similar contributions of genes to these parameters, with intratwin pair correlations ranging from 0.74 to 0.77 in monozygotic twins, corresponding to estimated heritabilities ranging from 0.64 to 0.77 for white matter hyperintensities lesions.²⁵ 

For retinal vascular parameters, we found intratwin pair correlations ranged from 0.49 to 0.71. This matches what was found by other studies, as estimated intra twinpair correlations of 0.51 to 0.88, with corresponding heritabilities ranging from 0.57 to 0.83, are reported.^20–22,64 

It was a strength of the study that all participants were very well characterized, in both the neurologic and ophthalmologic sense. Extensive screening was performed before inclusion of the participants for analysis. 

Another strength of the study was that we included the role of cardiovascular risk factors in the relationships between the eye and brain. Most other studies on this topic look at only the relationship between eye–brain, eye–systemic, or brain–systemic. In this study, we also looked at the relationship between retinal vascular parameters and small vessel disease when corrected for cardiovascular disease. 

We did not include dizygotic twins and therefore could not estimate the contribution of genetic factors and shared environmental factors to the twin associations. In older twins, however, the estimated contribution of shared environmental effects is lower, as these twins have spent most of their life apart. This is also the case for both white matter hyperintensities volume and retinal vascular parameters, as other studies have shown very low contributions of the shared environment on these parameters.^20,21,25 

We did not correct for multiple testing, as the nature of this study was mostly explorative. However, several of the relationships found in this study (venous tortuosity and deep white matter hyperintensities volume, fractal dimension, and deep white matter hyperintensities volume) were of such significance that they would survive a correction for multiple testing, suggesting a quite robust relationship. 

It is very possible, and even highly likely, that other systemic factors/cardiovascular risk factors we did not include in our analyses are responsible for causing the relations found within this study. The relations found are very unlikely to be of a causative nature (i.e., vascular changes in the eye do not cause changes in white matter hyperintensities volume in the brain). It is thus likely that a common factor causes changes in both these domains, explaining why a relation between the two is found. We have tried to elucidate what factor this could be by looking at some of the more obvious systemic parameters. There are, however, infinitely more (systemic) factors affecting the vascular state of the human body. It is impossible to look at all cardiovascular risk factors known thus far, so we limited our search to the more common ones (mostly the ones also used to compile the Framingham risk score). Additionally, our ultimate goal behind this study is to develop an easy and noninvasive biomarker to gain more insight into the vascular state of the brain. Regardless of what causes the relation between the eye and brain parameters within this study, if these eye parameters can be used to obtain information we would otherwise have to use an MRI for, we have already gained an invaluable biomarker. 

In conclusion, this study showed that there was a moderate to high correlation between monozygotic twins for retinal vascular parameters and cerebral small vessel disease, suggesting a relatively big contribution of genes to these parameters. Furthermore, both fractal dimension of arteries and veins as well as tortuosity of veins in the retina were related to deep white matter hyperintensities, and retinal venous width was related to periventricular white matter hyperintensities. The relationship between venous tortuosity and deep white matter hyperintensities volume was also significant on a cross-twin cross-trait as well as a twin difference analysis, giving us more insight into the nature of this relationship, but also strongly supporting its robustness. This relationship, in particular, may be of interest to explore further, to elucidate whether retinal venous tortuosity may have diagnostic properties in establishing those at risk of small vessel disease, as obtaining information using the eye can be a considerably more patient friendly and cheaper alternative to MRI scanning. 

Acknowledgments

Supported by the EU/EFPIA Innovative Medicines Initiative Joint Undertaking (EMIF Grant 115372). 

Disclosure: J.A. van de Kreeke, None; H.T. Nguyen, None; E. Konijnenberg, None; A. den Braber, None; M. ten Kate, None; C.H. Sudre, None; F. Barkhof, None; D.I. Boomsma, None; H.S. Tan, None; T.D. Verbraak, None; P.J. Visser, None