Many studies have explored, confirmed and quantified causal relationships between potential exposures and the risk of eye disease. To investigate the application of MR to eye disease, we systemically searched PubMed, EMBASE and Web of Science databases for studies using the search keywords, “Mendelian randomization” with any outcomes related to “eye OR ocular OR ophthalmology”. Database were searched from inception to 1st May 2021.
Age-related cataracts are defined by opacity of the lens due to lost optical clarity concomitant with ageing, and are common causes of vision loss worldwide [
16]. Energy and nutrient intake are involved in cataract formation and progression [
17]. Several epidemiological studies have also identified a link between obesity and cataracts [
18,
19]. An Asian population study used MR analysis to identify causal associations between fat mass, the obesity-associated gene (
FTO) and various types of senile cataracts [
20]. Accordingly, no causal associations were identified between obesity and cortical or posterior subcapsular (PSC) cataracts, whereas
FTO was putatively involved in nuclear cataract pathogenesis. However, this study did not test the three key aforementioned MR assumptions. It is noteworthy that genetic variations at the
FTO locus are associated with lipid metabolism pathways. Thus, further studies are warranted to investigate if lipid metabolism has a potential role in cataract pathogenesis. In another study, a two-stage MR approach was used to process Blue Mountains Eye Study data, and identified the single nucleotide polymorphism (SNP), rs9939609 as strongly related to exposure (obesity); the SNP was used as an IV to analyse the causal relationship between obesity and age-related cataracts [
21]. Furthermore, different regression models were applied to verify the MR hypothesis (assumption i and ii). Interactions between the
FTO SNP and protein intake were shown to increase the risk of PSC cataracts in individuals with dyslipidemia. However, these studies reflected a common issue with MR: no method has yet been identified to verify the existence of horizontal pleiotropy.
Globally, AMD is a leading cause of blindness in the elderly. Several studies have proposed that particular genes are involved in AMD pathogenesis and may participate in inflammatory and complement pathways [
22‐
24]. The Y402H SNP in the complement factor H gene,
CFH is associated with AMD [
25‐
27]. Increased
CFH expression was observed in vitro in cultured (dedifferentiated) human retinal pigment epithelium (RPE) cells by immunohistochemistry [
27].
CFH variations may lead to the improper control of the local complement system which triggers drusen formation [
25]. Based on this evidence, a North Indian population-based study used MR on
CFH rs1061170 as an IV to demonstrate the polymorphism affected AMD occurrence or progression by modulating serum CFH levels [
28]. From their findings,
CFH (rs1061170) C allele carriers had an increased risk of AMD due to decreasing serum CFH levels. Proteomic analyses subsequently confirmed that some major proteins (e.g., C-reactive protein, (CRP)) occur in drusen as hallmark lipid and proteinaceous deposits [
29]. Furthermore, to assess the causal relationship between circulating CRP levels and AMD, Han et al. selected SNPs associated with serum CRP levels as IVs for two-sample MR analysis [
30]. They performed IVW analysis, then multivariable MR analysis which regressed several potential AMD factors based on univariate analysis. Combined, these authors showed that inflammatory pathways were involved in AMD pathogenesis, and that CRP was causally related to AMD. Unfortunately, this study did not use the non-linear IV hypothesis to test non-linear relationships between circulating CRP levels and AMD. Thus, further investigations are required to manage non-linearity in similar MR analyses.
In clinical epidemiology, AMD is commonly classified into early, intermediate and late stages. Late-AMD stages often lead to visual impairment and even irreversible loss of central vision [
31,
32], and its treatment modality is different to other AMD stages. In early and intermediate AMD stages, pigmentary abnormalities and extracellular aggregate (drusen) accumulation occurs in the macular region [
24,
33]. Moreover, lipid is a structural component of drusen [
34]. Several observational studies explored relationships between lipid profiles and AMD risk and concluded that high levels of high-density lipoprotein cholesterol (HDL-C) increased disease risk [
35‐
38]. However, owing to different study designs and sample sizes, potential confounding factors and reverse causality cannot be excluded, thus conclusions remain largely inconsistent [
39,
40].
Importantly, GWAS have been instrumental in identifying associations between AMD and some genetic variants, such as hepatic lipase (
LIPC) [
41], cholesterol ester transfer protein (
CETP) [
42] and apolipoprotein E (
APOE) [
43]. Subsequently, several studies have exploited these genes to explore causal effects in different lipid pathways in AMD [
44‐
46]. A previous two-sample MR study conducted in European ancestry individuals from the International Age-Related Macular Degeneration Genomics Consortium, involving 185 lipid-related genetic variants as IVs, identified a causal relationship between HDL-C and AMD, with an odds ratio (OR) of 1.22 [
47]. In fact, some of the lipid deposition associated with AMD comes from the induction of the retina. Thus far, no validated IVs have been identified for retina lipids. Additionally, a multi-ethnic genetic study using three different MR models showed that HDL-C was a causal factor for advanced AMD [
48]. Both studies used multivariable MR analyses which advantageously adjusted pleiotropy caused by incomplete independence between genetic variations and tested exposure variables.
Due to the huge variety of lipids and different AMD stages, specific research on relationships between lipid subcomponents and early and intermediate AMD stages is lacking. Recently, a large-scale study using a two-sample MR approach investigated the causal effects of eight major serum lipid biomarkers toward different AMD subtypes [
6].Using a multivariable MR method based on Bayesian model averaging applied to exposure factors, the study identified lipid metabolism (apolipoprotein A1 and apolipoprotein B) functions in drusen formation, in particular at early and intermediate AMD development stages. MR approach based on Bayesian model averaging (MR-BMA) is a new modality for multivariate MR that incorporates more lipid particles using high-throughput experiments. Verena Zuber et al. used this method to prioritize risk factors by marginal inclusion probability (MIP), and then repeated analysis by modifying a posteriori probability, finally verified the strong causal effect of HDL on AMD, and also found the causal relationship between large and extra-large HDL particles and AMD [
49]. The verification idea of MR-MBA is also superior to the IVW method of biased estimation of risk factors. It believes that the risk factors that really play a causal role themselves are a minority of a large number of risk factors, and more risk causal factors can be found through screening and sorting.
Apart from lipid profiles, a previous systematic review summarised several other potential risk factors for AMD [
50] and showed inconsistent associations between refractive error (RE) and disease [
51,
52]. Wood et al. conducted a two-sample MR study with IVW, and selected 126 genetic variants associated with RE as IVs to verify if RE was a risk factor for AMD in individuals of European descent [
53]. Their analysis showed that hyperopia had a minimally positive causality with AMD, with an OR = 1.08. In addition, their sensitivity analyses, after excluding 31 variants closely related to non-RE traits, generated similar results (OR = 1.069). These inconsistent associations between RE and AMD may be due to non-causal factors such as parametric variations, stochastic fluctuations, potential confounders or selection bias.
RE is associated with many eye abnormalities and is one of the most common causes of visual dysfunction [
54] For example, the pathological consequences of myopia, including exudative myopic macular degeneration and rhegmatogenous retinal detachment (RRD), are common causes of irreversible blindness [
55]. Conventionally, observational epidemiological studies have postulated associations between some lifestyle exposures and myopia. For instance, education level was commonly regarded as a social factor leading to myopia onset, however, it is unclear if the relationship is causal [
56]. Therefore, Cuellar-Partida et al
. used GWAS education polygenic risk scores/data from the Social Science Genetic Association Consortium as IVs, and used a two-stage least squares approach to explore causal relationships [
57]. Their analyses suggested a causal relationship existed between education and myopia, with estimated effect sizes higher than observational study estimates. Above findings have advantages in terms of strengthening due to the mediating variables that may confuse causality mentioned in the previous article were also excluded in the process of MR analysis [
58,
59]. However, the other side of this causal relationship may be overestimated due to the existence of other education-associated markers (e.g., time spent on study). Therefore, more studies are required to investigate causal relationships between education and myopia, adjusted for more potential confounders.
To fully understand causal links between education and myopia, a bidirectional MR study based on the UK Biobank used a specific set of seven different assumptions and suggested exposure to more years in education increased myopia risk [
60]. Two analyses in bidirectional MR, only one proved the causal relationship between the two variables: the longer the education, the more likely to develop myopia. This study is significant for strengthening the myopia prevention because it replaced the traditional regression analysis to detect potential pleiotropy.
Despite evidence from observational studies showing associations between myopia and vitamin D levels, the causal evidence for vitamin D associations with myopia is lacking, and are independent of outdoors activity times [
61] or other related factors (e.g., time spent at work or exposure to natural light) [
62,
63]. In view of this, Cuellar-Partida et al. selected effect estimates from a large RE meta-analysis from the Consortium for Refractive Error And Myopia (CREAM) and three other young cohorts to calculate causal estimates between vitamin D levels and myopia using Wald-type ratio estimators in an MR framework [
64]. Valid and significant IVs identified by this study included 25-hydroxyvitamin D [25(OH)D] SNPs, as also determined by Afzal et al. [
65] and Mokry et al. [
66]. The meta-analysis identified no genetic causal links between 25(OH)D levels and myopia and further suggested associations were mediated by potential factors, including outdoor exercise times. Importantly, this analysis reflected a major MR benefit, i.e., unlike the temporary influence of exposure factors reflected by measured values in observational studies, MR IVs reflect the long-term exposure-induced state of the research factors.
In ophthalmologic anatomy, axial length (AL) and corneal radius of curvature (CR) are the two main physiological indicators of emmetropia. Previous studies reported that changes in children's growth trajectory could predict RE [
67] and that correlations existed between birth parameters and ocular biometrics [
68]. Tideman et al. conducted an MR analysis using weighted GRS as IVs and linear regression analysis to identify associations between prenatal and postnatal growth variables and AL and CR[
69]. These authors showed that body growth parameters were significant predictors for a higher AL and corneal curvature (CC) by the age of six. The same genetic pathways between birth parameters and ocular biological indices were related to the study of causal associations between foetal growth, infancy, early childhood and REs.
Additionally, a low birth weight (LBW) is considered a common cause of visual impairment, and its relationship with RE has attracted considerable attention [
70]. Some observational studies reported that LBW increased the risk of RE in school-age children (7–12 years) [
71,
72], however, it was unclear if individuals with LBW within the normal birth weight range experienced RE in adulthood. Therefore, a one-sample MR analysis with GRS’s as IVs and a linear regression analysis was conducted, and showed that LBW within the normal range was causally associated with a more myopic RE, while the impact of causal effects was modest in UK Biobank participants [
73]. In fact, race/ethnicity are factors associated with myopia; the myopia prevalence in East and South-East Asian is higher than European populations [
55,
56]. Thus, future research must identify causal relationships between LBW and refractive changes, especially in Asian populations.
Currently, it is difficult to distinguish the effects of different environmental factors on RE, which are primarily due to heterogeneous exposure in the population. A causal role for particular medications in promoting RE was described long before MR research commenced [
74]. For example, dorzolamide is a sulfonamide drug and was previously linked to transient myopia episodes [
75,
76], whereas the same conclusion was drawn from a recent MR study by Patasova et al. [
77]. Due to inevitable correlations between some medications and primary disease, Patasova et al. used SNPs associated with primary diseases as IVs and their effects over spherical equivalent (SE) for the outcome and suggested a causal effect of multisite chronic pain over hyperopic RE 74. One of the underlying causes of hyperalgesia is the lower threshold of sensory neuron firing, which may be related to genetic variations affecting proton-gated cation channels [
78]. Likewise, RE is related to genetic polymorphisms in deionised cation channels [
79]. Moreover, chronic pain may activate autonomic responses leading to pupillary dilation and accommodation relaxation [
80]. Interestingly, this latter MR study showed that higher intraocular pressure (IOP) caused a greater myopic degree (MR-Egger
P = 0.02), but without directional pleiotropy (intercept
P = 0.256). Notably, this relationship was not mediated by IOP-lowering medication [
81]. Furthermore, researchers observed that oral anti-diabetes drugs (metformin or gliburonide) may directly cause RE (MR-Egger
P = 0.003, intercept
P = 0.166). These findings suggest the effects of drugs on the ocular system must be accounted for when treating specific diseases.
Glaucoma is a common leading cause of irreversible blindness; by 2040, it is estimated that approximately 118 million people will suffer with the condition [
89]. Progressive retinal ganglion cell degeneration and unique visual field-loss are the main clinical manifestations of primary open-angle glaucoma (POAG) [
90]. Unfortunately, due to asymptomatic glaucoma, deriving a correct diagnosis at early disease stages in many patients is difficult [
91]. The 2013 US Preventative Services Task Force recommendation statement reported that IOP was a modifiable risk factor in patients with POAG; its decrease prevented disease progression and emphasised the importance of an early diagnosis and treatment [
92]. Notably, 10%–20% of POAG patients have normal or low IOP or "normal tension glaucoma" [
93]. Therefore, glaucoma control measures cannot use IOP as the only indicator; in fact, many other factors are associated with glaucoma, such as hyperglycaemia. To the best of our knowledge, diabetes cannot only cause micro-vascular lesion-diabetic retinopathy, but may be associated with glaucoma [
94,
95], however, epidemiological and clinical studies have not established a causal link between these conditions [
96,
97]. Most studies, using MR analysis, have focussed on relationships between POAG and type 2 diabetes (T2D) [
94,
95]. Previous studies screened relevant SNPs from an external GWAS database [
98‐
100], calculated different genetic risk indices (overall genetic risk index and mechanism-specific genetic risk indices) as IVs, and confirmed causality between T2D and POAG using mechanism-specific genetic IV analyses. Furthermore, these authors emphasised the key role of pancreatic β-cell functions in POAG. Since T2D is a binary variable, this conclusion suggested T2D patients with metabolic disorders caused by pancreatic β-cell dysfunction would have a higher glaucoma risk. These studies also reported that the pleiotropic melatonin receptor 1B gene (
MTNR1B) may have directly affected POAG via other mechanisms, independent of T2D. Accordingly, two loci (rs7708285 and rs6943153) were nominally significantly associated with POAG via fasting blood glucose, but the direction of effect was opposite. Thus, more studies on causal associations between fasting blood glucose and POAG, and the role of
MTNR1B in POAG susceptibility are warranted.
Apart from diabetes, relationships between other metabolic disorders and IOP have been expounded [
101,
102]. Hysi et al. reviewed the impact of different factors on IOP using metabolomics, Random Forest machine learning algorithms, and MR-IVW analyses to explore relationships between circulating metabolites and IOP [
103]. From three MR models, O-methylascorbate had a significant IOP-lowering effect. Specifically, raising the levels of this metabolite reduced IOP causally, while IOP had no causal effect on O-methylascorbic levels. O-methylascorbate is a circulating metabolite of vitamin C, which is an antioxidant vitamin. In future research, it will be important to extrapolate if vitamin C alters IOP through antioxidant stress mechanisms [
104].
Hyperlipidaemia is also a widely studied POAG biomarker [
105,
106]. A two-sample MR analysis using GWAS summary data as IVs, involving three plasma lipid components (low density lipoprotein cholesterol [LDL-C], HDL-C and triglyceride (TG)) as exposure variables, showed no significant causal relationships between hyperlipidaemia and POAG risk [
107]. However, only a causal association between routine lipid components and POAG was identified [
107]. In fact, several other lipoprotein particles defined by the internal lipid content affected the biological function of lipid metabolites [
108,
109]. Nusinovici et al., using a Bayesian network analysis, showed that total HDL-C was directly associated with POAG [
110]. SNPs that correlated with HDL-C were then selected as IVs for MR-IVW, weighted median, mode based estimator and configuration mixture (Remix) method analyses. The findings suggested a causal and specific relationship between HDL-C and POAG. Mechanistically, small, dense HDL sub-fractions could mediate the cholesterol outflow pathway via the ATP-binding cassette 1 transporter (ABCA1) pathway [
111,
112]. Also, imbalanced cholesterol levels can affect neurodegenerative disease [
113]. High cholesterol-24-hydroxylase levels were identified in the retinas of glaucoma patients which may have damaged optic ganglion cells [
114]. Additionally, HDL could contribute to POAG by influencing choroidal blood perfusion processes [
115]. Therefore, future functional studies on different lipid particles must examine if these molecules participate in glaucoma pathological mechanisms by affecting blood perfusion.
Corneal hysteresis (
CH) and corneal resistance factor (
CRF) are both quantitative genetic indicators of corneal biomechanics.
CH is a dynamic marker reflecting corneal structural and functional changes associated with IOP [
116,
117]. CRF is calculated using a coefficient that emphasizes the first applanation. These biomechanical properties affect corneal-compensated IOP (IOPcc) measurements, which are used as estimated IOP values [
118,
119]. Both indicators are genetically associated with ocular phenotypes and disease [
120‐
122]. Simcoe et al
. identified more than 200 loci related to corneal biomechanical properties using GWAS and the UK Biobank database, and conducted MR analysis to prove causal relationships between corneal biomechanics and IOP, and the directions of effect showed increased IOPcc reduced CH while simultaneously raising CRF [
123]. This study also identified an association between genetically induced
CRF and POAG, thus, future work could explore if the effects of
CRF on POAG are driven by IOP.
A previous study reported that acute IOP increases could be associated with caffeine intake [
124]. Although no links were identified between habitual caffeine intake and IOP changes in healthy subjects, increased caffeine consumption was associated with IOP increases in subjects with increased IOP susceptibility [
125]. Similarly, consistent outcomes were identified in individuals with a family history of glaucoma [
126]. Currently, causal evidence is lacking on whether reductions in long-term caffeine intake can control IOP and reduce glaucoma risks. Kim et al. conducted a two-sample MR analysis (UK Biobank database) using eight SNPs, which were significantly related to caffeine intake as IVs, to assess causal relationships between intake and IOP. Also, polygenic risk scores related to high IOP were calculated to test if genetically induced higher IOP altered associations between caffeine intake and IOP [
127]. These authors observed a weak effect of long-term caffeine consumption on reducing IOP, but no causal relationship was identified. In contrast, in subjects with a strong genetic susceptibility to higher IOP, greater caffeine consumption would increase IOP and lead to higher glaucoma prevalence.
Another important corneal biomechanical indicator, central corneal thickness (CCT) shows strong heritability, with variability dependent on ethnicity [
128,
129]. In a South African Eye Study, CCT in Africans was significantly lower than in other races [
130]. Furthermore, a thinner CCT was identified as a risk factor for POAG [
131]. Choquet et al. selected the main CCT related single nucleotide variants as IVs for a two-sample MR analysis, and showed a null causal association between thinner CCT and POAG [
132]. From these results, the phenotypic effect of a single SNP may have been reduced by the data inclusion criteria, or other factors such as the environment or epigenetics may have contributed.
A SMR analysis was conducted in Asian and European populations; multiple genes,
CLIC3, ILMN_1796423, PTGDS, ILMN_1664464, C9orf142 and
ILMN_1761138 were identified and showed pleiotropic associations with CCT, but only in European ancestry participants [
133]. Importantly, this study reflected the role of gene expression traits on outcomes in an MR framework. However, expression quantitative trait loci (eQTL) data were obtained from blood rather than eye tissue, and the same eQTL data were used in the analysis of different races. In the future, eQTL data must be sought from human eye tissue, and data analysis must be expanded to different races.
Optical coherence tomography (OCT) is an intravascular imaging technology widely used for the early diagnosis and management of ophthalmic diseases [
134,
135]. Previous research showed that the thickness of the two inner retina layers, the retinal nerve fibre layer (RNFL) and ganglion cell inner plexiform layer (GCIPL), could change at different glaucoma stages, suggesting these indices could function as biomarkers for predicting glaucoma [
136,
137]. A two-sample MR analysis using OCT images from 31,434 UK Biobank participants identified a relationship between morphological retinal phenotypes (RNFL and GCIPL) and POAG [
138]. Because the causal relationship between IOP and POAG was clear, this study concluded the combination of time and IOP made the inner layer of retina thinner. Also, GWAS outcomes showed that three chromosomal loci, 1q42.3, 17q25.3 and 21q22.13 were related not only to eye pigmentation which caused foveal hypoplasia, but also eye colour [
139]. These loci were also significantly related to visual acuity. Based on known genes, we can study associations between appropriate exposure factors related to them (such as changes in eye structure colour or the change force of skin and hair colour) and vision loss. OCT images combined with MR analyses could help unravel the biological function of different retina layers and provide insights on causal relationships between retinal morphology and eye disease.
Diabetic retinopathy (DR), plasma lipid and adiponectin (APN) correlations
Globally, DR is one of the leading causes of blindness in working-age individuals [
140], therefore, identifying and modulating risk factors is essential in preventing this condition. However, the underlying causal factors remain unclear. A previous study used GWAS summary data to investigate if lipid changes causally increased DR risk, using MR-IVW methods across three subgroups comprising all ethnicities, Caucasians and Chinese [
141]. No clear evidence suggested dyslipidaemia increased the risk of DR. However, an analysis of 12 pleiotropic TG-associated SNPs in the model suggested IVs could be associated with other lipid components. Three rs12678919, rs174546 and rs964184 SNPs were closely associated with the proliferator-activated receptor-α (
PPAR-α) gene which is used to treat DR by reducing TG levels [
142]. These observations suggested TGs may be involved in DR formation. However, as study data comprised summarised statistical data from large GWAS studies, no accurate blood lipid values were identified, therefore extrapolations to clinical settings are limited.
A correlation was previously identified between APN and DR. A previous meta-analysis showed APN levels were significantly associated with DR in patients with T2D [
143]. An observational cohort study also examined if elevated APN levels accelerated retinopathy progression in patients with T2D, and suggested APN could be a biomarker for DR pathogenesis [
144]. APN is an adipokine secreted by fat cells and has a variety of functions, including promoting blood vessel formation [
145] and endothelial cell inflammatory responses [
146]. To comprehensively ascertain if APN played an essential role in DR pathogenesis, Huang et al. performed an MR-IVW analysis on three SNP groups (GRS
ALL, GRS
Limited and GRS
APN) which were previously differentially correlated with APN levels, by setting different thresholds and calculating the GRS of each group as an IV [
147]. The reason for choosing GRS: through the calculation of F statistic, it was found that there were numerous weak IVs among APN-associated SNPs. A causal association between APN and DR was not fully proven, as only data from IVs- GRS
ALL suggested elevated APN levels could causally increase DR risk. Interestingly, this causal relationship between APN and DR only focused on T2D. It was previously suggested that the effects of APN on inflammatory markers in type 1 diabetes (T1D) or T2D was opposite [
148,
149]. Therefore, the effects of APN on DR in T1D patients must be comprehensively examined in future studies.
BD is a chronic, multi-system disease of unknown aetiology [
153,
154]. It causes pathological damage to various organs, comprising oral and genital ulcers and ocular and cutaneous lesions [
155]. Although BD pathogenesis is unclear, genetic susceptibility, infection triggers and immune abnormalities have been identified as the main pathogenic mechanisms [
156]. Previous studies reported that patients with BD could be treated by vitamin D supplementation to modulate immune and inflammatory responses [
157‐
159]. Zhong et al. used allele scores formed by four variants related to serum 25(OH)D levels as IVs to show that higher serum 25(OH)D levels increased BD risk [
7]. However, the study conclusions contrasted with the protective effects of vitamin D on BD in observational studies [
158,
159]. Therefore, potential BD development must be monitored after high-dose or long-term vitamin D intake. BD is also characterised by typical clinical features such as uveitis [
160], which has a prevalence of 38–714/100000 individuals [
161]. Uveitis causes serious damage to the eye tissue and leads to eye complications, including cataracts, glaucoma and ocular tuberculosis [
162]. Pathophysiologically,
Mycobacterium tuberculosis causes inflammatory immune response in the body [
163], but no evidence based on MR analyses have been reported. However, Zhong et al. conducted MR and
mycobacterium tuberculosis T cell spot test (T-SPOT.TB) analyses and showed that tuberculosis infection increased the risk of several non-infectious uveitis conditions [
164].
Opportunities for MR in ophthalmology
MR applies IV methods to observational data; IVs are genetic variants that provide reliable evidence inferring causal relationships between exposures and outcomes. Thus, MR has become instrumental in dissecting residual ophthalmology issues by combining genomics and other disciplines. Currently, large-scale GWAS have been used to identify gene loci related to eye traits and diseases, which have not only provided essential support for gene targeted therapies, but also identified preventable and modifiable environmental risk factors using MR. Indeed, specialised ocular trait measurement instruments are often required to identify apparent features in the eyes or physiological indices inside the eyeball. To measure eye traits that are not readily available, MR is a convenient tool as it analyses gene expression from blood tissue as IVs to identify exposure factors [
133]. In addition, MR is also used to analyse the causal role of specific pathological pathways involved in common ocular diseases (e.g. how the pancreatic β-cell-pathway in T2D increases the risk of POAG [
100]). These examples show how much underlying disease pathologies in ophthalmology have been clarified. With the development of technology, novel MR techniques have been applied to ophthalmology research, which can solve the directionality of causality and the interaction between phenotypes. In fact, there are three assumptions before implementing the MR method to make causal inferences between exposures and outcomes: 1. There is a causal relationship between exposures and outcomes; 2. There is no causal relationship between exposures and outcomes; 3. There is a bidirectional causal relationship between exposures and outcomes. When the basic MR method cannot fully explore these three situations, bidirectional MR becomes an auxiliary method to clarify the sequence of causality [
173,
174]. A typical example in ophthalmology research is that education level is related to myopia. In a study using bidirectional MR, it is found that the causal relationship between them is unidirectional, and there is no reverse causality (the idea that myopia takes longer to learn is not valid[
60]). In addition, the other advanced MR approach, multivariable MR analysis, which can be performed to identify associations between multiple categorical exposure factors with similar effects, allows genetic variants to be associated with more than one exposure in this analysis. At present, there are an increasing number of researches using multivariable MR analysis, which has played a unique advantage in the exploration of the etiology of mental illness and cardiovascular disease [
175,
176]. The advantage of multivariable MR is that eliminating the exposure factors with strong causality and exploring the causality values of the exposure factors with relatively weak correlation (for example, IOP, which is weakly correlated with SE, has less influence on RRD than IOP, but IOP still has causal influence on RRD[
86]).