Causal role of plasma liposome in diabetic retinopathy: mendelian randomization (MR) study

Abstract

Background

Research indicates that there may be an association between plasma lipidome levels and the incidence of diabetic retinopathy (DR) in patients. However, the potential causality of this relationship is yet to be determined. To investigate this matter further, we employed a two-sample Mendelian randomization (MR) analysis to comprehensively assess the causality between lipidome levels and DR.

Methods

Summary statistics for lipid levels and DR were obtained from the Genome-Wide Association Studies (GWAS) Catalog database and the FinnGen Consortium, respectively. We conducted a two-sample MR analysis, and statistical analysis were performed using the inverse variance weighted (IVW) with the addition of the MR-Egger, weighted median (WM), constrained maximum likelihood and model averaging (cML-MA) to test for causal associations between lipid levels and DR. Heterogeneity was checked using Cochran’s Q statistic. The MR Pleiotropy Residual Sum and Outlier (MR-PRESSO) global test and the MR-Egger regression were used to detect horizontal pleiotropy. The robustness of our findings was assessed using leave-one-out and funnel plots. To further assess the reliability of the results, linkage disequilibrium score regressions, colocalization analysis and reverse MR analysis were also performed.

Results

Analysis of the pooled MR results and after correction for the false discovery rate (FDR) revealed that five lipid levels were associated with DR risk. Phosphatidylcholine (16:0_16:0) levels [OR = 0.869 (0.810 to 0.933), P_fdr = 0.006], phosphatidylcholine (16:0_20:2) levels [OR = 0.893 (0.834 to 0.956), P_fdr = 0.043] and phosphatidylethanolamine (18:0_20:4) levels [OR = 0.906 (0.863 to 0.951), P_fdr = 0.006] were protective against DR, whereas sphingomyelin (d36:1) levels [OR = 1.120 (1.061 to 1.183), P_fdr = 0.006], and sphingomyelin (d40:1) levels [OR = 1.081 (1.031 to 1.134), P_fdr = 0.043] were associated with a greater risk of DR. Further sensitivity analysis did not reveal heterogeneity or horizontal pleiotropy.

Conclusion

In summary, genetic evidence suggests a causal relationship between the levels of specific lipid levels and DR. These findings may provide valuable insights into the causal relationships between lipid levels and DR, potentially informing future prevention and treatment strategies.

Background

Diabetes mellitus (DM) represents a significant long-term health challenge. According to the International Diabetes Federation, approximately 463 million individuals were affected globally in 2019, with projections indicating that this number could rise to 700 million by 2045 [1]. Diabetic retinopathy (DR) is a chronic pathological complication of diabetes and is recognized as the most prevalent and specific microvascular complication of DM [1]. It originates from microaneurysms in the retinal blood vessels, which cause retinal damage through abnormal vascular proliferation, swelling, and intravascular fluid leakage, potentially leading to vision loss or blindness. Notably, DR is a leading cause of blindness in the adult working-age population [2], significantly affecting individuals, families, and societies. Effective management of DM, focusing on prevention, early detection, and treatment, is crucial for reducing the risk of DR. Additionally, controlling blood glucose levels, managing blood pressure, ceasing smoking, and maintaining a healthy lifestyle are vital strategies to mitigate the risk of developing DR and its subsequent progression [3].

The association between lipids and DR has been a subject of extensive research since the 1950s [4]. Despite the fact that this relationship has been the focus of numerous studies, the findings remain inconclusive and controversial. For instance, the Early Treatment Diabetic Retinopathy Study reported that higher serum lipid levels were associated with an increased risk of retinal exudates [5]. Conversely, a study conducted by King demonstrated that reducing serum lipid levels resulted in a decrease in the prevalence of hard exudates in individuals with DR [6]. Moreover, several studies have identified a possible association between dyslipidemia and the onset of retinopathy in diabetic patients [7, 8], implying that lipids may play a role in the pathogenesis of DR. However, the relationship between lipids and DR is complex. Other studies have reported a lack of association between serum lipids and the presence of DR [9]. This inconsistency may be attributed to the intricate and multifaceted nature of the relationship, which is influenced by various factors such as diabetes duration, glycosylated hemoglobin, diastolic blood pressure, proteinuria, and body mass index [10]. Furthermore, the question of whether dyslipidemia is a primary cause or a consequence of DR remains unresolved, highlighting the need for further research to validate any potential causal links. Notably, recent studies have suggested that standard lipid measurements may not adequately capture the full picture of microvascular disease development in diabetic patients [11]. This underscores the importance of modern lipidomics techniques, which offer a more nuanced understanding of the diversity and range of circulating lipids [12]. These advances have the potential to enable a more comprehensive assessment of DR risk factors than traditional methods.

Mendelian randomization (MR) offers significant advantages in genetic epidemiology. By utilizing genetic variation that are strongly associated with specific exposures, MR enables the inference of causal relationships between these exposures and disease outcomes under specific assumptions. This approach provides new opportunities to test for causality and illustrates how investing in human genome research can aid in understanding and preventing the harmful effects of modifiable exposures on human health [13]. Furthermore, since genetic variants precede disease onset and this sequence is immutable, MR-based studies are inherently protected against reverse causality. Utilizing genetic variants that affect plasma lipidome levels as instrumental variables (IVs), we investigated their impact on the risk of developing or progressing DR. Overall, MR represents a powerful tool in genetic epidemiology that significantly advances our understanding of disease etiology and aids in shaping prevention and treatment strategies.

Methods

Study design

This study adhered to the guidelines established by the Strengthening the Reporting of Observational Studies in Epidemiology Mendelian Randomization (STROBE-MR) framework [14]. Using a two-sample MR approach, this investigation conducted a causal analysis of plasma lipidome levels and DR. The study was grounded in three principal assumptions of MR: firstly, the exposure variable was identified as plasma lipidome levels, with IVs being single nucleotide polymorphisms (SNPs) significantly associated with these levels; secondly, the genetic variants used as IVs were not associated with any confounders of the lipid-DR relationship; thirdly, the genetic variants were presumed to influence DR exclusively through lipid levels. By leveraging genetic information (i.e., SNPs) as IVs, we enhanced the robustness of causal inference. The methodology relies on publicly accessible Genome-Wide Association Studies (GWAS) for statistical summarization, thus bypassing the need for additional ethical approval or informed consent (Fig. 1).

Fig. 1

Study design of two-sample MR study of the association between 179 lipid species levels and Diabetic retinopathy. Genetic variants used as IVs fulfill three conditions: I there must be a clear and strong association between the genetic variant and the exposure; II the genetic variant that is the IV must be independent of any confounders of the association between the exposure and the outcome; and III the variant affects the outcome only through the exposure pathway and not through other biological pathways. In our study, the solid pathway is significant; the dashed pathway should not be present. Abbreviations: IVW-inverse variance weighted; WM-weighted median; cML-MA -constrained maximum likelihood and model averaging; MR-PRESSO-MR Pleiotropy Residual Sum and Outlier

Data source

This study utilized exposure data were meticulously selected from the GWAS data on human circulating lipids, kindly provided by Ottensmann [15]. The comprehensive dataset comprises genome-wide analyses of 179 lipid species, derived from 7,174 individuals in Finland. Utilizing advanced shotgun lipidomics techniques, the research team successfully detected 179 distinct lipid species, belonging to 13 lipid classes, encompassing four major lipid categories: glycerolipids, glycerophospholipids, sphingolipids, and sterols. The lipid species are systematically named using a standardized notation: class name followed by the sum of carbon atoms, the sum of double bonds, and the sum of hydroxyl groups. This notation facilitates clear and consistent communication among researchers. The annotation of lipid subspecies goes beyond basic identification, incorporating valuable information on their acyl moieties and, when available, details on their sn-positions. The acyl chains within the lipid molecules are separated either by an underscore (“_”) when the sn-position on the glycerol cannot be unambiguously determined or by a forward slash (“/”) when the position is resolvable. The GWAS catalogue serves as a valuable resource for the scientific community, providing GWAS summary statistics for lipid traits. These statistics, accessible through accession numbers ranging from GCST90277238 to GCST90277416, offer insights into the genetic basis of lipid metabolism and its association with various health outcomes. In the context of this study, GWAS summary data for DR were obtained from the FinnGen Consortium (R9 release). This dataset comprises 10,413 patients with DR and 308,633 controls. FinnGen is a public-private collaborative research project that combines imputed genotype data generated from newly collected samples from the Finnish Biobank and legacy samples with data from the Finnish Health Registry (https://www.finngen.fi/en) digital health records, with the aim of providing new insights into the genetics of the disease [16], which provides a solid basis for exploring the genetic contribution of DR. Table 1 shows the details of the exposure and outcomes analyzed in this MR study.

Table 1 Details of the exposure and outcome

Selection of IVs

First, to obtain more SNPs, we set a significance threshold for the plasma liposome data to p < 1 × 10^− 5, thereby identifying SNPs that were genetically associated with the traits of interest. Subsequently, we conducted a linkage disequilibrium (LD) test to ensure the independence of these clustered SNPs [17]. Since exposure and outcome data were derived from European populations, the LD of selected IVs for each trait was calculated based on the 1000 Genomes European reference panel [18].The LD criteria were set with a minimum r² value of 0.001 and a maximum window size of 10,000 kb. This step was crucial for eliminating any potential biases arising from correlated SNPs. Furthermore, we recognized the importance of detecting weak IVs in our analysis. To this end, weak IVs were detected using the F statistic, calculated as follows: F = $\:\frac{{\text{R}}^{2}\:(\text{N}-2)}{(1-{\text{R}}^{2})}$, where R² represents the cumulative explained variance, and N denotes the sample size for the selected SNPs [19,20,21]. To ensure the robustness of the findings, we used a threshold of F greater than 10 to indicate the absence of weak IVs. To mitigate the possibility of bidirectional effects confounding our results, we conducted a detailed investigation of the phenotype associated with each SNP. SNPs associated with phenotypes related to the study outcomes were excluded from further consideration. This rigorous screening process ensured the validity and reliability of our SNP set for further analyses. Notably, we leveraged information from the PhenoScanner database V2 (accessed on February 15, 2024) to inform our SNP selection and phenotype investigation. This comprehensive resource provided valuable insights into the genetic architecture underlying our traits of interest [22]. Finally, we performed the Steiger test to confirm that MR assumption of exclusivity was not violated in our dataset. This additional step further strengthened the confidence in our analytical approach and the subsequent findings [23].

MR analysis

Forward MR analysis

The study utilized four primary methodologies: inverse variance weighting (IVW), MR-Egger, weighted median (WM), and the constrained maximum likelihood and model averaging (cML-MA) approach, each with distinct significance and applications in MR analysis. IVW [24, 25] is a prevalent method in MR research that produces reliable results, particularly in the absence of cross-sectional pleiotropy. It leverages the inverse of each genetic variant’s variance as a weighting factor, minimizing measurement error-induced bias. MR-Egger [26, 27] is a technique that offers valid MR estimates, particularly in the presence of horizontal pleiotropy where genetic variant pleiotropic effects are unrelated to genetic exposure association. By utilizing the MR-Egger regression slope, unbiased causal effect estimates can be achieved. WM [28, 29] is considered a reliable approach that generates valid causal estimates even when up to 50% of IVs are invalid or influenced by pleiotropy. It ranks the effects of all genetic variants, selects the median, and thus minimizes sensitivity to outliers. The cML-MA [30] approach is a method that effectively identifies invalid IVs with either correlated, uncorrelated pleiotropic effects, or both, enhancing the consistency of causal effect estimations and inferences between exposures and outcomes. By implementing constraints in maximum likelihood estimation, this method ensures accuracy even when some IVs are invalid, significantly enhancing the robustness and precision of MR analysis.

Sensitivity analysis

Sensitivity analysis, including tests for heterogeneity and pleiotropy, were conducted to assess the reliability of the results. Heterogeneity in MR refers to variability in estimates derived from different IVs, which can undermine the validity of causal inferences. The Cochran Q statistic was used to measure heterogeneity in the causality estimations produced by both the IVW and MR-Egger approaches. In MR analysis, heterogeneity implies to changes in estimates obtained from distinct IVs, which might undermine the validity of causal inferences drawn from these IVs. A Cochran Q statistic resulting in P-values greater than 0.05 indicates a lack of significant heterogeneity within the study. Additionally, pleiotropy assessments were performed using the MR-Egger intercept [26] and the MR Pleiotropy Residual Sum and Outlier (MR-PRESSO) test [31]. P-values exceeding 0.05 in these tests suggest no substantial evidence of pleiotropy affecting the results. The reliability of our study’s outcomes was further evaluated through funnel plots and the leave-one-out sensitivity test.

LDSC regression analysis

MR analysis may violate causal effects in the context of genetic correlation between exposure and outcome [32]. Although SNPs associated with DR were excluded from the selection of IVs, uncorrelated SNPs may also mediate the genetics of DR. LDSC is a commonly used genetic correlation analysis method to estimate the genetic contribution of complex diseases and complex traits. Based on the concept of LD, the strength of association between each SNP and a complex trait is inferred by estimating the LD score for each SNP [33]. Therefore, to ensure the causal effects are not confounded by genetic correlations, LDSC was employed to assess the genetic correlation between the identified lipids and DR, providing a coherent framework for our analysis (https://github.com/bulik/ldsc).

Colocalization analysis

Using Bayesian colocalization analysis, we investigated whether lipids and DR share common causal variants within specific genomic regions [34, 35]. For lipids demonstrating evidence of a causal relationship with DR (P_fdr < 0.05), variables within a 50 kb range of the SNP of the corresponding to the instrumental variable were extracted and posterior probabilities (PP) were calculated. A PP.H4 value of 0.80 or higher was generally considered indicative of colocalization, suggesting shared genetic causality between the lipid and DR.

Reverse MR analysis

In order to elucidate whether the altered liposome levels observed in the positive results from the forward MR analysis were influence by DR, we performed a reverse MR analysis. The criteria for the selecting SNPs for this analysis illustrated in Fig. 1. The methodology employed in the reverse MR adheres to the same protocols as those described in our previous MR studies.

To determine the relationship between plasma liposome levels and the risk of DR, a P-value of less than 0.05 was considered nominally significant. A causal effect was inferred if the P-value obtained from the MR analysis met the false discovery rate (FDR) adjustment criteria. This study was conducted via R software (version 4.1.3), utilizing the “TwoSampleMR”, “MRPRESSO”, “MRcML”, “coloc” and “ldscr” packages.

Results

Details of the IVs

Based on the stringent IV screening criteria (P < 1 × 10^− 5), we excluded all weak IVs by calculating their F-statistics, which ranged from 18.80 to 644.19. This process ensured that the selected SNPs demonstrated robust IV effects, thereby minimizing the potential for weak instrumental bias. After removing potential confounders, we ultimately selected 3,030 SNPs associated with 179 lipid traits. To confirm that the exclusivity assumption of MR was maintained, we also conducted a Steiger test (Supplementary Tables 1–2).

Forward MR analysis

Among the 179 lipids examined for their causal association with DR, 20 were nominally significant (p < 0.05). After correcting for FDR, five lipids still maintained a strong association with DR. Specifically, phosphatidylcholine (16:0_16:0) levels (P_fdr = 0.006), phosphatidylcholine (16:0_20:2) levels (P_fdr = 0.043), phosphatidylethanolamine (18:0_20:4) levels (P_fdr = 0.006), sphingomyelin (d36:1) levels (P_fdr = 0.006) and sphingomyelin (d40:1) levels (P_fdr = 0.043) demonstrated significant associations. The first three lipids displayed negative correlations, while the last two exhibited positive correlations. The corresponding IVW-OR (95% CI) values were 0.869 (0.810 to 0.933), 0.893 (0.834 to 0.956), 0.906 (0.863 to 0.951), 1.120 (1.061 to 1.183), and 1.081 (1.031 to 1.133). In other MR analyses of these five lipids, the MR Egger, WM, and cML-MA results were in the same direction as the IVW results (Fig. 2 and Supplementary Table 3).

Fig. 2

Lipid levels and diabetic retinopathy - an overview of the main results of the two-sample MR study. Abbreviations: IVW-inverse variance weighted; WM-weighted median; cML-MA -constrained maximum likelihood and model averaging

Sensitivity analysis

We conducted sensitivity analysis on the five lipids to evaluate their reliability and robustness. Cochran’s Q test revealed no significant heterogeneity, and the MR-Egger intercept did not indicate the presence of horizontal pleiotropy. Furthermore, neither the MR-PRESSO test nor the leave-one-out analysis detected any outliers, underscoring the reliability of our results. The sensitivity analyses were thoroughly documented, confirming the robustness of the causal effects associated with the five lipids (Table 2 and Supplementary Figs. 1–4).

Table 2 Forward MR sensitivity analysis details

LDSC regression analysis

LDSC analysis was utilized assess the genetic correlation between the lipids demonstrating positive associations and DR. Subsequent estimations of SNP heritability for these lipids and DR, derived using LDSC, indicated that SNP heritability (the proportion of variance attributable to genome-wide SNPs) ranged from 0.023 to 0.185. The results revealed no genetic correlation between the lipids and DR, suggesting that the MR estimates are not confounded by shared genetic factors (Table 3 and Supplementary Table 4).

Table 3 Details of genetic correlations between DR and five significant liposomes

Colocalization analysis

Colocalization analysis was conducted to determine whether common regions of genetic variation exist among the five lipids identified in this study as being associated with DR risk. PP.H4 for all lipids were calculated, ranging from 0.007 to 0.265. These results, with all PP.H4 values being less than 0.80, indicate a lack of evidence for a shared genetic locus underpinning the association between these lipid phenotypes and DR (Table 4 and Supplementary Fig. 5).

Table 4 Details of gene colocalization analysis between DR and five significant liposomes

Reverse MR analysis

In order to address the potential for reverse causality, a reverse MR analysis was conducted. In this analysis, DR was treated as the exposure, and the five lipids previously identified with positive associations were considered as the outcomes. The results of the reverse MR demonstrated no evidence of causality. Additionally, there was no indication of heterogeneity or horizontal pleiotropy, thereby confirming the robustness of our findings (Supplementary Tables 5–8 and Supplementary Figs. 6–9).

Discussion

Our research demonstrated that five lipids remained strongly associated with DR after FDR correction. Notably, higher levels of phosphatidylcholine (16:0_16:0), phosphatidylcholine (16:0_20:2), and phosphatidylethanolamine (18:0_20:4) are associated with a reduced incidence of DR. Specifically, for every one standard deviation (SD) decrease in the levels of phosphatidylcholine (16:0_16:0), phosphatidylcholine (16:0_20:2), and phosphatidylethanolamine (18:0_20:4), the incidence rates decrease by 13.1%, 10.7%, and 9.4%, respectively. Conversely, higher levels of sphingomyelin (d36:1) and sphingomyelin (d40:1) are associated with an increased the risk of DR. For every one SD increase in sphingomyelin (d36:1) and sphingomyelin (d40:1) levels, the incidence rates increase by 12% and 8.1%, respectively. Furthermore, our results indicate no reverse causality. Colocalization analysis suggests that the observed associations we observed between lipid levels and DR risk are not influenced by common genetic loci. These findings elucidate the complex relationship between different lipid types and disease, offering new insights for public health interventions aimed at reducing DR risk factors.

Phosphatidylcholine plays a key role in a variety of physiological activities, including the regulation of inflammatory responses, leukocyte activation, memory and neural signaling, and cardiovascular protection [36]. While direct research linking phosphatidylcholine to DR pathogenesis remains limited, evidence suggests its potential role in retinal inflammation. The retina, a highly specialized organ, regulates lipid levels independently of systemic lipids. Dysregulation of these lipids can precipitate low-grade chronic inflammation, leading to increased endothelial permeability and cellular damage within the retina [37]. Notably, chronic low-grade retinal inflammation is a significant factor in DR development [38]. During retinal inflammation, reductions in phosphatidylcholine levels have been observed [39]. In vivo studies using an N-methyl-D-aspartate induced disease model have demonstrated that submicron amounts of phosphatidylcholine exert retinoprotective effects [40]. Recent clinical trials have shown that phosphatidylcholine is effective in treating eye diseases, positioning it as a valid candidate for managing inflammatory degenerative diseases, including age-related macular degeneration, DR, diabetic macular edema, retinal vein occlusion, uveitis and endophthalmitis [41]. Furthermore, studies on phosphatidylcholine liposomes in ischemia-reperfusion induced inflammation and neuronal death have shown significant reductions in the expression of proinflammatory genes such as IL-1 beta, IL-6, Ccl2, Cxcl10, and Icam1. These markers are often elevated in DR patients and play critical roles in disease progression [42]. Early-stage DR involves pathological changes in the vasculature, notably leukocyte-endothelial interactions, as hyperglycemia activates circulating leukocytes, leading to adherence to the vessel walls and gradual capillary obstruction, which ultimately results in permanent disruption and loss of endothelial lining. This pathology manifests in the late stages of DR as reduced capillary density and defects [43]. Phosphatidylcholine has also been effective in reducing iNOS expression inhibition and leukocyte activation, modulating inflammatory factors, and decreasing structural joint damage, pain, and angiogenesis in models of collagen-induced arthritis and systemic inflammation [44, 45]. The involvement of these proinflammatory markers in DR pathogenesis is well-documented. For instance, IL-1 beta and IL-6 contribute to oxidative stress during disease development [46,47,48,49], while chemokines such as Ccl2, Cxcl10, and Icam1 have been shown to mediate inflammatory cellular stasis and vasculopathy in the retina [50,51,52]. Based on our study, we may venture to speculate that phosphatidylcholine may play a protective role against DR by attenuating intravascular inflammation and thereby protecting the integrity of the blood-retinal barrier.

It was shown that phosphatidylethanolamine levels in lipoprotein (a) were significantly lower in non-DR patients than in DR patients. These findings appear to be inconsistent with previous research which suggested no significant differences [53]. The discrepancy might stem from the small sample sizes used in the functional assays, which limits the ability to conclusively determine the relationship between phosphatidylethanolamine and DR. In contrast, another observational study indicated that DR patients exhibited lower, statistically significant phosphatidylethanolamine levels in erythrocyte membranes than those without DR [54]. Further research by David A. Ford on retinal degeneration in a rat model of Smith-Lemli-Opitz syndrome revealed notable reductions in both phosphatidylcholine and phosphatidylethanolamine [55], emphasizing the significant connection between phosphatidylethanolamine and visual system impairment. This association was further supported by findings from a study on repetitive mild traumatic brain injury, which demonstrated a crucial link between phosphatidylethanolamine levels and visual damage [56]. As a multifunctional phospholipid essential in various cellular metabolic processes, phosphatidylethanolamine also plays a critical role in autophagy by binding with LC3-I to form LC3-II, a necessary component for autophagosome formation [57]. Alterations in phosphatidylethanolamine levels could influence intracellular autophagic activity. Experimental elevation of phosphatidylethanolamine through the provision of ethanolamine or overexpression of the phosphatidylethanolamine-producing enzyme Psd1 has shown to enhance autophagy in both yeast and mammalian cell cultures [58]. Autophagy helps mitigate oxidative stress and inflammation by removing damaged cellular components, thereby reducing oxidative damage and pro-inflammatory cytokine production. In addition, it alleviates endoplasmic reticulum stress by degrading its damaged components. The role of autophagy in cellular survival is multifaceted, as it not only protects retinal cells from apoptosis by supplying energy and nutrients but also, excessive autophagy may lead to cell death [59]. This underscores the complex interplay between phosphatidylethanolamine and retinopathy, highlighting the need for more mechanistic studies to elucidate this relationship.

Lipid analysis indicates that approximately 70% of total cellular sphingolipids reside within membrane microdomains [60]. This concentration of sphingolipids in membrane microdomains positions them as a key reservoir for ceramide production via sphingomyelinases present in the cell membrane [61]. Ceramides tend to spontaneously fuse, forming ceramide-rich platforms that can reorganize local membrane bilayers. These structural domains become focal points for protein concentration and oligomerization, facilitating receptor aggregation and amplifying inflammatory signaling across various receptors [62,63,64]. Additionally, increased ceramide levels have been correlated with reduced membrane fluidity in diabetic circulating angiogenic cells, impeding their migration to damaged areas in the retinal vasculature in a DR mouse model, which results in these areas remaining unrepaired [65]. As a central element of sphingolipid metabolism, ceramide acts as a critical mediator in the signaling cascade that regulates apoptosis [66]. The activation of caspase-3 by ceramide, alongside the inactivation of the anti-apoptotic protein Bcl-2 and the disruption of mitochondrial function, underscores its role in apoptosis induction. Further investigations have shown that ceramide specifically binds and activates cathepsin D, enhancing the substrate activity of its mature isoform through interaction with ceramide. Recent studies have proposed that lysosomal organizing proteases may serve as novel mediators in apoptosis [67]. Notably, DR is intricately linked to apoptosis, a predominant cause of this pathology. Apoptosis in this context follows two principal pathways: the intrinsic and extrinsic pathways [68]. The intrinsic pathway typically begins with oxidative stress that disrupts mitochondria, leading to cytochrome c release and triggering a cascade resulting in cell death. Concurrently, inflammatory agents like tumor necrosis factor-alpha can induce apoptosis through receptor binding and activation of extrinsic pathways.

Lipidomics holds significant potential for enhancing risk stratification in DR. By identifying specific lipid profiles associated with various stages of DR, clinicians can more accurately determine which patients are at elevated risk of disease progression. For example, a pioneering prospective observational registry study was the first to establish that serum phosphatidylcholine levels are significantly correlated with different stages of DR in an Asian cohort, positioning phosphatidylcholine as a metabolic marker for advanced DR [69]. Further research into the lipidomic profiles of DR patients revealed that phosphatidylethanolamine and ceramides are the most altered lipids in the DR group compared to controls. More detailed analysis indicated that triglycerides, sphingomyelin, phosphatidylcholine, and ceramides are substantially altered in proliferative DR relative to non-proliferative DR [70]. Additionally, lipidomic analysis of aqueous humor in patients with proliferative DR versus controls without DM indicated significant increases in phosphatidylethanolamine, sphingomyelin, and phosphatidylcholine [71]. These findings underscore the utility of biomarkers in offering a comprehensive understanding of DR, considering its complex pathogenesis. Due to the multifaceted nature of the disease, a combination of multiple biomarkers might prove more effective than a single biomarker for optimal disease management. Consequently, there is a pressing need for further extensive preclinical and clinical research to explore the mechanisms by which lipids influence DR.

In the present study, Bayesian colocalization analysis was employed to assess shared genetic loci. Although the PP.H4 value is below 0.80, it is essential to consider the implications of these results for the biological plausibility of causal relationships. The PP.H4 value quantifies the posterior probability that two features are both associated and share a single causal variant. A value below 0.80 suggests limited confidence in a shared causal variant between the two features. A number of factors may be contributory to this outcome. Firstly, genetic variants may not exhibit strong enough association signals in GWAS data, potentially due to small sample sizes, low effect sizes, or other factors that diminish the power to detect significant associations. Secondly, the traits under investigation may be influenced by multiple causal variants, rather than a single shared variant. This can result in low PP.H4 values, as the analysis assumes the presence of only one causal variant. Thirdly, the biological pathways involved in these traits may be more complex than initially assumed, and the involvement of multiple genes and variants must be considered, as must the possibility that the identified shared genetic loci do not fully encapsulate the underlying biology.

This MR study offers several advantages: Firstly, to the best of our knowledge, it is the first MR study to assess the relationship between serum circulating lipids and DR. This methodology leverages the three foundational principles of Mendelian inheritance to minimize confounding from factors typically associated with observational studies, thereby providing more robust evidence of a causal link between exposure and outcome. Secondly, the strength of this analysis lies in the utilization of the largest and most precise dataset of genetic variables available, which significantly reduces potential biases in the study results. Thirdly, the study population is based on a European ethnic group, which mitigates the impact of ethnic variability on the results. Fourthly, the genetic correlation between lipids and DR was evaluated using LDSC and colocalization analysis, enhancing the credibility of the MR findings.

Our study, while thorough in its methodology, is subject to several limitations. First, despite utilizing MR methods to minimize the impact of confounding variables, it was not possible to completely eradicate residual bias due to unmeasured confounders. Second, the study cohort predominantly consisted of European individuals, limiting the applicability of our findings to other ethnic groups. Third, the adoption of broader thresholds for assessing results may have led to an increased incidence of false positives, albeit allowing for a more comprehensive evaluation of the association between lipid levels and DR. Additionally, our results suggest a causal link between specific lipids and DR, though the precise mechanisms remain unclear. Lastly, the reliance on summary statistics rather than individual-level data restricted our capacity for more granular subgroup analysis.

Conclusion

In summary, this pioneering study serves as the first comprehensive MR analysis using genome-wide data to explore the causal relationship between circulating lipids and DR. The identification of five lipids potentially associated with DR offers promising prospects for both clinical practice and research. These findings could transform DR screening and prevention strategies by providing effective biomarkers that enhance early detection and improve treatment outcomes.