Human retinal microvascular endothelial cells (hRMECs) and ARPE-19 retinal pigment epithelial cells were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). hRMECs were cultured in endothelial cell medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin. ARPE-19 cells were maintained in DMEM/F12 (Gibco, USA) containing 10% FBS and 100 U/mL penicillin-streptomycin. All cells were incubated at 37°C with 5% CO₂. To induce hyperglycemia, cells were treated with 30 mM D-glucose for 48 hours (13, 14). Cells cultured in normal glucose (5–5.5 mM) served as controls.

Cell viability was assessed using the Cell Counting Kit-8 (CCK-8, Dojindo). hRMECs and ARPE-19 cells were seeded in 96-well plates at 1 × 10⁴ cells/well and incubated overnight. After 48-hour treatment, 10 μL of CCK-8 solution was added per well and incubated at 37°C for 1 hour. Absorbance was measured at 450 nm using a microplate reader.

Total RNA was extracted from ARPE-19 cells using TRIzol reagent (Invitrogen, USA) and reverse-transcribed into cDNA with PrimeScript RT Master Mix (Takara, Japan). The following cycling conditions: 95°C for 30 s, followed by 40 cycles of 95°C for 5 s and 60°C for 30 s. The following primers were used: LAT2 F 5′-GCTGGCTGTGATGTTCTGG-3′, R 5′-TGGTGGTAGAGGTGCTGATG-3′; GAPDH F 5′-GAAGGTGAAGGTCGGAGTC-3′, R 5′-GAAGATGGTGATGGGATTTC-3′.

This study employed a two-sample MR approach followed by mediation analysis to investigate the potential causal effects of plasma protein levels on diabetic retinopathy and to explore intermediary biological mechanisms. Exposure data were obtained from the UK Biobank Pharma Proteomics Project (UKB-PPP), which profiled 2,923 plasma proteins in 54,219 participants and identified 14,287 primary genetic associations, 85% of which were novel (15). These high-confidence protein quantitative trait loci (pQTLs) served as genetic instruments for the MR analysis. Outcome data for diabetic retinopathy were derived from the FinnGen Release 12 GWAS, including 96,429 individuals (14,142 cases and 82,287 controls) of Finnish ancestry (16). To further explore potential mediators, we conducted a mediation analysis using immune cell data from the SardiNIA cohort (17), which includes 3,757 individuals with immune profiling.

For MR analysis, IVs were selected based on three core assumptions: (1) they are strongly associated with the exposure (e.g., LAT2 expression); (2) they are not related to potential confounders; and (3) they affect the outcome (e.g., diabetic retinopathy) solely through the exposure (18, 19). For mediation MR, additional criteria were applied: the IV must influence the outcome only via the exposure and mediator, with no alternative pathways or unmeasured confounding between mediator and outcome (20, 21). SNPs were initially filtered using a genome-wide significance threshold of p < 5 × 10⁻⁸. In cases where fewer than 10 eligible SNPs were available, a relaxed threshold of p < 1 × 10⁻⁵ was adopted to ensure sufficient instrument strength. To ensure independence among SNPs, linkage disequilibrium (LD) pruning was performed using r² < 0.001 within a 10,000 kb window (22, 23). SNPs with an F-statistic < 10 were excluded to avoid weak instrument bias (24). All SNP harmonization, clumping, and F-statistic calculations were performed using the TwoSampleMR package in R (16).

We performed two-sample MR analysis to assess the causal effect of circulating protein levels on DR. The plasma pQTLs served as IVs. MR analyses were conducted using the R package TwoSampleMR (25, 26). For exposures with a single SNP, the Wald ratio method was applied, whereas for exposures with ≥2 SNPs, the inverse variance weighted (IVW) method was used. To account for multiple testing, P were adjusted using the false discovery rate (FDR) approach, and associations with FDR < 0.05 were considered significant.

To validate whether the genetic signal for LAT2 protein levels shared a common causal variant with DR, we conducted colocalization analysis using cis-pQTL SNPs located within ±1 Mb of the LAT2 gene region (11, 27). Posterior probabilities were calculated for five hypotheses (H0–H4), where H4 represents a shared causal variant between protein expression and disease. Colocalization was considered significant when the posterior probability for H4 (PP.H4) exceeded 0.8. Analyses were performed using the coloc package in R, based on summary statistics from GWAS and proteomic datasets.

To evaluate the potential broader effects of LAT2 expression across multiple phenotypes, we performed a phenome-wide MR (MR-PheWAS) analysis (28, 29). LAT2, identified as a candidate protective factor against diabetic retinopathy in the primary MR analysis, was selected as the exposure. The instrumental variables used for LAT2 were consistent with those applied in prior MR analyses. Summary-level outcome data were obtained from the Finngen R12 release (https://r12.finngen.fi/). The IVW or Wald ratio method was applied depending on the number of available SNPs. P values were corrected using the FDR method, and associations with Pfdr < 0.05 were considered statistically significant.

All MR analyses were conducted using the TwoSampleMR package in R (v4.3.3) (30). The IVW method was used as the primary analysis, and the Wald ratio was applied for single-SNP exposures (31–33). Sensitivity analyses included Cochran’s Q test for heterogeneity, MR-Egger intercept for pleiotropy, and leave-one-out analysis to assess the influence of individual SNPs (34, 35). In mediation analysis, a stepwise MR approach was used to evaluate the indirect effect of immune cells in the pathway from LAT2 to diabetic retinopathy. FDR correction was applied, and Pfdr < 0.05 was considered significant.

We conducted bidirectional two-sample MR analyses to evaluate the causal relationship between plasma protein levels and DR. The forward MR ( Figure 2A ) examined the effects of genetically predicted protein levels on DR risk, while the reverse MR ( Figure 2B ) tested whether genetic liability to DR affected protein expression levels. The IVW method was used as the primary method and was tested for heterogeneity and horizontal polytropy ( Supplementary Tables S1 , S2 ).

In the forward MR analysis, five proteins—Coagulation factor XIII B chain, Epithelial discoidin domain-containing receptor 1, Complement factor H-related protein 2, Protein PBMUCL2, and the newly added Linker for activation of T-cells family member 2 (LAT2), showed significant causal associations with DR. For example, higher genetically predicted levels of Epithelial discoidin domain-containing receptor 1 were associated with a reduced risk of DR (IVW: OR = 0.785, 95% CI: 0.726–0.849, p < 0.001), while increased levels of Coagulation factor XIII B chain were associated with higher DR risk (IVW: OR = 1.248, 95% CI: 1.162–1.340, p < 0.001). Similar significant associations were observed for Complement factor H-related protein 2 and PBMUCL2. For LAT2, only one instrumental SNP passed the genome-wide significance threshold; thus, the Wald ratio method was used. The result showed that elevated LAT2 levels were significantly associated with reduced DR risk (OR = 0.358, 95% CI: 0.215–0.597, p < 0.001), suggesting a potential protective effect of this immune-related protein.

The reverse MR analysis revealed no significant effects of DR on protein levels, including LAT2, supporting a unidirectional association. This step served to validate the directionality of the observed effects.

Building upon the protein-level associations identified in Figure 2 , we performed follow-up analyses to validate whether the observed MR signals could be supported by genetic colocalization and gene expression–based MR evidence. We systematically examined colocalization signals for the forward MR-identified proteins, but found that only LAT2 showed a high posterior probability of sharing a causal variant between protein levels and DR. Specifically, colocalization analysis for the LAT2 region ( Figure 3A ) revealed a strong posterior probability for hypothesis 4 (PP.H4 = 0.8546), suggesting a shared genetic signal. The lead SNP rs34200032 showed consistent effects in both the plasma proteomics and DR GWAS datasets.

To further assess whether genetically regulated expression of LAT2 contributes to DR risk, we conducted a SMR analysis ( Figure 3B ), which revealed a significant inverse association (b = –1.026, SE = 0.318, p = 0.00125), corresponding to an odds ratio of 0.36 (95% CI: 0.19–0.67). The lead SNP rs34200032 was significantly associated with LAT2 expression (eQTL p = 1.98 × 10⁻⁸) and DR risk (GWAS p = 8.04 × 10⁻⁵), further supporting its functional relevance.

To replicate this finding, we performed an eQTL-based two-sample MR using genome-wide significant variants (p < 5 × 10⁻⁸), a GWAS threshold of p < 0.001, and a 10,000 kb clumping window ( Figure 3C ). The results were consistent with the SMR analysis, again showing a protective effect of LAT2 expression on DR risk (IVW: p = 0.00051; OR = 0.799, 95% CI: 0.704–0.907).

To validate the findings from the MR and colocalization analyses, we next evaluated the expression and functional role of LAT2 in cellular models of DR. Based on previous results suggesting that LAT2 exerts a protective effect against DR, we hypothesized that its expression might be altered under hyperglycemic conditions.

As shown in Figure 4A , CCK-8 assays revealed that high glucose (HG) exposure significantly reduced cell viability in both hRMECs and ARPE-19 cells compared to their respective normal glucose (NC) controls (p < 0.05 or p < 0.01). This confirms the cytotoxic impact of hyperglycemia in retinal endothelial and epithelial cells. Interestingly, PCR analysis ( Figure 4B ) showed a significant upregulation of LAT2 mRNA levels under HG treatment in both hRMECs and ARPE-19 cells (p < 0.05), with a more pronounced elevation observed in ARPE-19. These results indicate a possible compensatory increase in LAT2 expression in response to glucose-induced stress, which aligns with the MR findings that suggest a protective effect of elevated LAT2 levels on DR risk.

Building on our findings that LAT2 is a protective factor against DR and upregulated under high-glucose conditions ( Figure 4 ), we further explored its potential immunological mediators. Given the role of immune dysregulation in DR, we hypothesized that LAT2 might exert its protective effect partly via B-cell–related mechanisms.

To test this, we first performed MR analysis of immune cell traits on DR risk ( Figure 5A ). Among the tested traits, CD27⁺ switched memory B cells, CD27⁺ memory B cells, and CD27⁺ IgD⁻CD38^dim B cells showed nominally significant associations. Notably, higher proportions of CD27⁺ switched memory B cells were significantly associated with reduced DR risk (IVW: OR = 0.951, 95% CI: 0.924–0.979, p < 0.001).

We then applied a three-step MR-based mediation framework ( Figure 5B ). LAT2 was significantly associated with both DR risk (Wald ratio: OR = 0.358, 95% CI: 0.215–0.597, p < 0.001) and CD27⁺ switched memory B cell abundance (OR = 3.528, 95% CI: 1.124–11.075, p = 0.031). The B cell subset was, in turn, inversely associated with DR (OR = 0.951, 95% CI: 0.924–0.979, p < 0.001). The estimated indirect effect was –0.0631 (p = 0.0474), corresponding to a mediation proportion of 6.2%. The negative sign of the indirect effect reflects consistent directionality: increased LAT2 leads to increased protective B cells, which lowers DR risk. This supports the interpretation of a biologically meaningful partial mediation.

Together, these findings suggest that the protective effect of LAT2 against DR may be partly mediated by CD27⁺ switched memory B cells, highlighting the importance of adaptive immune pathways in the pathophysiology of diabetic retinopathy.

To further investigate the phenotypic specificity of LAT2’s protective effect, we performed a MR-PheWAS analysis across a broad range of disease traits. As shown in Figure 6 , LAT2 expression exhibited significant associations with a limited number of phenotypes. Notably, diabetic retinopathy was among the top significant traits (p < 1×10⁻⁴), consistent with previous MR, colocalization, and eQTL-based results. In addition, we identified a potential association with restless leg syndrome, although its significance and biological plausibility require further investigation.