Univariable MR results of early AMD

We conducted univariable MR analyses for early AMD, summarizing the results and providing visualizations in Fig. 2. Detailed information on the number of SNPs, P-values, and ORs with 95% confidence intervals (CIs) is presented in Table 2. The SNPs ranged from three to 792, with explained variances spanning from 0.035% to 16.139%. Notably, the F-statistics for all considered traits exceeded 10, ranging from 36.164 to 191.577, indicating the absence of potential weak instrument bias.

Fig. 2

Univariable Mendelian randomization method showing causal effects of lifestyle factors, overall pigment, circulating parameters, and metabolic comorbidities on early AMD. All results described here can be found in Table 2. FA, fatty acids; LDL-C, low-density lipoprotein cholesterol; HDL-C, high-density lipoprotein cholesterol; IVW, inverse-variance–weighted; WM, weighted median; SNP, single nucleotide polymorphism. Red means OR > 1 and P < 0.0017. Pink means OR > 1 and 0.0017 < P < 0.05. Blue means OR < 1 and P < 0.0017. Light-blue means OR < 1 and 0.0017 < P < 0.05. White means P > 0.05

Lifestyle factors for the risk of early AMD

Regarding lifestyle factors, our findings (Fig. 2 and Table 2) revealed suggestive associations between genetically predicted “dairy smoothie intake”, “morning/evening person”, and an increased risk of early AMD, with ORs of 2.681 (95% CI 1.008–7.130, P = 4.811 × 10−2) and 1.202 (95% CI 1.030–1.402, P = 1.954 × 10−2) by the IVW method, respectively. While "use of sun/UV protection" was suggestively associated with a decreased risk of early AMD (OR = 0.682, 95% CI 0.490–0.948, P = 2.560 × 10−2) only by MR-Egger method.

Additionally, genetically predicted “never eat wheat products” was significantly associated with the increased risk of early AMD with ORs of 23.853 (95% CI 2.731–208.323, P = 4.122 × 10−3). Besides, we observed possible heterogeneity for “age started wearing glasses or contact lenses” (P_heterogeneity = 1.466 × 10–4). Thus, we performed the IVW (multiplicative random effects) model, and the relationship remained stable in this method. This result indicated that “age started wearing glasses or contact lenses” is a significant indicator of an increased risk of early AMD (OR = 1.605, 95% CI 1.269–2.030, P = 7.969 × 10−5).

Nevertheless, no significant causal association was observed between genetically predicted “fresh fruit”, “fish oil/cod liver oil”, “cigarettes per day”, and “Sleeplessness/insomnia” and early AMD.

Overall pigment level for the risk of early AMD

Regarding overall pigment level (Fig. 2 and Table 2), genetically predicted black hair color exhibited suggestive associations with an increased risk of early AMD (OR = 1.841, 95% CI 1.105–3.068, P = 1.92 × 10−2) by IVW method. Moreover, the significant association between dark brown hair color and the risk of early AMD was only observed by the MR-Egger method (OR = 1.281, 95% CI 1.064–1.543, P = 9.608 × 10−3).

The causal effect of light brown hair color on early AMD was verified in the IVW (multiplicative random effects) model (OR = 0.791, 95% CI 0.611–1.024, P = 7.557 × 10−2) and MR-Egger method (OR = 0.618, 95% CI 0.434–0.880, P = 9.477 × 10−3), but heterogeneity (P_heterogeneity = 1.047 × 10−2) and pleiotropy (P_pleiotropy = 5.350 × 10−2) were observed.

Furthermore, our IVW analyses provided strong evidence of liability towards early AMD based on the skin color trait (OR = 1.190, 95% CI 1.076–1.317, P = 7.041 × 10−4).

Circulating parameters for the risk of early AMD

We performed univariable Mendelian randomization for each serum lipid trait (Fig. 2 and Table 2). We used the IVW (multiplicative random effects) method and found that five serum lipid traits were significantly causally associated with early AMD risk, although possible heterogeneities exist (all five P_heterogeneity < 0.05). We observed that for a one-SD increase in high cholesterol, TG, HDL-C, LDL-C, ApoA1), and ApoB, the ORs of early AMD was 0.392 (95% CI 0.205–0.750, P = 4.682 × 10−3), 0.784 (95% CI 0.734–0.837, P = 3.661 × 10−13), 1.218 (95% CI 1.140–1.303, P = 6.842 × 10−9), 0.835 (95% CI 0.770–0.906, P = 1.362 × 10−5), 1.146 (95% CI 1.067–1.231, P = 1.814 × 10−4), and 0.843 (95% CI 0.788–0.901, P = 5.194 × 10−7), respectively.

Regarding circulating fatty acids levels, we observed possible heterogeneities for SFA, MUFA, and PUFA (all P_ heterogeneity < 0.05). The associations were estimated by the IVW (multiplicative random effects) method and genetically predicted higher SFA, MUFA, and PUFA levels were significantly associated with a decreased risk of early AMD, with the ORs (95%CI) of 0.772 (0.698–0.855, P = 5.909 × 10−7), 0.776 (0.706–0.852, P = 1.212 × 10−7), and 0.877 (0.798–0.963, P = 5.823 × 10−3), respectively. With respect to total FA, all five MR methods evidenced a significant association with early AMD, but possible pleiotropy was observed (P_pleiotropy = 2.705 × 10−2). No significant causal association was found between genetically estimated iron status, C-reactive protein level, and early AMD analyzed by all five MR methods.

Metabolic comorbidities for the risk of early AMD

Genetically instrumented metabolic-related factors, including BMI, diabetes, high blood pressure, and cardiovascular diseases, could not increase the risk for early AMD by all five MR methods. The ORs (95%CI) of the IVW were 1.000 (0.923–1.083, P = 9.922 × 10−3), 1.003 (0.356–2.825, P = 9.950 × 10−3), 1.047 (0.802–1.366, P = 7.382 × 10−3), and 1.024 (0.833–1.260, P = 0.8203) for a 1-SD increase, respectively.

Multivariable MR analysis of early AMD

We conducted multivariable MR analyses (MVMR-IVW method) to estimate the direct effects of serum lipid biomarkers on early AMD risk conditional on other serum lipid biomarkers and confounders. We further selected independent SNPs with other serum lipid biomarkers and confounders as instrumental variables. 107, 166, 146, and 39 SNPs related to ApoB, HDL-C, TG, and high cholesterol, four serum lipid biomarkers regressed by LASSO Cox, used as the genetic instruments (Fig. 3). For diabetes mellitus, high blood pressure, and BMI, we selected 28, 73, and 198 SNPs as the genetic instruments. We found that the associations of HDL-C (OR = 1.187, 95% CI 1.064–1.324, P = 2.138 × 10−3) and TG (OR = 0.838, 95% CI 0.738–0.950, P = 5.962 × 10−3) with early AMD risk were essentially unchanged in multivariable MR analyses compared with univariable MR analysis.

Fig. 3

Multivariable Mendelian randomization method showing causal effects of HDL-C and TG on early AMD. SNP, single nucleotide polymorphism; BMI, Body mass index; HDL-C, high-density lipoprotein cholesterol; OR, Odd ratio; CI, Confidence interval

Validation of the causal effects of TG and HDL-C on early AMD

To corroborate the reliability of the results about the causal effects of TG and HDL-C on early AMD, we further replicated MR analysis based on other different GWAS datasets about “Total triglycerides (GWASID: met-d-Total_TG)” and “HDL cholesterol (GWASID: met-d-HDL_C)” (Table 1). Here, 92 SNPs related to total triglycerides and 133 SNPs related to HDL cholesterol were included in this validation analysis, respectively. Consistently, we noticed that the significant associations of TG (OR = 0.795, 95% CI 0.732–0.864, P = 5.325 × 10−8) and HDL-C (OR = 1.201, 95% CI 1.101–1.310, P = 3.606 × 10−5) with early AMD risk were significant using the IVW (multiplicative random effects) method respectively, although possible heterogeneities were observed (P_heterogeneity < 0.05) (Table 2).

Furthermore, we performed a leave-one-out sensitivity test for the above risk factors, including “never eat wheat products”, “dairy smoothie intake”, “chronotype”, “age started wearing glasses/contact lenses”, “use of sun/UV protection”, “black hair color”, “dark brown hair colour”, “light brown hair color”, “skin colour”, “high cholesterol”, “TG”, “HDL-C”, “DLD-C”, “ApoA1”, “ApoB”, “total FA”, “SFA”, “MUFA”, and “PUFA”. We eliminated SNPs of each risk factor one by one and calculated the MR results for the remaining SNPs. All of them showed stability of the results (Fig. S1A–U).