2.1 Study design
The MR structure is based on following fundamental assumptions: (1) The IVs are connected to exposure; (2) They are unrelated to the outcome via confounding pathway; and (3) They have no direct impact on the outcome, their influence is potentially only indirect through the exposure factor. AP was the exposure factor, while SNPs were IVs. OC and six clinically common phenotypes (HGSC, LGSC, EnOC, CCC, and IMOC) were examined as outcomes. TSMR analysis was conducted to evaluate the causal effects between AP and OC. Additionally, MVMR analyses, considering confounders (obesity, smoking, and alcohol) and mediators, were performed to investigate the proximal causality (Fig. 1).
2.2 Data sources
AP served as the exposure variable. Data were obtained from the UK Biobank. This study offers publicly available information on phenotypic traits, genetic information, and genome-wide genetic analysis (11,12). AP data for European and East Asian populations were sourced from GWAS summary databases. In European populations, the GWAS summary datasets for PM2.5 (ukb-b-10817), PM2.5-10 (ukb-b-12963), PM10 (ukb-b-589), NO2 (ukb-b-2618), and NOx (ukb-b-12417) included 423,796, 423,796, 455,314, 456,380, and 456,380 participants, respectively. For East Asian populations, the datasets for PM2.5 (ukb-e-24006_EAS), PM2.5-10 (ukb-e-24008_EAS), PM10 (ukb-e-24005_EAS), NO2 (ukb-e-24016_EAS), and NOx (ukb-e-24004_EAS) included 2,505, 2,505,2,505, 2,625, and 2,625 participants, respectively (Table 1).
Table 1
Summary of the genome-wide association studies (GWAS)included in this two-sample MR study
Exposures/ outcomes | Dataset | Number of SNPs | Populatior | Consortium | Sex | Year |
Particulate matter (PM) | PM2.5 um | ukb-b-10,817 | 9851867 | European | MRC-IEU | Males and Females | 2018 |
PM2.5 um | ukb-e-24006_EAS | 8268350 | Asian (East Asia) | NA | Males and Females | 2020 |
PM2.5-10 um | ukb-b-12,963 | 9851867 | European | MRC-IEU | Males and Females | 2018 |
PM2.5-10 um | ukb-e-24008_EAS | 8268350 | Asian (East Asia) | NA | Males and Females | 2020 |
PM10 um | ukb-b-589 | 9851867 | European | MRC-IEU | Males and Females | 2018 |
PM10 um | ukb-e-24005_EAS | 8268350 | Asian (East Asia) | NA | Males and Females | 2020 |
Nitrogen dioxide | ukb-b-2,618 | 9851867 | European | MRC-IEU | Males and Females | 2018 |
Nitrogen dioxide | ukb-e-24016 EAS | 8268350 | Asian (East Asia) | NA | Males and Females | 2020 |
Nitrogen oxides | ukb-b-12,417 | 9851867 | European | MRC-IEU | Males and Females | 2018 |
Nitrogen oxides | ukb-e-24004_EAS | 8260777 | Asian(East Asia) | NA | Males and Females | 2020 |
ovaian cancer | leu-a-1120 | - | European | OCAC | Females | 2017 |
ovanan cancer | ebi-a-GCST90018668 | 12413202 | Asian(East Asia) | NA | Females | 2021 |
Endometrioid ovarian cancer | leu-a-1125 | - | European | OCAC | Females | 2017 |
High-grade serous | leu-a-1121 | - | European | OCAC | Females | 2017 |
Low grade serous | leu-a-1122 | - | European | OCAC | Females | 2017 |
Clear cell | leu-a-1124 | - | European | OCAC | Females | 2017 |
Invasive mucinous | leu-a-1123 | - | European | OCAC | Females | 2017 |
PM:Particulate matter;MR:Mendelian randomization;GWAS:Genome-wide association studies;SNPs:Single nucleotide polymorphisms;NA:No data |
OC served as the outcome.All GWAS data for ovarian cancer were obtained from the IEU open GWAS project (European population and East Asian population).GWAS summary dataset of OC in East Asian populations (ebi-a-GCST90018668) included 61,457 individuals (comprising 843 cases and 60,614controls) (13). Subgroup analyses were conducted with 13,037 high-grade serous, 1,012 low-grade serous, 1,366 clear cell, 2,810 endometrioid, and 1,417 invasive mucinous OC samples. Our study exclusively relied on published GWAS findings.
GWAS data such as BMI, alcohol consumption, and smoking initiation, were also obtained from the UK Biobank (12,14,15).
2.3 Selection criteria of instrumental
The significance threshold for exposure factors was set at p < 5 × 10− 8. Nevertheless, only PM2.5 (ukb-b-10817), PM10 (ukb-b-589), NO2 (ukb-b-2,618), and NOx (ukb-b-12,417) could identify a sufficient number of SNPs. For other exposures, we established a p-value threshold of 5e-6 to identify robust IVs based on the screening criterion for linear regression of genetic mutation on risk variables, ensuring unlikelihood of weak IV bias (16,17).
To eliminate SNPs in LD, we applied criteria of r2 < 0.001 and kb > 10,000 during IVs collection. Subsequently, we removed palindromic SNPs to guarantee that the allele's impact on exposure aligns with its effect on the outcome.
Finally, we computed the R² and F statistic for each SNP (18,19). R² quantifies the proportion of iron status variance elucidated by each SNP, while the F statistic is employed to evaluate the potential for weak IV bias. The F statistic for each selected SNP exceeded 10, indicating that genetic instruments chosen robustly predict exposure (20).
2.4 MR analysis
We assessed the cause-and-effect connection by the IVW approach. We also utilized alternative methods known for their robust estimation of horizontal pleiotropy. Subsequently, we opted for a fixed-effects model to provide a reliable estimation. This method is widely used in MR analysis (20,21). We also conducted MR regression, weighted median, weighted mode, and simple mode to further bolster the reliability and accuracy of our research findings (22).
2.5 Sensitivity analysis
Firstly, the leave-one-out method was utilized (23). Next, MR-PRESSO was employed to identify and adjust for any outliers. Cochran’s Q statistic (P < 0.1) was employed to assess heterogeneity (24). Thirdly, MR-Egger regression and MR-PRESSO global tests were employed to confirm that the IVs do not affect the likelihood of OC. Meanwhile, MR-PRESSO facilitates a systematic evaluation of pleiotropy (25). p > 0.05 represents the statistical criterion for ensuring pleiotropy absence in IVs.
2.6 MVMR
MVMR permits the examination of multiple exposures’ causal effects on the estimated outcomes (26,27), contingent upon the covariance between the impacts of IVs on included exposure (28,29). IVs are correlated with multiple risk factors, but they should satisfy the same assumptions as any other IVs (30). MVMR analysis solely integrates single AP type and an extra risk factor for each analysis. In this research, indirect effects were not calculated (31).
2.7 Statistical analysis
The “TwoSampleMR” (32) and “MR-PRESSO” (33) packages in R were utilized for analyses. The level of significance was determined to be p < 0.05.