MR employs genetic variants as instrumental proxies to assess causal relationships between exposures and outcomes. This method is based on three fundamental assumptions: instrument relevance, instrument independence, and exclusion-restriction criterion [15]. In accordance with these principles, our study employed an MR framework to investigate the bidirectional causal effects of mitochondrial proteins on the risk of POI, as illustrated in Fig. 1. This approach allows for the examination of the complex interplay between genetic and environmental factors, providing a more nuanced understanding of how mitochondrial proteins may influence POI development and offering insights that could guide future interventions.

Fig. 1.

Study design for MR analysis. The analysis is based on three core hypotheses: (1) a strong association between IVs and exposure factors, (2) no association between IVs and confounders, and (3) IVs influence the outcome solely through the exposure factors and not through other pathways. In this study, genome-wide statistical significance was as p 5 $\times$ 10^-6. Linkage disequilibrium (LD) was removed using the criteria of r² $>$0.001 and kb = 10,000. Furthermore, for the three main assumptions in the figure, the dotted lines represent impacts that should be avoided, while the solid lines represent assumptions that must be met. MR, Mendelian randomization; IVs, instrumental variables; IVW, inverse-variance weighted; NFKB1, Nuclear Factor Kappa B Subunit 1; NADH, nicotinamide adenine dinucleotide; SNPs, single nucleotide polymorphisms.

To investigate mitochondrial proteins, plasma protein data for 3622 proteins from 3301 healthy participants in the INTERVAL (https://www.intervalstudy.org.uk/) study were analyzed. The INTERVAL study is a prospective cohort of predominantly European-ancestry blood donors in the United Kingdom, as previously described [13]. This dataset includes 1927 genotype-protein associations (pQTLs), encompassing trans-associated loci for 1104 proteins [16]. This study provides new insights into the genetic regulation of protein expression. Subsequently, 66 datasets corresponding to mitochondrial proteins were screened. In parallel, POI was examined using the dataset labeled finn-b-E4_OVARFAIL, sourced from the Finnish FinnGen database (https://r8.risteys.finngen.fi/phenocode/E4_OVARFAIL), which includes data collected up to 2021 [17]. The analysis included 429,209 individuals of Finnish ancestry. After applying sex-specific filters (females only), the effective sample size was 239,906 individuals, comprising 542 POI cases and 239,364 controls. POI cases were identified based on nationwide health registries, including hospital discharge records, cause-of-death data, and drug reimbursement records, based on relevant International Classification of Diseases (ICD) codes (ICD-10: E28.3). The use of this dataset guarantees that our investigation focuses on the population most affected by POI, thus enhancing the relevance of the findings.

To further satisfy the MR independence and exclusion-restriction assumptions, each single nucleotide polymorphism (SNP) and its high-linkage disequilibrium (LD) proxies were systematically screened for prior genome-wide significant associations with potential confounders or traits related to the outcome. Specifically, for every IV retained after clumping and F-statistic filtering, we queried LDlinkR (LDtrait function) using European reference populations, defining high-LD proxies as those with r² $\ge$0.80. We then removed the original instrument if either the SNP itself or any high-LD proxy showed a genome-wide association study (GWAS) catalog association at p $\le$ 5 $\times$ 10^-8 with prespecified trait domains that could plausibly introduce horizontal pleiotropy with POI and mitochondrial proteins. Harmonization and subsequent MR analyses were performed on the filtered instrument sets using TwoSampleMR (version 0.6.17). The package was developed by the Medical Research Council (MRC) Integrative Epidemiology Unit at the University of Bristol and is publicly available from the Comprehensive R Archive Network (CRAN).

SNPs associated with mitochondrial protein levels were selected as IVs through rigorous screening. To ensure an adequate number of IVs while capturing relevant genetic variants associated with mitochondrial protein levels, IVs were required to show a genome-wide significant association with the exposure (p 5 $\times$ 10^-6), as described in previous studies [18, 19]. To guarantee independence and minimize confounding from LD, SNPs with r² $\ge$0.001 within 10,000 kb were excluded.

Furthermore, palindromic and proxy SNPs were removed to prevent inaccuracies in the MR analysis. To mitigate potential weak instrument bias, an F-statistic threshold of $\ge$10 was set, calculated as [(R² $\times$ (N–2))/(1–R²)], where R² is the variance in the exposure explained by the SNPs and N is the sample size [20]. These measures ensured the precision and reliability of the MR results [21].

Statistical analyses were conducted using R software (version 4.2.2; R Foundation for Statistical Computing, Vienna, Austria). The following R packages were used: ggplot2 (version 3.4.0; tidyverse), LDlinkR (version 3.1.3; National Cancer Institute, Division of Cancer Epidemiology & Genetics, Bethesda, MD, USA), MendelianRandomization (version 0.5.1), MRPRESSO (version 1.0; Broad Institute of MIT and Harvard, Cambridge, MA, USA), and TwoSampleMR (version 0.6.17; MRC Integrative Epidemiology Unit, University of Bristol, Bristol, UK). Analyses were repeated after excluding outlier SNPs (Supplementary Tables 1,2). The primary method used was the Inverse Variance Weighted (IVW) approach, which is known for its robustness in the absence of pleiotropy [22]. To address potential biases from horizontal pleiotropy, several sensitivity analyses were performed, including MR-PRESSO, MR-Egger, and Weighted Median regression. The MR-Egger intercept p-value $>$ 0.05 indicated no significant directional pleiotropy, reinforcing the validity of our causal estimates [23]. A two-tailed test was used for all statistical analyses, with a significance threshold of p 0.05. Heterogeneity among the causal estimates of individual SNPs was assessed using Cochran’s Q test, with a Q-statistic p 0.05 considered indicative of significant heterogeneity. In addition, the results of MR analyses were adjusted using the Benjamini-Hochberg false discovery rate (BH-FDR) correction to control for multiple testing. Given the exploratory nature of this study, a threshold of p 0.05 was still used to define nominal statistical significance.

MR analyses using the IVW method revealed significant associations between specific mitochondrial proteins and POI risk (Figs. 2,3). Among the 66 mitochondrial proteins tested, 4 demonstrated significant causal associations with POI risk. Coiled-coil domain-containing protein 90B was positively associated with POI risk (odds ratio [OR] = 1.80, 95% confidence interval [CI]: 1.02–3.19, p = 0.042), indicating increased risk. In contrast, 3 proteins demonstrated protective associations. For 39S ribosomal protein L14 (mitochondrial), the IVW method revealed a significant inverse association with POI (OR = 0.40, 95% CI: 0.20–0.80, p = 0.009), suggesting a protective effect. This association, with a beta ($\beta$) value of –0.905, remained robust across sensitivity analyses. Specifically, Cochran’s Q test indicated no significant heterogeneity (Q = 1.772, p = 0.778), and both the MR-Egger intercept test (p = 0.844) and the MR-PRESSO global test (p = 0.816) provided no strong evidence of directional pleiotropy, reinforcing the reliability of the causal estimate. These findings indicate that higher levels of 39S ribosomal protein L14 are associated with a reduced risk of developing POI.

Fig. 2.

Preliminary MR estimates of the effects of mitochondrial proteins on POI. From the inner to outer circles, the estimates represent: Beta-value, Q p-value, MRPRESSO global test p-value, Egger intercept p-value, WM p-value, MR-Egger p-value, and IVW p-value, respectively. The color intensity reflects the magnitude of the p-value. POI, primary ovarian insufficiency; WM, weighted median.

Fig. 3.

Forest plots illustrating the causal relationship between mitochondrial proteins and POI. Significant results (p 0.05) from the IVW method are displayed, as identified in the forward MR analysis. p.adj values represent p-values adjusted for multiple comparisons using the BH-FDR correction across all assessed proteins. 3 proteins (39S ribosomal protein L14, oligoribonuclease, and mitochondrial fission regulator 1) showed significant protective effects against POI, whereas 1 protein (Coiled-coil domain-containing protein 90B) showed a positive association with POI risk (IVW p 0.05). BH-FDR, Benjamini-Hochberg false discovery rate.

Similarly, oligoribonuclease (mitochondrial) was significantly associated with POI, with the IVW analysis yielding an OR of 0.64 (95% CI: 0.43–0.96, p = 0.031), suggesting a protective effect. Finally, mitochondrial fission regulator 1 showed a significant inverse association with POI using the IVW method (OR = 0.60, 95% CI: 0.36–0.99, p = 0.044), indicating that this mitochondrial protein may also reduce the risk of POI.

In summary, the IVW analysis consistently suggests that higher levels of 39S ribosomal protein L14, oligoribonuclease, and mitochondrial fission regulator 1 are associated with a reduced risk of POI, whereas higher levels of coiled-coil domain-containing protein 90B may increase the risk. However, after applying the BH-FDR correction, the strength of the associations was attenuated. Therefore, these findings shouldbe interpreted with caution. No significant heterogeneity, pleiotropy, reverse causality, or outlier IVs were detected for the effect of mitochondrial proteins on POI (Supplementary Table 1). Scatter, funnel, and forest plots illustrating these associations and sensitivity analyses are provided in Supplementary Fig. 1.

The impact of POI on several mitochondrial proteins was evaluated using the IVW method (Figs. 4,5). While associations were observed, the effect sizes were consistently small, suggesting a limited biological impact. For apoptosis-inducing factor 1 (mitochondrial), a significant negative association was found ($\beta$ = –0.03, 95% CI: –0.05 to –0.01, p = 0.007), although the small beta value indicates only a modest effect. Similarly, nicotinamide adenine dinucleotide (NADH) dehydrogenase [ubiquinone] iron-sulfur protein 4 exhibited a positive association ($\beta$ = 0.02, 95% CI: 0.00–0.04, p = 0.028), although the effect size remained minimal. For 39S ribosomal protein L14, a significant negative association was identified ($\beta$ = –0.03, 95% CI: –0.05 to –0.01, p = 0.011), yet the small effect size limits its biological significance. The association remained robust in sensitivity analyses, with no significant heterogeneity (Cochran’s Q p = 0.113) and no evidence of horizontal pleiotropy, as indicated by the MR-Egger intercept test (p = 0.350) and the MR-PRESSO global test (p = 0.145). Similarly, mitochondrial ubiquitin ligase activator of NFKB 1 demonstrated a weak association ($\beta$ = –0.02, 95% CI: –0.04 to –0.00, p = 0.031), further emphasizing the overall modest impact of POI on mitochondrial protein expression. The MR analysis examining the association of POI with mitochondrial proteins showed no significant evidence of heterogeneity or horizontal pleiotropy, supporting the robustness of the causal estimate (Supplementary Table 2). However, after applying the BH-FDR correction, the statistical significance of these associations was attenuated. Supporting plots illustrating these associations and the sensitivity analyses are shown in Supplementary Fig. 2.

Fig. 4.

Preliminary MR estimates of the effect of POI on mitochondrial proteins. From the inner to outer circles, the estimates represent: Beta-value, Q p-value, MRPRESSO global test p-value, Egger intercept p-value, WM p-value, MR-Egger p-value, and IVW p-value, respectively. The color intensity reflects the magnitude of the p-value.

Fig. 5.

Forest plots illustrating the causal relationship between POI and mitochondrial proteins. Significant results (p 0.05) from the IVW method are displayed, as identified in the inverse MR analysis. The p.adj values represent p-values adjusted for multiple comparisons using the BH-FDR correction across all assessed proteins. The result demonstrated the significant associations between POI and the 4 mitochondrial proteins (Apoptosis-inducing factor 1, NADH dehydrogenase iron-sulfur protein 4, 39S ribosomal protein L14, and Mitochondrial ubiquitin ligase activator of NFKB 1). BH-FDR, Benjamini-Hochberg false discovery rate; NADH, nicotinamide adenine dinucleotide.

Our analyses identified several drugs potentially involved in the regulation of mitochondrial proteins, focusing on their interactions with three protected genes: RNASEH1, MTFR1 and MRPL14. The principal findings of this prediction are summarized in Table 1, which lists the drugs along with their statistical significance, effect size, combined score, and associated cytokine targets [24]. This table highlights drugs that may modulate mitochondrial protein regulation, suggesting potential therapeutic applications in POI.

Although this study has several strengths, it is important to acknowledge the limitations inherent in its research design. First, the use of genetic data from European populations may limit the generalizability of our findings to other ethnic groups. It would be beneficial for future studies to include more diverse populations to enhance the applicability of these results. Second, although MR is a robust method for inferring causality, it relies on the assumption that the selected genetic variants are valid IVs. Despite efforts to mitigate potential pleiotropy, it is not possible to rule out the possibility of residual confounding. Third, although none of the associations remained statistically significant after FDR correction, the overall pattern of nominal associations suggests that mitochondrial health is closely linked to POI and may still reflect biologically relevant relationships. Finally, while our study identified an association between mitochondrial proteins and POI, it did not elucidate the underlying molecular mechanisms. Further functional studies are required to determine how these mitochondrial proteins influence ovarian biology.