Introduction

Osteoporosis, a skeletal disease, is marked by bone tissue’s microarchitectural degeneration and a reduction in bone mineral density (BMD), leading to heightened bone fragility and an increased risk of fractures1. Fractures secondary to osteoporosis are prevalent among older adults, affecting women over 55 and men over 65, and entail substantial morbidity, mortality, and considerable healthcare costs associated with their management2. BMD, a main contributor to osteoporosis, is known to be responsive to mechanical loading, with its variation attributable to not only environmental such as oxidative stress3 and inflammation4 and but also genetic factors such as family history of osteoporosis5 and copy number variation (CNV) of the human genome6, notably, genetic components explain a significant portion of the normal variance in bone mass, estimated to range from 50 to 85%2. Genome-wide association studies (GWAS) have uncovered approximately 100 genomic loci linked to bone density and related traits influencing bone strength and fracture susceptibility, however, the causal pathways for a majority of these loci remain to be elucidated7.

Mitochondria are pivotal in energy metabolism, primarily through the electron transport chain, a process that inherently generates reactive oxygen species (ROS) as a consequence8. Notably, mitochondrial DNA (mtDNA) is recognized for its heightened susceptibility to assault by free radicals8. Animal studies have demonstrated that elevated levels of mtDNA mutations in somatic cells of mtDNA-mutator mice precipitate the early manifestation of multiple age-associated phenotypes, including osteoporosis, alopecia, and decreased body weight9. In addition, a particular deletion in mtDNA has been reported to be linked to heightened oxidative stress observed in peripheral blood mononuclear cells (PBMCs) of men diagnosed with advanced osteoporosis10. Variants in mtDNA within PBMCs have been implicated in the pathogenesis of osteoporosis11. These findings collectively imply a potential connection between osteoporosis and dysfunctions in mitochondrial activity. Notably, research exploring the causal correlation between mitochondrial function and osteoporosis in humans remains limited. It has been postulated that mtDNA content, quantified through copy number, serves as an indicator of mitochondrial genomic stability and biogenesis, thereby offering insight into mitochondrial functionality12. In this viewer, further studies are warranted to elucidate the causal relationship between mitochondrial function, as inferred from mtDNA content, and osteoporosis.

Mendelian Randomization (MR) is an instrumental variable methodology that employs single nucleotide polymorphisms (SNPs) as natural instrumental variables to infer causal relationships between phenotypic traits13. This approach intrinsically minimizes biases introduced by confounding factors and circumvents the challenge of reverse causation14,15. This study used large-scale GWAS datasets to investigate mtDNA copy number (mtDNA-CN) potentially causally related to osteoporosis through a two-sample MR study. Findings of this study may help to illuminate the genetic underpinnings and biological pathways implicated in osteoporosis.

Materials and methods

Study design

In our study, single nucleotide polymorphisms (SNPs) from GWAS were selected as genetic IVs. This MR study was performed based on the following three assumptions (Fig. 1): (1) Relevance assumption: the genetic variant is directly related to the exposure of interest; (2) Independence assumption: genetic variant is not related to any confounders between the exposure and outcome; (3) Exclusion restriction assumption: genetic variant does not affect outcome except through its association with the exposure16,17. The two-sample MR analysis was used to assess the causal association of mtDNA-CN (exposure) with the risk of osteoporosis and its subtypes. No further ethical approval was required because the present study was based on publicly available GWAS data. Our study was reported according to the “STrengthening the Reporting of OBservational studies in Epidemiology using Mendelian Randomization (STROBE-MR)” checklist17.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Description of the study design. (A) three fundamental MR assumptions. (B) Flow chart of the study design.

Data source

The summary-level data on mtDNA-CN were obtained from three GWAS meta-analyses18,19,20. The sample sizes for the three studies were 395,781, 465,809, and 295,150, respectively. All participants in the three meta-analyses were of European ancestry. Summary-level data for mtDNA-CN were obtained from three large-scale GWAS meta-analyses of European-ancestry populations (Study 1: N = 395,781; Study 2: N = 465,809; Study 3: N = 295,150) investigating mitochondrial biology18,19,20.

Summary data on total BMD were acquired from a GWAS meta-analysis of 30 GWASs of total body BMD including 66,628 European individuals21. Data on lumbar spine BMD, femoral neck BMD, and forearm BMD were from a GWAS meta-analysis investigating novel non-coding genetic variants with large effects on BMD (N = 53,236) and fracture (N = 508,253) in individuals of European ancestry from the general population22. All datasets included mixed sex distributions with age ranges representative of osteoporosis risk populations (adults > 40 years).

Utilizing GWAS summary data from three European-ancestry meta-analyses (total N = 1,156,740 across exposure datasets; outcome data from 66,628 individuals for total BMD and 53,236–508,253 for site-specific BMD/fractures), four MR analytical methods. Instrumental variables were selected using genome-wide significant SNPs (P < 5 × 10−8) after LD clumping (R2 < 0.001, 10,000kb window), MAF > 0.01, and F-statistics > 10 to ensure strength and independence.

Instrumental variables (IVs) selection

In order to uphold the validity and precision of outcomes regarding the causal relationship between mtDNA-CN and osteoporosis, quality control measures were employed in the selection of optimal IVs: (1) Genome-Wide Significance: SNPs showing a strong association with mtDNA-CN and osteoporosis, with a significance level of P < 5 × 10−8, were initially considered23. (2) Minor Allele Frequency: Only SNPs exhibiting a minor allele frequency (MAF) greater than 0.01 were retained to ensure common variant representation23. (3) Linkage Disequilibrium Management: To avoid multicollinearity, SNPs in high linkage disequilibrium (LD; R2 < 0.001) within a 10,000-kb window were excluded24. (4) Proxy SNP Inclusion: In instances where an instrumental variable was missing from the outcome dataset, high-LD proxies (R2 > 0.8) were sought and employed as substitutes25. (5) Instrument Strength Assessment: The F statistic was computed for each SNP in the instrumental variable set to gauge instrument strength and mitigate weak instrument bias. The calculation followed the formula F = [R2 * (N − 2)]/(1 − R2), where R2 represents the proportion of exposure variance explained by the SNP, and N denotes the sample size26. SNPs with an F statistic greater than 10 were retained to ensure adequate instrument power26. During the instrumental variable selection process, the strong instrumental effect of the selected SNPs was ensured through the F-statistic (mean value of 91.02), and horizontal pleiotropy interference was excluded by the MR-PRESSO test. For missing SNPs (such as rs8176645, etc.) in BMD data from different skeletal sites, proxy SNPs with high LD values (R2 > 0.8) were used as substitutes. SNPs without suitable proxies were excluded.

Mendelian randomization

The MR analysis was primarily conducted using the Inverse-Variance Weighted (IVW) approach, which presumes the absence of an average pleiotropic effect27. The IVW method is deemed the most efficacious strategy to assess the causal relationship between the mtDNA-CN and osteoporosis using odds ratios (ORs) and the corresponding 95% confidence intervals (CIs)27. The IVW method calculates a weighted mean of effect sizes, allocating weights inversely proportional to the variance of each SNP27. To ensure the resilience of the findings in this study, supplementary methodologies such as MR-Egger28, weighted median29, and weighted mode30 analyses were also implemented. The MR-Egger methodology accommodates an intercept term, thereby furnishing unbiased estimates of causal effects even amidst horizontal pleiotropy28. The weighted median approach rests on the premise that at least half of the instrumental variables accurately gauge the causal link between exposure and outcome29. Conversely, the weight mode strategy identifies the causal impact by concentrating on the SNP subset exhibiting the highest concentration, achieved through clustering SNPs into groups based on similarity in causal effects30. All statistical analyses were executed using the ‘TwoSampleMR’ package in R software, version 4.3.231. Multiple testing correction for association P-values was conducted using the Bonferroni method, with statistical significance set at P < 0.01 (0.05 divided by 5*1) to account for multiple comparisons32. A scatter plot of was used to depict the association between mtDNA-CN and osteoporosis33. A forest plot was used to show the MR estimate and 95% CI for each SNP33.

Sensitivity analysis

Sensitivity analyses were conducted to examine the robustness of MR findings against potential pleiotropic influences. Heterogeneity among the instrumental variables (IVs) was tested using the Cochran’s Q test, where a P value < 0.05 signified significant heterogeneity34. Furthermore, the MR-Egger regression methodology was invoked to probe for horizontal pleiotropy, with an intercept near zero or statistically non-significant in the MR-Egger regression indicative of an absence of substantial pleiotropic interference35. A funnel plot was employed to visually depict horizontal pleiotropy31. Besides, the MR Pleiotropy Residual Sum and Outlier (MR-PRESSO) method was applied to assess horizontal pleiotropy and identify and exclude outliers36. Additionally, following the exclusion of outlying IVs, we assessed whether there was a statistically significant discrepancy from those prior to their removal. The leave-one-out analysis was applied to identify whether each SNP influenced the main causal association37.

Results

IVs selection

For MR analysis with mtDNA-CN as the exposure, 70 IVs with no LD were selected. The mean F statics was 91.02. The min and max of F statics were 29.54 and 473.58. For total BMD, lumbar spine BMD, femoral neck BMD, and forearm BMD as outcomes, the numbers of selected SNPs were 67, 62, 62, and 62 respectively. The respective numbers of SNPs that did not match the summary data for total BMD, lumbar spine BMD, femoral neck BMD, and forearm BMD were 1, 6, 6, and 6. In the analysis with total BMD as the outcome, one SNP (rs8176645) lacked corresponding information in the summary data, proxy SNPs were identified for rs5745582 (represented by rs75080135) and rs28665408 (substituted by rs1790964). For the MR analysis with femoral neck BMD as the endpoint, six SNPs (“rs111701081”, “rs1613662”, “rs73250616”, “rs392020”, “rs2977608”, “rs3002416”) were not matched in the summary data, and no proxy SNPs were found for these. When analyzing lumbar spine BMD, the same six SNPs were not located in the summary dataset, and no alternative proxy SNPs could be assigned. In the MR analysis with forearm BMD as the outcome, the absence of information persisted for the same set of six SNPs in the summary data, and no suitable proxy SNPs were identified for them. Details on matching information in the summary data were shown in Supplementary file (Tables S1–S4).

Mendelian randomization

MR findings revealed no statistically significant associations between mtDNA-CN and BMD (Table 1 and Figs. 2, 3). IVW showed an insignificant association between mtDNA-CN and total BMD (OR = 0.97, 95% CI 0.89–1.05, P = 0.427), lumbar spine BMD (OR = 1.00, 95% CI 0.87–1.15, P = 0.977), femoral neck BMD (OR = 1.00, 95% CI 0.89–1.12, P = 0.988), and forearm BMD (OR = 0.90, 95% CI 0.70–1.16, P = 0.421). Results of MR-Egger, Weighted Median, and Weighted Mode were consistent with the IVW method.

Table 1 Results of causal relationship between mtNDA-CN and Osteoporosis.
Full size table
Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Scatter plots of the association between mtDNA-CN and BMD. (A) mtDNA-CN and total BMD. (B) mtDNA-CN and lumbar spine BMD. (C) mtDNA-CN and femoral neck BMD. (D) mtDNA-CN and forearm BMD.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Forest plots of the association between mtDNA-CN and BMD. (A) mtDNA-CN and total BMD. (B) mtDNA-CN and lumbar spine BMD. (C) mtDNA-CN and femoral neck BMD. (D) mtDNA-CN and forearm BMD.

Sensitivity analyses

The results of Cochran’s tests indicated significant heterogeneity in the association between mtDNA-CN and lumbar spine BMD (Q = 96.36, P = 0.002), femoral neck BMD (Q = 91.42, P = 0.007), and forearm BMD (Q = 101.18, P = 0.001) (Table 2 and Fig. 4). Results of MR-Egge tests indicated insignificant pleiotropy in the relationship between mtDNA-CN and total BMD (P = 0.16), lumbar spine BMD (P = 0.39), femoral neck BMD (P = 0.58), and forearm BMD (P = 0.32) (Table 2). Results of the MR-PRESSO test showed no potential pleiotropy and outliers for all the MR estimates in this study (Table 3). Leave-one-out analysis suggested that the causal estimates of mtDNA-CN and BMD were not influenced by any single SNP. The leave-one-out analysis plots are depicted in Fig. 5.

Table 2 Pleiotropy and heterogeneity analyses.
Full size table
Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

Funnel plots of the association between mtDNA-CN and BMD. (A) mtDNA-CN and total BMD. (B) mtDNA-CN and lumbar spine BMD. (C) mtDNA-CN and femoral neck BMD. (D) mtDNA-CN and forearm BMD.

Table 3 Pleiotropy analyses and outlier detection using the MR-PRESSO method.
Full size table
Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.
Full size image

Leave-one-out plots of the association between mtDNA-CN and BMD. (A) mtDNA-CN and total BMD. (B) mtDNA-CN and lumbar spine BMD. (C) mtDNA-CN and femoral neck BMD. (D) mtDNA-CN and forearm BMD.

Discussion

This was the first study to investigate the causal association between mtDNA-CN and osteoporosis by conducting MR estimates. Results of this two-sample MR analyses did not show significantly causal relationship between mtDNA-CN and osteoporosis.

The possible reasons for the insignificant relationship between mtDNA-CN and osteoporosis may be that the pathogenesis of osteoporosis is intricate, influenced by a multitude of factors including genetics, environment, nutrition, hormonal elements, lifestyles, and the interaction between these factors, while mitochondrial function is pivotal to bone metabolism, mtDNA-CN may represent just one among several influencing factors, and its isolated effect may not be substantial enough to significantly contribute to the onset of osteoporosis38,39,40. Mitochondria regulate osteoblast differentiation and osteoclast activity via energy production and ROS signaling. While mtDNA-CN reflects mitochondrial biogenesis, it may not fully capture functional aspects (e.g., oxidative phosphorylation efficiency, mtDNA mutations) critical for bone homeostasis. It remains unclear whether the mtDNA-CN detected in peripheral blood can truly reflect the mitochondrial status within the bone microenvironment. Mitochondria in bone tissue may exhibit distinct biological characteristics from circulating cells due to microenvironment specificity, such as mechanical stress and local redox status. This tissue heterogeneity could potentially weaken the association between peripheral blood mtDNA-CN and bone phenotypes. Mitochondria regulate osteoblast differentiation and osteoclast activity via energy production and ROS signaling. While mtDNA-CN reflects mitochondrial biogenesis, it may not fully capture functional aspects (e.g., oxidative phosphorylation efficiency, mtDNA mutations) critical for bone homeostasis. It remains unclear whether the mtDNA-CN detected in peripheral blood can truly reflect the mitochondrial status within the bone microenvironment. Mitochondria in bone tissue may exhibit distinct biological characteristics from circulating cells due to microenvironment specificity, such as mechanical stress and local redox status. This tissue heterogeneity could potentially weaken the association between peripheral blood mtDNA-CN and bone phenotypes. Notably, existing evidence supports this tissue-specific mitochondrial regulation. In SAMP6 accelerated aging mice, ultrastructural mitochondrial abnormalities in bone tissue were independent of systemic changes41. Furthermore, studies demonstrate that PGC-1α/β-driven mitochondrial biogenesis in bone cells directly regulates osteoblast-osteoclast balance through energy metabolism and redox signaling46,47, while Nrf2 deficiency specifically impairs mitochondrial antioxidant capacity in bone, leading to trabecular bone loss49. These findings collectively suggest that bone-specific mitochondrial dysfunction may occur independently of peripheral blood mtDNA-CN measurements.

Furthermore, as a quantitative indicator, mtDNA-CN is unable to capture critical abnormalities in mitochondrial function, such as large-scale mtDNA deletions, heteroplasmic mutations, or defects in respiratory chain complex activity. While observational studies suggest associations, they cannot disentangle whether mtDNA-CN alterations precede bone loss or merely represent epiphenomena of aging-related metabolic dysfunction. These functional alterations may independently affect bone metabolism, regardless of changes in copy number. It is worth noting that the energy metabolism of bone cells, especially osteoblasts, demonstrates significant plasticity. Studies have shown that even in the presence of mitochondrial dysfunction, bone cells can maintain energy supply by enhancing glycolysis. This metabolic compensatory mechanism may decouple the association between mitochondrial genome stability (represented by mtDNA-CN) and bone phenotypes. For instance, exercise enhances mitochondrial biogenesis, potentially overriding genetic influences on mtDNA-CN. More studies are needed to verify the mechanism for this finding.

Studies on accelerated aging mice (SAMP6) have uncovered ultrastructural alterations in bone tissue, further implicating mitochondrial changes as a significant factor in the development of senile osteoporosis41. A cross-sectional investigation of postmenopausal women revealed a substantial, independent positive correlation between mtDNA-CN in peripheral blood and femoral neck BMD, even after adjusting for variables such as age, body mass index as a measure of obesity, levels of inflammatory markers, indices of bone metabolism, and exercise engagement8. However, the participants in this cross-sectional study were postmenopausal women, postmenopause has been proved to be associated with a heightened risk of fractures, with postmenopausal women constituting the predominant group affected by fracture incidents42. The participants in our study were not limited to postmenopausal women. Interestingly, mtDNA-CN was not connected with lumbar BMD in the aforementioned cross-sectional study, which was consistent with the current MR estimates. The underlying mechanisms are not elucidated and further evidence is needed to address the differential findings between mtDNA-CN and BMD in specific part of human body. Moreover, a recent MR study by He et al. pinpointed four mtDNA loci exhibiting indicative associations with heel BMD, notably including the variant m.16356T>C located in the MT-DLOOP region (P = 1.50 × 10−3) when considering the total sample cohort43. Of note, He et al.’s study focused on heel BMD, while the outcomes in our study were total BMD, lumbar spine BMD, femoral neck BMD, and forearm BMD. This may explain the different results between the current study and He et al.’s. To date, the existing evidence suggests that cells of the osteoblast lineage generate adenosine triphosphate (ATP) from glucose, employing both oxidative phosphorylation and glycolytic metabolic pathways, when mitochondrial function is impaired and cellular energy supply diminishes, cells can compensate for reduced ATP production by upregulating the activity of the glycolytic pathway44. Glycolysis, occurring in the cytoplasm independently of mitochondria, swiftly yields ATP albeit at a lower efficiency, proving particularly efficacious during periods of acute energy demand44. Furthermore, Peroxisome proliferator-activated receptor (PPAR)-γ coactivator-1 (PGC-1) enhances mitochondrial biogenesis and respiratory capacity through direct coactivation of multiple transcription factors central to cellular metabolic regulation, notably nuclear respiratory factor-1 (NRF-1) and nuclear respiratory factor-2 (NRF-2), elevated expression of PGC-1 provokes the transcription of both NRF-1 and NRF-2, which in turn bind to the promoter region of the nuclear-encoded mitochondrial transcription factor A (TFAM)45. This process initiates a regulatory cascade enhancing the expression of key mitochondrial genes encoding ATP synthase, Cytochrome C, and Cytochrome c oxidase subunit I, thereby augmenting mitochondrial biogenesis, this upregulation is crucial for meeting the heightened energy requirements during osteoblast and osteoclast formation46. Studies have demonstrated that the deletion of ERRα or ASXL2, both pivotal targets of PGC-1β, results in severe osteopetrosis, a condition characterized by impaired osteoclast differentiation and a consequential reduction in bone resorption47,48. Another possible mechanism for the insignificant causal relationship between mtDNA-CN and osteoporosis may be that under conditions of oxidative stress, Nrf2 is activated, thereby amplifying the expression of antioxidant genes and safeguarding osteocytes against oxidative damage49. A previous animal study has illustrated that mice lacking Nrf2 exhibit diminished trabecular bone mineral density in their femurs and a decrease in cortical bone area within their vertebrae, highlighting the critical role of Nrf2 in maintaining bone integrity49. Although mtDNA-CN reflects mitochondrial biosynthesis, it may not fully align with key parameters of mitochondrial function, such as oxidative phosphorylation efficiency and mtDNA mutation load. For instance, large deletions in mtDNA or defects in complex I/III activity can significantly impact ROS generation and apoptotic signaling, yet these functional abnormalities may occur within the normal range of mtDNA-CN, thus escaping detection by the genetic instrumental variables designed in this study.

There are strengths for our study. The findings of this study may provide evidence for the research on pathogenesis of osteoporosis. This study, for the first time, explored the causal relationship between mitochondrial DNA copy number (mtDNA-CN) and osteoporosis through Mendelian randomization, addressing the limitations of observational studies that cannot exclude confounding factors. Although the results showed no significant causal association, this conclusion suggests that caution should be exercised in clinical practice when considering mtDNA-CN as a biomarker for osteoporosis. Future research should focus on other mitochondrial functional indicators or genetic factors. Besides, based on the current evidence, mtDNA-CN may not serve as a reliable marker for the detection or prediction of risk for osteoporosis in the clinical setting. Meanwhile, there are limitations in this study. Firstly, the Mendelian randomization (MR) analysis was confined to a cohort of European ancestry, which restricts our ability to generalize these findings to more diverse populations. Secondly, while the possibility of participant overlap between exposure and outcome datasets cannot be excluded, accurately measuring the degree of this overlap presents a challenge. Nonetheless, the study’s use of highly robust instrumental variables, as evidenced by F statistics markedly exceeding 10, serves to effectively minimize bias that might arise from such overlaps50. Thirdly, the restricted number of single nucleotide polymorphisms (SNPs) employed as instrumental variables potentially captures only a small fraction of the exposure’s variability, possibly diminishing the statistical power of the inferred causal relationships. Consequently, given these constraints, interpretations of the study’s results should be made with caution. While mtDNA-CN is sensitive to oxidative stress and inflammation, these factors may act as confounders or effect modifiers. For example, chronic inflammation (e.g., elevated IL-6) accelerates mitochondrial dysfunction and bone loss independently of mtDNA-CN. Similarly, estrogen deficiency in postmenopausal women alters mitochondrial ROS production, which may overshadow mtDNA-CN’s role in BMD regulation. Our MR design, which isolates genetic effects, may not capture these dynamic interactions, necessitating future studies integrating environmental mediators.

While mtDNA-CN may not directly drive osteoporosis, our findings urge exploration of alternative mitochondrial mechanisms (e.g., mtDNA mutations, electron transport chain efficiency) and their interplay with lifestyle factors. Clinically, interventions targeting mitochondrial health (e.g., antioxidants, exercise) may still benefit bone metabolism via pathways beyond mtDNA-CN. Future research should prioritize gene-environment interactions and multi-omics approaches to unravel the complexity of osteoporosis etiology. The discrepancies between the findings of this study and previous observational research may stem from two aspects: First, mtDNA-CN solely reflects mitochondrial genome stability and cannot comprehensively characterize the functional status of mitochondria. Second, compensatory mechanisms exist in the energy supply for bone metabolism—when oxidative phosphorylation is impaired, cells may enhance the glycolytic pathway to sustain ATP production, which might attenuate the causal effect of mtDNA-CN on bone mineral density.

The findings from this two-sample MR analysis do not provide evidence for a genetically inferred causal link between mtDNA-CN and osteoporosis. Therefore, the correlation previously observed in observational studies between mtDNA-CN and osteoporosis might be attributed to the impact of confounding factors. The study population is limited to samples of European ancestry, while mitochondrial genome characteristics exhibit variations across different ethnic groups. This genetically informed approach allows us to circumvent the 'exposure measurement fallacy’, where blood mtDNA-CN inaccurately proxies bone mitochondrial status, by anchoring the analysis in germline genetic variants that systemically regulate mitochondrial biogenesis. For instance, Asian-specific mtDNA haplotypes may influence the association patterns between copy number and bone metabolism, which necessitates validation through cross-ethnic genome-wide association studies (GWAS) data. To solidify these study outcomes, future analyses should employ larger GWAS summary datasets based on multiple populations and an expanded repertoire of genetic instruments. Given the potential heterogeneity in frequency and effect size of the genome across different ethnic groups, the conclusions of this study should be cautiously generalized to other populations. We call for future studies to utilize multi-ethnic cohorts and cross-racial GWAS data to validate the association between mtDNA-CN and osteoporosis, in order to elucidate the potential impact of genetic background and environmental interactions.