Research Process and IVs Selection

Based on the clinical study that PCSK9 inhibitors can effectively reduce LDL-C levels, our research delved into the potential relationship between PCSK9 inhibitors and osteoporosis. We selected the LDL-C dataset (ieu-a-300) from the GWAS catalog (https://gwas.mrcieu.ac.uk/) as exposure factor. This dataset included a total of 173,082 individuals with 2,437,752 single nucleotide polymorphisms (SNPs). The dataset comprised 37 studies predominantly involving European individuals, with additional studies focusing on East Asian, South Asian, and African populations [12]. Samples for LDL-C analysis were collected from fasting individuals, with those taking lipid-lowering medication being excluded. Criteria for IVs selection included screening SNPs associated with LDL-C from the dataset (ieu-a-300), choosing SNPs meeting genome-wide significance (P < 5 × 10–8), and ensuring no significant linkage disequilibrium (r2 < 0.3). A total of 403 SNPs were initially identified, followed by further selection of 13 SNPs located within ± 100 clumping distance (kb) of the PCSK9 gene as IVs (Supplementary Table 1).

In this study, we utilized a dataset of coronary heart disease (CHD) as the positive control for drug-targeted MR analysis. The case status of CHD was defined using a comprehensive coronary artery disease (CAD) diagnosis, which encompassed acute coronary syndrome, coronary stenosis > 50%, chronic stable angina, or myocardial infarction (ieu-a-7). Utilizing CHD as the positive control not only ensured the reliability of the experimental outcomes but also highlighted the potential clinical implications of the drug-targeted MR analysis results. The dataset included a total of 184,305 individuals (case = 60,801, control = 123,504) and 9,455,779 SNPs. The initial focus was on assessing the causal relationship between PCSK9 inhibitors and CHD, with a PIVW < 0.05 indicating CHD as a positive control.

Following the verification of the causal relationship between PCSK9 inhibitors and CHD, we selected the osteoporosis dataset as the outcome factor (ukb-b-12141). This dataset comprised 462,933 individuals (case = 7,547, control = 455,386) with 9,851,867 SNPs. Osteoporosis diagnosis was established through a bone dual energy X-ray examination, with a T-score below -2.5 indicating the presence of osteoporosis [13]. Exclusion criteria involved the presence of bone destruction from neoplastic disease in all patients. The PHESANT (PHEnome Scan ANalysis Tool) was used to obtain the dataset from the UK Biobank [14]. To mitigate strong heterogeneity, all palindromic SNPs among the 13 IVs were removed using the “Two Sample MR” package. Subsequently, a “two sample MR” analysis was conducted, combining the IVs to investigate the relationship between PCSK9 inhibitors and osteoporosis. Heterogeneity, pleiotropy, sensitivity and bias were calculated.

Drug-targeted MR analysis was utilized to investigate the potential causal relationship between PCSK9 inhibitors and osteoporosis. This method was advantageous as it was not influenced by confounding factors, allowing for a more accurate exploration of mediation effects. To further examine the mediation effect, the “two step MR” method was employed, focusing on common indicators such as bone mineral density (BMD) levels, total 25-hydroxyvitamin D (T25(OH)D) levels, and calcium supplementations (Fig. 1).

Fig. 1

Study flow chart. This study was devided into MR analysis and mediation analysis. MR analysis was used to assess whether LDL-C level based on PCSK9 inhibitors had a causal relationship with osteoporosis. Mediation analysis was used to assess whether mediation effect was existing. MR, Mendelian randomization; SNPs, single nucleotide polymorphisms

Causal Effect and Heterogeneity

In this study, four algorithms (MR Egger, weighted median, inverse variance weighted and weighted mode) were utilized for MR analysis, with the inverse variance weighted (IVW) method being identified as the most important [15]. When the PIVW < 0.05, as long as the beta of other methods were in the same direction as IVW method, a causal relationship between the exposure factor and the outcome factor was existing [16]. To ensure the reliability of IVW method, heterogeneity was evaluated. We calculated the index of inconsistency (I2) and P-value by Cochran’s Q test. When P < 0.05, heterogeneity was considered and random-effects IVW was used. When P > 0.05, no heterogeneity was considered and fixed effect IVW was used [17].

Horizontal Pleiotropy and Sensitivity Analysis

Confounders were the exogenous factors that were related to both exposure and outcome factors in a study. If confounders were not identified or eliminated, the true relationship between exposure factor and outcome factor was hidden or exaggerated [18]. In this study, horizontal pleiotropy was calculated by the intercept of MR regression. When the intercept of MR regression was near to “0”, horizontal pleiotropy was ignored. And the results of the IVW were not affected by confounders [19]. The purpose of sensitivity analysis was to evaluate the reliability of MR. We conducted sensitivity analysis using the “leave-one-out” method. By eliminating each SNP one by one, the meta effects of the remaining SNPs were calculated to determine whether the SNP significantly altered the results. When all SNPs were on the same side of 0, the result of MR was robust.

Mediation Analysis of BMD Levels in Different Ages

BMD levels and age have been previously confirmed as important risk factors for osteoporosis. Bone mass growth goes through a steady increase in childhood and a rapid increase in adolescence, reaching its peak and remaining stable around age 30 years [20]. Then it gradually decreases around age 50 years, with a significant decline in postmenopausal women [21]. Dual energy X-ray absorptiometry (DXA) is considered the gold standard for detecting BMD levels. The International Society for Clinical Densitometry (ISCD) recommends the lumbar spine (LS), femoral neck (FN), and total hip as preferred measurement sites for individuals over 50 years old. For children and adolescents, the recommended sites for measurement include the total body (excluding the head) and LS (http://www.iscd.org/official-positions). Therefore, on the relationship between PCSK9 inhibitors and osteoporosis, we further explored whether BMD levels with different ages had a mediation effect on osteoporosis. We selected the largest dataset of BMD levels with different ages as the mediator [22]. The database included a total of 66,628 individuals, and was divided into five stages: 0–15 years old, 15–30 years old, 30–45 years old, 45–60 years old and > 60 years old. BMD levels were measured by DXA. Total body measurements were used in children. The LS and FN were used in adults.

To ensure the reliability of the odds ratio (OR) value, SNPs significantly associated with BMD levels were used as IVs to estimate the causal effect on osteoporosis. Based on definition of the dominant model and literature report, we categorized genotypes with mutant bases as the exposed group and wild homozygotes as the non-exposed group. Here we completed the conversion of BMD levels from continuous variables into binary categorical variables [22]. Then we conducted a “two sample MR” analysis on PCSK9 inhibitors, BMD level and osteoporosis, respectively. We calculated the mediation effect and direct effect. A PIVW < 0.05 was the prerequisite for mediation analysis, and any MR analysis that did not meet the requirements would not be subjected.

Mediation Analysis of Serum T25(OH)D Levels

The serum T25(OH)D level is an important laboratory test for evaluating vitamin D status and its significance in relation to osteoporosis was investigated in this study. A dataset from the GWAS catalog, specifically the largest dataset (ebi-a-GCST90000618) available, was selected as the mediation factor [23]. The participants were from 2006 to 2013 and ranged in age from 40 to 60. Serum T25(OH)D levels were measured using chemiluminescence immunoassay (CLIA). Data falling outside the validation range of 10–375 nmol/L were excluded. Finally, 417,580 individuals of European ancestry were obtained. Based on definition of the dominant model and literature report, we categorized genotypes with mutant bases as the exposed group and wild homozygotes as the non-exposed group [23]. We completed the conversion of serum T25(OH)D levels from continuous variables into binary categorical variables. A “two sample MR” analysis was conducted on PCSK9 inhibitors, serum T25(OH)D levels, and osteoporosis, respectively. We calculated the mediation effect using the same method as the BMD level. A PIVW < 0.05 was the prerequisite for mediation analysis.

Mediation Analysis of Calcium Supplements

The effect of calcium supplements on osteoporosis is being investigated. A balanced diet is important for bone health [24]. It was suggested that calcium and vitamin D supplements may reduce fracture risk modestly [25]. Therefore, we performed a mediation analysis on calcium supplements as a possible mediation factor in this study. We selected a dataset of calcium supplements from the GWAS catalog. The dataset included 336,314 participants (case = 22,047, control = 314,267). We conducted a “two sample MR” analysis on PCSK9 inhibitors, calcium supplements and osteoporosis, respectively. A PIVW < 0.05 was the prerequisite for mediation analysis.