Macrophage polarization-related genes with potential causal roles in osteoporosis: a multi-omics Mendelian randomization study Authors: Feng Xiong, Aoqi Zang, Ming Zhu, Bingxu Ma, Jun Zhang, Peitai Liu, Feng Qian, Jianzhong Guan, Hui Xu Published: 2026 ABSTRACT AbstractBackgroundEmerging evidence links macrophage polarization to osteoporosis pathogenesis, yet the specific causal genes and underlying molecular mechanisms remain poorly defined.MethodsWe used a multi-stage, summary-data-based Mendelian randomization (SMR) framework to integrate blood-derived methylation, expression, and protein quantitative trait loci (mQTL, eQTL, and pQTL) with large-scale osteoporosis genome-wide association study (GWAS) datasets. We performed SMR and colocalization analyses across predefined macrophage polarization-related genes, followed by multi-omics integration. To explore external expression patterns, we examined mRNA levels of selected genes in an independent femoral bone tissue microarray dataset (GSE230665; 12 postmenopausal osteoporosis cases and 3 controls), and conducted pathway enrichment and network analyses of prioritized genes.ResultsOur multi-omics analysis identified 66 genes at the methylation level, 14 at the expression level, and 5 proteins with SMR evidence for association with osteoporosis. Integrative analyses highlighted DNMT3A, CTSK, and CDCA4 at the methylation–expression interface, and AGER, F2, and HMOX1 at the protein level, as candidates with genetically predicted associations with osteoporosis risk. In the external femoral bone tissue dataset, AGER mRNA expression was lower in osteoporosis patients than in controls (p = 0.0044), whereas DNMT3A mRNA expression was higher in patients (p = 0.018), opposite to the protective association suggested by eQTL-based SMR, suggesting complex tissue-specific or compensatory regulation.ConclusionsThis blood QTL-based multi-omics SMR study prioritizes macrophage polarization-related genes and proteins with genetically predicted associations with osteoporosis risk and suggests links between peripheral immune signaling and bone metabolism. These findings are exploratory and hypothesis-generating and require validation in bone marrow-specific and functional studies.Clinical trial numberNot applicable.Supplementary InformationThe online version contains supplementary material available at 10.1186/s13018-026-06905-1.Keywords: Macrophage polarization, Osteoporosis, Summary-data-based mendelian randomization, Multi-omics analysis, Colocalization analysis Background Osteoporosis, a debilitating systemic skeletal disease characterized by diminished bone mass and microarchitectural deterioration, culminates in a heightened susceptibility to fragility fractures [1]. These fractures, particularly common in postmenopausal women, represent a significant global health crisis, inflicting substantial morbidity, loss of independence, and increased mortality upon millions worldwide [2]. While bone mineral density measurement serves as a cornerstone for diagnosis, accounting for roughly 70% of bone strength, the escalating incidence of osteoporotic fractures, driven by an aging global population [3, 4] and genetics [5, 6], imposes an immense socioeconomic burden [7], especially if the patients require complex procedures like vertebroplasty [8], which are not without risk [9]. Understanding the intricate interplay of factors, including significant genetic predispositions that govern bone metabolism and fracture risk, is therefore paramount for developing more effective preventative and therapeutic strategies to combat this pervasive condition [1]. Modern osteoporosis management has shifted toward individualized drug selection, biomarker‑guided monitoring, and precision approaches that consider comorbidities (e.g., glucocorticoid use) and genetic background [10–14]. Still, an in-depth understanding of the mechanisms involved in osteoporosis is necessary to formulate precise tailored treatments [15]. Macrophages and broader immune mechanisms are now recognized as central drivers and potential therapeutic targets in muscle, tendon, and joint pathologies, influencing both tissue damage and regeneration [16–18]. Furthermore, macrophage polarization may promote osteoblast differentiation and enhance bone formation, which is involved in osteoporosis and articular pathologies [19, 20]. Macrophages are essential components of the immune system and can polarize into various states in response to environmental signals, primarily M1 and M2 types [21, 22]. Activation of M1 macrophages increases pro-inflammatory factors that stimulate osteoclast formation and activity, leading to bone loss, whereas M2 macrophages secrete osteogenic factors that promote the differentiation and activation of pre-osteoblasts, thereby enhancing bone formation [23, 24]. Clinical studies have shown that in osteoporotic patients, there is a skewing towards M1 macrophage polarization, which correlates with increased bone resorption [25, 26]. Furthermore, targeting macrophage polarization through specific gene manipulations has shown promise as a therapeutic strategy [27]. The potential of these genes as targets or biomarkers for osteoporosis is being actively explored, highlighting their translational significance [28]. Autophagy also appears to be involved in osteoporosis [29]. However, there is currently a lack of systematic human genetic evidence on the causal roles of macrophage polarization-related genes in osteoporosis, particularly regarding their multi-omics regulatory mechanisms, from DNA methylation to protein expression levels. Summary Data-based Mendelian Randomization (SMR) is a method that uses genetic variants associated with methylation, gene expression, and protein abundance as instrumental variables to explore potential causal associations, combining traditional Mendelian randomization with multi-omics data to uncover complex connections between genes and diseases [30]. SMR analysis leverages summary-level data from genome-wide association studies (GWAS) to infer causal relationships between genetic variants and complex traits. This approach allows integration of various quantitative trait loci (QTL) data, including methylation QTL (mQTL), expression QTL (eQTL), and protein QTL (pQTL), to uncover the molecular mechanisms of disease and identify potential therapeutic targets [31]. Although the complex interactions between macrophages and osteocytes in osteoporosis occur primarily in the local bone microenvironment [25, 32], studying these processes in human cohorts is inevitably limited by the lack of large-scale, high-quality, bone tissue-specific multi-omics QTL data. However, osteoporosis is increasingly recognized as a systemic disease, in which circulating immune cells play a key role [33–35]. Crucially, peripheral blood monocytes are major circulating precursors that migrate to bone tissue and differentiate into osteoclasts and tissue-resident macrophages [36]. Thus, systemic genetic and epigenetic variants captured in whole blood can serve as valuable (albeit alternative) surrogate indicators reflecting the immune and osteoclast genesis potential that drives bone loss. Hence, although osteoporosis occurs in the bone microenvironment, systemic immunity and peripheral blood mononuclear cells (precursors of osteoclasts) play an important role in it. Due to the lack of large-scale bone-specific QTL data, this study used large-scale blood QTL data as proxies. Therefore, we capture systemic immune/epigenetic signatures that influence osteoporosis risk, rather than a direct response from local osteocytes. Unlike previous observational studies describing an M1‑skewed macrophage phenotype in osteoporosis, our multi‑omics SMR design exploits germline genetic instruments to reduce confounding and reverse causation and to identify specific methylation sites, genes, and proteins within macrophage polarization pathways that are genetically predicted to associate with osteoporosis risk. It provides a more granular and translationally relevant map of macrophage‑related targets than prior work. In this study, we aimed to employ a multi-omics SMR approach to systematically evaluate the potential causal associations between MPRGs and osteoporosis risk at the levels of DNA methylation, gene expression, and protein abundance, in order to identify key candidate genes and their putative regulatory pathways, thereby providing genetic evidence for elucidating the pathogenesis of osteoporosis and uncovering novel intervention targets. Methods Study design This study employed genetic variants (single-nucleotide polymorphisms (SNPs)) associated with methylation, gene expression, and protein abundance levels as instrumental variables. Independent SMR analyses were performed at each biological level to explore their associations with osteoporosis. The FinnGen_R10_M13_OSTEOPOROSIS dataset was used for primary discovery, while the GCST90086118 and UKB-b-17,796 cohorts provided validation. We confirmed that the primary outcome data (FinnGen) and exposure data (eQTLGen/Meta-analysis of European cohorts) were derived from independent populations with no participant overlap, as indicated in their respective cohort descriptions. To enhance causal inference, colocalization analysis was also conducted. By integrating the results from these three SMR analyses, we identified causal candidate genes and performed further tissue-specific validation. The workflow of SMR analysis for genetic variant selection and analysis methods was illustrated in Fig. 1. Finally, the results of the SMR analysis were validated within the GEO dataset. The research workflow and analysis methods Data sources Candidate genes related to macrophage function and inflammation were first collected using keyword‑based searches in GeneCards to ensure broad coverage of potentially relevant loci. This inclusive list was intended as a hypothesis‑generating starting set rather than a definitive pathway definition. For genes prioritized by our integrative analyses, we subsequently cross‑referenced existing literature on macrophage polarization, bone remodeling, and osteoporosis to confirm biological relevance and to contextualize their potential roles in the macrophage-bone axis. Genes associated with macrophage polarization were retrieved from the GeneCards database (https://www.genecards.org) by querying the keyword “Macrophage Polarization.” Restricting the search to the “Protein Coding” category yielded 731 protein-coding genes implicated in macrophage polarization mechanisms. The blood mQTL summary data were derived from a meta-analysis of two European cohorts [37], which includes gene expression data from the Brisbane Systems Genetics Study (n = 614) and the Lothian Birth Cohorts (n = 1366). Blood eQTL data were sourced from the eQTLGen consortium, which includes gene expression data from 31,684 individuals [38]. Blood pQTL summary data came from Sun et al. [39], covering 54,219 UKB participants. All methylation, expression, and protein quantitative trait loci used as instruments in this study were derived from whole blood cohorts and thus reflect genetic regulation in circulating or blood‑resident cells rather than in bone tissue or osteoclasts. To address the broad scope and potential bias of the GeneCards‑derived list, we additionally constructed curated macrophage polarization gene sets directly from the primary literature. First, we extracted differentially expressed genes from canonical in vitro M1 versus M2 macrophage polarization experiments reported in well‑characterized RNA‑seq and microarray studies, using thresholds of an absolute log2 fold change greater than 1 and an adjusted p-value less than 0.05 where available. Second, we compiled targets and downstream effectors of established macrophage polarization master regulators, including NF‑κB components and interferon‑ and interleukin‑responsive transcription factors, from published ChIP‑seq and perturbation studies. Third, where bone marrow–derived macrophage bulk or single‑cell RNA‑seq datasets were available, we identified genes enriched in inflammatory (M1‑like) versus reparative (M2‑like) clusters based on the authors’ annotations. These literature‑derived gene sets were used to perform pre‑specified sensitivity analyses in parallel with the GeneCards‑based list. Genome-wide association study (GWAS) summary statistics for osteoporosis were obtained from the FinnGen database, including 8017 cases and 391,037 controls. For validation, we used additional datasets from other GWAS catalogs (5399 cases and 51,238 controls) and the UKB (7426 cases and 55,627 controls). For external validation of transcriptomic findings, mRNA expression array data were obtained from the Gene Expression Omnibus (GEO) database under the accession number GSE230665. This dataset included samples from 12 postmenopausal patients with osteoporosis and 3 healthy controls. All summary statistics used for analysis were sourced from previously published studies (Table 1), all of which had received ethical approval. Table 1The GWAS queue informationTraitGWAS ID/GEO IDCase/controlSNPsOsteoporosisfinngen_R10_M13_OSTEOPOROSIS8017/ 391,03720,193,071OsteoporosisGCST900861185399/51,238144,850,81Osteoporosisukb-b-17,7967426/55,6273,161,666postmenopausal osteoporosis: healthy controlGSE230665 (mRNA array)12:3N/A The GWAS queue information Summary-data-based MR (SMR) analysis For each omics layer (mQTL, eQTL, pQTL), we first performed SMR screening across the predefined set of 731 macrophage‑ and inflammation‑related genes of interest. We performed Summary-data-based Mendelian Randomization (SMR v1.3.1) to identify pleiotropic associations between molecular traits (DNA methylation, gene expression, and protein levels) and disease phenotypes. This approach integrates multi-omics QTL data with GWAS summary statistics to test whether a shared causal variant influences both molecular levels and disease outcomes [40]. In this study, we selected the most significant cis-QTLs using a window centered on the corresponding genes (± 1000 kb) with a p-value threshold of 5.0 × 10−8. SNPs with allele frequency differences exceeding a specified threshold (set at 0.2 in this study) in any paired dataset (including LD reference samples, QTL summary data, and outcome summary data) were excluded. For eQTL, mQTL, and pQTL, the default allowed allele frequency difference ratio was 0.05 [22]. To distinguish pleiotropy from linkage disequilibrium (LD), we applied the Heterogeneity in Dependent Instruments (HEIDI) test. This test utilizes secondary SNPs in LD with the lead cis-QTL (P > 1.57 10−3, MAF > 0.01, and 0.05 < r2 < 0.9) to evaluate effect size heterogeneity. We also conducted multi-SNP SMR, which estimates joint effects by incorporating all cis-SNPs meeting the P < 5 × 10−8 and r2 < 0.9 criteria [40]. In this study, under the condition of p_SMR < 0.05, results that satisfy both p_SMR-multi < 0.05 and p-HEIDI > 0.01 will be used for subsequent eQTL, mQTL, and pQTL colocalization and integrative analyses. While we calculated Benjamini-Hochberg FDR-adjusted p-values (FDRSMR) for all primary screens, we categorized genes passing the multi-step integrative filter but not stringent single-layer FDR as suggestive causal candidates rather than definitively established causal genes [41, 42]. Finally, we integrated these omics layers to explore the regulatory cascade. Using methylation sites as exposures to identify epigenetic regulation of gene expression. Using gene expression as the exposure to validate downstream effects on protein abundance. Colocalization analysis We conducted colocalization analysis using the coloc R package to detect shared causal variants between identified macrophage polarization-related mQTLs, eQTLs, or pQTLs and osteoporosis disease. Specifically, when GWAS signals and QTLs colocalize, we infer that the GWAS signal loci may influence the phenotype by altering gene-related biological processes. In the colocalization analysis, five distinct posterior probabilities were reported, corresponding to five independent hypotheses: (1) no genetic association with either trait (H0); (2) genetic association only with gene expression (H1); (3) genetic association only with disease risk (H2); (4) both traits are associated with SNPs but with different causal variants (H3); (5) both traits share the same causal variant (H4). According to published literature, the colocalization region windows were set to ± 500 kb, ± 1000 kb, and ± 1000 kb, respectively, for colocalization analysis involving mQTL-GWAS, eQTL-GWAS, and pQTL-GWAS [43, 44]. The default prior probability that SNPs are associated with both exposure and outcome was set to p12 = 1 × 10⁻⁵. Although a posterior probability of colocalization (PPH4) > 0.8 has been shown to indicate strong Bayesian evidence for colocalization, Michael S. Breen et al. observed that many loci with PPH4 > 0.5 also qualitatively exhibit patterns consistent with colocalization [44]. Therefore, to enhance the sensitivity of the colocalization analysis and ensure the capture of more potential colocalization signals, especially during the exploratory phase of the study, we chose a posterior probability threshold of PPH4 > 0.5 for colocalization, which provides a more comprehensive initial clue. All statistical analyses were conducted using R (v 4.3.0). The R packages “ggplot2” and “ggrepe1” were used for Manhattan plot creation, and “forestplot” was used for forest plot generation. The code for creating SMR Locus Plots and SMR Effect Plots was sourced from Zhu et al. [45]. Transcriptomic validation and functional pathway analysis For the GSE230665 dataset, probe IDs were mapped to gene symbols. For genes corresponding to multiple probes, the average expression value was used. Differential expression analysis between the osteoporosis and control groups was performed using the “limma” package in R. p-values < 0.05 were considered statistically significant. To investigate the biological functions of the 77 unique MPRGs identified as significant in the SMR analyses, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed using the “clusterProfiler” R package. The Benjamini-Hochberg method was used to adjust p-values for multiple testing, and terms with an adjusted p-value < 0.05 were considered significantly enriched. To explore the interactions among the 77 candidate genes, a protein-protein interaction (PPI) network was constructed using the STRING database (v12.0) (https://string-db.org), with a required interaction score of > 0.4. The network was visualized in Cytoscape (v3.9.1), and central hub genes were identified using seven topological algorithms (Degree, Stress, Radiality, Closeness, MCC, MNC, and EPC) within the cytoHubba plugin. Genes ranked in the top 15 by all seven algorithms were defined as hub nodes. Additionally, the GeneMANIA web tool was used to construct a focused interaction network for the six primary candidate genes to predict additional interacting partners and functional associations based on physical interactions, genetic interactions, colocalization, and shared protein domains (https://genemania.org). For AGER, we further constructed a dedicated GeneMANIA network using AGER as the seed gene in the human database with default weighting, restricting the analysis to physical interactions, co‑expression, and shared pathway membership. The resulting network was exported to Cytoscape for visualization, and we calculated node degree and betweenness to qualitatively assess hub‑like behavior. Functional enrichment of the AGER‑centered network was evaluated using Gene Ontology and KEGG annotations provided by GeneMANIA, focusing on innate immune and pattern recognition receptor signaling categories. Results Methylation SMR analysis (mQTL → osteoporosis) Considering that overly rigorous multiplexing tests may lead to false negatives, we combine cross-omics consistency (i.e., showing significant nominal P at both methylation and expression levels) and strong colocalization evidence (PPH4 > 0.5) to retain specific genes/proteins as prioritized candidates rather than definite causal genes. Based on the criteria of “p_SMR-multi < 0.05, p_SMR < 0.05, p-HEIDI > 0.01”, we identified 145 methylation sites, corresponding to 66 MPRGs (all SMR analysis results are shown in Table S1). In the results obtained from the SMR analysis, seven sites (corresponding to the genes DCLK1, DNMT3A, F2, and TERT) were validated in the GCST90086118 cohort, and seven sites (corresponding to the genes CCL24, KLF6, and TUBB) were validated in the ukb-b-17,796 cohort (p_SMR_multi < 0.05, p_SMR < 0.05, p-HEIDI > 0.01) (all SMR analysis results are shown in Tables S2 and S3), providing strong evidence for the stability and consistency of these sites across different populations. In the discovery cohort, we also identified 23 sites (corresponding to 20 genes) that showed strong colocalization evidence (PPH4 > 0.5) (Fig. 2A). For example, the high methylation level of CpG site cg19994779, corresponding to the CDCA4 gene, was positively associated with an increased risk of osteoporosis [OR = 1.09, 95% CI (1.01–1.18)], supported by strong colocalization evidence (Figure S1A). The validation results of these genes further support their potential role in osteoporosis through the macrophage polarization pathway. Forest plot of potential causal associations between multi-omics QTLs and osteoporosis. The forest plot summarizes the odds ratios (OR) and 95% confidence intervals (CI) for the potential causal associations identified in the SMR analyses. Results are shown for A methylation QTLs (mQTLs), B expression QTLs (eQTLs), and C protein QTLs (pQTLs) with osteoporosis risk. Squares represent the OR, and horizontal lines represent the 95% CI Gene expression SMR analysis (eQTL → osteoporosis) Through screening, we identified that the expression of 14 genes was significantly associated with osteoporosis (p_SMR-multi < 0.05, p_SMR < 0.05, p-HEIDI > 0.01) (Fig. 2B, Table S4). Among these, high expression levels of five genes (CDCA4, CTSK, DNMT3A, JAG1, and ZBTB3) were associated with a decreased risk of osteoporosis, while high expression levels of the remaining genes were positively correlated with osteoporosis risk. Our findings further confirmed that the DNMT3A gene was validated in the GCST90086118 cohort, and the TUBB gene was validated in the ukb-b-17,796 cohort (p_SMR_multi < 0.05, p_SMR < 0.05, p-HEIDI > 0.01) (all SMR analysis results are shown in Tables S5 and S6). Additionally, in the discovery cohort, we identified strong colocalization evidence for the expression of six genes (PPH4 > 0.5) (illustrated using DNMT3A as an example in Figure S1B). Protein SMR analysis (pQTL → osteoporosis) At the protein abundance level, five proteins were found to be associated with osteoporosis (p_SMR-multi < 0.05, p_SMR < 0.05, p-HEIDI > 0.01) (Fig. 2C and Table S7). The findings suggest that these proteins may play crucial roles in the pathophysiology of osteoporosis, particularly through mechanisms involving macrophage polarization. SMR analysis further validated the association of the F2 in the GCST90086118 cohort and the AGER in the ukb-b-17,796 cohort (p_SMR_multi < 0.05, p_SMR < 0.05, P-HEIDI > 0.01) (All SMR analysis results are shown in Tables S8, S9). Moreover, in the discovery cohort, we identified one protein (HMOX1) with strong colocalization evidence (PPH4 > 0.5) (Figure S1C). To further investigate the role of AGER in macrophage polarization, we constructed an AGER‑centered interaction network using GeneMANIA. In this network, AGER occupied a highly integrated hub position, displaying strong physical interactions with classical M1‑type polarization triggers such as S100 family members and HMGB1, as well as dense connections with key NF‑κB transcription factors (RELA, NFKB1/2) through co‑expression and shared signaling pathways (Figure S3). Functional enrichment analysis of the AGER‑centered network revealed significant enrichment for pattern recognition receptor signaling and NF‑κB signaling pathways, supporting a mechanistic link between AGER activity, macrophage activation states, and osteoporosis‑related inflammatory signaling. Multi-omics integration Based on significant findings related to osteoporosis and macrophage polarization-associated mQTLs and eQTLs, to explore the functional mechanism of these loci, we performed SMR analysis to quantify the regulatory effects of DNA methylation on gene expression (mQTL→ eQTL). Table 2 details the significant methylation sites that strongly regulate the expression of the DNMT3A, CTSK, and CDCA4 genes (p_SMR-multi < 0.05, p_SMR < 0.05, p-HEIDI > 0.01) (Table 2 and Table S10). Considering DNMT3A gene as an example, hypermethylation levels at three CpG sites of DNMT3A were positively associated with osteoporosis risk: cg17742416 [OR = 1.23, 95% CI (1.04–1.45)], cg08485187 [OR = 1.17, 95% CI (1.04–1.32)], and cg13558695 [OR = 1.23, 95% CI (1.07–1.40)]. Conversely, DNMT3A gene expression was negatively associated with osteoporosis risk [OR = 0.62, 95% CI (0.46–0.85)]; the methylation levels at these sites were inversely correlated with DNMT3A gene expression: cg08485187 [OR = 0.70, 95% CI (0.63–0.77)], cg13558695 [OR = 0.67, 95% CI (0.59–0.75)], and cg17742416 [OR = 0.61, 95% CI (0.52–0.73)]. These results suggest that hypermethylation at CpG sites cg08485187, cg13558695, and cg17742416 may downregulate DNMT3A gene expression, thereby increasing the risk of osteoporosis. Similarly, our analysis indicated that increased methylation at CpG site cg20817941 (CTSK) and cg10124993 (CDCA4) was associated with both higher gene expression and a reduced risk of osteoporosis. Table 2Regulatory relationship between mQTL and eQTLExpo_IDOutco_GeneSymbolp_SMR_multifdr_SMR P-valuesOR_SMR (95% CI)cg20817941ENSG00000143387 CTSK 5.96E-152.39E-148.73 (5.07–15.04)cg17742416ENSG00000119772 DNMT3A 1.67E-083.08E-080.61 (0.52–0.73)cg08485187ENSG00000119772 DNMT3A 1.79E-126.13E-120.7 (0.63–0.77)cg13558695ENSG00000119772 DNMT3A 1.06E-113.19E-110.67 (0.59–0.75)cg10124993ENSG00000170779 CDCA4 2.85E-095.71E-092.38 (1.79–3.16) Regulatory relationship between mQTL and eQTL Further integration of eQTL and pQTL data did not yield significant additional results; however, due to strong colocalization evidence supporting the HMOX1 protein (PPH4 > 0.5) and the validation of the AGER and F2 proteins in independent datasets, the genes corresponding to these three proteins should also be considered for further investigation. We used Manhattan plots to illustrate the distribution of these key methylation sites, genes, and protein results across chromosomes (Fig. 3). Locus zoom plots illustrated the distribution of significant signals (Figure S2), while SMR effect plots demonstrated their impact on osteoporosis risk (Figure S4). Results of the Manhattan plot analysis for the FinnGen R10 cohort. A Results of the Manhattan plot analysis for mQTLs. B Results of the Manhattan plot analysis for eQTLs. C Results of the Manhattan plot analysis for pQTLs Preliminary transcriptomic exploration (GSE230665) To further validate our SMR findings and explore the underlying biological functions of the identified MPRGs, we performed differential expression analysis focused on the six prioritized core candidate genes using an independent transcriptomic dataset (GSE230665). We also conducted functional enrichment and network analyses on all 77 significant genes identified across the multi-omics SMR analyses (full results are shown in Table S11). In the external validation cohort of postmenopausal osteoporosis, we observed that AGER expression was significantly lower in osteoporosis patients than in healthy controls (p = 0.0044), consistent with the protective effect identified in our SMR analysis of AGER protein levels (Fig. 4A). Interestingly, DNMT3A expression was significantly higher in patients (p = 0.018), in a direction opposite to the protective association suggested by our eQTL SMR analysis (Fig. 4B); CTSK, CDCA4, HMOX1, and F2 did not reach statistical significance. Validation and Functional Analysis of Core Genes. A and B Boxplots showing the differential expression of AGER, DNMT3A, CTSK, CDCA4, HMOX1, and F2 in the GEO dataset (GSE230665) between postmenopausal osteoporosis patients (n = 12) and healthy controls (n = 3). p-values were calculated using the Wilcoxon test. C and D GO and KEGG pathway enrichment analysis for the 77 significant MPRGs. Top 5 enriched terms are displayed, with bar color indicating p-value and length representing gene count. E PPI network of the 77 MPRGs constructed using the STRING database. Hub genes are highlighted. This network highlights the complex, multi-functional connectivity among the 77 identified MPRGs. F GeneMANIA network predicting functional interactions for the six core candidate genes To elucidate the collective function of the candidate MPRGs, we performed GO and KEGG enrichment analyses. The GO analysis revealed significant enrichment in biological processes such as ‘response to oxygen levels’ and ‘regulation of epithelial cell proliferation’, as well as molecular functions including ‘DNA-binding transcription factor binding’ (Fig. 4C, complete results in Table S12). The KEGG analysis identified significant enrichment in several key pathways, most notably ‘Insulin signaling pathway’, ‘Cellular senescence’, and ‘Hippo signaling pathway’ (Fig. 4D, complete results in Table S13). Finally, to visualize the interplay among these genes, we constructed a PPI network, which highlighted a complex web of interactions (Fig. 4E). The 10 nodes, namely APOE, BMP4, CCND1, FOXO1, PARP1, PPARG, SMAD3, STAT1, TERT, and VIM, were all identified as the most important in the 7 algorithms. It is believed that these immune checkpoint-related genes play a core regulatory role in the interaction network. A functional association network for our primary candidate genes (DNMT3A, CTSK, CDCA4, HMOX1, F2, and AGER) further illustrated their connections through physical interactions, genetic interactions, and shared pathways, particularly in regulating coagulation and epigenetic processes (Fig. 4F). Hence, our findings suggest that genetically higher DNMT3A expression in blood‑derived cells is associated with lower osteoporosis risk, consistent with a potential role for systemic or hematopoietic pathways in modulating bone remodeling; however, given that our eQTL data are from whole blood, we cannot determine whether the relevant effects occur in circulating monocytes, bone‑resident macrophages, or other blood cell types. CTSK, a key osteoclast gene, showed a genetically predicted association with osteoporosis risk based on blood‑derived eQTLs. While CTSK is classically expressed in osteoclasts, our instruments come from whole blood and therefore capture systemic CTSK regulation in circulating or blood‑resident cells; they do not localize the causal effect to osteoclasts or other bone‑resident cells. Sensitivity analyses using curated macrophage polarization gene sets To test whether our findings depended on the initial GeneCards‑derived gene pool, we repeated the SMR screens, restricting the analysis to curated macrophage polarization gene sets assembled from primary experimental studies. Specifically, we re‑ran the mQTL‑, eQTL‑, and pQTL‑based SMR analyses using (i) genes differentially expressed in canonical M1 versus M2 polarization experiments, (ii) targets and effectors of macrophage polarization master regulators, and (iii) genes enriched in inflammatory versus reparative bone marrow–derived macrophage clusters. Across these more stringent gene sets, our key candidates identified in the primary analysis, including DNMT3A, CTSK, CDCA4, and the proteins AGER and F2, remained nominally associated with osteoporosis risk and showed consistent directions of effect. These sensitivity analyses suggest that the prioritization of these genes is not solely an artifact of the broad GeneCards query but is robust to more biologically curated definitions of macrophage polarization-related genes. Discussion Leveraging a multi-omics SMR framework, our study systematically identified numerous MPRGs at the DNA methylation, gene expression, and protein levels with potential causal associations with osteoporosis. Subsequent integration and validation pinpointed a core set of candidate molecules, including the genes DNMT3A, CTSK, and CDCA4, and the proteins AGER, F2, and HMOX1, which are highlighted for their potential causal associations linking macrophage polarization to disease risk. The identified genes and proteins are prioritized candidates whose genetically predicted associations with osteoporosis require further experimental and mechanistic validation before firm causal inferences can be drawn, although some insights can be gained from the literature. By integrating GWAS with mQTL, eQTL, and pQTL data in a summary‑data-based Mendelian randomization framework, we move beyond purely observational associations between macrophage polarization and osteoporosis. Using germline variants as instrumental variables helps mitigate confounding and reverse causation, allowing us to infer genetically predicted associations between specific macrophage polarization–related genes and osteoporosis risk. While this design still relies on Mendelian randomization assumptions and cannot provide definitive proof of causality, it offers substantially stronger evidence than conventional observational studies that the implicated genes may lie on causal pathways influencing bone metabolism. Because all QTL instruments in our SMR framework were derived from whole blood, our results primarily reflect genetically mediated variation in macrophage/immune and other blood‑cell pathways that is associated with osteoporosis risk. As such, we cannot determine from these data whether the relevant causal mechanisms operate in bone‑resident macrophages and osteoclast precursors, in circulating monocytes that traffic to bone, or in other immune or stromal cells. Our findings should therefore be viewed as implicating systemic macrophage/immune‑related axes in osteoporosis, rather than pinpointing a specific bone‑resident cell type. The present integrative multi-omics analyses identified DNMT3A as significantly associated with osteoporosis, finding that higher DNMT3A expression may decrease osteoporosis risk. DNMT3A is a gene that encodes a DNA methyltransferase enzyme responsible for establishing new DNA methylation patterns, which play a crucial role in gene expression regulation and cellular differentiation [46]. DNMT3A is involved in macrophage polarization, with loss-of-function mutations accelerating atherosclerosis and skewing macrophage phenotypes towards a pro-inflammatory state [47]. Epigenetic regulation by DNMT3A influences macrophage polarization, as evidenced by its involvement in the methylation of key genes such as PSTPIP2, thereby enhancing inflammation in liver injury [48]. DNMT3A mutations also alter immune cell function in heart failure, further implicating its involvement in macrophage polarization [49]. Comparative transcriptome analysis identified DNMT3A as a gene associated with osteoporosis [50]. DNMT3A mutations are prevalent in patients with clonal hematopoiesis, which is associated with an increased risk of osteoporosis [51]. At the cellular and animal model levels, DNMT3A has been shown to regulate osteoblast differentiation and alleviate osteoporosis via the PPARγ/SCD1/GLUT1 axis. Knockout models of DNMT3A demonstrate that its loss exacerbates osteoporosis, whereas its overexpression promotes osteoblast differentiation and bone formation [52]. These findings highlight DNMT3A’s role in bone metabolism and its potential as a therapeutic target for osteoporosis. Specifically, dysregulated DNMT3A activity may lead to aberrant DNA methylation patterns in these bone cells, affecting their differentiation, proliferation, and survival, thereby contributing to bone loss and increased fracture risk in osteoporosis. It is consistent with our findings, underscoring the gene’s critical importance in the prevention and treatment of the disease. Interestingly, while our SMR analysis suggests a lifelong protective effect of genetically higher DNMT3A expression, our validation in the GEO dataset revealed that DNMT3A mRNA levels were paradoxically elevated in patients with established postmenopausal osteoporosis. This apparent discrepancy may reflect complex underlying biology rather than a contradiction. One plausible explanation is a compensatory mechanism in which the osteoporotic microenvironment triggers a feedback loop that upregulates DNMT3A expression, an ultimately insufficient attempt to counteract bone loss. Alternatively, this difference may stem from the distinct nature of the analyses: SMR reflects a systemic, lifelong genetic predisposition based on blood QTLs, whereas the GEO data provide a snapshot of the disease state in a specific patient cohort. The role of DNMT3A may be context-dependent, with baseline expression protective, while other pathological factors alter its expression during active disease. It highlights the intricate regulatory dynamics of DNMT3A in bone health and disease, warranting further investigation into its cell-type-specific roles in the bone marrow microenvironment. A notable example of the tension between genetic and expression evidence in our study is DNMT3A. SMR analyses suggest that higher DNMT3A expression is associated with lower osteoporosis risk, whereas the GEO dataset shows higher DNMT3A expression in patients than in controls. Rather than viewing the transcriptomic data as a simple ‘validation’ of the genetic result, this pattern should be interpreted as a complex biological signal. SMR reflects the effect of lifelong germline variation on DNMT3A activity, whereas the GEO data capture a cross‑sectional disease state in individuals who already have osteoporosis. One possibility is that elevated DNMT3A expression in patients represents a compensatory response of the hematopoietic and immune compartments to ongoing bone loss, rather than a primary causal driver. However, an equally plausible explanation is that the SMR finding for DNMT3A is a false positive or highly context‑dependent, and that the transcriptomic pattern more accurately reflects disease‑related regulation. This discrepancy underscores that, for some loci, our multi‑omics framework cannot yet resolve directionality at the mechanistic level and highlights the need for longitudinal and functional studies to clarify the role of DNMT3A in the macrophage-bone axis. Our study identified a potential, novel association between higher methylation at cg20817941 in the CTSK gene and a potentially increased risk of osteoporosis. CTSK is a member of the peptidase C1 family and encodes a lysosomal cysteine protease primarily expressed in osteoclasts, a type of macrophage-derived cell specialized in bone resorption [53]. CTSK is implicated in osteoclast-mediated bone resorption and tumor progression, and is involved in mediating M2 macrophage polarization in castration-resistant prostate cancer [54]. Cathepsin K is highly expressed in osteoclasts and is a key collagenase mediating bone resorption; loss-of-function leads to high BMD but brittle bone, while pharmacologic CTSK inhibition reduces bone resorption and fractures in trials (e.g., odanacatib) [55–57]. In osteoclasts, hypomethylation of CTSK regulatory regions is associated with increased CTSK expression and osteoclast function, whereas hypermethylation can repress expression, analogously to other osteoclast genes [58, 59]. In inflammatory conditions, inhibition of Notch signaling stimulates osteoclastogenesis from common trilineage progenitors, suggesting an interplay between CTSK and macrophage polarization [60]. Other studies have found that the loss of tumor suppressor genes in CTSK cells can lead to the formation of periosteal osteosarcomas at the cellular level, indicating that CTSK cells are crucial for intramembranous ossification [61]. In our SMR analysis, higher methylation at the CTSK locus was associated with increased osteoporosis risk, which may reflect methylation changes at regulatory CpGs that do not simply downregulate CTSK or affect additional nearby genes with distinct roles in bone remodeling. Therefore, future investigations may focus on elucidating the mechanisms of action of CTSK to explain the potential causal association observed here, thereby providing potential therapeutic targets for clinical research. In this study, hypermethylation at the CpG site cg10124993 of CDCA4 upregulated CDCA4 gene expression, potentially reducing osteoporosis risk; however, to date, there is no direct experimental evidence implicating CDCA4 in osteoclast or osteoblast biology, bone remodeling, or osteoporosis. Existing reports link CDCA4 mainly to tumor cell proliferation and circRNA-mediated regulatory axes in osteosarcoma, which are only indirectly related to bone homeostasis [62, 63]. CDCA4 is a gene that encodes a protein involved in cell cycle regulation and is expressed in various tissues, including bone marrow-derived macrophages. CDCA4 encodes a cell-cycle-related transcriptional regulator induced by E2F that modulates proliferation via the E2F-RB pathway [64]. CDCA4 shows a positive correlation with tumors and is overexpressed in most cancer cell types [65]. Despite the lack of direct research substantiating a clear association between CDCA4 and osteoporosis, existing studies indicate that circRNAs can downregulate CDCA4 expression, thereby restraining the proliferation of osteosarcoma cells [62]. We therefore interpret CDCA4 as a statistically prioritized, hypothesis‑generating candidate emerging from our multi‑omics SMR pipeline rather than a gene with established relevance to osteoporosis pathophysiology, and we have reduced the strength of the claims around this locus. Prospective investigations may elucidate the intricate mechanisms underlying the link between this gene and osteoporosis. Our SMR analysis identified three proteins (AGER, F2, and HMOX1) potentially linking macrophage polarization to osteoporosis risk, findings supported by pQTL validation for AGER and F2 and by strong colocalization of HMOX1 in the discovery cohort. F2 is the gene encoding prothrombin. Prothrombin’s activated form, thrombin, is a potent signaling molecule acting via protease-activated receptors (PARs) on macrophages [66]. This activation is known to strongly promote a pro-inflammatory M1 phenotype [67]. M1 macrophages release key cytokines like TNF-α and IL-6, which are well-established drivers of osteoclastogenesis (bone resorption cell formation) and inhibitors of osteoblast function (bone formation cell activity) [68], thereby fostering a net bone loss environment. Thrombin itself may also directly stimulate osteoclast activity [69]. However, our analysis revealed a negative correlation between F2 expression and osteoporosis risk. The observed negative correlation thus presents a paradox. It might indicate that the systemic levels or genetic regulation captured by SMR/pQTL analysis do not fully reflect the local activity within the bone microenvironment, or that thrombin possesses other, context-dependent roles (perhaps stimulating osteoblast activity under certain conditions or via different pathways) that result in a net protective association in this population-level analysis. Alternatively, compensatory mechanisms or complex feedback loops involving the coagulation system and inflammation in bone might be at play. Coagulation factors and their receptors modulate bone cells, with experimental data showing that thrombin receptor F2R can act as a negative regulator of osteoclastogenesis via inhibition of Akt and NF-κB signaling, thereby reducing osteoclast formation and bone resorption [70, 71]. Genetically predicted higher F2 expression in bone or osteoclast‑lineage cells could reflect a compensatory or regulatory role limiting osteoclast activity and bone resorption, which would be consistent with a protective association against osteoporosis despite the pro‑thrombotic/pro‑inflammatory reputation of the coagulation cascade [70, 71]. Similarly, AGER expression also showed a negative correlation with osteoporosis risk. AGER, an inflammatory receptor [72], generally promotes M1 polarization [73], which would also be expected to increase bone loss. Like F2, this unexpected negative correlation for AGER suggests its role in bone biology is nuanced, possibly dependent on specific ligand availability or downstream signaling pathways within the bone niche [74], differing from its canonical pro-inflammatory functions observed elsewhere. RAGE signaling generally promotes osteoclast maturation and bone resorption, and RAGE deficiency in mice leads to impaired osteoclast function and increased bone mass [75–77]. Different forms and locations of RAGE are important: membrane RAGE on bone cells drives catabolic signaling, whereas circulating soluble RAGE (sRAGE), which can act as a decoy receptor, has been inversely associated with erosive bone disease and may have protective effects [76, 78, 79]. Our genetic signal may preferentially tag regulatory variation affecting RAGE expression or splicing in specific tissues or cell types (e.g., bone marrow macrophages vs. systemic circulation), or may be linked to higher sRAGE relative to membrane RAGE, which could plausibly yield an inverse association with osteoporosis risk [76, 78]. These context‑dependent mechanisms for F2 and RAGE are hypotheses consistent with current bone biology but remain to be tested with tissue‑specific and isoform‑specific functional studies in bone cells and in vivo models. Conversely, the observed negative correlation trend between HMOX1 expression and osteoporosis risk aligns well with its known functions. HMOX1, an antioxidant enzyme, skews macrophages towards an anti-inflammatory M2 phenotype associated with tissue repair and inflammation resolution [80, 81]. Furthermore, its role in mitigating oxidative stress, a known contributor to osteoporosis, provides a direct mechanism for bone protection [80, 82]. Thus, HMOX1 likely exerts a protective effect against osteoporosis through combined M2 polarization and antioxidant pathways. Collectively, these findings highlight AGER, F2, and HMOX1 as key proteins whose influence on macrophage polarization state may significantly impact osteoporosis susceptibility, warranting further investigation into their precise roles in bone pathophysiology. Taken together, our findings refine the broadly accepted concept of an M1 shift in osteoporosis by pinpointing specific upstream regulators within macrophage polarization pathways. Prior clinical and experimental studies have shown that osteoporotic patients tend to exhibit an increase in pro‑inflammatory, M1‑like macrophages; however, they did not identify which macrophage polarization-related genes are most likely to be in the causal pathway. By integrating methylation, expression, and protein data, we identified DNMT3A, CTSK, CDCA4, HMOX1, F2, and AGER as prioritized candidates with genetically predicted associations with osteoporosis, thereby providing molecular granularity previously lacking. In particular, DNMT3A and AGER emerge as key nodes linking epigenetic regulation, macrophage function, and bone metabolism, despite limited prior direct evidence connecting these genes to osteoporosis. These results generate concrete, testable hypotheses about how specific macrophage polarization-related genes, rather than macrophage activation in general, may modulate bone remodeling. Therefore, our study does not simply reiterate the known association between macrophage polarization and osteoporosis; it dissects this axis into distinct genetic and epigenetic drivers that can now be interrogated experimentally. The incorporation of pQTL data provides a direct translational bridge from genetic association to potential drug targets and biomarkers. Whereas much of the existing macrophage literature in osteoporosis focuses on cellular phenotypes or cytokine patterns, our pQTL‑based SMR analyses prioritize circulating proteins such as HMOX1, F2, and AGER whose genetically predicted levels are associated with osteoporosis risk. Proteins are inherently more tractable as therapeutic targets and clinical biomarkers than abstract pathway components, and our findings therefore offer a starting set of macrophage polarization-related proteins that could be evaluated in pharmacologic and biomarker development programs. In this way, the study adds a layer of specificity and druggability that goes beyond generic macrophage‑directed strategies. This study applied SMR analysis, using genetic variants as instrumental variables to effectively mitigate confounding factors, thereby enabling a more precise evaluation of potential causal associations. Moreover, by combining GWAS data with eQTLs, mQTLs, and pQTLs, this research explored the connections between MPRGs and osteoporosis at various biological levels, enhancing the depth and scope of the investigation [83]. The application of statistical methods, such as the SMR and HEIDI tests, improved the accuracy of causal inference and helped identify and exclude pleiotropic effects [84]. This study has several limitations. First, all analyses were conducted using summary-level QTL and GWAS data predominantly derived from European-ancestry cohorts, which may limit generalizability to other populations and preclude individual-level adjustment for potential confounders or exploration of non-linear and interaction effects. Second, our multi-omics SMR framework relies on key Mendelian randomization and colocalization assumptions, including the absence of unmeasured horizontal pleiotropy and the presence of a single shared causal variant per locus; despite using HEIDI and Bayesian colocalization, residual pleiotropy, linkage disequilibrium structure, and selection bias in SMR instrument choice may still bias effect estimates and lead to both false-positive and false-negative findings. Third, a central limitation of our study is that all mQTL, eQTL, and pQTL instruments were derived from whole blood and serve only as proxies for regulation in disease‑relevant tissues. This design precludes us from distinguishing whether the implicated genes act in bone‑resident macrophages or osteoclasts, circulating monocytes, other immune subsets, or non‑immune blood cells. Consequently, while we interpret our findings in the context of macrophage polarization and osteoclast biology based on prior experimental literature, our genetic data cannot specify the exact cellular compartment in which these causal mechanisms operate. Fourth, the exploratory choice of a relatively liberal colocalization threshold (PPH4 > 0.5) and SMR significance cut-offs was intended to increase sensitivity and prioritize a broader set of macrophage polarization-related genes, but this comes at the cost of an increased false-discovery rate and emphasizes that our prioritized genes should be viewed as hypothesis-generating rather than definitive causal loci. Fifth, the transcriptomic validation in GSE230665 used a small sample of femoral bone from postmenopausal women (12 cases and 3 controls), which limits statistical power, may not capture the full spectrum of osteoporosis phenotypes, and could contribute to discrepancies in the direction of effect between SMR-based genetically predicted associations and observed differential expression. Sixth, functional interpretation of several prioritized genes (e.g., CDCA4 and some coagulation- or immune-related loci) is constrained by limited prior experimental work in bone biology, meaning that some mechanistic links remain speculative and extrapolated from other disease contexts, underscoring the need for targeted in vitro and in vivo studies in bone and macrophage lineages. Seventh, the lack of statistical associations for CTSK, CDCA4, HBOX, and F2 in the validation analysis could be due to the small sample size of the validation dataset. Eighth, for certain genes, such as DNMT3A, we observe a mismatch between the direction of effect inferred from SMR and the differential expression pattern in cases versus controls. These discrepancies indicate that cross‑sectional transcriptomic data in patients may reflect compensatory or downstream responses, or that the MR signal is a false positive or strongly context‑dependent. Such inconsistencies limit our ability to draw definitive mechanistic conclusions for individual genes and should be addressed in future longitudinal and experimental work. Ninth, the integrative filtering approach reflects a trade‑off: strict FDR control across all tests would substantially reduce power and likely discard biologically plausible signals in this relatively focused but still high‑dimensional setting, whereas using nominal P‑values alone would inflate false positives. By combining statistical evidence with HEIDI, colocalization, cross‑omics consistency, and orthogonal validation, we aim to minimize arbitrary selection while explicitly acknowledging that our prioritized core genes remain hypotheses that require functional follow‑up. Finally, although multi-omics integration (mQTL, eQTL, pQTL, and GWAS) strengthens biological plausibility, our study cannot by itself establish definitive causality or therapeutic efficacy; the identified macrophage polarization-related genes and pathways should therefore be considered genetically predicted associations that require validation in independent cohorts, perturbation experiments, and ultimately clinical and pharmacologic studies before translation into precision osteoporosis management. These limitations, alongside the observational nature of SMR, underscore the need for further functional validation of our results. Conclusion In summary, this multi-omics SMR study integrates blood-derived QTLs and large-scale GWAS to prioritize a core set of macrophage polarization-related genes—notably DNMT3A, CTSK, CDCA4, HMOX1, F2, and AGER—with genetically predicted associations with osteoporosis risk. Rather than definitively established drivers, these prioritized candidates highlight a complex, context-dependent regulatory network linking systemic immune signatures to bone metabolism. These findings provide a highly targeted roadmap: translating these hypothesis-generating loci into precision therapeutics will require tissue-specific (bone/marrow) QTL mapping, longitudinal clinical validation, and mechanistic studies within the bone microenvironment. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4 Supplementary Material 5 Supplementary Material 6 Supplementary Material 7 Acknowledgements None. Abbreviations Macrophage polarization related genes Quantitative trait locus Genome-wide association study UK biobank Summary data-based Mendelian randomization Mendelian randomization Author contributions (I) Conception and design: Feng Xiong, Aoqi Zang. (II) Administrative support: Hui Xu and Jianzhong Guan. (III) Provision of study materials or patients: Feng Xiong, Aoqi Zang. (IV) Collection and assembly of data: Ming Zhu, Bingxu Ma, Peitai Liu. (V) Data analysis and interpretation: Feng Qian, Jun Zhang (VI) Manuscript writing: Feng Xiong. (VII) Final approval of manuscript: All authors. Funding This study was supported by the Scientific Research Project of Bengbu Health Commission (BBWK2023A202), the Natural Science Project of Bengbu Medical University (2023byzd196), and the Bengbu City science and technology innovation guidance project (2024ZD0042, 2024ZD0040). Data availability All data generated or analysed during this study are included in this published article. Declarations Ethics approval and consent to participate This article is a Mendelian randomization study. The data for this study were obtained from publicly available databases and published literature and do not require ethical approval or written informed consent. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Feng Xiong and Aoqi Zang contributed equally to this work. Contributor Information Feng Xiong, Email: Xf15056394440@126.com. Jianzhong Guan, Email: Guanjianzhong@bbmc.edu.cn. Hui Xu, Email: bbxuhui@hotmail.com. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Supplementary Material 4 Supplementary Material 5 Supplementary Material 6 Supplementary Material 7 Data Availability Statement All data generated or analysed during this study are included in this published article.