Skip to main content
NCBI home page
NPJ Parkinson's Disease logoLink to NPJ Parkinson's Disease
. 2026 Mar 18;12:111. doi: 10.1038/s41531-026-01325-8

Integrative clinical and genomic analyses reveal a causal role of GPNMB in the bone-brain axis of Parkinson’s disease

Xingzhi Guo 1,2,✉, Peiyao Wei 1,2, Wenzhi Shi 1, Rong Zhou 1,2, Rui Li 1,2,3,✉
PMCID: PMC13150002  PMID: 41851136

Abstract

Osteokines, primarily secreted by bone, have been implicated in brain function and Parkinson’s disease (PD) pathogenesis, yet their circulating levels in PD and potential role in the relationship between bone mineral density (BMD) and PD remain unclear. 80 participants (40 PD patients and 40 controls) were enrolled to measure plasma levels of eight osteokines (GPNMB, OPN, SOST, DKK1, RANKL, FGF23, BMP2, and BMP4) and assess their associations with clinical scales. Mendelian randomization (MR), SMR, and colocalization analyses were performed to evaluate causal relationships between osteokines and PD. Restricted cubic spline (RCS) models were applied to explore nonlinear associations between BMD, osteokines, and PD. GPNMB levels were significantly elevated in PD patients and showed a linear association with PD risk. Higher GPNMB levels were associated with worse cognitive performance and clinical severity, while higher SOST levels correlated with milder symptoms. Genetic analyses consistently supported a causal and colocalized relationship between GPNMB and PD. Total coxa BMD and T-score were lower in PD, but not statistically significant. RCS analysis revealed an “n-shaped” association between total coxa T-score and both PD and GPNMB levels. Overall, GPNMB appears causally linked to PD risk and may mediate the bone-brain axis connecting BMD with PD susceptibility.

Subject terms: Diseases, Genetics, Neurology, Neuroscience

Introduction

Parkinson’s disease (PD) is a progressive neurodegenerative disorder primarily characterized by motor symptoms such as bradykinesia and resting tremor1. PD is a complex disorder influenced by both genetic and environmental factors, with approximately 5-10% of cases caused by mutations in genes such as synuclein alpha (SNCA), leucine-rich repeat kinase 2 (LRRK2), and glucocerebrosidase (GBA)2,3. Numerous additional PD susceptibility loci have been identified by genome-wide association studies (GWAS)4,5, although the biological functions of most remain unclear6. Increasing evidence suggests that PD is also associated with alterations in bone metabolism7–9. Epidemiological and clinical studies have reported a higher prevalence of reduced bone mineral density (BMD) and osteoporosis among PD patients10–13. Moreover, recent prospective studies suggest that lower BMD or a diagnosis of osteoporosis is associated with an elevated risk of developing PD10. While factors such as vitamin D deficiency, immobility, and chronic inflammation have been proposed to underlie this link11,14, the biological mechanisms connecting bone health and PD remain poorly defined.

Osteokines, a class of bone-derived signaling molecules, such as sclerostin, osteopontin (OPN), bone morphogenetic proteins (BMPs), and receptor activator of nuclear factor kappa-Β ligand (RANKL), are mainly secreted by bone cells and are critical regulators of bone remodeling15,16. Emerging evidence suggests that osteokines may also influence the central nervous system (CNS), potentially contributing to neurodegenerative processes17,18. For example, elevated plasma and cerebrospinal fluid (CSF) OPN levels have been reported in PD and linked to disease severity19,20, while osteocalcin may exert neuroprotective effects21. OPN can activate inflammatory pathways such as NF-κB, MAPK, and ERK, which might contribute to neuroinflammation and neuronal dysfunction22. Sclerostin (SOST) has been reported to cross the blood–brain barrier (BBB) and influence brain function23, potentially through passive diffusion when the BBB is compromised or via active transport mechanisms involving endothelial receptors or extracellular vesicles24,25. Our previous studies also showed that osteocalcin exerted a protective effect on animal models of neurodegenerative diseases21,26. Glycoprotein non-metastatic melanoma protein B (GPNMB), also known as osteoactivin, is another osteokine involved in osteoclast and osteoblast differentiation and has also been implicated in neuroinflammation and PD pathogenesis27,28. Together, these findings suggest that osteokines might be implicated in the bone-brain axis in PD. However, most evidence linking BMD, circulating osteokines, and PD is obtained from observational studies and is therefore susceptible to confounding and reverse causation29, leaving their causal relationships poorly understood. By leveraging germline genetic variants from GWAS30, genetic approaches such as Mendelian randomization (MR) and colocalization analyses could mitigate limitations of observational studies and enable assessment of whether alterations in circulating osteokines are causally implicated in PD risk rather than simply reflecting disease-related changes.

In this study, we aim to identify osteokines that are differentially expressed in PD patients and investigate their associations with disease severity. We further assess whether these osteokines are causally related to PD risk using genetic approaches, including MR, summary-data-based MR (SMR), and colocalization analyses. In addition, we explore whether osteokines are implicated in the relationship between BMD and PD. By integrating clinical and genetic data, this study seeks to provide novel insights into the role of the bone-brain axis in the pathophysiology of PD.

Results

Baseline characteristics of the study participants

The demographic and clinical characteristics of the participants are summarized in Table 1. There were no significant differences between the PD and control groups in terms of age, sex, body mass index (BMI), fasting glucose, or lipid profiles. The mean age was 64.95 ± 5.8 years in the PD group and 65.45 ± 5.88 years in the control group, with males accounting for 47.5% and 50% for the PD group and control group, respectively. The mean disease duration was 4.68 ± 4.36 years, and the mean Hoehn & Yahr stage for the PD patients was 2.26 ± 0.88. PD patients exhibited significantly lower MoCA scores compared to controls (19.29 ± 6.04 vs. 22.64 ± 5.20, P = 0.018), indicating impaired cognitive function.

Table 1.

Population characteristics according to the baseline PD status

Variable Overall CON PD P-value
n 80 40 40
Age (mean (SD)) 65.20 (5.81) 65.45 (5.88) 64.95 (5.80) 0.703
Gender = 1 (%) 39 (48.8) 20 (50.0) 19 (47.5) 1.000
BMI (mean (SD)) 23.39 (3.05) 23.58 (2.41) 23.21 (3.60) 0.588
SBP (mean (SD)) 126.28 (14.44) 127.62 (15.43) 124.92 (13.44) 0.407
DBP (mean (SD)) 76.69 (9.22) 77.08 (8.14) 76.30 (10.28) 0.710
Disease duration (year, mean (SD) - - 4.68 (4.36) -
Smoke = 1 (%) 20 (25.0) 10 (25.0) 10 (25.0) 1.000
Alcohol = 1 (%) 16 (20.0) 5 (12.5) 11 (27.5) 0.162
Diabete = 1 (%) 19 (23.8) 12 (30.0) 7 (17.5) 0.293
Glu (mean (SD)) 5.11 (1.41) 5.26 (1.51) 4.97 (1.31) 0.363
TC (mean (SD)) 4.22 (0.99) 4.08 (1.00) 4.37 (0.98) 0.191
TG (mean (SD)) 1.34 (0.85) 1.44 (0.92) 1.23 (0.77) 0.278
HDL (mean (SD)) 1.22 (0.26) 1.17 (0.22) 1.29 (0.29) 0.039
LDL (mean (SD)) 2.37 (0.74) 2.29 (0.71) 2.45 (0.76) 0.347
Hcy (mean (SD)) 17.33 (14.68) 15.56 (8.18) 19.14 (19.16) 0.288
Hoehn Yahr (mean (SD)) - - 2.26 (0.88) -
MMSE (mean (SD)) 25.01 (4.43) 25.76 (4.37) 24.29 (4.43) 0.178
MoCA (mean (SD)) 20.94 (5.84) 22.64 (5.20) 19.29 (6.04) 0.018
Open in a new tab

PD Parkinson’s disease, BMI body mass index, SBP systolic blood pressure, DBP diastolic blood pressure, Glu glucose, TC total cholesterol, TG triglycerides, HDL high-density lipoprotein, LDL low-density lipoprotein, Hcy homocysteine, MMSE mini-mental state examination, MoCA Montreal cognitive assessment, SD standard error.

Plasma osteokine levels in PD and control groups

Plasma levels of GPNMB were significantly elevated in PD patients compared to healthy controls (28953.92 ± 5869.69 pg/ml vs. 24553.80 ± 5923.55 pg/ml, Padj = 0.005). In contrast, plasma BMP2 levels were markedly reduced in PD patients (21.37 ± 1.13 pg/ml vs. 22.49 ± 1.50 pg/ml, Padj = 0.002), with concentrations close to the assay detection limit. No significant group differences were observed for DKK1, SOST, RANKL, OPN, FGF23, or BMP4 (Fig. 1). To assess the robustness of our results, we further performed regression analyses adjusting for age and sex, showing that GPNMB and BMP2 remained significantly different between the PD and control groups (Supplementary Table S1). In additional sex-stratified analyses, plasma GPNMB levels remained significntly elevated in female PD patients (Padj = 0.031), whereas the increase did not reach statistical significance in males (P = 0.062, Padj = 0.252). In contrast, BMP2 levels were significantly reduced in male PD patients (Padj = 0.003) but not in females (P = 0.120, Padj = 0.320). Given the limited sample size, these subgroup findings should be interpreted with caution, and the detailed sex-specific results are presented in Supplementary Figs. S1 and S2. Due to BMP2 concentrations being close to the lower limit of detection, subsequent analyses focused primarily on GPNMB.

Fig. 1. Violin plots showing differences in plasma osteokine levels between patients with Parkinson’s disease and healthy controls.

Fig. 1

Open in a new tab

Each dot represents an individual sample. Abbreviations: GPNMB glycoprotein non-metastatic melanoma protein B, SOST sclerostin, DKK1 Dickkopf-related protein 1, OPN osteopontin, FGF23 fibroblast growth factor 23, RANKL receptor activator of nuclear factor κB ligand, BMP2 bone morphogenetic protein 2, BMP4 bone morphogenetic protein 4, Padj a corrected P-value for multiple comparisons.

Associations between osteokines, cognitive function, and PD severity

Cognitive function was assessed using the MMSE and MoCA, while PD severity was evaluated with the MDS-UPDRS scale (Parts 1–4). The results showed that higher plasma GPNMB levels were significantly associated with greater impairment in activities of daily living (higher MDS-UPDRS2: r = 0.354, P = 0.027). In contrast, higher plasma SOST levels were associated with better cognitive scores (higher MMSE scores: r = 0.268, P = 0.028, and higher MoCA scores: r = 0.416, P = 0.0005), less non-motor symptoms (lower MDS-UPDRS1: r = −0.408, P = 0.010), and milder motor symptoms (lower MDS-UPDRS3: r = −0.321, P = 0.043). Plasma BMP4 levels were negatively correlated with PD motor severity (higher MDS-UPDRS3: r = −0.347, P = 0.028) but not with cognitive function (Fig. 2). The correlation did not remain significant after adjustment for age and sex (Supplementary Fig. S3). The results of partial Spearman analyses adjusting for age and sex showed that GPNMB levels were also correlated with poorer cognitive performance (lower MMSE: r = −0.289, P = 0.019), greater impairment in activities of daily living (higher MDS-UPDRS2: r = 0.369, P = 0.024), and more severe motor symptoms (higher MDS-UPDRS3: r = 0.395, P = 0.014) (Supplementary Fig. S3). Other osteokines, including OPN, BMP2, RANKL, FGF23, and DKK1, showed no significant correlations with cognitive or clinical severity scores (Supplementary Tables S2 and S3).

Fig. 2. Spearman correlation analysis between plasma osteokine levels and clinical scores.

Fig. 2

Open in a new tab

The size of each circle represents the strength of the correlation, while the color indicates the direction (positive or negative). Abbreviations: GPNMB glycoprotein non-metastatic melanoma protein B, SOST sclerostin, DKK1 Dickkopf-related protein 1, OPN osteopontin, FGF23 fibroblast growth factor 23, RANKL receptor activator of nuclear factor κB ligand, BMP2 bone morphogenetic protein 2, BMP4 bone morphogenetic protein 4, MMSE Mini-Mental State Examination, MoCA Montreal Cognitive Assessment, MDS-UPDRS Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale. *P < 0.05.

Mendelian randomization and SMR analyses

Instrumental variables were only available for five osteokines (GPNMB, OPN, DKK1, SOST, and FGF23), while RANKL, BMP2, and BMP4 lacked valid genetic instruments passing IVs selection criteria described in the method section (Supplementary Table S4). MR analysis revealed a significant causal relationship between higher plasma GPNMB levels and increased PD risk. Using PD GWAS data from the IPDGC, the IVW method estimated an odds ratio (OR) of 1.429 (95% CI: 1.245–1.640, Padj = 3.8e-06) per SD increase in plasma GPNMB. This result was replicated using the FinnGen GWAS dataset (OR = 1.235, 95% CI: 1.092–1.397, Padj = 3.9e-03). Additional MR-Egger and weighted median methods yielded consistent results (Supplementary Table S5). No causal effects were observed between the other four osteokines (OPN, DKK1, SOST, and FGF23) and PD (Fig. 3A). SMR analysis further confirmed the association between plasma GPNMB (pQTL) and PD risk across both datasets (IPDGC: bSMR = 0.347, PSMR < 0.001, PHEIDI = 0.06; and FinnGen: bSMR = 0.203, PSMR = 0.008, PHEIDI = 0.104) (Fig. 3B–E). Similar findings were obtained using whole blood eQTL data from the GTEx project (Supplementary Fig. S4). The HEIDI tests indicate no obvious heterogeneity (P > 0.05), suggesting the associations were unlikely due to linkage disequilibrium.

Fig. 3. Causal relationship between circulating osteokine levels and PD risk based on MR and SMR analyses.

Fig. 3

Open in a new tab

A MR analysis identified a significant causal association between plasma GPNMB levels and increased PD risk (*P < 0.05), whereas no such associations were observed for OPN, DKK1, SOST, or FGF23. B–E SMR analysis further confirmed the association between GPNMB and PD using GWAS datasets from the IPDGC (B, C) and FinnGen consortium (D, E). Abbreviations: PD Parkinson’s disease, GPNMB glycoprotein non-metastatic melanoma protein B, SOST sclerostin, DKK1 Dickkopf-related protein 1, OPN osteopontin, FGF23 fibroblast growth factor 23, IPDGC International Parkinson’s Disease Genomics Consortium, MR Mendelian randomization, SMR summary-data-based Mendelian randomization, Padj a corrected P-value for multiple comparisons.

Colocalization analysis between GPNMB and PD

To determine whether plasma GPNMB levels and PD share a common genetic basis, colocalization analysis was performed using a ± 50 kb window centered on the GPNMB locus. The posterior probability for hypothesis 4 (PP.H4 = 0.76) indicated a high likelihood that the same causal variant drives both elevated plasma GPNMB levels and increased PD risk (Fig. 4). The lead SNP rs10250602 exhibited a high SNP-specific posterior probability (SNP.PP.H4 = 0.995, Supplementary Fig. S5). These findings support the hypothesis that genetic regulation of GPNMB contributes directly to PD susceptibility, indicating a shared genetic etiology and implicating GPNMB as a causal mediator of PD risk.

Fig. 4. Colocalization analysis between PD risk and GPNMB-associated genetic variants.

Fig. 4

Open in a new tab

Stacked plots display colocalization signals at the GPNMB locus, highlighting the lead variant rs10250602 with a posterior probability for shared causal association (SNP.PP.H4) > 0.99. The Y-axis represents the -log10 (p-values) from genome-wide association studies (GWAS), and the X-axis indicates genomic position. Abbreviations: GPNMB glycoprotein non-metastatic melanoma protein B, PD Parkinson’s disease.

Nonlinear associations among BMD, GPNMB, and PD

To assess the potential role of GPNMB in mediating the relationship between BMD and PD, we analyzed total coxa BMD data available from 44 participants (24 PD patients, 20 controls). Although BMD was lower in the PD group (0.81 ± 0.11 g/cm3) compared to controls (0.87 ± 0.17 g/cm3), the difference did not reach statistical significance. The same result was observed for T-scores (Fig. 5A). RCS analysis revealed a nonlinear, “n-shaped” association between total coxa T-scores and PD risk (P = 0.033), with the spline-derived nadir occurring at a T-score of -1.038, indicating individuals with moderately higher T-scores (threshold = −1.038) had a lower risk of PD (Fig. 5B). A similar “n-shaped” pattern was observed between total coxa T-scores and plasma GPNMB levels (P = 0.007), with a nadir at -0.725 (Fig. 5C). Given the limited sample size and to avoid overfitting a single spline-derived turning point, we used the average of these two model-derived nadirs (−0.85 = −(1.038 + 0.725)/2) as the post-hoc stratification threshold. Based on this threshold, 20 participants had T-scores > −0.85 and 24 participants had T-scores ≤ -0.85. Consistent with the RCS curves, Spearman correlation analysis showed that GPNMB levels were inversely correlated (r = −0.52, P = 0.022) with T-scores in the higher-T-score group (T-scores > −0.85), but positively correlated (r = 0.46, P = 0.020) in the lower-T-score group (T-scores ≤ −0.85) (Fig. 5D). These findings suggest a complex, non-linear interaction among BMD, GPNMB, and PD risk. No significant non-linear relationship was detected between plasma GPNMB levels and PD risk (Supplementary Fig. S6). All RCS models were adjusted for age, sex, and BMI.

Fig. 5. Associations among bone mineral density (BMD), plasma GPNMB levels, and Parkinson’s disease (PD).

Fig. 5

Open in a new tab

A Violin plots showing differences in total coxa T-scores and BMD (g/cm3) between PD patients and healthy controls. Each dot represents an individual sample (ns not significant). B Restricted cubic spline (RCS) analysis revealed a nonlinear association between total coxa T-scores and PD risk. C RCS analysis also demonstrated a nonlinear association between total coxa T-scores and plasma GPNMB levels. D Spearman correlation analysis between total coxa T-scores and plasma GPNMB levels stratified by average T-score threshold (−0.85). Abbreviations: RCS restricted cubic spline, GPNMB glycoprotein non-metastatic melanoma protein B, BMD bone mineral density, PD Parkinson’s disease.

Discussion

In this study, we comprehensively investigate the associations between circulating osteokines and PD, with a particular focus on the interplay between BMD, GPNMB, and PD risk. Our findings indicate that plasma levels of GPNMB, but not other osteokines, were significantly elevated in PD patients, and genetic analyses consistently supported a causal role of GPNMB in PD susceptibility. Additionally, BMD exhibited a nonlinear “n-shaped” association with both plasma GPNMB levels and PD risk, suggesting GPNMB as a potential molecular biomarker linking bone health to PD.

Previous studies have reported that the GPNMB levels were increased in PD patients31,32. Similarly, our data also showed that the plasma GPNMB levels were significantly increased in PD patients and correlated with the cognitive function and PD severity. Additionally, GPNMB was also found to be elevated in the CSF of PD patients, which was associated with microglial activation32. Among the eight osteokines assessed in this study, genetic analyses showed that only plasma GPNMB demonstrated robust and replicable associations with PD risk. Importantly, colocalization analysis further revealed that plasma GPNMB levels and PD share a common genetic variant, reinforcing its causal relevance. Our analyses identified rs10250602 as the top shared variant underlying both plasma GPNMB levels and PD risk. Although rs10250602 has not been previously reported as a PD risk SNP, its strong cis-pQTL effect on GPNMB and the observed colocalization with PD risk in our study suggest that this variant may influence PD susceptibility through GPNMB-mediated lysosomal and neuroinflammatory pathways, consistent with prior functional evidence on GPNMB biology27,28,33, warranting future experimental validation. Previous studies have indicated that GPNMB was involved in microglial activation, neuroinflammation, and lysosomal dysfunction, all of which are key features of PD pathogenesis27,32. Notably, GPNMB was found to be upregulated in activated microglia in the substantia nigra and has been implicated in α-synuclein-induced neurotoxicity by interaction with α-synuclein28,33. The lack of a nonlinear association between GPNMB levels and PD further supports a dose-dependent effect, suggesting that higher plasma concentrations are consistently associated with increased PD susceptibility. These findings raise the possibility of plasma GPNMB serving as a potential biomarker for early PD detection or a target for disease-modifying therapies.

Previous literature has reported alterations in several osteokines such as FGF23, OPN, and DKK1 in PD patients18,19,34. However, the present study did not observe significant differences in the plasma levels of these osteokines between PD patients and controls. The methodological differences in osteokine quantification and sample size might account for the discrepancies35. Nevertheless, it is also worth noting that some osteokines showed clinical relevance with PD. For instance, higher SOST levels were associated with better cognitive performance (higher MMSE and MoCA scores), fewer non-motor symptoms (lower MDS-UPDRS1 scores), and milder motor impairment (lower MDS-UPDRS3 scores) in PD patients. Indeed, previous studies have shown that SOST can cross the BBB and modulate Aβ production and synaptic plasticity in Alzheimer’s disease and is associated with cognitive decline and neuropathologies23,36. In addition, higher GPNMB levels were associated with worse motor performance, while higher BMP4 levels were associated with milder motor symptoms. Moreover, partial Spearman correlation analysis revealed that plasma GPNMB levels were associated with both cognitive function and PD severity after adjusting for age and sex, suggesting potential confounding effects of age and sex on the relationships between osteokine levels and PD. These findings suggest functional roles for these molecules in disease progression, even if their plasma levels are not altered on average in PD.

Although no significant difference was observed between PD patients and controls in BMD, RCS analysis revealed a significant nonlinear “n-shaped” association between BMD and PD risk. Specifically, when BMD T-scores were above approximately −0.85, higher BMD was associated with reduced plasma GPNMB levels and lower PD risk. These data suggest that preserving BMD within the normal range may have protective effects against PD. The non-linear relationship between BMD and GPNMB levels might be due to the biphasic role of GPNMB in bone metabolism37,38. For example, previous studies have demonstrated that GPNMB can promote osteoclast differentiation and bone resorption39. Conversely, GPNMB overexpression has also been shown to stimulate osteoblast progenitor differentiation and bone formation40. Furthermore, the lack of association between BMD and PD among individuals with T-scores below -1 might be attributable to the use of anti-osteoporosis medications, which have been shown to reduce PD risk in postmenopausal women10. However, this remains a hypothesis that requires confirmation through future studies incorporating detailed medication data and direct experimental validation. Overall, the observed “n-shaped” associations between BMD, GPNMB, and PD suggest that BMD might modulate systemic GPNMB levels, which in turn influence the risk of PD, but the underlying biological mechanisms remain to be determined. Future mechanistic and longitudinal studies will be necessary to clarify whether alterations in bone health modulate GPNMB signaling pathways relevant to PD pathogenesis, which might serve as a potential novel therapeutic strategy for PD.

While our study provides important insights, several limitations should be addressed here. First, given the modest sample size, especially the limited number of participants with available BMD data, this study may have been underpowered to detect moderate associations, thereby increasing the likelihood of Type II errors. Therefore, the non-significant findings involving osteokines other than GPNMB should be interpreted cautiously, and the apparent specificity of GPNMB in our results should not be overstated. Further studies with larger and well-characterized cohorts are essential to confirm these observations and clarify the contributions of different osteokines in PD. Second, the cross-sectional design of this study restricts its ability to infer temporal or causal relationships between BMD, plasma GPNMB levels, and PD risk. Longitudinal studies are warranted to investigate how dynamic changes in BMD and GPNMB levels relate to PD. Third, our findings from both observational and genetic analyses support a potential causal role for plasma GPNMB in PD and suggest its involvement in the bone-brain axis. However, experimental studies are required to explore the underlying biological mechanisms, which might open new avenues for therapeutic intervention. Fourth, our analyses were restricted to plasma GPNMB levels and did not assess the tissue-specific expression and effect of GPNMB in PD. Fifth, due to the inherent limitations of the Luminex Human Discovery Assay panel, we were only able to simultaneously quantify eight osteokines. Other potentially relevant bone-derived factors, such as osteocalcin and OPG, were not included in this study and warrant future investigation. Sixth, although sex-stratified analyses suggested possible sex-specific patterns in GPNMB and BMP2 alterations, the limited sample size substantially restricts the strength of these inferences. Thus, these sex-specific trends should be interpreted cautiously, and larger-scale studies are needed to determine whether these associations reflect true biological sex differences. Seventh, the genetic analyses in this study were based on GWAS, pQTL, and eQTL data from individuals of European ancestry, whereas our clinical participants were Chinese, which may limit the generalizability of the genetic findings across populations. Therefore, our genetic findings should be interpreted with caution when extrapolated to Chinese or other East Asian populations, and future studies using population-matched genetic and clinical data are warranted. Finally, lifestyle-related confounders, including dietary habits, physical activity, and vitamin D status, factors known to influence both BMD and PD risk, were not comprehensively adjusted in our analyses, which may introduce bias. Future studies should incorporate detailed assessments of lifestyle factors, such as dietary calcium and vitamin D intake, sunlight exposure, and physical activity levels, using validated questionnaires or wearable devices.

In summary, this study indicates a complex interplay between BMD, plasma GPNMB levels, and PD risk, suggesting that GPNMB might serve as a potential mediator linking bone health and PD. These findings support the existence of a bone-brain axis in PD pathophysiology and suggest GPNMB as a potential biomarker and therapeutic target. Future studies involving larger, longitudinal cohorts and mechanistic investigations are needed to confirm these associations and explore their translational relevance in clinical practice.

Methods

Participants

This study included 80 participants recruited in September 2024 from the clinic of Shaanxi Provincial People’s Hospital, comprising 40 patients diagnosed with PD and 40 healthy controls from the same hospital health screening programs without neurological disorders or a family history, and were not related to the PD patients. PD patients were diagnosed according to the established clinical criteria by the 2015 Movement Disorder Society clinical diagnostic criteria41. Participants with malignancies, stroke, thyroid or parathyroid disorders, renal dysfunction, or active infectious or systemic inflammatory diseases were excluded. Participants with a clinical diagnosis of major psychiatric disorders, such as major depressive disorder, bipolar disorder, or schizophrenia, were excluded. Cognitive function was evaluated using both the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA). PD severity was assessed using the Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS), including four parts: MDS-UPDRS1 (non-motor experiences of daily living), MDS-UPDRS2 (motor experiences of daily living), MDS-UPDRS3 (motor examination), and MDS-UPDRS4 (motor complications)42,43. Informed consent was obtained from all participants. This study was performed in accordance with the Declaration of Helsinki. The study protocol was approved by the Ethics Committee of Shaanxi Provincial People’s Hospital and is part of the Registration Research of Aging and Neurodegenerative Diseases (RAND), registered with the Chinese Clinical Trial Registry (ChiCTR2400087921).

Plasma osteokine measurement

Fasting venous blood samples (5 ml) were collected from each participant in the morning. Plasma was isolated by centrifugation at 3000 × g for 10 min and stored at −80 °C until analysis. Concentrations of eight osteokines, including GPNMB, SOST, OPN, BMP2, BMP4, RANKL, Dickkopf-related protein 1 (DKK1), and fibroblast growth factor 23 (FGF23), were quantified using the Luminex Human Discovery Assay panel (Catalog: LXSAHM-08, R&D Systems). All assays were run on a Luminex MAGPIX platform with xPONENT software (Merck & Millipore) following the manufacturer’s protocols.

Bone mineral density measurement

Bone mineral density (BMD) at the total hip (coxa) was measured using dual-energy X-ray absorptiometry (DXA) on an Osteocore system (Medilink, France) in Shaanxi Provincial People’s Hospital. Scans were performed by certified technicians following standardized procedures. BMD values were expressed in grams per square centimeter (g/cm3), and T-scores were calculated based on a reference population. Quality assurance procedures, including regular calibration of the DXA system, were implemented to ensure measurement accuracy.

GWAS data sources

Publicly available genome-wide association study (GWAS) summary statistics were used for genetic analyses (Supplementary Table S6). Cis-pQTLs associated with plasma GPNMB levels were derived from a GWAS of 35,698 individuals of European ancestry conducted by the deCODE consortium44. Cis-eQTLs for GPNMB expression in whole blood were obtained from the Genotype-Tissue Expression (GTEx) project (v8, N = 755)45. For PD outcome data, we used two large-scale GWAS datasets from the International Parkinson’s Disease Genomics Consortium (IPDGC) and the FinnGen consortium (release 12), comprising up to 482,730 and 500,348 individuals of European ancestry, respectively46,47. Genetic analysis was performed using publicly available data, and no ethical approval was required.

Statistical analysis

Clinical characteristics between PD patients and controls were compared using independent t-tests for continuous variables and chi-square tests for categorical variables. Continuous data are presented as mean ± standard deviation (SD), and categorical data as percentages. Restricted cubic spline (RCS) modeling was used to assess potential nonlinear relationships among BMD, GPNMB, and PD, adjusting for age, sex, and body mass index (BMI) using the rms R package (v7.0.0). To minimize variance and avoid overfitting a single model in a small sample size, the T-score threshold selected for post-hoc stratified analysis was based on the average values of turning points identified in the above RCS models48. Due to the small sample size for PD patients, Spearman correlation analysis was used to evaluate the association between osteokines and clinical scores49,50. Age- and sex-adjusted partial Spearman analyses were additionally performed to account for potential confounding effects. Missing values were handled by case-wise deletion, meaning that specific missing values were excluded from individual analyses without removing the entire participant from other relevant analyses51. To correct for multiple comparisons in plasma osteokine levels, the false discovery rate (FDR) correction was applied. A corrected P-value (Padj) < 0.05 was considered statistically significant.

Mendelian randomization analysis

Two-sample Mendelian randomization (MR) was performed using the TwoSampleMR package (v0.6.11) to investigate the causal relationship between plasma osteokine levels and PD risk. Instrumental variables (IVs) were selected from cis-regulatory single-nucleotide polymorphisms (SNPs) strongly (P < 5E-08) and independently (r2 < 0.001, 10,000 kb window) associated with osteokine levels. Steiger filtering was applied to exclude SNPs explaining more variance in the outcome than the exposure (P < 0.05), and SNPs with F-statistics less than 10 were also excluded. Palindromic SNPs with intermediate allele frequencies were removed during harmonization. Causal estimates were calculated using inverse-variance weighted (IVW) as the primary method, with MR-Egger and weighted median methods as sensitivity analyses.

SMR analysis

To further validate the causal association between GPNMB levels and PD, we performed summary-data-based Mendelian randomization (SMR) analysis integrating plasma GPNMB pQTL and whole blood eQTL data with PD GWAS summary statistics. Analyses were conducted using the SMR software (v1.3.1) with default settings. The heterogeneity in dependent instruments (HEIDI) test was applied to assess potential linkage. Genes with a significant SMR P-value (P < 0.05) and a non-significant HEIDI test (P > 0.05) were considered as putative causal genes.

Colocalization analysis

Bayesian colocalization analysis was performed using the coloc R package (v5.2.3) to determine whether the same genetic variant drives the associations between plasma GPNMB levels and PD risk. The analysis focused on a ±50 kb window around the GPNMB gene. Posterior probabilities (PP) were calculated for five hypotheses (H0-H4): no association with either trait (H0), association with only pQTL (H1), only with PD (H2), both but with different variants (H3), and both traits sharing a common causal variant (H4). A posterior probability (PP.H4) > 0.7 was considered strong evidence of a shared causal variant between GPNMB and PD. All statistical analyses were performed in R software (v4.4.3).

Supplementary information

Supplementary Tables and Figures (2.5MB, pdf)

Acknowledgements

This work was supported by the Natural Science Basic Research Plan in Shaanxi Province of China (No. 2024JC-YBQN-0799) and the Shaanxi Provincial Innovation of Healthcare Program (No. 2024PT-02). The authors thank the IPDGC, FinnGen, deCODE, and GTEx consortium, and all the participants and investigators who made the summary-level data publicly available.

Author contributions

X.G. and R.L. conceived and designed the project. P.W., W.S., R.Z., and X.G. collected samples and clinical data. X.G. and W.S. analyzed the data. X.G. drafted the manuscript. R.L. revised the manuscript. All authors reviewed and approved the final manuscript.

Data availability

The GWAS summary statistics data used in this MR study are publicly available which are included in the additional files. Clinical data are available from the corresponding author upon reasonable request, subject to scientific review and the completion of a material transfer agreement.

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.

Contributor Information

Xingzhi Guo, Email: guo.heinz@hotmail.com.

Rui Li, Email: rli@nwpu.edu.cn.

Supplementary information

The online version contains supplementary material available at 10.1038/s41531-026-01325-8.

References

  • 1.Morris, H. R., Spillantini, M. G., Sue, C. M. & Williams-Gray, C. H. The pathogenesis of Parkinson’s disease. Lancet403, 293–304 (2024). [DOI] [PubMed] [Google Scholar]
  • 2.Ortega, R. A. et al. Association of dual LRRK2 G2019S and GBA variations with parkinson disease progression. JAMA Netw. Open4, e215845 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Landoulsi, Z. et al. Large-scale copy number variant analysis in genes linked to Parkinson´s disease. npj Parkinsons Dis.11, 225 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Fiorini, M. R., Li, J., Fon, E. A., Farhan, S. M. K. & Thomas, R. A. Neural networks reveal novel gene signatures in Parkinson disease from single-nuclei transcriptomes. npj Parkinsons Dis.11, 304 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Sharma, K. et al. A Novel PINK1 p.F385S loss-of-function mutation in an indian family with Parkinson’s disease. Mov. Disord.39, 1217–1225 (2024). [DOI] [PubMed] [Google Scholar]
  • 6.Yang, J., Wu, X. & Song, Y. Recent advances in novel mutation genes of Parkinson’s disease. J. Neurol.270, 3723–3732 (2023). [DOI] [PubMed] [Google Scholar]
  • 7.Torsney, K. M. et al. Bone health in Parkinson’s disease: a systematic review and meta-analysis. J. Neurol. Neurosurg. Psychiatry85, 1159–1166 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Koo, H. Y. et al. Fracture risk in Parkinson’s disease according to its severity and duration. Osteoporos. Int.34, 81–89 (2023). [DOI] [PubMed] [Google Scholar]
  • 9.Liu, B. et al. Bone mineral density and related scores in Parkinson’s disease: a systematic review and meta-analysis. World Neurosurg.146, e1202–e1218 (2021). [DOI] [PubMed] [Google Scholar]
  • 10.Park, K. Y. et al. Bone mineral density and the risk of Parkinson’s disease in postmenopausal women. Mov. Disord.38, 1606–1614 (2023). [DOI] [PubMed] [Google Scholar]
  • 11.Van den Bos, F. et al. Bone mineral density and vitamin D status in Parkinson’s disease patients. J. Neurol.260, 754–760 (2013). [DOI] [PubMed] [Google Scholar]
  • 12.Kwon, M. J. et al. The occurrence of Alzheimer’s disease and Parkinson’s disease in individuals with osteoporosis: a longitudinal follow-up study using a National Health screening database in Korea. Front Aging Neurosci.13, 786337 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Feng, S. H., Huang, Y. P., Yeh, K. C. & Pan, S. L. Osteoporosis and the risk of Parkinson’s disease: a nationwide, propensity score-matched, longitudinal follow-up study. J. Clin. Endocrinol. Metab.106, e763–e771 (2021). [DOI] [PubMed] [Google Scholar]
  • 14.Lima, D. P. et al. Osteoporosis in Parkinson’s disease and the role of lean body mass: a cross-sectional study in a Brazilian tertiary center. Front. Endocrinol.15, 1326212 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Zaidi, M., Zaidi, S. & Yuen, T. Understanding osteokine biology. Cell Metab.36, 888–890 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Liang, W. et al. An integrated multi-omics analysis reveals osteokines involved in global regulation. Cell Metab.36, 1144–1163.e1147 (2024). [DOI] [PubMed] [Google Scholar]
  • 17.Gerosa, L. & Lombardi, G. Bone-to-brain: a round trip in the adaptation to mechanical stimuli. Front. Physiol.12, 623893 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Lin, Y. et al. Bone-derived factors as potential biomarkers for Parkinson’s disease. Front. Aging Neurosci.13, 634213 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Maetzler, W. et al. Osteopontin is elevated in Parkinson’s disease and its absence leads to reduced neurodegeneration in the MPTP model. Neurobiol. Dis.25, 473–482 (2007). [DOI] [PubMed] [Google Scholar]
  • 20.Cappellano, G. et al. The Yin-Yang of osteopontin in nervous system diseases: damage versus repair. Neural Regen. Res.16, 1131–1137 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Guo, X. et al. Causal effect of blood osteocalcin on the risk of Alzheimer’s disease and the mediating role of energy metabolism. Transl. Psychiatry14, 205 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Kahles, F., Findeisen, H. M. & Bruemmer, D. Osteopontin: a novel regulator at the cross roads of inflammation, obesity and diabetes. Mol. Metab.3, 384–393 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Shi, T. et al. Osteocyte-derived sclerostin impairs cognitive function during ageing and Alzheimer’s disease progression. Nat. Metab.6, 531–549 (2024). [DOI] [PubMed] [Google Scholar]
  • 24.Jiang, Y. L. et al. The Protective effects of osteocyte-derived extracellular vesicles against Alzheimer’s disease diminished with aging. Adv. Sci.9, e2105316 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Lajoie, J. M. & Shusta, E. V. Targeting receptor-mediated transport for delivery of biologics across the blood-brain barrier. Annu. Rev. Pharm. Toxicol.55, 613–631 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Guo, X. Z. et al. Osteocalcin ameliorates motor dysfunction in a 6-hydroxydopamine-induced Parkinson’s disease rat model through AKT/GSK3β signaling. Front Mol. Neurosci.11, 343 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Gillett, D. A., Wallings, R. L., Uriarte Huarte, O. & Tansey, M. G. Progranulin and GPNMB: interactions in endo-lysosome function and inflammation in neurodegenerative disease. J. Neuroinflammation20, 286 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Diaz-Ortiz, M. E. et al. GPNMB confers risk for Parkinson’s disease through interaction with α-synuclein. Science377, eabk0637 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Pingault, J. B. et al. Using genetic data to strengthen causal inference in observational research. Nat. Rev. Genet.19, 566–580 (2018). [DOI] [PubMed] [Google Scholar]
  • 30.Hemani, G., Tilling, K. & Davey Smith, G. Orienting the causal relationship between imprecisely measured traits using GWAS summary data. PLoS Genet.13, e1007081 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Qi, X. et al. Increased plasma GPNMB levels in patients with parkinson’s disease and cognitive impairment. Sci. Rep.15, 20684 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Zhu, X. C. et al. CSF GPNMB in Parkinson’s disease: a potential association with age and microglial activation. J. Parkinsons Dis.14, 1533–1542 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Moloney, E. B., Moskites, A., Ferrari, E. J., Isacson, O. & Hallett, P. J. The glycoprotein GPNMB is selectively elevated in the substantia nigra of Parkinson’s disease patients and increases after lysosomal stress. Neurobiol. Dis.120, 1–11 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Zimmermann, M. et al. The longevity gene Klotho and its cerebrospinal fluid protein profiles as a modifier for Parkinson´s disease. Eur. J. Neurol.28, 1557–1565 (2021). [DOI] [PubMed] [Google Scholar]
  • 35.Kjaergard, L. L., Villumsen, J. & Gluud, C. Reported methodologic quality and discrepancies between large and small randomized trials in meta-analyses. Ann. Intern Med135, 982–989 (2001). [DOI] [PubMed] [Google Scholar]
  • 36.Liu, Z. T., Liu, M. H., Xiong, Y., Wang, Y. J. & Bu, X. L. Crosstalk between bone and brain in Alzheimer’s disease: mechanisms, applications, and perspectives. Alzheimers Dement20, 5720–5739 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Hu, X., Zhang, P., Xu, Z., Chen, H. & Xie, X. GPNMB enhances bone regeneration by promoting angiogenesis and osteogenesis: potential role for tissue engineering bone. J. Cell Biochem.114, 2729–2737 (2013). [DOI] [PubMed] [Google Scholar]
  • 38.Huang, B. et al. Extracellular vesicles from GPNMB-modified bone marrow mesenchymal stem cells attenuate bone loss in an ovariectomized rat model. Life Sci.272, 119208 (2021). [DOI] [PubMed] [Google Scholar]
  • 39.Sheng, M. H., Wergedal, J. E., Mohan, S. & Lau, K. H. Osteoactivin is a novel osteoclastic protein and plays a key role in osteoclast differentiation and activity. FEBS Lett.582, 1451–1458 (2008). [DOI] [PubMed] [Google Scholar]
  • 40.Frara, N. et al. Transgenic expression of osteoactivin/gpnmb enhances bone formation in vivo and osteoprogenitor differentiation ex vivo. J. Cell Physiol.231, 72–83 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 41.Postuma, R. B. et al. MDS clinical diagnostic criteria for Parkinson’s disease. Mov. Disord.30, 1591–1601 (2015). [DOI] [PubMed] [Google Scholar]
  • 42.Martínez-Martín, P. et al. Parkinson’s disease severity levels and MDS-unified Parkinson’s disease rating scale. Parkinsonism Relat. Disord.21, 50–54 (2015). [DOI] [PubMed] [Google Scholar]
  • 43.Goetz, C. G. et al. Movement disorder society-sponsored revision of the unified Parkinson’s disease rating scale (MDS-UPDRS): scale presentation and clinimetric testing results. Mov. Disord.23, 2129–2170 (2008). [DOI] [PubMed] [Google Scholar]
  • 44.Eldjarn, G. H. et al. Large-scale plasma proteomics comparisons through genetics and disease associations. Nature622, 348–358 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 45.GTEx Consortium. The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science369, 1318–1330 (2020). [DOI] [PMC free article] [PubMed]
  • 46.Nalls, M. A. et al. Identification of novel risk loci, causal insights, and heritable risk for Parkinson’s disease: a meta-analysis of genome-wide association studies. Lancet Neurol.18, 1091–1102 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Kurki, M. I. et al. FinnGen provides genetic insights from a well-phenotyped isolated population. Nature613, 508–518 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Desquilbet, L. & Mariotti, F. Dose-response analyses using restricted cubic spline functions in public health research. Stat. Med.29, 1037–1057 (2010). [DOI] [PubMed] [Google Scholar]
  • 49.de Winter, J. C., Gosling, S. D. & Potter, J. Comparing the Pearson and Spearman correlation coefficients across distributions and sample sizes: a tutorial using simulations and empirical data. Psychol. Methods21, 273–290 (2016). [DOI] [PubMed] [Google Scholar]
  • 50.Schober, P., Boer, C. & Schwarte, L. A. Correlation coefficients: appropriate use and interpretation. Anesth. Analg.126, 1763–1768 (2018). [DOI] [PubMed] [Google Scholar]
  • 51.Schlomer, G. L., Bauman, S. & Card, N. A. Best practices for missing data management in counseling psychology. J. Couns. Psychol.57, 1–10 (2010). [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplementary Tables and Figures (2.5MB, pdf)

Data Availability Statement

The GWAS summary statistics data used in this MR study are publicly available which are included in the additional files. Clinical data are available from the corresponding author upon reasonable request, subject to scientific review and the completion of a material transfer agreement.

Articles from NPJ Parkinson's Disease are provided here courtesy of Nature Publishing Group

RESOURCES