A mechanistic model for genetic regulation of postmenopausal bone loss

Technical Summary: A Mechanistic Model for Genetic Regulation of Postmenopausal Bone Loss

Problem Statement
Postmenopausal bone loss is a complex physiological process driven by the disruption of the balance between bone-resorbing osteoclasts and bone-forming osteoblasts, primarily due to estrogen decline. While genetic factors significantly influence bone mineral density (BMD) loss, translating genetic associations into individualized risk predictions remains challenging. This difficulty arises from the small effect sizes of individual single nucleotide polymorphisms (SNPs) and the non-linear, polygenic interactions within the bone remodeling unit. Existing mathematical models of bone remodeling often lack explicit estrogen signaling mechanisms or fail to incorporate individual-level genetic variables, limiting their utility for personalized risk assessment.

Methodology
The authors developed a novel mechanistic model of bone cell population dynamics that integrates estrogen signaling and genetic polymorphisms. The framework builds upon established models (Lemaire et al., Pivonka et al., Farhat et al., Jørg et al.) and is formulated as a system of 27 ordinary differential equations (ODEs) with 64 parameters.

• Model Structure: The model simulates the Bone Multicellular Unit (BMU), tracking the dynamics of osteoclast progenitors (OCp), active osteoclasts (OCa), osteoblast progenitors (OBp), active osteoblasts (OBa), and osteocytes (OT). It explicitly models regulatory pathways including the RANK/RANKL/OPG axis, Wnt signaling (modulated by sclerostin and DKK1), and hormonal regulation by estrogen, parathyroid hormone (PTH), and TGF-β.
• Estrogen Mechanism: The model explicitly incorporates estrogen's regulation of RANKL, OPG, sclerostin, and, crucially, osteoclast apoptosis.
• Genetic Integration: The authors selected 22 SNPs across 8 genes (ESR1, CYP19A1, RANK, RANKL, OPG, SOST, LRP5, TGFB1) associated with BMD or estrogen levels. Genetic effects were mechanistically incorporated by mapping SNPs to their predicted functional impacts on molecular processes (e.g., modifying production rate constants for expression variants or association rate constants for binding variants). An additive genetic model was used, with marginal effect sizes adjusted for linkage disequilibrium (LD) to calculate conditional effects.
• Calibration and Validation: The model was calibrated using clinical data from healthy postmenopausal women and patients undergoing anti-osteoporotic therapy (denosumab, romosozumab, and micronized estradiol). To accurately represent drug interventions, minimal physiologically based pharmacokinetic (mPBPK) models with target-mediated drug disposition (TMDD) were developed for denosumab and romosozumab. Calibration involved fitting model parameters to longitudinal BMD and bone turnover metabolite (BTM) trajectories.
• Population Simulation: A virtual population of 1,000 individuals was simulated, stratified by ancestry (European, African, East Asian) with genotypes sampled to preserve ancestry-specific allele frequencies and LD patterns.

Key Contributions

1. Mechanistic Integration of Genetics: The study presents a proof-of-concept framework for integrating polygenic risk factors into a mechanistic disease model, moving beyond statistical associations to simulate biological trajectories.
2. Elucidation of Estrogen Mechanisms: Through ablation experiments, the model identified that estrogen's primary protective effect on bone is mediated by promoting osteoclast apoptosis, rather than solely through the regulation of RANKL, OPG, or sclerostin.
3. Drug Response Modeling: The development of coupled mPBPK-TMDD models allowed for the accurate simulation of therapeutic responses to RANKL and sclerostin inhibitors, enabling the estimation of target capacities and drug-specific parameters.
4. Genetic Interaction Analysis: The framework enabled the analysis of epistatic effects, revealing that while most genetic effects are additive, significant non-linear interactions exist between SNPs, particularly within the same loci and across different genes (e.g., ESR1 and SOST).

Results

• Model Performance: The calibrated model accurately reproduced healthy BMD decline in postmenopausal women (simulated rate: 0.62% per year) and captured the magnitude of BMD gain observed with denosumab (3.1% over 2 years), romosozumab (5.6% over 2 years), and micronized estradiol (1.9% over 3 years).
• Virtual Population Predictions: The median predicted 5-year BMD loss was 3.57% (95% prediction interval: 1.31–5.24%), consistent with literature values. The model accounted for approximately 25% of the assumed variance in BMD loss.
• Ancestry Differences: Individuals of African ancestry were predicted to experience the highest average 5-year BMD loss (3.86%) compared to European (3.46%) and East Asian (3.28%) ancestries.
• Genetic Risk Factors: The strongest predicted risk factors for bone loss were CYP19A1 rs727479 and OPG rs3102735. Conversely, LRP5 rs11228240 emerged as a protective factor.
• Epistasis: The analysis revealed that genetic interactions often attenuate additive effects. For instance, the bone-protective effect of RANK rs3018362 was only detectable in individuals carrying a risk allele at SOST rs188810925.

Significance and Claims
The authors claim this work provides a "proof-of-concept" for integrating genetic risk factors into mechanistic models of disease. They emphasize that the framework successfully bridges the gap between scientific discovery (GWAS findings) and clinical practice by enabling the simulation of personalized BMD trajectories based on genetic makeup. The study highlights the importance of non-linear genetic interactions and the necessity of mechanistic modeling to understand the combined genetic burden in polygenic conditions. The authors modestly note that while the model captures a significant portion of genetic variability, non-genetic factors (clinical, demographic, behavioral) play a major role in bone dynamics, and future work is required to incorporate these covariates and validate predictions at the individual level. The framework is proposed as extensible to other conditions with polygenic inheritance.