A laboratory analysis mapped four conventional antiepileptic drugs to different leading protein targets in an osteoporosis-related network, offering clues about possible links with bone biology but no clinical answer. Carbamazepine, phenobarbital, phenytoin and valproic acid each produced a separate leading signal, and the study then examined drug-target pairings with computer models. The evidence is computational, not a clinical finding about osteoporosis in people.
Across the four medicines, the analysis identified 47 potential osteoporosis-related targets for carbamazepine, 50 for phenobarbital, 46 for phenytoin and 48 for valproic acid. These are predicted intersections between drug targets and osteoporosis-related genes, not counts of patients, cases or measured bone changes.
Building a map of possible links
Researchers combined network toxicology, molecular docking and 100-nanosecond molecular-dynamics simulations with colony-formation and Western-blot assays in MC3T3-E1 pre-osteoblasts. For the target map, human protein targets were screened with STITCH and SwissTargetPrediction. STITCH pairs were retained at a combined confidence score of at least 0.4. SwissTargetPrediction candidates had to reach a probability of at least 0.1, after which the top 50 were kept.
Potential hubs were ranked using three network measures: degree, betweenness and closeness centrality. The researchers treated nodes in the top 10% for all three measures as candidates, then applied a degree-based filter to select the definitive hubs.
The leading signals split four ways
EGFR emerged as the primary carbamazepine hub, and the modeled carbamazepine-EGFR pair received a docking score of minus 8 kcal/mol. CASP3 was the top phenobarbital-associated hub, with a phenobarbital-CASP3 score of minus 6.0 kcal/mol. Both figures are docking estimates, not experimental binding measurements.
Phenytoin's primary hub was PPARG, with a predicted phenytoin-PPARG score of minus 7.4 kcal/mol. For valproic acid, PTGS2 was the most prominent hub, and the predicted valproic acid-PTGS2 score was minus 5.9 kcal/mol. Those numbers are computational estimates, not direct measurements of binding affinity.
What the simulations add
The researchers then simulated all four drug-target complexes for 100 ns and summarized the quantitative measures over the equilibrated 50-to-100-ns interval. Complex RMSD, a measure of how much the modeled arrangement shifts during a simulation, averaged 0.22 plus or minus 0.02 nm for PB-CASP3, 0.24 plus or minus 0.02 nm for PHT-PPARG and 0.18 plus or minus 0.01 nm for VPA-PTGS2. CBZ-EGFR was higher at 0.34 plus or minus 0.02 nm.
CBZ-EGFR also had the highest average RMSF, at 0.15 plus or minus 0.11 nm. RMSF captures fluctuations within a modeled complex, so this pattern points to more simulated movement in that pairing than in the other reported complexes. The measures are dynamic surrogates, not direct evidence of bone effects or experimental binding.
A laboratory clue, not a treatment result
The study also used colony formation and Western blotting in MC3T3-E1 pre-osteoblasts, but those assays do not turn the computer predictions into clinical evidence. The central verified results are target intersections, docking scores and simulation measures, rather than human bone-density, fracture or long-term treatment outcomes.
The findings support follow-up testing, not a clinical recommendation. The target counts are computational intersections, the docking scores are estimates, and the simulation measures are dynamic surrogates. Animal and clinical studies, along with replicated simulations and experimental binding tests, would be needed to determine whether any of the modeled links matter in living organisms.
Paper data and sources
Original title: Insights into osteoporosis triggered by common antiepileptic drugs via network toxicology, molecular docking and molecular dynamics simulations
Authors: Wang C, Gao S, Wu M et al.
Journal/Repository: Frontiers in pharmacology
Status: Peer-reviewed
First online: 2026-10-10
DOI: Not available
Original paper · Full text