Modeling Bone Metastasis: Digital Twins Reproduce Tumor Biology with Unprecedented Precision

Researchers at the Houston Methodist Academic Institute have unveiled a groundbreaking computational model that could transform how scientists study and treat bone metastasis, one of the deadliest complications of advanced cancers.

The study describes the process of creating a "digital twin" of bone metastasis — a high-fidelity computer model that replicates how tumors grow, spread and respond to therapies within the complex environment of bone tissue.

"To the best of our knowledge, this is the first computational model to accurately replicate how these tumors invade bone tissue and predict their response to therapy," said Dr. Stefano Casarin, Ph.D., the study's co-primary investigator and an assistant professor with the Center of Precision Surgery at Houston Methodist. "By merging quantitative imaging, histologic data and biological parameters from both prostate and renal cell carcinoma models, we have recreated a dynamic system that mirrors how metastatic tumors interact with bone and its vasculature in living organisms."

The work, published last month in the journal Cancer Research, represents a milestone in translational oncology modeling, combining the biological fidelity of animal studies with the scalability and predictive power of computational simulations.

Rethinking preclinical experimentation

Bone metastases remain one of the most lethal and least understood aspects of advanced cancer. They cause significant pain, fractures and spinal cord compression, and they're largely incurable. Traditional in vivo models, while invaluable, are limited in how many experimental conditions they can reproduce.

"The goal of our computational model isn't to replace animal research," said Dr. Casarin. "It's to complement it with something resource-efficient and infinitely repeatable. Instead of testing hundreds of therapies in animals, we can screen them in silico, refine the best candidates and then move only the most promising ones forward for in vivo validation."

The approach reflects a growing movement among NIH and regulatory agencies to reduce animal experimentation by integrating mechanistic computational models that can accurately predict biological behavior.

Building the digital bone

Bone metastases are notoriously difficult to study because they involve intricate communication between tumor cells, bone cells and blood vessels.

Dr. Casarin's model integrates years of biological data — from cellular behavior to bone geometry — into a virtual environment that mirrors reality down to the micrometer.

Designed to mimic the real architecture of the bone and its cellular ecosystem, the model combines multiphoton microscopy and high-resolution 3D images of murine bone microstructures, then reconstructs them to reproduce the cortical and trabecular compartments and the marrow cavity where metastases arise.

Each digital bone site contains simulated populations of tumor cells, blood vessels, osteoblasts (bone-forming cells) and osteoclasts (bone-resorbing cells).

"We spent more than seven years developing every detail of the bone tumor microenvironment," he said. "Our goal was to create a comprehensive system that captures all the interactions taking place — between the tumor, the bone, the vasculature, the osteoblasts and osteoclasts."

The interactions among these agents are governed by biologically derived rules that reflect real-world dynamics: how close a tumor cell is to a blood vessel affects its likelihood of dividing; how many osteoclasts are recruited determines how rapidly the bone matrix is resorbed.

"The power of the model lies in its spatial precision," said Dr. Casarin. "We don't assume an idealized bone. We reconstruct the true geometry from microscopy data, which allows us to model exactly where drugs accumulate, where the vasculature forms, and how mechanical or biochemical gradients shape tumor behavior."

Modeling tumor–bone crosstalk

A key innovation of the study is how it captures angiogenesis and osteolysis, two intertwined processes that drive bone metastasis progression.

Previous computational models of metastasis treated the bone as a uniform structure and overlooked how tumor-driven blood vessel growth (angiogenesis) fuels proliferation.

Dr. Casarin's model integrates angiogenic dynamics by simulating how tumor cells generate new vessels, how oxygen and nutrient gradients change over time, and how anti-angiogenic drugs disrupt this network.

One such drug, cabozantinib, was tested virtually within the model. The simulation predicted dose-dependent reductions in tumor volume that closely matched results seen in in vivo experiments, validating the model's predictive accuracy.

At the same time, the model simulates osteolysis — the pathological bone degradation caused by tumor-induced activation of osteoclasts.

"In healthy bone, osteoclasts and osteoblasts maintain equilibrium, but when a tumor grows near the mineralized surface, it releases signals that corrupt this balance," Dr. Casarin explained. "The osteoclasts go into overdrive and start digesting bone uncontrollably, weakening the entire structure."

The team's in silico system successfully recreated this destructive feedback loop, showing how osteolysis evolves as tumors expand and how it responds to anti-resorptive therapy, such as zoledronic acid.

"Osteolysis is a devastating complication for patients," said Dr. Casarin. "Our model allows us to see how and when that process starts — and to test out interventions before they reach the clinic."

Predicting the efficacy of complex therapy combinations

While the Cancer Research paper focuses on the digital twin for cabozantinib, Dr. Casarin's group is already expanding the model to include additional therapeutic modalities — most notably Radium-223, a radiopharmaceutical that emits alpha particles to destroy cancer cells from outside the bone.

"Cabozantinib attacks the tumor from the inside by cutting off its blood supply, while Radium-223 attacks from the outside by irradiating tumor cells that sit near the bone surface," explained Dr. Casarin. "The next step of our research is to integrate both mechanisms in silico to study optimal timing, dosing and potential synergistic effects."

This two-pronged model is part of a larger NIH/CPRIT-funded effort to use digital twins for precision therapy design, where simulations can predict how various regimens will behave before being tested in animals or patients.

The team is also collaborating with UTHealth engineers to quantify how these treatments affect bone mechanics — testing how irradiated or drug-treated bones respond to compression or bending to assess fracture risk.

Moving toward personalization

Although the present study used mouse data for calibration, the long-term goal is to translate the platform into patient-specific digital twins. Dr. Casarin envisions feeding CT or MRI data from human patients into the same computational framework to simulate personalized disease trajectories and therapy responses.

"The beauty of this approach is that it's adaptable," Dr. Casarin said. "With clinical imaging and biopsy-derived parameters, we can create digital twins that replicate an individual patient's bone and tumor. That's where oncology is heading — predictive, adaptive and deeply personalized."

As Houston Methodist continues to push the frontier of computational oncology, Dr. Casarin believes this convergence of biology, imaging and AI-driven modeling could become the foundation for a new generation of preclinical and clinical decision tools.

"Our hope is that these digital twins will accelerate discovery — helping scientists understand mechanisms that are impossible to observe in real time and guiding clinicians toward the most effective and least toxic treatments for patients with bone metastasis, he said."

A collaboration between Houston Methodist and MD Anderson Cancer Center, the study was co-led by Dr. Eleonora Dondossola from the Genitourinary Medical Oncology Department at MD Anderson. It is part of a long-standing international partnership with Politecnico di Milano and Università di Pavia, both Italian universities, supporting graduate students and Ph.D. candidates conducting research at Houston Methodist. All code and data supporting this work are available through the Code Ocean repository.