Radiation treatments typically assume uniform radiosensitivity across a whole population, but this is a major oversimplification. Extensive evidence exists that cancers of the same type can have very different radiosensitivities. Indeed, radiation response modelling of clinical data suggests that the dose needed to control 50% of tumours could vary by 20–25%.

The ability to predict the radiosensitivity of a patient's cancer could significantly impact treatment decisions, but direct measurement remains challenging, and the lack of predictive models is a significant problem. To address this shortfall, a research team from Queen's University Belfast and Massachusetts General Hospital has developed a mechanistic model of DNA damage repair to predict response to ionizing radiation.

The researchers have now assessed the predictive power of their model by comparing its radiosensitivity predictions with over 800 published experiments, including X-ray, proton and carbon ion data (Scientific Reports 7 10790).

Predictive power

The model uses initial distributions of DNA double strand breaks (DSBs) to calculate the probability of different types of DSB repair and predict the overall radiosensitivity. The authors emphasize that the model does not require any cell-specific fitting parameters. Instead, it defines parameters describing cellular processes common to a range of cell types, including various DNA repair and survival parameters.

"Predictions for a specific cell line are then achieved by combining these generic parameters with the particular characteristics of the cell – such as whether it is capable of homologous recombination DNA repair, or has a functional G1 checkpoint," explained first author Stephen McMahon.

For clinical use, many of these characteristics could be assayed in vitro from tumour biopsies. "But I think a much more exciting approach would be to link this model to tumour genomics," noted McMahon. "Many of the genetic pathways involved in radiation responses are well studied, and with the growing interest in genetic sequencing as a clinical tool to personalize radiotherapy, this opens the possibility to predict these phenotypic characteristics based on their underlying genetics."

To test the predictive power of their model for various cell lines exposed to X-rays, the researchers calculated mean inactivation doses (MIDs) for X-ray experiments reported in several datasets. For each response curve, they compared the observed MID to that predicted by the model. Even though these data represented a wide range of cell lines, good correlation (a correlation coefficient of 0.74) was seen across the entire range of radiosensitivities.

Particle irradiation

Next, McMahon and colleagues extended their model for use with charged particle therapies, by linking it with a Monte Carlo simulation of energy distribution around particle tracks. This adaptation only required the addition of a single charged-particle related fitting parameter: EDSB, the energy required to create one DSB on average.

By fitting existing experimental proton RBE data (from Phys. Med. Biol. 59 R419), the researchers determined a best-fitting value for EDSB of 60.7±14.0 keV. They then used this value to generate proton MID values for comparison with experimental observations. Again, the overall correlation was good, with a correlation coefficient of 0.66.

The mechanistic model can also be directly applied to heavier ions, simply by incorporating the appropriate radial energy distributions. To test this premise, the researchers used the model to calculate the LET-dependence of in vitro carbon RBE values. They used the EDSB value obtained previously to validate the model 's ability to extrapolate between different radiation types.

"The model assumes that one DSB on average is created for every 61 keV deposited. This relationship holds for all particles, with X-rays, protons and carbon ions only differing in their spatial distribution," McMahon explained.

A plot of modelled versus experimentally observed carbon RBE demonstrated a correlation coefficient of 0.77. "As protons and carbon ions span very different ranges of LET and RBE, the fact that a single parameter could effectively explain the effects of both particles suggests the model accurately reflects underlying mechanisms," said McMahon.

The team is now expanding their model to incorporate temporal effects, reflecting the fractionated exposures that are common in clinical radiotherapy. "In the longer term, we're looking to bridge the gap to potential clinical applications, by developing models linking cellular radiosensitivity phenotypes to the underlying genetics, to enable these models to be used to support the delivery of personalized radiotherapy," McMahon told medicalphysicsweb.