A general mechanistic model enables predictions of the biological effectiveness of different qualities of radiation

Stephen J Mcmahon,Aimee L Mcnamara,Kevin M Prise,Harald Paganetti,Jan Schuemann

doi:10.1038/s41598-017-10820-1

Stephen J Mcmahon, Aimee L Mcnamara + Show 3 more

Open Access

https://doi.org/10.1038/s41598-017-10820-1

Copy DOI

Abstract

Predicting the responses of biological systems to ionising radiation is extremely challenging, particularly when comparing X-rays and heavy charged particles, due to the uncertainty in their Relative Biological Effectiveness (RBE). Here we assess the power of a novel mechanistic model of DNA damage repair to predict the sensitivity of cells to X-ray, proton or carbon ion exposures in vitro against over 800 published experiments. By specifying the phenotypic characteristics of cells, the model was able to effectively stratify X-ray radiosensitivity (R2 = 0.74) without the use of any cell-specific fitting parameters. This model was extended to charged particle exposures by integrating Monte Carlo calculated dose distributions, and successfully fit to cellular proton radiosensitivity using a single dose-related parameter (R2 = 0.66). Using these parameters, the model was also shown to be predictive of carbon ion RBE (R2 = 0.77). This model can effectively predict cellular sensitivity to a range of radiations, and has the potential to support developments of personalised radiotherapy independent of radiation type.

Highlights

Cancer radiotherapy has long been a highly personalised treatment, with significant research and technical development being devoted into better identifying, localising, and treating cancers with ionising radiation
To test the predictive power of the model for a range of different cell lines exposed to X-rays, Mean Inactivation Doses (MIDs) were calculated for each of the unique X-ray experiments reported in two radiation response databases – Paganetti’s review of proton RBE25 and the Particle Irradiation Data Ensemble (PIDE)[26], together with those from a previous study[24]
For each X-ray response curve in these datasets, the Mean Inactivation Dose (MID) was calculated from experimental observations, and compared to the sensitivity predicted by the established mechanistic model

Summary

Introduction

Cancer radiotherapy has long been a highly personalised treatment, with significant research and technical development being devoted into better identifying, localising, and treating cancers with ionising radiation. The treatment dose and schedule are typically determined by the site (and potentially stage) of cancer alone, with the majority of patients receiving a ‘one size fits all’ treatment While these schedules are often based on the results of very large clinical trials[2, 3], their results typically assume a uniform radiation sensitivity across a whole population. Modelling of radiation response curves based on clinical data has suggested that the dose needed to control 50% of tumours (TD50) could have a standard deviation of 20–25%6, a range which is reflected in many in vitro studies of cellular radiation sensitivity[7,8,9] Such a range means that any dose selected on the population level would certainly over- or under-treat large numbers of patients, impacting negatively on clinical outcomes. A particular challenge with these approaches is the very large data sets which are required to generate meaningful fits, as they typically do not take advantage of our underlying knowledge of radiation effects at a biological level, instead focusing on purely statistical approaches to identify trends

Methods

Results

Conclusion