Radiotherapy is perhaps the most ubiquitous single treatment modality for patients with cancer. Despite its routine use, biomarkers of treatment response are critically lacking, and the biology underlying the heterogeneity in clinical response to radiation treatment remain largely unknown. In this issue of Cancer Research, Manem and colleagues seek to change this paradigm and detail the development of a radiation response computational platform (RadioGx) that allows for the integrative analysis of radiation response using radiogenomic data derived from preclinical, in vitro sources (1). This platform holds promise for novel hypothesis generation and may allow for the discovery of novel mediators of radiation response that may improve the therapeutic efficacy of radiation. Importantly, it also moves us closer to uncovering and validating radiation response biomarkers that may prove clinically useful.
See related article by Manem et al., p. 6227
Radiotherapy plays a critical role in the treatment of cancer, and over half of patients with cancer receive radiation as part of their treatment course (2). Despite the ubiquity of radiation in the management of patients with cancer, dosing of radiation has been relatively uniform for patients, often depending on the type of cancer they present with and the stage of disease. Personalization of dose based on individual patient and tumor factors has yet to enter the clinic, in large part due to the dearth of biomarkers predictive of radiation response. In an era of precision medicine and high-throughput data (genomic, transcriptomic, proteomic, etc.), opportunities now exist to alter this “one size fits all” paradigm of radiation dosing and treatment. Indeed, the concept that cell or tumor-intrinsic factors (like gene expression or tumor mutations) may inform radiation response and act as surrogate biomarkers of disease response is an area of active exploration. This comes as the radiobiology field has for decades relied on the response of normal and malignant cells grown in culture to radiation as the basis for describing tumor treatment response in vivo (3). These studies were instrumental in providing the data from which the linear quadratic model was derived, but it was incomplete in its ability to adequately model the heterogeneous response of tumors and normal tissue to ionizing radiation. On account of these known challenges, researchers have widened their investigation to develop more complete ways to both describe and predict the response of cancers to ionizing radiation. This widened investigation, together with advances that allow for the more complete evaluation of nucleic acids, peptides, epigenetic markers, and metabolites, has led to the development of “-omic” based approaches for the prediction of response of tumors to treatment (4–9).
It is within that context that Manem and colleagues combine “big data” approaches with the linear-quadratic equation and add to a growing body of literature that suggest that utilizing information about the intrinsic radiosensitivity of cancers may inform clinical dosing and treatment of the disease these models represent. More importantly, they provide a novel informatics tool, RadioGx, that allows for in silico modeling of cellular response that may lead to the generation of novel radiosensitizing strategies in the near term and allows for a more personalized dosing of radiation for patients in the clinic in the long term. Working with colleagues at the Cleveland Clinic, they develop a computational platform for integrative radioresponse analysis based on radiogenomic databases by fitting in vitro cancer cell lines' dose response to the linear-quadratic model. Utilizing this platform, the authors utilize gene expression differences between radioresponsive and resistant populations, or baseline mutations, in the cell lines used for modeling to explore the biology underlying radiation response and unbiasedly nominate pathways known to influence radiation response, most notably the nonhomologous end joining (NHEJ) pathway. They also identify mutations associated with radioresponse and highlight the utility of the platform by using it to address two intriguing questions related to oxygen's effect on radiation response and druggable targets for radiosensitization. In summary, they provide RadioGx, a computational toolbox that facilitates comparative and integrative analysis of radiogenomic datasets.
RadioGx also allows for new biological insights and predictions. Perhaps the greatest strength of the tool is the potential utility in uncovering mechanisms that may underlie radioresistance or sensitivity based on modeled perturbations of the system. For example, the authors highlight how modeling hypoxia may identify genes that mediate the radioresistance seen experimentally under hypoxic conditions. This opens an avenue for further validation to confirm that the modeled data matches experimentally derived data and could strengthen the rationale for using RadioGx for this purpose. Indeed, similar modeling could be done to explore the effect of signaling pathway inhibition, growth factor depletion, or drug treatment on radiation response and offers an exciting opportunity to develop hypotheses that can be experimentally tested.
One additional major implication of the study is that radiation response is tumor-type specific. While this observation has been noted in the clinic for decades, recent efforts by several groups have focused on a “pan-cancer”–based approach to radiation response modeling in which data from all cancer models (cell lines, patient-derived xenografts, etc.) are treated similarly in a tumor-type agnostic way. The compelling data presented by Manem and colleagues argues that such approaches are likely fundamentally flawed, and that radiation prediction modeling is likely to be more accurate and therefore clinically relevant when based on the cancer type of origin. Thus, one can envision using these data to derive a signature that is specific for breast, prostate, pancreatic, and any other type of cancer.
Despite the many strengths of this work, there are foundational limitations to the model system used that may ultimately limit the translation of this work into the clinic. This arises from limitations in the reliance on the linear-quadratic model and the assay used to model cell survival. It also relies on the use of cell lines from heavily pretreated patients and cells obtained from metastatic, rather than primary, sites with no microenvironment or immune system present. In addition, high-dose radiation is increasingly utilized in clinical radiation delivery, with many diseases now treated routinely with doses above 2 Gy per fraction. Indeed, stereotactic body radiotherapy (SBRT) and stereotactic radiosurgery (SRS) routinely utilize doses of 8–24 Gy in one or just a few fractions to treat many types of cancer with >90% tumor control in many clinical situations. The challenges with modeling outcomes with these high doses using the linear-quadratic model have previously been described (10), and RadioGx similarly was less reliable in the most radioresistant cell lines where higher doses of radiation are required to kill cells and the surviving fraction after 2 Gy (SF-2Gy) is no longer relevant. In addition, foundational to the RadioGx platform is the 9-day viability assay as a surrogate for radiation response. While this concept is intriguing as it allows for higher throughput characterization of cell lines, organoids, or patient-derived xenograft samples, the more tedious clonogenic survival assay still remains the gold standard. Additional confirmatory studies comparing the viability assay to the clonogenic survival assay are needed. This platform, of necessity, ignores much of the complexity of tumor biology and radiation response and cannot incorporate microenvironment, immune, or environmental factors that are known to influence radiation response. It also is dependent on in vitro cell line data that comes from serially passaged, laboratory grown cell lines derived from metastatic sites that are often a far cry from the early-stage, treatment-naïve patients that generally receive radiotherapy. Finally, the RadioGx model validation is circular in some aspects. The model relies on the linear-quadratic model, one derived from in vitro cell survival, and is based on the concept of DNA repair. They then model cell survival in vitro and identify key contributors to radiosensitivity involving DNA repair. However, the linear-quadratic model repeatedly has fallen short in randomized trials. DNA repair alterations or pathway expression have not consistently correlated to radiation response, the immune system plays a critical role in tumor response to radiotherapy, and thus the translatability of this work requires further validation. These limitations, however, do not necessitate that we “throw the baby out with the bathwater” and RadioGx does hold the promise of more rapid generation of testable hypotheses and novel combination therapies for preclinical testing. Indeed, in an era of high-throughput data generation and the need for more personalized therapy recommendations, RadioGx represents a potentially valuable tool, allowing for a future in which radiation treatment recommendations are not based on population-based responses, but on the individual patients' predicted response to radiation.
Radiotherapy is a mainstay of clinical management of cancer. Biomarkers of treatment response are now used routinely to guide clinical decisions regarding targeted therapy, immunotherapy, or systemic chemotherapy usage. Despite these advances, there are currently very few predictive biomarkers of radiation response used clinically. The development of RadioGx has the potential to hasten the discovery, development, and validation of radiation response biomarkers as well as uncover the mediators or radiation resistance and represents continued progress towards the goal of truly personalized precision medicine in radiation oncology.
Disclosure of Potential Conflicts of Interest
C. Speers has ownership interest (including patents) and is a consultant/advisory board member (unpaid) PFS Genomics. D.E. Spratt is an advisor at Janssen and Blue Earth. No other potential conflicts of interest were disclosed.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.