Advertisement

Molecular Biomarkers in Radiation Oncology

  • Brita Singers SørensenEmail author
  • Christian Nicolaj Andreassen
  • Jan Alsner
Living reference work entry

Abstract

Molecular biomarkers in radiation oncology could enable the selection of patients for individualization of radiation therapy and concomitant treatment with optimizing agents and are an important tool in precision medicine. The biomarkers are directed toward either determining the tumor radiosensitivity or as markers for the normal tissue response to radiotherapy and are focused on the biological characteristics that affect response to radiation. Although a lot of focus has been directed on identifying suitable molecular biomarkers for radiation oncology, almost none has made it into clinical use.

This chapter will introduce the basic definitions of biomarkers and the differences between prognostic and predictive biomarkers. It will go through the biological factors that influence differences in response to radiation, such as intrinsic radiosensitivity, hypoxia, HPV status and number of cancer stem cells (CSCs), and candidate biomarkers for these causes of radioresistance.

The ability to predict normal tissue complication risk prior to radiotherapy has been a long sought goal in radiobiology. For more than 25 years, various attempts have tried to establish a predictive assay for normal tissue radiosensitivity, and these efforts will be summarized. Different strategies for identifying predictive markers for the normal tissue response to radiotherapy will be described, and the current status of radiogenomics is evaluated.

Validation of identified biomarkers is a crucial step in getting a biomarker implemented in the clinic. The different levels of validation, analytical validity, clinical validity, and clinical utility, will be explained, and examples of biomarkers that have gone through this validation process are included.

Keywords

Molecular biomarkers Prognostic biomarkers Predictive biomarkers Radiation response Radiosensitivity Hypoxia HPV Cancer stem cells Radioresistance Validation of biomarkers Normal tissue response 

Introduction

Biomarkers are biological features that provide information of the biological state of a given tissue or organism. Examples include levels of particular proteins in tissue or body fluids, nucleic acid-based biomarkers such as gene expression levels, gene mutations or polymorphisms, or peptides, lipid metabolites, and other small molecules. The US National Institutes of Health (NIH) Biomarkers and Surrogate Endpoint Working Group defines a biological marker, a biomarker, as a characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacological responses to therapeutic intervention (Atkinson et al. 2001).

Molecular biomarkers in radiation oncology are focused on the biological characteristics that have been shown to influence differences in response to radiation for both the tumor and the surrounding normal tissue. These characteristics include intrinsic radiosensitivity, hypoxia, HPV status, number of cancer stem cells (CSCs), and repopulation between radiotherapy fractions (Baumann et al. 2016). In the last 20 years, remarkable progresses in the knowledge of the biological factors influencing radiation response and the causative molecular basis, such as the DNA damage response and repair mechanisms, signaling pathways, and tumor microenvironmental factors, have been made and have generated a large number of potential biomarkers. Biomarkers are an important tool in individualization of patient treatment and can potentially guide treatment and stratify patients who may benefit from a specific treatment (Bibault et al. 2013).

Compared to surgery, radiotherapy has the advantage of being potentially organ and function preserving. Nevertheless, it may lead to severe acute or late toxicity (Kerns et al. 2015). Radiation-induced normal tissue reactions represent a dose-limiting factor in radiotherapy. Typically, treatment regimens are designed to ensure that the risk of severe late effects does not exceed 5–10% (Scaife et al. 2015). This means that a small fraction of radiosensitive patients limits the dose that can be given to the entire patient population, though the majority of patients could potentially tolerate a higher dose. If the variability in normal tissue response could be taken into account in the treatment-planning phase, the therapeutic strategy could be individualized accordingly (Kerns et al. 2014). Patients being relatively sensitive to the effects of ionizing radiation could (when possible) be offered a treatment strategy that does not include radiotherapy, whereas the resistant patients could be dose escalated to some extent. This would potentially lead to a substantial improvement in the therapeutic index for radiotherapy (Kerns et al. 2015). Consequently, the ability to predict normal tissue complication risk prior to therapy has been a long sought goal in radiobiology (Russell and Begg 2002; Scaife et al. 2015).

Although a lot of focus has been directed on identifying suitable molecular biomarkers for radiation oncology, almost none has made it into clinical use. This has been due to either time-consuming methods, heavy technical demand, or an insufficient level of evidence or validation of the biomarker. This chapter will focus on a section of molecular biomarkers in radiation oncology, and the biological rationale, as well as some fundamental aspects of biomarkers in general.

Biomarkers: Definitions

Prognostic Versus Predictive Biomarkers

Biomarkers are classified as either prognostic or predictive based on their association with outcome and their ability to predict the effect of different classes of, or even individual, therapeutic agents (Ballman 2015, reviewed by Hayes 2015).

Prognostic biomarkers can be defined as biomarkers that can inform on the future behavior of an established cancer or the risk of developing treatment-induced morbidity. The term “prognostic factor” refers to a biomarker test that infers a low or high risk of a cancer- or treatment-related event, assuming the patient receives no more therapy than he/she has already received, if any. The most commonly used prognostic factors for cancer-related events are the TNM classification system. Examples of molecular tumor biomarkers include PSA level at the time of a prostate cancer diagnosis and estrogen receptor (ER) level in breast cancer.

In contrast, a biomarker is predictive if the treatment effect (experimental compared with control) is different for biomarker-positive patients compared with biomarker-negative patients. Importantly, there must be at least two comparison groups available (e.g., two different treatment arms in a randomized trial) to determine whether a biomarker is predictive. A classic example of a predictive molecular tumor biomarker is again the ER level in breast cancer, where only patients with ER-positive tumors will benefit from endocrine (antiestrogen) therapy. Thus, ER level is an example of a biomarker that is both prognostic and predictive. In patients treated without endocrine therapy, ER-positive tumors are associated with a poor outcome (prognostic), and ER positivity identifies patients that will benefit from endocrine therapy (predictive).

There can be quite some confusion about the distinction between prognostic and predictive biomarkers. When trying to predict outcome for a patient, mathematical modeling is performed looking at the relationship between explanatory variables (independent) and an outcome variable (dependent). The explanatory variables are often biomarkers, and these are most often prognostic factors. However, in statistical terms, the explanatory variables are usually referred to as predictive variables, which may generate some confusion. Avoiding the term “predictive factor,” for what is in fact a prognostic factor, can help circumvent some of the confusion. An often-used terminology is therefore to describe these factors as being associated with outcome, rather than predicting outcome.

Figure 1 illustrates the differences between prognostic and predictive biomarkers in theoretical examples of patients treated with radiotherapy +/− a radiation sensitizer. Figure 1a shows a typical prognostic factor associated with outcome in patients treated with standard radiotherapy outside a clinical trial. Figure 1b shows the same marker in patients treated with radiotherapy plus a sensitizer, again outside a clinical trial. In patients treated with the sensitizer, an association as seen in Fig. 1b is sometimes mistaken as a demonstration that the biomarker-positive patients benefit from the sensitizer. This is not possible to determine outside the context of a clinical trial, and when comparing Fig. 1a, b, the biomarker is most likely a prognostic factor without any predictive value. When looking in another example, the same difference in outcome is seen in patients treated with the sensitizer (Fig. 1d). In this example, however, patients are from a randomized trial. When comparing patients treated with (Fig. 1d) and without (Fig. 1c) the sensitizer, it is seen that in this example, the marker is strictly a predictive marker, identifying patients that will benefit from the sensitizer, and not a prognostic marker (as there is no difference in Fig. 1c). The marker could be a specific mutation that in itself does not affect outcome after radiotherapy but makes the tumor cells sensitive to the sensitizer. The fact that Fig. 1b, d are identical, but the marker is prognostic in Fig. 1b and predictive in Fig. 1d, illustrates the need of randomized clinical trials when studying predictive markers.
Fig. 1

Illustrations of the differences between prognostic and predictive biomarkers in theoretical examples of patients treated with radiotherapy +/− a radiation sensitizer. A prognostic marker investigated in cohorts outside randomized clinical trials (a, b). A predictive marker (c, d) and a prognostic/predictive marker (eh) investigated in randomized clinical trials

Figure 1e–h illustrate data from a different randomized trial. In this case the marker is both prognostic, i.e., associated with outcome in patients treated with standard radiotherapy (Fig. 1e), and predictive (Fig. 1f). The observation that a predictive marker can show no difference in outcome among patients treated with the sensitizer further illustrates the need for a randomized trial. The ideal way to present data on predictive biomarkers is to divide patients on biomarker status and look at the effect of the experimental treatment in both groups. This is illustrated in Fig. 1g, h which include the exact same data as Fig. 1e, f but organized by biomarker status. When plotted this way, it can be seen that the sensitizer does not have an effect in biomarker-negative patients but improves outcome in biomarker-positive patients. An example of such a biomarker is a 15-gene signature for hypoxia, which, as described below, is prognostic in patients treated with standard radiotherapy and predictive for the benefit of the sensitizer nimorazole (Toustrup et al. 2011, 2012a).

When analyzing data on the predicative value of a biomarker, it is tempting to conclude that if there is no association with outcome in the biomarker-negative cohort, but a significant association in the biomarker-positive group (as in Fig. 1g, h), then the biomarker is predictive for the benefit of the experimental arm. According the REMARK guidelines (Altman et al. 2012), a significant effect in one group and a nonsignificant effect in the other are not sound evidence that the effect of the marker differs by subgroup. A test of interaction is required to rigorously assess whether effects are different in subgroups. In the examples of Fig. 1g, h, a test for interaction (also known as effect modification) between biomarker and treatment can be performed by testing the multiplicative term for significance in a Cox model (see Knol and Vander Weele (2012) for details on how to perform and report tests for interaction).

Tumor Biomarkers: Biomarkers of Resistance to Radiation Therapy

Intrinsic Radiosensitivity

Tumor radiosensitivity varies to a considerable degree both across tumor types and also between the same tumor type, independent of the impact of microenvironment and of fractionation parameters (Baumann et al. 2016). The possibility of identifying which tumors are particularly sensitive or resistant to radiation therapy could be of great benefit for treatment. As biomarkers for radiation sensitivity, the subnuclear accumulation (also referred to as radiation-induced foci, RIF) of proteins in the DNA damage response (DDR), such as 53BP1 (tumor suppressor p53-binding protein 1), and the phosphorylation of H2AX (γH2AX), a histone protein which is phosphorylated during DNA repair, have been used. The γH2AX foci are formed at the site of the DNA double-strand breaks (DSBs) and can be visualized in tissue sections by immunofluorescence staining. Timing of measurements of radiation-induced foci is of great importance, as the level of γH2AX foci is time dependent. The maximum level is reached after approximately 10–60 min and correlates closely to the number of DSBs induced by the radiation, whereas the level of foci at later time points, the residual foci, indicates the number of yet unrepaired DSBs. The rate of decrease of the level of radiation-induced γH2AX foci correlates with the rate of DNA repair. Residual foci still detectable after 24 h or more after irradiation represent misrepaired or incompletely repaired lethal DSBs (Fig. 2). This has shown to be the level that correlates best with cell survival and with radiosensitivity (Rothkamm et al. 2015; Willers et al. 2015).
Fig. 2

Anti-γH2AX staining on human skin fibroblasts. (a) Without irradiation. (b) 10 min after a 2Gy irradiation. (c) Typical γH2AX foci kinetics curve after a 2Gy irradiation with X-rays. (Courtesy of Larry Bodgi)

Measurement of DNA damage foci as a marker for radiosensitivity can be employed in an ex vivo assay, where tissue samples from patients are cultured prior to ex vivo treatment and the DDR is detected at certain time points after treatment (Siddiqui et al. 2015). The γH2AX foci levels have clinically been tested as markers for normal tissue reactions, but also the use of γH2AX as a biomarker for sensitivity of tumors to radiotherapy, e.g., in lung tumors, has been the subject of studies (Rothkamm et al. 2015).

HPV Status

Human papillomavirus (HPV) has been shown to have an impact on radiosensitivity and outcome in especially head and neck squamous cell carcinoma (HNSCC), but also in other tumor entities, such as in anal cancer. HPV-associated HNSCC represents a distinct subgroup of HNSCC characterized by a distinct molecular biology, epidemiology, and better prognosis (Lassen 2010). One of the distinct molecular biology features of HPV is an altered cell cycle control, which leads to an upregulation of p16. As biomarkers for HPV status, various detection methods have been employed, e.g., detection of HPV DNA and detection of HPV E6/E7 mRNA (from transcriptionally active virus) or p16 expression as a surrogate marker for HPV (Fig. 3). There is an ongoing discussion on which assay to use universally, as all the methods have their pros and cons, e.g., for p16 expression the advantages are simple and very sensitive assays, which detect all high-risk types of HPV. Furthermore, p16 expression represents those tumors in which HPV has been involved in the carcinogenic process. The disadvantage is suboptimal specificity where p16 expression can be driven by some non-viral mechanism. This encompasses a risk of false-positive results (Westra 2014).
Fig. 3

p16 staining as pseudomarker for HPV status. (a) Tonsillar carcinoma with strong specific tumor cell immunohistochemical staining of p16 expression. (b) The influence of HPV-associated p16 expression on response to conventionally fractionated radiotherapy. Actuarial estimated locoregional tumor control rates in patients with p16-positive and p16-negative carcinoma of the pharynx and supraglottic larynx. (Modified from Lassen (2010) with permission)

Cancer Stem Cells (CSCs)

Cancer stem cells are the cells within a tumor that are capable of self-renewal and generate the heterogeneous lineages of cancer cells that comprise the tumor. The level of CSC in the tumor is an important factor to the needed radiation dose to achieve local tumor control, as surviving CSCs determine tumor recurrences. Thus for a higher CSC number, a higher radiation dose is necessary to cure the tumor. Furthermore, CSC-related radiobiological mechanisms, for example, repopulation during treatment and recovery from radiation-induced damage between the single fractions, have been shown to increase tumor resistance against radiotherapy.

For detection of CSC, flow sorting techniques have been used to isolate cell populations on the basis of cell surface markers that are differentially expressed in tumor cell subpopulations. For the determination of levels of stem cell markers, both immunohistological techniques and qPCR have been employed. The expression level of CSC markers has been suggested as a potential surrogate of CSC density.

There are a range of markers that have been explored, such as CD133 in colon, brain, and pancreatic tumors, as well as CD24 and CD44 in breast, prostate, colon, and pancreatic cancer. The specific markers vary considerably depending on tumor type or tissue of origin, and there is so far no universal marker for CSCs. A concern is that inconsistency in the data on the effect of high stem cell populations, both in preclinical and clinical data, points toward issues on the specificity of the currently identified cancer stem cell markers. It can also be difficult to distinguish the different factors contributing from each other, for example, CSCs have been shown to reside in hypoxic niches in the tumor, and the hypoxia may contribute to increased radioresistance of cancer stem cells (CSCs) (Baumann et al. 2008, 2009).

Tumor Hypoxia

Tumor hypoxia is a well-known trait in most solid tumors, which is mainly due to an inadequate and heterogeneous vascular network (Vaupel et al. 1989). Tumor hypoxia causes resistance to radiation therapy, as well as more aggressive and invasive tumors, and has been associated with poor prognosis in different cancers.

A lot of effort has gone into identifying usable biomarkers for tumor hypoxia, as the clinical impacts of being able to characterize patients in terms of prognosis or select patients for hypoxia-modifying treatments are remarkable. One of the used approaches is to use hypoxia-upregulated genes as endogenous markers. Gene expression is highly affected by cellular oxygen levels, and the majority of the affected genes are orchestrated by hypoxia-inducible factor, HIF1. HIF1 is a transcription factor, which is the main regulator of the transcriptional response to hypoxia. Some of these genes, such as CAIX and GLUT1, have been employed as hypoxic markers, but with varying success. There has been a great focus on developing gene expression signatures, which is a group of genes, instead of single markers. More than 30 hypoxia gene expression signatures have been published, which have been developed in different ways and with different characteristics. Currently, there is no consensus as to the best way to derive hypoxia gene expression signatures. No single gene is represented in all signatures, but some genes are heavily represented in the gene signatures, such as BNIP3, F3, and LOX. The signatures are at different levels in respect to clinical usability and validation.

In HNSCC more signatures have been developed. The Toustrup 15-gene classifier was based on an in vitro study in a panel of HNSCC cell lines, identifying genes that were upregulated by low oxygen concentration, independent of pH. It was developed in a HNSCC training cohort, consisting of 58 patients with the pO2 level measured by oxygen electrode measurements, and validated in the DAHANCA 5 cohort, a Danish study randomizing patients to either receiving nimorazole, a hypoxia modifier, or placebo together with radiotherapy. In this cohort, the 15-gene classifier was shown to be both prognostic and have predictive impact for hypoxic modification (Fig. 3) (Toustrup et al. 2012a). Another hypoxia signature in HNSCC is the 26-gene classifier by Eustace et al., which is based on a metagene signature developed for patients with head and neck, breast, and lung cancers. In the cohort from the Dutch ARCON trial, the patients classified as “more hypoxic” according to the 26-gene classifier showed a significantly improved locoregional control when treated with carbogen and nicotinamide, two hypoxia-modifying agents, in addition to radiotherapy compared to radiotherapy alone in laryngeal cancer (Harris et al. 2015) (Fig. 4).
Fig. 4

Predictive impact of the 15-gene hypoxia classifier in “more” hypoxic tumors (a) versus “less” hypoxic tumors (b) in terms of locoregional control. (Modified from Toustrup et al. (2012a) with permission)

For all signatures, at the current level, there is a need for a continued validation, especially in the decisive step of advancing from retrospective to prospective classification of patients and their allocation to hypoxia-modifying therapies in the clinic.

Besides mRNA-based biomarkers, microRNAs (miRNAs) have also been shown to be inducible by hypoxia, with hsa-mir-210 having shown a prognostic significance in HNSCC treated with postoperative radiotherapy (Baumann et al. 2016; Harris et al. 2015; Toustrup et al. 2012b).

Tumor Heterogeneity

Intra-tumor heterogeneity is a well-known feature of most cancers, which, among others, originates from genomic instability, one of the hallmarks of cancer (Cyll et al. 2017). Genomic instability covers a variety of alterations to the genome, ranging from small to large structural and numerical alterations, and potentially leads to diverse populations of cells. In the case of tumor hypoxia, the tumor heterogeneity originates from the heterogeneous vascular network, leading to areas with low oxygen concentration, whereas other areas are well perfused and with an adequate oxygen concentration. Recognizing the biological heterogeneity of tumors, there is a potential for sampling error when a small biopsy is taken from a much larger tumor. This can potentially also affect biomarker analyses, as a single sample from one small tumor region might not be optimal for characterizing the tumors’ radiosensitivity, level of CSCs, or hypoxic status. However, sampling multiple regions from the same tumor is impractical for most cancer types, and an analysis within a relatively small sample might also be of predictive value. This should be taken into consideration in the validation procedure, which is discussed at a later point.

Circulating Biomarkers

Circulating biomarkers include factors that can be measure in the blood, such as circulating tumor cells, RNA, DNA, or proteins. As blood samples are easy obtainable and can be attained in a serial manner, as before, during, and after treatment, circulating biomarkers are a very promising area. One interest in the use of circulating biomarkers has been on circulating biomarkers of hypoxia, where the proteins evaluated include osteopontin (OPN) and interleukin-8 (IL8). The abovementioned hypoxia biomarkers, the hypoxia gene expression signatures, are measured in tumor tissue, e.g., in the diagnostic biopsy, which is obtained at baseline, before treatment starts. Using circulating biomarkers could be an advantage in measuring hypoxia biomarkers at different time points during treatment to monitor for reoxygenation effects using sequential blood samples.

The use of circulating DNA and circulating tumor cells (CTCs) as biomarker has in popular terms been named as liquid biopsy. Circulating tumor DNA (ctDNA) is a DNA that is released by the tumor cells into the circulation. The mechanisms of release of nucleic acid into the blood are not yet explained. It is thought to result from necrosis, apoptosis, and release of DNA by phagocytes that have engulfed tumor cells. Of the overall circulating cell-free DNA (cfDNA) in the blood, the fraction that is derived from tumors is for some tumor types very low, as in colorectal cancer where ctDNA as a fraction of total cfDNA was found to range from 0.01% to 1.7% and from 0.02% to 3.2% in NSCLC, whereas it was up to 47% in multiple myeloma. The level of ctDNA is thought to correlate to some degree with tumor size and the rate of tumor cell death. The half-life of ctDNA in the circulation is approximately 0.5–2 h, so only tumor cells that died several hours before sample collection contribute to ctDNA levels. The use of ctDNA as biomarker includes both the pretreatment level of ctDNA and monitoring the kinetics of the levels of ctDNA, as acute changes in ctDNA concentrations during radiotherapy or immediately following treatment might also have a prognostic or predictive value. Another use is noninvasive tumor genotyping via the plasma, which can be an advantage in the lack of tumor biopsy material.

For ctDNA analysis, a technical challenge has been to find techniques that are sensitive and specific enough to detect the ctDNA. The methodology to discriminate tumor-derived DNA from normal cell-free DNA is using the unique genetic profile of the tumor. Used techniques for detection of specific mutations within ctDNA include PCR-based methods, which can preferentially amplify low levels of mutant alleles in the pool of wild-type DNA. Especially digital PCR (dPCR), which has a very high sensitivity and specificity, is often used for ctDNA analysis (Chaudhuri et al. 2015; Rostami and Bratman 2017).

Predictive Markers for the Normal Tissue Response to Radiotherapy

As mentioned in the introduction, the ability to predict normal tissue complication risk prior to radiotherapy has been a long sought goal in radiobiology (Russell and Begg 2002). For more than 25 years, various attempts have tried to establish a predictive assay for normal tissue radiosensitivity (3B). In the following, these efforts will be summarized.

In Vitro Assays

In the early 1990s, radiobiology was heavily influenced by the so-called target-cell hypothesis according to which normal tissue toxicity was assumed to be the result of depletion of certain target cells. Under this hypothesis, it was investigated if the risk of normal tissue toxicity could be predicted from the study of cell irradiated in vitro. Associations were sought between in vitro radiosensitivity of fibroblasts and risk of radiation-induced fibrosis. In addition, some studies looked at the in vitro radiosensitivity of lymphocytes or lymphoblasts (West et al. 2001). Generally, no significant associations were found for acute toxicity. A number of relatively small studies reported significant associations between the in vitro radiosensitivity of fibroblast and risk of late toxicity. Nevertheless, these associations could not be confirmed in subsequent larger studies (Russell and Begg 2002). Interest was also taken in the development of rapid assays measuring cellular or subcellular damage such as DNA lesions, chromosome aberrations, or apoptosis. A classic example is the so-called alkaline comet assay. The results of these studies have been inconsistent. A major challenge in the interpretation of these results has been the limited size of the studies and a substantial heterogeneity concerning the investigated assays (Russell and Begg 2002). Thus, it has not been possible to establish an in vitro assay suitable for routine clinical use.

Radiogenomics: The Search for Genetic Predictors of Radiosensitivity

In the late 1990s, increasing interest was taken in the hypothesis that normal tissue radiosensitivity is under genetic control and that normal tissue complication risk could be predicted from genetic analyses (Andreassen et al. 2016a). This concept received support from the observation that patients suffering from certain rare genetic syndromes like ataxia telangiectasia, Bloom’s syndrome, Fanconi’s anemia, and Nijmegen breakage syndrome experience devastating normal tissue reactions if treated with radiotherapy. However, these syndromes characterized by Mendelian inheritance are extremely rare and probably of little relevance when addressing the average cancer patients. It was hypothesized that heterozygous carriers of truncating mutations in genes such as ATM, BRCA1, and BRCA2 could make up a radiosensitive subpopulation. Nevertheless, a number of relatively small studies did not provide indications that such sequence alterations are overrepresented among patients with excessive normal tissue reactions nor that carriers of these mutations exhibit a higher risk of normal tissue complications (Andreassen et al. 2016a).

Basic Hypotheses

At the turn of the millennium, substantial efforts were made to unravel the genetics underlying a variety of different biomedical phenotypes. This research provided important insights into the human genome: The human genome is made up by three billion base pairs. In approximately 1 in every 300 base pairs, variation is present. Ninety percent of this variation is made up by approximately 11 million single nucleotide polymorphisms (SNPs). An SNP is defined as a single base substitution in which the least common allele has a population-based abundance of at least 1%. The remaining 10% is made up by a heterogeneous group of other sequence alterations (e.g., point mutations, copy-number variants, deletions, insertions, inversions, translocations, etc.) often referred to collectively as “rare sequence alterations.” The number of different rare sequence alterations in the human genome is likely to substantially exceed that of SNP. By means of these insights, a more comprehensive hypothesis about the genetics of normal tissue radiosensitivity could be formulated (Andreassen et al. 2016a). The hypothesis can be encompassed in the following three paragraphs:
  1. 1.

    Normal tissue radiosensitivity is as a complex trait dependent on the combined influence of sequence alteration in several genes.

    The mere observation that normal tissue radiosensitivity is characterized by a continuous spectrum rather than falling into distinct categories is per se indicative of a polygenic inheritance. The biological response to irradiation is presumably very complex and involves a multitude of different pathways and genes. Given the widespread existence of sequence alterations in the genome, numerous loci with potential impact on clinical radiosensitivity are likely to exist.

     
  2. 2.

    Single nucleotide polymorphisms may make up a proportion of the genetics underlying differences in clinical normal tissue radiosensitivity.

    SNPs account for approximately 90% of the inter-individual sequence variation within human populations. SNPs in coding regions may alter protein function, whereas SNPs in regulatory regions may affect gene expression/protein synthesis rate. Thus, SNPs have the potential to affect various phenotypes including normal tissue radiosensitivity.

     
  3. 3.

    Some genetic alterations are expressed selectively through certain types of normal tissue reactions, whereas others exhibit a “global” impact on radiosensitivity.

    Clinical studies addressing normal tissue radiosensitivity have indicated that the risks of different types of adverse reactions after radiotherapy are not strongly associated with each other. This observation indicates that some genetic alterations have to be expressed selectively through certain types of normal tissue reactions. However, patients suffering from one of the aforementioned “radiosensitive syndromes” appear to experience a general enhancement of clinical radiosensitivity apparently affecting both acute and late morbidity. Based on this, it seems likely that some genetic alterations affect clinical radiosensitivity in a generalized manner, whereas others exhibit a differential expression through different types of normal tissue reactions (Andreassen et al. 2016a).

     

Studies in Radiogenomics

As of May 2018, a total of 168 studies have been carried out in order to establish associations between genetic sequence alterations and the risk of normal tissue toxicity after radiotherapy (Fig. 5). One hundred sixty studies were based on a candidate gene approach, whereas only eight used a genome-wide approach. The vast majority of the studies have addressed SNPs. The primary reason for this is presumably that SNPs are catalogued, they are easy to asses, and they have an abundance that makes them easy to work with in terms of statistical power. In contrast complete sequencing will usually be needed in order to search for rare sequence alterations. This is still relatively costly, and the low population frequency of the alterations makes up a major challenge from a statistical point of view. Table 1 summarizes some of the most compelling associations identified in these studies. A comprehensive overview of the studies is provided in Herskind et al. (2016).
Fig. 5

Overview of 168 original papers (including individual patient data meta-analyses) published in radiogenomics as of May 2018. Blue columns indicate candidate gene studies, whereas orange columns*indicate GWASs

Table 1

Overview of compelling associations in radiogenomics. Criteria for selection of SNPs: Candidate gene study; p-value below 0.05 in a study with several independent cohorts with total size of at least 1,000 patients. GWAS: p-value below 5.5 × 10−8

Author, publication year

Study type, N

SNP

Gene

Cancer site, normal tissue endpoint(s)

Effect size

p value

Talbot, 2012

Candidate gene, 2036

Rs1800629

SNP near TNFα

Breast, overall late toxicity

OR 2.45

0.003

Seibold, 2015

Candidate gene, 2636

Rs2682585

XRCC1

Breast, overall late toxicity

OR 0.77

0.02

Andreassen, 2016

Candidate gene, 5456

Rs1801516

ATM

Breast and prostate, various endpoints

OR ≈ 1.2–1.5

0.0001

Kerns, 2010

GWAS, 79

Rs2268363

FSHRH

Prostate, erectile dysfunction

OR 7.03

5.46 × 10−8

Kerns, 2013

GWAS, 1149

rs7120482 rs17630638

Intergenic

Prostate, rectal bleeding

OR 3.1–6.7

5.40 × 10−8

Barnett, 2014

GWAS, 2011

rs2788612

KCND3

Prostate, rectal incontinence

RR 9.91

1.05 × 10−12

Fachal, 2014

GWAS, 1742

Rs264663

TANC1

Prostate, overall late toxicity

Per allele OR ≈ 6

4.46 × 10−11

Kerns, 2016

GWAS, 1564

rs17599026

Intergenic

Prostate, variuous endpoints

OR 3.2

4.16 × 10−8

rs7720298

OR 2.7

3.21 × 10−8

Rs11230328

Beta 0.3

6.27 × 10−10

Candidate Gene Studies

In the candidate gene approach, a limited number of sequence alterations are selected for investigation based on the hypothesis that they are likely to affect the phenotype of interest. As mentioned previously, a few rather small studies have addressed the impact of truncating ATM and BRCA mutations upon normal tissue radiosensitivity. Apart from this, the candidate gene studies have exclusively focused on SNPs. These studies typically investigated SNPs in genes involved in processes such as detection of DNA damage (e.g., ATM), DNA repair (e.g., XRCC1, XRCC3, and APEX), tissue remodeling (e.g., TGFB1 and TIMP), and scavenging of reactive oxygen species (e.g., SOD2 and GSTP1). More than 100 different genes were investigated as part of this research. With a median sample size around 150 patients, most of the studies were very small. Around two thirds of these studies have reported significant associations. Nevertheless, the findings were often inconsistent and independent replication rarely took place. In hindsight, it seems obvious that the vast majority of these studies have suffered from methodological shortcomings such as lack of statistical power and unattended multiple testing problems. Nevertheless, a few candidate gene studies have actually reported compelling SNP associations (Herskind et al. 2016; Scaife et al. 2015) (Table 1). These studies included more than a thousand patients and involved independent replication cohorts. A study published in 2012 investigated 43 SNPs in 35 genes related to TGF-β signaling pathways in 2,036 breast cancer patients (Talbot et al. 2012). It showed an increased risk of late toxicity (OR = 2.45, 95% CI 1.52–3.98) for patients being homozygous for the minor allele of the rs1800629 SNP. This SNP is located a few hundred bases upstream of the transcription start site of the TNFα gene and relatively close to the lymphotoxin-alpha gene (LTA). More recently, a candidate gene study investigated 305 SNPs in 59 genes related to oxidative stress and DNA repair pathways in a discovery set of 753 breast cancer patients (Seibold et al. 2015). Subsequently, the top ranking 10 SNPs were tested in several replication cohorts with a total of 1,883 patients. This study identified a SNP (rs2682585) in the DNA repair gene XRCC1 for which carriers of the minor allele had a reduced risk of late skin toxicity (multivariate OR 0.77, 95% CI 0.61–0.96, p = 0.02) and overall toxicity (regression coefficient − 0.08, 95% CI −0.15 to −0.02, p = 0.02). Finally, in 2016 an “individual patient data meta-analysis” with more than 30,000 recordings of normal tissue toxicity from 5,456 breast and prostate cancer patients showed that the possession of the minor allele at the ATM rs1801516 SNP increases the risk of radiation-induced toxicity with an odds ratio of approximately 1.5 for acute toxicity and 1.2 for late toxicity (Andreassen et al. 2016b).

The Genome-Wide Association Study

The genome-wide association study (GWAS) provides a radical alternative to the candidate gene approach (Andreassen et al. 2016a). The GWAS makes use of the fact that SNPs cluster into haplotypes due to linkage disequilibrium. This means that the majority of the estimated 11 million SNPs in the human genome can be indirectly assessed by means of a microarray that genotypes around 500,000 to a few million well-chosen “tagging SNPs.” Thus, the GWAS provides the opportunity to conduct a hypothesis-free search for associations across the entire genome. In a genome-wide SNP experiment, the effective number of independent variables is approximately one million. Thus, the GWAS has an inherent severe multiple testing problem. In order to adjust for that, the significance threshold is usually set at 5 × 10−8 which is simply the result of a Bonferroni correction applied to a million parallel experiments. In order to maintain statistical power under such stringent conditions, large patient cohorts are needed: As shown in Fig. 6, a study of 1,000 patients will be well powered to detect SNPs with a genotype relative risk of 2.0, whereas it will take a study of 10,000 patients to detect GRRs of 1.25 across a reasonable span of risk allele frequencies.
Fig. 6

Doing a GWAS is fishing in the ocean of trait-associated SNPs. The larger the study, the deeper the net will reach. Blue curves indicate the sample size needed to obtain 80% power for different genotype relative risks (GRRs) according to the risk allele frequency. The figure is derived from data generated by CaTS power calculator for GWASs (http://csg.sph.umich.edu//abecasis/CaTS/interface.html). Model assumptions: Case-control ratio 1:1, phenotype prevalence 25%, significance threshold 10−7, and multiplicative inheritance. (Modified from Andreassen et al. (2016) with permission)

GWASs in Radiogenomics

As of May 2018, eight GWASs addressing normal tissue radiosensitivity have been published (Fig. 5). Below, findings reaching or approaching genome-wide significance will be summarized. The first GWAS in normal tissue radiobiology investigated the risk of erectile dysfunction among 79 African Americans treated with radiotherapy for prostate cancer (Kerns et al. 2010). It identified a SNP (rs2268363) in the gene of the follicle-stimulating hormone receptor (FSHR) with a relative risk of 7.03 and a p-value of 5.46 × 10−8. Another GWAS included 1,149 prostate cancer patients in a two-stage GWAS (Kerns et al. 2013). This study showed an association between two SNPs in “a gene desert” (rs7120482 and rs17630638) and risk of rectal bleeding after radiotherapy (OR 6.7 and 3.1 in first and second stage, respectively, combined p-value = 5.4 × 10−8).The largest GWAS conducted so far in normal tissue radiobiology was published in 2014 (Barnett et al. 2014). This two-stage GWAS included more than 3,500 breast and prostate cancer patients. The study looked for associations with regard to overall toxicity (STAT score) as well as a number of “individual endpoints” (e.g., telangiectasia, rectal bleeding, nocturia, and rectal incontinence). Even though the study had substantially more power to detect associations for overall toxicity, the strongest associations were shown for individual endpoints. For the rs2788612 SNP in the gene KCND3, a significant association with risk of rectal incontinence was found (RR 9.91, p = 1.05 × 10−12). KCND3 encodes a voltage-gated potassium channel involved in the electrochemical processes underlying smooth muscle contraction. A GWAS published in 2014 was conducted as a collaborative project involving three different patient cohorts from Spain, the UK, and North America (Fachal et al. 2014). It included 1,742 prostate cancer patients. The study identified a locus in the TANC1 gene that was associated with risk of overall late toxicity with a per allele odds ratios around 6 and a p-value of 4.46 × 10−11. TANC1 is involved in regeneration of damaged muscle fibers. In 2016, a meta-analysis of individual level data from four genome-wide association studies including 1,564 prostate cancer patients was published (Kerns et al. 2016). It identified two SNPs: rs17599026 on 5q31.2 that was associated with urinary frequency (OR 3.12, 95% CI 2.08–4.69, p-value 4.16 × 10−8) and rs7720298 on 5p15.2 that was associated with decreased urine stream (OR 2.71, 95% CI 1.90–3.86, p-value = 3.21 × 10−8).

As accounted for above, the GWASs conducted so far in normal tissue radiobiology have identified a number of SNPs reaching or approaching genome-wide significance. Interestingly, these SNPs are not related to genes involved in radiobiology in a narrow sense. Instead, they are related to other physiological phenomena with a plausible relationship to the investigated normal tissue complications. This experience underlines the exploratory nature of genome-wide approaches and indicates that GWASs are likely to broaden our understanding of the mechanisms underlying radiation-induced normal tissue complications (Andreassen et al. 2016a). Furthermore, several of the SNPs were associated with specific types of normal tissue toxicity rather than radiosensitivity in general. This supports the aforementioned hypothesis that some genetic variants are expressed selectively through certain types of normal tissue reactions (Andreassen et al. 2016a).

Lessons Learned in Other Scientific Fields

During the last decade, substantial efforts were made to unravel the genetics of various phenotypes that are assumed to have a complex polygenic background. As of March 2019, more than 3,800 GWASs addressing various biomedical phenotypes were published. These studies have provided more than 130,000 trait-associated SNPs (updated lists are available at http://www.ebi.ac.uk/gwas/). This research has shed important new light on the allelic architecture presumably underlying the most complex traits (Andreassen et al. 2016a). The insights gained can be summarized as follows:

The Candidate Gene Approach May Not Be of Much Use

Most of the SNPs identified by GWAs were not located in genes anticipated to be of major importance for the phenotype. Often, the SNPs were located in noncoding sequences. Generally, this experience questions the value of the candidate gene approach (Andreassen et al. 2016a).

Noncoding Sequence Is of Major Importance

Less than 3% of the human genome is made up by coding sequence (i.e., sequence that is expressed as protein). The vast majority of traits with a Mendelian pattern of inheritance are related to alterations in coding sequence. Nevertheless, around 88% of SNPs affecting various complex traits are located in noncoding sequence. It is to an increasing extent recognized that the noncoding sequence harbors important regulatory functions. Sequence variants in such regulatory elements have the potential to affect phenotype through altered gene expression (Andreassen et al. 2016a).

The Typical Impact of each SNP Is Relatively Small

It is a common experience in most GWASs that the majority of the identified trait-associated SNPs only exhibited a modest impact on phenotype. Usually, relative risks well below 1.2 were found. Many of the GWASs were well powered to detect associations with GRRs above 1.5, but such associations were rarely reported. This probably implies that the typical impact of SNPs is rather small. As shown in Table 1, some of the SNPs identified by GWASs addressing normal tissue toxicity had substantially larger effect sizes. This may indicate that normal tissue radiosensitivity has an allelic architecture that differs from that of other complex traits. However, it may also reflect that GWASs tend to overestimate the effect size due to the so-called winner’s curse phenomenon (Andreassen et al. 2016a).

Missing Heritability

The genetic background of a number of different biomedical traits has been quite extensively investigated using GWASs. Regardless of that, the identified SNPs typically only accounted for a limited proportion of the expected heritability. For instance, a GWAS meta-analysis with more than 250,000 subjects identified a total of 697 SNPs that explain around 20% of the heritability of height (Wood et al. 2014). Large GWASs addressing breast cancer susceptibility have identified 148 SNPs that account for approximately 18% of the familial risk of breast cancer (Fachal and Dunning 2015; Michailidou et al. 2017). Thus, missing heritability is present. Some of the missing heritability is likely to be made up by SNPs with effect sizes below the detection threshold of the current GWASs. For breast cancer susceptibility, it has been estimated that another 1,000 trait-associated SNPs with relative risks below 1.05 may exist (Fachal and Dunning 2015). Furthermore, the missing heritability presumably also indicates that other types of sequence variants than SNPs (i.e., various “rare sequence variants”) contribute to complex trait genetics. For instance, this is the case for breast cancer susceptibility where the known high- and intermediate-risk alleles (i.e., BRCA1, BRCA2, PTEN, and ATM mutations) make up around 35% of the expected heritability (Fachal and Dunning 2015). Thus, it seems likely that the genetics underlying most complex traits is made up by a spectrum ranging from comment alterations (i.e., SNPs) with a small impact on phenotype to rare sequence alterations with a stronger impact (Andreassen et al. 2016a) (Fig. 7).
Fig. 7

A proposed model for the relationship between allele frequency and relative risk for genetic variants associated with normal tissue toxicity after radiotherapy. The model is inspired from studies addressing breast cancer susceptibility. The actual sequence variants affecting breast cancer susceptibility can be seen in Wood et al. (2014). (Modified from Andreassen et al. (2016a) with permission)

The Genomic Challenge

As accounted for above, the human genome has some inherent characteristics that make it challenging to deal with. Common alterations are likely to have a small impact on phenotype, whereas alterations with major impact on phenotype are likely to be rare. Furthermore, the number of variants to choose from is immense (11 million SNPs and a presumably much higher number of rare alterations). Even though, we do not know the exact allelic architecture underlying normal tissue radiosensitivity, it is not unreasonable to assume that it might resemble that of height or cancer susceptibility. Thus, the task we are faced with can presumably be described as follows: We have to put together a jigsaw puzzle made up by several hundred small pieces (Fig. 8). Most likely, we do not have to identify all the pieces in order to establish a gene-based predictive model, but we probably have to identify a substantial proportion of the pieces (Scaife et al. 2015). Some of the pieces (i.e., the SNPs) are located in a black box that also contains roughly a million pieces that do not belong to the jigsaw puzzle. The remaining pieces (i.e., the rare alterations) are located in a different box in which we have barely started looking. Within this box, an even larger number of pieces are located that do not belong to the jigsaw puzzle (Andreassen et al. 2016a).
Fig. 8

The genomic challenge. Normal tissue radiosensitivity is likely to be determined by the combined influence of large number different loci. These are to be selected from a very large pool of sequence variants of which the vast majority is not associated with the trait. (Reprinted from Andreassen et al. (2016a) with permission)

Future Directions in Radiogenomics

The aforementioned challenges can be dealt with in different ways. In genome-wide approaches, large sample size is key to success. Therefore, collaborative research projects should be encouraged in order to obtain sufficiently large patient cohorts. As accounted for, SNP-based experiments may need to be complimented by sequencing in order to identify rare sequence alterations affecting radiosensitivity. In addition, a better understanding of the processes and pathways underlying normal tissue toxicity could potentially be used to narrow in the focus to a smaller number of sequence variants, thereby getting rid of some of the multiple testing burdens that relate to genome-wide approaches (Andreassen et al. 2016a).

The International Radiogenomics Consortium (RgC)

The International Radiogenomics Consortium (RgC) was established in 2009 in order to foster large-scale collaborative research projects (West et al. 2010). It has around 200 members from 110 institutions in 26 countries. The consortium has published a set of reporting guidelines in radiogenomics. In 2013, the RgC obtained an EU grant of 6,000,000 Euros for the so-called REQUITE project that will prospectively collect outcome data and biological material from 5,300 lung, prostate, and breast cancer patients in order to establish and validate predictive tests for normal tissue radiosensitivity (West et al. 2014). As part of this project, the predictive test based on lymphocyte apoptosis mentioned below will be validated. The consortium currently undertakes a number of large GWAS meta-analyses in radiogenomics.

Should Alternative Strategies Be Considered?

Despite recent technological advances, it obviously remains a massive undertaking to pursue a comprehensive understanding of the genetics assumed to underlie differences in normal tissue radiosensitivity. In particular, the large sample sizes needed in genome-wide approaches represent a major challenge in radiogenomics. This may favor research into predictive assays that are not based on genotype. As described above, several attempts have previously been made to establish predictive assays based on in vitro radiosensitivity of various cell types (Russell and Begg 2002). Such approaches are still being explored (Andreassen et al. 2016a).

For instance, it has been shown that an inverse association may exist between lymphocyte apoptosis and risk of normal tissue toxicity (Azria et al. 2008). Initiatives are in progress to validate this finding (West et al. 2014). Another study has reported a relatively strong association between the amount of the autophosphorylated ATM protein after in vitro irradiation of fibroblasts and risk of radiation-induced toxicity (Pereira et al. 2018). Predictive assays based on gene expression have proven successful in various settings and could potentially become useful in normal tissue radiobiology (Andreassen et al. 2016a). An example of such a test is a classifier based on the expression pattern of nine genes assessed in fibroblasts after irradiation in vitro. This classifier identified a subset of patients (around 20%) that exhibited substantial radioresistance (Alsner et al. 2007). The test has been independently validated (Andreassen et al. 2013). The clinical utility of this particular test is limited by the fact that it is based on fibroblasts and that it takes several weeks to culture. Nevertheless, the classifier represents an important proof of principle and indicates that predictive assays based on gene expression may represent a rewarding alternative to genotype. Compared to cell survival (or apoptosis), gene expression has the advantage that distinct expression profiles could potentially be identified for different types of normal tissue toxicity. Therefore, this type of classifier may have the ability to reflect the multidimensional nature of clinical normal tissue radiosensitivity (i.e., that the phenotype is made up by a number of sub-phenotypes that are not necessarily strongly correlated) (Andreassen et al. 2016a).

Validation of Biomarkers

When discussing validation of biomarkers, it is important to distinguish between biomarker and biomarker test (Hayes 2015). A biomarker test is used to identify or measure the changes reflected by the biomarker. There may be one or more tests that give an indication of the status of the biomarker. They may measure the same thing, or they may measure very different aspects. An example is the biomarkers for hypoxia which may be measured on multiple different levels as described above.

Before they can be applied in a clinical setting, biomarker test must be validated on three different levels, analytical validity, clinical validity, and clinical utility (Hayes 2015). Analytical validity means that the test for the biomarker is accurate and reliable in the type of specimen to which it will be applied. Examples of such technical validation studies are rare for molecular biomarkers in radiotherapy, but two examples have been performed on hypoxia gene signatures (Betts et al. 2013; Toustrup et al. 2016).

Clinical validity addresses whether the biomarker test divides a population into two or more distinct groups, based on biology or clinical outcomes, with statistical significance. Using the hypoxia gene signatures, this is exemplified in (Eustace et al. 2013; Toustrup et al. 2011, 2012a).

Finally, clinical utility requires that high levels of evidence demonstrate that the use of the biomarker test improves clinical outcomes or that clinical outcomes are identical with less cost or toxicity. Ideally, this level of evidence should come from prospective clinical trials in which the clinical utility for a specific use of the biomarker is the main objective. Such clinical trials are rare in oncology in general and even more so in radiotherapy. One of the hypoxia gene signatures, the previously mentioned 26-gene classifier by Eustace et al., is currently being tested retrospectively within the NIMRAD trial on the hypoxia radiosensitizer nimorazole (ClinicalTrials.gov Identifier: NCT01950689). Another hypoxia gene signature, the Toustrup 15-gene classifier, is currently being tested prospectively in DAHANCA 30, a non-inferiority trial of hypoxia-profile guided hypoxic modification with nimorazole (NCT02661152).

Cross-References

References

  1. Alsner J, Rødningen OK, Overgaard J. Differential gene expression before and after ionizing radiation of subcutaneous fibroblasts identifies breast cancer patients resistant to radiation-induced fibrosis. Radiother Oncol. 2007;83(3):261–6.PubMedCrossRefGoogle Scholar
  2. Altman DG, McShane LM, Sauerbrei W, Taube SE. Reporting recommendations for tumor marker prognostic studies (REMARK): explanation and elaboration. PLoS Med. 2012;9(5):e1001216.PubMedPubMedCentralCrossRefGoogle Scholar
  3. Andreassen CN, Overgaard J, Alsner J. Independent prospective validation of a predictive test for risk of radiation induced fibrosis based on the gene expression pattern in fibroblasts irradiated in vitro. Radiother Oncol. 2013;108(3):469–72.PubMedCrossRefGoogle Scholar
  4. Andreassen CN, Schack LMH, Laursen LV, Alsner J. Radiogenomics – current status, challenges and future directions. Cancer Lett. 2016a;382(1):127–36.PubMedCrossRefGoogle Scholar
  5. Andreassen CN, Rosenstein BS, Kerns SL, Ostrer H, De Ruysscher D, Cesaretti JA, Barnett GC, Dunning AM, Dorling L, West CML, Burnet NG, Elliott R, Coles C, Hall E, Fachal L, Vega A, Gómez-Caamaño A, Talbot CJ, Symonds RP, et al. Individual patient data meta-analysis shows a significant association between the ATM rs1801516 SNP and toxicity after radiotherapy in 5456 breast and prostate cancer patients. Radiother Oncol. 2016b;121(3):431–9.PubMedPubMedCentralCrossRefGoogle Scholar
  6. Atkinson AJ, Colburn WA, DeGruttola VG, DeMets DL, Downing GJ, Hoth DF, Oates JA, Peck CC, Schooley RT, Spilker BA, Woodcock J, Zeger SL. Biomarkers and surrogate endpoints: preferred definitions and conceptual framework. Clin Pharmacol Ther. 2001;69(3):89–95.CrossRefGoogle Scholar
  7. Azria D, Ozsahin M, Kramar A, Peters S, Atencio DP, Crompton NEA, Mornex F, Pèlegrin A, Dubois J-B, Mirimanoff R-O, Rosenstein BS. Single nucleotide polymorphisms, apoptosis, and the development of severe late adverse effects after radiotherapy. Clin Cancer Res. 2008;14(19):6284–8.PubMedPubMedCentralCrossRefGoogle Scholar
  8. Ballman KV. Biomarker: predictive or prognostic? J Clin Oncol. 2015;33(33):3968–71.PubMedCrossRefGoogle Scholar
  9. Barnett GC, Thompson D, Fachal L, Kerns S, Talbot C, Elliott RM, Dorling L, Coles CE, Dearnaley DP, Rosenstein BS, Vega A, Symonds P, Yarnold J, Baynes C, Michailidou K, Dennis J, Tyrer JP, Wilkinson JS, Gómez-Caamaño A, et al. A genome wide association study (GWAS) providing evidence of an association between common genetic variants and late radiotherapy toxicity. Radiother Oncol. 2014;111(2):178–85.PubMedCrossRefGoogle Scholar
  10. Baumann M, Krause M, Hill R. Exploring the role of cancer stem cells in radioresistance. Nat Rev Cancer. 2008;8(7):545–54.PubMedCrossRefGoogle Scholar
  11. Baumann M, Krause M, Thames H, Trott K, Zips D. Cancer stem cells and radiotherapy. Int J Radiat Biol. 2009;85(5):391–402.PubMedCrossRefGoogle Scholar
  12. Baumann M, Krause M, Overgaard J, Debus J, Bentzen SM, Daartz J, Richter C, Zips D, Bortfeld T. Radiation oncology in the era of precision medicine. Nat Rev Cancer. 2016;16(4):234–49.PubMedCrossRefGoogle Scholar
  13. Betts GNJ, Eustace A, Patiar S, Valentine HR, Irlam J, Ramachandran A, Merve A, Homer JJ, Möller-Levet C, Buffa FM, Hall G, Miller CJ, Harris AL, West CML. Prospective technical validation and assessment of intra-tumour heterogeneity of a low density array hypoxia gene profile in head and neck squamous cell carcinoma. Eur J Cancer. 2013;49(1): 156–65.PubMedCrossRefGoogle Scholar
  14. Bibault JE, Fumagalli I, Ferté C, Chargari C, Soria JC, Deutsch E. Personalized radiation therapy and biomarker-driven treatment strategies: a systematic review. Cancer Metastasis Rev. 2013;32(3–4):479–92.PubMedCrossRefGoogle Scholar
  15. Chaudhuri AA, Binkley MS, Osmundson EC, Alizadeh AA, Diehn M. Predicting radiotherapy responses and treatment outcomes through analysis of circulating tumor DNA. Semin Radiat Oncol. 2015;25(4):305–12.PubMedPubMedCentralCrossRefGoogle Scholar
  16. Cyll K, Ersvær E, Vlatkovic L, Pradhan M, Kildal W, Avranden Kjær M, Kleppe A, Hveem TS, Carlsen B, Gill S, Löffeler S, Haug ES, Wæhre H, Sooriakumaran P, Danielsen HE. Tumour heterogeneity poses a significant challenge to cancer biomarker research. Br J Cancer. 2017;117(3):367–75.PubMedPubMedCentralCrossRefGoogle Scholar
  17. Eustace A, Mani N, Span PN, Irlam JJ, Taylor J, Betts GNJ, Denley H, Miller CJ, Homer JJ, Rojas AM, Hoskin PJ, Buffa FM, Harris AL, Kaanders JHAM, West CML. A 26-gene hypoxia signature predicts benefit from hypoxia-modifying therapy in laryngeal cancer but not bladder cancer. Clin Cancer Res. 2013;19(17): 4879–88.PubMedPubMedCentralCrossRefGoogle Scholar
  18. Fachal L, Dunning AM. From candidate gene studies to GWAS and post-GWAS analyses in breast cancer. Curr Opin Genet Dev. 2015;30:32–41.PubMedCrossRefGoogle Scholar
  19. Fachal L, Gómez-Caamaño A, Barnett GC, Peleteiro P, Carballo AM, Calvo-Crespo P, Kerns SL, Sánchez-García M, Lobato-Busto R, Dorling L, Elliott RM, Dearnaley DP, Sydes MR, Hall E, Burnet NG, Carracedo Á, Rosenstein BS, West CML, Dunning AM, et al. A three-stage genome-wide association study identifies a susceptibility locus for late radiotherapy toxicity at 2q24.1. Nat Genet. 2014;46(8): 891–4.PubMedCrossRefGoogle Scholar
  20. Harris BHL, Barberis A, West CML, Buffa FM. Gene expression signatures as biomarkers of tumour hypoxia. Clin Oncol. 2015;27(10):547–60.CrossRefGoogle Scholar
  21. Hayes DF. Biomarker validation and testing. Mol Oncol. 2015;9(5):960–6.PubMedCrossRefGoogle Scholar
  22. Herskind C, Talbot CJ, Kerns SL, Veldwijk MR, Rosenstein BS, West CMLL. Radiogenomics: a systems biology approach to understanding genetic risk factors for radiotherapy toxicity? Cancer Lett. 2016;382(1):95–109.PubMedPubMedCentralCrossRefGoogle Scholar
  23. Kerns SL, Ostrer H, Stock R, Li W, Moore J, Pearlman A, Campbell C, Shao Y, Stone N, Kusnetz L, Rosenstein BS. Genome-wide association study to identify single nucleotide polymorphisms (SNPs) associated with the development of erectile dysfunction in African-American men after radiotherapy for prostate cancer. Int J Radiat Oncol Biol Phys. 2010;78(5): 1292–300.PubMedPubMedCentralCrossRefGoogle Scholar
  24. Kerns SL, Stock RG, Stone NN, Blacksburg SR, Rath L, Vega A, Fachal L, Gómez-Caamaño A, De Ruysscher D, Lammering G, Parliament M, Blackshaw M, Sia M, Cesaretti J, Terk M, Hixson R, Rosenstein BS, Ostrer H. Genome-wide association study identifies a region on chromosome 11q14.3 associated with late rectal bleeding following radiation therapy for prostate cancer. Radiother Oncol. 2013;107(3):372–6.PubMedPubMedCentralCrossRefGoogle Scholar
  25. Kerns SL, West CML, Andreassen CN, Barnett GC, Bentzen SM, Burnet NG, Dekker A, De Ruysscher D, Dunning A, Parliament M, Talbot C, Vega A, Rosenstein BS. Radiogenomics: the search for genetic predictors of radiotherapy response. Future Oncol. 2014;10(15):2391–406.PubMedCrossRefGoogle Scholar
  26. Kerns SL, Kundu S, Oh JH, Singhal SK, Janelsins M, Travis LB, Deasy JO, Janssens ACJE, Ostrer H, Parliament M, Usmani N, Rosenstein BS. The prediction of radiotherapy toxicity using single nucleotide polymorphism-based models: a step toward prevention. Semin Radiat Oncol. 2015;25(4):281–91.PubMedPubMedCentralCrossRefGoogle Scholar
  27. Kerns SL, Dorling L, Fachal L, Bentzen S, Pharoah PDP, Barnes DR, Gómez-Caamaño A, Carballo AM, Dearnaley DP, Peleteiro P, Gulliford SL, Hall E, Michailidou K, Carracedo Á, Sia M, Stock R, Stone NN, Sydes MR, Tyrer JP, et al. Meta-analysis of genome wide association studies identifies genetic markers of late toxicity following radiotherapy for prostate cancer. EBioMedicine. 2016;10:150–63.PubMedPubMedCentralCrossRefGoogle Scholar
  28. Knol MJ, Vander Weele TJ. Recommendations for presenting analyses of effect modification and interaction. Int J Epidemiol. 2012;41(2):514–20.PubMedPubMedCentralCrossRefGoogle Scholar
  29. Lassen P. The role of human papillomavirus in head and neck cancer and the impact on radiotherapy outcome. Radiother Oncol. 2010;95(3):371–80.PubMedCrossRefGoogle Scholar
  30. Michailidou K, Lindström S, Dennis J, Beesley J, Hui S, Kar S, Lemaçon A, Soucy P, Glubb D, Rostamianfar A, Bolla MK, Wang Q, Tyrer J, Dicks E, Lee A, Wang Z, Allen J, Keeman R, Eilber U, et al. Association analysis identifies 65 new breast cancer risk loci. Nature. 2017;551(7678):92–4.PubMedPubMedCentralCrossRefGoogle Scholar
  31. Pereira S, Bodgi L, Duclos M, Canet A, Ferlazzo ML, Devic C, Granzotto A, Deneuve S, Vogin G, Foray N. Fast and binary assay for predicting radiosensitivity based on the theory of ATM nucleo-shuttling: development, validation, and performance. Int J Radiat Oncol Biol Phys. 2018;100(2):353–60.PubMedCrossRefGoogle Scholar
  32. Rostami A, Bratman SV. Utilizing circulating tumour DNA in radiation oncology. Radiother Oncol. 2017;124(3): 357–64.PubMedCrossRefGoogle Scholar
  33. Rothkamm K, Barnard S, Moquet J, Ellender M, Rana Z, Burdak-Rothkamm S. DNA damage foci: meaning and significance. Environ Mol Mutagen. 2015;56:491–504.PubMedCrossRefGoogle Scholar
  34. Russell NS, Begg AC. Editorial radiotherapy and oncology 2002: predictive assays for normal tissue damage. Radiother Oncol. 2002;64(2):125–9.PubMedCrossRefGoogle Scholar
  35. Scaife JE, Barnett GC, Noble DJ, Jena R, Thomas SJ, West CML, Burnet NG. Exploiting biological and physical determinants of radiotherapy toxicity to individualize treatment. Br J Radiol. 2015;88(1051): 20150172.PubMedPubMedCentralCrossRefGoogle Scholar
  36. Seibold P, Behrens S, Schmezer P, Helmbold I, Barnett G, Coles C, Yarnold J, Talbot CJ, Imai T, Azria D, Koch CA, Dunning AM, Burnet N, Bliss JM, Symonds RP, Rattay T, Suga T, Kerns SL, Bourgier C, et al. XRCC1 polymorphism associated with late toxicity after radiation therapy in breast cancer patients. Int J Radiat Oncol Phys. 2015;92(5):1084–92.CrossRefGoogle Scholar
  37. Siddiqui MS, François M, Fenech MF, Leifert WR. Persistent γH2AX: a promising molecular marker of DNA damage and aging. Mutat Res Rev Mutat Res. 2015;766:1–19.PubMedCrossRefGoogle Scholar
  38. Talbot CJ, Tanteles GA, Barnett GC, Burnet NG, Chang-Claude J, Coles CE, Davidson S, Dunning AM, Mills J, Murray RJS, Popanda O, Seibold P, West CML, Yarnold JR, Symonds RP. A replicated association between polymorphisms near TNFα and risk for adverse reactions to radiotherapy. Br J Cancer. 2012;107(4):748–53.PubMedPubMedCentralCrossRefGoogle Scholar
  39. Toustrup K, Sørensen BS, Nordsmark M, Busk M, Wiuf C, Alsner J, Overgaard J. Development of a hypoxia gene expression classifier with predictive impact for hypoxic modification of radiotherapy in head and neck cancer. Cancer Res. 2011;71(17):5923–31.PubMedCrossRefGoogle Scholar
  40. Toustrup K, Sørensen BS, Lassen P, Wiuf C, Alsner J, Overgaard J. Gene expression classifier predicts for hypoxic modification of radiotherapy with nimorazole in squamous cell carcinomas of the head and neck. Radiother Oncol. 2012a;102(1):122–9.PubMedCrossRefGoogle Scholar
  41. Toustrup K, Sørensen BS, Alsner J, Overgaard J. Hypoxia gene expression signatures as prognostic and predictive markers in head and neck radiotherapy. Semin Radiat Oncol. 2012b;22(2):119–27.PubMedCrossRefGoogle Scholar
  42. Toustrup K, Sørensen BS, Metwally MAH, Tramm T, Mortensen LS, Overgaard J, Alsner J. Validation of a 15-gene hypoxia classifier in head and neck cancer for prospective use in clinical trials. Acta oncol. 2016;55(9):1091–8.PubMedCrossRefGoogle Scholar
  43. Vaupel P, Kallinowski F, Okunieff P. Blood flow, oxygen and nutrient supply, and metabolic microenvironment of human tumors: a review. Cancer Res. 1989;49(23): 6449–65.PubMedGoogle Scholar
  44. West CM, Davidson SE, Elyan SA, Valentine H, Roberts SA, Swindell R, Hunter RD. Lymphocyte radiosensitivity is a significant prognostic factor for morbidity in carcinoma of the cervix. Int J Radiat Oncol Biol Phys. 2001;51(1):10–5.PubMedCrossRefGoogle Scholar
  45. West C, Rosenstein BS, Alsner J, Azria D, Barnett G, Begg A, Bentzen S, Burnet N, Chang-Claude J, Chuang E, Coles C, De Ruyck K, De Ruysscher D, Dunning A, Elliott R, Fachal L, Hall J, Haustermans K, Herskind C, et al. Establishment of a radiogenomics consortium. Int J Radiat Oncol Phys. 2010;76(5): 1295–6.CrossRefGoogle Scholar
  46. West C, Azria D, Chang-Claude J, Davidson S, Lambin P, Rosenstein B, De Ruysscher D, Talbot C, Thierens H, Valdagni R, Vega A, Yuille M. The REQUITE project: validating predictive models and biomarkers of radiotherapy toxicity to reduce side-effects and improve quality of life in cancer survivors. Clin Oncol (R Coll Radiol). 2014;26(12):739–42.CrossRefGoogle Scholar
  47. Westra WH. Detection of human papillomavirus (HPV) in clinical samples: evolving methods and strategies for the accurate determination of HPV status of head and neck carcinomas. Oral Oncol. 2014;50(9):771–9.PubMedPubMedCentralCrossRefGoogle Scholar
  48. Willers H, Gheorghiu L, Liu Q, Efstathiou JA, Wirth LJ, Krause M, von Neubeck C. DNA damage response assessments in human tumor samples provide functional biomarkers of radiosensitivity. Semin Radiat Oncol. 2015;25(4):237–50.PubMedPubMedCentralCrossRefGoogle Scholar
  49. Wood AR, Esko T, Yang J, Vedantam S, Pers TH, Gustafsson S, Chu AY, Estrada K, Luan J, Kutalik Z, Amin N, Buchkovich ML, Croteau-Chonka DC, Day FR, Duan Y, Fall T, Fehrmann R, Ferreira T, Jackson AU, et al. Defining the role of common variation in the genomic and biological architecture of adult human height. Nat Genet. 2014;46(11):1173–86.PubMedPubMedCentralCrossRefGoogle Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  • Brita Singers Sørensen
    • 1
    Email author
  • Christian Nicolaj Andreassen
    • 1
    • 2
  • Jan Alsner
    • 1
  1. 1.Department of Experimental Clinical OncologyAarhus University HospitalAarhusDenmark
  2. 2.Department of OncologyAarhus University HospitalAarhusDenmark

Personalised recommendations