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

Abstract

Purpose

Several small studies have indicated that the ATM rs1801516 SNP is associated with risk of normal tissue toxicity after radiotherapy. However, the findings have not been consistent. In order to test this SNP in a well-powered study, an individual patient data meta-analysis was carried out by the International Radiogenomics Consortium.

Materials and methods

The analysis included 5456 patients from 17 different cohorts. 2759 patients were given radiotherapy for breast cancer and 2697 for prostate cancer. Eight toxicity scores (overall toxicity, acute toxicity, late toxicity, acute skin toxicity, acute rectal toxicity, telangiectasia, fibrosis and late rectal toxicity) were analyzed. Adjustments were made for treatment and patient related factors with potential impact on the risk of toxicity.

Results

For all endpoints except late rectal toxicity, a significantly increased risk of toxicity was found for carriers of the minor (Asn) allele with odds ratios of approximately 1.5 for acute toxicity and 1.2 for late toxicity. The results were consistent with a co-dominant pattern of inheritance.

Conclusion

This study convincingly showed a significant association between the ATM rs1801516 Asn allele and increased risk of radiation-induced normal tissue toxicity.

Since 2003, more than 100 published studies have tried to link single nucleotide polymorphisms (SNPs) to the risk of normal tissue toxicity after radiotherapy [1]. Except for 7 studies [2–8], all took a candidate gene approach. One of the main purposes of this research is to identify genetic factors that can be used to establish a predictive assay for normal tissue radiosensitivity [9]. The vast majority of these studies has been underpowered to detect the small effect sizes usually found for SNPs [10]. Even though numerous significant associations were reported [10–12], only very few of these have been independently confirmed [13,14]. The gene product of ATM plays a crucial role in biological response to ionizing radiation. It is involved in the detection of DNA double strand breaks and initiation of pathways that lead to cycle arrest followed by DNA repair or apoptosis [15]. Patients having a truncating mutation in both copies of the ATM gene suffer from the rare syndrome ataxia telangiectasia [16]. It has been reported that these patients develop devastating normal tissue toxicity if given radiotherapy [17]. At the turn of the millennium, ATM mutations were among the first to be investigated for a possible association with normal tissue radiosensitivity. Several studies looked for an association between heterozygocity for truncating ATM mutations and risk of normal tissue toxicity after radiotherapy (reviewed in [18]). These studies did not show any significant association. However, the limited size of the studies should be taken into account in the interpretation of these results. Numerous studies have investigated the impact of SNPs in ATM upon normal tissue toxicity. The SNP designated rs1801516 (c.5557G > A, p.Asp1853Asn) is among the ATM SNPs that have been most extensively investigated. This SNP results in a non-conservative amino acid substitution from an aspartic acid to an asparagine in exon 37 (NM_000051.3). A number of relatively small studies indicated that the minor allele of this SNP increases the risk of normal tissue toxicity after radiotherapy (reviewed in [19–21]). Nevertheless, the findings have not been entirely consistent. Three literature based meta-analyses have addressed the impact of the rs1801516 SNP upon radiation-induced normal tissue toxicity. Two of these meta-analyses demonstrated a significantly increased risk of acute toxicity (N = 1588) [20] and radiation-induced fibrosis (N = 2000) [21] respectively among carriers of the minor (Asn) allele whereas the largest of these meta-analyses (N = 2127) did not show a significant association between the SNP and risk of normal tissue toxicity in broader terms [19]. In order to bring this SNP to a test in the setting of a well-powered investigation, the International Radio-genomics Consortium (RgC) [22] conducted an individual patient data meta-analysis. With the intention to minimize the effect of publication bias, the analysis included published as well as unpublished data. In this paper, we intended the best possible adherence to the STROGAR guidelines for reporting results of genetic association studies in radiogenomics [23].

Material and methods

Study cohorts

Between January 2013 and September 2014, members of the RgC were encouraged to submit patient series (published as well as unpublished) for a meta-analysis addressing the impact of the ATM rs1801516 SNP upon normal tissue radiosensitivity. As part of this process, 17 cohorts (6 for prostate cancer and 11 for breast cancer) were submitted and all of these were included in the present meta-analysis. The cohorts summed up to 5624 patients. However, 168 patients were excluded from the analysis due to missing genotype data (n = 31) or missing toxicity data (n = 137) leaving 5456 patients to be included in the analysis (Table 1). Consort diagram of the analysis is provided in Supplementary Fig. S1. For three of the cohorts, the rs1801516 SNP was indirectly genotyped by means of the rs4988023 SNP (Table 1). This SNP is in strong linkage with the rs1801516 SNP (r² = 1) [11] and we therefore consider it an appropriate surrogate for the rs1801516 SNP. A brief description of the cohorts, genotyping procedures and scoring of the normal tissue reactions is given in Supplementary Text File 1 and in [5,6,8,13,14] and [24–30]. In addition to genotype and toxicity data, recordings of covariates with potential influence on radiation response were collected.

Table 1.

Data included from each cohort with total number of observations (Obs) and number of patients (Ptn). (b) = breast cohort, (p) = prostate cohort. Ref. indicates the reference in which the cohort is described.

Cohort	Ref.	N (all)	STAT global		STAT acute		Acute skin		Acute rectal		STAT late		Telengiectasia		Fibrosis		Late rectal
			Obs	Ptn	Obs	Ptn	Obs	Ptn	Obs	Ptn	Obs	Ptn	Obs	Ptn	Obs	Ptn	Obs	Ptn
CIHR1 (b)†	14	158	632	158							632	158	158	158	158	158
CIHR2 (b)†	14	78	465	78							465	78	78	78	78	78
DBCG I (b)†	24	41	123	41	41	41	41	41			82	41	41	41	41	41
DBCG II (b)	25	234	702	234	234	234	234	234			468	234	234	234	234	234
GenePARE (b)	8,26	179	358	179							358	179	179	179	179	179
LeND (b)	13	583	1165	583							1165	583	583	583	582	582
MARIERAD (b)†	14	273	545	273							545	273	273	273	272	272
Pre START (b)	27	52	104	52							104	52			52	52
RAPPER (b)†*	6	940	11651	940	925	925	925	925			10726	929	858	858	737	737
SANT PAU (b)†	30	101	390	101	99	99	99	99			291	101			96	96
VHEBRON (b)†		58	86	58	58	58	58	58			28	28			28	28
AERT RADIOGEN (p)†	28	724	4304	724	2896	724			724	724	1408	470					468	468
CCI (p)†	5	155	465	155							465	155					155	155
GenePARE (p)	5	976	1920	976							1920	976					970	970
Ghent (p)†	29	281	1964	281	1964	281			1964	281
RAPPER CHHiP (p)†*	6	411	4483	411							4483	411					411	411
RAPPER RT01 (p)†*	6	212	1649	212							1649	212					211	211
All breast		2697	16221	2697	1357	1357	1357	1357			14864	2656	2404	2404	2457	2457
All prostate		2759	14785	2759	4860	1005			2688	1005	9925	2224					2215	2215
All		5456	31006	5456	6217	2362	1357	1357	2688	1005	24789	4880	2404	2404	2457	2457	2215	2215

^†Indicates cohorts from which data on covariates with potential impact on the normal tissue outcome were available.

^*Indicates that the rs1801516 SNP was indirectly genotyped by means of the rs4988023 SNP.

Statistical analysis

Overall analytical strategy

The strategy for statistical analysis was determined prior to data analysis. It was decided to analyze the dataset with regard to so called STAT scores (see below) for overall toxicity, acute toxicity and late toxicity. In addition, it was decided to analyze the dataset with regard to separate endpoints for which data were available from at least 1000 patients. This was the case for acute skin toxicity (n = 1357), acute rectal toxicity (n = 1005), telangiectasia (n = 2404), fibrosis (n = 2457) and late rectal toxicity (n = 2215).

The following criteria were used to determine which data should be included in the analysis:

• Toxicity recorded during treatment or within a period of 3 months after treatment was included in the STAT scores for acute toxicity.

• Toxicity recorded 2 years or more from the end of treatment was included in the STAT scores for late toxicity.

• For cohorts that were scored more than once, the recordings closest to the end of treatment were used for the analysis of acute toxicity and the recordings at least 2 years after treatment that had the highest proportion of non-missing values were used for late toxicity.

• All toxicity recordings were included in the STAT score calculations except for endpoints with a frequency of events below 5% (an event was defined as a toxicity score different from zero).

• For the 5 ‘separate endpoints’, data series with event frequencies below 5% were accepted.

Using these criteria, 31,006 individual toxicity measurements were eligible for the meta-analysis. An overview of the data that were included in the analysis is provided in Table 1 and a more elaborate description is given in Supplementary Table S1. Recordings of covariates with potential influence on the normal tissue outcome were available for 3666 patients from 13 cohorts (Supplementary Table S2). A total of 50,458 covariate observations were eligible for the analysis.

Z score conversion

In order to obtain a summary measure of toxicity across different endpoints, scoring systems and institutions, all analyzed toxicity recordings were converted into Z scores using the equation

$${\mathrm{Z}}_{\mathrm{k}}=({\mathrm{S}}_{\mathrm{k}}-\text{mean})/\text{standard deviation}$$

in which Z_k designates the Z score for a patient k having the toxicity score S_k and the mean and standard deviation taken over all cases in the cohort with non-missing scores for that endpoint. This procedure produces a standardized normal variable, i.e. a toxicity score with a mean of zero and a standard deviation of one. By doing so, a measure is obtained of where the patient's toxicity score fits into the sensitivity spectrum of his or her original study cohort. Patients having a toxicity score that was in the right-hand tail of the distribution would be assigned a relatively high Z score whereas a patient with a score close to the population average for the endpoint in question would be assigned a Z score close to zero.

Calculation of STAT and Z STAT scores

Standardized Total Average Toxicity (STAT) scores [31] were calculated for each patient by taking the mean of the (non-missing) Z scores for different endpoints, thereby providing an overall measure of the patient's radioresponsiveness. STAT scores were calculated for overall toxicity (STAT global), acute toxicity (STAT acute) and late toxicity (STAT late).

For some of the cohorts, the STAT scores were based on only one or a few toxicity recordings whereas for others numerous recordings were used for the calculation. The distribution of the mean STAT score depends on the number of toxicity items used in forming the average (i.e. STAT scores derived from a large number of observations tended to have smaller standard deviations than STAT scores derived from a single or few toxicity recordings). To produce a standardized variable, we converted the STAT scores into Z scores (hereafter termed Z STAT). In this way, all Z scores and Z STAT scores have a common format characterized by a mean of zero and a standard deviation of one, making them directly comparable across different endpoints and cohorts.

Statistical test of the association between genotype and normal tissue toxicity

Analysis based on Z STAT scores and Z scores

For each of the three possible ATM rs1801516 genotypes we calculated the means of Z STAT global, Z STAT acute, Z STAT late, Z acute skin, Z acute rectal, Z telangiectasia, Z fibrosis and Z late rectal. Statistical significance was evaluated using the non-parametric Mann–Whitney U test (Wilcoxon rank-sum test) comparing carriers of one or two minor alleles with the common allele homozygotes. In addition, statistical significance was tested using Spearman's rank correlation coefficient as a test for trend over all three possible SNP genotypes.

Analysis of odds ratios for having a score in the upper quartile

In order to provide a measure of radiosensitivity that is easier to translate into a clinical context, we calculated odds ratios for having a Z STAT or Z score in the upper quartile comparing carriers of one or two minor alleles with the common allele homozygotes. Since the data typically had many ties, the proportion of patients having a score in the upper quartile occasionally deviated from 25%. The data were meta-analyzed and Forest plots created using the DerSimonian–Laird random effects-model [32]. The I-squared test was used to test for heterogeneity between cohorts [33]. Funnel plots [34] and Peters test for ‘small study effects’ [35] were used to test for indications of publication bias.

Multivariate analysis, calculation of residuals and analysis of odds ratios for having a residual in the upper quartile

As mentioned previously, information on covariates with potential impact on normal tissue outcome was available for 13 of the cohorts. For each of these cohorts, ordinal logistic regression was used to test for associations between the recorded patient and treatment related factors and the 8 toxicity scores. The tested covariates together with regression coefficients and p-values are listed in Supplementary Table S2. Multivariate analysis of the Z STAT and Z scores with the rs1801516 genotype was then performed on each cohort, including covariates associated with the toxicity score on univariate analysis with a p-value < 0.2. For covariates that were closely related (e.g. various dose volume parameters for the same organ), the most significant one was chosen for the multivariate analysis. After multivariate analysis, residuals were calculated for each patient, for the STAT and Z scores (hereafter termed R-STAT and R-Z scores respectively). The analysis was carried out using STATA (StataCorp, Chicago, USA). A residual is the difference between the observed and the estimated toxicity score and provides a measure of the toxicity not explained by available patient and treatment related factors. Patients with residuals of zero have toxicity entirely accounted for by these factors. Patients with negative or positive residuals have less or greater toxicity respectively than is explained by known factors. In parallel with the analysis of the Z STAT and Z scores, we calculated odds ratios for having an R- STAT or R-Z score in the upper quartile comparing carriers of one or two minor alleles with the common allele homozygotes. The data were meta-analyzed and Forest plots created using the Der-Simonian-Laird random effects-model [32].

Results

Genotype distribution

The genotype distribution of the patients included in the analysis is shown in Supplementary Table S3. Approximately 76% were common allele homozygotes, 22% were heterozygotes whereas just below 2% were minor allele homozygotes. This corresponds to a MAF of 12.9%. These results are consistent with those reported in other studies and the genotype distribution did not deviate from Hardy Weinberg equilibrium using a significance threshold of 0.01.

Analysis based on Z STAT scores and Z scores

Fig. 1 shows the mean Z STAT and Z scores as per genotype for the eight toxicity endpoints. Generally, patients with the variant allele had higher toxicity scores. The results were mostly consistent with a co-dominant pattern of inheritance. Furthermore, the strongest associations were observed for acute endpoints. Using the Mann–Whitney U test (Wilcoxon rank-sum test) as well as the Spearman rank correlation test, significant results were obtained for Z STAT global, Z STAT acute, Z STAT late, Z acute skin and Z acute rectal. For Z telangiectasia and Z fibrosis, the results did not reach statistical significance with p-values around 0.1. However, the results had the same direction as those reaching statistical significance. For late rectal toxicity, the analysis did not provide any indication that the SNP affects the normal tissue complication risk.

Fig. 1.

Mean Z STAT and Z scores as per genotype for the eight normal tissue endpoints. Error bars indicate standard error of the mean (SEM).

Analysis of odds ratios for having a score in the upper quartile

Fig. 2 shows the Forest and funnel plots for the meta-analysis of the odds ratios for having a Z STAT global score in the upper quartile comparing patients with one or two minor alleles with the common allele homozygotes. Supplementary Fig. S2 shows the Forest plots and funnel plots for the other endpoints. Table 2 provides a summary of the meta-analysis odds ratios for the 8 endpoints. In this analysis, significant associations were found for five of these endpoints but not for Z STAT late, Z acute rectal and Z late rectal. Generally, the associations were stronger for acute endpoints (with odds ratios around 1.5) than for the late endpoints (with odds ratios around 1.2). The I-squared test did not provide indications of significant heterogeneity between the cohorts for any of the endpoints. The funnel plots and Peters test for ‘small study effects’ did not provide indications of publication bias (Fig. 2 and Supplementary Fig. S2).

Fig. 2.

Forest plot and funnel plot showing the meta-analysis result for having a Z STAT global score in the upper quartile (Asp/Asp and Asp/Asn vs. Asn/Asn).

Table 2.

Overview of the meta-analysis odds ratios for having a Z or Z STAT score or a residual of these scores in the upper quartile (Asp/Asp and Asp/Asn vs. Asn/Asn). Significant findings are marked in bold.

Endpoint	N	Meta-analysis OR	95% CI	Endpoint	N	Meta-analysis OR	95% CI
Z STAT global	5456	1.20	1.04–1.38	R-STAT global	3339	1.24	1.03–1.48
Z STAT acute	2362	1.49	1.17–1.88	R-STAT acute	2300	1.51	1.21–1.88
Z acute skin	1357	1.71	1.11–2.66	R-Z acute skin	1304	1.60	1.06–2.41
Z acute rectal	1005	1.38	0.93–2.03	R-Z acute rectal	996	1.25	0.70–2.24
Z STAT late	4880	1.16	0.99–1.34	R-STAT late	2773	1.18	0.97–1.45
Z telangiectasia	2004	1.31	1.05–1.65	R-Z telangiectasia	1593	1.23	0.82–1.84
Z fibrosis	2457	1.27	1.02–1.58	R-Z fibrosis	1602	1.18	0.86–1.62
Z late rectal	2215	1.12	0.77–1.64	R-Z late rectal	1225	1.24	0.90–1.72

Multivariate analysis, calculation of residuals and analysis of odds ratios for having a residual in the upper quartile

Fig. 3 shows the Forest and funnel plots for R-STAT global. Supplementary Fig. S3 shows the Forest plot and funnel plots for the other endpoints. Table 2 provides an overview of the meta-analysis results. The adjustment for the included covariates only slightly altered the results compared to the unadjusted scores. Only the associations for R-STAT global, R-STAT acute and R-Z acute skin reached statistical significance.

Fig. 3.

Forest plot and funnel plot showing the meta-analysis result for having an R-STAT global score in the upper quartile (Asp/Asp and Asp/Asn vs. Asn/Asn).

Discussion

Methodological aspects

The primary strength of the present study is its very large sample size. Furthermore, the data were analyzed at the level of the individual patient and adjustments were made for treatment related factors with potential impact on toxicity risk. With more than 5000 patients included in the analysis and a total number of toxicity recordings exceeding 30,000, the study is by far the largest of its kind. Nevertheless, this came at the expense of the cohorts included in the analysis being rather heterogeneous. Particularly for the STAT scores, very diverse data were pooled for the analysis. In order to allow for a comparison across different endpoints, scoring systems, treatment regimens and institutions, we converted the toxicity data into Z-scores. This method is generally accepted as a pragmatic solution to the problems that relate to heterogeneous study cohorts (31) and has often been used in collaborative radiogenomics studies (e.g. [6,13,14] and [30]). Furthermore, the method mitigates the so-called center effect [36] wherein the institution at which the patient was treated may act as a confounder. Concerning the multivariate analysis, it was reassuring that the covariates included in the multivariate linear regression model were often factors known to affect the risk of normal tissue toxicity (such as body mass index/breast size, radiation dose and fractionation). Nevertheless, the adjustments made for these covariates only altered the results marginally. Since recordings of treatment and patient related covariates were only available for some of the cohorts, the number of patients included in the multivariate analysis was somewhat reduced. Presumably due to this, fewer of the findings reached statistical significance. The present meta-analysis included published as well as unpublished patient series. The funnel plots showed no obvious signs of publication bias, and the Peters test for ‘small study effects’ was negative for all the meta-analyses conducted. It is therefore unlikely that the results are significantly affected by publication bias. One base downstream of the rs1801516 a SNP designated rs1801673 (c.5558A>T, p.Asp1853Val) is located. This SNP is relatively rare with a minor allele frequency below 1% and was not taken into account in our meta-analysis. Our meta-analysis involved multiple comparisons. Despite this, we did not adjust for multiple testing. In most instances, the tests were not statistically independent and a standard adjustment for multiple comparisons would therefore be overly conservative. Furthermore, there was a high degree of consistency of the results across endpoints that were not scored in the same patients. Consequently, we consider it highly unlikely that the positive findings of our meta-analysis could be the result of random fluctuations due to multiple testing.

Interpretation of the results

The present meta-analysis shows, with a high level of confidence, that the ATM rs1801516 SNP is associated with risk of normal tissue toxicity after radiosensitivity. Furthermore, the analysis indicates that the SNP has a stronger association with acute than late toxicity. The results are indicative of a co-dominant pattern of inheritance. With an odds ratio around 1.3, the phenotypic impact of this SNP is fairly small. This observation is in agreement with the results of many genome wide association studies (GWAS) addressing various biomedical phenotypes in which most of the identified SNPs only had a modest impact on phenotype [1,37,38]. The effect size of the rs1801516 SNP demonstrated by our meta-analysis is below the detectable effect size in the radio-genomics GWASs published so far (2–8). Furthermore, the small effect size explains why the candidate gene studies conducted did not report consistent findings for this SNP [19–21]. As mentioned in the introduction, three literature based meta-analyses have addressed the impact of the rs1801516 SNP upon risk of normal tissue toxicity after radiotherapy. One of these included 2127 patients from 12 cohorts and addressed normal tissue toxicity in a broader sense [19]. The patients were treated for breast, prostate and head and neck cancer and investigated a mixture of different acute and late normal tissue reactions. This study did not show any significant association between the rs1801516 SNP and normal tissue outcome. Nevertheless, with a per-allele odds ratio of 1.11 (95% CI 0.77–1.59) for the Asn allele, our value is within the 95% confidence interval of the meta-analysis. Another meta-analysis included 1588 patients from 5 different cohorts and addressed acute reactions after radiotherapy for prostate and breast cancer [20]. Consistent with our findings, this study reported an odds ratio of 1.33 (95%CI 1.04–1.77) for carriers of the minor (Asn) allele. The most recent meta-analysis included 2000 patients from 9 different cohorts [21]. The patients were given radiotherapy for breast cancer, prostate cancer, head and neck cancer and brain tumours. This study addressed radiation-induced fibrosis. In agreement with our analysis, it showed a significantly increased risk of fibrosis (OR = 1.78, 95% CI: 1.07, 2.94) among carriers of the Asn allele. It should, however, be noted that all these meta-analyses had a number of cohorts in common with our study.

The rs1801516 SNP is located in a highly conserved element in the ATM gene (LOD score 250) (Supplementary Text File 2). It results in a non-conservative amino acid substitution from the acidic aspartic acid to the polar asparagine at codon 1853 in exon 37 (NM_000051.3). Thus, the SNP may potentially have a functional impact. Nevertheless, the SNP is classified as ‘benign’ by several different missense prediction algorithms (Align GVGD, SIFT, NCBI ClinVar and PROVEAN) (Supplementary Text File 2). Likewise, no change in natural splice sites is predicted by several splicing prediction tools (Splice Sequence Finder, MaxEntScan, NNSplice, and Human Splicing Finder). A cryptic and alternative acceptor splice site, located at the nucleotide 5558, is predicted to be removed by the rs1801516 SNP minor allele (Supplementary Text File 2). However, the consequence of this on alternative splicing is unknown. To our knowledge, only one study has investigated the functional impact of the rs1801516 SNP [39]. In this study, no significant differences were observed in the constitutive ATM protein level, cell survival or TP53 protein induction after radiation in vitro of lymphoblastoid cell lines from patients with each of the three possible rs1801516 SNP genotypes. Nevertheless, these investigations only included ‘a limited number of patients’ [39] and were therefore probably not powered to detect small differences in these parameters. The rs1801516 SNP is in linkage disequilibrium with numerous other SNPs [40] (Supplementary Text File 2). Consequently, it cannot be ruled out that the SNP acts as a tagging SNP for another sequence variant with functional impact. Heterozygosity for a truncating mutation in ATM is a well-known risk factor for breast cancer [38]. Given that the rs1801516 SNP has a functional impact (or tags a variant with functional impact), one could expect this SNP to affect breast cancer risk. However, even though breast cancer susceptibility has been investigated in several very large GWASs that were powered to detect small effect sizes, the rs1801516 SNP was not identified as a risk factor [38]. Nor did a large meta-analysis (N = 7971) provide evidence that the rs1801516 affects breast cancer risk [40]. A search in the NHGRI-EBI GWAS database (http://www.ebi.ac.uk/gwas/) identifies only one hit for the rs1801516 SNP. This involves a large multi-center GWAS with several replication cohorts addressing risk of malignant melanoma [41]. This study included a total of 10,422 subjects and identified an association for the rs1801516 SNP that reached genome wide significance with a p-value of 3.4 × 10⁻⁹. Nevertheless, in this study the minor allele was protective with an odds ratio of 0.84. A more recent meta-analysis of data from 11 GWASs addressing malignant melanoma with a total of 15,990 cases and 26,409 controls confirmed the association with a p value of 6.6 × 10⁻¹² [42]. As mentioned previously, the gene product of ATM is involved in the detection of DNA double strand breaks and initiation of pathways that leads to cell cycle arrest followed by DNA repair or apoptosis [15]. Thus, ATM plays a crucial role in the biological response to ionizing radiation and influences the fate of the cell following irradiation [16]. Acute radiation toxicity typically takes place in skin or mucosal membranes and its pathogenesis is assumed to involve depletion of rapidly dividing cells in the basal layer [12]. In our meta-analysis, the rs1801516 SNP had a stronger impact on acute than late toxicity. Based on these observations, it can be hypothesized that the Asn allele in codon 1853 (or another variant tagged by this allele) may result in an increased tendency to cell cycle arrest or apoptosis. This would possibly lead to increased acute toxicity but at the same time be protective against radiation-induced malignancy. Our meta-analysis shows that it is possible to extract useful information from study cohorts that are very heterogeneous. However, it does at the same time illustrate the methods and large sample sizes required to detect a relatively weak signal in a noisy dataset. Thus, our study warrants further cooperative research initiatives in radiogenomics.

Conclusion

This very large individual patient data meta-analysis conducted by the RgC provides convincing evidence of an association between the Asn allele in the ATM rs1801516 SNP and an increased risk of normal tissue toxicity after radiotherapy with and odds ratio of approximately 1.5 for acute toxicity and 1.2 for late toxicity. The findings of this study warrant research into the functional impact of this SNP (or another sequence variant tagged by this SNP). In addition, further large studies quantifying the effect size of this SNP in the setting of a less heterogeneous study population should be encouraged.