A genome-wide association study on radiation induced fibrosis in breast cancer patients

Graphical abstract

Highlights

• •A genome-wide-association-analysis testing genetic variants and radiotherapy-induced fibrosis in 869 breast cancer patients.
• •A locus on chromosome 10 reached a P-value = 5.90 × 10^-8. The strongest association was represented by the SNP rs75542274.
• •Meta-analysis of summary data with one or more independent and comparable breast cancer cohort(s) is warranted.

Abstract

Background and purpose

Radiation-induced fibrosis affects women having undergone post-lumpectomy radiotherapy to a varying degree and remains a significant late morbidity for breast cancer patients. While Single Nucleotide Polymorphisms (SNPs) have been linked to fibrosis after radiotherapy (RT), the genetic architecture remains incompletely understood. We aimed to identify new genetic variants in a cohort of early breast cancer patients representing two cohorts treated within the Danish Breast Cancer Group (DBCG) protocols.

Material and methods

A genome-wide association study (GWAS) was conducted on 869 patients treated with lumpectomy and adjuvant radiotherapy within the DBCG trials hypo- versus normofractionated radiotherapy trial, DBCG-HYPO, and the DBCG partial versus whole breast irradiation trial, DBCG-PBI. After genotyping and imputation, we tested associations between common variants and grade 2–3 fibrosis (LENT-SOMA) using a per-risk allele log-additive model. The threshold for genome-wide significance was set at P < 5 × 10⁻⁸.

Results

After adjusting for principal component outliers, we identified a suggestive association on chromosome 10 (P = 5.90 × 10⁻⁸). The lead variant was rs75542274. While this locus approached the pre-defined genome-wide significance threshold, no other variants reached significance.

Conclusion

This exploratory GWAS identified a potential susceptibility locus for radiation-induced fibrosis on chromosome 10. Given the near-significant nature of this finding, independent validation or meta-analysis is required to confirm the role of rs75542274 in the development of RT-induced fibrosis. Further analysis is warranted.

Introduction

Fibrosis induced by radiotherapy (RT) is a dose-limiting late morbidity across cancer sites. In early breast cancer, the risk of subcutaneous fibrosis is associated with several different treatment and patient related factors. Treatment related factors include RT-associated parameters like dose, volume, fractionation and boost as well as the use of chemotherapy have also been reported. Patient related factors include parameters like breast volume, age, post-operative oedema and smoking [1], [2], [3], [4]. A study investigated covariates predictive of RT-induced fibrosis in the two cohorts included in this GWAS study and identified the three following factors: Smoking, RT-Dose in young boost patients below 50 years of age and irradiated volume of the breast for patients at or above 50 years of age [5], [6].

Normal tissue complication probability (NTCP) models including the above-mentioned treatment and patient related factors can explain some of the variance in risk of radiation-induced fibrosis [1], [3], [4]. The field of radiogenomics research has emerged to identify additional patient related factors not previously accounted for in NTCP models, i.e. genetic variants associated with radiation-induced morbidity [7].

The need to create data for associations between genetic variants and RT outcome – both in terms of cancer outcome and treatment-induced morbidity − has been pointed out by the research communities in radiation oncology as an additional way to personalize and improve clinical decision making in cancer treatment [8], [9]. Appropriate steps have been taken to provide guidelines for study design and interpretation of findings to facilitate and streamline the way for future findings to enter patient care in a clinically meaningful way [10], [11].

In particular, the genetic variations of single nucleotide polymorphisms (SNPs) have been studied so far [12]. In 2012, Barnett et al. performed a validation study on 92 previously published SNPs in a large, independent cohort of 1613 patients [13]. Since then, several additional SNPs have been identified, as reviewed by Herskind et al. [14]. Later, a single additional SNP has been validated through a large meta-analysis by the Radiogenomics Consortium (RgC) [15]. In recent years, one meta-analysis found common genetic factors across different cancer types linked to radiation-induced acute toxicities [16], another meta-analysis confirmed previous findings between genetic markers and the development of acute and late skin toxicity after radiation in breast cancer patients [17]. The meta-analysis identified variants in genes ATM, XRCC2, IFNG, TGFB1 and PER3 to be associated with increased risk of acute and late skin side effects after RT. However, this study also stated that findings should be interpreted cautiously due to heterogeneity and possible publication bias [17]. The research field of radiogenomics requires large sample sizes and well-conducted trials with prospective scoring of endpoints to obtain the necessary power to identify genetic variants associated with RT-related endpoints as well as it requires cohorts for validation of exploratory findings. This GWAS contributes a new cohort with solid clinical data to the field of radiogenomic research.

In the present study, we undertook a genome-wide association study on a cohort consisting of 869 patients with well-characterized treatment data, patient characteristics, and prospective follow-up data from the Danish Breast Cancer Group (DBCG). The cohorts included represent two Randomised Clinical Trials (RCTs) i.e. the DBCG hypo- versus normofractionated radiotherapy trial, DBCG-HYPO [18] and the DBCG partial (PBI) versus whole breast irradiation (WBI) trial, DBCG-PBI [19].

Materials and methods

Cohorts

DBCG-HYPO: In a clinically controlled, randomized trial of hypo fractionated versus standard fractionated RT of early lymph-node negative breast cancer, patients were treated with curative intent in an adjuvant setting after breast conserving surgery with either 50 Gy in 25 fractions or 40 Gy in 15 fractions [18], [20]. Patients were enrolled between 2009 and 2014, and follow-up data was collected yearly until 5 years after RT. The primary endpoint was grade ≥ 2 breast induration at 3 years after RT. Patients were eligible by the following criteria: Invasive early breast cancer or DCIS, age ≥ 41 years, pTis-pT2, pN0-pN1(mic), any histology/ER/HER2, grade I, II or III, boost allowed, any breast size, any systemic therapy. The trial was organized through the Danish Breast Cancer Group (DBCG). It comprised 1549 patients, of these 661 patients had available biological material and were enrolled in the present study.

DBCG-PBI: In a clinically controlled, randomized trial, low risk breast cancer patients were treated with 40 Gy in 15 fractions and randomized for WBI or PBI after radical breast conserving surgery. None of the patients received lymph node irradiation. Patients were eligible by the following criteria: Unifocal, unilateral, non-lobular breast cancer, age ≥ 60 years, pT1, pN0, grade I/II, ER+, HER2-. The primary endpoint was grade ≥ 2 breast induration at 3 years after RT [19]. Patients were enrolled between 2009 and 2015, and follow-up data was collected yearly up until 5 years after RT. The trial was organized through DBCG. It comprised 880 patients, of these 230 patients had available biological material and were enrolled in the present study.

Table 1a, Table 1b summarise the patient and treatment characteristics of the cohorts. For both cohorts, clinical data was available from the DBCG database [21] and biological material was available from the National Danish Cancer Biobank.

Table 1a.

Characteristics of the DBCG-HYPO + DBCG-PBI RCTs and the subset of patients included in the present GWAS.

	DBCG-HYPO (n = 644)		DBCG-PBI (n = 225)
	n/median	(%/range)	n/median	(%/range)
Randomisation arms	All WBI: 50 Gy/25 fx/5w vs. 40 Gy/15 fx/5w		All 40 Gy/15 fx/5w: PBI vs. WBI
Primary Endpoint	3-year Grade 2–3 Breast Induration		3-year Grade 2–3 Breast Induration
RT-induced fibrosis rate	9.0% in 40 Gy arm vs. 11.8% in 50 Gy arm		5.1% in PBI arm vs. 9.7% in WBI arm
Age (years) Median (range)	59	(41–83)	66	(60–86)
RT-boost n (%) None 10 Gy 16 Gy	558 61 25	(87) (9) (4)	225	(100)
Systemic adjuvant treatment	Chemotherapy, Trastuzumab, Endocrine in accordance with DBCG guidelines		Endocrine in accordance with DBCG guidelines
Breast size limit	Any breast size allowed		Any breast size allowed

RCT: Randomised Clinical trial. WBI: Whole Breast Irradiation. PBI: Partial Breast Irradiation. Gy: Gray. Fx: Fractions of RT. W: Weekly. RT: Radiotherapy. DBCG: The Danish Breast Cancer Group.

Table 1b.

Patient and treatment characteristics in the joint DBCG-HYPO + DBCG-PBI GWAS cohort.

	DBCG-HYPO + PBI (n = 869)
	n/median	(%/range/IQR)
Age (years)
Median (range)	62	(41–86)
Follow-up time (years)
Median (range)	5.0	(2.5–5.5)
Induration
Grade 0–1	774	(89)
Grade 2–3	95	(11)
Dose (Gy/fx) / Field (Protocol)
50/25 / Whole breast (HYPO)	320	(37)
40/15 / Whole breast (HYPO)	324	(37)
40/15 / Whole breast (PBI)	117	(13)
40/15 / Partial breast (PBI)	108	(12)
Breast volume (ml)
Median (inter quartile range, IQR)	682	(427–978)
Irradiated breast volume (ml)
Median (IQR) − Whole breast	412	(259–570)
Median (IQR) − Partial breast	163	(91–231)
Boost
No boost	783	(90)
10 Gy	61	(7)
16 Gy	25	(3)
Adjuvant treatment
No	409	(47)
Chemotherapy	189	(22)
Antihormone	219	(25)
Chemotherapy + antihormone	52	(6)
Smoking
Never/Prior	676	(78)
Current	193	(22)

Endpoints

The DBCG-HYPO and DBCG-PBI protocols specified “induration” as the primary endpoint. The term “fibrosis” has been widely used historically, but a growing awareness of the distinction between fibrosis as a histopathological diagnosis and induration as a clinical measure has led to the increased specification of terms. We chose to respect the original terminology in the publications we refer to in this paper and therefore use “fibrosis” and “induration” synonymously as an endpoint representing a clinical assessment.

Induration/fibrosis was scored by the LENT-SOMA scale for post radiation fibrosis (0 = none, 1 = barely palpable/increased density, 2 = definite increased density and firmness, 3 = marked density, retraction and fixation). This score was dichotomized in “no fibrosis”: grade 0–1 and “grade ≥ 2 fibrosis”: grade 2–3. Evaluation of morbidity was performed by a limited number of specially trained nurses who also participated in yearly workshops examining and discussing participating patients with radiation-induced morbidity. An event was defined as the first time the patient had breast induration grade 2–3 only if the patient also had grade 2–3 at one or more visits afterwards or if grade 2–3 was recorded at the last visit. Induration data at 3 years were used in accordance with the primary endpoint of the two Randomized Clinical Trials [18], [19].

Covariates

In previous studies, a strong correlation between the specific volume of breast irradiated to 40 Gy, boost, age and smoking was strongly correlated to fibrosis. This was identified in the DBCG-PBI cohort and validated in the DBCG-HYPO cohort [5], [6]. Based on these results a logistic regression model for predicting fibrosis was developed based on these four parameters: a constant, an interaction term between age and irradiated volume (without boost), an interaction term between age and irradiated volume (with boost), and smoking (for details, see Supplementary Document 1). Age was included as three groups (40–49, 50–64, 65+), smoking as never/prior vs current smokers, and the irradiated breast volume with no boost was defined as the volume of CTVp_breast irradiated to RT dose (D) ≥ 40 Gy or D ≥ 50 Gy in the 40 Gy and 50 Gy arms (V_40Gy/50Gy), respectively. For boost patients, irradiated breast volume was defined as the volume of the breast receiving the prescribed dose plus 5 Gy. Thus, the irradiated CTVp_breast volume was V_45Gy/55Gy. Patient and treatment characteristics are available in Table 1a + b.

Genotyping, imputation, quality control and bioinformatics

Normal tissue (germ line) whole-genome DNA was extracted from buffy coats and SNP genotyping was done using the Infinium OncoArray-500 K Bead Chip (Illumina Inc., San Diego, USA). Pre-imputation quality control adhered to OncoArray guidelines [22] and is described in detail in Supplementary Document 2. In short, Whole-genome DNA was extracted from buffy coats, quantified and normalized. Initially, samples and variants with call rates < 80% were excluded. This was repeated for samples and SNPs with call rates < 95%. Samples with heterozygosity < 5% or > 40% and a heterozygosity P < 10^-6 were excluded. Variants with MAF < 0.01 were excluded. Gender errors were checked by using a set of 51 well performing chromosome Y SNPs and 10,361 chromosome X SNPs. Number of chromosome Y SNPs and chromosome X heterozygosity was compared and checked for concordance with clinical data for gender. A P-value of 10^-7 was set as a cut-off for Hardy-Weinberg equilibrium. Autosomal and chromosome X SNPs were checked separately. Duplicates and relatives were checked with a set of 7,822 autosomal SNPs known to perform well. In case of concordance > 0.80 as estimated by the identity by state function, the sample with the highest number of well genotyped SNPs was kept in the dataset.

The combined DBCG-HYPO and DBCG-PBI cohort was imputed simultaneously as a single dataset. Haplotype estimation (pre-phasing) of genotyped SNPs was done using SHAPEITv2 (r837) [23]. Imputation was done using Minimac3 [24] and TOPMED Version R2 [25] mapped to GRCh38 as reference.

After imputation, a set of filtered genotyped SNPs were used for relatedness pruning and removal of genetic outliers using an initial principal component analysis (PCA), see Supplementary Document 2 for details. A second PCA was performed on the relatedness pruned samples excluding outliers of the initial PCA, principal components 1–10 of the second PCA were used for adjusting for any remaining population substructure in downstream analyses.

Statistical analysis

DBCG-HYPO and DBCG-PBI cohorts were analysed together as uniform covariates were available and RT treatment regimens comparable.

Genome-wide associations between genotypes and fibrosis were tested using logistic regression adjusted for the above-mentioned covariates and principal component 1–10 of the second PCA using plink v2. [26]. Genotypes were included as number of effect alleles or imputed genotype dosage. A log-additive genetic model was assumed in all analyses, resulting in a per-allele beta ($\mathrm{\beta }$) and odds ratio (OR) for each genotype. A P-value < 5 × 10^-8 was considered genome-wide significant [27].

Data preparation and statistical analyses was carried out using Stata13 and Stata18 (StataCorp, TX USA) and R Statistical Package v 4.3.2 [28]. The study is reported in compliance with the STROGAR guidelines for radiogenomic studies [29].

Data availability

The full list of summary data is available in supplementary table 2 and will further be made available in the NHGRI-EBI GWAS Catalog [30] upon publication.

Ethics

The study was approved by the Data Protection Agency (j.no. 1–16-02–627-15/2012–58-006) and the Scientifical Ethics Committee (j.no. 1–10-72–212-15) in Central Denmark Region.

Results

Patient characteristics

The cohort initially included 1041 patients including duplicate controls. Nine hundred and ninety-three passed genotyping and imputation quality control and of these, 112 were excluded due to missing clinical data or withdrawal of consent as shown in Fig. 1. Eight hundred and eighty-one patients with follow-up data and well imputed genotypes were available for analysis and finally after QC, 869 patients remained in analysis. Patient and treatment characteristics are summarized in Table 1a, Table 1b.

Fig. 1.

Patient selection. The diagram in red represents the DBCG-HYPO cohort, the blue represents the DBCG-PBI cohort, and the green box represents the combined cohorts with available germ-line genotype data. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Genotypes

A total number of 533,631 SNPs were genotyped on the OncoArray, 10.7^x10⁶ were successfully imputed and passed quality control (Supplementary Document 2). The imputation quality was good with a r² mean score = 0.9381 (95% CI 0.9380–0.93813), minimum r² = 0.6, maximum 1.0, and the following quantiles: Q₁ = 0.9162, Q₂ = 0.9705 and Q₃ = 0.9882.

Violin- and boxplot of MiniMac3 imputation scores is available in Supplementary Fig. 5.

Number of samples and SNPs excluded in quality control process are listed in Supplementary Table 1.

GWAS

Manhattan plot and QQ-plot for the HYPO/PBI cohort are presented in Fig. 2, Fig. 3 respectively. The QQ plot and the genomic inflation factor $\left(\lambda =0.939\right)$ showed no indication of genomic inflation, thus suggesting that any potential population substructure, cryptic relatedness or technical confounding was properly accounted for. There was, however, signs of deflation $\left(\lambda <1.0\right)$, which is often caused by overcorrection or, more likely in the current study with a relatively low sample size of 869 patients, statistically underpowered tests. The genomic inflation factor standardized to an effective sample sizes of 1000 cases and 1000 controls $\left({\lambda }_{1000}=0.82\right)$, showed even stronger deflation, most likely caused by the relative low number of cases (N_case = 95) compared to controls (N_ctrls = 774).

Fig. 2.

Manhattan plot of GWAS results after relatedness pruning $(\widehat{\pi }<0.2)$ and removal of PCA outliers falling outside a four-dimensional ellipsoide 5x standard deviations from the mean. With adjustment for 11 covariates (including PC 1–10 from a second PCA). X-axis shows genomic position of SNPs and y-axis is −log₁₀ of the logistic regression P-value. The horizontal red line indicates a P-value of 5x10^-8 used as cut-off for genome-wide significance, while the horizontal grey line indicates a P-value of 7x10^-8. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3.

Quantile-quantile-plot (Q_Q plot) of −log₁₀ of the expected P-values under the null hypothesis of no association versus −log₁₀ of the observed P-values from the GWAS, with genomic inflation factor $\left(\lambda \right)$ and the genomic inflation factor rescaled to a sample size of 1000 $\left({\lambda }_{1000}\right)$ using the following formular: ${\lambda }_{1000}=1+\left(\lambda -1\right)\bullet \left(1000/N_e\right)$, where $N_e$ is the effective sample size: $N_e=4/\left(1/N_{\mathit{cases}}+1/N_{\mathit{ctrls}}\right)$. N_P is the number of plottet P-values. Red diagonal line indicates $-{\mathit{log}}_{1o}\left(P_{\mathit{expected}}\right)=-{\mathit{log}}_{1o}\left(P_{\mathit{observed}}\right)$, dashed grey lines indicated 95% confidence interval of expected P-values. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The strongest association was OR = 5.14 (95% CI 2.85–9.30) P = 5.9^x10^-8, for the minor allele C in rs75542274, an imputed SNP (imputation score of r² = 0.96) located on chromosome 10 in an intergenic region. The over all MAF of allele C in rs75542274 was 0.0424741 and distributed in cases: MAF(C)_cases = 0.119967 and in controls: MAF(C)_controls = 0.0329627.

Six other variants in the same linkage disequilibrium (LD)-block reached P-values below 10^-4 and 401 variants reached P-values below 10^-3. The top 10 entries in different LD-blocks are listed in Table 2. The P-values of SNPs 500 kb down- and upstream of rs75542274 on chromosome 10 are shown in Supplementary Fig. 4. A Locus Zoom plot of SNPs 500 kb down- and upstream of index SNP rs75542274 on chromosome 10 is available in Supplementary Fig. 6.

Table 2.

Top ten lead SNPs and their imputation score (r²). Number of variants within LD-blocks for different P-intervals. With adjustment for 11 covariates: Risk factor, PC1, PC2, PC3, PC4, PC5, PC6, PC7, PC8, PC9 and PC10.

lead SNP	r2	Region	P	all	>0.05	0.01–0.05	0.001–0.01	0.0001–0.001	<0.0001	Genes within LD-block
rs75542274	0.96	chr10:109292242..109459050	5.90E-08	411	0	0	4	401	6	−
rs143983742	0.88	chr1:159802501..159957347	1.86E-07	5	0	0	0	2	3	[C1orf204, CCDC19, CRP, DUSP23, FCRL6, IGSF9, KCNJ10, LINC01133, PIGM, SLAMF8, SLAMF9, TAGLN2, VSIG8]
rs142016012	0.96	chr3:184063892..184375716	2.52E-07	7	0	0	1	2	4	[ABCC5, ABCC5-AS1, ABCF3, ALG3, AP2M1, CAMK2N2, CHRD, CLCN2, DVL3, ECE2, EIF2B5, EIF4G1, FAM131A, HTR3C, HTR3D, HTR3E, MIR1224, POLR2H, PSMD2, SNORD66, THPO, VWA5B2]
rs62455836	0.96	chr7:70582211..70602042	4.28E-07	10	0	1	1	1	7	[AUTS2]
rs117376869	0.92	chr12:129256221..129256221	1.193E-06	2	1	1	0	0	0	[LOC101927735, TMEM132D]
rs12454355	0.77	chr18:9770328..9807324	1.215E-06	8	1	0	1	3	3	[RAB31, TXNDC2]
rs114150329	0.84	chr3:124255956..124378620	1.285E-06	9	0	2	4	0	3	[KALRN, MIR6083]
rs71329756	0.99	chr20:57872903..57872903	1.642E-06	4	1	2	1	0	0	[MIR4532]
rs72832716	0.76	chr10:122370570..122470881	1.652E-06	64	1	1	1	15	46	[ARMS2, BTBD16, DMBT1, HTRA1, MIR3941, PLEKHA1]
rs4840481	0.98	chr8:10387585..10411909	2.547E-06	26	0	1	13	7	5	[MSRA]

Supplementary Table 2 contains summary data for the full lists of SNPs in the GWAS.

The summary statistics for SNP and covariates in the logistic regression analysis for lead SNP rs75542274 are listed in Table 3 further showing a strong correlation between the non-genetic covariate risk factor and the end point of RT-induced fibrosis reaching a P-value = 7.98^x10^-11.

Table 3.

Summary statistics in logistic regression for SNP and covariates for lead SNP rs75542274.

TEST	N	OR	L95	U95	Z_STAT	P
rs75542274	869	5.14	2.85	9.3	5.42	5.9E-08
Covariate risk factor	869	1071.23	130.73	8777.73	6.5	8.0E-11
PC1	869	0	0	9.4E + 13	−0.87	0.39
PC2	869	8.1E + 05	0.32	2.0E + 12	1.81	0.07
PC3	869	0.01	0	1079.75	−0.81	0.42
PC4	869	4214.14	0.27	6.5E + 07	1.7	0.09
PC5	869	0.05	0	50.88	−0.84	0.4
PC6	869	52.72	0.05	57830.18	1.11	0.27
PC7	869	53.65	0.06	49501.45	1.14	0.25
PC8	869	0.01	0	10.14	−1.3	0.19
PC9	869	0.01	0	5.75	−1.47	0.14
PC10	869	19.01	0.02	19177.81	0.83	0.4

Discussion

This study represents an original cohort with no overlap of patients from any previous GWAS. We aimed to perform an exploratory study of genome-wide associations between SNPs and radiation-induced fibrosis in breast cancer patients in two DBCG cohorts with a total of 882 patients.

After removal of 13 patients, 869 patients remained in analysis and by relatedness-pruning and removal of PCA outliers, a locus reaching near-significance in an intergenic position on chromosome 10 was identified in 95 cases and 774 controls. The index SNP (rs75542274) of the locus is a known SNP with a minor allele frequency of 0.032 in both 1000 Genomes Project Phase 3 and genomAD genomes v4.1 and without any known effect on phenotype. This SNP may, however, be in linkage disequilibrium with other inter- or intragenic variants of functional importance. Intergenic regions can be involved in regulation of specific gene functions, or variants in intergenic regions may be in LD with functional variants in protein coding exons. The identified region is surrounded by pseudogenes and the region thus without any known function (see Supplementary Fig. 6). As with many single-cohort GWAS the current study is underpowered, and genome-wide significant variants (if any) correspond to only a small fraction of the truly associated variants. Regions with multiple non-significant but low P-value variants in LD, may hint at genome-wide significant regions of future large-scale meta-analysis of multiple cohorts.

One may speculate that a genome with 3 billion base pairs and genes expressed variously in trillions of cells requires analysis more sophisticated analysis than logistic regression models and dichotomization of endpoints that clinically appear in a continuum. Machine-learning prediction modelling in RT-induced morbidity have been published [31], [32]. We still found that the approach taken in this study was justified, based on considerations of feasibility and a consensus on how to conduct these studies until now, rendering comparison, validation and meta-analysis more available. In addition, and to the best of our knowledge, there is a yet emerging field on how to apply, interpret and validate different AI- or machine-learning models in genomic analyses.

The study did not identify variants reaching the threshold for genome-wide significance which is likely explained by the number of samples available. Power in the present GWAS was limited by the cohort size which rendered identification of low-penetrance variants unlikely. This is additionally for a large part due to a significance level reflecting a strict correction for multiple testing, i.e. Bonferroni correction for 1 million independent SNPs in the genome, thus accounting for linkage disequilibrium.

Previous studies have identified several loci associated with RT-induced morbidity. A GWAS on RT-induced morbidity in breast cancer patients identified several loci [33]. However, only few have been validated and yet none have entered clinically useful tools [34]. High-quality cohorts in terms of both clinical and genotype data are a natural prerequisite for the development of any such tool. The strength of this study very much lies in the quality of the prospectively collected clinical data and the uniform RT dose planning causing a low amount of heterogeneity in endpoint scoring and factors related to RT (e.g. fields, hot spots, low-dose volume).

Collaborative studies of combining or meta-analysing cohorts with comparable endpoints and genotype data is an evident way to meet the need for well-powered analyses.

In conclusion

The exploratory GWA study identified a locus in an intergenic region on Chromosome 10 that was close to genome-wide significant association with risk of radiation-induced fibrosis. This locus cannot yet be considered a true positive finding but should be tested in comparable cohorts of breast cancer patients. Meta-analysis of summary data with one or more independent and comparable breast cancer cohort(s) is warranted.

Ethics declarations & trial registry information

The study was approved by the Data Protection Agency (j.no. 1-16-02-627-15/2012-58-006) and the Scientifical Ethics Committee (j.no. 1-10-72-212-15) in Central Denmark Region.

Web resources

dbSNP, https://www.ncbi.nlm.nih.gov/snp/docs/about/

ENSEMBL, https://www.ensembl.org/index.html.

ENSEMBL index SNP rs75542274,

https://www.ensembl.org/Homo_sapiens/Variation/Explore?r = 10:109458550-109459550;v = rs75542274;vdb = variation;vf = 655119188.

Michigan TOPMED Imputation server, https://imputation.biodatacatalyst.nhlbi.nih.gov/#

National Institute of Health LD analysis and plot, https://analysistools.nci.nih.gov/LDlink/

PLINK, https://www.cog-genomics.org/plink2.

R Software, https://www.r-project.org/

SHAPEIT, https://mathgen.stats.ox.ac.uk/genetics_software/shapeit/shapeit.html.

SNPedia, https://www.snpedia.com/index.php/SNPedia.

Trans-Omics for Precision Medicine | TOPMed, https://topmed.nhlbi.nih.gov/

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: