DNA methylation at quantitative trait loci (mQTLs) varies with cell type and nonheritable factors and may improve breast cancer risk assessment

Abstract

To individualise breast cancer (BC) prevention, markers to follow a person's changing environment and health extending beyond static genetic risk scores are required. Here, we analysed cervical and breast DNA methylation (n = 1848) and single nucleotide polymorphisms (n = 1442) and demonstrate that a linear combination of methylation levels at 104 BC-associated methylation quantitative trait loci (mQTL) CpGs, termed the WID™-qtBC index, can identify women with breast cancer in hormone-sensitive tissues (AUC = 0.71 [95% CI: 0.65-0.77] in cervical samples). Women in the highest combined risk group (high polygenic risk score and WID™-qtBC) had a 9.6-fold increased risk for BC [95% CI: 4.7-21] compared to the low-risk group and tended to present at more advanced stages. Importantly, the WID™-qtBC is influenced by non-genetic BC risk factors, including age and body mass index, and can be modified by a preventive pharmacological intervention, indicating an interaction between genome and environment recorded at the level of the epigenome. Our findings indicate that methylation levels at mQTLs in relevant surrogate tissues could enable integration of heritable and non-heritable factors for improved disease risk stratification.

Fig. 1. Identification of the most informative surrogate sample for breast-variable DNA methylation indicates that cervical samples exhibit higher variability in the top-variable breast CpGs compared to matched buccal and blood samples.

a Higher variability indicates the potential presence of more information, opposed to CpGs which are homogeneously methylated or unmethylated across samples. We explored this by identifying variability of CpGs in tissue at risk of breast cancer ( = breast tissue) and assessing variability of these top breast-variable CpGs in three non-invasive surrogate tissues, utilising matched buccal, blood, and cervical samples from the same individuals (n = 222 per tissue). b Standard deviation of the top variable breast CpGs (1, 2, 5, 10, 15, and 20 percentiles, respectively) in matched cervical, buccal, and blood samples (all and separated by inferred immune cell composition). c Variability of all CpGs versus mQTL CpGs in the three matched tissues (all or separated by inferred immune cell composition). The dashed line shows median variability of all CpGs in all cervical samples while the dotted line shows median variability of mQTLs in cervical samples with an immune cell composition (ic) < 25%. Boxplot boxes indicate median (centre line), interquartile range (bounds of box), and 95% confidence interval (whiskers). DNAme DNA methylation, mQTL methylation quantitative trait locus, SNP single nucleotide polymorphism, ic immune cell (proportion).

Fig. 2. The WID™-qtBC index distinguishes breast cancer cases and controls in cervical and breast samples.

a Outline of classifier development and validation strategy. b The WID™-qtBC index is increased in cervical samples from current breast cancer cases compared to controls in the validation set (p = 1.2e-07 in two-tailed Student’s t-test). c The WID™-qtBC has an AUC of 0.71 and is unaffected by estrogen receptor (ER) status. d Permutation analysis of index training featuring randomly selected CpGs indicates that the WID™-qtBC AUC in the validation set is higher than expected by chance. Shaded area indicates 95% confidence interval from permutation testing. e The WID™-qtBC index is increased in normal tissue adjacent to breast cancer compared to normal breast tissue. f ROC curve for discrimination of normal to normal-adjacent tissue based on the WID™-qtBC index. Boxplot boxes indicate median (centre line), interquartile range (bounds of box), and 95% confidence interval (whiskers). AUC area under the receiver operating characteristic curve, ER- estrogen receptor negative breast cancer, ER+ estrogen receptor positive breast cancer.

Fig. 3. Association of the WID^TM-qtBC with the 313 SNP polygenic risk score and cancer characteristics.

a Pearson correlation of the WID^TM-qtBC index with the polygenic risk score (PRS₃₁₃) in the Validation set. b ROC curve analysis of the WID^TM-qtBC index stratified by median PRS group (higher or lower than median) in the Validation set. c Odds ratio of the PRS₃₁₃, WID^TM-qtBC, or their combination, across different percentile categories for the risk scores. Shading indicates 95% confidence intervals. d Assessment of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2 status), nodal and tumour stage, and tumour grade amongst cases in the Validation set, comparing different risk groups defined in Table 1. Numbers in bars indicate n for each group. p = 0.0349 for an association of T2-4 tumours with increasing PRS₃₁₃ and WID™-qtBC, p = 0.0167 for increasing Grade III tumours with increasing PRS₃₁₃ and WID™-qtBC. P values were derived from logistic regression model using ER status, PR status, HER2 status, nodal stage, tumour stage, and tumour grade as independent variables and risk group as the dependent variable. ER status was not significant. Abbreviations: PRS₃₁₃ polygenic risk score (313 SNPs). AUC area under the receiver operating characteristic, ER estrogen receptor, PR progesterone receptor.

Fig. 4. Dynamic changes of the WID^TM-qtBC and association with non-genetic characteristics.

a Matched breast biopsy samples in healthy women before and after two months of vitamin or mifepristone treatment. 7/9 (77.8%) women in the mifepristone group showed a reduction in the WID^TM-qtBC while only 3/11 (27.3%) showed a reduction in the vitamin group. Association of the WID™-qtBC scores in cervical samples in the validation set with (b) age, (c) body mass index, (d) age at menarche, and (e) age at menopause in cancer cases and controls (R indicates Pearson correlation coefficient).