Breast cancer (BC) is the second leading cause of cancer deaths in women worldwide [1–3], and is an increasing public health problem [4–7]. Global BC cases are increasing by more than 3% annually with over one third of world BC cases currently diagnosed under the age of 50 [4, 8–11]. The rate of BC is also increasing in the US, especially for women under the age of 40 [12, 13], and it is the leading cause of death for women in their 40s [1–3].
The high and increasing rate of BC cases worldwide emphasizes the need for sensitive and specific BC risk assessment tools. Standard questionnaire-based BC risk prediction models generally have quite poor performance with AUCs (Area Under receiver-operator Curve) rarely above 0.65 [14] for general population cohorts. Consequently there have been a number of factors proposed to augment and improve them, such as AI-based mammographic information, polygenic risk scores and pathogenic variants [15–20]; however, even with these additions, current BC risk prediction is still, at best moderate, with AUCs rarely above 0.7 [15–20].
This modest performance makes it challenging to target effective primary prevention options (e.g., chemoprevention) for the great majority of women who are not known to have pathogenic variants in breast cancer susceptibility. Further, the efficiency of secondary prevention options (e.g., onset, frequency, and method of BC screening by mammography or other supplemental methods) would also benefit from more accurate BC risk prediction. Based on these consideration, a desirable target for breast cancer screening performance in the context of screening, counseling, prevention would be an AUC of at least 0.8 [14], and thus the current study is designed to investigate a different class of BC predictor which may be combined with present models to further improve BC predictor performance.
In this context DNA damaging agents such as smoking, ionizing radiation, exogenous estrogens, alcohol intake and diets that increase the risk of obesity [21–27] have all been associated with breast cancer risk. When comparing tumors with nearby non-tumor tissues, breast tumors have significantly more DNA damage [28–32] and DNA adducts [28, 29, 33, 34]. For these reasons, we and other groups have been assessing predictive biomarkers that could enhance BC risk prediction: several reports, including those from our own team [35–37], have suggested that decreased DNA repair capacity (DRC) is a major risk factor for development of cancer at many sites including lung, bladder, as well as breast [38]. The goal of the present study is to develop this approach with the goal of improving personalized BC predictive models.
DNA repair is a crucial mechanism for maintaining genomic stability in cells. Defects in the DNA repair machinery increase cell vulnerability to DNA-damaging agents and accumulation of mutations in the genome, and lead to the development of various disorders including cancer. Studies that have measured DRC, including our own, have estimated that DRC is associated with a much higher risk of BC (3-15-fold) [37], than most other established BC risk factors, with the exception of highly penetrant mutations in genes like BRCA1 and BRCA2, genes that are themselves critical to DNA repair. This lack of inclusion of a major risk factor – DRC – may be one of the reasons that current clinical BC risk models have only modest performance.
DRC can be assessed with genotypic, proteomic, or phenotypic approaches. A concern with genomic or proteomic approaches is that mammalian DNA damage repair mechanisms are extraordinarily complex, in humans involving ~ 450 genes, 13 different pathways, with over half the proteins interacting with other proteins from different pathways [39]. Most BC risk models now try to capture genetic risk variants in DNA repair but that genotype typically only explains a small portion of the variation in phenotype [40–44]. Consequently, any specific genomic or proteomic methodology may not reflect overall DRC.
By contrast to genotypic and proteomic approaches, phenotypic approaches such as inducing DNA damage and then measuring the rate of DNA damage repair or the amount of unrepaired/misrepaired DNA damage, or both, have the potential to be more reflective of overall DRC. To date, however, such phenotypic approaches have been laborious to perform resulting in low throughput, and there have been no large-scale prospective studies of the relationship between BC and DRC. In the current study, we have overcome this by using an automated phenotypic approach to DRC characterization, based on measuring the kinetics of DNA double-strand break (DSB) damage repair [45–47] after a radiation challenge.
Specifically, measurement of phosphorylation of histone H2AX at Ser 139 at the site of DNA DSB has become one of the most common DSB assays [48–50] and here we measure the repair of γ-ray induced DNA DSB over time by observing the disappearance of γ-H2AX signals using an image-flow cytometer [45–47]. Of course PBMCs consist of a number of different cell subtypes and it is known that DSB induction sensitivity differs in different PBMC subtypes [51], and thus their DRCs may also vary. For this reason we separately analyzed the different PBMC subtypes, identified using standard surface markers. Using the automated high-throughput RABiT (Rapid Automated Biodosimetry Technology) approach [45, 52–60], time-dependent measurements of the γ-H2AX DSB biomarker after a radiation challenge, together with use of surface markers for different PBMC cell types and a multi-channel image-flow cytometer, allowed us to directly quantitate DRC in each PBMC subtype.
We quantitatively characterized DRC in blood samples from 92 women, 46 of whom were diagnosed with BC, and 46 matched healthy controls. For the women with BC, the blood samples were drawn and stored before BC diagnosis, and the controls were matched by age and time since blood draw. Our goal was to investigate whether parameters derived from the blood sample DRC measurements could be predictive of the risk of BC diagnosis.