PRS-BC313 integration for tailored breast cancer prevention in female patients and their healthy relatives.

This study aims to evaluate how the inclusion of the PRS based on 313 genomic variants (PRS-BC 313 ) influences BC risk estimation by refining existing risk prediction models. The study cohort comprised three groups of women: Healthy female relatives under the age of 50 years of BC index cases with no identified pathogenic or likely pathogenic variants (genetically unremarkable). Healthy women carrying pathogenic or likely pathogenic variants in genes associated with hereditary BC. Women with HBOC (affected) carrying pathogenic or likely pathogenic variants in genes associated with hereditary BC. Participants were recruited through the genetic counselling unit and clinical genetics services of the Institute of Human Genetics at Leipzig University Hospital. Recruitment occurred during routine counselling sessions in 2024 as well as through re-contact of eligible patients up to 2 years after their initial presentation. Over this period, all individuals meeting inclusion criteria for one of the three groups were consecutively selected. Information on BC status was based on clinical history obtained during genetic counselling. Because detailed histopathological reports were not available for all affected participants, we could not systematically distinguish between invasive BC and in situ disease in all cases. Accordingly, all self-reported or clinically documented BC diagnoses were included. Regarding ancestry, the cohort consisted of individuals of European descent, which aligns well with the reference population used for the validation of the PRS-BC 313 model ( online supplemenfigure S1 ). In total, 547 women were approached for study participation and invited to provide written informed consent. Of these, several groups were excluded based on predefined criteria. Eighteen individuals diagnosed exclusively with ovarian cancer (OC) were excluded, as the CanRisk model does not currently offer validated parameters for estimating BC risk in OC-affected patients. Two additional women were excluded because their age exceeded the model’s applicable prediction range (limited to individuals up to 80 years). A further 301 probands were excluded due to incomplete data on key lifestyle, hormonal or reproductive risk factors required for comprehensive BC risk estimation within the CanRisk framework. Of these, approximately 10 individuals actively declined participation, while the remaining exclusions were due to missing information essential for model-based risk calculation. One individual of East Asian ancestry was excluded. The final cohort composition and the sequential exclusion process are illustrated in online supplemental figure S2 . The pedigrees analysed in this study were generated using the PhenoTips web tool during initial patient consultations. Each pedigree spans at least three generations and includes the following information for all listed individuals: Life status. Year of birth and current age. Cancer status. For all index cases, the age at first disease onset was known, as well as the status and results of genetic testing. Furthermore, direct gene test and panel diagnostics were specified for the risk calculation. Where mandatory, year of birth and death of affected relatives was either specified by the patient or when unknown by the patient estimated. In addition, all participants underwent a standardised CanRisk questionnaire. For CanRisk calculations, pedigrees were formatted using PhenoTips as BOADICEA files. Gene panel results were incorporated into the analysis where applicable. For participants in groups 2 and 3, individual gene panel results from diagnostic testing ( BRCA1 , BRCA 2 and other HBOC-associated genes listed in CanRisk) were included directly in the risk calculation via the CanRisk model. For women in the variant-negative group, gene panel information from the tested index patient (or the counselled proband, if tested) was incorporated to confirm the absence of LP/P variants in the family. Thus, monogenic findings were systematically integrated into the risk model for the index, ensuring accurate representation of carrier status in PRS-combined risk estimation. To assess individual BC risk with higher precision, participants completed a standardised CanRisk-based questionnaire that collected both clinical and lifestyle-related data, as specified in the CanRisk model. The questionnaire was completed at the time of risk consultation or per telephone and included the following components: General information: Name, date of birth, height (cm) and weight (kg). Lifestyle factors: regular alcohol consumption, including type, frequency (daily, weekly, monthly) and quantity (eg, wine, beer, spirits). Reproductive and hormonal history: Participants reported age at menarche, history and duration of oral contraceptive use (including use in the past 2 years), menopausal status and use of hormone replacement therapy (HRT), including the type (oestrogen monotherapy vs combination therapy) and duration. HRT use within the last 5 years was also assessed. Breast cancer screening history: Participants indicated whether they had previously undergone mammography, and if breast density was reported using the American College of Radiology (ACR) classification (A–D). Breast density was self-reported based on prior clinical findings. Medical history: Participants were asked about a prior diagnosis of endometriosis, history of sterilisation, bilateral oophorectomy and/or bilateral mastectomy. Women from families with suspected HBOC either underwent in-house panel diagnostics, which included PRS-BC 313 testing, or were referred with an external report of targeted Sanger testing or panel diagnostics. For women where a pathogenic variant (PV) was known within the family, targeted testing was conducted as part of the in-house panel diagnostics. In cases where external diagnostics had been performed or a healthy female relative from families with suspected HBOC had not yet been genetically tested, PRS-BC 313 measurement was performed via a panel. Of the known lower-penetrance CHEK2 variants, p.I157T was included, as it is incorporated into the PRS-BC 313 model and contributes to risk estimation within CanRisk. In contrast, p.S428F—classified in Germany as likely benign—was not included in the monogenic category, nor in the PRS, and therefore was not reported. p.T476M, a recognised risk allele, was reported as a monogenic variant. Accordingly, only p.I157T was treated as a contributory variant within the polygenic and combined risk assessment framework. Genomic DNA was extracted from whole blood using the MagCore Kit 101 and MagCore instrument (RBC Bioscience, New Taipei City, Taiwan). DNA concentration was measured using NanoDrop 2000 (Thermo Scientific, Waltham, Massachusetts, USA) and Qubit (Thermo Scientific). NGS was performed after sample preparation with the Twist Library Preparation EF Kit 2.0, and the Twist Universal Adapter System—TruSeq Compatible, 96 Samples Plate A–D, along with enrichment using the Twist Custom Panel (design name: Cancer_PRS_HUGV6; Twist Design ID: TE-96674869 or HousePanel v2/v2.1, Twist Design ID TE-96574338/TE-91078956, Twist Bioscience, South San Francisco, California, USA). All panels covered the coding sequence of all known genes predisposing for HBOC ( BRCA1 , BRCA2 , PALB2 , TP53 , ATM , CHEK2 , RAD51C , RAD51D , BARD1 , BRIP1 , SMARCA4 , STK11 , PTEN , CDH1 as well as genomic variants to calculate the PRS-BC 313 and 79 genomic variants to determine the ancestry). Sample identification was conducted using the Nimagen RC-PCR assay. Sequencing was performed on a NextSeq500/550 Mid Output v2.5 kit (Illumina, San Diego, California, USA; sequencer: Illumina NextSeq550, Illumina). The mean coverage was at least 300×, with all target regions covered at least 20×. Both single nucleotide variants (SNVs) and copy number variants (CNVs) were detected. The raw data were analysed using the Varfeed software (Limbus, Rostock, Germany), and variants (SNVs and CNVs) were annotated with the Varvis software (Limbus). All variants were described according to GRCh38 and classified following the latest ACMG criteria and ClinGen VCEPs, if available. Databases such as ClinVar, HGMD and HerediCare were used for classification, considering factors such as gene and variant attributes, population frequency (gnomAD), predicted impact on protein function, in silico prediction tools (primarily BayesDel, Revel and SpliceAI-lookup), conservation and phenotype. Raw reads were quality checked using fastqc, and remaining adapter sequences and low-quality data were removed using trimmomatic. The processed data were aligned to hg19 using minimap2, and visual duplicates were flagged with samtools. Haplotypes for the BCAC-313 PRS model were called using freebayes, and positions with coverage below 20× or with conflicting haplotype signals were imputed using twice the allele frequency as described in the BCAC-313 model. The resulting haplotypes were then used to calculate the normalised z-score via the CanRisk API. The final z-scores were combined with the corresponding BOADICEA file to estimate the 10-year risk for each participant. It is important to note that the original PRS was derived from SNP-array data; however, the simultaneous analysis of monogenic variants and PRS via NGS is more efficient. This approach requires careful identification and exclusion of array-specific artefacts from the PRS calculation to ensure accuracy. Currently, imputation adheres to the GC-HBOC consortium’s standard protocol. 12 16 Risk calculations were performed via the CanRisk API, incorporating available clinical, familial, ancestral and genetic data. For each case, risk was calculated both with and without inclusion of PRS-BC 313 , allowing for direct comparison. The analogue CanRisk questionnaire was digitised using an Excel template, in which a trained study staff member manually entered participants’ responses. These digital entries were then imported into the CanRisk interface for automated risk calculation. The transfer from the analogue to the digital format was carefully monitored and checked to ensure accuracy and consistency. Due to the frequent unavailability of PRS data from relatives, we were unable to include PRS information of relatives in the model. Additionally, genetic data from family members carrying pathogenic or likely pathogenic variants were often missing. There was also inconsistency in whether relatives had undergone targeted single-gene testing or broader multigene panel diagnostics. For these reasons, detailed genetic test results for family members were not incorporated into the model (but for the index). Instead, familial risk was included in a simplified form, based solely on whether relatives were reported as ‘affected’ or ‘not affected’, along with their age at diagnosis. Pedigree information was available for at least three generations, depending on participant knowledge. Birth and death years were either provided or estimated when unavailable. Only relatives with sufficient age information were included in CanRisk calculations, as the model requires this for accurate risk estimation. Gene panel data were incorporated wherever available. For participants in groups 2 and 3 (carriers of pathogenic or likely pathogenic variants), individual diagnostic results from gene panels—including BRCA1 , BRCA2 and other HBOC-associated genes listed in CanRisk—were directly included in the risk calculations. For women in the variant-negative group, gene panel information from the tested index patient or counselled proband was used to confirm the absence of pathogenic or likely pathogenic variants in the family. In this manner, monogenic findings were systematically integrated into the CanRisk model for the index participant, ensuring accurate representation of carrier status in combined PRS and monogenic risk estimation. Data analysis was performed with GraphPad Prism, SPSS and Excel software. CIs were computed using the R base library. Genotypes of 79 variants (online supplemental information) that allow for a differentiation between the superpopulations of European, East Asian, South Asian and African ancestry were used as an input to create a principal component analysis model. The genotyping of the samples was done using FreeBayes V.1.3.7. The genotypes were mapped to the principal component analysis model, followed by logistic regression to predict the ancestry of the sample. To benchmark the performance of our ancestry prediction we used the 2156 samples from the 1000 Genome project with known ancestry ( https://www.internationalgenome.org/ ). Samples from the 1000 Genome project were filtered to the regions of the 79 SNPs after that the ancestry of the sample was predicted with our approach. For the superpopulations of European, East Asian, South Asian and African, the model correctly predicted 2115 out of 2156 (98.09%) of the 1000 genome samples. For each sample included in the presented study, the ancestry was predicted using the described model. Additionally, we visualised the results of ancestry prediction for each sample together with the labelled data of 1000 genome project samples as a reference. Discrepancies between the allele frequencies observed in our cohort and those reported in population reference databases (gnomAD v3/v4) were noted for a subset of SNPs. These deviations most likely reflect a combination of low-complexity genomic regions at certain loci and array-specific technical artefacts rather than true biological divergence from the reference population. As detailed by Baumann et al , NGS-based genotyping of PRS loci is subject to well-characterised technical limitations that can contribute to such discrepancies. Importantly, affected SNPs were systematically excluded from analysis across all participants; their impact on the overall validity of risk estimates is therefore expected to be minimal. 12 Distribution equivalence between our local cohort and the PRS base cohort was evaluated by testing reported PRS percentiles against the standard uniform distribution (on the interval (0, 1)) using the Kolmogorov-Smirnov test.