Ethical approval and consent to participate

All participants gave their informed consent, and the study was approved by the local ethics committee of the National University of Colombia (study number 2019–5738). Although in the present study the secondary findings report was not performed due to the approved protocol restrictions by the ethics committees, the technical and bioinformatics pipeline feasibility was demonstrated. This study was conducted in accordance with the ethical standards described in the Declaration of Helsinki.

Study cohorts

The first study cohort consists of 25 individuals with hereditary cancer syndromes who met clinical criteria and had a molecular diagnosis, i.e., a germline P/LP variant that explains the patients’ phenotype. A total of 25 cases meet these criteria and are hereafter referred to as the “pipeline validation cohort.” Our second group, called the “TNBC discovery cohort,” includes 24 patients with a confirmed pathological diagnosis of TNBC. Both groups underwent whole exome sequencing (WES) after signature and acceptance of written informed consent for research.

Whole exome sequencing of germline DNA from blood

Genomic DNA was extracted from peripheral blood, and the DNA sample was quantified by spectrophotometry (Nanodrop OneC) and fluorometry (Qubit 4) to normalize the DNA concentration across samples. Whole Exome Sequencing (WES) libraries were prepared using enrichment kits according to the manufacturer’s instructions (Agilent SureSelect All Exon V6 for the TNBC discovery cohort and Illumina DNA Prep with Exome 2.5 Enrichment for the pipeline validation cohort). Briefly, DNA was randomly fragmented into 180–280 bp fragments. The resulting fragments were end-repaired and ligated with Illumina molecular adapters. Adapter-containing fragments were amplified by PCR, size-selected, and purified. Hybridization capture of the libraries was performed using a buffer containing biotin-labelled probes, and streptavidin-coated magnetic beads were used to capture the gene exons. Subsequently, unhybridized fragments were washed away, and the probes were digested. The captured libraries were further enriched by PCR amplification. Libraries were analysed with Qubit and a fragment bioanalyzer to determine size distribution. They were then quantified, pooled, and sequenced on Illumina platforms using a 2x150 bp paired-end read strategy.

Development of a pipeline for germline variant detection based on GATK best practices

The study evaluates the feasibility and proof of concept for implementing an end-to-end bioinformatics pipeline. This workflow integrates the recommendations of GATK Best Practices [7] to establish a robust and reproducible pipeline for germline variant identification, covering everything from initial data quality assessment to variant annotation and prioritization. First, the quality assessment was performed with FastQC [8], which detects potential contamination or artifacts before proceeding with alignment. For sequence alignment, the cleaned data are mapped to the reference genome (hg38) using the Burrows-Wheeler Aligner BWA-MEM [9] software, and the output files are converted and sorted by chromosomal position (SAM/BAM) with Samtools (view, sort, index). GATK then performs critical steps for variant calling optimization, including duplicate marking (MarkDuplicatesSpark), alignment sorting (SortSam), base quality recalibration (BaseRecalibrator, ApplyBQSR), and finally, germline variant calling with HaplotypeCaller [10]. After separating SNPs and INDELs with SelectVariants, quality filters are applied using the VariantFiltration tool to obtain a set of reliable variants. The SNP and InDel filter parameters are shown below: SNP: QD < 2.0, FS > 60.0, MQ < 40.0, HaplotypeScore>13.0, MappingQualityRankSum < −12.5, ReadPosRankSum < −8.0; InDel: QD < 2.0, FS > 200.0, ReadPosRankSum < −20.0. For general metrics, variants with a depth > 20x and an allele frequency>15% were considered valid. In a complementary manner, alignment metrics (CollectAlignmentSummaryMetrics) and library fragment size metrics (CollectInsertSizeMetrics) are generated with GATK to evaluate the consistency of the process. The resulting variants are functionally annotated with Ensembl VEP [11] and the Franklin Genoox platform, incorporating information on the allele frequency of the variants, molecular effects, and clinical databases. *Databases used in variant annotation include: dbSNP, gnomAD (Exome), gnomAD (Genome), Sift, Polyphen2, Mut Taster, Mut Assessor, Fathmm, ClinVar, Gerp, Revel, SpliceAI, Vep, dbNSFP, dbscSNV. **Genomic variant classification criteria: The guidelines of the American College of Medical Genetics and Genomics (ACMG) [12] and the Association for Clinical Genomic Science (ACGS) [13] were followed.

To ensure reproducibility, this workflow was implemented in a Docker container with specific versions of the aforementioned bioinformatics tools. Finally, we tested the bioinformatics pipeline using a validation set of independent samples previously sequenced and analysed, with platforms developed for a clinical diagnostic environment (Euformatics Genomics Hub: https://www.euformatics.com, Emedgene:https://emg.emedgene.com, Franklin: https://franklin.genoox.com), based on fastq files from patients for whom a germline variant previously identified, responsible for a hereditary cancer syndrome, had been previously identified. Then, we ran the raw data (fastq) from all these patients through the new germline pipeline, and 100% concordance was obtained in the results, confirming the variants previously identified with the pipeline and the clinical exome analysis platform. The concordance strictly refers to the ability of the bioinformatics pipeline to accurately detect and annotate the causal variants of the “gold standard” phenotype present in the pipeline validation cohort. These results confirm the robustness and reliability of this pipeline for germline variant identification in the context of precision genomic diagnosis.

Code availability and hybrid Docker orchestration (high reproducibility): The exact versions of the bioinformatics tools, the complete source code, the command-line parameters, and the computational environments are available in the GitHub repository at https://github.com/GatoconBata-07/Germinal_Pipeline_GATK_BEST.

Consolidation of an expanded and comprehensive panel of cancer susceptibility genes

To prioritize germline variants with potential impact on cancer predisposition, genes were selected from a panel of cancer susceptibility genes (CSG) compiled from the literature [14–17], and from gene panels used in clinical diagnosis (Illumina TruSight Oncology, Pan-cancer panel CD Genomics). We selected the genes from the IntOGen database that were present in at least 1% of cancer patients. We also selected genes that were consensus in at least three sources in the OnkoKB database, and were cited as cancer drivers. We finally obtained a comprehensive and expanded panel of 813 genes (pancancer panel).

Prioritization of germline variants

After exome annotation, germline variants with a sequencing depth of ≥20 reads and an allelic frequency >15% were selected from the comprehensive panel.The following criteria were used for variant prioritization: i) variants predicted to cause loss of function (nonsense, frameshift, and canonical splice sites: + /-3 bp); ii) missense variants and in-frame indels; iii) synonymous variants. The terms were entered into the Franklin platform using the Human Phenotype Ontology (HPO) nomenclature: Tumor (HP:0002664), Carcinoma (HP:0030731), Breast cancer (HP:0003002). Finally, the prioritized variants were interpreted according to the American College of Medical Genetics and Genomics (ACMG) criteria [12].

The developed pipeline enables precise identification of causal variants in patients with various types of cancer

Using complete exome data from cancer patients, we developed a pipeline (Fig 1) to identify germline variants in accordance with the Broad Institute´s best practices (GATK), and we confirmed the results using a pipeline validation cohort comprising samples previously characterized by a pipeline used in clinical diagnostic practice. This validation set identified the causal P/LP variants for various hereditary cancers in 25 patients, including breast cancer, tuberous sclerosis, juvenile myelomonocytic leukemia, colorectal cancer, ovarian cancer, multiple osteochondromas, schwannomatosis, neurofibromatosis, familial adenomatous polyposis, and Li-Fraumeni syndrome. It revealed a spectrum of P/LP variants in the BRCA1, BRCA2, TSC2, MLH1, TP53, PALB2, KRAS, STK11, PMS2, EXT1, BAP1, NF1, and NF2 genes.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Bioinformatics pipeline for the identification of germline variants in cancer susceptibility genes.

https://doi.org/10.1371/journal.pone.0356762.g001

Based on validation cohort results (n = 25 patients), summarized in S3 Table, we identified 25 causal variants associated with cancer in patients, of which 23 (92%) are classified as pathogenic and 2 (8%) are likely pathogenic. Truncating events that result in loss of function, dominated—frameshift 11/25 (44%) and nonsense 7/25 (28%)—followed by missense 5/25 (20%) and canonical splice site 2/25 (8%). Most pathogenic variants were found in genes involved in the homologous recombination DNA repair pathway (HDR): (BRCA1/BRCA2/PALB2 (13/25: 52%), followed by genes involved in the mismatch repair pathway (MLH1/PMS2, 3/25: 12%). and pathogenic variants were less frequently identified in other cancer predisposition syndromes, such as Li-Fraumeni syndrome, Peutz-Jeghers syndrome, tuberous sclerosis, and neurofibromatosis (TP53, STK11, TSC2, NF1/NF2, BAP1, EXT1, KRAS) among others. This pattern—characterized by the inactivation of tumor suppressors and the prevalence of truncating variants in BRCA1/BRCA2/PALB2—is consistent with population studies indicating that protein-truncating variants in these genes significantly increase cancer risk and form the core of susceptibility to hereditary breast and ovarian cancer [18].

Overall, the results of the pipeline developed for germline variant identification show an exact match in detecting P/LP variants previously identified by a clinical diagnostic bioinformatics platform. These findings have significant clinical implications, including genetic counselling and monitoring family members for early follow-up and intervention for carriers of the identified P/LP variants.

The whole-exome approach enables the detection of germline variants in cancer-susceptibility genes in TNBC patients

After achieving 100% concordance in the pipeline validation cohort, we sequenced the whole exome of TNBC patients (TNBC discovery cohort) and processed the data through the pipeline. We analyzed all germline variants in a panel of 813 cancer genes (pancancer panel) (Fig 2 and S4 Table). A total of 13,781 variants were identified in these patients within the panel, including 10,694 single-nucleotide variants (SNVs) and 3,087 insertions/deletions (InDels) (S5 Table). Among these variants, 11,072 are benign or probably benign, 2,702 are variants of uncertain clinical significance, and 7 are P/LP variants, with only 3 explaining the hereditary cancer phenotype in patients with TNBC. We reviewed all variants in the pancancer panel and established the functional consequences for all patients (Fig 3).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Genomic distribution of variants found in the pancancer panel.

(A) Among all variants, 48.25% are in intronic regions, 21.74% in exons, and 17.63% in intronic regulatory regions. Variants in splicing regions make up 4.13% and variants in the 3′ and 5′ UTRs account for 4.12% and 2.08%, respectively, potentially affecting RNA maturation and translation. An additional 1.16% and 0.35% are located in upstream and downstream regulatory regions; other variants are mapped to these regions (0.29% and 0.13%). Canonical splice sites contain the fewest variants (acceptor, 0.10%; donor, 0.01%). Overall, most germline variants are found in non-coding regions. The high number of variants annotated as benign or likely benign (11,072/13,781) within pancancer genes indicates they are polymorphisms without direct roles in tumor development. (B) Counts and types of alterations in pancancer genes. SNV, single-nucleotide variant; indel, insertion/deletion.

https://doi.org/10.1371/journal.pone.0356762.g002

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Analysis of variant distribution based on their functional effects in cancer genes identified in TNBC patients.

The analysis shows that most variants are synonymous (1,413), followed by missense (1,348), and promoter region variants (1,051). Variants in promoter-flanking regions (815) and CTCF-binding sites (392) may influence epigenetic and transcriptional regulatory mechanisms. Additionally, variants were found in enhancers (182), open chromatin regions (126), and TF binding sites (71), suggesting a potential effect on chromatin accessibility. Non-frameshift variants (148), frameshift variants (67), and stop-gain mutations (13) can alter the length or integrity of the protein reading frame. In contrast, start-loss (6) or start-gain variants involve the loss or gain of the translation start codon. The remaining variants are mostly located in deep intronic regions, making their functional effects unclear. These findings emphasize the importance of studying both coding variants and those in genome regulatory regions.

https://doi.org/10.1371/journal.pone.0356762.g003

We identified three pathogenic variants associated with an increased breast cancer risk in three patients: in high-risk genes BRCA1 p.Gly559Valfs*13 (Patient 21: P21), RAD51D p.Val32Phefs*38 (Patient 9: P9), and a moderate-risk gene in ATM p.Lys750= (Patient 6: P6). These variants affect genes involved in the homologous recombination repair (HRR) pathway. Recent studies show that BRCA1 and RAD51D are high-risk genes that predispose individuals to breast cancer. ATM mutations are linked to an increased risk of breast, pancreatic, and lymphoma cancers due to loss of function, which can lead to the accumulation of mutations [19].

The variant in the ATM gene (ATM:c.2250G > A, p.Lys750 = NM_000051.4) was particularly interesting to us because, despite its moderate risk classification, it has been observed in patients with hereditary cancer. A case reported in the literature [20] describes a 49-year-old female diagnosed with breast cancer who carries this variant. Expert panels such as ClinGen and Decipher have identified associations of P/LP variants in the ATM gene with hereditary breast carcinoma and familial ovarian cancer, as supported by publicly available resources [21,22]. Additionally, this variant has 28 entries in the ClinVar database [23], all classified as P/LP (Accession: VCV000003044.76). The variant affects a critical position at the splice donor site in exon 14 (S7 Information). In silico splice prediction tools indicate a strong disruptive or deleterious effect (SpliceAI = 0.94). Consistent with this, functional studies [24] demonstrate that ATM:c.2250G > A alters normal splicing and results in complete skipping of exon 14, producing the r.2125_2250del transcript and a 42-amino acid frameshift deletion (p.Glu709_Lys750del).

Classic studies in lymphoblasts show that variants affecting this canonical splicing site reduce ATM protein expression, decrease kinase activity, and increase chromosomal instability, supporting a deleterious effect at the molecular level [25]. Overall, the evidence combines robust in silico predictions, RNA validation, and functional data pointing to a pathogenic mechanism of splicing alteration caused by a synonymous variant at the end of exon 14. This is a pathogenic mechanism inferred from the literature, not a functional validation derived directly from the ATM variant patient.

After reviewing the ACMG criteria for variants classified as P/LP and recategorizing them, we identified variants that do not causally explain the phenotype of TNBC patients (S6 Table). These variants include SDHA c.1064 + 2T > A (LP), RUNX1 p.Arg427Profs*174 (LP), FANCA p.Arg1187Glufs*28 (P), and FBXO11 p.Gln21Serfs*116 (LP). Deleterious variants in the SDHA gene are linked to pheochromocytomas, paragangliomas, and gastrointestinal stromal tumors [26]. P/LP variants in the RUNX1 gene are linked to hematologic malignancies [27], such as myelodysplastic syndromes, while biallelic variants in the FANCA gene are associated with Fanconi anemia and a significant predisposition to leukemia and squamous cell carcinomas [28]. Alterations in the FBXO11 gene have been linked to diffuse large B-cell lymphomas [29]. These findings are not part of the set of genes with causal evidence for breast cancer risk and are therefore considered unrelated to the development of TNBC at this time.