The present study examined samples from five adult females (designated CM09, CM10, CM27, CM30 and CM64) with a median age of 61.5 years and an age range of 35-88 years. A total of two pathologists independently assessed the histopathological features of these specimens.

Participants were recruited from April 2010 to May 2013 from the Oncology Hospital of the XXI Century National Medical Center (Mexico City, Mexico). Inclusion criteria were as follows: Adult female patients with a confirmed histopathological diagnosis of TNBC, adequate tumor cell content (≥70% neoplastic cells in the biopsy), and availability of tumor, adjacent non-tumor and peripheral blood samples with sufficient DNA concentration and purity for downstream analyses. Exclusion criteria included the absence of a confirmed TNBC diagnosis, diagnosis of other malignancy or cases of cancer recurrence. Samples with insufficient DNA concentration or purity for downstream analyses were also excluded. A total of 14 samples were analyzed, comprising TUM (n=5), ADJ (distance, 1 cm; n=5) and peripheral blood (LEU, n=4). Histological assessment of the biopsies was performed by two trained pathologists independently. They assigned the phenotypic diagnosis of TNBC. Only biopsy samples for which both pathologists independently reached the same TNBC diagnosis were included in the study.

The present study was approved (approval no. 2008-785-001) by the Scientific and Ethics Committees of Health Research Coordination of Mexican Social Security Institute, Mexico City, Mexico. All participants were informed about the nature of the study and provided written informed consent. The study was conducted in accordance with best clinical practice and all participant identities were anonymized. Immunohistochemistry data were obtained from the pathology and clinical records of the patients. All samples were TNBC.

DNA was extracted from TUM, ADJ and LEU using a commercial kit (QIAamp^® Micro kit, cat. no. 56304, Qiagen GmbH) as previously described (13). Samples were digested with proteinase K for 3 days at 56˚C in a water bath, with fresh enzyme added every 24 h.

The extracted DNA was quantified by spectrophotometry using a Nanodrop 2000 instrument (Thermo Fisher Scientific, Inc.). DNA quality was assessed by Multiplex PCR (Multiplex PCR kit, Qiagen GmbH) employing primers designed to amplify different regions of GAPDH gene as previously described (14). PCR amplification was performed as follows: Initial denaturation at 95˚C for 15 min, followed by 40 cycles of 94˚C for 30 sec, 57˚C for 90 sec, and 72˚C for 90 sec and final extension at 72˚C for 10 min. PCR products were visualized by electrophoresis on a 2% agarose gel, stained with RedGel^® Nucleic Acid Gel Stain (Biotium, Inc.) and documented using an ultraviolet transilluminator system (Syngene). The dye was placed in the molten agarose during gel preparation.

Samples were analyzed using Affymetrix^® CytoScan^™ microarrays according to the manufacturer's protocol, starting with 250 ng DNA. An additional five PCR cycles were included to increase DNA yield. PCR products (90 µg) were fragmented and labeled through an additional PCR labeling step using the Cytoscan™ HD Suite which included Cytoscan HD matrices and Cytoscan reagent kit (cat. no. 901835); Thermo Fisher Scientific, Inc.

Raw intensity files (.CEL) obtained from the Affymetrix^® CytoScan^™ HD Array platform (Thermo Fisher Scientific, Inc.) (13) were analyzed using the proprietary Chromosome Analysis Suite (ChAS) software (version 4.3.0.71; Thermo Fisher Scientific, Inc.). The CytoScanHD_Array.na36.annot.db file was employed for annotation, with the GRCh38 (February 2013) human genome assembly (Genome Reference Consortium, 2013; https://www.ncbi.nlm.nih.gov/assembly/GCF_000001405.26/) serving as the reference model.

Data processing was performed using the Circular Binary Segmentation algorithm embedded in ChAS software (version 4.3.0.71; Thermo Fisher Scientific, Inc.) (15), in which the Log2 ratio for each marker was calculated relative to the reference signal profile. CNV analysis was performed following normalization to baseline reference intensities based on the Chromosome Analysis Suite (ChAS) software (version 4.3.0.71; Thermo Fisher Scientific, Inc.) reference model, which included 284 HapMap samples and 96 healthy control individuals (15). The hidden Markov model (16) implemented in ChAS software (version 4.3.0.71; Thermo Fisher Scientific, Inc.) was applied to determine CN-state and corresponding breakpoints.

A customized high-resolution filter was applied for detection of CNVs, defining CN-gains as regions containing ≥50 markers and ≥400 kbp and CN-losses as regions containing ≥50 markers and ≥100 kbp. Quality control was performed using the median absolute pairwise difference (MAPD) and single nucleotide polymorphism-quality control (SNP-QC) score. Only samples with MAPD <0.25 and SNP-QC >15 were included in the subsequent analyses. Other parameters were as follows: Gene coverage >36,000 RefSeq genes with one marker/880 bases; backbone (non-gene) coverage of one marker/1,737 bases; highest density of SNPs and CN-probes for whole-genome coverage; 2.67 million markers; 750,000 SNPs; 1.9 million non-polymorphic probes; enriched gene coverage for cancer and germline or inherited genetic variation; X chromosome genes (one marker/486 bases); covering 12,000 Online Mendelian Inheritance in Man (OMIM) genes (one marker/659 bases); cancer gene coverage (one marker/553 bases); CN-resolution <25 kb; 99.9% sensitivity.

A custom Perl script was developed to process the CNV segment data files generated by ChAS software (version 4.3.0.71; Thermo Fisher Scientific, Inc.) for each sample. The script compared all individual CNV files to compile a comprehensive gene table including the frequency of affected patients, CNV event type (gain or loss), chromosomal position, cytogenetic band and gene function, disease associations and inheritance patterns retrieved from the OMIM database (omim.org) (17). Additional annotations were integrated from Haploinsufficiency Prediction version 3 from the DECIPHER database v11.35 (deciphergenomics.org) (18), and gene information from dbEMT 2.0 (dbemt.bioinfo-minzhao.org) (18,19).

Genes altered in ≥3 patients with TNBC were included for further analysis and visualization. In the absence of an established measurement (gold standard), an arbitrary value was determined to provide guidance in exploring relevant alterations. The present study analyzed CNV type (gain or loss), chromosomal location, cytoband, cumulative length, number of affected patients, associated genes, haploinsufficiency scores, EMT-gene status and disease associations associated with CNVs of TUM, ADJ and LEU. Genes altered in ≥3 in TUM, ADJ and LEU samples were included for analysis and visualization. A karyotype was generated (Fig. 1), using the KaryoploteR package (version 1.30.0; bioconductor.org/packages/karyoploteR/) (20) in combination with biomaRt (version 2.60.1; bioconductor.org/packages/biomaRt/) (21). Gene overlap analysis was conducted by generating Venn diagrams using the Jvenn web server (22). Hallmarks of Cancer enrichment analysis (adjusted P<0.05) was performed using 6,763 genes (23) (Fig. 2). Metabolic pathway enrichment analysis was performed with Reactome (version 88) (24), considering results with a false discovery rate (FDR) <0.05 as statistically significant. To identify the Hallmarks of Cancer associated with the TUM, ADJ and LEU profiles, an interaction network was constructed by CNV types (gains and losses), integrating genetic and physical interactions, biological pathways and predicted associations (Fig. 3). Network generation and visualization were performed using STRING (25), GeneMANIA (26) and Cytoscape (version 3.10.3) (27). Network connections were established based on direct interactions, shared neighbors and weight scores to identify the most significant gene associations. Manual annotation of the corresponding Hallmarks of Cancer, including ‘adhesion’, ‘angiogenesis’, ‘inflammation’, ‘migration’, ‘metastasis’, ‘morphogenesis’, ‘proliferation’ and ‘survival’ (2) was performed using databases including The Human Protein Atlas (28), the NCBI Gene database (ncbi.nlm.nih.gov/gene/), UniProt (https://www.uniprot.org/) and GeneCards (genecards.org/).

CNV profiles identified in TUM, ADJ and LEU samples were compared against publicly available copy number data from The Cancer Genome Atlas [(TCGA) TNBC cohort (n=139), accessed through cBioPortal (cbioportal.org), TCGA dataset accession: TCGA-BRCA; study ID: brca_tcga; (29); Cancer Genome Atlas Network, 2012(30)], to identify shared and potentially population-specific chromosomal alterations.

CNV data are presented as categorical genomic events (gains or losses) with associated genomic coordinates, affected genes and frequency across patients (cut-off, ≥3 patients). Data are summarized by frequency counts, cumulative genomic lengths (cl) in megabase pairs (Mbp) and enrichment statistics. P<0.05 was considered to indicate a statistically significant difference. Enrichment significance was evaluated using Fisher's exact test with Benjamini-Hochberg correction for multiple comparisons. Pathway enrichment analyses were performed in Reactome (v.88) and the Hallmarks of Cancer tool (23,24), both of which apply hypergeometric testing with FDR correction. Data analysis was performed using R software (version 4.4.0; R-project.org) and the RStudio environment (version 2025.05.0+496; Posit PBC; posit.co), incorporating packages from Bioconductor (version 3.21; bioconductor.org): KaryoploteR (v1.30.0; bioconductor.org/packages/karyoploteR), biomaRt (v2.60.1; bioconductor.org/packages/biomaRt), ReactomePA (v1.50.0; bioconductor.org/packages/ReactomePA) (31) for pathway enrichment, and ggplot2 (v3.5.1; cran.r-project.org/web/packages/ggplot2) (32) for data visualization.

Samples were collected from five patients with TNBC (third-generation Mexicans), aged 35-88 years, including four treatment-naïve individuals and one patient who had received prior therapy (Table I). Four samples corresponded to infiltrating ductal carcinoma (IDC) and one to lobular carcinoma (LC; Table I). The proportion of neoplastic cells in the tumor tissues ranged from 70 to 100%. The present study analyzed samples from five female patients diagnosed with TNBC, including both IDC and LC. Clinical staging was T2N0M0 (stage IIA) for CM09 and CM64, T2N2M0 (stage IIIA) for CM10, T2N1M0 (stage IIB) for CM27; data was not available for CM30. All patients were treatment-naïve at the time of sample collection, except CM64, whose prior treatment records were unavailable.

The immunohistochemical analysis of the samples revealed a heterogeneous profile of biomarkers (Table II). All five cases were negative for HER2, estrogen receptor (ER) and progesterone receptor (PR) expression. Ki67 expression was positive in CM27 and CM64, suggesting a higher proliferative index in these tumors. CK5 was positive in two cases (CM09 and CM64), indicating a possible basal phenotype. Overall, the cohort displayed a predominance of TNBC features, with variability in proliferation and basal markers.

The CNVs were estimated based on regions where the majority of SNPs do not exhibit heterozygosity. Table III summarizes the chromosomes with the highest frequency of CNV events, reflecting recurrent alterations, although exact chromosomal coordinates were not identical across samples.

To identify the most relevant CNVs in TNBC, alterations detected in ≥3 patients were analyzed. A similar distribution pattern was observed across all CNVs, with the highest number of events detected in tumor (TUM) samples (n=1,717), followed by adjacent non-tumor tissue (ADJ; n=132) and leukocytes (LEU; n=13) samples (Table SI). In addition, Table SII presents the accumulated CNV length per chromosome to assess whether larger losses were associated with increased genomic disruption.

In TUM samples, the chromosomes most affected based on cumulative CNV length (Mbp-cl) and the number of CNV events were chromosome 3 (gains) and chromosome 5 (losses). In ADJ samples, chromosomes 8 (gains) and 4 (losses) were predominantly altered, whereas in LEU samples, chromosome 4 exhibited both gains and losses (Table SII). Notably, in TUM and ADJ samples, the most frequent CNV sizes ranged from 200 to 500 kbp for gains and from 100 to 200 kbp for losses (Table SIII).

In TUM, recurrent gains were observed in 1q23.3 (12.77 Mbp-cl) and 1q32.1 (7.03 Mbp-cl), as well as in the 8q24.22-8q24.3 region (20.32 and 16.25 Mbp-cl, respectively), while a loss was noted in 6q25.2 (3.18 Mbp-cl; Table SIV). ADJ displayed gains in cytobands 1p36.32, 4p16.1, 5p15.33, 8q24.3 and 20q13.33 (Table SIV). LEU presented both gains and losses, including a loss at 8p11.22 and 14q11.2 and gains at 14q32.33 and 22q11.22 (Table SIV). The total number of gains and losses per chromosome across all three tissue types is summarized in Tables SII and Table IV.

These findings indicate that CNVs were not restricted to TUM samples but also present in ADJ and LEU, potentially reflecting early or systemic genomic instability. Such alterations may serve as potential biomarkers and provide insight into the interactions between the tumor and its microenvironment.

A Venn diagram was constructed to examine CNV-TNBC-relevant genes present in ≥3 patients across TUM, ADJ and LEU samples (Fig. 2A). The present study identified 1,615 CNV unique genes in TUM, 25 in ADJ and one in LEU (Fig. 2; Table SV). Of the 95 genes shared between TUM and ADJ, six carried Hallmarks of Cancer annotations [ADGRB1, CHRNA4, RecQ-like helicase (RECQL)4, SSTR5, TONSL and UVSSA; Table SV], suggesting functional relevance across both tissue compartments. These six shared genes exhibited a greater number of enriched hallmarks compared with LEU (Fig. 2B-D).

The genes altered in TUM were associated with ‘sustaining proliferative signaling’, ‘tumor-promoting inflammation’, ‘tissue invasion and metastasis’, ‘evading immune destruction’, ‘resisting cell death’, ‘evading growth suppressors’ and ‘reprogramming energy metabolism’. In ADJ, altered genes were linked to ‘evading growth suppressors’. In LEU, the single altered gene (IGLL5) was annotated with ‘genome instability’ as a Hallmarks of Cancer category (Fig. 2); however, no statistically significant pathway enrichment was identified in this sample type (Table V).

Comparison with TCGA-TNBC data (Table SVI) revealed both shared and potentially population-specific patterns. Alterations in 8q24.3 (MYC amplification) were common in both the present study cohort and the TCGA-TNBC cohort, consistent with this being a canonical TNBC alteration (29,30). However, the present study observed a higher frequency of gains in 1q23.3 and losses in 6q25.2, although the small sample size (n=5) precluded statistical testing of these differences (30,33).

Metabolic pathways were predicted in Reactome using Homo sapiens as the model organism based on the CNV genes associated with Hallmarks of Cancer (Table V). In TUM, the most significantly enriched pathways included ‘DNA repair’ (NER and TP53 signaling), ‘extracellular matrix remodeling’ (collagen assembly) and ‘cell cycle checkpoints’ (G2/M). In ADJ, enrichment was observed in ‘NOTCH signaling’ and ‘RAC3 GTPase cycle’ pathways, while LEU showed limited enrichment, with IGLL5 associated with ‘antigen binding’ and ‘immune activation’.

CNV genes associated with Hallmarks of Cancer and linked to metabolic pathways were used to construct interaction networks using STRING (Fig. 3). Edges represent associations based on metabolic pathways, gene expression, localization, inferred interactions, genetic interactions, data mining and neighborhood associations. Each node was annotated with flags corresponding to the reported hallmarks. In the LEU network, a single node representing IGLL5 was observed. In ADJ, the genes with the highest number of connections were E1A-binding protein p300 and RECQL4. The TUM network contained 39 nodes, with ERCC4, EGFR and HSP90AB1 showing the highest connectivity. Notably, EGFR, telomerase reverse transcriptase, neurotrophic receptor tyrosine kinase 1 and ribosomal protein S6 kinase B1 were associated with the largest number of hallmarks within the network (Table VI) (39,49,53-64).