13q14.2 loss is a recurrent event associated with poor breast cancer survival.
Loss of chromosome cytoband 13q14.2 has emerged as one of the most frequent aberrations in patients with CLL (38). However, its frequency and clinical im- plications in other cancers remain largely unexplored. We initially investigated the prevalence of 13q14.2 loss across cancer types in the pan-cancer dataset from The Cancer Genome Atlas (TCGA) (39). We found that loss of this cytoband is widespread, with the highest frequencies respectively observed in non-small cell lung cancer, ovarian cancer, soft tissue sarcoma, and BCa (Fig. 2).
Then, given the notable occurrence of 13q14.2 loss in BCa, particularly within a substantial patient cohort (n = 210), we further explored its potential role in BCa. To achieve this, we conducted a series of analyses. First, we examined the prevalence of 13q14.2 loss in comparison to other focal copy number (CN) losses using gene-based SNP6 array data in two large and well-annotated bulk BCa patient cohorts: Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) (40) and TCGA. Of note, 13q14.2 loss is the 33rd most lost cytoband in the METABRIC dataset and is part of the 6th most common sub-chromosome arm-level CN loss (Fig. 3a). In TCGA, 13q14.2 is the 7th most common Supplementary sub- chromosome arm-level CN (Supplementary Fig.S1a). Loss of 13q14.2 was predominantly heterozygous (-1), with 13q14.2 homozygous loss (-2) only observed in a small fraction of patients in both cohorts.
Secondly, to statistically confirm the previous data, we evaluated the frequency of 13q14.2 CNAs in comparison to other focal CNAs, encompassing both losses and gains, across the METABRIC and TCGA cohorts. Remarkably, in both datasets, the incidence of 13q14.2 loss significantly surpassed that of other losses, whereas its gain was less prevalent compared to gains observed in other regions (Fig. 3b and c). These observations indicate that the loss of 13q14.2 may be subject to positive selection pressure within tumor populations.
Thirdly, we explored the frequency of 13q14.2 CNAs within distinct BCa subtypes in METABRIC and TCGA, delineated by the Prediction Analysis of Micro-array 50 classification (PAM50) (41). Notably, approximately 50% of patients classified as Basal, Her2-enriched, and lumB subtypes exhibited 13q14.2 heterozygote loss, while about 25–30% of patients identified as LumA and Claudin-like subtypes, manifested this loss in both cohorts (Fig. 3d and e). Less than 2% of all patients exhibited amplification of 13q14.2 in these cohorts.
Lastly, we utilized the clinical data from the METABRIC and TCGA datasets to explore the impact of 13q14.2 loss on patient overall survival. Patients with 13q14.2 loss demonstrated markedly poorer survival outcomes compared to their wildtype counterparts within the METABRIC cohort (Fig. 3f). Further stratification by molecular subtypes unveiled significant survival discrepancies, particularly within the HER2- and ER + subgroups (Fig. 3d and e). Although statistical significance was not reached, a trend toward poorer survival was observed for patients harboring 13q14.2 loss compared to those with 13q14.2 diploidy in other subtypes of the METABRIC cohort (Supplementary Fig. S1b-d) and across the TCGA cohort (Supplementary Fig. S1e-i).
These collective findings underscore the prevalent occurrence of 13q14.2 loss in BCas, its association with adverse survival outcomes in specific molecular subtypes, and its potential as a clinically relevant biomarker.
13q14.2 loss in breast cancer is part of a larger repertoire of genomic alterations rather than an isolated event.
In cancer cells, multiple evolutionary pathways are pursued to pro- mote survival and growth. Oncogenic drivers compete under selective pressure, leading to the dominance of certain drivers while others are outcompeted, resulting in mutual exclusivity (42, 43). However, certain oncogenic drivers may also be selected concomitantly, most likely due to a co-regulatory function during tumorigenesis (43). To ascertain whether 13q14.2 loss represents an independent oncogenic event or if it coincides with other genomic alterations in BCas, we performed co-occurrence and mutual exclusivity analyses with both mutated genes and other re- current CNAs in TCGA and METABRIC BCa cohorts. Firstly, we conducted an analysis of mutual exclusivity or co-occurrence using Fisher’s exact test, which involved computing p-values and odds ratios for each gene among those with a mutation frequency greater than 2% against the loss of 13q14.2 in both the TCGA and METABRIC cohorts. Mutated genes exhibiting a positive -log10(p value) * Direction, such as TP53, strongly co-occurred with 13q14.2 loss (Fig. 4a and b). The observed co-occurrence of TP53 mutations with loss of 13q14.2 is particularly noteworthy, given that TP53 serves as the gatekeeper of the cell cycle (44). Loss of the 13q14.2 cytoband, which includes the RB1 gene that controls the cell cycle, combined with TP53 mutation, synergistically leads to a significant reduction in the cell’s ability to regulate its growth and repair DNA damage. This identifies a potential treatment target for tumors harboring loss of the 13q14.2 cytoband. Conversely, mutated genes exhibiting negative -log10(p value) * Direction, including MAP3K1 and PIK3CA, were mutually exclusive with 13q14.2 loss in both cohorts (Fig. 4a and b). This suggests that the regulatory effect of 13q14.2 loss might represent both alternative or overlapping pathways for tumor development with that of involving mutations in MAP3K1 and PIK3CA.
Secondly, we conducted a similar analysis to assess the co- occurrence or mutual exclusivity of 13q14.2 loss with the loss or gain of other top altered cytobands in both the TCGA and METABRIC cohorts. Notably, loss of 13q14.2 often coincides with gains of 8q12-q24 and 1q22-43 (Fig. 4c), as well as losses of cytobands adjacent to 13q14.2 and 17p11-p13, 11q22-q25, and 8p12-p23 (Fig. 4d) across both patient cohorts. It is noteworthy that the tumor suppressor gene TP53 (17p13.1) and its negative regulator MDM4 (1q32.1) are located within the co-occurring loci.
To further explore this, we conducted additional analyses of the genomic alterations involving these genes in TCGA BCa samples, comparing those with and without 13q14.2 loss. This revealed that TP53/MDM2/MDM4 alterations occurred 11% more frequently in cases where 13q14.2 was lost (Fig. 4e; p = 3.2x10-10, Fisher’s exact test). This suggests a potential functional relationship between the loss of 13q14.2 and TP53/MDM2/MDM4, involving both mutations and CNAs. The co-occurrence of these alterations may indicate a synergistic effect or shared underlying mechanisms in BCa development or progression.
These observations collectively suggest that 13q14.2 loss does not act alone but interacts with other genomic alterations to drive tumor development.
13q14.2 loss affects breast cancer cells and their tumor microenvironment.
To study the impact of 13q14.2 loss on cancer cells, we conducted differential gene expression analysis (DEA) using mRNA data from patient samples with 13q14.2 loss compared to those with diploidy for this region, utilizing the TCGA BCa dataset. As expected, the expression of multiple genes located in the 13q14.2 region were significantly downregulated (Fig. 5a, orange circles). Interestingly, other genes located on the 13q chromosome arm were also downregulated, suggest- ing that a considerable number of patients have lost larger fractions of the 13q chromosome arm in addition to 13q14.2 loss (Fig. 5a, purple circles). In contrast, multiple genes involved in cell cycle-related pathways were remarkably upregulated, including but not limited to E2F1, DSN1, GINS1, PCNA, AURKA, TPX2 and DSCC1 (Fig. 5a, red circles).
To complement this gene expression analysis, we also performed differential protein expression analyses in the same patient samples, where available. It is worth noting that the number of statistically significant downregulated (phospho-) proteins on 13q14.2 (n = 1) and 13q (n = 3 including the one on 13q14.2) was considerably lower than at the mRNA level. This difference can be attributed to the fact that fewer (phospho-)proteins were analyzed, and post-transcriptional compensatory mechanisms that may affect protein levels. Consistent with the gene expression data, protein levels of many cell cycle-related proteins were significantly increased, including CyclinB1, FoxM1, and CyclinE1. Additionally, we observed upregulation of several pro-apoptotic components, including Caspase-7-cleaved D198, BAK, BIM, and Caspase-3 (Fig. 5b, red circles). Conversely, anti- apoptotic proteins like BCL2 were downregulated (Fig. 5b).
These observations prompted us to perform pathway analysis using the combined mRNA and protein differential expression data. To avoid potential biases introduced by downregulated genes on 13q14.2 due to CN loss that may not reflect true functional differences, we focused on upregulated genes and proteins. By analyzing the top significantly upregulated mRNAs and proteins, using sources including ’MSigDB Hallmark’(45) ’Reactome’(46) and ’Panther’(47), we identified two prominent upregulated pathways: ’Cell cycle’ and ’Apoptosis’ (Fig. 5c). Interestingly, the loss of the 13q14.2 region is associated with up- regulation of the PI3K/AKT/mTOR and p53 pathways.
To explore the potential impact of 13q14.2 loss on the cellular composition in the tumor microenvironment, we estimated the fractions of different cell types within the breast tumor microenvironment. Employing our established pipeline (23), we utilized high-resolution scRNA-seq data (24) as a reference for the breast TME and employed CIBERSORTx as a deconvolution method (22). The scRNA-seq data encompassed a diverse array of cell types, including cancer cells divided into seven recurrent gene modules (GenMod1-7). Each module is enriched in specific biological pathways, such as Jun, Fos, ER, RTKs, and TP53 for GenMod1; Myc and OxPhos for GenMod2; and EMT, IFN, and Complement for GenMod3, among others (24). Our analysis revealed that loss of 13q14.2 resulted in a higher fraction of GenMod4, characterized by proliferating cancer cells (Fig. 5d). This finding aligns with our DEA, which indicated that loss of 13q14.2 led to enrichment of cell cycle-related proteins. Additionally, we observed a significant decrease in the fractions of healthy epithelial and endothelial cells, such as mature luminal cells and myoepithelial cells, associated with 13q loss, consistent with the presence of fewer differentiated cells (Fig. 5d). Furthermore, 13q14.2 loss led to a higher fraction of natural killer (NK) cells and macrophages in METABRIC cohort but not in the TCGA cohort (Fig. 5d). Analysis of the cell fractions enumerated from spatial omics data in the METABRIC cohort confirmed a positive association between 13q14.2 loss and an increased number of macrophages (Supplementary Fig.S2).
These findings collectively suggest that 13q14.2 loss contributes to alterations in the composition of the tumor microenvironment, favoring the proliferation of cancer cells.
Loss of 13q14.2 sensitizes breast cancer cells to BCL2 inhibitors. Following a Sanger Institute pipeline (26, 48), we previously applied a machine learning approach to a pan-cancer drug screen dataset to identify mutations and focal or chromo- some arm-level CNAs associated with increased drug sensitivity or resistance (27). Here, we leveraged an expanded version of this dataset to apply this approach on a breast cancer-specific subset of these data, including 30,394 half-maximum inhibitory concentration (IC50) values from 542 unique drugs across 52 BCa cell lines (28). Notably, using this unbiased approach, we identified a novel CNA-drug association with robust impact: loss of 13q14.2 strongly associates with increased sensitivity of BCa cells to BCL2-targeting drugs (Fig. 6a, blue circles). This association was also identified by the Sanger Institute analysis (Fig. 6a, orange triangle) using a prior version of the dataset, albeit at a lower statistical significance level (27).
We note that our analysis also identified genomic amplification of the HER2 gene, located on chromosome 17q12, as strongly associated with sensitivity to HER2-targeting drugs in BCa cells (Fig. 6a, yellow squares). Identification of this well-established interaction (49) thus validating our approach.
To delve deeper into the association between 13q14.2 loss and BCL2 inhibitors, we investigated which drugs exhibited the most substantial effect sizes and increased sensitivity to 13q14.2 loss in the BCa cell lines across 673 tested drugs. Intriguingly, BCL2 inhibitors, including venetoclax, ABT737, and navitoclax, dis- played the most significant associations with increased sensitivity in BCa cell lines with 13q14.2 loss (Fig. 6b, green circles). However, we noted that 13q14.2 loss did not statistically significantly associate with other BCL2 inhibitors. In addition, 13q14.2 loss did not statistically significantly associate with resistance to any of the 673 drugs (Fig. 6b).
Classifying the BCa cell lines into two groups, those with loss of 13q14.2 and those wild type (diploid) for this region, we then compared their sensitivities to these 7 BCL2 inhibitors. This showed that loss of 13q14.2 resulted in significantly lower IC50 values (indicating greater sensitivity) for venetoclax, navitoclax, ABT-737 and WEHI-539, including in two independent screens for the former two (Fig. 6c-i), but not for and MIM1, sabutoclax, TW37, AZD5991 and UMI-77 (Supplementary Fig.S3a- f). We observed a marginally increased sensitivity for obatoclax in one screen (Fig. 6i) but not in another screen (Supplementary Fig.S3c).
To further characterize the sensitivities of BCa cell lines with 13q14.2 loss to BCL2 inhibitors, we determined the effect size magnitude of each inhibitor. Venetoclax exhibited the largest effect size with the lowest p value < 0.00001, followed by navitoclax and ABT-737 with p value < 0.0001. WEHI-539 and obatoclax ranked third with pvalues < 0.01 and < 0.05, respectively (Fig. 6k, left). Notably, comparison of the selective inhibitory ef- fects of each drug towards different BCL2 family member pro- teins strongly suggests that these increased sensitivities are primarily, if not exclusively, due to inhibition of BCL2 protein, rather than other BCL2 family member proteins (Fig. 6k-right).
To validate the increased sensitivity of BCa cells with 13q14.2 loss to BCL2 inhibitors in vitro, we treated a panel of 10 BCa cell lines with and without 13q14.2 loss venetoclax, the most specific inhibitor of BCL2. The presence or absence of 13q14.2 loss and its approximate size within the genome was initially confirmed by SNP6 microarray (Supplementary Fig.S4b-h) and shallow sequencing (Supplementary Fig.S5). Subsequently, the cell lines were treated with serial log10-fold concentration dilutions of venetoclax, and their viability was measured after 72 hours. Importantly, cells with 13q14.2 loss exhibited significant sensitivity to venetoclax compared to their wildtype counterparts (Fig. 6l and Supplementary FigS3g-p) (One-tailed Mann-Whitney u-test, p value = 0.025). It is worth noting that the response to venetoclax varied across the cell lines harboring a loss of 13q14.2 with EVSA-2 being the most sensitive and HCC2218 the least sensitive to venetoclax (Fig. 6l). The variation was also seen in the cell lines with diploid or amplified 13q14.2 with MFM223 the most resistant and BT20 the least resistant cell lines to venetoclax (Fig. 6l).
To evaluate whether the presence or absence of BCL2 protein affects the association between loss of the 13q14.2 region and sensitivity to venetoclax in cell lines, we measured the protein levels of BCL2. However, our data suggest that the relationship between 13q14.2 deletion and venetoclax sensitivity appears to be independent of BCL2 expression (Supplementary Fig.S3q).
To investigate whether amplification of this region influences the response to BCL2 inhibitors, we employed two BCa cell lines: PMC42-ET, possessing two copies of chromosome 13, and PMC42-LA, derived from PMC42-ET with three copies of chromosome 13 (37) (Supplementary Fig.S6a-b). We evaluated the chemotherapeutic sensitivities of PMC42-ET and PMC42-LA to Venetoclax (ABT-199) and two pan-specific Bcl-2/Bcl-XL inhibitors, ABT737 and Navitoclax (ABT263). The IC50 values for each drug were determined by treating the cell lines with serial 10-fold dilutions of the drugs and measuring their viability after 72 hours. Our drug assays consistently showed that PMC42-LA cells, with an extra copy of chromosome 13, exhibited significantly higher resistance to all three drugs compared to PMC42- ET (Fig. 6m). The IC50 values for ABT199 were 2.60 µM and 9.12 µM for parental PMC42-ET and derivative PMC42-LA, respectively; for ABT737, the values were 0.83 µM and 1.23 µM; and for ABT263, the values were 7.46 µM and 14.36 µM (Supplementary Fig.S6c-e). Consequently, our observation that cells harboring an extra copy of chromosome 13 exhibit increased resistance to three BCL2 inhibitors aligns with our findings that the loss of the 13q14.2 region enhances the sensitivity of BCa cell lines to these drugs.
Considering that patient-derived xenograft models more faith- fully replicate the molecular characteristics and tumor environment of patients, we expanded our analysis from cell lines to data obtained from a biobank of BCa patient-derived xenograft explants (50). We aimed to determine the effect of 13q14.2 loss on the sensitivity of patient-derived xenografts (PDXs) to navitoclax compared to their wildtype counterparts. Notably, the in- creased sensitivity to the drug resulting from 13q14.2 loss was also evident in these PDX models (Fig. 6n), providing further sup- port for the significant role of this CNA in the treatment response to BCL2 inhibitors.
These data collectively suggest that loss of 13q14.2 enhances the therapeutic response to BCL2 inhibitors. Consequently, loss of 13q14.2 could potentially serve as a predictive biomarker for the efficacy of BCL2 inhibitors in treating BCa.