1q21.3 amplification modulates therapy sensitivity in Triple Negative Breast Cancer

The contrast in therapy sensitivity and response across triple negative breast cancer (TNBC) patients suggest underlying genotypic heterogeneity. Using publicly available data, we found significant associations between DNA-level copy number alterations of 1q21.3 locus and therapy sensitivity. We show that in spite of their aggressive nature, 1q21.3 amplified tumors are more responsive to commonly used cytotoxic therapies, highlighting the relevance of 1q21.3 copy number status as a genetic marker for risk stratification, therapy selection and response.


INTRODUCTION
Breast cancer (BRCa) is the most common malignancy among women in the US and is expected to affect 1 in 8 (13%) American women in their lifetime. Despite the availability of effective treatments and an advanced understanding of the disease biology, unaccountable differences in the therapy response across patients continue to exist. The triple negative (TNBC) subtype is the most aggressive form of BRCa, and is characterized by the lack of estrogen receptor (ER), progesterone receptor (PR), and HER2 receptor expression. It is generally assumed to carry a high likelihood of recurrence and rapid progression. Nevertheless, TNBC has long been considered a heterogeneous disease due to its diversity in genotype and therapy response as a subset of patients are very responsive to chemotherapy (1). Studies have demonstrated that~40% of TNBC patients undergo pathological complete response (pCR) when treated with neoadjuvant chemotherapy and these patients have improved overall and event-free survival (2). This contrast in therapy response supports the possibility of underlying genetic modulators among TNBC patients. Thus, there is a great need for novel molecular markers for an evidence-based disease management of TNBC (3). Using publicly available data, we uncovered genotypic characteristics modulating the therapy outcomes in TNBC.
Genomic alterations are generally a consequence of replication stress (DRS) and xenobiotics like hydroxyurea and cisplatin may act as a trigger for such events (4,5). It is also known that some of the standard chemotherapeutic agents like doxorubicin can set off endoduplication events in cancer cells (6). However, several studies have also demonstrated the effect of copy number alterations on therapeutic efficacy and survival outcomes (7). Although there is increasing evidence suggesting the associations between xenobiotic efficacy (particularly cytotoxic agents) and genomic instability, it is not clear why particular alterations are much more common than others and how they modulate cellular dynamics. Furthermore, understanding the effect of cytogenetic abnormalities on therapy response and vice-versa would help understand the underlying biology of drug efficacy and help in the development of improved treatments.
As one of the largest chromosomes with~3000 genes, chromosome 1 is often genetically altered.
Some of these alterations are already linked to numerous medical conditions like cancer, diabetes and schizophrenia (8,9). Despite the wealth of information regarding these alterations, the context and time dependent changes in these alterations have been less explored in cancer. In particular, 1q amplification is a very common alteration in prostate (neuroendocrine and exocrine), breast, pancreas, and biliary system cancers. In BRCa, the alteration frequency is around 8-20% and the frequency is even higher (30-40%) in TNBC (10). Interestingly, 1q amplification is often termed as "regional amplification" due to its discontinuous amplification across the chromosomal arm or even within a particular locus of the arm (11).
Numerous studies have demonstrated the association between high expression of individual 1q genes like MUC1, S100 & SPRR family of genes with disease progression and therapy resistance (12). However, the major determinant of mRNA expression of most genes is inherently their copy number status (13). Therefore, elucidating the role played by DNA copy number changes in addition to mRNA expression can provide a more complete view. Although previous studies indicate that gain of DNA copy number of 1q genes is associated with poor prognostic outcomes, here we show that 1q21.3 amplification imparts selective sensitivity to cytotoxic therapies.

Characterization of 1q21.3 amplified TNBC tumors using TCGA and METABRIC datasets
We used TCGA and METABRIC TNBC copy number alteration data to study the association between 1q21.3 amplification and survival of patients who received both chemotherapy and radiotherapy. Due to regional amplification in the locus, we used a set of three representative/proxy genes (S100A8, HORMAD1, and VPS72) for analysis. The genes were selected based on their chromosomal location, amplification incidence and the correlation between the mRNA levels and their DNA-level copy number (CN). Used in combination, the mean mRNA level of these genes is fairly associated with CN status (figure 1a). High incidence frequency of amplification events in the representative genes was observed in Basal (35%), Luminal (A = 28%, B = 22%) and HER2 (16%) molecular subtypes obtained by PAM50 + low claudin classifier (14) (figure 1b).
TNBC patients with amplification events in either of the representative genes exhibited significantly better overall and progression-free survival. Larger significant differences in survival were observed after stage homogenization (using only Stage II) (figure 2). Similar results were observed when the analysis was repeated using basal and low-claudin samples as defined by the PAM50 classification scheme (14). Pathway analysis using gProfiler on differentially over-expressed genes in amplified samples showed overrepresentation of mitotic To further ascertain the role of this locus alteration in disease progression and prognosis, we evaluated the distribution of positive lymph node counts stratified by CN status across the treated samples using the METABRIC dataset. We observed that amplified samples have considerably lower lymph node involvement (p = 0.053) relative to unamplified samples (figure 1e).  figure 3a). Further, using GSE110153 we observed up-regulation of mitotic pathways in docetaxel sensitive xenografts models relative to the resistant models. We also noticed over-expression of numerous 1q genes in sensitive models, suggesting 1q copy number gain/amplification (figure 3b).

1q21.3 amplified tumors show differential immune cell enrichment
We used the CIBERSORT to perform immunophenotyping using METABRIC RNA-seq data on samples stratified by both CN and treatment status (16). We observed several differences in the immunological landscape across the cohorts. In particular, higher monocyte and M1 macrophage content and lower M2 macrophage content were observed in treated, amplified samples ( figure   4a).
Although M1 macrophage infiltration was shown to promote anti-tumor activity, a recent study of trastuzumab treated metastatic HER2+ breast cancer tumors demonstrated that higher infiltration of M1 macrophages and CD8+ T-cells is significantly associated with improved overall survival, with longer progression free state after response to treatment (17,18). To confirm this finding, we performed immunophenotyping using gene expression data of matched pre-chemotherapy and post-chemotherapy breast cancer samples (NCBI GEO accession: GSE28844) that were either treated with anthracyclines or taxanes. We observed higher fractions of monocytes in post-chemotherapy samples and M1 macrophages in pre-chemotherapy samples with higher Miller and Payne grades (grades 4 and 5).
In previous studies, it was shown that chemotherapy associated elevations in blood monocytes confers a survival advantage (19). However, we propose that increases in monocytes with higher infiltration of M1 macrophages during/after chemotherapy can elicit robust anti-tumor responses.
Irrespective of treatment status, amplified samples demonstrated significantly higher infiltration of T-follicular helper cells ( Tfh) . Tfh cells are shown to intensify the humoral response and to promote anti-tumor activity. Especially in aggressive breast cancer subtypes like TNBC and HER2+/HR-, expression of Tfh signatures shows an almost linear relationship with survival (20). ). There were no significant differences in stage distribution, cellularity and age across the groups. Almost all samples in the C1 cluster were ER- (Table 1). Approximately 70% of C1 cluster samples were basal subtype compared to 45% in C2 and 25% in C3. 87% of the C1 samples possessed TP53 mutations compared to 82% of C2 and 66% of C3. All the samples in the C1 cluster possessed 1q21.3 amplification, while a negligible (<7%) number of C2 and C3 samples had these alterations. 8q alterations were predominantly observed in the C2 group (~70%) and to a lesser extent in the C1 group (~45%) (figure 4c). We observed that a considerable fraction (~50%) of the C1 samples also bore 8q23 and 8q24 amplification events in addition to 1q21.3 amplification (q < 0.05). The enrichment of over-expressed genes in the C1 cluster showed up-regulation of mitotic pathways like activation of NIMA kinases and polo-kinase mediated events.

Concurrence of amplification events between 1q21.3 and neighboring loci
To evaluate any concurrence of amplification events between 1q21.3 and other (1q or 8q) loci, we performed co-occurrence analysis on chromosomal locus level alterations of 173 treated patients (without regard to histological/molecular subtype). We observed a significant concurrence of amplification events in 1q21.3 adjacent loci (adj-p < 0.05) (figure 4d). None of the 8q loci seemed to show significant amplification concurrence with 1q21.3 locus after p-value correction. These results were reproducibly observed when using the entire METABRIC cohort (n = 2105), irrespective of treatment and subtype status.

1q21.3 amplified breast cancer cell line models show higher sensitivity to anthracyclines
We used NCI60 data of all five primary BRCa cell lines to assess the correlation between copy number values of the 1q21. higher chemotherapeutic responses of this cell line (21,22). We used CCLE 2019 copy number data to identify and stratify the ER-/HER-cell lines with respect to 1q and 8q copy number statuses. Further, we used Genomics of Drug Sensitivity in Cancer (GDSC) dataset to ascertain the sensitivities of 1q gain/amplification (n =2), 1q+8q gain/amplification (n=2) and without 1q amplification TNBC cell lines (n = 3) across ten cytotoxic drugs (Table 3). Although we noticed high sensitivity of 1q amplified/gained cell lines for most of the drugs, significant differences (p < 0.05) in sensitivity were observed with doxorubicin, cisplatin, and 5-fluorouracil (figure 5c).

Chemo-radiotherapy treated samples have higher incidence of 1q21.3 amplification
We observed a two-fold increase in the 1q21.3 amplification incidence in treated (36%; n = 100) relative to untreated METABRIC TNBC samples (17%; n = 48). We hypothesize that this difference in amplification rates may be due to treatment induced chromosomal breaks and its association with resulting amplicons. However, it should be noted that no differences in amplification rates were observed between TCGA samples with or without prior cancers.
Moreover, we used the MSKCC 2017 targeted sequencing BRCa data to evaluate the incidence of amplification events across genes belonging to loci 1q21.2, 1q22 & 1q23.1 between primary (n = 481) and metastatic tumors (n = 783) and no significant differences in the amplification rates were observed across metastatic and non-metastatic primary tumors. We used genes in 1q21.3 adjacent loci as none of the genes in the 1q21.3 locus were profiled for copy number alterations in this study and also our earlier analysis showed significant concurrence of amplification events between the 1q21.3 locus and the neighboring loci.
Although there are isolated studies showing evidence of a strong association between chemotherapy and genomic instability, only a few studies highlight the association between chromosomal instability, 1q21.3 alterations, and therapy response (23). To determine whether chromosomal breaks follow a non-random pattern, we used HumCFS database to identify most common chromosomal fragile sites (24). We observed that the highest number of fragile sites are present on Chromosome 1 and 2 (13 sites each). In particular, the FRA1F fragile site induced by aphidicolin, an inhibitor of DNA replication, spans most of the 1q21 region (including 1q21.3 locus).
It is already known that aphidicolin acts synergistically to improve the sensitivity and cytotoxic potential of purine analogs and platinum compounds (25). It is also established that entinostat, a HDAC inhibitor, when combined with low-dose doxorubicin with or without All Trans Retinoic
Most of the "CN high" subtype samples have shared membership with "Mitotic" mRNA subtype and are characterized by highest transcriptional activity in addition to increased dysregulation of the cell cycle pathways (27

DISCUSSION
It is long known that genotypic characteristics of the cell inherently possess the potential to modulate the therapy response. Our results exhibit that 1q21.3 amplified tumors possess increased cell cycle dysregulation alongside apparent mitogenic phenotype and thus show improved sensitivity to cytotoxic therapies. We also observed intensified pro-inflammatory tumor response in these tumors.
Over-expression of mitogens, cyclins and cell cycle regulators indicate the high proliferative potential of untreated 1q21.3 amplified tumors. In particular, the increased expression of FOXM1, CDK, CCNB1 and CCNE proteins in these samples further points to the high mitotic index of these tumors. In treated samples, however, high mRNA expression of AURORA and BUB1 suggests mitotic checkpoint arrest. Mitotic catastrophe is a well-established phenomenon whereby a cell-when exposed to anti-cancer treatments-undergoes apoptotic cell death after aberrant mitosis. Improved sensitivity of cytotoxic agents towards amplified tumors may be due to enhanced DNA fragmentation ability.
Our results exhibit the contrariety of 1q21.3 amplified tumors: chemo-radiosensitivity and high aggressive potential. We hypothesize that amplified tumors demonstrate better histological response to standard cytotoxic agents and likely achieve remission. However, in case of late diagnosis or deferred treatment, the risk of rapid progression and invasiveness may be very high.
Based on our cell line drug sensitivity results, we assume that 1q21.3 amplified tumors are more responsive to anthracyclines. It is shown in earlier studies that neoadjuvant anthracyclines are more effective in basal and HER2+ subtypes for achieving pCR (29). Interestingly, these subtypes have the highest incidence of 1q21.3 amplification events and therefore, we postulate that patients with amplified tumors are most likely to achieve pCR with neoadjuvant/preoperative chemotherapy. Our analysis involving tumors/cell lines without amplification events hint higher risk of treatment refractoriness with the use of anthracyclines.
Taxanes may be useful in the optimal management of these tumors due to their negligible differences in their sensitivity across the amplified and unamplified cell lines. Moreover, in current clinical practice, taxanes are generally preferred over anthracyclines in the treatment of refractory and advanced cases (30). Our results also indicate that HDAC inhibitor (HDACi) entinostat like its cousin vorinostat, facilitates chromosomal breaks while preventing DNA repair and thus sensitizing tumor cells to cytotoxic agents (31). This action likely enhances the susceptibility of unamplified tumors to DNA-damaging agents. However, their potential use in the treatment of refractory tumors is yet to be ascertained.
Although other BRCa subtypes like luminal cancers showed high incident rates of 1q21.3 amplification, there were no considerable prognostic differences observed with respect to the 1q21.3 CN status across these subtypes. On the contrary, the shared prognostic and molecular features of 1q21.3 amplified tumors across TNBC, HGSOV and serous UCEC suggest the possibility of a unified treatment approach.

Data retrieval
We

TCGA and METABRIC BRCa analyses
Only primary tumors were used for the analysis. Samples with missing clinical information regarding HR/HER receptor status, treatment status, sample type and neoadjuvant therapy status were discarded prior to the stratification. Only treatment naive TCGA TNBC primary samples were considered for the analysis. TNBC samples were identified by HR and HER receptor negativity provided in the clinical information.
Unless specified, the classification of copy number alterations are based on widely accepted copy number value thresholds where deep/homozygous deletion (-2), arm-level/shallow/heterozygous deletion (-1), diploid (0), arm-level/gain (1) and amplification (2). We focused most of our analyses on amplification (focal/high level gain of gene copies) rather than copy number gain as their contribution can be determined clearly. Nevertheless, we used arm-level alterations sporadically to evaluate the consistency of the results.
We used cbioportal ( www.cbioportal.org ) to retrieve significantly over-expressed genes (P < Immunophenotyping was performed on previously normalized METABRIC data using CIBERSORT tool with LM22 signatures and 500 permutations. Quantile normalization was disabled for all the analyses. We only considered samples with CIBERSORT p-value < 0.05 for further analysis. We profiled the immune cell content across samples grouped by treatment and 1q amplification status. Kruskal Wallis test (p < 0.05) was used to evaluate the significance of differential immune cell enrichment.
The ggplot2 , ggpubr, survminer and factoextra R packages were used to make boxplots, violin plots and barplots. Survival plots were made using cbioportal's in-built custom analysis feature.
Network and chromosome distribution plots of differential genes were made using ShinyGO web browser (32). GO based enrichment analysis using GO biological processes was performed using FUNSET web browser (33).

Co-occurrence analysis
Pairwise co-occurrence analysis on METABRIC 1q and 8q copy number data of treated samples (irrespective of the histological subtype) (n = 173) was carried out using the DISCOVER python package (34) . The gene-level copy number alterations were transformed to locus-level alteration by using the median copy number values across the locus. Arm-level copy number alterations were not considered for this analysis. An independent analysis of similar scope was also performed using the entire METABRIC cohort (n = 2105). FDR corrected p-values (q value < 0.05) of pairwise comparisons were used to infer concurrence of amplification events.

Hierarchical clustering analysis
We used the gene-level processed copy number values of all 173 chemo-radiotherapy treated METABRIC samples (irrespective of their histological subtype) for the analysis. Hierarchical clustering was performed using hclust R base function with complete linkage. For all the three analyses involving all, TNBC and HER2 treated samples, we used the root node to separate the C1 cluster from C2 and C3. We seperated C2 and C3 using the "all" treated cohort by making a cut at level 3 of the tree. We made a cut at level 4 to separate clusters C2 and C3 of "TNBC" clusters. C2 and C3 clusters of HER2 treated samples were obtained with a cut at level 1 of the tree.

NCI60 and GDSC Cell line analyses
Spearman correlation was performed between gene level copy number values (1q and 8q representative genes to identify amplification events across the cell lines.

Gene expression datasets
We used CIBERSORT tool for immunophenotyping analysis on GSE28844 gene expression data to identify differential immune cell enrichment in pre and post chemotherapy treated samples (n = 56) with higher Miller and Payne grades (grade 4 & 5). We used Kruskal Wallis test (p < 0.05) for assessing the significance of differential enrichment.
To establish the association between up-regulation of mitotic pathways and treatment sensitivity,     a) Network plot illustrating upregulated GO biological processes in pCR achieved cohort (GSE25066). The circumference of the node represents the size of the gene sets. The color intensity of the node represents their enrichment significance. The thickness of the edges represents the extent of the overlapping gene sets. c) Schematic representation of amplification incidence of 1q and 8q loci across all (n = 173), TNBC only (n = 100) and HER2 only (n = 42) treated samples. Differential analysis of the cohorts were done using the cbioportal custom analysis feature. Significant gene level amplification incidence rates (q < 0.05) of 1q and 8q genes are transformed to locus-level incidence using the median incidence across the genes of their corresponding locus. Only loci with incidence proportion of > 0.70 in the clusters were included for the heatmap.  b) Principal component plot illustrating the association between 1q and 8q copy number status and 11 chemotherapeutic agents (NCI60 data). The spearman correlation values between screened compounds and copy number values (p < 0.10) are used to perform principal component analysis. Strong associations between 1q21.3 copy number status and sensitivity of fluorouracil, gemcitabine, anthracyclines, alkaloids, fulvestrant and chlorambucil can be discerned.

Figure 6
Chromosomal representation of significantly overexpressed genes (Adjp < 0.05) in Doxorubicin + Entinostat treated group compared to Doxorubicin treated group in MDA-MB-231 cells.