Home Browse Integrated analysis of -omic landscapes in breast cancer subtypes
ALL Metrics
-
Views
-
Downloads
Get PDF
Get XML
Cite
Export
Track
▬
✚
Research Article

Integrated analysis of -omic landscapes in breast cancer subtypes

[version 1; peer review: 2 approved with reservations]
Suren Davitavyan
https://orcid.org/0000-0002-3414-5553
1,2, Gevorg Martirosyan3, Gohar Mkrtchyan3, [...] Andranik Chavushyan3, Ani Melkonyan3, Hovsep Ghazaryan
https://orcid.org/0000-0001-5612-8933
3, Hans Binder
https://orcid.org/0000-0002-2242-4678
4,5, Arsen Arakelyan
https://orcid.org/0000-0002-6851-1056
1,2
Suren Davitavyan
https://orcid.org/0000-0002-3414-5553
1,2, Gevorg Martirosyan3, [...] Gohar Mkrtchyan3, Andranik Chavushyan3, Ani Melkonyan3, Hovsep Ghazaryan
https://orcid.org/0000-0001-5612-8933
3, Hans Binder
https://orcid.org/0000-0002-2242-4678
4,5, Arsen Arakelyan
https://orcid.org/0000-0002-6851-1056
1,2
PUBLISHED 03 Jun 2024
Author details Author details
OPEN PEER REVIEW
REVIEWER STATUS

Abstract

The subtypes of breast cancer exhibit diverse histology, molecular features, therapeutic response, aggressiveness, and patient outcomes. Multi-omics high-throughput technologies, which are widely used in cancer research, generated waste amounts of multimodal omic datasets calling for new approaches of integrated analyses to uncover patterns of transcriptomic, genomic, and epigenetic changes in breast cancer subtypes and connect them to disease clinical characteristics.

Here, we applied multi-layer self-organizing map (ml-SOM) algorithms to PAM50-classified TCGA breast cancer samples to disentangle the diversity of the effects of gene expression, methylation, copy number, and somatic single nucleotide variation in the disease subtypes. Furthermore, we studied the association of perturbed gene modules with survival, prognosis, and other clinical characteristics.

Our findings highlight the power of multi-omic analyses to offer a better understanding of the molecular diversity of breast cancer subtypes compared to single-omic analyses. Moreover, they highlight the complex subtype-characteristic associations between gene expression and epigenetic/genomic factors and their implications for survival and clinical outcomes.

Keywords

Breast cancer, multi-omics study, self-organizing maps

Corresponding authors: Suren Davitavyan, Arsen Arakelyan
Competing interests: No competing interests were disclosed.
Grant information: This study was funded by a research grant from the Committee of Higher Education and Science of the Ministry of Education and Science of the Republic of Armenia (21AG-1F021, PI: AA).
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Copyright:  © 2024 Davitavyan S et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
How to cite: Davitavyan S, Martirosyan G, Mkrtchyan G et al. Integrated analysis of -omic landscapes in breast cancer subtypes [version 1; peer review: 2 approved with reservations]. F1000Research 2024, 13:564 (https://doi.org/10.12688/f1000research.148778.1) First published: 03 Jun 2024, 13:564 (https://doi.org/10.12688/f1000research.148778.1) Latest published: 03 Jun 2024, 13:564 (https://doi.org/10.12688/f1000research.148778.1)

Introduction

Breast cancer is the most common cancer worldwide, with 7.8 million women alive as of the end of 2020 who had received a diagnosis within the previous five years. It represents a diverse group of cancers of mammary glands characterized by considerable heterogeneity of morphological, genomic, epigenetic, transcriptomic, and proteomic features that drive tumor progression, treatment resistance, and aggressiveness.1–4

The current molecular classification of breast cancer includes stratification by expressed markers (ER, PR, HER2).5 Recently, a new gene expression-based molecular subtyping of breast cancers has been proposed (PAM50).6 These classification approaches are linked to treatment selection and prognosis. Other classification approaches including TNM staging,7 MammaPrint,8 and Oncotype DX9 were used for breast cancer subtyping. There are also new approaches that build on these clinical classifications and use various mathematical algorithms to improve their accuracy, such as PCA-PAM50 which combines the breast cancer clinical annotation of PAM50 and improves the accuracy of the approach by integrating with the primary component analysis.10 This underscores the significance of employing interdisciplinary methodologies.

The studies into molecular characteristics of BC subtypes have shown variability of subtype-associated biological and pathophysiological processes. Multiple studies showed considerable differences in the gene expression patterns in cancer subtypes with implications for survival and prognosis (for example reviewed in Refs. 11, 12). Our previous study also provided insight into the variability of transcriptomic landscape in cancers with somatic and germline mutations in BRCA1 and BRCA2 genes suggesting that uncovering the molecular heterogeneity of breast cancers can provide clues for developing more efficient therapeutic approaches.13

The advances in -omics technologies and efforts in collecting large multi-omic datasets have opened new opportunities for integrated analyses to deepen our understanding of molecular features of breast cancers, the mechanisms that drive the molecular diversity, and fine-graining disease molecular subtypes.14 This, in turn, paved the way for developing approaches for informed treatment selection.15

Several studies have already explored this direction.16 The study based on CNAs, mRNA, miRNA abundance, methylation, and protein abundance data already shows efficiency in dividing breast cancer into invasive lobular carcinoma (ILC) and invasive ductal carcinoma (IDC) subtypes.17 The novel integrated multi-omics approaches led to the identification of a new hybrid subtype with a poor survival prognosis.18 Also, survival and drug response-based frameworks show high effectiveness for diagnosing cancer patients in comparison with single-omics approaches.19 These and other results underline the importance of integrative studies that can open a new dimension in the analysis of molecular heterogeneity of cancers, address the complexity of the interplay of omic features in different subtypes of disease by linking genetic defects, epigenetics reprogramming, and perturbed transcription factor networks.

Previously, we have developed a self-organizing maps (SOM)-based bioinformatics pipeline for analysis, functional characterization, and visualization of omic datasets.20,21 It was successfully applied in many studies, including molecular characterization of various cancers, such as low-grade gliomas,22 B-cell lymphomas,23 etc. In this paper, we used its multilayer-SOM approach to analyze the genomic, epigenetic, and transcriptomic features of The Cancer Genome Atlas Breast Invasive Carcinoma (TCGA-BRCA) data collection.

Methods

Study datasets

In this study, we used available omic datasets of the TCGA-BRCA project.24 Total, RNA-seq counts, microarray promoter and gene body methylation, microarray CNV, and SNV were obtained for 996 samples. PAM50 molecular classification data for these samples were obtained from the publication by Chia et al.6

The patient’s clinical data was retrieved from the GDC database and contains variables, such as tumor pathologic stage, information about treatment, survival, etc.

Data preprocessing

In the GDC Data Portal, raw star count files are available for transcriptomic data for each sample within the TCGA-BRCA dataset. We have downloaded them by using gdc-client25 algorithms and merged all files across samples. Accordingly, the TPM values for each sample were obtained. In the succeeding step of the analysis, they were subjected to normalization and converted to log counts using variant stabilization transformation from the DeSeq2 package.26

The GDC-supplied methylation data contains betta values for samples, in which a subset is calculated with Illumina human methylation 450 microarray, while others with Illumina human methylation 27 microarray. DNA methylation data was aggregated by merging the data generated with two microarrays. In the case of overlapping CpG islands, the mean value per gene was used. Promoter methylation data was converted from betta to m values.27

CNV data was obtained from the GDC Data Portal, which provides Gene Level Copy Number files generated through the genotyping array method using the Affymetrix SNP 6.0 platform. To avoid constant values in data, we have normalized CNV data by adding small numbers (mean=0, sd=0.001) to each sample on a gene-wise basis.

Somatic mutation data resides within GDC in maf files and has been used for generating an SNV matrix. We have summarized all single nucleotide variation counts by genes. To avoid constant values in data, we have normalized SNV data by adding small numbers (mean=0, sd=0.001) to each sample gene-wise.

Integrated analysis of cancer molecular features with multi-layer SOM

To conduct an integrative analysis of omic datasets of breast cancer, we employed a refined multilayer self-organizing maps (ml-SOM) approach built on our previous developments.22,28 The self-organizing map (SOM) algorithm is a neural network-based technique for dimensionality reduction and clustering. SOM topology is driven by co-variance of gene expression according to the selected weight factors.

Furthermore, the SOM implementation in the oposSOM package is enriched with robust function mining capabilities, facilitating the assignment of biological functions to gene clusters. This capability efficiently reduces the high-dimensional gene space into numerous differentially expressed functional modules. As a result, this approach enables a transition from analyzing individual genes to conducting systems-level analyses while preserving the integrity of the original information.

For ml-SOM, we organized all omic datasets into four distinct layers and trained them collectively on a single SOM grid, similar to a classical single-layer SOM (sl-SOM).20 The key distinction between the training processes of sl-SOM and ml-SOM lies in how the best matching unit (BMU) is selected within the SOM grid.

In sl-SOM, the BMU is chosen based on the distance between the input vectors and the weight vectors of SOM nodes. However, in the case of ml-SOM, we calculate these distances separately for each layer and then combine them into a single value, taking into account the respective layer weights as follows:

D=∑inωi∗di,
where n - number of layers, ωi - weight if ith layer, di - distance to the SOM node on ith layer.

The weight factor ω scales the effect of each of the layers on the topology of the ml-SOM. In this particular analysis, we used the following weights: 1 for RNA-seq, 0 - for promoter methylation, 0 - for CNV, and 0 - for SNV. In this way, the arrangement of genes on the SOM grid is driven by the transcriptome layer.

The downstream analysis of ml-SOM is similar to the oposSOM pipeline.20 Following the Self-Organizing Map (SOM) training, we partitioned the resulting metagene map into discrete regions referred to as “spots”. These spots represent clusters of genes that exhibit co-expression patterns, particularly genes that are perturbed in at least one of the omic layers. Spot identification within the individual SOM portraits was based on a “k-means” and “variance” criterion.21 These spots were subsequently combined into an overexpression summary map, offering a comprehensive view of the transcriptomic landscape across all layers. The expression values of each spot, detected across various layers, collectively constituted the corresponding spot profile.

To further elucidate the biological significance of these spots, we subjected the genes associated with the spots to functional annotation, including overrepresentation and Gene Set Z-score analyses. We also employed the Enrichr resource29 for over-representation analysis using additional gene sets.

Analysis of the association between molecular features

To assess the relationships between gene expression, methylation, CNV, and SNV across various spots (gene modules), we utilized linear regression with gene expression as the outcome and the other genomic markers as explanatory variables. The model was further enhanced by incorporating PAM50 subtypes as an interaction term, allowing us to examine the variability in these relationships across different disease subtypes and omic layers. For the statistical analysis of these interactions and their visualization, the ‘emmeans’30 and ‘interactions’31 packages in R were employed. Additionally, we applied the Dunnett’s Test to evaluate pairwise differences in mean levels of expression and methylation, CNV, and SNV gene modules in PAM50 subgroups compared to true normal tissue.

Analysis of survival

Overall survival analysis was performed using the Cox proportional hazards regression using survival (https://cran.r-project.org/web/packages/survival/index.html) and survminer (https://cran.r-project.org/web/packages/survminer/index.html) R packages. The events were defined based on vital status information (“Dead” or “Alive”). The time to event was defined as “days to death” or “days to last follow-up”. We included survival as a dependent variable and spot omic profiles and breast cancer subtype information as predictors. We used the contsurvplot (https://cran.r-project.org/web/packages/contsurvplot/index.html) R package to visualize survival curves.

Analysis of the association between SOM clusters and clinical characteristics

Previously we introduced so-called SOM phenotype portraits by mapping the association of clinical and phenotypic characteristics on the SOM grid.32 Accompanying phenotype data, such as medication, disease stage, and grade for TCGA-BRCA samples was obtained from the GDC portal. Phenotype maps were created for all available data types. To create a phenotype map, we constructed a linear regression model for each metagene as a dependent variable and clinical parameters as an ordinal variable. Then we mapped the corresponding regression coefficient for the predictor variable to a corresponding position of that metagene on the SOM grid. The visualization of weight coefficients allows evaluation of the association of corresponding clinical characteristics and the levels of functional gene modules on different omic layers.

Results

The multi-omics landscape of breast cancers

In this study, we performed an integrative analysis of the TCGA-BRCA dataset (Figure 1) using four molecular data types – gene expression, promoter methylation, copy number variation, and somatic single nucleotide variation. We organized each omic data type into a data layer and analyzed them with multi-layer SOM training.

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure1.gif

Figure 1. Distribution of TCGA-BRCA samples according to PAM50 transcriptomic subtypes.

Lum A (luminal A) - ER+ and PR+, HER2- and low levels of the Ki-67; Lum B (luminal B) - ER+ and HER2- and has either high levels of Ki-67; HER2-enriched - ER-, PR- and HER2+; Basal triple-negative and CK5+ and CK17+; CLOW - triple-negative and high expression of immunity genes; Normal-like - triple-negative and CK5- and EGFR-; True Normal - normal tissue.

The multi-layer SOM training organized multi-omic high dimensional data on the two-dimensional grid of 40x40 size. During the training phase, ml-SOM combines the genes having similar profiles of expression, methylation, CNV, and SNV across samples into adjacent nodes according to the weight factors on the SOM grid thus forming gene clusters (also referred to as spots or gene modules). These clusters can be visualized for each sample by the expression value, which allows direct comparison between samples. Additionally, the expression variance for each cluster can be calculated and visualized as a map as well.

We combine samples in groups for downstream analysis according to the PAM50 classification to assign cancer samples to molecular subtypes. The average multi-omic SOM portraits showed considerable variations in expression (Gex), methylation (Gmx), CNV, and SNV both across PAM50 subgroups as well as compared to true normal tissue (Figure 2).

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure2.gif

Figure 2. Visualization of SOM group portraits according to PAM50 classification.

The color scheme indicates the nature of the change on each omic layer. On the transcriptome layer (Gex) blue to red scale represents under- to overexpression. On the methylation layer (Gmx) blue to red scale represents hypo- and hypermethylation. On the CNV layer blue to red scale represents CNV loss and CNV gain, respectively. On the SNV layer blue to red scale represents low to high mutation burden, respectively. Consequently, the green areas on SOM portraits represent invariants gene clusters or areas.

Visualizing variance within layers showed regions of gene clusters with high and low variance in brown and blue, respectively (Figure 3A-D). Regions with high variance likely capture significant differences in gene expression (Gex), promoter methylation (Gmx), copy numbers (CNV), and mutational load (SNV), respectively, characterizing the distinct tumor subtypes. We used the k-means33 segmentation of the map criterion to dissect the variance maps into 20 spots annotated by letters “A” to “T” (presented in Supplementary_Figure1.pdf from “Extended Data”, see Data and software availability). For further analyses, we selected highly variable ones (spots A, C, E, F, L, R, Q, S) that contained genes with highly variant feature profiles across samples at least in one omic layer.

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure3.gif

Figure 3. Metagene Variance Maps for gene expression (Gex, A), promoter methylation (Gmx, B), copy number variation (CNV, C), and single nucleotide variants (SNV, D) SOM layers.

These maps offer an overview of the variance patterns across different data types, highlighting the variability of specific gene clusters (spots) across omic layers. The blue regions represent areas of low variance, while red regions indicate high variance. The letters correspond to highly variant spots as identified with k-means clustering (presented in Supplementary_Figure1.pdf from “Extended Data”, see Data and software availability).

On average, each spot comprised 155±60 genes. The largest gene counts (207 genes) were associated with spot R showing high variance for Gex, Gmx, and CNV, while spot F contained only 9 genes mostly associated with high CNV-variance. Due to the self-organizing properties of the SOM algorithm, genes of similar expression profiles were grouped in spot-like clusters across the omic layers, suggesting they may share common biological functions or regulatory mechanisms.21 To explore this further and map functional signatures to the spots, we performed Gene Set Enrichment Analysis using built-in gene sets20 and an over-representation analysis tool that provided additional gene sets covering multiple domains.29

Spot A contained 114 genes, associated primarily with DNA replication (padj = 3.3e-05), E2F targets (padj = 5.3e-09), retinoblastoma pathway (padj = 3.23e-05), and cell cycle activity (padj = 5.4-e04). Notably, spot genes were also associated with EMT markers taken from Sarrió et al,34 as well as markers for the basal BC subtype taken from Smid et al.35

Spot C contained 165 genes mostly involved in protein transport (padj = 2.0e-02), SMARCA2 antiproliferative targets (padj = 2.2e-06),36 and DNA repair. (padj = 4.34E-03).

Spot E contained 118 genes enriched with luminal cancer gene signatures (padj = 0.005)37 and genes associated with the amplification of chromosome 16p13 (padj = 0.02).38

Spot F contained only 9 genes, however, they were implicated in vitamin D signaling (padj = 0.038), palmitoyl-CoA Hydrolase Activity (padj = 0.025), Androgen Receptor/NKX3-1 Signaling (padj = 0.01) and ICGC transcription factor target genes (padj < 0.01).

Spot L contained 67 genes strongly associated with the immune system process (padj = 1.2e-12).

Spot Q contained 141 genes enriched with stromal (padj = 7.84E-03)39 and stem cell gene signatures (padj = 8.81E-14),40 genes involved in accelerated proliferation (padj = 6.1e-05), inflammation (padj = 0.03), RAS signaling (padj = 1.1e-05), hypermethylation of tumor suppressor genes (padj = 3.2e-04).

Spot R contained 207 genes associated with RNA splicing (padj = 0.005) and mitochondrial gene signatures (padj = 0.004).

Spot S contained 106 genes enriched with luminal cancer signatures (padj = 4.3e-06), ESR1 signatures (padj = 3.5e-15), and metastasis-suppressing signatures (padj = 2.1e-03).

The full list of annotations associated with the spots is available in Supplementary Tables S1 - S16 in Supplementary_Tables_S1-S16.xlsx file of “Extended data” (see Data and software availability).

Based on the functional annotation we assigned each spot to the functional processes that best describe the genes located in that module (Table 1).

Table 1. List of selected spots, along with the top 3 genes from each spot and their respective Pearson correlation coefficients.

The Spot Assignment column presents a generalized spot description based on the gene set enrichment results.

SpotTop correlated genes on transcriptome SOM layer (Pearson correlation coefficient, r)Spot Assignment
AKIF2C (r = 0.81), RAD51AP1 (r = 0.76), DSCC1 (r = 0.76)cell cycle, metastasis, EMT
CHLTF (r = 0.83), GIT2 (r = 0.80), ACAP2 (r = 0.80)DNA repair/miRNA targets
EROGDI (r = 0.77), RAB26 (r = 0.75), HAGH (r = 0.75)luminal cancer
FTATDN3 (r = 0.64), THEM4 (r = 0.62), DCAF8 (r = 0.58)VDR signaling
LFERMT3 (r = 0.93), PARVG (r = 0.86), FMNL1 (r = 0.88)immune response
QCAV1 (r = 0.81), TGFBR2 (r = 0.77), RBMS1 (r = 0.75)stroma/stem cells
RSSNA1 (r = 0.85), DRAP1 (r = 0.82), SURF2 (r = 0.82)RNA splicing
SSCUBE2 (r = 0.82), ESR1 (r = 0.82), ABCC8 (r = 0.76)ESR1 signaling

The feature profiles of the spots in the different omic layers represent the “averaged” pattern of the respective genes across the samples and serve as a surrogate level for the functional signatures associated with the respective spot.21 We calculated two-way clustered similarity matrices for the different omic layers using the spot profiles to estimate the relatedness between the tumor subtypes as seen by the different omic layers (Figure 4). Gene expression (Figure 4A) and methylation (Figure 4B) clustering of the breast cancer subtypes showed remarkable similarity indicating marked variance correlations between Gex and Gmx data as seen also in the similar Gex and Gmx variance maps (Figure 3). Both omic layers express roughly three clusters collecting normal-like, basal, and HER2E as well as luminal subtypes, respectively. On the CNV layer, one finds two major clusters of normal tissue, Lum A, Lum B, and HER2E tumors, and of normal-like and basal tumors (Figure 4C). No notable clusters were observed on the SNV layer (Figure 4D).

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure4.gif

Figure 4. Correlation-based clustering of cancer subtypes on group SOM portrait and sample level.

The left part of the figure shows the Pearson correlation-based heatmaps across data categories, where red indicates a positive correlation and blue indicates a negative correlation, respectively. Color intensity reflects the strength of the correlation. The right part displays correlations between individual samples, with colors indicating their respective PAM50 subtype memberships. The edges between samples indicate the Pearson correlation coefficient > 0.5.

The similarity net images in the right part of Figure 4 visualize the correlation of omic profiles on a single tumor level. It indicates the most pronounced clustering in the Gex layer and the weakest one for SNV data with CNV and Gmx taking intermediate positions.

Integrated module/spot analysis across the omics landscape

Next, we analyzed the genes collected from most variant spots across the omic landscapes to identify mutual correlations, possible driver events, and functional associations with underlying cancer-related biological processes. For this, we used linear regression/ANOVA to evaluate the mean differences of spot levels in disease subtypes across the four omic layers (for example see Figure 5A). We further used post hoc Dunnett’s test to assess the differences in mean spot levels in cancer subtypes compared with normal tissues. The actual levels are indicated as boxplots (for example see Figure 5B).

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure5.gif

Figure 5. Multi-omic analysis of EMT and cell cycle gene signature (spot A) in PAM50 cancer subtypes.

A) ANOVA/linear regression coefficient plots indicate the magnitude and significance of differences in mean levels in cancer subtypes compared to normal tissue. B) Boxplots of spot levels in cancer subtypes across SOM layers. C) Visualization of PAM50 subtype-characteristic association (regression slopes and confidence intervals) between expression (dependent variable) and epigenetic (methylation) and genomic (CNV, SNV) factors.

To understand the association between transcriptomic, epigenetic, and genetic features in each cancer subtype we used linear regression with expression spot levels as the dependent variable and other omic layers as well as their interactions with cancer subtypes. In this case, the slope coefficient for each interaction will show the level and direction of association in a subtype-characteristic manner. Thus, we could use this as a proxy for the impact of genomic/and epigenetic elements on the expression of spot-associated genes in cancer subtypes (for example, see Figure 5C). The samples of the CLOW subtype we excluded from further analyses because of their low number (n = 4).

Particularly, we considered cell cycle and EMT (spot A), DNA repair (spot C), luminal cancer signature (Spot E), vitamin D signaling (Spot F), immune response (Spot L), stroma and stem cells (Spot Q), RNA-splicing (Spot R) and estrogen receptor signaling (Spot S) activities (Figures 5-12, respectively).

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure6.gif

Figure 6. Multi-omic analysis of DNA repair gene signature (spot C) in PAM50 cancer subtypes.

A) ANOVA/linear regression coefficient plots indicate the magnitude and significance of differences in mean levels in cancer subtypes compared to normal tissue. B) Boxplots of spot levels in cancer subtypes across SOM layers. C) Visualization of PAM50 subtype-characteristic association (regression slopes and confidence intervals) between expression (dependent variable) and epigenetic (methylation) and genomic (CNV, SNV) factors.

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure7.gif

Figure 7. Multi-omic analysis of luminal cancer gene signature (spot E) in PAM50 cancer subtypes.

A) ANOVA/linear regression coefficient plots indicate the magnitude and significance of differences in mean levels in cancer subtypes compared to normal tissue. B) Boxplots of spot levels in cancer subtypes across SOM layers. C) Visualization of PAM50 subtype-characteristic association (regression slopes and confidence intervals) between expression (dependent variable) and epigenetic (methylation) and genomic (CNV, SNV) factors.

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure8.gif

Figure 8. Multi-omic analysis of VDR signaling gene signature (spot F) in PAM50 cancer subtypes.

A) ANOVA/linear regression coefficient plots indicate the magnitude and significance of differences in mean levels in cancer subtypes compared to normal tissue. B) Boxplots of spot levels in cancer subtypes across SOM layers. C) Visualization of PAM50 subtype-characteristic association (regression slopes and confidence intervals) between expression (dependent variable) and epigenetic (methylation) and genomic (CNV, SNV) factors.

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure9.gif

Figure 9. Multi-omic analysis of immune/inflammatory response gene signature (spot L) in PAM50 cancer subtypes.

A) ANOVA/linear regression coefficient plots indicate the magnitude and significance of differences in mean levels in cancer subtypes compared to normal tissue. B) Boxplots of spot levels in cancer subtypes across SOM layers. C) Visualization of PAM50 subtype-characteristic association (regression slopes and confidence intervals) between expression (dependent variable) and epigenetic (methylation) and genomic (CNV, SNV) factors.

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure10.gif

Figure 10. Multi-omic analysis of stromal/stem cell gene signature (spot Q) in PAM50 cancer subtypes.

A) ANOVA/linear regression coefficient plots indicate the magnitude and significance of differences in mean levels in cancer subtypes compared to normal tissue. B) Boxplots of spot levels in cancer subtypes across SOM layers. C) Visualization of PAM50 subtype-characteristic association (regression slopes and confidence intervals) between expression (dependent variable) and epigenetic (methylation) and genomic (CNV, SNV) factors.

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure11.gif

Figure 11. Multi-omic analysis of RNA splicing gene signature (spot R) in PAM50 cancer subtypes.

A) ANOVA/linear regression coefficient plots indicate the magnitude and significance of differences in mean levels in cancer subtypes compared to normal tissue. B) Boxplots of spot levels in cancer subtypes across SOM layers. C) Visualization of PAM50 subtype-characteristic association (regression slopes and confidence intervals) between expression (dependent variable) and epigenetic (methylation) and genomic (CNV, SNV) factors.

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure12.gif

Figure 12. Multi-omic analysis of ESR1 gene signature (spot S) in PAM50 cancer subtypes.

A) ANOVA/linear regression coefficient plots indicate the magnitude and significance of differences in mean levels in cancer subtypes compared to normal tissue. B) Boxplots of spot levels in cancer subtypes across SOM layers. C) Visualization of PAM50 subtype-characteristic association (regression slopes and confidence intervals) between expression (dependent variable) and epigenetic (methylation) and genomic (CNV, SNV) factors.

Cell cycle and EMT (spot A)

Spot A contained 114 genes (Supplementary Table S17 in Supplementary_Tables_S17-S24.xlsx file of “Extended data”, see Data and software availability) mostly associated with cell cycle and EMT. The Pearson’s correlation coefficient of the spot expression profile and its genes ranged from 0.21 to 0.81. The top correlated genes in the spot were KIF2C, DSCC1, RAD51AP1, DONSON, CDC123, MCM6, CLSPN, A2ML1, DKC1, and SKP2, which previously were implicated in breast cancers.41–47

The expression profiles of this gene module were significantly upregulated in all cancer groups compared to true normal tissue. Furthermore, the expression levels were gradually increased with the highest values in basal cancers compared to all other groups.

On the methylation layer, the Lum A and Lum B profiles were hypermethylated, while the basal cancers showed hypomethylation compared to true normal samples. Meanwhile, no differences were observed in the CNV and SNV layers (Figure 5A and 5B).

Mutual associations of expression with methylation and the genetic features across groups showed that the expression levels of spot A negatively correlate with methylation profiles in all cancer subtypes (except Lum A) being significant in Basal and Normal-like groups, while a significant positive correlation with CNV was observed for Lum A, Lum B, and HER2E tumors. No significant trend was observed for SNV in all cancer subtypes. Interestingly, in the true normal samples, we observed a significant positive correlation between methylation and expression (Figure 5C).

DNA repair (spot C)

DNA repair gene module (spot C) contained 165 genes (Supplementary Table S18 in Supplementary_Tables_S17-S24.xlsx file of “Extended data”, see Data and software availability) with a correlation coefficient in the range of 0.13-0.83. The top correlated genes in this module were HLTF, GIT2, ACAP2, OTUD4, ALG11, IREB2, EEA1, DCAF17, SEC24B, and GCC2.48–54

The expression profile of the spot was significantly downregulated in Lum A, basal and normal-like cancers. Methylation levels of the genes showed a significant decrease in Lum A, Lum B, and HER2E cancers compared with the true normal group. In normal-like cancers, there was also a significant decrease in CNV levels. SNV levels were not significantly different among all studied groups (Figure 6A and 6B).

Analysis of regression trends showed that the expression of spot C was positively correlated with methylation in all cancer groups, however only in Lum A, basal and normal-like groups the trends were significant (padj < 0.05). No association of expression with CNV was observed. A significant positive trend of expression-SNV counts association was observed only for normal-like cancers (Figure 6C).

Luminal cancer signature (Spot E)

The luminal gene signature (spot E) contained 118 genes (Supplementary Table S19 in Supplementary_Tables_S17-S24.xlsx file of “Extended data”, see Data and software availability) with a correlation coefficient in the range of 0.17-0.77, with the top genes being ROGDI, RAB26, HAGH, ZNF688, METRN, NPAS1, OCEL1, HDAC11, RSPH1, SLC22A18.55–60 The expression levels of this module were significantly upregulated in all cancer groups compared to true normal, with the highest values in Lum A and Lum B subtypes. Meanwhile, the methylation of these genes was significantly downregulated in the Lum A, Lum B, and HER2E cancers. The increase in CNV levels was observed only in HER2E cancers, while the SNV profile was decreased in Lum A (Figure 7A and 7B). Methylation was significantly negatively associated with expression in Lum A tumors paralleled with a positive correlation with CNV. No significant associations were recorded for SNVs in all cancer subtypes (Figure 7C).

Vitamin D signaling (Spot F)

Spot F contained only 9 genes (TATDN3, THEM4, DCAF8, DCAF6, QSOX1, SETMAR, PPA2, ATXN10, SNW1,61–65 Supplementary Table S20 in Supplementary_Tables_S17-S24.xlsx file of “Extended data”, see Data and software availability) mainly related to vitamin D signaling processes (Pearson’s correlation coefficient: 0.24-0.64). The gene expression profiles associated with this gene module didn’t change across most cancer subtypes compared with true normals except for downregulation in normal-like cancers. Furthermore, all cancers were characterized by significantly increased methylation levels (except normal-like cancers). Finally, CNV profiles were significantly downregulated in normal-like cancers. No difference in SNV levels was observed (Figure 8A and 8B).

The expression profiles were associated negatively with methylation in luminal B and normal-like cancers, and positively with CNV profiles in all groups except the true normal and normal-like groups (Figure 8C).

Immune response (Spot L)

The immune response gene module (spot L) contained 67 genes (Pearson’s correlation coefficient: 0.92-0.14). The top correlated genes in the module were FERMT3, PARVG, FMNL1, TMC8, C1QA, GMFG, LST1, BIN2, TNFRSF1B, ABI3 (Supplementary Table S21 in Supplementary_Tables_S17-S24.xlsx file of “Extended data”, see Data and software availability).66–72

The immune response signature expression profiles were upregulated in all cancers. In addition, methylation levels were significantly increased in Lum B cancer and decreased in basal cancers compared to normal tissue. CNV profile was significantly decreased only normal-like cancers (Figure 9A and 9B).

The expression of immune system genes was strongly negatively correlated with methylation profiles in all cancer subtypes. In addition, there was a positive association of expression with CNV in luminal B cancers as well as a positive association with SNV profiles in luminal A and luminal B breast cancers (Figure 9C).

Stroma and stem cells (Spot Q)

Stromal and stem cell gene module (141 genes, Pearson correlation coefficient: 0.21-0.81) with the top correlated CAV1, TGFBR2, RBMS1, CRYAB, CX3CL1, RCAN1, ETS1, ADAMTS9, LIMS2, GPX3 genes (Supplementary Table S22 in Supplementary_Tables_S17-S24.xlsx file of “Extended data”, see Data and software availability).73–80 Spot Q was characterized by downregulated expression and upregulated methylation profiles in all cancer subtypes (except basal cancers). CNV profiles were significantly upregulated in HER2E and normal-like cancers and SNV profiles were significantly downregulated in luminal A cancers (Figure 10A and 10B).

Methylation was negatively associated with the expression levels in Lum A and basal cancers. CNV levels were negatively correlated with expression in Lum A cancers. No notable correlation with expression profiles was observed for SNV (Figure 10C).

RNA splicing (Spot R)

RNA splicing gene module consisted of 207 genes (Pearson correlation coefficient: 0.14-0.85). The top correlated genes in this module were SSNA1, DRAP1, SURF2, PTGES2, RBM42, FASTK, BAX, LSM4, GUK1, and ZNHIT1 (Supplementary Table S23 in Supplementary_Tables_S17-S24.xlsx file of “Extended data”, see Data and software availability).81–85

The expression levels of the RNA splicing gene module were massively overexpressed in all cancers. In addition, CNV levels of this gene module were significantly decreased in normal-like cancers, while no other differences were observed across other omic profiles in the rest of the breast cancer subtypes (Figure 11A and 11B). No significant association between expression and other omic profiles was observed in the dataset, except for a significant positive association of SNVs in luminal A cancers (Figure 11C).

Estrogen receptor signaling (Spot S)

Finally, the expression profiles of the ESR1 signature (spot S, Pearson correlation coefficient: 0.24-0.82, top correlated genes: SCUBE2, ESR1, ABCC8, RALGPS2, APH1B, BBS4, MYB, GALNT10, LMX1B, HHAT,86–90 Supplementary Table S24 in Supplementary_Tables_S17-S24.xlsx file of “Extended data”, see Data and software availability) were significantly downregulated in Lum B, HER2E, basal, and normal-like cancers, while were significantly upregulated in the Lum A subtype. In addition, the methylation profiles in the basal subtype were significantly increased compared to the true normal tissue. CNV profiles were significantly upregulated in Lum B cancers, while no differences in SNV levels were observed (Figure 12A and 12B).

The expression profiles were negatively correlated with the methylation profiles in all cancers but were significant only for Lum A, Lum B, and HER2E subtypes. CNV profiles for HER2E and Lum B cancers also were significantly negatively correlated with the expression of ESR1 signature genes. Finally, we observed negatively correlated expression with SNV profiles in Lum A and Lum B cancers.

Multi-omic summary of deregulated modules in breast cancer subtypes, survival, and clinical phenotypes

We aimed to summarize findings from multi-omic analyses based on breast cancer subtypes, focusing on gene modules, survival, and phenotypic characteristics. For this purpose, we constructed Cox regression models for the interaction of continuous expression, methylation, CNV, and SNV levels for each gene module and each cancer subgroup. We also generated so-called phenotype maps visualizing the association between clinical phenotype parameters with different omic layers as described in our previous publication.13

We found that gene signatures associated with EMT/cell cycle, luminal, immune system, and RNA splicing were upregulated across all cancer subtypes compared to normal tissue. Conversely, stromal/stem cell signatures were downregulated across all cancer subtypes. The expression levels of RNA splicing genes remained consistent across all cancer subtypes. Immune signature genes were notably higher in HER2E, basal, and normal-like cancers than in luminal A and B subtypes. For other gene modules, the extent of expression was varied along with cancer subtypes. Specifically, the expression of the EMT/cell cycle module progressively increased from luminal A through normal-like, luminal B, HER2E, to basal cancers, with the highest expression noted in basal cancers. Similarly, for luminal gene signature, the expression gradually increased from basal through normal-like, HER2E, luminal A to luminal B cancers. Finally, gradual downregulation of stromal/stem cell signature was observed from normal-like through basal, luminal A, HER2E to luminal B subtypes. Interestingly, these changes were paralleled with the increase in methylation levels. In addition, we observed consistently increased methylation levels of VDR genes across all cancer types, except the normal-like category. However, this was not associated with any notable changes in their expression levels. In addition to these changes shared by all cancer subtypes, there were more subtype-characteristic perturbations (Figures 5-12).

Thus, luminal A cancers were additionally, characterized by downregulation of expression and methylation of DNA repair genes, and overexpression of ESR1 signature genes as the most dominant feature for this cancer subtype. Finally, this subtype showed decreased counts of SNVs in the immune system, stromal/stem cells, and RNA splicing genes. Interestingly, EMT/cell cycle gene expression in this subgroup was upregulated despite increased methylation levels; however, their expression levels were positively correlated with CNVs. The survival in luminal A cancers was associated with several gene modules on different omic layers, with the highest impact of the low expression levels of EMT and DNA repair genes on favorable survival prognosis. The luminal A cancers showed multiple significant associations with clinical phenotypes (Figure 14). Particularly, the overexpression of DNA repair genes was associated with poor prognosis. Moreover, increased SNV profiles and decreased methylation of these genes in this cancer type were associated with advanced stages of American Joint Committee on Cancer’s (AJCC) tumor pathologic assessment; particularly with pathologic M (metastasis) and pathologic N (lymph nodes). Furthermore, the increased expression of luminal cancer gene signature was associated with the presence of prior malignancies.

The luminal B subtype exhibited expression changes similar to luminal A cancers, except for unchanged expression DNA repair genes and a downregulated ESR1 signaling gene signature (spot S). This specific downregulation in luminal B was not linked to significant changes in methylation or CNV when compared to luminal A cancers. However, it showed a negative correlation with the SNV profile, not observed in luminal A cancers. Moreover, the methylation profiles in these two luminal subtypes closely resemble, except for increased methylation of immune system genes in the luminal B cancers. We did not observe any significant association with survival in this cancer subtype, except for methylation of immune system genes with borderline significance (p=0.0678) (Figure 13). Phenotype portraits for this cancer subtype showed a positive association of advanced AJCC pathologic staging, and, in particular AJCC pathologic M with expression and methylation of RNA splicing genes as well as decreased expression and methylation of DNA repair genes. AJCC pathologic T (primary tumor) was positively associated with the increased CNV profiles of the EMT/cell cycle and immune response genes.

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure13.gif

Figure 13. Association of omic features of SOM gene modules with survival in PAM50 subtypes.

Survival analysis was performed using the Cox proportional hazards regression with the inclusion of spot levels as continuous variables using “contsurvplot”, “survival”, and “survminer” packages. Survival curves were visualized with range values with 5 intervals (Q1: minimum, Q2: 25th percentile, Q3: 50th percentile, Q4: 75th percentile, Q5: maximum). Only plots with survival-spot association with p-value ≤ 0.1 are displayed.

The transcriptome profiles for the HER2E subtype were closely aligned with those of luminal B, differing only in the magnitude of changes. Unlike in the luminal B subtype, methylation levels of EMT/cell cycle genes in the HER2E subtype remained unchanged compared to the true normal samples. Furthermore, this subtype exhibited increased CNV profiles for luminal and stromal/stem cell gene signatures. Notably, the survival impact for this cancer subtype was most influenced by the underexpression of the VDR gene signature and the overexpression of immune system genes (Figure 13). In agreement with the survival data, the negative association of vital status with the overexpression of immune system genes was observed. Moreover, the overexpression of the luminal cancer signature was positively correlated with advanced tumor staging (Figure 14).

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure14.gif

Figure 14. The phenotype portraits of the association of omic features of SOM gene modules (spots) with clinical parameters in PAM50 subtypes.

Phenotype portraits show the -log10(p) regression model of SOM metagene levels and clinicopathological stages, vital status, and treatment variables. Deregulated gene modules (spots) are indicated on the maps and coloring shows the significance of their association with evaluated parameters.

Unlike Lum A, Lum B, and HER2E cancers, in basal cancers, a strong negative correlation existed between the overexpression of EMT/cell cycle gene signatures and decreased methylation levels. A similar relationship between underexpression and methylation levels was noted for DNA repair genes. This cancer subtype’s survival was linked to various gene modules across omic layers, with hypomethylation of the ESR1 gene signature having the most significant positive impact on prognosis (Figure 13). Furthermore, the prior malignancy was positively associated with increased methylation of DNA repair genes and decreased methylation of RNA splicing genes (Figure 14).

Finally, the normal-like cancers were characterized with additional underexpression of VDR signaling signature and a decrease of CNV profiles in almost all gene modules (DNA repair, luminal cancer, VDR, immune response, stromal/stem cells, RNA splicing) compared to other cancer subtypes. No specific gene modules were significantly associated with survival for this cancer subtype, however, we observed a significant association of CNV increase in VDR signaling, EMT/cell cycle, and immune system genes with prior malignancy in this cancer subtype (Figure 14).

Discussion

This study analyzed the omic landscapes in breast cancer PAM50 subgroups. Our results confirmed the multi-omic molecular heterogeneity of cancer subtypes as well as showed that this heterogeneity results in subtype-characteristic associations between transcriptomic, genomic, and epigenetic features as well as with clinical parameters and survival across disease subtypes (graphical representation is given in Figure 15).

4b2ff5a6-6685-440d-be8c-336d4a844b31_figure15.gif

Figure 15. Schematic summary of the multi-omic analysis of breast cancer PAM50 subtypes.

The results show that the molecular diversity of breast cancer is most comprehensively captured through a multi-omic overview of perturbed gene modules. The study also found subtype-specific associations between omic features, as well as distinct relationships between these features and subtype survival rates and clinical characteristics.

The changes in gene expression across cancer subtypes were more qualitative rather than quantitative. All breast cancer subtypes showed upregulation of EMT/cell cycle, luminal signature, immune response, and RNA splicing genes and downregulation of stromal/stem cell signature genes. These gene modules represent the fundamental processes related to breast cancers and were extensively studied elsewhere.91 However, the extent of deregulation of the modules showed considerable subtype association. Indeed, previous studies clearly demonstrated the differences in the expression of proliferative and metastatic signatures in basal cancers compared with the luminal cancers and HER2E cancers, indicating the more aggressive nature of the latter subtype.92,93 Immune/inflammatory gene signatures were lowest in luminal cancer subtypes and highest in normal-like, HER2E, and basal subtypes, which also corresponds to the previous findings.93,94 The luminal gene signature, in contrast, was reversed being highest in the luminal A and B cancers and lowest in the basal subtype. The most striking difference on the transcriptome layer was the overexpression of ESR1 genes in the luminal A subtype, while in other PAM50 subtypes including luminal B cancers, they were considerably lower. This finding is, however, confirmatory, since the higher expression of ER-related genes and lower expression of proliferative genes in luminal A cancers has been described previously.95,96

Methylation levels of perturbed gene modules showed higher diversity compared to expression levels. Indeed, EMT/cell cycle genes were hypomethylated in basal cancers which well agrees with their overexpression. However, overexpression of the same genes in HER2E cancers was not paralleled with significant changes in methylation compared to normal tissue; moreover, in luminal A and luminal B cancers, the same genes were significantly hypermethylated despite their increased expression. Immune response genes were significantly hypomethylated in luminal A, luminal B, and HER2E cancers, but no differences were observed for basal and normal-like cancers. We observed hypermethylation of the ESR1 gene signature in basal cancers, consistent with previously reported results.97 On the other side, the underexpression of the same gene module in other cancer subtypes was not associated with changes in the methylation levels. The differential methylation in PAM50 subtypes was reported previously,98 moreover, the incorporation of methylation caused improvement in class prediction of subtypes.99

CNV changes are common in breast cancers,100 moreover, they show subtype-characteristic amplifications or deletions.101 The CNV profiles of practically all gene modules were lower in normal-like breast cancers compared to other cancer subtypes. This subtype closely resembles normal breast tissue however11 and shows low levels of expression of genes associated with cell proliferation and higher expression of genes related to adipose tissue and normal mammary gland function.102 Despite their “normal-like” gene expression pattern, these cancers can still behave aggressively.103 Interestingly HER2E was characterized by a CNV increase in luminal gene signature that includes oncogenes on chromosome 9.104

Our study also indicated that somatic alteration counts were not much different between breast cancer subtypes. These findings corroborate prior research that reports no difference in somatic mutation load across PAM50 subtypes,105 although differences in the types of mutations are observed in the previous studies. For instance, TP53 mutations are more prevalent in basal subtypes, PIK3CA mutations are common in HER2-enriched subtypes, and mutations in GATA3, FOXA1, XBP1, and MYB are typically found in the luminal A subtype.98 Thus, the mutation counts in genes may not be a good characteristic for cancer subtypes.

The abovementioned results specifically demonstrate that similarities in expression levels do not necessarily reflect across other genomic and epigenetic layers. This raises the question if associations of omic data are similar or different across breast cancer subtypes. Multi-layer SOM approach allowed exploring this issue with considerable detail. It is well known that promoter methylation and CNVs are important regulators of gene expression.106,107 The general understanding is that methylation drives the silencing of gene expression,108 however, it has been demonstrated that many genes are active in the methylated state both in normal and cancer cells.109 Somatic CNVs, as well, can have both activating and silencing effects in cancer.100 The somatic single nucleotide variations (SNVs), which predominantly impact protein structure and function rather than regulatory mechanisms110 usually showed minimal effects on the transcriptome. Our results clearly demonstrated that expression-methylation-CNV-SNV associations substantially vary in different breast cancer subtypes. Perhaps, the most striking example is the expression of cell cycle and EMT genes overexpressed in all cancer subtypes. However, only in basal and normal-like cancers it was strongly associated with their hypomethylation,111 while, in Lum A, Lum B, and HER2E cancers CNVs were increased along with upregulated expression.112

The diversity of omic data associations across subtypes was reported previously in cancers,113,114 including breast cancer.115 Analysis of how different omic layers interact in cancers remains limited, but this is a crucial aspect of therapeutic development. Different approaches may be needed to target the same genes across various subtypes based on the understanding of these interactions.

Although the overall survival rate within the TCGA breast cancer cohort is relatively high,116 we demonstrated that within subtypes, survival rates could be linked with specific deregulations in gene expression, methylation, or copy number variation of described functional gene modules. This suggests that molecular alterations can have a significant impact on the prognosis within individual breast cancer subtypes and multi-omic type predictors can significantly improve assessment of survival and prognosis.117

The majority of omic data integration studies are focused on building multi-omic classifiers rather than understanding the relations between expression, methylation, and CNVs.118–121 An exemplary study was done by Ochoa and de Anda-Jáuregui who analyzed the gene expression regulation of 50 genes in the PAM50 subtypes and found a unique set of predictors for the expression of genes in the PAM50 signature associated with each of the molecular subtypes.122 However, only a few studies focus on the analysis of intricate omic features of breast cancer subtypes98 or try to understand the complexity of their interactions.123,124 In this sense, our study offers a significant advancement in understanding the complexity of molecular events and their interactions in breast cancer subtypes with the implications for survival and clinical parameters.

There are several limitations worth pointing out in this study. First, we didn’t integrate other regulatory elements such as transcription factors,125 miRNAs,126 or chromatin modifiers,127 which were shown to be implicated in breast cancers. Furthermore, the sample size per breast cancer subtype is imbalanced in the TCGA-BRCA cohort. The largest number of samples have Lum A subtype (480 samples), which implies higher statistical power to detect significant associations between different omic layers compared to normal-like (32 samples) or HER2E (74 samples) cancers.

Conclusions

Multi-omics SOM analysis allowed characterization of molecular diversity of breast cancer subtypes in term of perturbed gene modules across expression, methylation, copy number and single nucleotide variations. Moreover, the results highlight the complex subtype-characteristic associations between gene expression and epigenetic/genomic factors and their implications for survival and clinical outcomes.

Ethics and consent

Ethical approval and consent were not required.

Author contributions

Conceptualization - AA, HB; Data Curation - SD, GeM, GM, ACh, AM, HG; Formal Analysis - SD, AA; Funding Acquisition - AA; Investigation - SD; Methodology - AA, HB; Project Administration - AA, HB; Resources - AA; Software - SD, AA; Supervision - AA, HB; Validation - GeM, GM, ACh, AM, HG; Visualization - SD; Writing – Original Draft Preparation - SD, AA, HB; Writing – Review & Editing - SD, AA, HB. All authors have read and agreed to this version of the manuscript.

Data and software availability

Underlying data

Zenodo: Integrated analysis of “-omic” landscapes in breast cancer subtypes: Supplementary Dataset. https://doi.org/10.5281/zenodo.10947982. 128

The project contains the following underlying data:

  • - BRCA_mlSOM_Dataset.zip (The Data folder contains TCGA-BRCA omic (RNA-Seq, methylation, CNV, and SNV) processed data and multi-SOM pipeline scripts. The BRCA-TCGA-mlSOM folder contains scripts for multi-SOM downstream analysis including functional annotation of gene modules (spots), comparison of their levels in cancer subtypes with true normal tissue, regression analysis for assessment of the association between omic layers, survival, and clinical parameter analysis. It also contains the file representing the results of the SOM analysis for each omic layer (expr - Results folder - expression, prom - Results folder - promoter methylation, cnv - Results folder - copy number variations, amd snv - Results folder - somatic single nucleotide variation), as well as phenotype portraits per omic laver (organized in pheno.exp, pheno.met, pheno.cnv, and pheno.snv) folders. The Supplementary data folder contains supplementary tables and figures cited in the text.)

Extended data

Zenodo: Integrated analysis of “-omic” landscapes in breast cancer subtypes: Supplementary Dataset. https://doi.org/10.5281/zenodo.10947982. 128

The project contains the following extended data:

  • - BRCA_mlSOM_Dataset.zip (The Data folder contains TCGA-BRCA omic (RNA-Seq, methylation, CNV, and SNV) processed data and multi-SOM pipeline scripts. The BRCA-TCGA-mlSOM folder contains scripts for multi-SOM downstream analysis including functional annotation of gene modules (spots), comparison of their levels in cancer subtypes with true normal tissue, regression analysis for assessment of the association between omic layers, survival, and clinical parameter analysis. It also contains the file representing the results of the SOM analysis for each omic layer (expr - Results folder - expression, prom - Results folder - promoter methylation, cnv - Results folder - copy number variations, amd snv - Results folder - somatic single nucleotide variation), as well as phenotype portraits per omic laver (organized in pheno.exp, pheno.met, pheno.cnv, and pheno.snv) folders. The Supplementary data folder contains supplementary tables and figures cited in the text.)

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).

Acknowledgments

Grammar checks and text style revisions were conducted with the support of ChatGPT, an AI-based tool.

References

  • 1.  Turner KM, Yeo SK, Holm TM, et al.: Heterogeneity within molecular subtypes of breast cancer. Am. J. Physiol.-Cell Physiol. 2021 Aug 1 [cited 2024 Mar 6]; 321(2): C343–C354. PubMed Abstract | Publisher Full Text | Free Full Text
  • 2.  Guo L, Kong D, Liu J, et al.: Breast cancer heterogeneity and its implication in personalized precision therapy. Exp. Hematol. Oncol. 2023 Jan 9 [cited 2024 Mar 6]; 12(1): 3. PubMed Abstract | Publisher Full Text | Free Full Text
  • 3.  Shipitsin M, Campbell LL, Argani P, et al.: Molecular Definition of Breast Tumor Heterogeneity. Cancer Cell. 2007 Mar [cited 2024 Mar 6]; 11(3): 259–273. Publisher Full Text Reference Source
  • 4.  Makki J: Diversity of Breast Carcinoma: Histological Subtypes and Clinical Relevance. Clin. Med. Insights Pathol. 2015 Jan [cited 2024 Mar 6]; 8: CPath.S31563–CPath.S31531. PubMed Abstract | Publisher Full Text | Free Full Text
  • 5.  Rakha EA, Reis-Filho JS, Ellis IO: Combinatorial biomarker expression in breast cancer. Breast Cancer Res. Treat. 2010 Apr [cited 2024 Mar 6]; 120(2): 293–308. Publisher Full Text
  • 6.  Chia SK, Bramwell VH, Tu D, et al.: A 50-Gene Intrinsic Subtype Classifier for Prognosis and Prediction of Benefit from Adjuvant Tamoxifen. Clin. Cancer Res. 2012 Aug 15 [cited 2024 Mar 6]; 18(16): 4465–4472. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 7.  Sawaki M, Shien T, Iwata H: TNM classification of malignant tumors (Breast Cancer Study Group). Jpn. J. Clin. Oncol. 2019 Mar 1 [cited 2024 Mar 6]; 49(3): 228–231. PubMed Abstract | Publisher Full Text Reference Source
  • 8.  Slodkowska EA, Ross JS: MammaPrintTM 70-gene signature: another milestone in personalized medical care for breast cancer patients. Expert. Rev. Mol. Diagn. 2009 Jul [cited 2024 Mar 6]; 9(5): 417–422. PubMed Abstract | Publisher Full Text
  • 9.  Rath MG, Uhlmann L, Fiedler M, et al.: Oncotype DX® in breast cancer patients: clinical experience, outcome and follow-up—a case–control study. Arch. Gynecol. Obstet. 2018 Feb [cited 2024 Mar 6]; 297(2): 443–447. PubMed Abstract | Publisher Full Text
  • 10.  Raj-Kumar PK, Liu J, Hooke JA, et al.: PCA-PAM50 improves consistency between breast cancer intrinsic and clinical subtyping reclassifying a subset of luminal A tumors as luminal B. Sci. Rep. 2019 May 28 [cited 2024 Mar 6]; 9(1): 7956. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 11.  Yersal O: Biological subtypes of breast cancer: Prognostic and therapeutic implications. World. J. Clin. Oncol. 2014 [cited 2024 Mar 6]; 5(3): 412. Reference Source
  • 12.  Nelson DJ, Clark B, Munyard K, et al.: A review of the importance of immune responses in luminal B breast cancer. Onco Targets Ther. 2017 Mar 4 [cited 2024 Mar 6]; 6(3): e1282590. PubMed Abstract | Publisher Full Text | Free Full Text
  • 13.  Arakelyan A, Melkonyan A, Hakobyan S, et al.: Transcriptome Patterns of BRCA1- and BRCA2- Mutated Breast and Ovarian Cancers. Int. J. Mol. Sci. 2021 Jan 28 [cited 2024 Mar 7]; 22(3): 1266. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 14.  Menyhárt O, Győrffy B: Multi-omics approaches in cancer research with applications in tumor subtyping, prognosis, and diagnosis. Comput. Struct. Biotechnol. J. 2021 [cited 2024 Mar 7]; 19: 949–960. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 15.  Ahmed Z: Multi-omics strategies for personalized and predictive medicine: past, current, and future translational opportunities. Emerg. Top Life Sci. 2022 Apr 15 [cited 2024 Mar 7]; 6(2): 215–225. PubMed Abstract | Publisher Full Text Reference Source
  • 16.  Omondiagbe DA, Veeramani S, Sidhu AS: Machine Learning Classification Techniques for Breast Cancer Diagnosis. IOP Conf. Ser. Mater. Sci. Eng. 2019 Jun 7 [cited 2024 Mar 7]; 495: 012033. Publisher Full Text
  • 17.  Sivadas A, Kok VC, Ng KL: Multi-omics analyses provide novel biological insights to distinguish lobular ductal types of invasive breast cancers. Breast Cancer Res. Treat. 2022 Jun [cited 2024 Mar 7]; 193(2): 361–379. PubMed Abstract | Publisher Full Text
  • 18.  Zhen WZ, Hua LX, Ling WX, et al.: Integration of multi-omics data reveals a novel hybrid breast cancer subtype and its biomarkers. Front. Oncol. 2023 Mar 21 [cited 2024 Mar 7]; 13: 1130092. Publisher Full Text
  • 19.  Malik V, Kalakoti Y, Sundar D: Deep learning assisted multi-omics integration for survival and drug-response prediction in breast cancer. BMC Genomics. 2021 Dec [cited 2024 Mar 7]; 22(1): 214. PubMed Abstract | Publisher Full Text | Free Full Text
  • 20.  Löffler-Wirth H, Kalcher M, Binder H: oposSOM: R-package for high-dimensional portraying of genome-wide expression landscapes on bioconductor. Bioinformatics. 2015 Oct 1 [cited 2024 Mar 7]; 31(19): 3225–7. PubMed Abstract | Publisher Full Text Reference Source
  • 21.  Wirth H, Von Bergen M, Binder H: Mining SOM expression portraits: feature selection and integrating concepts of molecular function. BioData Min. 2012 Dec [cited 2024 Mar 11]; 5(1): 18. PubMed Abstract | Publisher Full Text | Free Full Text
  • 22.  Binder H, Schmidt M, Hopp L, et al.: Integrated Multi-Omics Maps of Lower-Grade Gliomas. Cancers. 2022 Jun 4 [cited 2024 Mar 7]; 14(11): 2797. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 23.  Hopp L, Nersisyan L, Löffler-Wirth H, et al.: Epigenetic Heterogeneity of B-Cell Lymphoma: Chromatin Modifiers. Genes. 2015 Oct 21 [cited 2024 Mar 7]; 6(4): 1076–1112. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 24.  The Cancer Genome Atlas Research NetworkWeinstein JN, Collisson EA, et al.: The Cancer Genome Atlas Pan-Cancer analysis project. Nat. Genet. 2013 Oct [cited 2024 Mar 11]; 45(10): 1113–1120. Publisher Full Text Reference Source
  • 25.  Grossman RL, Heath AP, Ferretti V, et al.: Toward a Shared Vision for Cancer Genomic Data. N. Engl. J. Med. 2016 Sep 22 [cited 2024 Mar 11]; 375(12): 1109–1112. PubMed Abstract | Publisher Full Text | Free Full Text
  • 26.  Michael Love SA: DESeq2. [object Object].2017 [cited 2024 Mar 8]. Reference Source
  • 27.  Du P, Zhang X, Huang CC, et al.: Comparison of Beta-value and M-value methods for quantifying methylation levels by microarray analysis. BMC Bioinformatics. 2010 Dec [cited 2024 Apr 9]; 11(1): 587. PubMed Abstract | Publisher Full Text | Free Full Text
  • 28.  Loeffler-Wirth H, Hopp L, Schmidt M, et al.: The Transcriptome and Methylome of the Developing and Aging Brain and Their Relations to Gliomas and Psychological Disorders. Cells. 2022 Jan 21 [cited 2024 Mar 19]; 11(3): 362. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 29.  Kuleshov MV, Jones MR, Rouillard AD, et al.: Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res. 2016 Jul 8 [cited 2024 Mar 7]; 44(W1): W90–W97. PubMed Abstract | Publisher Full Text | Free Full Text
  • 30.  Searle SR, Speed FM, Milliken GA: Population Marginal Means in the Linear Model: An Alternative to Least Squares Means. Am. Stat. 1980 Nov [cited 2024 Mar 7]; 34(4): 216–221. Publisher Full Text
  • 31.  Bauer DJ, Curran PJ: Probing Interactions in Fixed and Multilevel Regression: Inferential and Graphical Techniques. Multivar. Behav. Res. 2005 Jul [cited 2024 Mar 7]; 40(3): 373–400. PubMed Abstract | Publisher Full Text
  • 32.  Wang B, Li R, Perrizo W, editors. Big Data Analytics in Bioinformatics and Healthcare.Azar AT, editor. Advances in Bioinformatics and Biomedical Engineering. IGI Global; 2015 [cited 2024 Mar 8].Publisher Full Text
  • 33.  Binder H, Hopp L, Cakir V, et al.: Molecular phenotypic portraits - Exploring the “OMES” with individual resolution. Proceedings of the 6th International Symposium on Health Informatics and Bioinformatics. Izmir, Turkey: IEEE. 2011 [cited 2024 Mar 20]; pp. 99–107. Reference Source
  • 34.  Sarrió D, Rodriguez-Pinilla SM, Hardisson D, et al.: Epithelial-Mesenchymal Transition in Breast Cancer Relates to the Basal-like Phenotype. Cancer Res. 2008 Feb 15 [cited 2024 Mar 7]; 68(4): 989–997. PubMed Abstract | Publisher Full Text Reference Source
  • 35.  Smid M, Wang Y, Zhang Y, et al.: Subtypes of Breast Cancer Show Preferential Site of Relapse. Cancer Res. 2008 May 1 [cited 2024 Mar 7]; 68(9): 3108–3114. PubMed Abstract | Publisher Full Text Reference Source
  • 36.  Shen H, Powers N, Saini N, et al.: The SWI/SNF ATPase Brm Is a Gatekeeper of Proliferative Control in Prostate Cancer. Cancer Res. 2008 Dec 15 [cited 2024 Mar 7]; 68(24): 10154–10162. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 37.  Charafe-Jauffret E, Ginestier C, Monville F, et al.: Gene expression profiling of breast cell lines identifies potential new basal markers. Oncogene. 2006 Apr 6 [cited 2024 Mar 7]; 25(15): 2273–2284. PubMed Abstract | Publisher Full Text Reference Source
  • 38.  Nikolsky Y, Sviridov E, Yao J, et al.: Genome-Wide Functional Synergy between Amplified and Mutated Genes in Human Breast Cancer. Cancer Res. 2008 Nov 15 [cited 2024 Mar 7]; 68(22): 9532–9540. PubMed Abstract | Publisher Full Text Reference Source
  • 39.  Finak G, Bertos N, Pepin F, et al.: Stromal gene expression predicts clinical outcome in breast cancer. Nat. Med. 2008 May [cited 2024 Mar 11]; 14(5): 518–527. Publisher Full Text Reference Source
  • 40.  Lim E, Wu D, Pal B, et al.: Transcriptome analyses of mouse and human mammary cell subpopulations reveal multiple conserved genes and pathways. Breast Cancer Res. 2010 Apr [cited 2024 Mar 7]; 12(2): R21. PubMed Abstract | Publisher Full Text | Free Full Text
  • 41.  Liu S, Ye Z, Xue VW, et al.: KIF2C is a prognostic biomarker associated with immune cell infiltration in breast cancer. BMC Cancer. 2023 Apr 4 [cited 2024 Apr 9]; 23(1): 307. PubMed Abstract | Publisher Full Text | Free Full Text
  • 42.  Bridges AE, Ramachandran S, Pathania R, et al.: RAD51AP1 Deficiency Reduces Tumor Growth by Targeting Stem Cell Self-Renewal. Cancer Res. 2020 Sep 15 [cited 2024 Apr 9]; 80(18): 3855–3866. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 43.  Zhang W, Cao L, Sun Z, et al.: Skp2 is over-expressed in breast cancer and promotes breast cancer cell proliferation. Cell Cycle. 2016 May 18 [cited 2024 Apr 9]; 15(10): 1344–1351. PubMed Abstract | Publisher Full Text | Free Full Text
  • 44.  Elsharawy KA, Mohammed OJ, Aleskandarany MA, et al.: The nucleolar-related protein Dyskerin pseudouridine synthase 1 (DKC1) predicts poor prognosis in breast cancer. Br. J. Cancer. 2020 Nov 10 [cited 2024 Apr 9]; 123(10): 1543–1552. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 45.  Issac MSM, Yousef E, Tahir MR, et al.: MCM2, MCM4, and MCM6 in Breast Cancer: Clinical Utility in Diagnosis and Prognosis. Neoplasia. 2019 Oct [cited 2024 Apr 9]; 21(10): 1015–1035. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 46.  Song N, Deng L, Zeng L, et al.: USP9X deubiquitinates and stabilizes CDC123 to promote breast carcinogenesis through regulating cell cycle. Mol. Carcinog. 2023 Oct [cited 2024 Apr 9]; 62(10): 1487–1503. PubMed Abstract | Publisher Full Text
  • 47.  Erkko H, Pylkäs K, Karppinen SM, et al.: Germline alterations in the CLSPN gene in breast cancer families. Cancer Lett. 2008 Mar [cited 2024 Apr 9]; 261(1): 93–97. PubMed Abstract | Publisher Full Text Reference Source
  • 48.  Zhao B, Song X, Guan H: CircACAP2 promotes breast cancer proliferation and metastasis by targeting miR-29a/b-3p-COL5A1 axis. Life Sci. 2020 Mar [cited 2024 Apr 9]; 244: 117179. PubMed Abstract | Publisher Full Text Reference Source
  • 49.  Zhao X, Su X, Cao L, et al.: OTUD4: A Potential Prognosis Biomarker for Multiple Human Cancers. Cancer Manag. Res. 2020 Feb [cited 2024 Apr 9]; Volume 12: 1503–1512. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 50.  Debauve G, Nonclercq D, Ribaucour F, et al.: Early expression of the Helicase-Like Transcription Factor (HLTF/SMARCA3) in an experimental model of estrogen-induced renal carcinogenesis. Mol. Cancer. 2006 Dec [cited 2024 Apr 9]; 5(1): 23. PubMed Abstract | Publisher Full Text | Free Full Text
  • 51.  Zhu T, Xiao Z, Yuan H, et al.: ACO1 and IREB2 downregulation confer poor prognosis and correlate with autophagy-related ferroptosis and immune infiltration in KIRC. Front. Oncol. 2022 Aug 17 [cited 2024 Apr 9]; 12: 929838. PubMed Abstract | Publisher Full Text | Free Full Text
  • 52.  Tang Z, Chen T, Ren X, et al.: Identification of transcriptional isoforms associated with survival in cancer patient. J. Genet. Genomics. 2019 Sep [cited 2024 Apr 9]; 46(9): 413–421. PubMed Abstract | Publisher Full Text Reference Source
  • 53.  Liu Z, Zhou J, Wang Z, et al.: Analysis of SEC24D gene in breast cancer based on UALCAN database. Open. Life Sci. 2019 Dec 31 [cited 2024 Apr 9]; 14(1): 707–711. PubMed Abstract | Publisher Full Text | Free Full Text
  • 54.  Gong PJ, Shao YC, Yang Y, et al.: Analysis of N6-Methyladenosine Methyltransferase Reveals METTL14 and ZC3H13 as Tumor Suppressor Genes in Breast Cancer. Front. Oncol. 2020 Dec 9 [cited 2024 Apr 9]; 10: 578963. PubMed Abstract | Publisher Full Text | Free Full Text
  • 55.  Song HS, Ha SY, Kim JY, et al.: The effect of genetic variants of SLC22A18 on proliferation, migration, and invasion of colon cancer cells. Sci. Rep. 2024 Feb 16 [cited 2024 Apr 9]; 14(1): 3925. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 56.  Liu H, Zhou Y, Qiu H, et al.: Rab26 suppresses migration and invasion of breast cancer cells through mediating autophagic degradation of phosphorylated Src. Cell Death Dis. 2021 Mar 17 [cited 2024 Apr 9]; 12(4): 284. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 57.  Hu R, Peng G, Dai H, et al.: ZNF668 Functions as a Tumor Suppressor by Regulating p53 Stability and Function in Breast Cancer. Cancer Res. 2011 Oct 15 [cited 2024 Apr 9]; 71(20): 6524–6534. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 58.  Akkus G, Koyuturk LC, Yilmaz M, et al.: Asprosin and meteorin-like protein immunoreactivity in invasive ductal breast carcinoma stages. Tissue Cell. 2022 Aug [cited 2024 Apr 9]; 77: 101855. PubMed Abstract | Publisher Full Text Reference Source
  • 59.  Yi C, Mu L, De La Longrais IAR, et al.: The circadian gene NPAS2 is a novel prognostic biomarker for breast cancer. Breast Cancer Res. Treat. 2010 Apr [cited 2024 Apr 9]; 120(3): 663–669. PubMed Abstract | Publisher Full Text | Free Full Text
  • 60.  Martin: Loss of occludin leads to the progression of human breast cancer. Int. J. Mol. Med. 2010 Sep 21 [cited 2024 Apr 9]; 26(5): 723–734. PubMed Abstract | Publisher Full Text Reference Source
  • 61.  Uddin MH, Pimentel JM, Chatterjee M, et al.: Targeting PP2A inhibits the growth of triple-negative breast cancer cells. Cell Cycle. 2020 Mar 3 [cited 2024 Apr 9]; 19(5): 592–600. PubMed Abstract | Publisher Full Text | Free Full Text
  • 62.  Furrer D, Dragic D, Chang SL, et al.: Association between genome-wide epigenetic and genetic alterations in breast cancer tissue and response to HER2-targeted therapies in HER2-positive breast cancer patients: new findings and a systematic review. Cancer Drug Resist. 2022 [cited 2024 Apr 9]; 5: 995–1015. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 63.  Chen HH, Fan P, Chang SW, et al.: NRIP/DCAF6 stabilizes the androgen receptor protein by displacing DDB2 from the CUL4A-DDB1 E3 ligase complex in prostate cancer. Oncotarget. 2017 Mar 28 [cited 2024 Apr 9]; 8(13): 21501–21515. PubMed Abstract | Publisher Full Text | Free Full Text
  • 64.  Knutsvik G, Collett K, Arnes J, et al.: QSOX1 expression is associated with aggressive tumor features and reduced survival in breast carcinomas. Mod. Pathol. 2016 Dec [cited 2024 Apr 9]; 29(12): 1485–1491. PubMed Abstract | Publisher Full Text Reference Source
  • 65.  Sato N, Maeda M, Sugiyama M, et al.: Inhibition of SNW 1 association with spliceosomal proteins promotes apoptosis in breast cancer cells. Cancer Med. 2015 Feb [cited 2024 Apr 9]; 4(2): 268–277. PubMed Abstract | Publisher Full Text | Free Full Text
  • 66.  Castellvı́-Bel S, Castells A, Johnstone CN, et al.: Evaluation of PARVG located on 22q13 as a candidate tumor suppressor gene for colorectal and breast cancer. Cancer Genet. Cytogenet. 2003 Jul [cited 2024 Apr 9]; 144(1): 80–82. PubMed Abstract | Publisher Full Text Reference Source
  • 67.  Zhang Q, Yang H, Tang C, et al.: FMNL1 promotes growth and metastasis of breast cancer by inhibiting BRCA1 via upregulation of HMGA1. Trop. J. Pharm. Res. 2022 Feb 16 [cited 2024 Apr 9]; 20(8): 1559–1564. Publisher Full Text Reference Source
  • 68.  Song J, Tang Y, Luo X, et al.: Pan-Cancer Analysis Reveals the Signature of TMC Family of Genes as a Promising Biomarker for Prognosis and Immunotherapeutic Response. Front. Immunol. 2021 Nov 9 [cited 2024 Apr 9]; 12: 715508. PubMed Abstract | Publisher Full Text | Free Full Text
  • 69.  Azzato EM, Lee AJX, Teschendorff A, et al.: Common germ-line polymorphism of C1QA and breast cancer survival. Br. J. Cancer. 2010 Apr [cited 2024 Apr 9]; 102(8): 1294–1299. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 70.  Yang Y, He X, Tang QQ, et al.: GMFG Has Potential to Be a Novel Prognostic Marker and Related to Immune Infiltrates in Breast Cancer. Front. Oncol. 2021 Jul 23 [cited 2024 Apr 9]; 11: 629633. Publisher Full Text
  • 71.  Sánchez-Barrena MJ, Vallis Y, Clatworthy MR, et al.: Correction: Bin2 Is a Membrane Sculpting N-BAR Protein That Influences Leucocyte Podosomes, Motility and Phagocytosis. Soldati T, editor. PLoS One. 2013 Aug 8 [cited 2024 Apr 9]; 8(8). Publisher Full Text
  • 72.  Xu F, Zhou G, Han S, et al.: Association of TNF-α, TNFRSF1A and TNFRSF1B Gene Polymorphisms with the Risk of Sporadic Breast Cancer in Northeast Chinese Han Women. Lee SG, editor. PLoS One. 2014 Jul 10 [cited 2024 Apr 9]; 9(7): e101138. PubMed Abstract | Publisher Full Text | Free Full Text
  • 73.  Qian XL, Pan YH, Huang QY, et al.: Caveolin-1: a multifaceted driver of breast cancer progression and its application in clinical treatment. Onco. Targets Ther. 2019 Feb [cited 2024 Apr 9]; Volume 12: 1539–1552. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 74.  Busch S, Acar A, Magnusson Y, et al.: TGF-beta receptor type-2 expression in cancer-associated fibroblasts regulates breast cancer cell growth and survival and is a prognostic marker in pre-menopausal breast cancer. Oncogene. 2015 Jan 2 [cited 2024 Apr 9]; 34(1): 27–38. PubMed Abstract | Publisher Full Text Reference Source
  • 75.  Zhang J, Zhang G, Zhang W, et al.: Loss of RBMS1 promotes anti-tumor immunity through enabling PD-L1 checkpoint blockade in triple-negative breast cancer. Cell Death Differ. 2022 Nov [cited 2024 Apr 9]; 29(11): 2247–2261. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 76.  Tardáguila M, Mira E, García-Cabezas MA, et al.: CX3CL1 Promotes Breast Cancer via Transactivation of the EGF Pathway. Cancer Res. 2013 Jul 15 [cited 2024 Apr 9]; 73(14): 4461–4473. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 77.  Rashidieh B, Bain AL, Tria SM, et al.: Alpha-B-Crystallin overexpression is sufficient to promote tumorigenesis and metastasis in mice. Exp. Hematol. Oncol. 2023 Jan 9 [cited 2024 Apr 9]; 12(1): 4. PubMed Abstract | Publisher Full Text | Free Full Text
  • 78.  Deng R, Huang JH, Wang Y, et al.: Disruption of super-enhancer-driven tumor suppressor gene RCAN1.4 expression promotes the malignancy of breast carcinoma. Mol. Cancer. 2020 Dec [cited 2024 Apr 9]; 19(1): 122. PubMed Abstract | Publisher Full Text | Free Full Text
  • 79.  Kim GC, Lee CG, Verma R, et al.: ETS1 Suppresses Tumorigenesis of Human Breast Cancer via Trans-Activation of Canonical Tumor Suppressor Genes. Front. Oncol. 2020 May 14 [cited 2024 Apr 9]; 10: 642. PubMed Abstract | Publisher Full Text | Free Full Text
  • 80.  Chen J, Cheng L, Zou W, et al.: ADAMTS9-AS1 Constrains Breast Cancer Cell Invasion and Proliferation via Sequestering miR-301b-3p. Front. Cell Dev. Biol. 2021 Nov 24 [cited 2024 Apr 9]; 9: 719993. PubMed Abstract | Publisher Full Text | Free Full Text
  • 81.  Zhu C, He L, Zhou X, et al.: Sulfatase 2 promotes breast cancer progression through regulating some tumor-related factors. Oncol. Rep. 2016 Mar [cited 2024 Apr 9]; 35(3): 1318–1328. PubMed Abstract | Publisher Full Text | Free Full Text
  • 82.  Saindane M, Rallabandi HR, Park KS, et al.: Prognostic Significance of Prostaglandin-Endoperoxide Synthase-2 Expressions in Human Breast Carcinoma: A Multiomic Approach. Cancer Inform. 2020 Jan [cited 2024 Apr 9]; 19: 117693512096969. PubMed Abstract | Publisher Full Text | Free Full Text
  • 83.  Sun X, Hu Y, Wu J, et al.: RBMS2 inhibits the proliferation by stabilizing P21 mRNA in breast cancer. J. Exp. Clin. Cancer Res. 2018 Dec [cited 2024 Apr 9]; 37(1): 298. PubMed Abstract | Publisher Full Text | Free Full Text
  • 84.  Das S, Yeung KT, Mahajan MA, et al.: Fas Activated Serine-Threonine Kinase Domains 2 (FASTKD2) mediates apoptosis of breast and prostate cancer cells through its novel FAST2 domain. BMC Cancer. 2014 Dec [cited 2024 Apr 9]; 14(1): 852. PubMed Abstract | Publisher Full Text | Free Full Text
  • 85.  Kholoussi NM, El-Nabi SEH, Esmaiel NN, et al.: Evaluation of Bax and Bak Gene Mutations and Expression in Breast Cancer. Biomed. Res. Int. 2014 [cited 2024 Apr 9]; 2014: 1–9. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 86.  Dustin D, Gu G, Fuqua SAW: ESR1 mutations in breast cancer. Cancer. 2019 Nov [cited 2024 Apr 9]; 125(21): 3714–3728. PubMed Abstract | Publisher Full Text | Free Full Text
  • 87.  Zhang H, Suo B, Sun XP, et al.: Bardet-Biedl Syndrome 4 in Early Diagnosis and Prognosis of Breast Cancer. Indian J. Pharm. Sci. 2021 [cited 2024 Apr 9]; 83. Publisher Full Text Reference Source
  • 88.  Cicirò Y, Sala A: MYB oncoproteins: emerging players and potential therapeutic targets in human cancer. Oncogenesis. 2021 Feb 26 [cited 2024 Apr 9]; 10(2): 19. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 89.  Lee RS, Sad K, Fawwal DV, et al.: Emerging Role of Epigenetic Modifiers in Breast Cancer Pathogenesis and Therapeutic Response. Cancers. 2023 Aug 7 [cited 2024 Apr 9]; 15(15): 4005. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 90.  Özcan G: SCUBE2 as a Marker of Resistance to Taxane-based Neoadjuvant Chemotherapy and a Potential Therapeutic Target in Breast Cancer. Eur. J. Breast Health. 2022 Dec 27 [cited 2024 Apr 9]; 19(1): 45–54. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 91.  Elmi A, McDonald ES, Mankoff D: Imaging Tumor Proliferation in Breast Cancer. PET Clin. 2018 Jul [cited 2024 Mar 10]; 13(3): 445–457. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 92.  Hicks DG: Molecular Pathology of Breast Cancer. Cell and Tissue Based Molecular Pathology. Elsevier; 2009 [cited 2024 Mar 11]; pp. 360–378. Reference Source
  • 93.  Bertucci F, Finetti P, Birnbaum D: Basal Breast Cancer: A Complex and Deadly Molecular Subtype. Curr. Mol. Med. 2012 Jan 1 [cited 2024 Mar 11]; 12(1): 96–110. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 94.  Ding R, Wang Y, Fan J, et al.: Identification of immunosuppressive signature subtypes and prognostic risk signatures in triple-negative breast cancer. Front. Oncol. 2023 Jun 12 [cited 2024 Mar 11]; 13: 1108472. PubMed Abstract | Publisher Full Text | Free Full Text
  • 95.  Sørlie T, Tibshirani R, Parker J, et al.: Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc. Natl. Acad. Sci. 2003 Jul 8 [cited 2024 Apr 8]; 100(14): 8418–8423. PubMed Abstract | Publisher Full Text | Free Full Text
  • 96.  Sørlie T, Perou CM, Tibshirani R, et al.: Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc. Natl. Acad. Sci. 2001 Sep 11 [cited 2024 Apr 8]; 98(19): 10869–10874. PubMed Abstract | Publisher Full Text | Free Full Text
  • 97.  Roll JD, Rivenbark AG, Sandhu R, et al.: Dysregulation of the epigenome in triple-negative breast cancers: Basal-like and claudin-low breast cancers express aberrant DNA hypermethylation. Exp. Mol. Pathol. 2013 Dec [cited 2024 Mar 10]; 95(3): 276–287. PubMed Abstract | Publisher Full Text Reference Source
  • 98.  The Cancer Genome Atlas Network. Comprehensive molecular portraits of human breast tumours. Nature. 2012 Oct [cited 2024 Mar 8]; 490(7418): 61–70. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 99.  Huang S, Xu W, Hu P, et al.: Integrative Analysis Reveals Subtype-Specific Regulatory Determinants in Triple Negative Breast Cancer. Cancers. 2019 Apr 10 [cited 2024 Mar 8]; 11(4): 507. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 100.  Kumaran M, Cass CE, Graham K, et al.: Germline copy number variations are associated with breast cancer risk and prognosis. Sci. Rep. 2017 Nov 7 [cited 2024 Mar 10]; 7(1): 14621. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 101.  Li X, Zhou J, Xiao M, et al.: Uncovering the Subtype-Specific Molecular Characteristics of Breast Cancer by Multiomics Analysis of Prognosis-Associated Genes, Driver Genes, Signaling Pathways, and Immune Activity. Front. Cell Dev. Biol. 2021 Jul 1 [cited 2024 Mar 10]; 9: 689028. PubMed Abstract | Publisher Full Text | Free Full Text
  • 102.  Kothari C, Diorio C, Durocher F: The Importance of Breast Adipose Tissue in Breast Cancer. Int. J. Mol. Sci. 2020 Aug 11 [cited 2024 Mar 10]; 21(16): 5760. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 103.  Sieuwerts AM, Kraan J, Bolt J, et al.: Anti-Epithelial Cell Adhesion Molecule Antibodies and the Detection of Circulating Normal-Like Breast Tumor Cells. JNCI. J. Natl. Cancer Inst. 2009 Jan 7 [cited 2024 Mar 11]; 101(1): 61–66. PubMed Abstract | Publisher Full Text | Free Full Text
  • 104.  Wu J, Liu S, Liu G, et al.: Identification and functional analysis of 9p24 amplified genes in human breast cancer. Oncogene. 2012 Jan 19 [cited 2024 Mar 8]; 31(3): 333–341. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 105.  Lesurf R, Aure MR, Mørk HH, et al.: Molecular Features of Subtype-Specific Progression from Ductal Carcinoma In Situ to Invasive Breast Cancer. Cell Rep. 2016 Jul [cited 2024 Mar 10]; 16(4): 1166–1179. PubMed Abstract | Publisher Full Text Reference Source
  • 106.  Sun W, Bunn P, Jin C, et al.: The association between copy number aberration, DNA methylation and gene expression in tumor samples. Nucleic Acids Res. 2018 Apr 6 [cited 2024 Mar 8]; 46(6): 3009–3018. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 107.  Gamazon ER, Stranger BE: The impact of human copy number variation on gene expression: Figure 1. Brief Funct. Genomics. 2015 Sep [cited 2024 Mar 10]; 14(5): 352–357. PubMed Abstract | Publisher Full Text | Free Full Text
  • 108.  Ma L, Li C, Yin H, et al.: The Mechanism of DNA Methylation and miRNA in Breast Cancer. Int. J. Mol. Sci. 2023 May 27 [cited 2024 Mar 10]; 24(11): 9360. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 109.  Singhal SK, Usmani N, Michiels S, et al.: Towards understanding the breast cancer epigenome: a comparison of genome-wide DNA methylation and gene expression data. Oncotarget. 2016 Jan 19 [cited 2024 Mar 10]; 7(3): 3002–3017. PubMed Abstract | Publisher Full Text | Free Full Text
  • 110.  Fragoza R, Das J, Wierbowski SD, et al.: Extensive disruption of protein interactions by genetic variants across the allele frequency spectrum in human populations. Nat. Commun. 2019 Sep 12 [cited 2024 Mar 15]; 10(1): 4141. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 111.  Urbanova M, Buocikova V, Trnkova L, et al.: DNA Methylation Mediates EMT Gene Expression in Human Pancreatic Ductal Adenocarcinoma Cell Lines. Int. J. Mol. Sci. 2022 Feb 14 [cited 2024 Mar 8]; 23(4): 2117. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 112.  Zhao M, Liu Y, Qu H: Expression of epithelial-mesenchymal transition-related genes increases with copy number in multiple cancer types. Oncotarget. 2016 Apr 26 [cited 2024 Mar 8]; 7(17): 24688–24699. PubMed Abstract | Publisher Full Text | Free Full Text
  • 113.  Haider Z, Landfors M, Golovleva I, et al.: DNA methylation and copy number variation profiling of T-cell lymphoblastic leukemia and lymphoma. Blood Cancer J. 2020 Apr 28 [cited 2024 Mar 8]; 10(4): 45. Reference Source
  • 114.  Kim SY, Choe EK, Shivakumar M, et al.: Multi-layered network-based pathway activity inference using directed random walks: application to predicting clinical outcomes in urologic cancer. Bioinformatics. 2021 Aug 25 [cited 2024 Mar 8]; 37(16): 2405–2413. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 115.  Sammut SJ, Crispin-Ortuzar M, Chin SF, et al.: Multi-omic machine learning predictor of breast cancer therapy response. Nature. 2022 Jan 27 [cited 2024 Mar 15]; 601(7894): 623–629. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 116.  Liu J, Lichtenberg T, Hoadley KA, et al.: An Integrated TCGA Pan-Cancer Clinical Data Resource to Drive High-Quality Survival Outcome Analytics. Cell. 2018 Apr [cited 2024 Mar 11]; 173(2): 400–416.e11. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 117.  Liu MC, Pitcher BN, Mardis ER, et al.: PAM50 gene signatures and breast cancer prognosis with adjuvant anthracycline- and taxane-based chemotherapy: correlative analysis of C9741 (Alliance). Npj Breast Cancer. 2016 Jan 6 [cited 2024 Mar 8]; 2(1): 15023. Reference Source
  • 118.  Reel PS, Reel S, Van Kralingen JC, et al.: Machine learning for classification of hypertension subtypes using multi-omics: A multi-centre, retrospective, data-driven study. EBioMedicine. 2022 Oct [cited 2024 Apr 9]; 84: 104276. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 119.  Wu F, Yin YY, Fan WH, et al.: Immunological profiles of human oligodendrogliomas define two distinct molecular subtypes. EBioMedicine. 2023 Jan [cited 2024 Apr 9]; 87: 104410. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 120.  Pang J, Liang B, Ding R, et al.: A denoised multi-omics integration framework for cancer subtype classification and survival prediction. Brief. Bioinform. 2023 Sep 20 [cited 2024 Apr 9]; 24(5): bbad304. PubMed Abstract | Publisher Full Text
  • 121.  Liu W, Wang W, Zhang H, et al.: Development and Validation of Multi-Omics Thymoma Risk Classification Model Based on Transfer Learning. J. Digit. Imaging. 2023 Jun 2 [cited 2024 Apr 9]; 36(5): 2015–2024. PubMed Abstract | Publisher Full Text | Free Full Text
  • 122.  Ochoa S, De Anda-Jáuregui G, Hernández-Lemus E: Multi-Omic Regulation of the PAM50 Gene Signature in Breast Cancer Molecular Subtypes. Front. Oncol. 2020 May 22 [cited 2024 Mar 8]; 10: 845. PubMed Abstract | Publisher Full Text | Free Full Text
  • 123.  Ochoa S, Hernández-Lemus E: Functional impact of multi-omic interactions in breast cancer subtypes. Front. Genet. 2023 Jan 5 [cited 2024 Apr 9]; 13: 1078609. PubMed Abstract | Publisher Full Text | Free Full Text
  • 124.  Chen YX, Chen H, Rong Y, et al.: An integrative multi-omics network-based approach identifies key regulators for breast cancer. Comput. Struct. Biotechnol. J. 2020 [cited 2024 Apr 9]; 18: 2826–2835. PubMed Abstract | Publisher Full Text | Free Full Text Reference Source
  • 125.  Zacksenhaus E, Liu JC, Jiang Z, et al.: Transcription Factors in Breast Cancer—Lessons From Recent Genomic Analyses and Therapeutic Implications. Advances in Protein Chemistry and Structural Biology. Elsevier; 2017 [cited 2024 Mar 10]; pp. 223–273. Reference Source
  • 126.  Chen H, Xie G, Luo Q, et al.: Regulatory miRNAs, circRNAs and lncRNAs in cell cycle progression of breast cancer. Funct. Integr. Genomics. 2023 Sep [cited 2024 Mar 8]; 23(3): 233. PubMed Abstract | Publisher Full Text
  • 127.  Zhuang J, Huo Q, Yang F, et al.: Perspectives on the Role of Histone Modification in Breast Cancer Progression and the Advanced Technological Tools to Study Epigenetic Determinants of Metastasis. Front. Genet. 2020 Oct 29 [cited 2024 Mar 10]; 11: 603552. PubMed Abstract | Publisher Full Text | Free Full Text
  • 128.  Arakelyan A, Davitavyan S: Integrated analysis of “-omic” landscapes in breast cancer subtypes: Supplementary Dataset. [Data set]. Zenodo. 2024. 2024. Publisher Full Text

Comments on this article Comments (0)

Version 1
VERSION 1 PUBLISHED 03 Jun 2024
Comment
Author details Author details
Competing interests
Grant information
Article Versions (1)
Copyright
Download
 
Export To
metrics
Views Downloads
F1000Research - -
PubMed Central
Data from PMC are received and updated monthly.
 - -
Citations
SEE MORE DETAILS
CITE
how to cite this article
Davitavyan S, Martirosyan G, Mkrtchyan G et al. Integrated analysis of -omic landscapes in breast cancer subtypes [version 1; peer review: 2 approved with reservations]. F1000Research 2024, 13:564 (https://doi.org/10.12688/f1000research.148778.1)
NOTE: If applicable, it is important to ensure the information in square brackets after the title is included in all citations of this article.
track
receive updates on this article
Track an article to receive email alerts on any updates to this article.

Open Peer Review

Current Reviewer Status: ?
Key to Reviewer Statuses VIEW
ApprovedThe paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approvedFundamental flaws in the paper seriously undermine the findings and conclusions
Version 1
VERSION 1
PUBLISHED 03 Jun 2024
Views
4
Cite
Reviewer Report 26 Aug 2024
Brian D. Lehmann, Vanderbilt University Medical Center, Tennessee, USA 
Approved with Reservations
VIEWS 4
https://doi.org/10.5256/f1000research.163134.r310839
In the manuscript by Davitavyan et. al, the authors apply self-organizing map algorithms to integrate gene expression, methylation, copy number and somatic mutations to divide breast cancers from the TCGA into new classifications. They characterized genes and pathways associated with ... Continue reading
CITE
CITE
HOW TO CITE THIS REPORT
Lehmann BD. Reviewer Report For: Integrated analysis of -omic landscapes in breast cancer subtypes [version 1; peer review: 2 approved with reservations]. F1000Research 2024, 13:564 (https://doi.org/10.5256/f1000research.163134.r310839)
The direct URL for this report is:
https://f1000research.com/articles/13-564/v1#referee-response-310839
NOTE: it is important to ensure the information in square brackets after the title is included in all citations of this article.
Report a concern
Views
15
Cite
Reviewer Report 20 Jun 2024
Stepan Nersisyan, Thomas Jefferson University,, Philadelphia, USA 
Approved with Reservations
VIEWS 15
https://doi.org/10.5256/f1000research.163134.r288456
The paper by Suren Davitavyan with co-authors presents a multi-omics analysis of breast cancer subtypes. The authors trained multi-layer self-organizing maps on four data modalities: gene expression, methylation, CNV, and mutation data from TCGA. They comprehensively characterize genes and ... Continue reading
CITE
CITE
HOW TO CITE THIS REPORT
Nersisyan S. Reviewer Report For: Integrated analysis of -omic landscapes in breast cancer subtypes [version 1; peer review: 2 approved with reservations]. F1000Research 2024, 13:564 (https://doi.org/10.5256/f1000research.163134.r288456)
The direct URL for this report is:
https://f1000research.com/articles/13-564/v1#referee-response-288456
NOTE: it is important to ensure the information in square brackets after the title is included in all citations of this article.
Report a concern

Comments on this article Comments (0)

Version 1
VERSION 1 PUBLISHED 03 Jun 2024
Comment

Open Peer Review

Reviewer Status

Alongside their report, reviewers assign a status to the article:

Approved
The paper is scientifically sound in its current form and only minor, if any, improvements are suggested
Approved with reservations
A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
Not approved
Fundamental flaws in the paper seriously undermine the findings and conclusions

Reviewer Reports

Invited Reviewers
1 2
Version 1
03 Jun 24 		read read
  1. Stepan Nersisyan, Thomas Jefferson University,, Philadelphia, USA
  2. Brian D. Lehmann, Vanderbilt University Medical Center, Tennessee, USA

Comments on this article

All Comments(0)

Add a comment
Sign up for content alerts

Browse by related subjects

    keyboard_arrow_left Back to all reports
    keyboard_arrow_left Back to all reports
    Alongside their report, reviewers assign a status to the article:
    Approved - the paper is scientifically sound in its current form and only minor, if any, improvements are suggested
    Approved with reservations - A number of small changes, sometimes more significant revisions are required to address specific details and improve the papers academic merit.
    Not approved - fundamental flaws in the paper seriously undermine the findings and conclusions
    Sign In
    If you've forgotten your password, please enter your email address below and we'll send you instructions on how to reset your password.

    The email address should be the one you originally registered with F1000.
    Email address not valid, please try again

    You registered with F1000 via Google, so we cannot reset your password.

    To sign in, please click here.

    If you still need help with your Google account password, please click here.

    You registered with F1000 via Facebook, so we cannot reset your password.

    To sign in, please click here.

    If you still need help with your Facebook account password, please click here.

    Code not correct, please try again
    Email us for further assistance.
    Server error, please try again.