The authors have declared that no competing interests exist.
Edited the MS: FL LC CLS SC. Coordination of tissue collection and clinical databases: CLS. Conceived and designed the experiments: KH CLS PG. Performed the experiments: GH PG FL SC. Analyzed the data: LC AP AF. Contributed reagents/materials/analysis tools: HA AV CMC. Wrote the paper: KH PG.
¶ These authors also contributed equally to this work.
Triple negative breast cancer (TNBC) is a heterogeneous disease at the molecular, pathologic and clinical levels. To stratify TNBCs, we determined microRNA (miRNA) expression profiles, as well as expression profiles of a cancer-focused mRNA panel, in tumor, adjacent non-tumor (normal) and lymph node metastatic lesion (mets) tissues, from 173 women with TNBCs; we linked specific miRNA signatures to patient survival and used miRNA/mRNA anti-correlations to identify clinically and genetically different TNBC subclasses. We also assessed miRNA signatures as potential regulators of TNBC subclass-specific gene expression networks defined by expression of canonical signal pathways.
Tissue specific miRNAs and mRNAs were identified for normal
Our findings suggest that miRNAs play a key role in triple negative breast cancer through their ability to regulate fundamental pathways such as: cellular growth and proliferation, cellular movement and migration, Extra Cellular Matrix degradation. The results define miRNA expression signatures that characterize and contribute to the phenotypic diversity of TNBC and its metastasis.
miRNAs are small (19–25 nucleotides), non-coding RNAs that reduce the abundance and translational efficiency of mRNAs and play a major role in regulatory networks, influencing diverse biological processes
Triple-negative breast cancers are defined by a lack of expression of estrogen receptor (ESR1), progesterone receptor (PGR1), and ERBB2 receptor. This subgroup accounts for 15% of all types of breast cancer and is an aggressive form with limited treatment options.
We have used the nanoString nCounter platform (Seattle, WA, USA) to profile both miRNA and mRNA expression (nanoString Cancer Reference panel) using the same RNA sample from each breast cancer patient. Our analyses confirmed some observations from previous studies
We also assessed, through target prediction analysis, if the miRNA signatures can potentially regulate and thus define pivotal pathways in mRNA defined TNBC subclasses. Moreover we were able to correlate tissue specific miRNA expression signatures, defined by comparisons among normal, tumor and metastatic tissues, with specific dysregulated mRNAs in the same comparisons. The analyses confirm the heterogeneity of TNBC and provide a basis for further molecular studies to develop miRNA-based early detection markers and novel therapeutic targets for triple negative breast cancer.
The profiles of 224 samples, 165 primary cancer-derived RNAs and 59 normal RNAs from patients with a median age of 51 years were considered. Hierarchical clustering represented in the heat map in
miR-106b
Hierarchical clustering of the subset of 55 matched tumor and normal tissue RNAs is shown in
Several miRNAs have been shown to be involved in breast metastasis induction and progression, through processes such as epithelial-mesenchymal transition (EMT), extracellular matrix modification (ECM) and mesenchymal-epithelial transition (MET)
The Venn diagram (
Venn diagram (
Examples of up-regulated miRNAs through the three tissue classes are: miR-128 (logFC 1.51, 1.15, 2.66) targeting
We investigated associations between miRNA expression levels and survival, for the entire TNBC cohort, as well as for the >50 years of age patients (50+, mostly post-menopausal) and the 50 years and under patients (50−, mostly premenopausal, but including some patients who had hysterectomies). Here, we present a detailed analysis of the 50− cohort, while further investigations for the 50+ subset are ongoing. For the 50− patients, the median follow-up was 79 months (range 9–194 mo), the median age was 43 (range 20–50 yrs). Censoring occurred at the date of death from any causes (overall survival, OS), first evidence of distant recurrence (distant-disease free survival, DDFS) or at time of the last known follow-up, whichever occurred first. All expressed miRNAs (n = 133) in tumor samples were considered. For both OS and DDFS, we used Cox proportional hazards models and identified sets of miRNAs that are significantly related to outcomes. We then performed permutation tests in which the times and censoring indicators were randomly permuted among samples. Permutation P values for significant miRNAs were computed based on 10,000 random permutations. Hazard ratios (HR) were computed for a 2-fold change in the miRNA expression level. 4 miRNAs were significantly associated with OS, as determined by univariate and multivariate analysis. Of these, 3 were up-regulated and 1 down-regulated in the normal
Overall survival (OS) of TNBC patients of 50 yrs and younger patients due to differentially expressed miRNAs in the three classes. (
For DDFS the median follow-up was 75 months (range 6–194 mo.). 7 miRNAs were significantly associated with DDFS, as determined by univariate and multivariate analysis. These 7 miRNAs were significantly differentially expressed in the normal
The nanoString GX Human mRNA Cancer Reference panel was used to profile expression in 158 tumors, 40 adjacent normal tissues and 54 lymph node mets. The profiles discriminated non-tumor tissue from TN tumors and mets. Hierarchical clustering represented in the heat maps (
The heat maps represent hierarchical clustering of differentially expressed genes in normal and tumor-derived RNAs. mRNA profiles are clustered in 4 different subgroups (orange, blue, yellow, pink) defined by the mRNA expression patterns. Overlapping Gene Ontology terms for top canonical pathways represented by the differentially expressed genes in each subgroup, as determined by IPA-ingenuity software, are shown on the right for each of the normal/tumor comparison-defined subgroups.
miRNA Expression | Gene Expression | microRNA∶mRNA | ||
miRNA | logFC | Gene Symbol | logFC | Expression Correlation |
hsa-miR-16 | 5.73 |
|
−0.48 | −0.13 |
|
−1.54 | −0.14 | ||
|
−1.31 | −0.16 | ||
|
−2.27 | −0.34 | ||
|
−3.38 | −0.35 | ||
|
−0.70 | −0.32 | ||
|
−1.31 | −0.25 | ||
|
−1.47 | −0.39 | ||
|
−1.53 | −0.47 | ||
|
−0.95 | −0.38 | ||
|
−1.88 | −0.36 | ||
|
−1.99 | −0.37 | ||
|
−2.00 | −0.31 | ||
|
−1.34 | −0.31 | ||
|
−1.75 | −0.48 | ||
|
−1.66 | −0.37 | ||
|
−3.63 | −0.41 | ||
|
−2.09 | −0.37 | ||
|
−2.34 | −0.43 | ||
|
−3.34 | −0.45 | ||
hsa-miR-125 | −2.03 |
|
0.75 | −0.18 |
hsa-miR-374a | 2.83 |
|
−0.48 | −0.15 |
|
−1.28 | −0.36 | ||
|
−1.17 | −0.36 | ||
|
−1.88 | −0.22 | ||
|
−0.70 | −0.27 | ||
|
−1.31 | −0.10 | ||
|
−1.43 | −0.34 | ||
|
−3.38 | −0.21 | ||
|
−0.85 | −0.21 | ||
|
−1.95 | −0.27 | ||
|
−1.27 | −0.26 | ||
|
−3.22 | −0.24 | ||
|
−2.07 | −0.18 | ||
|
−1.45 | −0.21 | ||
|
−1.08 | −0.15 | ||
|
−1.54 | −0.14 | ||
hsa-miR-374b | 1.92 |
|
−0.48 | −0.16 |
|
−1.28 | −0.43 | ||
|
−1.17 | −0.37 | ||
|
−1.88 | −0.30 | ||
|
−0.70 | −0.27 | ||
|
−1.31 | −0.16 | ||
|
−1.43 | −0.35 | ||
|
−3.38 | −0.32 | ||
|
−0.85 | −0.21 | ||
|
−1.95 | −0.42 | ||
|
−1.27 | −0.36 | ||
|
−3.22 | −0.40 | ||
|
−2.07 | −0.32 | ||
|
−1.45 | −0.29 | ||
|
−1.08 | −0.31 | ||
|
−1.54 | −0.25 | ||
hsa-miR-421 | 3.36 |
|
−1.17 | −0.14 |
|
−1.34 | −0.26 | ||
|
−0.70 | −0.21 | ||
|
−0.79 | −0.23 | ||
|
−0.99 | −0.25 | ||
|
−1.20 | −0.27 | ||
|
−3.34 | −0.45 | ||
|
−0.49 | −0.16 | ||
|
−2.41 | −0.48 | ||
|
−2.12 | −0.45 | ||
|
−1.14 | −0.24 | ||
|
−1.83 | −0.40 | ||
hsa-miR-655 | −1.56 |
|
0.75 | −0.07 |
The degree of anti-correlation among the mRNA–miR pairs is calculated by Pearson correlation. The predicted targeted-anti-correlated genes are shown.
We note that the ESR1 probe is in the nanoString cancer mRNA panel and very low expression of this mRNA was detected in 7 TNBCs, as shown in
Hierarchical clustering of a subset composed of 40 normal and 50 regional lymph node mets RNAs (
Venn diagram (
Stratification of TNBC into subclasses using new markers will identify new screening methods, prognostic factors, methodologies and perhaps targets for personalized therapies. Several recent studies have correlated miRNA expression with outcomes in TNBC using microarray or other high throughput technologies. mRNA expression profiles that sub-classify TNBCs have also been reported in association with investigations of outcome, new molecular pathways and possible chemotherapy alternatives
We first profiled miRNAs expressed in specific tissue classes for the entire cohort. Subsequent focus on the 50 and under subset allowed identification of 2 miRNA signatures, distinguishing ‘protective’ and ‘risky’ prognostic miRNAs; these prognostic signatures require validation on an independent group of TNBC patients and this work is in progress. The primary tumor mRNA expression profiles of the entire cohort clustered into 4 molecular subgroups. Through miRNA target prediction and use of IPA software, we determined that the prognostic signature miRNAs were connected with the 4 molecular subclasses, in which they are predicted to target many transcripts and participate in control of their canonical signal pathways.
13 miRNAs (
In tumor
miR-125a, the only miRNA significantly dysregulated only in the normal
Using univariate and multivariate Cox analysis we investigated the correlation of differentially expressed miRNAs in TNBC samples with OS and DDFS, in a subset of cases (a somewhat homogeneous cohort of patients aged 20–50 yrs) and observed 4 miRNA and 7 miRNA signatures prognostic for OS and DDFS, respectively, with ‘protective’ miR-16, 374a and ‘risk-associated’ 125b included in both signatures. In invasive breast cancers, association between down-regulation of miR-125b and poor survival has been reported
Seven miRNAs were prognostic for DDFS. miR-497, a down-regulated member of the miR-16 family, has a ‘protective’ function. It was reported to be down-regulated in breast cancer, partially due to DNA methylation
The small amount of RNA available necessitated assessment of mRNA expression of a limited set of genes represented by the nanoString Cancer Reference panel, a limitation in this molecular profiling study. Nevertheless, we observed dysregulated expression profiles among these 230 mRNAs in TNBC. The 124 mRNAs that were significantly deregulated clustered into 4 color-coded molecular subgroups representing the most specifically perturbed pathways in TNBC.
Gene expression clusters and IPA Ingenuity analysis gave an overview of critical pathways involved in this TN cohort. As expected in a tumor cohort (orange subgroup), processes such as cell cycle, DNA damage check point, anti-apoptosis, ECM degradation, cell growth and proliferation are turned on; more surprising was the down-regulation of known cancer-associated pathways detected in the other three subgroups. The oncogenes
Analysis of the mRNA-defined molecular subgroups, together with the OS and DDFS prognostic miRNAs, exploiting the Pearson correlation coefficient, allowed definition of correlations of most of the prognostic miRNAs (excepting miR-155 and 497) with their predicted gene targets lending strong support to the results. Though these miRNAs are reportedly involved in breast cancer, none has previously been associated with OS or DDFS. Only two, miR-125b and -16, have been shown to contribute to drug resistance in breast cancer; further investigation of the signature miRNAs is ongoing.
In comparisons of mRNA expression patterns in the three tissue classes, the metastatic tissue group gave interesting results. Up-regulation of
Interestingly, our novel findings concerning expression levels of
In conclusion, in TNBC, integrated miRNA-mRNA profiling can distinguish the different breast tissue “stages”: tumor adjacent, tumor and lymph node metastasis. miRNA signatures can be prognostic markers for OS and DDFS, suggesting that therapies tailored to these markers may contribute to improved survival.
mRNA profiling also emphasized the heterogeneity of TNBC, as exemplified by the varying signal pathways driving tumors and metastasis of the mRNA-defined subgroups.
An IRB-approved OSU protocol for this research linked clinical features, treatment and outcome data of breast cancer patients in the OSU National Comprehensive Cancer Network breast cancer database/tumor registry with archrival breast cancer pathology specimens stored in the OSU Tissue Archive Service using the Information Warehouse at OSUMC to serve as “honest broker” and provided de-identified clinic-pathological information. From 1995–2005, a cohort of 365 consecutive triple negative localized breast cancer patients were identified. After pathology review for tumors with sufficient sample for study, the only selection criterion, 173 paraffin blocks for TNBCs were identified for preparation of a tissue microarray and cores for RNA preparation, with the characteristics shown in the demographics summary in
Characteristic | All (173) | >50 yr (87) | <50 yr (86) | |
|
Caucasian | 153 | 82 | 71 |
African American | 16 | 4 | 12 | |
Other | 4 | 1 | 3 | |
|
Pre-menopausal | 64 | 3 | 61 |
Post-menopausal | 103 | 81 | 22 | |
Unknown | 6 | 3 | 3 | |
|
I | 2 | 0 | 2 |
II | 15 | 10 | 5 | |
III | 148 | 76 | 72 | |
IV | 2 | 0 | 2 | |
Unknown | 6 | 1 | 5 | |
|
Yes | 78 | 42 | 36 |
No | 95 | 45 | 50 | |
|
Positive | 102 | 72 | 30 |
Negative | 62 | 12 | 50 | |
Unknown | 9 | 3 | 6 | |
|
< = 40 | 34 | ||
41–50 | 52 | |||
> = 51 | 87 | |||
|
Yes | 59 | 30 | 29 |
No | 114 | 57 | 57 | |
|
No | 126 | 67 | 59 |
Yes | 47 | 20 | 27 | |
|
In situ | 1 | 0 | 1 |
Local/Regional | 3 | 1 | 2 | |
Distant | 35 | 15 | 20 | |
Type unknown | 8 | 4 | 4 |
Does not include the cases that were never disease-free, those unknown if ever disease-free and those with missing recurrence information. The demographic features of the original 365 TNBCs were nearly identical to those shown above for the 173 TNBCs analyzed.
RNA was isolated from formalin-fixed paraffin-embedded tissue of 165 tumor, 59 tumor-associated, adjacent normal and 54 associated lymph node mets tissues, using the Recover ALL kit (Ambion). Due to small amounts of RNA available from the 251 formalin-fixed paraffin-embedded cores, we have used the nanoString nCounter human miRNA expression profiling v1 panel and mRNA cancer panel, which allows profiling from 100 ng of RNA per sample
RNAs were processed with the nanoString nCounter system (nanoString, Seattle, Washington, USA) in the Nucleic Acid Shared Resource of The Ohio State University. The miRNA panel detects 664 endogenous miRNAs (with 654 probes), 82 putative viral miRNAs, and five housekeeping transcripts.
For analysis of mRNA expression, the nanoString GX Human mRNA Cancer Reference panel, that includes tags specific for 230 cancer-related mRNAs (
The ‘Core Analysis’ function included in the Ingenuity Pathways Analysis (IPA) software (
Raw expression data were log-transformed and normalized by the quantile method after application of a manufacturer-supplied correction factor for several miRNAs. Data were filtered to exclude relatively invariant features (IQR = 0.5) and features below the detection threshold (defined for each sample by a cutoff corresponding to twice standard deviation of negative control probes plus the means) in at least half of the samples. Using R/Bioconductor and the filtered dataset, linear models for microarray data analysis were employed with a contrast matrix for the following comparisons: normal vs tumor, normal vs mets, tumor vs mets. P values were used to rank miRNAs of interest, and correction for multiple comparisons was done by the Benjamini-Hocheberg method. Correlations were determined using the Pearson correlation coefficient (r). The mean time to first relapse was compared between groups using the rank sum test. Analysis of OS and DDFS, was performed by the Kaplan-Meier method, and comparisons of outcomes among subgroups were performed by using the long-rank test. Two-tailed tests were used for univariate comparisons. For univariate and multivariate analysis of prognostic factors (using tumor grade and age as covariates), a Cox proportional hazard regression model was used. Data processing and analysis were conducted using BRB-ArrayTools
To investigate the differences among the gene expression profiles detected by the nanoString GX Human mRNA Cancer Reference Kit, we performed hierarchical clustering using the 124 dysregulated genes (P-value<0.05) in the entire Normal versus Tumor samples dataset. Two-dimensional average-linkage hierarchical clustering of a Spearman rank correlation similarity matrix of the primary tumors and normal samples was performed. All gene expression analyses were performed using R software (version 2.13.0). As expected, we identified two distinct patient clusters (Normal and Tumor samples) and four distinct gene clusters designated by different colors (
All fold-changes associated with these analyses are represented in log2 scale (logFC) and we show only data with a P-value of <0.05, considered to indicate statistical significance.
The miRNA and mRNA expression data have been submitted to the Gene Expression Omnibus (GEO) with accession number GSE 41970.
To validate the study findings, three approaches were used: first we validated the deregulated miRNAs “
Lastly, in a subset of samples (randomly chosen based on availability of RNAs) we were able to validate the expression levels of a subset of miRNAs (7 differentially expressed miRNAs reported in
cDNA was reverse transcribed from 10 ng of total RNA of each sample using specific miRNA primers from the TaqMan® MicroRNA Assays and reagents from the TaqMan® MicroRNA reverse Transcription Kit, Life Technologies (Grand Island, NY). Subsequently, in the PCR step, PCR products are amplified from cDNA samples using the TaqMan® MicroRNA Assays together with the TaqMan® Universal PCR Master Mix. All the assays were performed in triplicate according to the manufacturer's instructions.
Hierarchical clustering of miRNA expression patterns of tumor and normal samples. Heat map representing miRNA profiles of 165 tumor and 59 normal samples using average linkage clustering and Spearman Rank method as distance metrics. Bar above the dendrogram identifies the samples, normal shown in light blue and tumors in yellow. Samples are shown in columns, miRNAs in rows. Heat map from blue to red represent relative miRNA expression as indicated in the key bar at the top.
(PDF)
Clustering of miRNA expression patterns of paired tumor and normal samples. Heat map representing miRNA profiles of 55 tumor and 55 paired normal samples using average linkage clustering and Spearman Rank method as distance metrics. A bar above the dendrogram identifies the samples, tumors shown in yellow and normal light blue. Samples are shown in columns, miRNAs in rows.
(PDF)
Clustering of miRNA expression patterns of normal and metastatic RNAs. Heat map representing miRNA profiles of 54 metastatic and 59 normal samples using average linkage clustering and Spearman Rank method as distance metrics. Samples are shown in columns, miRNAs in rows. A bar above the dendrogram identifies the samples, metastases in purple and normal in light blue.
(PDF)
Comparison of mRNA expression profiles of normal vs metastasis-derived RNAs. The heat map representing expression patterns of 120 mRNAs in 40 normal and 50 metastasis-derived RNAs, using average linkage clustering and Spearman Rank methods as distance metrics. A bar above the dendrogram identifies the samples, Metastatic RNAs shown in purple and normal in light blue.
(PDF)
qRT-PCR validation. Box plots represent expression of 7 deregulated miRNAs in a representative subset of samples of the three tissue groups, assayed by TaqMan® qRT-PCR.
(PDF)
20 deregulated miRNAs in matched Normal
(XLSX)
103 deregulated miRNAs in the comparison Normal
(XLSX)
15 miRNAs differentially expressed across the three tissue group comparisons.
(XLSX)
Dysregulated mRNAs in Normal
(XLSX)
We thank Jason Bacher of the OSU Pathology Core Facility for preparation of cores for RNA isolation, David Kellough of the OSU Human Tissue Resource Network, Department of Pathology, for excellent organization of tissue resources for this study. We also thank Paolo Fadda for excellent assistance in preparing RNAs for and execution of nanoString nCounter assays and Teresa Druck for assistance with MS editing and illustration preparation.