Introduction
MicroRNAs (miRNAs) are a group of small (20 to 25 nt) noncoding RNAs able to regulate gene expression posttranscriptionally by binding to the 3'-untranslated region (UTR) of target mRNAs [
1‐
3]. Since the initial discovery in
Caenorhabditis elegans, more than 1,000 human miRNAs have been described, each of them targeting about 100 different mRNA molecules [
4‐
6]. In this way, approximately 30% of all human genes are regulated by miRNAs [
7,
8], thereby influencing several different pathways and processes in the cell, including development, differentiation, apoptosis, and cell proliferation [
9‐
11].
As miRNAs are involved in fine-tuning gene expression in the cell [
1,
2], deregulation of miRNA expression could lead to altered gene expression, which might contribute to the development of cancer [
12]. Several studies have shown a differential miRNA-expression profile in cancer as compared with normal controls [
13‐
15]. Although specific miRNAs can be upregulated in cancer [
16], global miRNA downregulation is a common trait of human malignancies [
13,
17]. Furthermore, miRNAs are involved in the metastatic cascade, which is the most dismal feature of tumor biology with respect to patient prognosis.
MiRNA-expression profiling of primary tumor samples and their associated metastases identified both prometastatic and metastasis-suppressor miRNAs [
15]. These miRNAs modulate the expression of metastasis-associated genes [
18,
19], both directly and indirectly, by influencing the epigenetic machinery [
20].
Breast cancer is the most frequent carcinoma and the second most common cause of cancer-related mortality in women [
21]. In the past decade, it has been repeatedly shown that breast cancer is a heterogeneous condition consisting of at least five [
22] but possibly more [
23,
24] molecular subtypes. These molecular subtypes (Luminal A, Luminal B, Basal-like, ErbB2+, and Normal-like) are characterized by specific mRNA-expression profiles. Blenkiron and colleagues [
14] showed that these specific mRNA-expression profiles are at least partially attributable to differential miRNA expression. Also, Iorio and colleagues [
25] identified a global pattern of miRNA deregulation in breast cancer tissue when compared with normal breast tissue, hinting at the importance of miRNA deregulation in the development of breast cancer in general.
As miRNAs appear to be critical regulators of tumor biology, their potential as prognostic and predictive biomarkers has recently been given attention. In addition, their great stability when compared with mRNA molecules, both in blood samples and in formalin-fixed, paraffin-embedded tissue samples, offers a great advantage [
26,
27]. Levels of miRNAs do not substantially change when serum or plasma samples are subjected to freeze-thaw cycles, boiling, or maintenance at room temperature [
28,
29]. As the bloodstream is easily accessible, blood-borne miRNAs or circulating miRNAs hold the potential to serve as noninvasive biomarkers in oncology.
Recently, Heneghan and colleagues [
30] showed that miRNA expression is detectable in whole blood, plasma, and the serum of cancer patients and healthy controls. In addition, miRNA-195 was identified as a potential biomarker for detecting noninvasive and early-stage breast disease [
30].
The goal of this study was twofold. First, we aimed to identify patterns of miRNA deregulation in breast cancer. Therefore, we compare miRNA-expression patterns between breast tumor samples classified according to the molecular subtypes and between breast tumor samples and normal breast samples. We hypothesize that such profiles can be informative for breast cancer detection and prognosis and might assist in defining specific targets for future therapy.
Second, we investigated whether the expression levels of miRNAs are measurable in blood samples from patients with breast cancer and healthy volunteers and if such expression profiles are potentially useful for the detection and staging of breast cancer.
Materials and methods
Patients and samples collection
Tumor and blood samples were obtained from patients with breast adenocarcinoma treated in the Breast Clinic of the General Hospital Sint-Augustinus (Antwerp, Belgium). Tissue and serum samples were derived from two entirely independent populations. Each patient gave written informed consent. This study was approved by the Institutional Review Board. Clinicopathologic data are stored in a database in accordance with hospital privacy rules and are summarized in Table
1. All tissue samples were stored in liquid nitrogen within 15 minutes after excision (median delay of 9 minutes). Healthy control tissue was obtained from breast-reductive surgery. None of the control samples showed pathologic changes. In total, 84 tumor samples and eight healthy control samples were included.
Table 1
Clinicopathologic data
T status | 1 | 27 (32%) | 31 (41%) |
| 2 | 27 (32%) | 22 (29%) |
| 3 | 6 (7%) | 4 (5%) |
| 4 | 24 (29%) | 18 (25%) |
N status | 0 | 35 (42%) | 28 (37%) |
| 1 | 21 (25%) | 15 (20%) |
| 2 | 14 (17%) | 10 (13%) |
| 3 | 13 (15%) | 9 (12%) |
| 4 | 1 (1%) | 13 (18%) |
M status | 0 | 70 (83%) | 4 (5%) |
| 1 | 14 (17%) | 71 (95%) |
ER status | Negative | 25 (30%) | 23 (31%) |
| Positive | 59 (70%) | 52 (69%) |
PR status | Negative | 45 (54%) | 36 (48%) |
| Positive | 39 (46%) | 39 (52%) |
ErbB2 status | Negative | 62 (74%) | 49 (65%) |
| Positive | 22 (26%) | 26 (35%) |
Grade | 1 | 9 (11%) | 14 (19%) |
| 2 | 32 (38%) | 29 (39%) |
| 3 | 43 (51%) | 32 (42%) |
Stage | I | 24 (29%) | 19 (25%) |
| II | 20 (24%) | 14 (19%) |
| III | 27 (32%) | 22 (29%) |
| IV | 13 (15%) | 20 (27%) |
Disease status | Progressive | 29 (35%) | 47 (63%) |
| Nonprogressive | 55 (65%) | 28 (37%) |
The collection of serum samples was described previously [
31]. In brief, samples were prospectively obtained from 75 patients with breast cancer and 20 healthy volunteers. Patients were divided into three groups: four patients with localized breast cancer (group A), 55 patients with metastatic breast cancer receiving treatment (group B), and 16 patients with untreated metastatic breast cancer (group C). The blood samples of patients with metastatic disease were taken during the course of treatment. For all these samples, circulating tumor cells (CTCs) were enumerated by using the CellSearch system (Veridex, Warren, NJ, USA), CK19, and mammaglobin mRNA expression was recorded, the ADNAgen test for detection of CTCs was performed, and levels of total plasma DNA and serum methylated DNA for
ESR1,
RASSF1A, or
APC1 were measured in earlier studies [
31,
32]. Disease status was assessed by using the RECIST (Response Evaluation Criteria in Solid Tumors) criteria without knowledge of the patient's CTC or circulating DNA results [
33]. Stable disease was measured up to 8 weeks after the initiation of therapy. In addition, we collected blood samples from an additional series of 18 unselected patients to evaluate which blood medium (that is, serum, plasma, platelet-rich plasma, whole blood, or peripheral blood mononuclear cells (PBMCs)) was best suited for extraction of small RNAs (sRNAs).
After tissue disruption, total RNA was extracted by using the mirVana miRNA Isolation Kit (Ambion, Austin, TX, USA) according to the manufacturer's instructions for total RNA isolation. In brief, the sample was homogenized in a denaturing lysis solution, followed by an acid-phenol:chloroform extraction. Thereafter, the sample was purified on a glass-fiber filter and quantified by using the Nanodrop ND1000 (NanoDrop Technologies, Waltham, MA, USA). Total RNA (100 ng) was converted to cDNA by priming with two pools of stem-looped RT primers (Megaplex RT Primers, Human Pool A & B; Applied Biosystems, Foster City, CA, USA) in combination with the TaqMan MicroRNA Reverse Transcription Kit (Applied Biosystems), allowing the simultaneous transcription of 377 unique miRNAs and six endogenous controls per primer pool. In brief, 3 μl of total RNA was supplemented with RT primer mix (×10), dNTPs with dTTP (100 mM), Multiscribe Reverse Transcriptase (50 U/μl), RT buffer (×10), MgCl2 (25 mM), and RNase inhibitor (20U/μl) in a total reaction volume of 7.5 μl.
Thermal-cycling conditions were as follows: 40 cycles at 16°C for 2 minutes, 42°C for 1 minute, and 50°C for 1 second, followed by reverse transcriptase inactivation at 85°C for 5 minutes. The Megaplex RT product (2.5 μl) was preamplified by using the TaqMan PreAmp Master Mix (Applied Biosystems) and preamplification primers in a 25-μl PCR reaction. For each pool of stem-looped RT primers in the cDNA reaction, a different pool of PreAmp Primers (Human Pool A & B; Applied Biosystems) was used. Thermal-cycling conditions were as follows: 95°C for 10 minutes, 55°C for 2 minutes, and 75°C for 2 minutes, followed by 12 cycles of 95°C for 15 seconds and 60°C for 4 minutes. MiRNA quantification was performed with the TaqMan Human MicroRNA Array sets A & B (Applied Biosystems), each containing 384 TaqMan miRNA assays. The PreAmp product was diluted fourfold. Each of the eight wells was loaded with 100 μl of PCR reaction mix, containing 50 μl of TaqMan Universal PCR Master Mix, no AmpErase uracil N-glycosylase (UNG) (Applied Biosystems), 1 μl of diluted PreAmp product, and 49 μl of nuclease-free water. Thermal-cycling conditions were as follows: 94.5°C for 10 minutes, followed by 40 cycles at 97°C for 30 seconds and 59.7°C for 1 minute. All PCR reactions were performed on a 7900HT Fast Real-Time PCR System (Applied Biosystems).
To test the efficiency of the miRNA assays, we compared the Ct-values of an undiluted sample with those of a 10-fold diluted sample. To evaluate the linearity of the preamplification, we compared the Ct-values of all miRNAs on both array cards for one sample before and after preamplification. The reproducibility of the arrays was tested by analyzing four samples in duplicate. The robustness of the TaqMan RT-PCR method was investigated by comparing the qRT-PCR miRNA expression profile of 12 samples with their miRNA expression profile obtained by using the nCounter Analysis System (Nanostring Technologies, Seattle, WA, USA). This system is a medium-high throughput gene-expression quantification system with PCR sensitivity that uses a novel digital technology based on direct multiplexed measurement of miRNA expression. Besides a direct quantification, the workflow incorporates only one enzymatic step (ligase step to enable tagging of the miRNAs) instead of three enzymatic steps in the qRT-PCR workflow, thereby substantially reducing the possibility for technical bias. The nCounter experiment was performed in collaboration with the VIB MicroArray Facility (O&N, UZ Gasthuisberg, Leuven, Belgium).
First, we evaluated which blood medium was best suited for the extraction of sRNA molecules. Therefore, plasma, platelet-rich plasma, serum, whole blood, and PBMCs were obtained from 18 patients with breast cancer. Peripheral blood was collected in a 9-ml EDTA tube, from which 3 ml of whole blood was transferred into a cryovial while the remaining blood was centrifuged slowly (150 g, 20 minutes) at 4°C to generate platelet-rich plasma. Plasma and PBMCs were obtained in an 8-ml CPT tube, which was centrifuged (1,650 g, 20 minutes) at room temperature. Plasma and PBMC aliquots were transferred into separate cryovials. Finally, 8 ml blood was collected in serum separator tubes (Becton Dickinson, Franklin Lakes, NJ, USA) and centrifuged (2,000 g, 10 minutes) at room temperature. All samples were stored at -80°C until use.
sRNA was isolated from 200 μl of each medium by using the microRNA Isolation Kit (BioChain Institute Inc, Hayward, CA, USA) according to the manufacturer's instruction for sRNA purification. In brief, after adding lysis buffer to the sample for homogenization, 20 μl of Proteinase K solution (Qiagen, Valencia, CA, USA) was added and incubated for 10 minutes at 75°C to digest the excess of proteins released after addition of the lysis buffer. This was followed by an acid-phenol:chloroform extraction. Small and large RNAs were separated by using a centrifugation step, after which the large RNAs were retained on a glass-fiber filter. The sRNA molecules were recovered from the flow-through by purifying them on a second glass-fiber filter, and their concentration and purity (A260/A280 and A260/A230) was recorded by using the NanoDrop ND1000 (NanoDrop Technologies, Waltham, MA, USA). The concentrations were compared by using a Kruskal-Wallis test with Tukey HSD post hoc testing.
To evaluate circulating miRNA expression in blood samples from 20 healthy volunteers and 75 patients with breast cancer, we isolated total RNA, as described before. Isolated total RNA was reverse transcribed to produce cDNA by using the TaqMan MicroRNA Reverse Transcription Kit (Applied Biosystems) by priming with TaqMan MicroRNA Assays (Applied Biosystems) directed at 4 miRNAs identified by comparing tumor tissue with normal breast tissue (vide supra). In addition, miR-16 expression was determined as a normalization factor. In brief, each 15-μl reaction contained 0.15 μl 100 mM dNTPs with dTTP, 1.0 μl Multiscribe Reverse Transcriptase (50 U/μl), 1.50 μl RT Buffer (×10), 0.19 μl RNase Inhibitor (20 U/μl), 4.16 μl nuclease-free water, 5.0 μl total RNA, and 3.0 μl RT primer. Thermal-cycling conditions were 30 minutes at 16°C, 30 minutes at 42°C, and 5 minutes at 85°C. Each 20-μl reaction for the real-time quantitative PCR contained 1.0 μl real-time primer, 1.33 μl product from RT reaction, 10.0 μl TaqMan Universal PCR Master Mix, no AmpErase UNG (Applied Biosystems), and 7.67 μl nuclease-free water. The reactions were performed in duplicate on a 7900HT Fast Real-Time PCR System in the 9600 emulation mode, with conditions of 10 minutes at 95°C, followed by 40 cycles of 15 seconds at 95°C and 1 minute at 60°C. The mirVana miRNA Reference Panel (Ambion, Austin, TX, USA) was included in each PCR plate in a 2,000-fold dilution to correct for between-plate differences.
All subsequent analyses were performed by using BioConductor in R. To reduce technical variation, the miRNA assays with a PCR efficiency outside the range of 2log(10) or 3.32 ± 25% [
34] and those with Ct values above 35 in at least 25% of the cases were filtered out. By using efficient and informative miRNA assays only, we calculated the mean difference between the Ct values of one sample before and after preamplification. To avoid technical bias, we excluded miRNA assays with a difference in Ct values before and after preamplification outside the range of the mean value ± 25%. For the final set of miRNAs, we calculated the mean expression level per sample and used this value as a normalization factor to account for differences in input material [
35]. Relative miRNA expression levels were calculated by using the ΔCt-method [
36] and log2-transformed to obtain a normal distribution. To investigate assay reproducibility, we correlated the expression profiles of the duplicate samples by using the Spearman correlation coefficient.
An additional technical validation was done by performing a pairwise correlation analysis between the miRNA profiles obtained by qRT-PCR and the nCounter Analysis System for the 12 samples analyzed on both platforms. Both correlation analyses were done by using the normalized expression profiles of the 327 common miRNAs only.
Unsupervised hierarchical cluster analysis (UHCA), with the Manhattan distance as similarity metric and Ward clustering as the dendrogram drawing method, was performed to visualize global themes in the expression data. We classified samples according to the miRNA-centroids for molecular subtypes published by Blenkiron
et al. [
14]. Therefore, we correlated the molecular subtype-specific miRNA-expression profiles of each sample with each of the five miRNA-based expression centroids by using the Spearman correlation coefficient.
The resulting classification was compared with the UHCA result. For 66 samples with available Affymetrix-profiles, we compared the correlation coefficients between the samples grouped according to the SSP (single-sample predictor)-defined molecular subtype classification [
37] obtained through mRNA-expression profiling reported in earlier studies [
38,
39]. Significance was assessed by using the Mann-Whitney
U tests.
Next, we aimed to identify molecular subtype-specific miRNAs. Therefore, we performed a pairwise comparison of the different molecular subtypes, defined through mRNA-expression profiling, by using regression analysis with the limma-package. False Discovery Rate (FDR) correction was performed by using the Benjamini and Hochberg step-up procedure. For each subtype, we crossed the lists of differentially expressed miRNAs resulting from the pairwise comparisons involving the desired subtype in search for common miRNAs.
By using regression analysis, we identified differentially expressed miRNAs between normal and tumor samples. Resulting P values were corrected for false discovery, as described earlier. To investigate global over- or underexpression in normal samples, we calculated the median expression level of the differentially expressed miRNAs per sample. These median expression values were compared by using Mann-Whitney U testing. The top four differentially expressed miRNAs by fold-change were selected for further analysis. For these miRNAs, we identified target mRNAs in at least two of three public databases (PicTar, TargetScan, and Miranda) by using the RmiR-package. These target-gene lists were subjected to Ingenuity Pathway Analysis (IPA) to study the implications of the identified miRNAs in cancer biology.
Expression levels of circulating miRNAs were calculated with miR-16 as normalization factor. Raw Ct values measured in the miRNA Reference Panel were subtracted from the Ct values measured in the samples, yielding a between-plate corrected expression value for each miRNA per 200 μl of serum. The miR-16 normalized expression value was calculated by subtracting the between-plate corrected expression value for miR-16 from the between-plate corrected expression values for the remaining miRNAs (ΔΔCt-method). Relative expression values were calculated by using the 2
-ΔΔCt method [
36]. To compare the expression data with categoric variables, the Mann-Whitney
U test was performed. To compare expression data with continuous variables, Spearman correlation coefficients were calculated.
Discussion
We attempted to identify a panel of deregulated miRNAs in breast cancer and investigated their potential as biomarkers for the detection and staging of breast cancer by using blood-based testing. Before analyzing the miRNA-expression data, we first evaluated the performance of the PCR technology used throughout this study. To reduce the technical variation in our data set, we included only informative miRNAs assays (Ct values smaller than 35 in at least 25% of the cases) with similar PCR efficiencies and similar differences in Ct values before and after preamplification. The boundaries for PCR efficiency were defined as described in earlier studies [
34], and the boundaries for preamplification efficiency were set alike. The expression data recorded by the final set of 373 selected miRNAs proved to be reproducible, and no between-array card difference was observed. Moreover, we noticed an above-moderate agreement between the qRT-PCR-based miRNA profiles of 12 samples with the miRNA profiles measured by using the nCounter Analysis System.
This is important for two reasons. First, the nCounter Analysis System incorporates only one enzymatic step (that is, a ligase treatment for attachment of the reporter tags) in its workflow and is therefore less prone to technical bias than is the PCR-based protocol that incorporates three enzymatic steps.
A second reason for comparing the miRNA expression profiles by using alternative profiling techniques is related to the fact that good normalization procedures for miRNA expression data are currently still lacking. The qRT-PCR-based miRNA-expression data in this study were normalized relative to the mean expression value of all miRNAs per sample, as proposed by Mestdagh
et al. [
35]. However, we think that this normalization procedure might have a major drawback because of the role of DICER1, a miRNA-preprocessing enzyme, in breast cancer. Recent reports have shown that the expression of DICER1 is different across the different molecular subtypes [
14,
40‐
42]. As DICER1 is involved in cleaving the precursor miRNAs into mature miRNAs, variation in DICER1 expression might result in altered turnover rates of the precursor miRNAs and, hence, higher concentrations of mature miRNAs in those tumor samples with higher DICER1 expression. Therefore, we reason that the mean miRNA-expression levels can vary depending on DICER1 expression and that normalization relative to the mean miRNA-expression level might obscure between-sample differences, particularly in breast cancer. The alternative approach would be to use the reference miRNAs provided on the array cards or miR-16, which is often suggested as reference miRNA. However, the CVs for these reference miRNAs were about threefold higher than the CV of the mean miRNA expression level per sample. In addition, about 20% of the miRNA assays on both array cards yielded more-robust expression data (data not shown).
Therefore, in spite of our previously raised concerns, we decided to normalize our expression data relative to the mean Ct value per sample and compare the results with the data obtained by using nCounter Analysis System, which uses a panel of five mRNA assays (RPLPO, RPL19, ACTB, B2M, and GAPDH) for data normalization. The above-moderate agreement between the miRNA-expression data obtained by using both profiling techniques lends credit to the biologic validity of our qRT-PCR-based miRNA-expression profiles.
Further evidence that the applied normalization procedure did not obscure molecular subtype-specific differences is derived from the UHCA, which showed that the molecular subtypes govern global themes in our miRNA expression data set. Also, the miRNA-based molecular subtype classification is in agreement with the classification resulting from the application of a more-validated algorithm on mRNA data (SSP) [
37]. For example, the comparison of the miRNA-based expression profile of SSP-defined Basal-like breast tumors with the miRNA-based expression centroid for Basal-like breast cancer results in more-elevated Spearman correlation coefficients than when compared with the results obtained for non-Basal-like breast tumor samples. Although the classification error rate was substantial, we must keep in mind that the miRNA-based expression centroids reported by Blenkiron and colleagues [
14] are based on a limited series of samples. Therefore, it is arguable that the expression centroids are not very stable, which affects the classification accuracy. When performing a supervised analysis, we were able to identify sets of specific miRNAs for each molecular subtype, except for the ErbB2+ breast tumor samples. Overall, our results are in line with previously reported data [
14,
43‐
45], except for the results with respect to the ErbB2+ subtype, for which an miRNA signature has been defined in the past [
45]. Of note is the concordant overexpression of miRNAs belonging to the polycistronic miR-17-92 cluster (miR-17, miR-18a, miR-19a, miR-20a, and miR-92) and its paralogs (miR-18b and miR-106a) in Basal-like breast tumors. The miR-17-92 cluster is known to downregulate ERα in a MYC-dependent manner and inhibits the protein translation of AIB1, an ERα transcriptional coactivator [
46]. Also, the miR-17-92 cluster is known to regulate cell migration, invasion, and metastasis in breast cancer by regulating ROCK [
47] and the HBP1/β-catenin pathway [
48].
Although the sample-clustering pattern based on the expression of 373 miRNAs demonstrated that the global themes in our expression data set are related to the presence of the classic molecular subtypes in breast cancer, we did identify one sample cluster without any connection to the classic molecular subtypes. This sample cluster originated early in the dendrogram, indicative of a specific miRNA-expression profile. Indeed, the heatmap did reveal an miRNA cluster, including members of the miR-200 family, members of the let-7 family, and NFκB-regulating miRNAs [
49], that is overexpressed in this group of tumor samples, at a level exceeding the expression level observed in the Luminal-like sample cluster. The latter observation is at least remarkable, as all these miRNA families are known to inhibit stem cell-specific pathways, epithelial-to-mesenchymal transition, cell proliferation, and other global oncogenic processes [
16,
50‐
53]. Hence, their overexpression would induce a more-differentiated, less-proliferative, less-mesenchymal, and less-migratory/invasive cell phenotype. The presence of this tumor sample cluster with its particular molecular characteristics warrants further investigation. When focusing on the Normal-like samples, a clear and distinct miRNA profile was observed. In addition, the true normal breast samples constituted a coherent group inside the cluster of the Normal-like samples, suggesting vast differences in miRNA expression between tumor samples and normal breast samples. Indeed, supervised analysis revealed high numbers of differentially expressed miRNAs with nominal (uncorrected)
P values less than 0.05.
The huge difference in miRNA expression between normal and tumor samples underlines the important role of miRNA deregulation in the development of breast cancer. After correction for false discovery, we observed that the majority of the differentially expressed miRNAs have attenuated expression levels in the tumor samples. The global repression of miRNAs in cancerous tissue relative to normal tissue has been reported previously and suggests that most miRNAs have a tumor-suppressive function [
13]. This view is corroborated by reports on the cellular functions of the top four (miR-215, miR-299-5p, miR-411, and miR-452) differentially expressed miRNAs by fold-change. Song
et al. [
54] demonstrated that miR-215 overexpression in a colon cancer cell line reduced the proliferation rate and led to improved cell-cycle control, probably due to an increased expression of the cell-cycle control genes
p53 and
p21. Duan
et al. [
55] showed that miR-299-5p expression increased downstream of the tumor suppressor PRDM5 in HEK293 cells. In contrast, Fang
et al. [
56] showed that
SOX2, a gene with tumor-promoting activity involved in cell proliferation and colony formation of LN229 glioblastoma multiforme cells, represses miR-452. The evidence for a role of miR-411 as tumor suppressor is less clear, but this miRNA is located at the 14q32.31 locus, which is known to harbor many tumor-suppressive miRNAs [
57]. The biologic processes regulated by miR-215, miR-299-5p, miR-411, and miR-452, identified through the analysis of their respective target-gene lists, are in line with their role in maintaining cellular homeostasis.
Because of the marked overexpression of miR-215, miR-299-5p, miR-411, and miR-452 in normal breast samples, in addition to the fact that miRNAs have a proven stability in blood samples [
29], we hypothesized that this panel of miRNAs might be suitable for the detection of breast cancer by using blood-borne testing. The reason for using serum samples for this purpose is twofold. First, we argued that miRNA-expression profiles in whole blood, platelet-rich plasma, and PBMCs would be dominated by host miRNA expression, and therefore would be less suitable for the detection of tumor-specific miRNA expression. Conversely, reports have shown that miRNA expression is also detectable in serum and plasma samples from healthy donors [
58]. Second, we noticed a slightly higher and more consistent sRNA yield in serum as compared with plasma. When evaluating the relative expression profiles of miR-215, miR-299-5p, miR-411, and miR-452 in serum samples from patients with breast cancer and healthy volunteers, we recorded comparable expression profiles in tissue and blood samples, except for miR-452. Of note, when comparing absolute CT values, the expression differences between samples from patients with breast cancer and healthy volunteers were maintained, however, at a higher fold-change level. In addition, we observed that the reduction of miRNA expression was particularly obvious in serum samples from patients with untreated metastatic breast cancer, whereas the expression profiles "normalized" with treatment.
No associations between blood-borne miRNA expression in serum samples from patients with breast cancer and the classical clinicopathologic variables were observed, except for the patient's age at diagnosis. However, this should not be surprising, as our miRNA panel was not selected to make this distinction. Of note is the lack of associations between circulating miRNA expression and the presence of CTCs, measured by three alternative techniques. This observation suggests that recorded serum miRNA profiles are not CTC derived and that the mechanisms responsible for the release of miRNAs in the circulation are unrelated to the extravasation of tumor cells. Indeed, several reports have suggested that miRNAs are selectively released in the bloodstream by tumor cells either via exosomes or attached to lipoprotein complexes or within a functional RISC complex [
59‐
63]. These mechanisms of secretion offer an explanation for the marked stability of miRNAs in the blood stream, due to shielding of the associated miRNAs from RNAse-activity. In addition, miRNAs secreted as such are functionally active and have been found to regulate gene expression in target cells, thereby providing alternative ways of cell-cell communication. This opens the possibility that miRNAs, secreted by tumor cells, evoke a response in host cells altering their expression profile, which explains how subtle differences in tumor-specific expression are measurable in a background of nontumorigenic expression. More specific in the context of our results, one could envision that tumor-driving miRNAs, secreted by tumor cells, affect the expression profile of host cells, which is reflected in the serum profile of breast cancer patients and explains the observed decrease in miRNA expression.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
EVS, MCAW, and IVDA acquired all data. IVDA and SJVL performed data analysis. EVS, MCAW, IVDA, and SVL assisted in data interpretation. IVDA, SJVL, PBV, and LD contributed to the conception and design of the study. EVS, MCAW, IVDA, and SVL were involved in drafting the manuscript. All authors were involved in revising the manuscript critically for important intellectual content. All authors gave final approval of the version to be published.