Robust expression of tumor suppressor miRNA’s let-7 and miR-195 detected in plasma of Saudi female breast cancer patients

Approval and written consent was obtained from all study participants for the use of their blood samples for research purposes. The study was approved by the Ethical Research Committee and Basic Research Committee on Clinical Research at KFSHRC, Riyadh, Saudi Arabia and was carried out under the terms of the Helsinki Declaration.

A total of 127 females, disease-free individuals (n = 34), triple-negative breast tumors (TNBC) (n = 36) and patients with luminal tumors (n = 57) were recruited for the study. All women were of Saudi background and recruited at the King Faisal Specialist Hospital (KFSHRC). Details of the study subjects with respect to the age of diagnosis, grade and lymph node status are reported in (Additional file 1: Table S1). All blood samples were obtained from patients before any cancer therapy was administered. Blood samples were collected by vene-section in EDTA blood collection tubes (BD Vactainer, Plymouth, UK) and kept at 4 °C. The blood was then centrifuged within 2 h at 1500 g at 4 °C for 15 min to isolate the plasma. The plasma was collected and centrifuged again at 2500 g at 4 °C for 15 min (Heraeus multifug 3S-R-UK) to eliminate the debris. All samples were stored at -80 °C.

RNA was isolated from the plasma using the miRNeasy Serum/Plasma Kit (Qiagen, Hilden, Germany), following the manufacturer’s protocol with some modifications. RNA isolation was performed in duplicate. For the Qiagen kit assay, 1 ml of QIAzol lysis reagent (Qiagen) was added to 200 μl of plasma together with 1 μg of carrier MS2 RNA (Roche). The samples were mixed and incubated for 5 min at room temperature. To monitor the RNA isolation before purification 3.5 μl (1.6 × 10⁸ copies/μl) of C.elegans-miR-39 miRNA mimic spike-in control was added and 12 μg of pure glycogen was added as a carrier/co-precipitant. Next, 200 μl of chloroform was added to the starting sample and mixed for 15 s. Next, samples were incubated for 3 min at room temperature, and centrifuged for 15 min at 12,000 g at 4 °C. 600 μl of the aqueous phase was transferred to a new tube and 900 μl of 100% ethanol was added to the spin column to allow all the RNA molecules to reach the binding condition. After mixing, the samples were transferred to RNeasy MinElute spin columns in a 2 ml collection tube and spun for 15 s at ≥8000 g. The columns were washed with two buffers from 700 μl RWT buffer and then with 500 μl RPE buffer (Qiagen, Hilden, Germany), with a short spin of 15 s at ≥8000 g each time. Next, each column was washed with 500 μl of 80% ethanol and centrifuged for 2 min at ≥8000 g. Following 5 min high speed centrifugation, 14 μl of RNase-free water could be used for elution. All RNA samples were frozen at -80 °C until further analysis.

Quality of RNA was assessed by Nanodrop ND-2000 (Wilmington, DE, USA). Chromatographic characteristics and integrity of all RNA samples were determined, included interpretation of the peak detection of different profiles, by means of RNA 6000 Nano LabChip, (Agilent Technologies, Waldbronn, Germany), Agilent 2100 Bioanalyzer system (Agilent Technologies, Santa Clara, USA) and the 2100 expert software tool (Agilent Technologies, Santa Clara, USA).

250 ng of the eluted RNA sample was used to make cDNA using a miScript RT II kit with miScript HiSpec buffer from Qiagen (Qiagen, Hilden, Germany). Briefly, the reaction was set up in the thermal cycler for 60 min at 37 °C followed by 5 min at 95 °C. The cDNA was then diluted with 200 μl of RNase-free water. For RT-qPCR a total of 2750 μl was prepared, made up of (2 x QuantiTect SYBR green PCR master mix, 10 x miScript universal primers with a cDNA template and RNase free water). MicroRNAs screening was performed using miScript miRNA PCR Array Human Breast Cancer-MIHS-109Z (Qiagen, Hilden, Germany). The miRNA PCR Array Panel contains 84 mature miRNAs most relevant to breast cancer tumorigenesis. The final reaction volume was 25 μl per well, enough to provide 1 ng cDNA per well. The plates were run following a thermal cycling protocol: 95 °C for 15 min to activate the HotStar Taq DNA polymerase, 40 amplification cycles of 15 s at 94 °C, 30 s at 55 °C, 30 s at 70 °C and at the end a melting curve program. All qPCR reactions were run in duplicate.

All samples passed ‘Positive PCR Controls’ (PPC) in which the acceptable range of Ct values was set to 19 ± 2. Both Reverse Transcription Controls (RTC) and (PPCs) were used to assess whether there had been any inhibition during the reverse transcription reaction. Avg Ct^miRTC – Avg Ct^PPC was calculated for each sample (Avg “average”). A difference of greater than seven indicated impurities and reaction inhibition, cellular contamination was assessed using the mean Ct of the SNORNA (SNORD) controls. Only the non-zero values were considered. SNORD72 was excluded from computations as it performed poorly across samples. A sample with a mean Ct < 32 was taken to indicate cellular contamination. All samples used in the qRT-PCR analysis were tested for quality and neither indication of cellular contamination nor reaction inhibition was detected. All samples collected were retained and none discarded. For miRNA to be within the detection limit, the Ct values were recommended to be between zero and 35. Cel-correction to correct for technical variations that arise during extraction procedure, exogenous spike-in controls from C.elegans was used. The Qiagen miScript miRNA PCR Array Human Breast Cancer array panel contains two Cel-miR-39 spiked-in controls. The average Ct value recorded in each sample for the spiked-in controls from C.elegans, cel-39 was recorded and the median of average Ct values (of all samples) was found. The normalizing factor for each sample was calculated by subtracting the median of all average Ct^cel values from the average Ct^cel value for the sample. The ddCt computation (ΔΔCt) for each of the three groups, the triple negative (TNBC) and the luminal tumors, average dCt (ΔCt) was calculated for each miRNA. The miRNAs with average dCt values (ΔCt) below 15 and/or above 35 were excluded. ddCt (ΔΔCt) of a given miRNA for a given pair of groups was computed by finding the difference between the average dCt (ΔCt) of the respective groups; for example, ΔΔCt(ddCt _{(Triple Negative vs. Controls))} = [Average dCt (TNBC)] – [Average dCt (Controls)]. The relative expression of a given miRNA between any two groups was assessed by computing 2^(−ddCt) (2^(−ΔΔCt)). A differential set was identified using the ddCt (ΔΔCt) method proposed by Livak et al. [30]. The Mann-Whitney unpaired test and Benjamin-Hochberg multiple testing corrections were used to determine significant differences in miRNA expression levels between groups [31]. All qPCR reactions were run in duplicate.

The miRNA targets and the biological pathways they were involved in were predicted using the microT-CDS algorithm and mirPath v.2.0 available on the web-based server DIANA. The micro-T threshold for target prediction was set at 0.8 and targeted pathways were considered significant at a p-value < 0.05 [32, 33]. Hierarchical clustering was performed using GeneSpring GX 14.5.

Receiver Operating Characteristic (ROC) curves were generated using the web-based tool ROCCET [34] for finding two sets of miRNAs that could best differentiate (i) triple-negative tumour samples from normal (control) samples and (ii) luminal tumour samples from normal (control) samples. ROC (Receiver Operating Characteristic) curves were then generated by Monte-Carlo Cross Validation (MCCV). The procedure was performed repeatedly to calculate the performance and confidence interval of each model.

Since plasma levels of miRNA are not necessarily a reflection of tissue levels [35‐38] and tumor tissue was not available from patients recruited for this study as plasma was collected pre-cancer therapy, publicly available data from the TCGA (The Cancer Genome Atlas) [39] was used to determine whether circulating plasma levels were distinct from breast cancer tissue and surrounding non-cancerous tissue levels. We compared the observed miRNA plasma levels with the tissue levels of corresponding miRNA precursors from the TCGA study. The tissue level expression of miRNA precursors, available as RPKM (reads per kilobase of transcript per million) values was obtained for control samples (n = 87), luminal samples (n = 120) and Triple Negative samples (n = 38) from TCGA. The RPKM values of the precursors in tissue and the expression value for miRNAs in the current study were log₂ transformed and auto-scaled, to ensure the datasets are comparable.

The tissue RPKM and the plasma cel-corrected C_t values are normalized to the mean (μ) and standard deviation (δ) for each of the data. Since a miRNA precursor can give rise to an active form from each arm, we compared both the 3′ and 5′ active forms were matched to the same precursor. The 18 differentially expressed active miRNA forms mapped to 17 precursor miRNAs from TCGA. Then we analyzed the expression trends of a given miRNA across tissue and plasma samples.

Comparative analysis identified an initial set of 18 circulating miRNAs (Table 1), which because of their differential presence between the patient groups and healthy controls, were further examined by cancer type. Figure 1 illustrates relative expression of these 18 miRNAs subdivided into three groups: TNBC plasma vs. disease free plasma (Group A; n = 8 miRNAs: hsa-miR-29c-3p, hsa-miR-195-5p, hsa-miR-210-3p, hsa-miR-19b-3p, hsa-miR-19a-3p, hsa-miR-22-3p, hsa-miR-7-5p, hsa-miR-15a-5p); luminal patient plasma vs. disease free plasma (Group B; n = 5 miRNAs: hsa-let-7c-5p, hsa-miR-489-3p, hsa-miR-340-5p, hsa-miR-199a-5p, hsa-miR-328-3p); lastly, all breast cancer patients (irrespective of subtype) vs. disease free plasma (Group C; n = 5 miRNAs: hsa-let-7i-5p, hsa-miR-25-3p, hsa-miR-16-5p, hsa-let-7b-5p, hsa-miR-199a-3p).

Next, the tissue expression trends of the corresponding 17 miRNA precursors in relation to their active forms in plasma were measured (Fig. 2). Since plasma collected from all patients enrolled in this study occurred prior to chemotherapy administration and/or tumor biopsy, miRNA breast cancer tissue and non-cancerous tissue expression values were obtained from publically available TCGA RNASeq data; non-cancerous tissue samples (n = 87), luminal samples (n = 120) and Triple Negative samples (n = 38). Expression trends for some miRNAs (hsa-miR-19a, hsa-miR-19b, hsa-miR-210, hsa-miR-16, hsa-miR-7 and to a certain extent hsa-miR-15a) were similar in both tissue and plasma. Tissue and plasma levels of non-diseased controls compared against both cancer patient groups showed a broad reversal of the trend. For example, tumor repressor miRs hsa-let-7c and hsa-miR-195 were significantly decreased in both luminal and TNBC breast cancer tissue levels (TCGA) and increased in non-tumor tissue samples. Slight variations of this pattern were observed for hsa-miR-489, hsa-miR-328, hsa-miR-25, hsa-let-7i, hsa-let-7b, hsa-miR-29c, hsa-miR-199a, hsa-miR-340 and hsa-miR-22.

Lastly, a pathway analyses was performed on the target genes of the 18 differential miRNAs identified in this study (Additional file 1: Figure S1). Not surprisingly, many of them were involved in signaling functions, namely the PI3K-Akt, mTOR, p53, TGF-beta, Wnt, FoxO, estrogen signaling and Hippo signaling pathways. However, the most significantly enriched pathways were the ECM-receptor interaction (Extracellular Matrix) and the fatty acid biosynthesis (FASN) pathways. Gene targets of the let-7 family were found to be involved in ECM receptor interaction while the fatty acid biosynthesis pathway (FASN) was shown to be enriched mainly by hsa-miR-16-5p, hsa-miR-15a-5p and hsa-miR-195-5p. Both hsa-miR-15a-5p and hsa-miR-195-5p were enriched exclusively in the plasma of TNBC patients (see Fig. 1).

In order to assess the potential of each of these miRNAs as cancer biomarkers, we generated ROC curves. Since Univariate AUC ROC curves looked promising, whether a more robust prediction could be made using multiple miRNAs was explored. A panel of seven miRNAs consisting of hsa-miR-199a-3p, hsa-miR-15a-5p, hsa-let-7c-5p, hsa-miR-7-5p, hsa-miR-195-5p, hsa-miR-489-3p and hsa-let-7i-5p showed the maximum discriminatory potential between TNBC patient plasma and disease-free plasma (Fig. 3a). Similarly, a panel of five miRNAs consisting of hsa-miR-328-3p, hsa-miR-199a-3p, hsa-let-7i-5p, hsa-miR-195-5p and hsa-miR-25-3p best predicted luminal tumor patients from disease-free individuals (Fig. 3b). The Univariate AUC statistic for each miRNA is provided (Additional file 1: Table S2).