This study was reviewed and approved by the Valley Hospital Institutional Review Board. Written informed consent was obtained from all study participants. The pancreatic cancer patient group included stage IIA/IIB patients whose stage was confirmed post-operatively by pathologists from the Valley Hospital. Eight patients diagnosed with ductal adenocarcinoma were used in the microarray screening and real-time quantitative reverse transcription polymerase chain reaction (RT-qPCR) confirmation study. A second group of eleven ductal adenocarcinoma patients was used in the validation study. The control group for all studies was comprised of eleven healthy participants with no family history of pancreatic cancer. The high risk group included eleven healthy participants who had a strong family history of pancreatic cancer, including ten participants with at least two “first degree relatives” and one participant with two “second degree relatives” having a pancreatic cancer diagnosis. The participant demographics are shown in Table 1.

Patients were excluded from the study if they had prior pancreatic cancer surgery, had other concomitant cancers other than non-melanoma basal cell skin cancer, or had a history of HIV infection.

Patient blood was drawn by peripheral venipuncture into BD Vacutainer^® CPT™ (Cell Preparation Tubes) with Sodium Citrate (Becton Dickinson, Franklin Lakes, NJ). Blood processing was typically done within two hours of collection and performed according to the manufacturer’s protocol. Harvested plasma was stored at -80 °C.

Plasma samples were processed at Ocean Ridge Biosciences (ORB, Palm Beach Gardens, FL) for analysis using custom multi-species microarrays containing 1209 probes covering 1220 human mature miRNAs present in the miRBase version 16.0 database released in 2010. The sensitivity of the microarray is such that it could detect as low as 20 amoles of synthetic miRNA being hybridized along with each sample. The microarrays were produced by Microarrays Inc. (Huntsville, Alabama), and consisted of epoxide glass substrates that have been spotted in triplicate with each probe.

Sample processing: Samples were isolated from 0.7 to 1.0 mL of plasma using TRI Reagent^® BD (Molecular Research Center, Cincinnati, OH) as per manufacturer instructions. For quality control, a mixture of 10 synthetic miRNAs were added (spike-in) at a mass of 12.5 fmoles/mL of plasma to each plasma sample during isolation and one miRNA was added at 200 amoles per sample prior to labeling and hybridization. Total RNA was 3’-end labeled with Oyster-550 fluorescent dye using the Flash Tag RNA labeling Kit (Genisphere, Hatfield, PA). Labeled RNA samples were hybridized to the miRNA microarrays according to conditions recommended in the Flash Tag RNA labeling Kit manual. The microarrays were scanned on an Axon Genepix 4000B scanner, and data was extracted from images using GenePix v4.1 software.

Data pre-processing: Spot intensities were obtained for the 8816 features on each microarray by subtracting the median local background from the median local foreground for each spot. The 95^th percentile of the negative control spots was also calculated for each array. The spot intensities and 95^th percentile of negative controls (TPT95) were transformed by taking the Logarithm base 2 (indicated as log₂) of each value. The normalization factor (N) for each microarray was determined by obtaining the 20% trimmed mean of the human probe intensities that were detected one log₂ unit above TPT95 (TPT95 + 1) in all samples and with standard deviation of probe intensities among all samples less than 1.25. The log₂-transformed spot intensities for all 8816 features were normalized, by subtracting N from each spot intensity, and scaled by adding the grand mean of N across all microarrays. The mean probe intensities for each of the 1209 human probes on each of the 20 arrays were then determined by averaging the triplicate spot intensities. Spots flagged as poor quality during data extraction were omitted prior to averaging. The 1209 human non-control log₂-transformed, normalized, and averaged probe intensities were filtered to obtain a panel of 290 human miRNA probes showing probe intensity greater than one log₂ unit above TPT95 (> TPT95 + 1) in at least 10% of the samples.

Quality control: Sensitivity of the microarray hybridization was confirmed by detection of hybridization signal for all 11 spikes that were added during isolation and labeling well above TPT95. The array also contains a set of specificity control probes complementary to three different miRNAs. Each specificity control includes a perfect match, single mismatch, double mismatch, and shuffled version of the probe. Specificity of the hybridization was confirmed by detection of hybridization signal on the microarray for the perfect match probes and not the double mismatch and shuffled version of the probes. Reproducibility of the arrays was determined by monitoring the hybridization intensity for the triplicate human spots on each array.

Differential expression analysis: For statistical analysis, samples were binned in two groups: Healthy Controls and Pancreatic Cancer. The log₂-transformed and normalized spot intensities for the 290 detectable human probes were examined for differences between the groups by 1-way ANOVA using National Institute of Ageing Array Analysis software[30]. This ANOVA was conducted using the Bayesian Error Model and 50 degrees of freedom. A total of 116 probes exhibited significant differences between the healthy controls and the pancreatic cancer patients. Statistical significance was determined using the False Discovery Rate (FDR) method[31]; an FDR < 0.15 was considered significant in this study.

Hierarchical clustering of miRNA array data: Data for the 290 detectable human probes were clustered using Cluster 3.0 software[32]. Genes were median-centered prior to hierarchical clustering. Hierarchical clustering was conducted using Centered Correlation as the similarity metric and Average Linkage as the clustering method. Intensity scale shown is arbitrary.

Total RNA was extracted from 0.9 to 2.1 mL of plasma using TRI Reagent^® BD (Molecular Research Center, Cincinnati, OH) according to the manufacturer’s instructions with minor modifications. For down-stream quality control monitoring, a mixture of 10 synthetic miRNAs were added at a mass of 12.5 femtomoles/mL of biofluid following homogenization of the samples in Trizol-BD.

RNA isolated from plasma was diluted and used in 10 μL reverse transcription (RT) reactions using ABI (Applied Biosystems by Life Technology) miRNA-specific RT primers (Life Tech, Carlsbad, CA). For each miRNA and sample, RNA equivalent to 50 μL of biofluid was used in each RT reaction except for the probe sets for miR-642b-3p and miR-7, in which case RNA equivalent to 100 μL was used. The cDNA product was diluted 1:10 in water and 4.5 μL of the diluted product was combined in triplicate PCR-plate wells with 1 × ABI Universal PCR amplification mix and ABI miRNA-specific Taqman PCR primers in a final volume of 10 μL/well. The PCR plate was subjected to thermal cycling in an ABI StepOne Plus real-time PCR instrument. The cycling conditions were an initial incubation for 10 min at 95 °C, followed by 40 cycles of: 15 s at 95 °C, 1 min at 60 °C. The Cq (Quantification Cycle according to Minimum Information for Publication of Quantitative Real-Time PCR Experiments guidelines[33], also known as threshold cycle (C_T), or number of cycles required for the well to reach a specified threshold of fluorescence intensity) was determined for each well using ABI software and a threshold setting of 0.05. In the initial pilot experiment, only miRNA with Cq value at 36 or above were used for confirmation study. Triplicate Cq values were averaged to obtain the “mean Cq value.”

MiRNA was extracted from 2.0 to 4.0 mL of plasma using QIAamp circulating Nucleic Acid Kit (Qiagen, Germantown, MD) with QIAvac Connecting System (Qiagen), following manufacturer instructions. The concentration of extracted miRNA was determined using Agilent Small RNA kit with 2100 Bioanalyzer (Agilent, Santa Clara, CA), and was in the range of 10 to 100 ng.

RT-qPCR experiments were performed using TaqMan MiRNA Assays (Life Tech, Carlsbad, CA), for each miRNA, following manufacturer instructions. The TaqMan miRNA assays used were: hsa-miR-885-5p (ID#002296), hsa-miR-22-3p (ID#000398), hsa-miR-642b-3p (ID#462949_mat), and hsa-miR-3196 (ID#241941_mat). For RT, 1 to 10 ng RNA isolated from plasma was diluted and used in 5 μL of RT reactions with each miRNA-specific RT primers and MiRNA Reverse Transcription kit (Life Tech), without pre-amplification. The cDNA product was diluted 1:3 in water, and 5 μL of diluted cDNA was used in triplicate PCR-plate wells with miRNA-specific TaqMan PCR primer and Universal Master mix II (Life Tech). The qPCR reaction was performed on 7500 real-time PCR system (Life Tech). The Cq value required for the well to reach a specific threshold of fluorescence intensity was determined for each well using 7500 Software (Life Tech) with threshold setting at 0.05. All Cq values were 36 or less and were included for subsequent calculation. Triplicate Cq values were averaged to obtain the “mean Cq value”.

The normalization control used for the RT-qPCR experiments was hsa-miR-3196. The endogenous control has been selected based on both microarray and pilot RT-qPCR experiments. MiR-3196 demonstrated relatively consistent and stable expression in microarray analysis and also in a series of pilot RT-qPCR. The use of the same extracted plasma miRNA (patients and controls, n = 19), with the same miRNA input, yielded consistent Cq values at 29.33 ± 0.50.

Fold change calculation was performed following the Comparative C(T) Method[34]. The “mean Cq value” of hsa-miR-3196 was subtracted from the mean Cq value for each miRNA to obtain a “normalized mean ΔCq value” for each miRNA. The “normalized mean ΔCq value” for each miRNA was then converted to 2^-(ΔCq). Fold change of each miRNA was determined by calculating the ratio of the mean of 2^-(ΔCq) of all pancreatic cancer patients to the mean of 2^-(ΔCq) of all healthy controls; and the mean of 2^-(ΔCq) of all high risk controls to the mean of 2^-(ΔCq) of all healthy controls. The fold change values were transformed to a log₂ scale for the purpose of plotting the values on a continuum.

In order to further evaluate the RT-qPCR experimental results against those observed from the microarray analysis, microarray data for eight miRNAs (miR-642b-3p, miR-762, miR-4253, miR-885-5p, miR-18a, miR-486-5p, miR-7, and miR-22) were re-analyzed following the Comparative C(T) Method[34]. The spot intensities of all data points (in log₂) were normalized by subtracting the values for each miRNA from miR-3196 (as normalization control) to generate “normalized log₂ values”, which were used as the equivalent of “normalized mean ΔCq value” in RT-qPCR for further calculation. The fold change values are then calculated as stated above for RT-qPCR data analysis.

The CA19-9 serum marker test was performed by the Valley Hospital Histology Laboratory following the routine diagnostic laboratory testing protocol using Tosoh A1A-360 Immunoanalyzer (Tosoh Bioscience, Inc.). CA19-9 range above 47 U/mL is considered to be a positive test result.

The Wilcoxon-Mann-Whitney test was used to compare the fold changes observed among the patient groups for each miRNA. Receiver operating characteristic (ROC) curves were generated to evaluate the sensitivity and specificity of predicting cancer cases and control cases for each miRNA and for the combination of miRNAs. Area under the curve and their respective 95%CI were calculated for all ROC curves. P-values ≤ 0.05 were considered to be statistically significant. All analyses were done using IBM-SPSS software (Version 19).

The miRNA expression signature was profiled using custom miRNA microarray chips covering 1220 human miRNAs derived from the miRBase database, version 16, released in 2010[29]. The unsupervised hierarchical clustering of the 290 miRNAs with acceptable detection intensities is shown in Figure 1A. Remarkably, as depicted in the hierarchical dendrogram, the clustering pattern clearly separates the pancreatic cancer from most healthy controls.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 The microarray signature profile of circulating microRNA in pancreatic cancer patients and healthy controls. A: The unsupervised hierarchical clustering of 290 miRNAs that are differentially expressed among pancreatic cancer patients (n = 8) and healthy controls (n = 11). The dendrogram on top indicates the hierarchical clustering relationship between pancreatic cancer patients (in red) and healthy controls (in blue); B: Heat map depicting the subset of 42 miRNAs chosen for confirmation RT-qPCR study, arranged as indicated. Heat map color scale represents fold increase (red) or decrease (green) from baseline. MiRNA: MicroRNA.

To compile a set of targeted miRNA markers for further investigation and confirmation, the list of 290 miRNAs was narrowed down to 31 miRNAs with significant differential expression and combined with an additional 22 miRNAs, which have been shown to be candidate biomarkers for pancreatic cancer by other investigators[14,17-20,25]. The latter criterion was added to ensure the inclusion of miRNAs which may not have exhibited significant differential expression in this experimental setting but have otherwise been shown by others to be good candidates. As shown in Table 2, among the final set of 42 miRNAs, 11 miRNAs exhibiting the most significantly different expression levels (FDR ≤ 0.15) in this study (miR-194, miR-18a, miR-7, miR-26b, miR-301a, miR-106b, miR-16, miR-93, miR-106a, miR-19b, and let-7i) as well as in the previously published literature. The heat map of the 42 miRNAs in Figure 1B demonstrates a differential expression signature between pancreatic cancer patients and controls similar to the results observed from the unsupervised clustering in Figure 1A.

The panel of 42 miRNAs were subjected to further investigation using real time quantitative RT-PCR (RT-qPCR). In the initial pilot experiment using available miRNA TaqMan probes and testing the relative abundance of expression, 8 of the 42 miRNAs gave acceptable and most consistent signals (data not shown) and therefore were chosen for the subsequent confirmation study.

As shown in Figure 2, when comparing microarray and TaqMan RT-qPCR results side-by-side after normalizing both data sets with the same control, miR-3196, the relative expression level of each miRNA varied due to the differing nature of the experimental technology. However, three of the eight miRNAs, miR-642b-3p, miR-885-5p, and miR-22-3p, demonstrate consistent outcomes across the two methodologies with higher expression seen in pancreatic cancer patients than in healthy controls. Therefore, they were chosen for further validation study as a potential diagnostic panel.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Confirmation study of the expression profile of eight microRNAs. Microarray (solid bar) and RT-qPCR (open bar) data sets were analyzed based on the “Comparative C(T) Method”, normalized to miR-3196, and calculated as the ratio of mean fold change of pancreatic cancer patients (n = 8) to healthy controls (n = 11). MiRNA: MicroRNA.

The validation study was performed on a new independent cohort of 11 pancreatic cancer patients, 11 healthy controls and 11 high risk controls. As shown in Figure 3A, the three miRNAs, plotted using the raw data of “normalized mean ΔCq value”, differentiate pancreatic cancer patients from both the healthy and high risk controls. Likewise, as shown in Figure 3B, when determining the relative fold expression change by calculating the ratio of pancreatic cancer to healthy controls, expression of the three miRNAs is clearly up regulated in pancreatic cancer patients. For the high risk controls, on the other hand, the expression level is either comparable or lower when compared to the healthy controls. Furthermore, the relative expression levels of all three miRNAs are remarkably consistent in direction and magnitude between confirmation RT-qPCR (Figure 2) and validation RT-qPCR (Figure 3B), despite the use of different patient samples and independent experimental procedures. This lends support to the suggestion that they could play potential roles as diagnostic biomarkers for pancreatic cancer.

Open in New Tab   Full Size Figure   Download Figure

Figure 3 Validation of three-microRNA panel using quantitative real-time reverse transcription polymerase chain reaction. A: Box plot of relative expression of three miRNAs, based on normalized mean ΔCq values [mean quantification cycle (Cq) normalized to miR-3196] of pancreatic cancer patients (PC, n = 11), healthy controls (HC, n = 11), and high risk controls (HR, n = 11). The whiskers extend to the observations which are no more than 1.5 times the length of the box (interquartile range) away from the box. More extreme observations are considered outliers and are indicated as dots. "a" indicates P value ≤ 0.017 (0.05/3, a Bonferroni-adjusted α-level based on the 3 multiple comparisons performed among patient groups); B: Fold change expression levels of three miRNAs calculated as the ratio of mean fold change of PC/HC and HR/HC. MiRNA: MicroRNA.

To assess the potential use of the three miRNA as a diagnostic panel for pancreatic cancer, ROC analysis was performed on the validation data set for each of the individual miRNAs and the combination of the three. As summarized in Table 3 and shown in Figure 4B and 4C, miR-885-5p and miR-22-3p each demonstrated high sensitivity of 82% for cancer case identification and relatively high specificity of 73% and 82%, respectively, for identifying healthy controls. For miR-642b-3p (Figure 4A), a high sensitivity of 82% for identifying cancer cases was demonstrated, but a lower specificity for identifying healthy controls of 55% was observed. When using the optimal cut-point, as shown in Figure 4D, the composite of all three miRNAs yielded both a sensitivity and specificity of 91%.

Open in New Tab   Full Size Figure   Download Figure

Figure 4 Receiver operating characteristic curve analysis of three microRNAs. Validation study with 11 stage IIA/IIB pancreatic cancer patients and 11 healthy controls. A: Receiver operating characteristic (ROC) curve for the miR-642b-3p data alone [area under the curve (AUC) = 0.79]; B: ROC curve for the miR-885-5p data alone (AUC = 0.84); C: ROC curve for the miR-22-3p data alone (AUC = 0.86); D: ROC curve for all three miRNAs (miR-642b-3p, -885-5p, 22-3p) as a composite panel (AUC = 0.97). MiRNA: MicroRNA.

Given that the CA19-9 serum marker is the only marker currently available for routine diagnostic testing, we sought to compare CA19-9 results with the three-miRNA panel for predicting pancreatic cancer. The CA19-9 test has been performed on our cancer patient sample (n = 11) and non-cancer patient samples (healthy and high-risk controls, n = 22), and the sensitivity was observed to be 73% (8 out of 11 patients), and the specificity was observed to be 100% (all controls were below the reference point).

Pancreatic cancer is one of the most lethal human cancers with a mere 6% 5-year survival rate. Studies have shown that early detection is the best option available for controlling this disease. The goal of this study, therefore, is to explore and compile a diagnostic biomarker panel, based on microRNA (miRNA) in the circulating blood, for detection of pancreatic cancer at earlier stages.

A diagnostic panel of biomarkers for cancer is highly desirable because it would help to fight cancer at the earliest possible stage when the disease is still curable. However, for the past fifteen to twenty years, despite numerous studies and publications, this goal remains elusive with only scanty numbers of risk assessment and prognostic panels eventually developing into clinical tests. One of the major problems is study reproducibility. The results of one study cannot be readily reproduced by another study. In addition, gene expression profiles can vary greatly among the different stages for each type of cancer, as well as between blood-based markers and tissue-specific markers. Therefore, it is critical for each investigator to be clear on their general experimental strategy which can, in turn, address issues pertaining to the development of future clinical tests.

The current study focuses on identifying a panel of blood-based biomarkers for pancreatic cancer. If it is effective, pancreatic cancer detection can be performed by a simple non-invasive blood draw instead of an invasive procedure. The approach used in this study is innovative due to the fact that, in addition to the general strategy of microarray screening followed by polymerase chain reaction-based confirmation, the authors employed a second layer of validation experiments, using different experimental procedures, instrumentation, and lab personnel at an independent location. The results of the two experimental strategies are remarkably similar even with a new cohort of patient specimens, suggesting high validity of this diagnostic panel. Furthermore, the authors have included a group of high risk individuals as controls. High risk controls are subjects with a strong family history of pancreatic cancer (at least two first degree relatives with the disease). Therefore, they have inherited genetic susceptibility to developing pancreatic cancer and hence are genotypically closer to individuals with disease than to normal healthy controls. Remarkably, the three-miRNA panel identified in this study can differentiate pancreatic cancer patients from both normal and high risk controls, demonstrating its high sensitivity and specificity for pancreatic cancer.

The identified panel of three miRNAs can potentially be used as a diagnostic detection set for early stage pancreatic cancer.

MiRNA are small non-coding RNA approximately 18-24 nucleotides in size. The abnormal expression of miRNA found in patient blood is known to be associated with cancer progression. Hence, the utilization of miRNA biomarkers was proposed as a way to potentially detect pancreatic cancer.

This is a well written paper examining a potentially useful means of screening for pancreatic carcinoma. The authors would recommend publication of the manuscript.