Serum MicroRNAs as Biomarkers in Hepatitis C: Preliminary Evidence of a MicroRNA Panel for the Diagnosis of Hepatocellular Carcinoma

Early diagnosis of cirrhosis and hepatocellular carcinoma (HCC) due to chronic Hepatitis C (CHC) remain clinical priorities. In this pilot study, we assessed serum microRNA (miRNA) expression to distinguish cirrhosis and HCC, alone and in combination with the aminotransferase-to-platelet ratio (APRI), Fibrosis 4 (FIB-4), and alpha-fetoprotein (AFP). Sixty CHC patients were subdivided into 3 cohorts: Mild disease (fibrosis stage F0-2; n = 20); cirrhosis (n = 20); and cirrhosis with HCC (n = 20). Circulating miRNA signatures were determined using a liver-specific real-time quantitative reverse transcription PCR (qRT-PCR) microarray assessing 372 miRNAs simultaneously. Differentially-expressed miRNA candidates were independently validated using qRT-PCR. Serum miRNA-409-3p was increased in cirrhosis versus mild disease. In HCC versus cirrhosis, miRNA-486-5p was increased, whereas miRNA-122-5p and miRNA-151a-5p were decreased. A logistic regression model-generated panel, consisting of miRNA-122-5p + miRNA-409-3p, distinguished cirrhosis from mild disease (area under the curve, AUC = 0.80; sensitivity = 85%, specificity = 70%; p < 0.001). When combined with FIB-4 or APRI, performance was improved with AUC = 0.89 (p < 0.001) and 0.87 (p < 0.001), respectively. A panel consisting of miRNA-122-5p + miRNA-486-5p + miRNA-142-3p distinguished HCC from cirrhosis (AUC = 0.94; sensitivity = 80%, specificity = 95%; p < 0.001), outperforming AFP (AUC = 0.64, p = 0.065). Serum miRNAs are differentially expressed across the spectrum of disease severity in CHC. MicroRNAs have great potential as diagnostic biomarkers in CHC, particularly in HCC where they outperform the only currently-used biomarker, AFP.


Patient recruitment and characteristics
Sixty HCV-positive patients were retrospectively subdivided into 3 cohorts based on expert clinical assessment, transient elastography (FibroScan TM ; Echosens, Paris, France) and medical imaging as follows: mild disease without advanced fibrosis (F0-2; n=20); cirrhosis (F4; n=20); and cirrhosis with HCC (HCC; n=20). The diagnosis of HCC was established according to currentlyaccepted professional guidelines. [1] Patient demographics, biochemistry and relevant medical information were obtained from patients' medical records. Serum samples were obtained through Pathology Queensland collection centres. All samples were processed within 8 hours of blood draw and stored at -80˚C. A serum sample was collected from each patient and corresponding APRI (applying a cut-off of 1.0 to exclude cirrhosis),[2] FIB-4 (using a cut-off of 1.45 to exclude advanced fibrosis),[3] Child-Turcotte-Pugh (CTP) and model for end-stage liver disease (MELD) scores were calculated.

RNA extractions and reverse transcription.
For the screening phase, RNA was extracted from 200 µl of serum using the miRNeasy Serum/Plasma Kit (Qiagen; Hilden, Germany). Isolations were performed according to manufacturer's instructions with minor modifications to optimise results. Following the phenolchloroform phase separation, the interphase and organic layers were rehydrated with RNAse free water equal to the volume removed during the aqueous phase collection.[4] Thus, a second aqueous phase was obtained maximizing RNA recovery. Both aqueous phases were combined into one RNeasy MiniElute spin column and RNA isolation was continued.
[4] Following RNA elution, a second elution with identical settings was performed using 14µl RNAse free water in a new collection tube. All samples were assessed for ethanol contamination and RNA yield by Nanodrop TM (Thermo Fisher Scientific; Waltham, MA, USA). Serum extracted RNA was reverse transcribed using the miScript II RT Kit (Qiagen) following the manufacturer's instructions and RNA input recommendation for miScript miRNA PCR Arrays with 250 ng RNA. cDNA products were diluted 10-fold prior to microRNA PCR array run. During the validation phase, serum RNA was extracted using the Plasma/Serum RNA Purification Mini Kit (Norgen Biotek Corp; Thorold, Ontario, Canada) following the manufacturer's instructions with minor modifications. Prior to transferring the sample onto the Micro Spin Column the mixture was passed through a syringe with a 26G needle to break down lysis debris and prevent column clogging. RNA was eluted in 15 µl of RNAse free water and assessed on a Nanodrop TM . Extracted RNA was reversed transcribed using the miRCURY LNA TM universal RT microRNA PCR Kit (Exiqon, Vedbaek, Denmark) following manufacturer's instructions. A fixed RNA volume input of 4 µL was used (at the manufacturer's suggestion) due to limitations in quantifying circulatory microRNAs. cDNA products were diluted 1:40 prior to qRT-PCR use.

MiRNA PCR Array, qRT-PCR and data analysis.
During the screening phase, a miRNA PCR Array (Human Liver miFinder miScript miRNA PCR Array MIHS-3116ZG; Qiagen) was used to simultaneously measure expression of 372 liver-related miRNAs in all 60 samples. The most significant differentially-expressed miRNAs, (>2-fold change and P<0.05) were selected for further validation. Leading miRNA candidates were independently validated by qRT-PCR (miRCURY LNA™ miRNA kit and Exiqon primers, Supplementary Table 2). The miRNA PCR Array and miRCURY miRNA qRT-PCR were performed using the Lightcylcer480 (Roche; Basel, Switzerland) and CFX384 (Bio-Rad; Hercules, California) thermal cycler, respectively. MiRNA PCR Array and miRCURY qRT-PCR data were reported as crossing points (CP) and quantification cycles (Cq), respectively.
Statistical analysis of patient demographics was undertaken using the chi-square test for categorical variables, and either an ANOVA, Kruskal-Wallis test, unpaired t-test or Mann-Whitney test for continuous variables as informed by the D'Agostino & Pearson normality test. Significance was defined as P<0.05.
The endogenous reference miRNAs let-7i-5p and miRNA-23a-3p were selected for further validation studies based on a distance metric ranking combining the P-values from the pairwise comparisons of comparing mild disease (F0-2) vs cirrhosis (F4) and comparing cirrhosis (F4) vs HCC. The distance metric was calculated as the summation of the absolute value of the log transformed Pvalues. Validation qRT-PCR data were analysed using the 2 ΔCT method and expression values normalized to let-7i-5p and miRNA-23a-3p.

Panel design and k-fold cross validation.
Stepwise logistic regression using forward selection and backward elimination was used to derive microRNA panels for i) cirrhosis (F4) vs mild disease (F0-2), and ii) HCC vs cirrhosis (F4). Pairwise correlations between each of the miRNA were assessed to exclude any significant correlations within the models at the 5% level of significance. Model selection was based on the Akaike Information Criteria, the likelihood ratio test based on the change of the residual deviance, and by assessing the stability of the coefficient estimates.
The resulting model equation is as follows: To obtain the , the probability of the i subject having the outcome of cirrhosis (function A) or HCC (function B), use the following formula where function is the right hand side of the equation in Function A or B: The probability, , will range from 0 to 1.
The Youden's index is a technique used to determine the most appropriate cut-off value, which corresponds to a point on the ROC curve with the highest vertical distance from the 45% diagonal line. At this point, the true positive rate and the false positive rate is at the maximum possible.
[5] Using the Youden's index a panel specific cut-point for was determined. For the probability derived using function A, a value above the cut-point of 0.44 would classify the subject as cirrhosis. For the probability derived using function B, a value above the cut-point of 0.65 would classify the subject as HCC. K-fold cross-validation (5-fold) was used to assess the performance of the selected microRNA panels. Univariate and multivariable analyses were used to perform receiver operating characteristic (ROC) curve analysis resulting in area under the curve (AUC) with 95% confidence interval (CI), positive predictive value (PPV), negative predictive value (NPV), sensitivity, specificity and accuracy. All calculations were performed using the cvAUC and the ROCR package on R version (version 3.