For this work, we used complete Hepatitis C virus (HCV) reference strain H77 genomic sequences (genotype 1a, EMBL accession number: NC_004102), which included the ORFs and the 5’ and 3’-UTRs of the viral RNA. Figure 1 shows a flowchart of the computational prediction process. Complete antigenomic sequences were obtained in silico by reverse complement using software from the MEGA program [22] .

VMir program [23] was used to analyze the HCV H77 strain sequences. VMir is an ab initio prediction program specifically designed to identify pre-miRNA in viral genomes. Using this approach, HCV H77 antigenome sequences were analyzed for possible pre-miRNA hairpin structures, using default as well as stringent filtering parameters (minimum hairpin size of 60 nucleotides (nt), maximum hairpin size of 120 nt, minimum hairpin score of 115, minimum window count of 25). VMir scores were calculated according to Grundhoff [23] . Only

pre-miRNA hairpin structures with a minimum score of 140 were considered.

To discriminate real pre-miRNAs from other hairpin structures (pseudo hairpins) we employed MiPred [24] (available at: http://www.bioinf.seu.edu.cn/miRNA/), which decides whether each hairpin is either a pseudo-pre- miRNA-like hairpin or a real pre-miRNA using random forest prediction model.

With the purpose of extracting mature miRNA: miRNA* duplexes from pre-miRNA hairpins, we employed the MaturePred Web Tool (available at: http://nclab.hit.edu.cn/maturepred/). This software uses a model based on Support Vector Machine (SVM) that predicts the starting position of an miRNA by performing discriminant analysis against the query hairpin structure using various features of known real/pseudo miRNA: miRNA* duplexes as a training set (position-specific features, energy-related features, structure-related features, and stability- related features) [25] .

Similarity among predicted miRNA sequences in HCV H77 strain (genotype 1a) and corresponding sequences of all HCV types and sub-types isolated elsewhere were observed by means of the use of a curated reference alignment (which includes all HCV types and subtypes), obtained from the HCV Database (available at: http://hcv.lanl.gov/content/sequence/NEWALIGN/align.html) [26] .

The RNAfold web server [27] (available at http://rna.tbi.univie.ac.at/cgi-bin/RNAfold.cgi) was used to predict the secondary structure of pre-miRNAs. Only default parameters were used. In all cases, folding structures with centroid were depicted.

In order to identify the regulatory relationships between predicted miRNAs and their targets gene transcripts we employed miRTar (available at: http://mirtar.mbc.nctu.edu.tw) [28] . The parameters assigned for the hybridization were set to −14 kcal/mol of minimum free energy (MFE) for all hybridizations as a cut-off value and an alignment score ≥ 140. We study the relation among predicted miRNAs and their targets 3’untranslated regions (3UTR) gene transcripts involved in six different metabolic pathways (apoptosis, chemokine signaling pathway, cytokine-cytokine receptor interaction, Jak-STAT signaling pathway, mTOR signaling pathway and T cell receptor signaling pathway). In order to enrich the identified genes with connection to specific functional terms, the potential targets were analyzed using KEGG pathway maps from miRTar [28] to elucidate the biological roles of miRNAs in biological pathways. The p-values of the miRNA target genes in the metabolic pathway were also determined. Only genes with a p-value ≤ 0.05 were included in the analysis.

In order to confirm an effective hybridization between 3’UTR gene transcript targets and predicted HCV miRNA, we employed the RNAhybrid tool [29] (available at: http://bibiserv2.cebitec.uni-bielefeld.de/rnahybrid/). RNA hybrid is a tool for finding the minimum free energy hybridization of a long and a short RNA and widely used for miRNA target prediction. Pairing energy or minimum free energy (MFE) indicates the stability of the hybridization. The MFE was set at −14 kcal/mol for all hybridizations, as a cut-off value.

Computational prediction represents a widely used and effective strategy to identify novel miRNAs that can be further examined and validated by experimental approaches. RNA virus antigenomes are present in the cytoplasm and form various secondary structures that constitute potential pre-miRNAs. In order to observe whether HCV antigenome could be folded into putative pre-miRNA stem-loop structures, we first analyzed its putative miRNA-encoding capacity using full-length antigenome sequences of HCV H77 reference strain were screened (Figure 2). As default settings were used, 251 candidate hairpins (Figure 2(a)) were identified. To avoid bona fide pre-miRNA hairpin, we filtered VMir output using custom setting, establishing a cut-off value of 60 nucleotides (nt) for minimum hairpin size, 120 nt for maximum hairpin size, and 115 minimum hairpin score. These settings revealed 82 pre-miRNA hairpins (Figure 2(b)) that were selected as potential hairpins. These results suggested the potential miRNA-encoding capacity of the antigenome of this HCV strain.

From these 82 potential pre-miRNAs, potential stem-loop structures with VMir scores higher of 140 were selected for further analysis. Then, in order to confirm the presence of these pre-miRNA structures, we classified

them into real or pseudo pre-miRNAs. Ten different pre-miRNA structures identified by VMir, nominated MD11, MD36, MD19, MD18, MD3, MD39, MD30, MD1, MD26 and MD14, were classified as real pre-miRNAs (data not shown). These structures are present in different different HCV genomic regions (see Table 1).

Comparable to other RNA viruses previously studied, HCV H77 strain miRNAs can be located on either of the two arms in the secondary hairpin structure (see Table 1). Of the ten miRNAs identified in the antigenome of HCV H77 strain, four are located in the 5’-arm of the stem loop hairpin structure, while six are in the 3’-arm (Table 1).

HCV presents a high degree of genetic variability. The high error rate of RNA-dependent RNA polymerase and the pressure exerted by the host immune system, has driven the evolution of HCV into 7 different genotypes and a growing number of subtypes. In order to observe if mature miRNA sequences present in the predicted miRNAs found in the antigenome of HCV H77 strain are conserved among different HCV types and subtypes, we search for the similarity among that sequences and corresponding sequences of HCV strains included in a cu-

^aPositions relative to HCV strain H77 antigenome. ^bThe region of the HCV genome is indicated. ^cThe VMir score is shown. Mature miRNA sequences are shown in bold.

rated HCV sequence dataset that includes all types and subtypes. The results of these studies revealed that all mature miRNA sequences present in the 10 real pre-miRNA structures found in the antigenome of HCV H77 strain are conserved only among HCV sub-genotype 1a. Interestingly, mature miRNA sequences present in pre-miRNA-MD39 share 100% similarity among HCV strains belonging to genotypes 1a, 1b, 1c, 2a, 2c, 2k, 5, 6 and 7, and 95% similarity to genotype 2a, 3a, 3b, 3k and 4 (data not shown). This revealed a high degree of conservation of the mature miRNA sequences of this predicted miRNA among HCV strains. In order to observe if this pre-miRNA structure is conserved among HCV genotypes, the secondary structure of pre-miRNA-MD39 was predicted using corresponding sequences of pre-miRNA-MD39 of different HCV types and subtypes. The results of these studies are shown in Figure 3.

As it can be seen in the figure, very similar secondary structures are obtained using HCV strains of all types. This revealed that this structure is conserved among HCV genotypes.

To understand the dynamics between viral miRNAs and their targets is extremely important to understand the complexity of biological regulation and virus-host interaction. In silico prediction of miRNA targets provides a suitable approach for identifying potential target sites based on their complete or partial complementarity with the miRNAs. For these reasons, pairwise comparison of human 3’-UTR gene transcripts involved in six different metabolic pathways and miRNA-MD39 mature sequences were performed. We observed 17 transcript targets for the predicted HCV-miRNA-MD39 (Table 2). These transcripts are mostly involved in apoptosis as well as immune responses pathways of the host cell (Table 2).

Gene ontology is a useful tool for the mining of gene datasets and their functional annotations. In order to gain insight into these matters, we performed a functional enrichment analysis using KEGG pathway maps. A total of 38 genes were identified (Figure 4).

In order to reconfirm effective hybridizations among the identified 3’UTR gene transcript targets and HCV- miR-MD39, we observed their hybridization patterns and calculated the minimum free energy of each hybridization. We have used a MFE of −14 kcal/mol as a cut-off value for potential miRNA pairing. The results of these studies are shown in Table 3. As it can be seen in the table, effective hybridizations were found in all cases included in this analysis.