Conserved RNA secondary structures and long-range interactions in hepatitis C viruses

Abstract

Hepatitis C virus (HCV) is a hepatotropic virus with a plus-strand RNA genome of ∼9.600 nt. Due to error-prone replication by its RNA-dependent RNA polymerase (RdRp) residing in nonstructural protein 5B (NS5B), HCV isolates are grouped into seven genotypes with several subtypes. By using whole-genome sequences of 106 HCV isolates and secondary structure alignments of the plus-strand genome and its minus-strand replication intermediate, we established refined secondary structures of the 5′ untranslated region (UTR), the cis-acting replication element (CRE) in NS5B, and the 3′ UTR. We propose an alternative structure in the 5′ UTR, conserved secondary structures of 5B stem–loop (SL)1 and 5BSL2, and four possible structures of the X-tail at the very 3′ end of the HCV genome. We predict several previously unknown long-range interactions, most importantly a possible circularization interaction between distinct elements in the 5′ and 3′ UTR, reminiscent of the cyclization elements of the related flaviviruses. Based on analogy to these viruses, we propose that the 5′–3′ UTR base-pairing in the HCV genome might play an important role in viral RNA replication. These results may have important implications for our understanding of the nature of the cis-acting RNA elements in the HCV genome and their possible role in regulating the mutually exclusive processes of viral RNA translation and replication.

INTRODUCTION

Hepatitis C virus (HCV) is a hepatotropic plus-strand RNA virus causing acute or chronic liver disease that often leads to liver cirrhosis and hepatocellular carcinoma (Yamane et al. 2013; Zeisel et al. 2013). As the prototype member of the genus Hepacivirus within the family Flaviviridae, HCV is a representative of a large group of plus-strand RNA viruses that encode their own RNA-dependent RNA polymerases (RdRp). Since these enzymes lack 3′–5′-exonuclease proofreading activity, RNA replication of these viruses gives rise to a large number of progeny genomes with replication errors. Accordingly, a huge number of HCV isolates has been found worldwide, grouped into seven genotypes with several subtypes (Simmonds 2013; Jackowiak et al. 2014). For these reasons, HCV can be used as a model system to analyze the conservation of functional RNA sequences and secondary structures as well as long-range RNA–RNA interactions in a viral genome.

The single-stranded HCV genome of ∼9.6 kb is translated at the rough endoplasmatic reticulum in the cytosol. A single open reading frame (ORF) encodes for a single polyprotein that is co- and post-translationally processed by host and viral proteases into 10 cleavage products. These are classified as structural (core, envelope glycoproteins E1, E2, and p7) and nonstructural proteins (NS2, NS3, NS4A, NS4B, NS5A, and the RNA-dependent RNA polymerase [RdRp] residing in NS5B) (Lohmann 2013; Moradpour and Penin 2013).

The ORF encoding the polyprotein is flanked by highly structured 5′ and 3′ untranslated regions (UTRs), which are indispensable for translation and replication of the viral genome (Liu et al. 2009; Lohmann 2013). The 5′ UTR consists of four structured loop domains (SLI–SLIV), which have been extensively studied for genotypes 1 and 2 (Brown et al. 1992; Honda et al. 1999a; Friebe and Bartenschlager 2009; Lohmann 2013; Niepmann 2013). The SLI and SLII domains (Fig. 1) are essential for RNA replication (Friebe et al. 2001). This region contains two binding sites for the liver-specific microRNA-122 (miR-122) that stimulates HCV RNA translation and replication (Jopling et al. 2005; Henke et al. 2008) and enhances RNA stability (Shimakami et al. 2012; Sedano and Sarnow 2014) and translation (Henke et al. 2008).

FIGURE 1.

Overview of previously known (black) and novel (gray) RNA stem–loop structures of the HCV RNA genome. (A) 5′ UTR and core region SLV and SLVI of the viral plus-strand. (B) CRE region and 3′ UTR of the viral plus-strand. (C) 3′ End of the viral minus-strand. (D) 5′ End of the viral minus-strand.

Partially overlapping with these replication signals is the internal ribosomal entry site (IRES) which is formed by the domains SLII–SLIV along with the first 20–40 nt of the coding region (Friebe et al. 2001). This IRES confers cap-independent RNA translation (Brown et al. 1992; Honda et al. 1996).

Domains SLII, SLIIIa, d, e, and f as well as SLIV of the IRES are important for the recruitment of the ribosomal 40S subunit (Pestova et al. 1998; Spahn et al. 2001; Kieft et al. 2001, 2002). For this process the pseudoknot of the IRES has been shown to be essential (Wang et al. 1995; Berry et al. 2011). Additionally, the initiation factor eIF3 is recruited to the IRES mainly by determinants in domains SLIIIb and c (Buratti et al. 1998; Sizova et al. 1998; Kieft et al. 2002; Siridechadilok et al. 2005). Based on the crystal structure of the complete HCV IRES, a model has been proposed how the preinitiation complex might assemble to allow efficient RNA translation (Pérard et al. 2013).

The 3′ UTR is essential for viral minus-strand RNA synthesis (Yanagi et al. 1999; Friebe and Bartenschlager 2002; Yi and Lemon 2003; You and Rice 2008). It consists of a variable region (VR), a poly(U/UC) tract, and a conserved structure called 3′X or X-tail (Tanaka et al. 1995; Lohmann 2013). The variable region is poorly conserved among HCV isolates; this region is not essential, but enhances RNA replication (Friebe and Bartenschlager 2002). In the variable region resides the conserved stem–loop (SLIV) containing the stop codon of the polyprotein (Lohmann 2013). Conservation of a second stem–loop in the 3′ UTR among HCV isolates was not convincingly shown. The poly(U/UC) tract has a variable length, ranging between 30 and 90 nt (Lohmann 2013). The highly conserved X-tail can form two different conformations. One conformation consists of three experimentally verified stem–loop structures (SLI, SLII, SLIII, numbered in 3′ to 5′ direction with respect to the plus-strand) (Blight and Rice 1997) and is crucial for RNA replication (Friebe and Bartenschlager 2002). In an alternative conformation of the X-tail, only two stem–loops are formed. Here, SLI remains the same, but SLII and SLIII are merged to form a single stem–loop which includes a so-called dimerization linkage sequence (DLS) in its apical loop that was proposed to be necessary for viral RNA genome dimerization and possibly for genome encapsidation (Ivanyi-Nagy et al. 2006; Shetty et al. 2010).

It has been reported that the adjacent NS5B coding region might contain five structured stem–loops (Tuplin et al. 2002; Diviney et al. 2008) designated: 5BSL1 (also called SL9033), 5BSL2 (SL9132), 5BSL3.1 (SL9217), 5BSL3.2 (SL9266), and 5BSL3.3 (SL9324). Of these, 5BSL3.2 has been identified as a cis-acting replication element (CRE) that is important for viral replication (Lee et al. 2004; Friebe et al. 2005; Diviney et al. 2008). 5BSL3.2 is involved in several long-range interactions (LRIs). Its apical loop interacts with the SLII region of the X-tail, while its internal bulge can interact either with the IRES SLIIId apical loop (Romero-López and Berzal-Herranz 2009) or alternatively with a sequence called CRE 9110 which is located between 5BSL1 (SL9033) and 5BSL2 (SL9132); see Fig. 1 (Diviney et al. 2008).

The minus-strand of HCV is a replication intermediate that serves as template for the synthesis of progeny plus-strand RNAs. This minus-strand contains several CREs at its 3′ end, but owing to different G–U base-pairing and single-stranded regions, the secondary structures in this region (Smith et al. 2002) are very different from their complementary counterpart in the 5′ UTR of the plus-strand. In fact, the 3′ end of the minus-strand comprises several stem–loop structures, including stem–loops SL-I′, SL-IIz′, and SL-IIy′ that are crucial for RNA replication, and SL-III′ and SL-IV′ that contribute less to RNA replication efficiency (Friebe and Bartenschlager 2009).

Only little is known about the 5′ end of the minus-strand. Smith and Wu (2004) have shown that this part can form a mirror image of the structure of SLI of the X-tail, called SLα. The authors suggest that this structure protects the minus-strand from degradation by cytoplasmic exonucleases. However, thus far no structures corresponding to SLII and SLIII of the plus-strand could be identified in the minus-strand RNA (Smith and Wu 2004).

In this study, we present refined secondary structures of the 5′ UTR, the CRE-region in the NS5B RdRp coding region and of the 3′ UTR. We show alternative RNA secondary structures, flexible regions and several novel possible long-range interactions which are conserved among all HCV isolates analyzed. Based on all available HCV genomes, we hypothesize a 62-bp long 5′-UTR SLII/DLS interaction as the long-sought putative circularization sequence.

RESULTS AND DISCUSSION

This work is based on computational predictions which were performed without applying prior knowledge about secondary structures of HCV (except the pseudoknot in the 5′ UTR). Obtained results were then compared with results from the literature. We are able to confirm and extend most of the previously described secondary structures of 5′ UTR, CRE and 3′ UTR for all HCV genotypes (Fig. 1). The results of our comprehensive secondary structure analysis, including novel possible long-range interactions, are summarized in Figure 7A below.

FIGURE 7.

Long-range interactions in 5′ UTR, CRE, VR, and X-tail. (A) Overview. (Gray) Known interactions (No. 13–17 from the literature, validated by our analysis for all examined isolates); (green) new interactions (from our calculations). Numbers indicate different interactions shown in detail in B. (B) Interactions for all available sequences labeled with numbers corresponding to A. Bottom numbers indicate number of sequences in the alignment used for the interaction. Interaction No. 1 can be extended for a possible circularization. (C) Possible circularization of the HCV RNA. The interaction of SLII and DLS of HCV plus-strand RNA can be extended to at least 62 bp in all available 19 isolates. (D) Corresponding interaction of minus-strand RNA.

We used a full-genome alignment of 106 isolates from all known genotypes with an average pairwise sequence identity (PSI) of 70.02%. Interestingly, the 5′-UTR alignment is based on a much higher sequence identity of 94.2%, showing the high functionality of this compact and essential region, including the IRES and the miR-122 binding sites. The 3′-UTR alignment (based on 19 isolates) has a PSI of 95.9%. This extremely high degree of conservation may reflect overlapping functions of the 3′-UTR sequences that do not allow for sequence variability like covariations in stem regions. Moreover, single mutations in the 3′ UTR can lead to losses of crucial binding sites (Niepmann 2013). The CRE-region is part of the coding region but remains with 73.6% PSI above the genome wide average, which may to be due to the multiple functions of this region.

5′-UTR-IRES region

For 5′-UTR secondary structure prediction we used 86 out of 106 isolates covering 60 out of 63 subtypes. The remaining 20 isolates comprise incomplete 5′ UTRs. The prediction based on structure conservation is very similar to previously proposed secondary structures (Honda et al. 1999a; Pang et al. 2011). However, there are some differences compared with previous studies (shown in Fig. 2A):

1. The predicted SLII is contradicting the structure of Honda et al. (1999a) and studies referring to it (Masante et al. 2008; Friebe and Bartenschlager 2009). However, our prediction is in agreement with the work of Zhao and Wimmer (2001) and Lukavsky et al. (2003) and with the SHAPE analysis of Pang et al. (2011) (Supplemental Fig. 1C).

2. The predicted apical stem of SLIIIb contains a shift of nucleotides compared with the data of Honda et al. (1999a). However, the unpaired nucleotides in the loop region which are part of the binding site for initiation factor eIF3 (Fraser and Doudna 2007) are identical.

3. We confirm the secondary structure of SLIIId as previously described and validated by experimental data (Supplemental Fig. 1A; Brown et al. 1992; Kolupaeva et al. 2000; Lukavsky et al. 2000; Odreman-Macchioli et al. 2000; Romero-López and Berzal-Herranz 2009; Pang et al. 2011; Hajdin et al. 2013). These secondary structures are additionally stabilized by an LRI introduced in Romero-López and Berzal-Herranz (2012) (No. 14 in Fig. 7, below).

FIGURE 2.

(A) Alignment based secondary structure prediction of the 5′ UTR of 86 isolates. The alignment was calculated by LocARNA, the consensus sequence and structure by RNAalifold. The sequence for SLI is contained in only nine isolate sequences and therefore appears here as white. SLII is different from that described in Honda et al. (1999a), but in agreement with Zhao and Wimmer (2001) and SHAPE analysis of Pang et al. (2011) (see Supplemental Fig. 1C). Here, we show an alternative secondary structure for SLIIId with a slightly better MFE and conservation compared with the known structure. (Blue sequences) Forbidden to pair by RNAalifold in order to build the pseudoknot that is essential for the recruitment of the ribosome and therefore for viral protein translation (Berry et al. 2011). (Gray) Secondary structure different from those proposed by Honda et al. (1999a). (B) Consensus sequences and secondary structures of the CRE, VR, and X-tail in two alternative foldings. CRE and VR based on 96 isolates, X-tail based on 19 isolates. Apical regions of 5BSL1, 5BSL2, and 5BSL3.1 show up as described previously (Diviney et al. 2008). Even though it seems that the apical part of the stems can vary between isolates, the consensus secondary structure leaves little variability (see dotplot in Fig. 3). SLIV is present in all tested isolates but appears white since the stem is variable (see supplement, STK file for variable region). (Gray) Differences in secondary structures of 5BSL1 (SL9033), and 5BSL2 (SL9033) compared with data of Diviney et al. (2008), of 5BSL3.1 (SL9132) compared with data of Diviney et al. (2008) and Romero-López et al. (2014), of SLIV (Kolykhalov et al. 1996), of SLII (Kolykhalov et al. 1996; Blight and Rice 1997; Ito and Lai 1997), and of DLS (Ivanyi-Nagy et al. 2006; Shetty et al. 2010; Romero-López et al. 2014). (Rectangle) Consensus of RNAalifold to establish 5BSL3.2, instead of the interaction in Figures 3 (oval) and 7, No. 15 (Diviney et al. 2008). A detailed explanation of the base pair color code is available in the Supplemental Material.

In addition, we propose an alternative secondary structure SLIIId* (Fig. 2A). This proposed secondary structure has a slightly better alignment minimum free energy (consisting of the average MFE and a covariance term, hereafter named MFE) and can be formed by all isolates tested. In this alternative structure, the position of SLIIId is shifted: The previously proposed apical loop region is base-pairing, whereas the region between SLIIId* and SLIIIe is base-pairing differently. We hypothesize the possibility of a coexistence of both structures in equilibria (Supplemental Fig. 1B).

Rep-Core

The secondary structure of the first 177 nt of the Core region is known to be involved in replication. The secondary structure yielded by our analysis (Supplemental Fig. 2C) is in agreement with the previously known structures SLIV, SLV, and SLVI (Wang et al. 2000; McMullan et al. 2007; Vassilaki et al. 2008).

CRE/Variable Region (VR)

The cis-acting replication element (CRE), located at the 3′ end of the NS5B coding region, plays an important role in RNA synthesis. The CRE-region consists of at least five structured stem–loops: 5BSL1 (SL9033), 5BSL2 (SL9132), 5BSL3.1 (SL9217), 5BSL3.2 (SL9266), and 5BSL3.3 (SL9324) (Tuplin et al. 2002; Lee et al. 2004; Friebe et al. 2005; Diviney et al. 2008). According to our analysis, all 96 isolates can form these five stem–loops (see Fig. 2B). Comparing the CRE/VR and X-tail region to previous literature, we obtain the following new features:

1. Here, we present for the first time a consensus secondary structure of 5BSL1, 5BLS2, and 5BLS3.1 based on 96 isolates. The three apical loop regions are identical to those proposed by Diviney et al. (2008); however, the stem regions differ significantly (see Fig. 2B, gray parts). In particular, the region shown as “SL9110” in Romero-López et al. (2014) that interacts with the internal loop in 5BSL3.2 appears not as a hairpin loop in our consensus calculations (Fig. 3, oval). In contrast, this sequence is located in the left basal region of 5BSL2. The basal regions of 5BLS1 and 5BSL2 may also assume alternative secondary structures or may be even single-stranded (Fig. 3); in the latter case, the sequence that interacts with 5BSL3.2 would be located in a single-stranded region between the core parts of 5BSL1 and 5BSL2.

2. We show that the NS5B stop codon is contained in a stem–loop (SLIV) in all available sequences (Supplemental Material, Stockholm file Dat. 1). This was unexpected due to the fact that the primary sequences downstream from the NS5B stop codon are highly variable among isolates, indicated by white colored base pairs in Figure 2B.

3. Several studies propose a second stem–loop downstream from the 3′SLIV (Kolykhalov et al. 1996; Ito and Lai 1997; Yi and Lemon 2003) but they did not provide structure probing data. From analyzing this region using secondary structure alignments, we were not able to find a suboptimal consensus structure that confirms this loop. However, when folding each sequence separately, an extremely variable stem–loop can be found in genotypes 1, 2, and 4, but rarely in genotypes 3 and 6 (Supplemental Material, Stockholm file Dat. 1).

FIGURE 3.

RNAalifold base-pairing probability matrix of the CRE and VR region, based on 96 isolates. The upper half of the matrix shows all base pair probabilities <10⁻⁶. The lower half shows the base pair probabilities of the MFE structure. (Oval) The interaction of the internal bulge of 5BSL3.2 with basal region of 5BSL2 (Fig. 7, No. 15), described previously in Diviney et al. (2008) and Tuplin et al. (2012). Dotplot and structure of CRE, VR, and X-tail region (19 isolates) without constraints are available in the Supplemental Material (Supplemental Fig. 3A,B).

Thus, we can only speculate that formation of this putative stem–loop may not be of general functional importance.

X-tail

According to the dotplot of the X-tail region based on analysis of the available 19 isolates from 13 subtypes (Fig. 4), we find up to four possible secondary structures of the X-tail (Fig. 4A,B are in agreement with the SHAPE analyses of Romero-López et al. 2014), which demonstrates a high flexibility of the very conserved sequences in particular in the region of 3′SLII and 3′SLIII which can form a variety of different secondary structures. These structure variants which are compatible with all considered sequences include:

1. Formation of the terminal 3′SLI and the known upstream stem–loop that includes the dimer linkage sequence (DLS) (Fig. 4A; Ivanyi-Nagy et al. 2006; Shetty et al. 2010).

2. The commonly accepted stem–loop structures 3′SLI, 3′SLII, and 3′SLIII (Fig. 4B; Blight and Rice 1997; Friebe and Bartenschlager 2002). However, in our analysis the apical part of 3′SLII is predicted different from Friebe et al. (2005). Taking into account the known long-range interaction of 5BSL3.2 with 3′SLII, we consider that the entire apical region may be locally unpaired.

3. Another, previously unknown, predicted secondary structure corresponds to the SLI and an alternative stem–loop containing the DLS (Fig. 4C).

4. Finally, another alternative structure can be obtained that contains SLI, SLIII, and an alternative version of SLII that is shifted by 4 nt compared with the previously known version of SLII (Fig. 4D).

FIGURE 4.

Alignment based secondary structures of the X-tail using sequences of 19 isolates. (Top) RNAalifold base-pairing probability matrix of the X-tail region. (A) Consensus secondary structure without applying any constraints. (B) Consensus secondary structure using constraints that force the postulated structure of Blight and Rice (1997). (C,D) Previously unknown secondary structures of the X-tail. All alternative structures contain SLI but show shifted hairpins compared with DLS or SLII.

5′ End of minus-strand

Our prediction for the 5′ end of the minus-strand (based on 19 isolates) returns an overall structure similar to the complementary plus-strand (Fig. 5A), consistent with previous studies (Smith and Wu 2004; Ivanyi-Nagy et al. 2006), see Supplemental Figure 5 for dotplot, although structures mirroring 3′SLII and 3′SLIII were previously not identified by structural probing (Smith and Wu 2004; Ivanyi-Nagy et al. 2006). We forced RNAalifold to calculate this structure, resulting in an only slightly higher MFE (Fig. 5B, −58.18 kcal/mol). Interestingly, the probing data of Ivanyi-Nagy et al. (2006) also do not contradict the latter structure. Additionally, we were able to detect a possible new secondary structure (similarly proposed in Smith and Wu 2004), displayed in Figure 5C, with a MFE of −60.13 kcal/mol. In conclusion, different structures seem possible for the 5′ end of the minus-strand.

FIGURE 5.

Alignment based secondary structure of the 5′ end of the minus-strand with 19 isolates. (A) Without constraints; (B) with constraints aiming at a similar structure like the mirror plus-strand; (C) suboptimal structure according to dotplot (Supplemental Fig. 5).

3′ End of minus-strand

Our analysis comprising 86 isolates returns a consensus MFE structure with eight stem–loop structures (Fig. 6A). This secondary structure contains the SLI′, SLIIz′, SLIIy′, SLIIIa′, SLIIIb′ (Smith et al. 2002; Dutkiewicz et al. 2008; Mahias et al. 2010) and SLIV′ (Smith et al. 2002) which are already known from experimental data. However, there is no unique structure between SLIIIb′ and SLIV′ (Fig. 6A).

FIGURE 6.

Alignment based secondary structure of the 3′ end of the minus-strand with 86 isolates. (A) RNAalifold base-pairing probability matrix. The upper half of the matrix shows all base pair probabilities <10⁻⁶. The lower half shows the base pair probabilities of the MFE structure. (B) Consensus secondary structure without constraints. (C) Consensus secondary structure with constraints: The last 3 nt at the very 3′ end were forced to be unpaired (blue). The resulting structure mirrors the secondary structure of the domain SLIII from RNA plus strand. (D) Consensus secondary structure with constraints forcing the postulated structure of Smith et al. (2002). (E) Consensus secondary structure with constraints forcing the postulated structure of Dutkiewicz et al. (2008). (F) Consensus secondary structure with constraints forcing the postulated structure of Schuster et al. (2002). (Dots) Nucleotides base-pairing with an extended version of the 3′ end; for details see Schuster et al. (2002). (Connected rectangles) Further interactions calculated by LocARNA. White base-pairings are derived from isolates being sequenced only partially.

The following secondary structures appear with a similar MFE: Figure 6B shows the consensus structure with best MFE for all available isolates. In order to make this secondary structure comparable with those known from the literature, in a separate approach we disallowed only the last 3 nt to bind. Surprisingly, this approach returned a secondary structure of the domain SLIII (Fig. 6C) that mirrors the upper regions of the SLIII in the plus strand, with a similar MFE score. When forcing the experimentally verified secondary structures from Schuster et al. (2002), Smith et al. (2002), and Dutkiewicz et al. 2008), we obtained the structures shown in Figure 6D–F which are also very conserved but have MFEs considerably worse than that of the calculated optimal structure (Fig. 6B).

When we compare the structures from Figure 6B–F with available probing data from different sources (Table 1), the secondary structure of Figure 6D (Smith et al. 2002) appears most accurate among the proposed probing data. We suppose that (in analogy to RNAs from other organisms, e.g., bacteria) not a rigid structure is present in solution, but dynamic changes between closely related structures can occur in solution even in one molecule (Narberhaus 2010).

TABLE 1.

Number of nucleotides contradicting experimental T1/T2/A/Pb probing data in proposed secondary structure models Fig. 6B–F

Long-range interactions

When searching for long-range interactions in 5′ UTR, Rep-Core, CRE, VR, and X-tail, we could verify several known long-range interactions in every HCV genome analyzed:

1. The interaction of SLIIId–5BSL3.2 (No. 14 in Fig. 7A,B, obtained in 104 out of 104 sequences) (Romero-López and Berzal-Herranz 2012) possibly blocks HCV IRES-dependent translation and therefore may have a critical role in the viral replication cycle (Romero-López and Berzal-Herranz 2009; 2012). Furthermore, we were able to extend this interaction (Supplemental Fig. 6).

2. The interaction of the sequence called CRE “9110” with the internal bulge of 5BSL3.2 (No. 15 in Fig. 7A,B; Tuplin et al. 2012; Romero-López et al. 2014) is consistent in 106 out of 106 isolates and was verified for isolate Con1 (genotype 1b) by SHAPE analysis (Tuplin et al. 2012).

3. The interaction of the apical loop of 5BSL3.2 and 3′SLII (No. 16 in Fig. 7A,B, consistent in 19 out of 19 isolates) (Friebe et al. 2005; Liu et al. 2009) is necessary for replication (Friebe et al. 2005) and was speculated to be involved in a control switch (Romero-López et al. 2014).

4. The interaction of the SLVI (Core) and the single-stranded region between 5′-UTR SLI and SLII (No. 17 in Fig. 7, in 77 out of 77 isolates) has been described before in several studies (Honda et al. 1999b; Wang et al. 2000; Kim et al. 2003; Beguiristain et al. 2005). The interaction of 5′ UTR and Rep-Core is displayed in Supplemental Figure 2A. Apart from this interaction the 5′ UTR and Rep-Core seem to be independent.

In addition to the above known interactions, we identified several novel possible long-range interactions, described in the following sections.

Collapse of 3′-UTR structure

When folding the entire CRE/3′-UTR region we obtain a conserved interaction of 7 nt of the basal 5BSL2 and of the basal 3′SLII (No. 8 in Fig. 7A,B). This new interaction results in an additional hairpin directly adjacent to the 3′SLIII that is equivalent to the apical part of the DLS. This finding reinforces the assumption that the secondary structures are in a dynamic equilibrium and may provide a “switch” between translation and replication (Ivanyi-Nagy et al. 2006; Shetty et al. 2010; Romero-López et al. 2014). This switch might even be initiated by miRNA-122, binding to the downstream seven interacting nucleotides of 5BSL2 (Nasheri et al. 2011).

The interacting 3′SLIV

When folding the complete CRE/VR/X-Tail region, we obtain a strong conserved long-range interaction (No. 10 in Fig. 7A,B) between the region basal to 5BSL2 (called region “9110” in Romero-López et al. 2014) and the apical loop of the 3′SLIV containing the NS5B stop codon. This interaction would compete interaction No. 15. Furthermore, we can show a possible interaction of 3′SLIV to 5BSL1 (No. 9 in Fig. 7A,B) that would compete with interaction No. 10. Thus, these interactions may perhaps dynamically change during translation of the genomic RNA.

Other long-range interactions

In addition to the interactions described above, we were able to find about 20 more possible interactions (Supplemental Fig. 7), of which the most reliable eight are also presented in Figure 7 (No. 2–No. 7 and No. 11–No. 12). Interestingly, some conserved hairpins show no direct interaction with the remaining regions of the HCV genome investigated in this study, but they might possibly be sites for protein interactions (e.g., 5BSL3.1, 3′SLI), whereas other parts of the genome seem highly reactive to the examined genomic regions (e.g., SLII, 3′SLIV).

Possible circularization of HCV

The most interesting long-range interaction of this HCV study is the genome circularization between 5′-UTR SLII and 3′-UTR DLS. We initially identified a conserved interaction between the apical loops of SLII and DLS (No. 1 in Fig. 7A,B). This initial interaction can be extended to include 62 interacting base pairs (Fig. 7C). For such a possible circularization interaction, SLI and SLII of the 5′ UTR would need to unfold and interact with the completely unfolded X-tail region.

Viral genome circularization has been described in several RNA plus-strand viruses to be important for translation, transcription and viral replication (Edgil and Harris 2006). Such a genome circularization might be also involved in 5′–3′-end communication of the HCV RNA genome to facilitate RNA translation (Song et al. 2006) and/or RNA replication (Isken et al. 2007). Importantly, since both processes are mutually exclusive, genome circularization might be one mechanism to regulate both processes in a temporal manner. Accordingly, we propose that also in case of HCV this genome circularization might be involved in RNA replication. In a similar way, also both ends of the minus-strand are predicted to form such an extended long-range circularization interaction (Fig. 7D), that could be involved in the regulation of plus-strand RNA synthesis initiation.

Genome circularizations are known to be necessary for the RNA replication in Human immunodeficiency virus (HIV) and in Poliovirus (Herold and Andino 2001; Ooms et al. 2007; Beerens and Kjems 2010). Also in the related members of the Flaviviridae family, which include West Nile virus (WNV), yellow fever virus (YFV) and dengue virus (DENV), such a genome circularization has been described to be essential for RNA replication (You et al. 2001; Corver et al. 2003; Alvarez et al. 2005; Zhang et al. 2008; Friebe et al. 2011). Moreover, in the case of DENV this RNA circularization has been observed by atomic force microscopy in the absence of viral or cellular proteins, demonstrating that such a RNA circularization can be formed solely by RNA base-pairing (Alvarez et al. 2005). Nevertheless, cellular and/or viral proteins could facilitate this process. For example, the helicase NS3 could unwind the stem–loops structures to facilitate this circularization. Several cellular RNA binding proteins have been identified by proteomics that could be involved in this process and in different steps of HCV replication. For instance, 26 host proteins were found to specifically interact with the IRES (Lu et al. 2004) and more than 70 cellular factors were identified to interact with the 3′ UTR of the plus-strand (Harris et al. 2006). Accordingly, it has been speculated that a RNA circularization could be a general replication mechanism for all plus-strand RNA viruses (Herold and Andino 2001).

CONCLUSION

Using a comprehensive bioinformatical approach, we confirmed several important previously experimentally proven RNA secondary structures and interactions. Moreover, we provide strong evidence for new and alternative, yet unknown structures and interactions based on their conservation in all seven genotypes, suggesting possible biological functions. Among the newly identified interactions, in particular the possible circularization of the plus-strand and of the minus-strand RNA may have important biological implications for the initiation of RNA synthesis on the viral RNA genome and its replication intermediate.

From 950 downloaded isolate sequences we selected the two longest per group and retrieved 86 sequences containing a nearly complete 5′-UTR sequence and 19 genomes with a complete 3′ UTR. Although in the future we need more 5′/3′ ends of HCV sequenced, we applied no constraints to the foldings to identify possible RNA secondary structures in an unbiased approach (an exception was a constraint applied to the plus-strand 5′-UTR sequence which forces formation of the pseudoknot; nucleotides marked blue in Fig. 2A).

Among the newly identified secondary structures and interactions, some may have biological relevance, such as the alternative structure of SLIIId* in the newly proposed structure of the 5′ UTR (Fig. 2A), in addition to the known structure (Supplemental Fig. 1A). Some cryo-electron microscopy analyses do not provide sufficient resolution to unambiguously identify the known SLIIId structure (Spahn et al. 2001; Boehringer et al. 2005; Siridechadilok et al. 2005). However, experimental data strongly argue for the classical known SLIIId structure to be present at least when the IRES RNA is in solution (Brown et al. 1992; Odreman-Macchioli et al. 2000; Romero-López and Berzal-Herranz 2009; Pang et al. 2011; Hajdin et al. 2013), and nuclease footprints in the apical loop of the classical version of SLIIId disappear upon binding of the small 40S ribosomal subunit (Kolupaeva et al. 2000; Lukavsky et al. 2000). Moreover, the classical version of SLIIId is strongly supported by data showing a four-way junction including SLIII, SLIIIe, and SLIIIf (Berry et al. 2011; Yamamoto et al. 2014). Thus, we can only speculate if the newly identified highly conserved alternative SLIIId* structure has a biological function. For example, refolding of the IRES to the newly proposed SLIIId* structure could favor dissociation of the IRES from the ribosome to allow genome RNA encapsidation into new viral particles. Our analysis of the 5′ UTR also clearly confirms (in both shown 5′-UTR structures) that the SLII sequence is optimized to assume the conformation experimentally validated by Zhao and Wimmer (2001).

Our study suggests that other regions of the HCV genome may also be very dynamic. The 3′ UTR of the plus-strand as well as the minus-strand 5′- and 3′-ends display different structures with each very close MFEs. Thus, these structures may exist in equilibrium in the HCV RNA and adopt different structures easily to serve different functions in the HCV life cycle. The SLII of the 3′ UTR (Fig. 4B) was reported to interact with the 5BSL3.2 in a kissing loop interaction that has a role in replication; mutations disabling this interaction impair HCV replication that can be restored by complementary mutations restoring base-pairing (Friebe et al. 2005). The alternative structure (Fig. 4A) exposes the DLS sequence that is suspected to have a role in genome dimerization and perhaps also encapsidation (Ivanyi-Nagy et al. 2006; Shetty et al. 2010; Romero-López et al. 2014). However, we find two other very highly conserved structures (Fig. 4C,D) with stabilities comparable with the two known structures. We do not know a function of these structures, but their high degree of conservation and stability suggests that they may have a biological role. Currently, we can only speculate if the different conserved structures found for the 5′ end of the minus-strand (Fig. 5) just result from the constraints shaping plus-strand sequences, or if these minus-strand structures may have genuine functions.

The alternative proposed structures of the 3′ end of the minus-strand (Fig. 6) have two different features. On one hand, they have in common that the terminal stem–loops SL-I′ and SL-IIz′ essentially form in all proposed models, underlining the importance of these sequences as shown in various experimental approaches (Astier-Gin et al. 2005; Friebe and Bartenschlager 2009; Mahias et al. 2010). On the other hand, the conservation of the sequences upstream of SL-IIz′ allows varying conserved secondary structures. Besides our consensus structure model (Fig. 6B) we found that their structure is also well compatible with the model of Smith et al. (2002). Thus, both the consensus structure prediction and the experimental results suggest that these structures are of minor importance for replication.

The variety of known and newly identified long-range interactions (LRIs) shown in Figure 7 suggests that the interactions between different regions of the HCV RNA may be very dynamic. These LRIs may be classified into four groups. First, there are interactions within the 5′ region (already known: No. 13, new: 11 and 12). Among these, interaction No. 13 is known to be involved in dynamic changes in IRES structure that augment entry of the HCV sequence containing the AUG start codon into the entry channel of the 40S ribosomal subunit (Filbin and Kieft 2011). Interaction No. 17 was reported to be involved in repressing HCV translation. This interaction can be relieved in the presence of microRNA-122 which by that can indirectly stimulate translation (Díaz-Toledano et al. 2009; Goergen and Niepmann 2012). In contrast, a possible biological role of the proposed interactions No. 11 and 12 is not yet known. Second, the interactions within in the NS5B CRE region (already known: No. 15; new: 9 and 10) might stabilize the local aggregation of secondary structures. These interactions could be involved, e.g., in exposing the apical loop of 5BSL3.2 for binding of the NS5B polymerase (Lee et al. 2004). Third, there are interactions between the 3′-terminal NS5B coding region and the 3′ UTR (already known: No. 16; new: 8). We do not know what the biological role of these interactions is, but they might transfer the NS5B polymerase trapped by 5BSL3.2 to the 3′ UTR, the actual site of minus-strand initiation.

The fourth class is long-range interactions between the two genome ends. These interactions involve one already known interaction between the IRES SLIIId and the 5BSL3.2 CRE element (No. 14) and several newly identified possible interactions between the sequences in the NS5B coding region and the IRES (No. 3, 4, 6, and 7), as well as interactions between 3′ UTR and 5′ UTR (No. 1, 2, and 5). In addition to the one known interaction, we identified in total seven new interactions between the two genome ends. Some of these interactions are mutually exclusive, like interaction 1 versus 2 plus 5, since they depend on two different conformations of the 3′ UTR. Nevertheless, these hybridizations suggest intense and perhaps dynamic interactions between 3′-terminal and 5′-terminal sequences of the viral genome which might serve different functions in the replication cycle.

Surprisingly, the interaction between the IRES SLII and the 3′ UTR DLS (No. 1) could be extended up to a 62 nt long base-pairing, resulting in extensive interaction between both genome ends (Fig. 7C). A corresponding interaction could be also shown for the minus-strand (Fig. 7D), although slightly different due to different G–U base-pairing. Such a genome circularization of the plus-strand can be hypothesized to have an important role in the initiation of minus-strand RNA synthesis and/or regulation of a switch from RNA translation to genome replication. These results strongly support earlier reports which show that the HCV genome ends interact with each other (Isken et al. 2007) to facilitate RNA replication. This circularization may be promoted by proteins, in particular by RNA helicases (Ariumi et al. 2007; Isken et al. 2007; Jangra et al. 2010) that might be involved in resolving and forming base-pairing interactions between RNA secondary structures at the genome ends. Furthermore, other studies showed that the 3′ UTR stimulates RNA translation (Song et al. 2006 and references therein), and a recent study suggested that the 3′ UTR is also associated with the ribosomal 40S subunit (Bai et al. 2013). Currently, we can only speculate what occurs first, if binding of both genome ends to the ribosome facilitates genome end hybridization, or if hybridization of the 3′X region to those regions of the 5′ UTR which are not directly involved in ribosome binding facilitates stimulation of translation by the 3′ UTR and paves the way for minus-strand initiation by genome circularization.

In this context, it is important to note that the circularization of the plus-strand ends results in a situation in which the very 3′ end of the plus-strand, which was assumed to be hidden in the stable 3′-SLI structure, now is presented to the viral replicase with a free single-stranded conformation (see Fig. 7C). In analogy to studies with the related Dengue Virus (Alvarez et al. 2005, 2008), we propose that this circularization may have an important role in HCV replication.

Taken together, our results propose several new RNA–RNA interactions in the HCV RNA, leading to the concept of highly dynamic local and long-range interactions in the viral RNA genome that may serve multiple functions in the viral life cycle.

MATERIALS AND METHODS

Alignment generation

We downloaded 950 HCV genomes from NCBI (17.01.2014, Nucleotides, Search term: (Hepatitis C virus[porgn:__txid11103] AND complete genome[Title])) and the HCV database (v. 2008 [Kuiken et al. 2005] http://hcv.lanl.gov/content/sequence/NEWALIGN/align.html), see Supplemental Material (“Initial data”). Sequences which occur in both data sets were only included once. Nucleotides “Y,” “S,” “W,” “R” have been replaced by corresponding valid nucleotides. An alignment with MAFFT –auto (v.6.8 [Katoh et al. 2002]) was performed and a phylogenetic tree was built with Geneious (v.6.1 [Kearse et al. 2012]) Neighbor-Joining method Tamura-Nei (Tamura and Nei 1993). Based on this tree and on the annotations from NCBI and the HCV database the data set was reduced to two of the longest genomes from each subtype if available (see Supplemental Material, “Selected data”). With the reduced data set of 106 sequences from 65 subtypes, we performed a second MAFFT –maxiterate 1000 –localpair alignment. This alignment was used to define the ranges of 5′ UTR, Rep-Core, CRE, 3′ UTR and protein coding regions and is used in the following (see Supplemental Material, “Selected data”).

Alignment based secondary structures

For 5′ UTR, Rep-Core, CRE, and 3′ UTR we calculated a secondary structure based alignment with LocARNA (v.1.7.2 [Will et al. 2012]). We compared the quality of the secondary structure alignment from LocARNA, DAFS (Sato et al. 2012), and Lara (Bauer et al. 2007) based on the CRE-region. Although all programs present nearly the same results, we decided to use LocARNA, because it has a lower running time and refining of the CRE-region alignment performed best (see Supplemental Fig. 8).

We define the combined 5′-UTR-IRES region of the plus-strand (referred to as “5′ UTR” hereafter) to range from MAFFT alignment position 0–358 (based on 86 isolates, see Supplemental Material, “Selected data”). The latter position refers to 12 nt downstream from the start codon of the polyprotein coding region to allow the formation of SLIV (Honda et al. 1999a; Niepmann 2013). The Core region part known to be needed for replication (Rep-Core) is defined from position 330–520 of the MAFFT alignment (based on 106 isolates). For interaction of 5′ UTR and Rep-Core we used positions 1–520 (based on 77 isolates). The 3′ end of minus-strand is defined from 0–366 to form known possible hairpins (Smith et al. 2002; Dutkiewicz et al. 2008). We used 86 sequences for the above analyses since 20 sequences lack most of the 5′ UTR. The CRE ranges from MAFFT alignment position 9399–9775, and its structure is calculated together with that of the variable region of the 3′ UTR (9399–9820) followed by a poly-U region and the remaining X-tail, from 10,000–10,101. For the latter calculation we included all available sequences covering the complete region (96 sequences for CRE/VR and 19 sequences for X-tail). We performed alignments for the whole region (9399–10,101) as well as for CRE/VR and X-tail separately. Positions for the 5′ end of the minus-strand refer to the plus-strand. The consensus structures are calculated with RNAalifold -r -p –color –MEA (v.2.0.7 [Lorenz et al. 2011]) and -C when constraints were used, as described in the main text. Each minimum free energy we use herein is the sum of the average minimum free energy of all sequences in the given alignment plus the covariance term as used in RNAalifold.

Long-range interactions

For each of the calculated secondary structures (Fig. 2A, Supplemental Figs. 1, 2B, 4A,B) all alignment regions containing at least three consecutive nonpairing nucleotides, added with two flanking nucleotides, were concatenated to a new alignment, separated with “NNN.” For each of the artificial alignments 5′ UTR/CRE, 5′ UTR/X-Tail A and 5′-UTR/X-Tail B a base pair probability matrix (Supplemental Fig. 7) was calculated. Based on this matrix we picked the most probable interactions, recalculated the selected interactions with LocARNA and filtered the results manually from dotplots of, e.g., the 5′ UTR (Supplemental Fig. 1B) and CRE (Supplemental Fig. 3B,D). We used the manual approach for LRI identification, as alignment based secondary predictions tools alone—such as RNAalifold, RNAaliduplex (Lorenz et al. 2011), LocARNA, or PETcofold (Seemann et al. 2011)—are not able to predict all of the previously known LRIs from the full defined regions (5′ UTR, CRE, 3′ UTR).

The z-score, based on RNAalifold MFE was calculated as

$$Z=\frac{X-\mu }{\sigma }$$

whereas X is the MFE of the corresponding interaction, μ is the mean MFE and σ the corresponding standard deviation of the 1000 randomly dinucleotide shuffled alignments (generated by multiperm) (Anandam et al. 2009).

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article at http://www.rna.uni-jena.de/supplements/hcv/.