Eighteen sRNA libraries representing three biological replicates of pooled DENV2-exposed and un-exposed control mosquitoes were mined to identify significantly modulated miRNAs (Hess et al., 2011). sRNAs were aligned to miRBase hairpin release 17 (Griffiths-Jones, 2006; Kozomara & Griffiths-Jones, 2011). Mapped reads in the miRNA size range (18–23 nt) showed a marked predominance of forward strand reads, whereas reads <18 nt showed a more balanced representation of both forward and reverse strands (data not shown). This evidence supports the current understanding of miRNA biogenesis mechanisms, wherein the guide strand is retained and the complementary strand is degraded. In miRNA biogenesis, this process occurs via a two-step RISC-loading process, wherein the partial complementarity of the double-stranded precursor is sensed by the RISC, one strand is nicked by Argonaute-2, and the guide strand is loaded into a second RISC, with concomitant loss of the passenger strand and subsequent cleavage of target mRNAs (Preall & Sontheimer, 2005; O'Toole et al., 2006; Diederichs & Haber, 2007), In contrast, siRNA biogenesis relies on a single cleavage-dependent RISC loading event of dsRNA precursors that presumably results in either strand serving as guide strand.
DENV2-exposed Ae. aegypti sRNA libraries showed modulation of miRNA profiles compared with un-exposed controls at 2, 4 and 9 dpe. Age-matched DENV2-fed and unexposed controls were analysed for each timepoint. Only those miRNAs homologous to previously reported mature −5p and −3p miRNAs, previously termed miRNAs and *miRNAs, respectively, were analysed further (MirBase.org)(Griffiths-Jones, 2006). Conserved miRNAs from 31 miRNA genes showed significant modulation (edgeR, P < 0.05) (Fig. 1, Table 1). Of these, 26 unique miRNAs have orthologues in insects, and 24 have been reported for Ae. aegypti (Mirbase.org).
Figure 1. Significantly modulated miRNAs and indicated LogFC at each timepoint, relative to unexposed controls (P < 0.05).
Download figure to PowerPoint
Table 1. MicroRNA LogFC. Table lists miRNA sequence, length and edgeR output, log fold-change, and P value
|miRNA||Sequence||nt #||Control*||DENV2*||Total count||logFC||P|
Over the course of infection, out of a total of 35 modulated miRNAs, the number of unique modulated miRNAs increased from 5 and 3 at 2 dpe and 4 dpe, respectively, to 23 by 9 dpe (Fig. 1, Table 1). At all timepoints, DENV2 viral small RNAs (viRNAs) were detectable in DENV-exposed pools (Hess et al., 2011). Importantly, the libraries were constructed from whole mosquitoes that had received a virus-laden bloodmeal, and 50% of the mosquitoes sampled did not retain detectable virus by plaque titration at 9 dpe (Hess et al., 2011). The retention of significantly modulated miRNAs at 9 dpe is remarkable, considering that 50% of any given pool would have no detectable virus by plaque titration. In our previous work, we found that significant differences in overall host sRNA profiles were largely abrogated by 9 dpe, probably because of the mixed infection status. Persistent changes to the miRNA pathway have implications for lifelong alteration of mosquito metabolism. Hypotheses with similar implications have been tested previously to assess behavioural changes in DENV-infected aedines, but this is the first evidence of a regulatory response at the global level (Platt et al., 1997; Sim et al., 2012).
Although 3′ untranslated regions (UTRs) were originally identified as being the key sites of miRNA-target interactions, recent target prediction studies set a precedent for probing full transcriptomes rather than merely 3′ UTRs (Hafner et al., 2010; Fang & Rajewsky, 2011); this approach seems especially suitable for an organism whose genome is not extensively annotated, such as Ae. aegypti. Putative miRNA targets were identified from transcriptome release 1.3 (Vectorbase.org) using the Miranda, PITA and TargetScan prediction methods (Enright et al., 2003; Lewis et al., 2005; Kertesz et al., 2007). miRNA targets are commonly identified by comparing the base complementarity of a portion of the miRNA to that of a given target sequence. In animals, miRNAs show partial rather than complete complementarity to targets (Brennecke et al., 2005). This partial complementarity is most important at the 5′ end of the miRNA and typically spans nt positions 2–8, thus defining the miRNA seed regions that are used in target prediction algorithms. The Targetscan prediction algorithm relies on base complementarity of seed region positions 2–8, whereas PITA also considers the free-energy of association between the miRNA and target along the length of the seed (Enright et al., 2003; Lewis et al., 2005; Kertesz et al., 2007). Miranda, one of the earliest prediction methods, considers free energy of association along the entire length of the miRNA-target region (Enright et al., 2003; Lewis et al., 2005; Kertesz et al., 2007). Putative targets were cross-validated with genes previously reported to be associated with DENV2 infection in Ae. aegypti (Guo et al., 2010; Behura et al., 2011; Colpitts et al., 2011; Bonizzoni et al., 2012; Sim et al., 2012). Only those targets contained within the intersection of the three target prediction methods and cross-validated by RNA-Seq reports were analysed further.
The 35 modulated miRNAs share 4076 in silico-validated transcriptome targets among all three prediction methods (Fig. 2A). PITA and TargetScan identified many more sites per target than Miranda. The Miranda prediction algorithm with stringent cut-off criteria produced the most inclusive target set of all approaches used (see Experimental procedures).
Figure 2. Context-specific canonical miRNAs common to the target prediction methods PITA, Miranda, and TargetScan. (A) Context-specific targets for each prediction method. (B) Context-specific targets identified in 3′ UTR and coding sequences (CDS) using the Miranda package.
Download figure to PowerPoint
Target sites in both the 3′UTR and the coding sequence (CDS) have been suggested to be synergistic for miRNA regulation (Fang & Rajewsky, 2011). Upon interrogation of 3′ UTRs alone, we found 464 unique gene targets (Fig. 2B). Of these, 365 targets contain miRNA-binding sites in both the CDS and the 3′ UTR, further strengthening their predicted significance as targets.
To characterize miRNA target biological attributes over the course of infection, 4076 targets common to the three target prediction methods were graphed according to miRNA and functional group (Table S2). Targets were classified by processes known to be associated with viral replication (Le Breton et al., 2011; Perera et al., 2012). The functional groups chosen account for ≥85% of all targets (data not shown). The processes involving transport, transcription/translation and cytoskeletal/structural components are required for successful DENV2 replication and dissemination. Transport, signal transduction, cytoskeletal/structural and metabolism make up the four most abundant target functional groups (Table S2). It is important to acknowledge the limitations inherent to studies of miRNA expression analysis. Any post-transcriptional or post-translational modifications of predicted targets would not be detected.
Implications for viral replication
The DENV2 miRNAi response could be a manifestation of several possible host responses. The following hypotheses are founded upon a variety of vector host response studies (Xi et al., 2008; Behura et al., 2011; Colpitts et al., 2011; Bonizzoni et al., 2012; Chauhan et al., 2012; Ocampo et al., 2013). For example, modulation of miRNA activity could regulate cell autonomous responses, such as those that occur within infected cells, or be indicative of signal transduction events that influence processes in distal tissues. Alternatively, modulation of miRNA levels could be evidence of viral exploitation of cellular processes. Some proteins, such as Loquacious, interact with proteins involved in both the miRNAi and siRNAi pathways (Fukunaga et al., 2012; Martinez & Gregory, 2013).
Importantly, because of the mixed infection status by 9 dpe, another possibility is that the significantly changed miRNAs at this timepoint represent responses to repair mechanisms in uninfected mosquitoes. To explore the possibilities, we compared our results with other host response studies. We used data from two recent reports describing the comparison of gene expression responses in DENV2-resistant and -susceptible Ae. aegypti strains (Chauhan et al., 2012; Ocampo et al., 2013). For example, Chauhan et al. (2012) reported that expressed genes representing glycolysis, gluconeogenesis, and the Wingless (Wnt) signalling functions are enriched in resistant strains, while transcripts with functions related to ER protein processing, nucleotide excision repair, the pentose phosphate pathway and the proteasome, are enriched in susceptible mosquitoes. In another study, Ocampo et al. (2013) described increased caspase expression in DENV2-infected tissues of resistant mosquitoes. To determine whether the predicted targets in the present work are supportive of resistant or susceptible phenotypes described by these two studies, we looked at the number of targets listed in Table S1 in each of these categories, There are 43 targets associated with ER protein processing and 84 associated with proteasome function. There are 11 targets associated with glycolysis/gluconeogenesis and two for Wnt signalling genes. The caspases Caspase-16, Dronc, and Dredd were not present on the target list. Of the 35 miRNAs modulated, four are enriched and the remaining 31 are depleted. Upon miRNA depletion, given that the predicted target mRNAs would be expected to be retained rather than degraded, and therefore expected to subsequently produce more protein, we speculate that our results are consistent with a susceptible rather than a resistant phenotype.
Modulation of miRNAi at early timepoints in DENV2 infection suggests that upstream regulatory processes are altered early during infection and raises the important question of whether the miRNAi host response is a defence response or a result of viral exploitation. Alternative explanations for changes to miRNA levels could include the removal or repair of infected cells containing these noncoding RNAs. To explore the possible implications of modulation to miRNA levels, targets orthologous to human flavivirus host response proteins were identified among the predicted targets. A subset of the targets is orthologous to a group of human proteins that physically interact with flavivirus non-structural proteins NS3 and NS5 (Le Breton et al., 2011). Human chromatin structural modification proteins also directly interact with flavivirus NS3 and S5 (Le Breton et al., 2011). In the present study, we identified 141 putative targets in the chromatin structure and dynamics functional category. DENV2 NS3 and NS5 are essential for successful flavivirus replication in endoplasmic reticulum-associated replication vesicles (Luo et al., 2008; Welsch et al., 2009). Importantly, 50 predicted miRNA targets are identical to a subset of human proteins that physically interact with flavivirus NS3 and NS5 (Table S1) (Le Breton et al., 2011). Fifteen of these are targets of miRNAs modulated at 2 dpe. Moreover, we found 33 unique targets with innate immunity or defence descriptors (Table S1). As discussed above, because 31 of the 35 miRNAs are depleted rather than enriched, any such mRNA targets would be expected to be retained rather than depleted. These results suggest that suppression of the innate immune response in vectors, should it occur, may be mediated through other response mechanisms. Clearly, these preliminary indications require closer investigation to clarify possible gene-for-gene interactions between viral proteins and the host response.
Mitochondrial function is essential for cellular energy production and often associated with DENV2 infection (El-Bacha et al., 2007; Sessions et al., 2009; Perera et al., 2012). We found 235 mitochondrial protein targets within the common target set (Tables S1 and S2). The most likely explanation for these target effects is that mitochondrial function increases to accommodate increased demand for cellular energy needs. A major component of dipteran mitochondrial membranes, phosphatidylethanolamine, is significantly altered during DENV2 infection of mosquito cells (Chan, 1970; Perera et al., 2012). It is tempting to speculate that the predicted mosquito host response produces changes to mitochondrial membrane composition, as well, although this hypothesis remains to be validated experimentally.
The presence of modulated miRNAs in DENV2-exposed mosquitoes suggests that miRNA pathway activity is altered during infection, although the causal factors remain unknown. The characteristics of that involvement must rely on characterization of miRNA activity on a gene-by-gene level in future studies. The importance of this is illustrated in a recent work, wherein it was demonstrated that a bloodmeal-induced miRNA, miR-375, stimulates enrichment of one mRNA and depletion of another (Hussain et al., 2012). In our dataset, alteration of mIR-375 was depleted in DENV2-exposed mosquitoes at 9 dpe, but the change was not significant (−1.6 LogFC, P = 0.08, edgeR).