Before designing a method of exploiting miRNA machinery to suppress DENV replication, we sought to confirm DENV infection does not block miRNA biogenesis or function. To assess effective knockdown of a miRNA target, human fibroblasts expressing a green fluorescent protein (GFP) targeted by miR-142 (pEGFP-142t) were transfected with either vector or a plasmid expressing miR-142 (p142), a miRNA normally absent in these cells. In uninfected cells, expression of miR-142 resulted in a dramatic reduction of visible GFP expression (), equating to a 75% loss of fluorescence (). Furthermore, despite a multiplicity of infection (MOI) that would ensure high levels of virus replication, verified by quantitative RT-PCR (qRT-PCR) (), the presence of DENV-2 did not alter the efficiency of miR-142-mediated PTS. In accordance with the lack of DENV-2-mediated disruption of miR-142-targeting, northern blots for the expression of other exogenous (miR-124) and endogenous (miR-93) miRNAs also demonstrated no discernable differences during virus replication (Figure S1
). Taken together, these results suggest DENV-2 does not interfere with miRNA-mediated PTS or miRNA biogenesis, and that cellular miRNA machinery may be exploited for DENV targeting.
DENV-2 does not block miRNA function.
During natural infection, DENV displays tropism for cells of hematopoietic lineage, specifically, monocytes, macrophages and DCs 
. As miR-142 is one of the most abundant hematopoietic-specific miRNAs 
, it served as an ideal candidate for mediating cell-specific attenuation. To generate a DENV-2 strain that demonstrated miR-142-susceptibility, we incorporated four target sites into the 3′ UTR of the virus (). Given the highly structured secondary conformations important for both transcription and translation in the 3′ UTR 
, we incorporated the target sites downstream of the NS5 coding frame, in the variable region believed to be more amenable to nt insertions 
). Following construction of both a miR-142-targeted (142t) virus and a parental control (ctrl) virus, encoding reverse target sites, we next sought to determine their replicative properties in cells lacking miR-142. To this end, we performed multi-cycle growth curves in mosquito Aedes Albopictus
larvae cells (C6/36) and baby hamster kidney (BHK) cells, neither of which express miR-142 (). By qRT-PCR and western blot, both ctrl and 142t strains grew to similar levels in C6/36 cells () reaching equal growth at 84 hrs post infection (hpi), but demonstrating an approximate one-log reduction as compared to wild type, unmodified virus (wt). This data was further validated by plaque assay, demonstrating peak titers of ~1×103
PFU/mL for both ctrl and 142t viruses as compared to ~1×104
PFU/mL for wt virus (Figure S3
). Furthermore, experiments in BHK cells, another cell devoid of miR-142, demonstrated peak virus loads at 48 hpi, a time at which cytopathic effects reduced overall titers and NS5 expression (). Consistent with the data from C6/36 cells, wt virus exceeded both ctrl and 142t strains, suggesting that the 157 nt insertion lowered the overall fitness of the virus, independent of sequence.
Generation and characterization of miR-142-targeted DENV-2.
To determine whether DENV-2 transcripts could be targeted by the miRNA machinery, we first attempted to see whether the presence of miR-142 could prevent virus production from the in vitro
transcription (IVT) of the 142t cDNA clone. BHKs, expressing vector or p142, were electroporated with IVT products from wt, ctrl, or 142t DENV-2 cDNA clones. Twenty-four hrs post transfection (hpt), miR-142 production was evident by northern blot, albeit lower than observed in bone-marrow derived primary macrophages (BMMs) (). Given the production of miR-142, we evaluated NS5 protein levels from the IVT product to determine the degree of transcript targeting (). Consistent with previous work demonstrating miRNA-mediated virus attenuation 
, NS5 expression derived only from 142t IVT product was abrogated in a miRNA-specific manner (). This result would suggest that virus synthesis from the 142t IVT product was blocked in the presence of miR-142.
In vitro knockdown of miR-142-targeted DENV-2 during exogenous and endogenous miR-142 expression.
Next, we sought to determine whether miR-142 could attenuate 142t in the context of infection. To this end, fibroblasts expressing vector or p142 were infected with wt, ctrl, or 142t strains, and assessed for virus products. As evident by NS5 expression, miR-142 abolished replication of the 142t strain, while exuding no impact on wt or ctrl strains (). It is noteworthy that residual levels of NS5 for the 142t strain likely reflect replication in cells not successfully transfected with p142, a constraint that would not be in place during an endogenous infection of a hematopoietic cell. Taken together, these data clearly demonstrate that insertion of miRNA target sites into the 3′UTR of DENV-2 renders the virus susceptible to miRNA expression.
In an attempt to ascertain the mechanism underlying miR-142-mediated attenuation, we modified the miR-142 target sites to render the DENV genome resistant to canonical miRNA binding and Ago2 cleavage. To this end, the nucleotides complementary to positions 3 and 10 of miR-142 were mismatched in the virus targets to destroy the seed sequences and cleavage sites, respectively 
. To assess whether this mutated 142t strain, herein referred to as 142tm, was targeted by miR-142, we compared NS5 synthesis in the presence and absence of the target miRNA (). Surprisingly, virus production from the 142tm strain was still abrogated in a miR-142-specific manner, despite harboring two critical mismatches in the miRNA targeting sites. As loss of NS5 in the 142tm strain maintained miR-142-specificity, these results would suggest that RISC was still engaging on the mutated targets in a non-canonical manner, perhaps as a result of the remaining extensive complementarity. To determine the underlying mechanism responsible for attenuation of the 142tm strain, we examined the levels of genomic viral RNA to discern between RNA degradation and translational inhibition. Quantitative RT-PCR analysis demonstrated that, at the level of RNA, the amount of miR-142-mediated repression of the 142tm virus was significantly (p
0.0178) lower than that of 142t, suggesting the mode of attenuation may no longer be RNA cleavage, but may involve a form of translational repression (). This translational repression could be the result of extensive 3′ miRNA complementarity acting in a canonical manner or could be the result of miR-142/RISC sterically blocking the association of the 5′ and 3′ ends, preventing cyclization and amplification of the virus genome. While future studies will be required to ascertain the exact mechanism of 142tm attenuation, these data strongly demonstrate that inhibition of virus replication occurs in a miR-142-dependent manner. As the 142tm virus showed decreased silencing activity, subsequent characterization focused only on the comparison between ctrl and 142t strains.
As cells of hematopoietic lineage are the natural sites of DENV infection and express high levels of miR-142 () 
, we aimed to determine whether we could observe differential cell-specific attenuation between the ctrl and 142t strains. To this end, we compared infections of the recombinant DENV-2 strains in two hematopoietic lineages, B cells and BMMs, as well as a non-hematopoietic fibroblast lineage, HEK293s (). Consistent with exogenous targeting by miR-142, infection of hematopoietic cell types demonstrated selective attenuation of the 142t strain, while infection of the non-hematopoietic cells demonstrated no change in NS5 levels between ctrl- and 142t-virus infected samples (). Taken together, these data demonstrate that the incorporation of miR-142 target sites into the 3′UTR of DENV-2 confers endogenous attenuation of the virus in a cell-specific manner.
As human DENV isolates currently lack an adequate animal model that recapitulates human pathogenesis during infection, we chose Ifnar1−/−/Il28r−/−
mice to investigate our recombinant viruses in vivo
. These mice lack the ability to respond to type I and III interferon and are, thus, more susceptible to virus infection 
. To ensure that the in vitro
and ex vivo
data reflected in vivo
attenuation, we first infected mice with ctrl and 142t viruses and isolated CD11b+
macrophages and DCs, respectively (). qRT-PCR analysis of RNA derived from these cells demonstrated minimal detection of NS5 transcripts derived from the 142t strain and a significant decrease compared to ctrl virus. These data support our in vitro
data, where 142t virus replication is excluded in hematopoietic cell types.
In vivo knockdown of miR-142-targeted DENV-2.
To examine in vivo
replication of the 142t virus in non-hematopoietic compartments, we sorted splenocytes from infected mice for CD45-expressing cells, and assessed viral gene expression in each population (Figure S4A
). The nature of this experiment however is complicated by the fact that, should hematopoietic cells be required for virus dissemination, we would anticipate lower virus titers in non-hematopoietic fractions as well. As anticipated, 142t virus growth was attenuated in the CD45+
hematopoietic fraction. In addition, the CD45−
, non-hematopoietic fraction demonstrated a decrease in 142t virus growth as compared to ctrl, supporting that dissemination of the virus is dependent on its ability to replicate in hematopoietic cell types. To account for this lack of dissemination, we compared the relative level of virus growth between hematopoietic and non-hematopoietic populations, and determined that the 142t virus displayed enhanced replication in CD45−
, non-hematopoietic cells (). Taken together, these data suggest that the 142t virus is not attenuated in non-hematopoietic cells, but that titers are impaired as a result of decreased hematopoietic replication.
Following verification of in vivo
targeting, we further assessed dissemination of the ctrl and 142t strains in Ifnar1−/−/Il28r−/−
mice. To this end, three different routes of inoculation were administered to assess virus replication in both liver and spleen. Different cohorts of mice were administered virus by either intraperitoneal (IP), intraveneous (IV), or subcutaneous (SC) injection (, Figure S4B
). For IP administration, qRT-PCR analysis of the spleen revealed an ~1.5-log reduction in viral transcript levels of the 142t strain as compared to those given ctrl virus (). This attenuation was further enhanced by IV and SC injection, routes that better simulate natural inoculation. Interestingly, decreased amounts of NS5 transcript was also evident in the liver (Figure S4B
), where miR-142 expression is reduced compared to the spleen 
. Taken together, these in vivo
results suggest that virus levels in the liver reside predominantly in resident macrophages and DCs, and that perhaps these cells are needed to maintain basal levels of virus replication in non-hematopoietic cells. To further validate the reduction of viral growth in vivo
, viral titers from infected mice corroborated qRT-PCR data by demonstrating an approximate three-log reduction in titers from spleen and liver (, Figure S4C
). Viral titers were undetectable in heart, lung, and brain from these mice, demonstrating that the spleen and liver were the primary sites of virus replication in the context of this animal model.
As low levels of 142t virus were evident in both the spleen and the liver, we next sought to determine whether this reflected a non-hematopoietic reservoir for the virus or whether this was evidence of virus escape. In an effort to characterize the genotype of the recombinant DENV viruses over the course of infection, we amplified the 3′UTR of the ctrl and 142t viruses derived from in vitro
and in vivo
infections (Figure S5
and ). To this end, non-hematopoietic fibroblasts expressing varying levels of miR-142 were infected for 3′UTR sequence analysis. In vitro
, 142t-specific transcripts were only detectable in conditions where miR-142 expression was decreased, a result that could reflect lower transfection efficiency or insufficient miRNA levels. To assess the 3′UTR sequence in vivo
, splenocytes from infected mice were analyzed, demonstrating abundant transcript levels of the ctrl virus and significantly lower levels of the 142t virus (). Furthermore, the only product derived from the 142t virus migrated faster during gel electrophoresis, suggesting it had been truncated. Sequence analyses of transcripts from these two infections demonstrated an unbiased mutation frequency in vitro
that was comparable between the two viral cohorts, suggesting the mutations were a reflection of low polymerase fidelity, rather than a result of evolutionary constraint. In contrast, sequence analyses of the transcripts found in vivo
demonstrated a complete absence of 142t virus. Rather, virus species remaining displayed loss of all four miRNA target sites either by complete excision or by replacement with a small host RNA fragment (). The sequence analysis performed here explains why low levels of 142t virus was detected by qRT-PCR in , as this assay measured levels of NS5, which encompasses any escape mutants present during infection. Our data in reveals that the percentage of 142t virus represented is, in fact, much lower. Furthermore, this is unlikely attributable to the presence of quasi-species in our virus stocks, as these mutations were not observed in vitro
In vivo escape mutants of miR-142-targeted DENV-2.
Previous efforts aimed at defining the tropism of DENV infection have focused on the detection of virus components in various tissues of the host. Unfortunately, distinguishing between active DENV replication within a cell versus the presence of virus due to cellular engulfment has been difficult. Here we take an innovative approach to addressing this question with the generation of a virus that is selectively attenuated in a cell-type specific manner. Through the exploitation of hematopoietic-specific miR-142, we were able to exclude the replication of DENV at its major sites of replication in vivo, and demonstrate that these cell populations are critical for dissemination of the virus to other tissues. Our initial in vitro work clearly demonstrated the specificity of the system and provided a foundation for the subsequent studies performed in mice. The technology enabled a precise mechanism to eliminate primary target cells of DENV replication and study the effects in the context of a dynamic in vivo infection.
An additional noteworthy finding that resulted from this study centers on the mechanism of miRNA-mediated attenuation. While attenuation of the 142t strain is presumably the result of target RNA cleavage by Ago2, the repression of 142tm is an enigma. The data supports that 142tm attenuation occurs less at the level of RNA than it does at the level of protein, making the underlying molecular biology responsible for this repression unclear. While the extensive complementarity may be responsible for some level of post-transcriptional silencing, the complete loss of the seed sequence makes this unlikely. An alternative hypothesis would be that the level of complementarity is sufficient to recruit miR-142-containing RISC, but not sufficient to confer post- transcriptional silencing. In this model, RISC-binding may do nothing more but provide a steric hinderance to the virus. To undergo replication, DENV, as well as other flavivirus genomes, cyclize through complementarity at conserved elements in the 5′ and 3′ ends of the viral RNA 
. As such, RISC function may be at the level of physically disrupting proper folding of the virus genome. This latter model is supported by the isolation of escape mutants. Rather than identifying single point mutations in the seed or center sites of each target, the only escape mutants isolated were viruses in which the complete targeting cassette was excised. Taken together, this suggests that miRNA-mediated attenuation of viruses may be occurring both through transcriptional silencing, and through the steric interference of RNA folding. Future work to address this possibility is ongoing.
In closing, this technology has become a unique and effective tool to study cell populations involved in the DENV life cycle, and can be easily applied to other viruses to examine the relevance of cellular subsets involved in virus replication. In this regard, it is interesting that restricting hematopoietic replication of DENV-2 prevents overall dissemination of the virus suggesting that non-hematopoietic primary cells may not be productively infected in vivo or that macrophages and DCs are essential for viral spread. Altogether, the successful attenuation of DENV in a cell-specific manner suggests this technology may be exploited for studying the relative contributions of cell subsets to virus pathogenesis in vivo.