Design of antiviral AGO2-dependent short hairpin RNAs

Abstract

The increasing emergence and re-emergence of RNA virus outbreaks underlines the urgent need to develop effective antivirals. RNA interference (RNAi) is a sequence-specific gene silencing mechanism that is triggered by small interfering RNAs (siRNAs) or short hairpin RNAs (shRNAs), which exhibits significant promise for antiviral therapy. AGO2-dependent shRNA (agshRNA) generates a single-stranded guide RNA and presents significant advantages over traditional siRNA and shRNA. In this study, we applied a logistic regression algorithm to a previously published chemically siRNA efficacy dataset and built a machine learning-based model with high predictive power. Using this model, we designed siRNA sequences targeting diverse RNA viruses, including human enterovirus A71 (EV71), Zika virus (ZIKV), dengue virus 2 (DENV2), mouse hepatitis virus (MHV) and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), and transformed them into agshRNAs. We validated the performance of our agshRNA design by evaluating antiviral efficacies of agshRNAs in cells infected with different viruses. Using the agshRNA targeting EV71 as an example, we showed that the anti-EV71 effect of agshRNA was more potent compared with the corresponding siRNA and shRNA. Moreover, the antiviral effect of agshRNA is dependent on AGO2-processed guide RNA, which can load into the RNA-induced silencing complex (RISC). We also confirmed the antiviral effect of agshRNA in vivo. Together, this work develops a novel antiviral strategy that combines machine learning-based algorithm with agshRNA design to custom design antiviral agshRNAs with high efficiency.

Keywords: AgshRNA; Antiviral strategy; Machine learning model.

Fig. 1

Prediction of high efficiency antiviral siRNAs by using a machine-learning model. A Schematic for training the model to develop an efficacy predication algorithm. Set of siRNAs with known efficiency are collected from Huesken et al.'s work (Huesken et al., 2005). The siRNAs with silencing efficiency higher than 0.7 are considered as high efficiency (1) while others low efficiency (0). Dataset are randomly split into training set (1999 siRNAs) and testing set (432 siRNAs). The features of siRNAs in the dataset are extracted to build a feature matrix (Supplementary Table S1). Logistic regression was used to train the model and 5-fold cross-validation was performed to adjust model's parameter. Final model was tested using testing set. B Model performance evaluation using ROC curve (upper) and confusion matrix (lower). C Strategy to predict virus-targeting siRNAs with high efficiency. EV71 was used here as an example. Viral genome was scanned to generate all possible siRNAs, which were then aligned to host genome to remove potential off-target siRNAs. The remaining non-off-target siRNAs were evaluated using our model and the score of each sequence was ranked (Supplementary Table S2).

Fig. 2

The antiviral strategy that combines machine learning-based siRNA prediction with agshRNA design. A Schematic for transformation of siRNA sequences into agshRNA. B Design of agshEV71-h1, shEV71-h1, siEV71-h1 based on the same siRNA sequences with the highest score. Guide strands were shown in red and passenger strands in blue. C–E RD cells were transfected with siEV71-h1, shEV71-h1 and agshEV71-h1 with different concentrations indicated for 2 ​h, and then infected with EV71 (MOI ​= ​0.1). At 24 hpi, total RNAs were extracted and subjected to qRT-PCR to detect the levels of EV71 genomic RNA. The IC₅₀ value of each RNA molecule was determined. Data were means ​± ​SEM from triplicate samples. F–H RD, A549 and L2 cells were transfected with the corresponding agshRNAs with top 3 highest, median 3 and lowest 3 scored siRNA sequences (agshEV71-h1, -h2 and -h3, agshEV71-m1, -m2 and -m3, agshEV71-l1, -l2 and -l3, 10 ​nmol/L for each) for 2 ​h, and then infected with EV71, ZIKV and MHV (MOI ​= ​0.1 for each), respectively. AgshNC was used as the control in each group. At 24 hpi, total RNAs were extracted and subjected to qRT-PCR to detect the levels of viral genomic RNA, and the level of viral RNAs in cells with agshNC treatment was defined as 100%. Data were means ​± ​SEM from triplicate samples. I–J A549 and Vero E6 cell were transfected with the corresponding agshRNAs with top 3 highest scored siRNA sequences (agshDENV2-h1, -h2 and -h3, agshSARS-CoV-2-h1, -h2 and -h3, 10 ​nmol/L for each) for 2 ​h, and then infected with DENV2 and SARS-CoV-2 (MOI ​= ​0.1 for each), respectively. AgshNC was used as the control in each group. At 24 hpi, total RNAs were extracted and subjected to qRT-PCR to detect the levels of viral genomic RNA, and the level of viral RNAs in cells with agshNC treatment was defined as 100%. Data were means ​± ​SEM from triplicate samples. ∗P ​< ​0.1; ∗∗P ​< ​0.01; ∗∗∗P ​< ​0.001; n.s., no significant.

Fig. 3

Antiviral effect of agshRNA is dependent on AGO2-processed guide RNA. A Normal, NoDice and AGO2-KO 293T cell were transfected with agshEV71-h1, agshEV71-h2 and agshNC (10 ​nmol/L for each), respectively, and then infected with EV71 at an MOI of 0.1. At 24 hpi, total RNAs were extracted and subjected to qRT-PCR to detect the levels of EV71 genomic RNA, and the level of viral RNAs in infected cells with agshNC treatment was defined as 100%. Data were means ​± ​SEM from triplicate samples.∗∗∗P ​< ​0.001; n.s., no significant. B Normal, NoDice and AGO2-KO 293T cell were transfected with agshEV71-1 (10 ​nmol/L). At 24 ​h post transfection, total RNAs were extracted and northern blot was performed to detect the 5′arm (5p) and the 3′arm (3p) of agshEV71-h1. The synthetic 21- and 25-nt RNAs were used as size markers. C 293T cell were transfected with increasing amounts of the plasmid encoding Flag-tagged PARN (0.18, 0.37, 0.75, 1.5 ​μg) together with agshEV71-h1 (10 ​nmol/L) for 24 ​h. Northern and western blottings were performed to detect the indicated RNAs and proteins, respectively.

Fig. 4

The guide RNAs derived from agshRNAs load into the RISC. Human 293T cells were transfected with agshEV71-h1, agshMHV-h1, agshZIKV-h1 respectively (10 ​nmol/L for each). At 24 ​h post-transfection, cell lysates were subjected to RNA immunoprecipitation with an anti-AGO antibody or mouse IgG. Input and AGO-bound RNAs were prepared and subjected to deep sequencing. A The size distribution of agshRNAs-derived small RNAs in indicated input samples. B Processing of agshRNAs determined by deep-sequencing. Guide strands are shown in red. Dash lines show the 3′ end of guide strands, and the relative abundance is indicated next to the lines. C Relative abundance of endogenous small non-coding RNAs in input and IP samples. The relative abundances of small RNAs are shown in plat plot. This sequencing experiment was repeated twice independently, and the representative data was graphed (Supplementary Table S5, “agshEV71-h1#1-Input” and “agshEV71-h1#1-IP”, “agshMHV-h1#1-Input” and “agshMHV-h1#1-IP”, “agshZIKVh1#1-Input” and “agshZIKV-h1#1-IP”). D Northern and Western blottings were performed to detect input and precipitated small RNAs or proteins.

Fig. 5

AgshMHV protects mice from lethal MHV infection. A Four-week-old C57BL/6 female mice were tail vein injected with LNP-agshMHV-h1 (n ​= ​6) or LNP-agshNC (n ​= ​6) at 2 ​mg/kg of body weight. After 24 ​h, mice were challenged with 1 ​× ​10⁶ ​PFU of MHV-A59 via i.p. injection. Mouse survival was observed and recorded daily until 8 dpi. Statistical significance was evaluated via using Log-rank test. B Body weight changes of the different groups of mice in (A). C MHV-infected C57BL/6 mice treated with LNP-agshMHV-h1 or LNP-agshNC were detected via qRT-PCR. The level of MHV RNAs in LNP-agshNC-treated mice was defined as 100%. Two-way ANOVA was employed to analyze the overall differences in body weight change between the two groups of mice. ∗P ​< ​0.1; ∗∗∗P ​< ​0.001.

figs1

Transcriptome sequencing analysis of gene expressions in cells transfected with agshEV71-h1 and agshNC. Human 293T cells were transfected with agshEV71-h1 and agshNC, respectively. At 24 h post transfection, total RNAs were extracted to perform RNA sequencing. Reads align to human genome (GRCh38) were annotated using htseq-count. Genes with counts per million reads (CPM) lower than 5 were excluded. Different expression genes (DEGs) were defined by |log2FC| > 1, P.adj < 0.05. Up-regulated DEGs were showed in red and down-regulated in blue.