key: cord-0754235-rbhr1qso
authors: Besson, Benoit; Kim, Seonhee; Kim, Taehee; Ko, Yoonae; Lee, Sangchul; Larrous, Florence; Song, Jihwan; Shum, David; Grailhe, Regis; Bourhy, Hervé
title: Kinome-Wide RNA Interference Screening Identifies Mitogen-Activated Protein Kinases and Phosphatidylinositol Metabolism as Key Factors for Rabies Virus Infection
date: 2019-05-22
journal: mSphere
DOI: 10.1128/msphere.00047-19
sha: cb04b7111c59c623bd68cf0441f18ccfcddb5b40
doc_id: 754235
cord_uid: rbhr1qso

Throughout the rabies virus (RABV) infectious cycle, host-virus interactions define its capacity to replicate, escape the immune response, and spread. As phosphorylation is a key regulatory mechanism involved in most cellular processes, kinases represent a target of choice to identify host factors required for viral replication. A kinase and phosphatase small interfering RNA (siRNA) high-content screening was performed on a fluorescent protein-recombinant field isolate (Tha RABV). We identified 57 high-confidence key host factors important for RABV replication with a readout set at 18 h postinfection and 73 with a readout set at 36 h postinfection, including 24 common factors at all stages of the infection. Amongst them, gene clusters of the most prominent pathways were determined. Up to 15 mitogen-activated protein kinases (MAPKs) and effectors, including MKK7 (associated with Jun N-terminal protein kinase [JNK] signalization) and DUSP5, as well as 17 phosphatidylinositol (PI)-related proteins, including PIP5K1C and MTM1, were found to be involved in the later stage of RABV infection. The importance of these pathways was further validated, as small molecules Ro 31-8820 and PD 198306 inhibited RABV replication in human neurons. IMPORTANCE Rabies virus relies on cellular machinery for its replication while simultaneously evading the host immune response. Despite their importance, little is known about the key host factors required for rabies virus infection. Here, we focused on the human kinome, at the core of many cellular pathways, to unveil a new understanding of the rabies virus infectious cycle and to discover new potential therapeutic targets in a small interfering RNA screening. The mitogen-activated protein kinase pathway and phosphatidylinositol metabolism were identified as prominent factors involved in rabies virus infection, and those findings were further confirmed in human neurons. While bringing a new insight into rabies virus biology, we also provide a new list of host factors involved in rabies virus infection.

discovery of a RABV vaccine that is effective as a postexposure treatment (3) , reaching a better understanding of RABV biology and finding new therapeutic targets remain priorities.

RABV is a single-stranded negative RNA virus (order Mononegavirales, family Rhabdoviridae) with a 12-kb genome coding for 5 proteins. The nucleoprotein (N) encapsulates the genomic viral RNA, which constitutes a ribonucleoprotein (RNP). The phosphoprotein (P) and the RNA-dependent RNA polymerase (L) bind to the RNP, forming the replication and transcription complex (4, 5) . The glycoprotein (G) is inserted into the envelope of the viral particle and acts as a binding and fusion protein for the entry of the virus into the host cell. The matrix protein (M) links the N protein involved in the RNP to the internal domain of the G protein, shaping the viral particle in a "bullet" shape as well as playing a crucial role in the entry and budding phases, notably hijacking the endosomal sorting complex required for transport or endosomal sorting complexes required for transport (ESCRT) (6) . Together with the N protein, the M protein also regulates the balance between replication and transcription (7) . Moreover, all viral proteins harbor secondary functions to hijack the cellular machinery. The N and L proteins are notably involved in hiding viral RNAs from cell sensors (8, 9) . Further, both the G and M proteins participate in the regulation of cell death, interfering with MAST2, protein tyrosine-phosphatase nonreceptor 4 (PTPN4), TRAIL, and cytochrome c oxidase signaling (10) (11) (12) . Finally, the P and M proteins hijack the IRF (IB kinase [IKK] , TBK1, IRF3), NF-B (RelAp43, p105/p50, ABIN2), JAK-STAT (JAK1, STAT1, STAT2, STAT3), and possibly the mitogen-activated protein kinase (MAPK) (TPL2) pathways in order to control the immune response (13) (14) (15) (16) (17) (18) (19) .

Large-scale screening approaches allow a better understanding of the complex nature of virus-host interactions through the identification of key host factors. Transcriptomic (20) (21) (22) (23) and proteomic (24) (25) (26) experiments identified differentially expressed genes and proteins in RABV-infected cells but generally lacked a clear functional relevance to pathogenesis (23) . A yeast two-hybrid screening determined that the focal adhesion kinase (FAK) interacts with the P protein, and a cell-free protein synthesis screening identified ATP-binding cassette family E1 (ABCE1) as a possible factor for capsid formation such as is observed with HIV (27, 28) . Recently, a first loss-of-function screening was performed on 3,200 mutant neurons differentiated from murine embryonic stem cells using the SPBN-Nfu-GFP laboratory strain (29) . The spreading capacity of green fluorescent protein (GFP)-labeled RABV was investigated at 24, 48, and 72 h, and 63 hits were found, among which 2 targets were validated: Unc13d and Bbs4 (the 61 other hits were neither confirmed nor invalidated). A gene ontology (GO) analysis assigned the 63 hits to the following GO terms: induction of apoptosis, G-protein-coupled receptor binding, and perinuclear region of cytoplasm and ubiquitin-specific protease/thioesterase activity. Further, a genome-wide screening focusing on the late stage of several negative-strand RNA viruses (30) identified nearly 72 important genes for vesicular stomatitis virus (VSV; Rhabdoviridae), among which 25 were also identified as important for lymphocytic choriomeningitis virus (LCMV; Arenaviridae) and human parainfluenza virus type 3 (HPIV3; Paramyxoviridae). A second genome-wide screening was performed on the early stages of VSV infection (31) and identified 300 host genes important for VSV, among which 23 were validated. Taken together, the following 8 common hits were identified as key factors for the early and late VSV infection: ARCN1, COPB1, COPG, COPZ1, MAT2A, NHP2L1, SYVN1, and UTP6.

In the light of the many interactions between rabies virus proteins and cell signaling pathways, we designed a functional genomic approach based on high-content screening using a kinase and phosphatase small interfering RNA (siRNA) library on the fluorescent protein-recombinant field isolate Tha RABV to identify cellular targets involved in the RABV infectious cycle. A total of 106 key host factors were identified in HEK293T cells as "viral helpers" or "viral inhibitors" at early (18 h) or late (36 h) stages of the infection. Taking the results together, the screening provided a list of targets involved in the early and late-stage replication cycles of field isolate RABV.

RNA interference (RNAi) screening for kinase and phosphatase factors in field isolate RABV infection. To identify key host factors for RABV at both the early and late stages of the infection, a kinase and phosphatase siRNA screen was performed with readout at 18 h and 36 h postinfection. The GFP-recombinant field isolate Tha virus (Tha-GFP), engineered in our laboratory, was screened against a focused library of 3,024 duplexes composed of three independent siRNAs for 710 kinase genes and 298 phosphatase genes. HEK293T cells were transfected with a single siRNA and counted 72 h later using Hoechst staining and automated microscopy (see Fig. S2A in the supplemental material). We chose to infect cells at a multiplicity of infection (MOI) of 5 in order to reach 20% of infected cells at 18 h and 70% at 36 h (Fig. S2 ). At the experimental endpoint, cells were fixed, nuclei were subjected to Hoechst staining, and images were acquired for both Hoechst fluorescence and GFP fluorescence by automated microscopy. Images (Fig. S2B ) were analyzed to quantify the number of cell per well (Fig. S2C) , the Tha-GFP intensity in each well (Fig. S2D) , and the percentage of Tha-GFP positive cells (Fig. S2E) .

As a functional control measuring RNAi transfection, we targeted polo-like kinase 1 (PLK1)-an essential gene for cell viability-which reduced the count of nuclei to less than 5% ( Fig. S2B and C) of the count determined for the scramble siRNA (siNEG). The effective RNA interference of Tha replication was confirmed with small interfering GFP (siGFP) and with siRNA targeting the nucleolin (NCL), which was previously described as a key host factor for RABV (32) . Targeting the GFP reduced the levels of Tha-GFP expression to 30% at 18 h and 7% at 36 h compared to siNEG, while the siNCL reduced cell viability to 70% only at 18 h ( Fig. S2B and C) and Tha-GFP expression to 52% of the intensity at 18 h and 82% at 36 h (Fig. S2B , D, and E).

In order to define relevant and specific hits, we analyzed both nucleus counts and Tha-GFP intensity distributions of the population of siRNA duplexes. All siRNAs affecting cell viability at 18 h (Fig. 1A ) and 36 h (Fig. 1B) were labeled toxic (threshold at 30% of siNEG count of nuclei). The mean of the siNEG Tha-GFP intensity was then used to define hits under a 3 threshold as viral helpers and hits over a 2 threshold as viral inhibitors at 18 h (Fig. 1C ) and 36 h (Fig. 1D ). The specificity of the 3 threshold was confirmed by analysis of the controls. At 18 h postinfection, the Tha-GFP signals corresponding to 95% of the level seen with the siGFP-treated well and 45% of the level seen with the siNCL-treated well were under the 3 threshold. At 36 h, 100% of the siGFP-treated wells but none of the siNCL-treated wells had a Tha-GFP signal under the 3 threshold. Therefore, NCL is a key factor for Tha-GFP only in the early stages of the infection, which further validates our system. Finally, only genes confirmed with at least two of three single siRNAs were considered high-confidence key host factors (listed in Table 1 and 2) whereas genes with only one effective siRNA were labeled low-confidence factors in further analysis (listed in Table S1 and S2 in the supplemental material). Using this approach, we were able to identify 54 high-confidence viral helpers for the early stages of the infection and 66 for the late stages of the infection (Fig. 1E ). In comparison, only a few high-confidence viral inhibitors were identified (3 at 18 h and 7 at 36 h) (Fig. 1F) . Interestingly, 24 genes were commonly identified as high-confidence viral helpers at all stages of the infection (Fig. 1G) . Furthermore, 7 high-confidence viral helpers were found to be systematically affecting RABV replication with all three siRNAs, including CDC25C, CHTF18, EPHA7, PPP2CA, and PTPRN at 18 h postinfection and DUSP5 and PRPF4B at 36 h postinfection (Table 1 and 2) .

Identifying key functions and pathways involved in RABV infection. In order to identify the main functions and pathways that are taken advantage of during RABV infection, we next used STRING to form a network of genes identified at 18 h (Fig. 2 ) and 36 h (Fig. 3) postinfection. For this purpose, we computed high-confidence hits and low confidence hits, but none of the hits were labeled toxic. Further, only low-confidence hits showing a direct interaction with a high-confidence hit(s) were displayed; where possible, biological functions were attributed to gene clusters by the use of DAVID.

At the early stages of the infection, two metabolic pathways were identified as key host factors (Fig. 2) . Among 4 hits involved in sphingolipid metabolism, 2 lipid phosphate phosphohydrolases (PPAP2A and PPAP2B) were defined as a high-confidence viral inhibitor and a viral helper, respectively. Fructose metabolism and mannose metabolism were also identified as important for RABV replication as, in a cluster of 4 genes, 2 phosphofructokinase-biphosphatases (PFKBP1 and PFKBP3) were highconfidence viral helpers. Two genes encoding mitotic checkpoint kinases (AURKA and BUB1B) in a cluster of 5 genes involved in the mitotic spindle were high-confidence viral helper genes (Fig. 2) . Interestingly, numerous other hits defined as viral helpers are involved in mitotic regulation. In particular, CDC25C was shown to be important at both 18 h and 36 h postinfection and all three single siRNAs used against it were shown to affect RABV replication. Further, a significant number of MAPK (MEK1/MAP2K1)associated phosphatases (DUSP5, DUSP7, and DUSP22) and downstream effectors (PAK4 and PAK6) were defined as high-confidence viral helpers among a total of 12 MAPK-associated proteins (Fig. 2) . Interestingly, PAK6 interference was the most effective with respect to RABV replication at 18 h, reducing the Tha-GFP signal to 24% of levels seen with the controls ( Table 1 ). The largest cluster of genes involved as viral helpers of RABV corresponded to inositol phosphate metabolism, with 5 highconfidence hits (INPPL1, INPP5E, MINPP1, MTMR4, and PIK3C2G) among 13 hits (Fig. 2) . This result was linked to two downstream effectors of phosphatidylinositol as well as connecting with the previously described MAPK and inositol phosphate pathways. It is also noteworthy that SRC interference had an inhibitory effect on Tha-GFP at 36 h. However, it was not considered for further analysis as it was also shown to be cytotoxic (Table S2) . At the later stages of the infection, only four gene clusters were associated with a function or pathway (Fig. 3) , and all were already identified at the early stages (Fig. 2) . First, mannose metabolism and fructose metabolism were also shown to be important in the later stages, as a phosphofructokinase-biphosphatase (PFKBP2), a fructosebiphosphatase (FBP2), and pyruvate kinase (PKLR) were high-confidence viral helpers. If we consider the hits labeled "low confidence," FBP2 had already been identified at 18 h and the phosphofructokinase PFKM was identified at both 18 h and 36 h (Fig. 2) . Interestingly, mannose and fructose metabolism interference reduced Tha-GFP signal by 50% at 18 h and by more than 70% at 36 h (Table 2; see also Table S2 ). In addition, high-confidence MAPK (MKK7/MAP2K7) and a downstream effector (PAK6) as well as an associated phosphatase (DUSP5) were again identified as viral helpers at 36 h (Fig. 2) . If we consider the hits labeled "low confidence," PAK4 and MAP2K1 (MEK1) were identified with only one siRNA after 18 h (Fig. 2 ; see also High-confidence (Ն2 siRNA, black) and low-confidence (1 siRNA, gray) hit interactions were computed using STRING v10.5 and were visualized using Cytoscape v3.4. Only low-confidence hits directly interacting with at least one high-confidence hit are displayed. The Tha-GFP score corresponds to the average GFP signal measured in each active single siRNA normalized to the mean of the data from the siNEG control. Hits were retrospectively marked as validated (*) or OTE (#). tyrosine-phosphatase nonreceptor (PTPN7) as well as many MAPK regulators labeled low confidence (i.e., RPS6KA1, RPS6KA4, ZAK, JAK1, and KSR1) were identified as viral helpers. PAK6 interference was also strongly affecting RABV replication, reducing Tha-GFP signal by 87%, as well as RPS6KA5 (RSK) interference, reducing Tha-GFP signal by 85% (Table 2 ). It should be noted that JNK/MAPK8 was identified as a viral helper with only 1 siRNA at both 18 and 36 h while extracellular signal-regulated kinase (ERK)/MAPK1 silencing was associated with cell toxicity (Table S1 and S2). Finally, genes associated with inositol phosphate metabolism also constituted the largest cluster observed at 36 h with 7 high-confidence hits (IMPAD1, NPP1, INPP5E, ITPKB, MINPP1, PIK3C2G, and PIP5K1C) among 17 hits (Fig. 3) . Among them, ITPKB interference reduced Tha-GFP signal by 70% whereas the other genes were associated with a 50% to 60% decrease of Tha-GFP signal ( Table 2) .

To provide a comprehensive overview of the most prominent pathways identified as key factors for RABV infection cycle, all the high-confidence hits for both the early and late infection stages were mapped on the MAPK signaling (Fig. 4) , PI metabolism (Fig. 5 ), PI signaling system (Fig. S3) , and mannose and fructose metabolism (Fig. S4 ) pathways using KEGG Pathways.

Confirmation of key host factors. First, we investigated a selection of highconfidence hits using the C911 method (33) to determine whether the inhibition of RABV could have been due to off-target effects (OTE) or if it was specific to the siRNA/target identified (Table S2) . Briefly, the C911 method is based on the design of mismatch siRNAs that carry OTE but lose on-target effects. We selected a single siRNA for high-confidence hits that are present at different levels in the MAPK pathway (MKK7/MAP2K7, DUSP5, and RPS6KA5) or the PI3K pathway (PIP5K1C, MTM1, MINPP1, and PIK3C2G) or that are involved in mRNA capping (PRPF4B and RNGTT) or are associated with various other cellular functions (NRBP2, NEK4, ERN1, and FYN). As the C911-modified siRNA should not affect the targets, we were able to demonstrate that the following hits were clearly subject to OTE: RPS6KA5, MINPP1, PIK3C2G, PRPF4B, RNGTT, and FYN (Fig. 6 ). Both siMAP2K7 (MKK7) and siDUSP5 of the MAPK pathway reached 100% inhibition compared to siGFP (Fig. 6) , a level significantly higher than that seen with their C911 target (P Ͻ 0.001). Both siPIP5K1C and siMTM1 of the phosphatidylinositol 3-phosphate (PI3P) pathway were seen to inhibit Tha-GFP compared to their C911 target (P Ͻ 0.05), providing 25% and 45% inhibition, respectively. Furthermore, NEK4 but not its C911 target (P Ͻ 0.05) inhibited 39% of Tha-GFP replication and both NRBP2 and ERN1 inhibited ϳ20% of Tha-GFP but the levels were not statistically significantly different from those seen with their C911 target (P ϭ 0.054 and P ϭ 0.103, respectively). NEK4, involved in replicative senescence, has never been found to be associated with host-virus interactions (34) . This confirmed that the hits identified in the RNAi screening-in particular, the hits from both the MAPK and phosphatydilinositol-3-phosphate (PI3P) pathways-are important for RABV replication.

The importance of the most prominent pathways was assessed with a small-scale compound screening experiment based on MAPK and PI3P pathway inhibitors cherrypicked in our compound libraries (Sigma and Selleckchem). In total, 10 MAPK inhibitors (Table S4) were tested using 10 concentration points ranging from 10 M to 5 nM.

The compound screening was first run on HEK293T cells infected with Tha-GFP to validate the system used for the RNAi screening. At 18 h (Fig. S5A) , we were able to fit dose-response curves to 8 inhibitors (Ro 31-8220 mesylate, D-609, idelalisib, NSC663284, PD98059, SD169, SL327, and wortmannin), among which 1 (Ro 31-8220 mesylate) was cytotoxic and 2 (NSC663284 and SD169) showed a trend toward significance but gave results that were above the level seen with the control (DMSO [dimethyl sulfoxide]). At 36 h (Fig. S5B) , only PD198603 and Ro 31-8220 mesylate data were fitted to a dose-response curve. As expected, Ro 31-8220 mesylate appeared to be cytotoxic in HEK293T cells. PD198306 was significantly inhibiting Tha-GFP replication by 25% at the highest concentration tested, 10 M, after 36 h (GFP intensity of 3.10 7 opposed to 4.10 7 in DMSO).

To confirm our hits in a more relevant model for RABV infection, we treated human neurons differentiated from induced pluripotent stem cells (IPSCs) with the selected compounds. Using a newly generated Tha-Crimson recombinant virus that provides a better signal/noise ratio in neurons than Tha-GFP, we observed consistent results 18 h (Fig. S6A) and 36 h (Fig. S6B) postinfection. Among 14 compounds, one (NSC663284) was cytotoxic at concentrations above 5 M (Fig. S6A and S6B ) and 3 compounds had an inhibitory effect on Tha-Crimson at 18 h (Fig. S6A) and 36 h (Fig. 7 ; see also Fig. S6B ). PD198306 inhibited Tha-Crimson replication at 10 M (Fig. 7A) , reducing it by 30% at 18 h (Fig. S6A ) and 50% at 36 h (Fig. 7) . Furthermore, PD198306 5 M reduced Tha-Crimson replication significantly by 20% at 18 h (Fig. S6A ) but such a result could not be confirmed at 36 h. In addition, Ro 31-8220 mesylate at 1.25 M inhibited Tha-Crimson replication significantly by 12% at 18 h (Fig. S6A) and by 28% at 36 h (Fig. 7) . At 2.5 M, Ro 31-8220 mesylate inhibited Tha-Crimson replication by 23% at 18 h (Fig. S6A) and by 55% at 36 h (Fig. 7) . Finally, wortmannin inhibited Tha-Crimson replication by 10% from 2.5 M at 18 h (Fig. S6A ) and by 20% from 5 M at 36 h (Fig. 7) .

Several only at a high concentration, suggesting that it could be dependent on the presence of secondary targets, including ERK and PI3K. The lack of cytotoxicity of PD 198306 suggests that it is not related to SRC (Table S1 and S2). As several cyclin-dependent kinases (CDKs) were shown to be involved in RABV replication (Table S1 and S2), it is possible that CDKs mediate PD 198306 inhibition of RABV. Ro 31-8220 mesylate, which targets PKC and MAPK effectors, was the most effective inhibitor of RABV. It should be noted that in the siRNA screening (Table S1 and The data corresponding to the latter were fitted using nonlinear regression. Each value represents the mean of data from 2 separate wells, and error bars represent standard deviations. Data corresponding to differences between the results seen with the compounds and the results seen with DMSO were assessed using ANOVA in Prism 7 (GraphPad). The P values and images are annotated for the lowest dose with a significant effect. ***, P Ͻ 0.001; **, P Ͻ 0.01; *, P Ͻ 0.05. Dotted line ϭ average Tha-Crimson value in DMSO.

Representing only 2% of the genome, protein kinases regulate the structures and functions of 30% of all cellular proteins and are involved in a wide spectrum of cellular processes (35) . As viruses such as RABV exploit many assets of the cellular machinery to their own benefit throughout their infectious cycle, the host kinome represents a target of choice to identify various key host factors in a large-scale RNAi screening.

Several host factors for RABV had already been identified previously in individual studies, including Hsc70 (36), CCT␣ (37) , and NCL (32) . The latter was shown to be important for viral protein expression and infectious virus production after 24 h of infection. According to our results, NCL is indeed important, although not essential, for viral replication at the early stages of the infection; however, it is not required at the later stages (see Fig. S2 in the supplemental material). In a large-scale loss-of-function screening of murine differentiated neurons (29), the following 2 kinases among 63 hits were involved in later stages of RABV infection: Pick1 (PRKCA-binding protein) and FN3KRP (fructosamine 3 kinase-related protein). In our assay, only Pick1 was found to be a viral helper at both early and later stages of RABV infection with a single siRNA. As a protein adaptor for PDZ domain proteins playing a role in synapse trafficking, Pick1 could therefore be involved in the interaction of G protein with PDZ domain proteins MAST1/2 and PTPN4 in the control of cell survival (10) . Further, several PTPN proteins were identified as viral inhibitors in the later stages and that effect could be linked to the capacity of the G protein to interact with PTPN4. Finally, the proteins identified as involved in mitotic regulation (AURKA and BUB1B) are notably important for microtubule regulation and could have an impact on rabies virus transport (38) .

In the light of the different proteins and pathways known to be directly affected by RABV, the present screening provided new insights. It should be noted that although HEK293T cells lack Toll-like receptors (39), they have functioning RLRs and downstream pathways, and they are able to build up an IFN response (40) (41) (42) . The silencing of TPL2 and TBK1, both of which are linked to RABV escape of the immune system (18, 19) , had no effect on RABV replication. MAP2K6/MEK6, whose expression was shown to be repressed by Tha virus (15) , had no significant impact on viral replication. MAP3K14, or NIK, which is crucial for the induction of the noncanonical NF-B pathway (43) , is essential for RABV replication at later stages, while the canonical pathway was shown to be targeted by the M protein to escape the immune response (14, 15, 19) . This indicates that, unlike the canonical pathway, which is inhibited during RABV infection, the noncanonical NF-B pathway is important for RABV replication and the effect could be linked to its disturbance by M-RelAp43 interaction. Indeed, in a previous mass spectrometry analysis of RelAp43 interacting proteins, we found that the M-protein strikingly increases RelB and p100/p52 interaction with RelAp43 (19) . Finally, if we consider a less stringent 2 threshold for the selection of viral helpers, all proteins of the Janus kinase family (JAK1/2/3 and TYK2) had at least 1 siRNA effective at 18 h postinfection. This suggests a role in RABV replication of this protein family triggering the JAK/STAT pathway, inhibited downstream of JAK proteins by the P protein of RABV (16, 17) , as well as in regulating MAPK (44) .

The MAPK pathway is involved in many physiological functions such as cellular growth, differentiation, and survival as well as neuronal functions and the immune response (45) . Here, the presence of several MAPK actors such as MEK1 and RSK2 of the ERK pathway and MLK3 and MKK7 of the JNK pathway as well as MAPK inhibitors (PTPN and DUSP phosphatases) appears to be highly important for RABV replication, suggesting the need of a finely tuned MAPK pathway to ensure the survival of infected cells without triggering the immune response. This correlates with the activation of the ERK/MAPK pathway previously observed in macrophages and microglia infected with RABV, triggering the production of CXCL10 and reactive nitrogen species (46, 47) .

PI metabolism consist of a wide range of phosphorylated lipids involved in events of membrane fusion between cellular compartments (48) and cell signaling, centered around the PI3 kinases (49) . If we have demonstrated here a role of PI metabolism in the RABV infectious cycle for the first time, it has been previously found to be associated with both entry and egress for several viruses. Phosphatydilinositol-3,5biphosphate [PI(4,5)P 2 ] is required for foot-and-mouth disease virus (FMDV; Picornaviridae) and VSV (Rhabdoviridae) internalization (50) as well as Ebola virus (EBOV; Filoviridae) egress (51) , and PI(3)P was shown to affect influenza virus (Orthomyxoviridae) propagation (52) . PI kinases PI3K and PI4K3 are also required for EBOV and severe acute respiratory syndrome coronavirus (SARS-CoV; Coronaviridae) entry, respectively (53, 54) . Further, PI(4)P and PI4K3 are also essential for the formation of the viral replication complex of Aichi virus (Picornaviridae) in organelles (55, 56) . In regard to closely related viruses such as VSV or EBOV, we can hypothesize that (i) several PI-related proteins playing an essential role at the early stages (MTM1, MTMR4, INPPL1, MINPP1, INPP5E, and PI3KC2G) are involved in RABV entry and that (ii) other PI-related proteins only affecting the later stages (IMPAD1, PIP5K1C, INPP1, and ITPKB) are involved in RABV egress. In particular, the PI-related proteins appear to all converge in the production of myo-inositol through I(3)P or I(4)P or I(3,4)P 2 or some combination thereof in the later stages (Fig. 5) .

Ro 31-8220 is a protein kinase C (PKC) inhibitor and is also active with respect to effects on several MAPKs or MAPK effectors (MAPKAP-K1b, MSK1, GSK3b, and S6K1). Interestingly, PKC has been previously shown to be important for the phosphorylation of RABV P protein (57) . Therefore, in accordance with the high sensitivity of the virus to PKC inhibitor Ro 31-8220, the importance of this mechanism for RABV replication should be considered and further investigated. PD 198306 is a MEK1/2 inhibitor and is also active with respect to effects on other cellular kinases (ERK, c-Src, CDKs, and PI3-kinase). Finally, wortmannin primarily targets PI3K but also cross-reacts with other effectors of the phosphatidyl-inositol metabolism (DNA-PKCs, ATR, and PI4K). As many of the secondary targets of Ro 31-8820, PD 198306, and wortmannin were also identified in the RNAi screening as part of the MAPK pathway (MAPKAP-K1b, MSK1, GSK3b, and ERK), the PIP pathway (S6K1, PI3K, and PI4K), and other pathways (c-Src, CDKs), it is difficult to conclude which affects RABV replication, and the answer is likely to involve a combination of several targets. However, the results from this assay were consistent with the idea of MAPK and PI3K being important for RABV infection in differentiated human neurons.

Loss-of-function screenings using technologies such as RNA interference represent powerful approaches allowing systematic analysis of a system at a focused or genomic scale (58) . Here, targeting the kinome provided an insightful view of the RABV infectious cycle, identifying new key host factors and pathways involved at different stages of the infection and supported by the literature. Further, specific compounds confirmed the importance of MAPK and PI3K in differentiated human neurons, a highly relevant model for the study of RABV, opening the way toward understanding the underlying mechanisms.

Small molecules used in this study. A list of the small molecules used in this study and their known targets according to the manufacturer is provided in Table S4 in the supplemental material.

Cell culture. The human embryonic kidney HEK293T cell line (HEK-293T/17; ATCC CRL-11268) was grown at 37°C in 5% CO 2 . HEK293T cells were propagated in complete growth medium containing Dulbecco's modified Eagle's medium supplemented with 10% heat-inactivated fetal bovine serum. All cell culture supplies were from Wellgene (Gyeongsan, South Korea) and Invitrogen (CA, USA).

The neural progenitor cell line (NPC) used in this study carries a neuron differentiation Map2-Nluc marker and was provided by Jihwan Song (CHA Stem Cell Institute, CHA University). NPCs were cultured at a 1:1 ratio of N2 and B27 media with 20 ng/ml of basic fibroblast growth factor (bFGF). N2 medium consisted of DMEM/F12 plus GlutaMAX, 1ϫ N2, 1ϫ nonessential amino acids, 5 g/ml insulin (Sigma), 100 M ␤-mercaptoethanol (Sigma), penicillin, and streptomycin. B27 medium consisted of Neurobasal medium, 1ϫ B27 with or without vitamin A, 1ϫ GlutaMAX, penicillin, and streptomycin (59) . NPCs were maintained on a poly-L-ornithine/laminin-coated dish or plate. All materials for media were supplied by Gibco unless mentioned otherwise.

Recombinant virus construction and reverse genetics. A Tha-GFP recombinant virus was built based on the wild isolate Thailand RABV (isolate 8743 THA, EVAg collection, Ref-SKU: 014 V-02106). The complete sequence of Tha was cloned in Tha-pSDI-HH-flash-SC vector (Tha-rec) as previously described (15) . Using the same approach, the E2-Crimson gene was inserted into the virus to obtain a Tha-Crimson virus. Fragment F3-2 of the genome delimited by SanD1 and Sma1 (see Fig. S1A in the supplemental material) was subcloned in Topo-TA vector (Invitrogen). A PacI restriction site was created by mutagenesis in the M-G intergenic region by replacing 2 bases (Fig. S1A) . The sequence encoding E2-Crimson was amplified from the pEGF-C1 vector (Invitrogen), with simultaneous addition of a short intergenic region that included the initiation and termination sequences of the M gene as indicated in Fig. S1A . The F3-2 fragment was then replaced in Tha-pSDI-HH-flash-SC vector.

Tha-GFP and Tha-Crimson virus were rescued as previously described (15) . Full-length viral cDNA (2.5 g) and plasmids N-pTIT (2.5 g), P-pTIT (1.25 g), and L-pTIT (1.25 g) were transfected in 10 6 BSR T7/5 cells (60) in 6-well plates. Cells were then passaged every 3 days until 100% of the cells were infected. The supernatant was harvested and titrated on BSR cells to determine the MOI. The insertion of the enhanced green fluorescent protein (eGFP) gene was shown to have no effect of viral replication compared to the results seen with the original Tha virus (Fig. S1B) .

Liquid-dispensing and automation system. Several liquid-dispensing devices were used throughout this study. The siRNA duplexes were plated and transferred using a grade 384 stainless steel head with disposable low-volume polypropylene tips on a PP-384-M personal pipettor (Apricot Designs, Monrovia, CA). The addition of cell suspensions and solutions containing RABV was performed using a WellMate (Thermo Fisher Scientific, MA, USA), and cell fixation and staining were performed using an ELx405 automated washer (Biotek, VT, USA). Assay plates (Greiner Bio-One, NC, USA) were incubated in the tissue culture incubator (Thermo Fisher Scientific) under conditions of controlled humidity at 37°C and 5% CO 2 . The assay was performed in a biosafety level 3 (BSL-3) facility with adherence to safety and handling guidelines regarding infectious pathogens.

Kinome-focused siRNA screening for host factors involved in lyssavirus infection. The arrayed kinase-and-phosphatase-focused siRNA library (Ambion Silencer Select v4.0; Thermo Fisher Scientific) contains 3,024 siRNA duplexes covering 1,008 genes. The siRNA library was diluted from 20 M stock solutions to a concentration of 0.25 M, and 5-l volumes were transferred into assay plates for a final testing concentration of 25 nM. The screening was performed in a simple biological and technical replicate, considering that for each gene, 3 independent siRNA duplexes were tested at two different time points. Internal references were included in each assay plate with Silencer Select Negative Control no. 1 siRNA (4390843) as a negative control (siNEG), PLK1 siRNA (s449) as a phenotypic control, and GFP siRNA as a functional control. For the reverse transfection procedure, 5 l of a transfection mix containing 0.5 l Dharmafect1 (GE Life Sciences, CO, USA)-Opti-MEM (Invitrogen) was added to each well and the reaction mixture was incubated for 20 min to promote complex formation. Next, HEK293T cells were dispensed into the assay plates at 2,000 cells per well in 40 l of growth media and incubated with siRNA complexes for 72 h. One plate was used to determine the cell number present before infection. RABV solution was prepared in media for infection at an MOI of 5, and 10 l was dispensed into the assay plates. To measure different time points of infection, the first set of assay plates were processed after 18 h for events occurring in the early stage (first round of infection) and the second set of assay plates were processed after 36 h for events occurring in the late stage (second round of infection and late infection). Cells were fixed in 2.5% paraformaldehyde (PFA) for 20 min followed by staining of nuclei in a solution of 1 M Hoechst stain for 30 min. Images were acquired on an Operetta system (Perkin Elmer, MA, USA) for analysis of GFP fluorescence intensity and Hoechst-stained nuclei. A total of 4 images per well were acquired using a 10ϫ magnifying objective and analyzed using Columbus software (Perkin Elmer).

RNAi screening data analysis and selection. Raw data were analyzed based on the biased discriminant analysis (BDA) method (61) . Briefly, the BDA method was employed as follows. To identify host factor modulators of RABV, the active siRNA duplexes were scored using a threshold based on standard deviations of the means of GFP fluorescence intensity values from the siNEG control. Host factors that are important for RABV infection ("viral helpers") were scored based on the Ϫ3 threshold and the mean, and suppressor genes ("viral inhibitors") were scored based on the ϩ2 threshold and the mean. Thresholds were determined empirically. Cytotoxic siRNA duplexes were scored based on the siNEG, and those showing a reduction in the count of nuclei greater than 70% (below the number of cells initially seeded) were deemed toxic. Host factors obtained with two or more cytotoxic siRNAs were labeled essential and not considered in the analysis.

MAPK-focused and PI3K-focused compound screening and analysis. NPCs were dissociated by the use of 3 ml Accutase (StemPro) and resuspended in NPC medium (1:1 N2 and B27 media, with 20 ng/ml bFGF, without vitamin A), and 3,000 cells were plated per well in 384-well plates. The next day, bFGF was withdrawn to initiate neuronal differentiation, and the medium was refreshed (using 1:1 N2 and B27 media, with vitamin A) every 2 to 3 days for 2 weeks. Differentiation of the NPCs into neurons was confirmed by Nluc quantification (data not shown), and they were further processed after they reached a high density combined with a low level of cell clustering. HEK293T cells were plated in 384-well plates with 4,000 cells/well and incubated for 24 h before compound treatment was performed.

Compounds were prepared in 2-fold dilutions in the 384-well format at 10 points, starting at 10 mM. On the day of experiment, cells were pretreated with compounds 6 h before infection. The compounds were tested in technical duplicate at 10 different concentrations and at two separate time points. One plate was used to determine the cell number before infection with Tha-E2 Crimson virus was performed at an MOI of 3 for neurons and with Tha-eGFP virus at an MOI of 5 for the 293T cells. During the following days, cells were fixed at 18 h and 36 h postinfection with 2.5% PFA and analyzed as described above.

Bioinformatic analysis. For each gene with a positive hit, the GFP intensity measured with each active single siRNA was normalized to the mean of the data from the siNEG population to provide a "GFP score" (%).

Networks were built using STRING v10.5 as follows: for hits to be considered high-confidence hits (i.e., with at least two positive single siRNA results), a "high-confidence score" (0.7) was required over 4 sources (text mining, experiments, databases, and coexpression), and for hits to be considered lowconfidence hits (i.e., with only one positive single siRNA result), a "very-high-confidence score" (0.9) was required over 3 sources (text mining, experiments, and databases). The networks were then merged in Cytoscape v3.4, and only the genes not labeled "essential," the genes labeled "high confidence," and the genes labeled "low confidence" involved in a direct interaction with a gene labeled "high confidence" were displayed.

Signaling pathways were obtained using KEGG Mapper on the KEGG database (62) (63) (64) . Statistical analysis. Group comparisons of data were performed by one-way or two-way analysis of variance (ANOVA) using GraphPad Prism 7 software.

Supplemental material for this article may be found at https://doi.org/10.1128/ mSphere.00047-19. 

Global Alliance for Rabies Control Partners for Rabies Prevention

Re-evaluating the burden of rabies in Africa and Asia

Methode pour prevenir la rage apres morsure

Structure, interactions with host cell and functions of rhabdovirus phosphoprotein

Rabies virus transcription and replication

PPEY motif within the rabies virus (RV) matrix protein is essential for efficient virion release and RV pathogenicity

Rabies virus matrix protein regulates the balance of virus transcription and replication

Rabies virus nucleoprotein functions to evade activation of the RIG-I-mediated antiviral response

Critical role of K1685 and K1829 in the large protein of rabies virus in viral pathogenicity and immune evasion

Attenuation of rabies virulence: takeover by the cytoplasmic domain of its envelope protein

Lyssavirus matrix protein induces apoptosis by a TRAIL-dependent mechanism involving caspase-8 activation

Mitochondrial dysfunction in lyssavirus-induced apoptosis

Identification of the rabies virus alpha/beta interferon antagonist: phosphoprotein P interferes with phosphorylation of interferon regulatory factor 3

RelAp43, a member of the NF-B family involved in innate immune response against Lyssavirus infection

The matrix protein of rabies virus binds to RelAp43 to modulate NF-B-dependent gene expression related to innate immunity

Interaction of rabies virus P-protein with STAT proteins is critical to lethal rabies disease

Rabies virus P protein interacts with STAT1 and inhibits interferon signal transduction pathways

Contribution of the interaction between the rabies virus P protein and I-kappa B kinase ⑀ to the inhibition of type I IFN induction signalling

Regulation of NF-B by the p105-ABIN2-TPL2 complex and RelAp43 during rabies virus infection

Effect of rabies virus infection on gene expression in mouse brain

Attenuated rabies virus activates, while pathogenic rabies virus evades, the host innate immune responses in the central nervous system

Virus infection switches TLR-3-positive human neurons to become strong producers of beta interferon

Screening of upregulated genes in suckling mouse central nervous system during the disease stage of rabies virus infection

Proteomic profiling reveals that rabies virus infection results in differential expression of host proteins involved in ion homeostasis and synaptic physiology in the central nervous system

Proteomics analysis of BHK-21 cells infected with a fixed strain of rabies virus

Proteome analysis of virus-host cell interaction: rabies virus replication in Vero cells in two different media

Focal adhesion kinase is involved in rabies virus infection through its interaction with viral phosphoprotein P

Rupprecht CE. 2013. Host-rabies virus protein-protein interactions as druggable antiviral targets

High-throughput differentiation and screening of a library of mutant stem cell clones defines new host-based genes involved in rabies virus infection

RNAi screening reveals requirement for host cell secretory pathway in infection by diverse families of negative-strand RNA viruses

A genome-wide small interfering rna screen identifies host factors required for vesicular stomatitis virus infection

Identification of a role for nucleolin in rabies virus infection

C911: a bench-level control for sequence specific siRNA off-target effects

Nek4 regulates entry into replicative senescence and the response to DNA damage in human fibroblasts

Mechanisms of specificity in protein phosphorylation

Rabies viruses leader RNA interacts with host Hsc70 and inhibits virus replication

The chaperonin CCT␣ is required for efficient transcription and replication of rabies virus

Double-labeled rabies virus: live tracking of enveloped virus transport

Reconstruction of Toll-like receptor 9-mediated responses in HEK-Blue hTLR9 cells by transfection of human macrophage scavenger receptor 1 gene

Modulation of innate immune signalling by lipid-mediated MAVS transmembrane domain oligomerization

Correction: ATP hydrolysis by the viral RNA sensor RIG-I prevents unintentional recognition of self-RNA

Activation of the RLR/MAVS signaling pathway by the L protein of Mopeia virus

Induction of p100 processing by NF-Binducing kinase involves docking I B kinase (IKK) to p100 and IKKmediated phosphorylation

Intracellular signalling: putting JAKs on the kinase MAP

MAP kinase signalling cascades and transcriptional regulation

Rabies virus-induced activation of mitogenactivated protein kinase and NF-kappaB signaling pathways regulates expression of CXC and CC chemokine ligands in microglia

Rabies virus stimulates nitric oxide production and CXC chemokine ligand 10 expression in macrophages through activation of extracellular signalregulated kinases 1 and 2

Phosphatidylinositol metabolism and membrane fusion

PI3K signalling: the path to discovery and understanding

Plasma membrane phosphatidylinositol 4,5 bisphosphate is required for internalization of foot-and-mouth disease virus and vesicular stomatitis virus

The Ebola virus matrix protein, VP40, requires phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2) for extensive oligomerization at the plasma membrane and viral egress

Effect of the phosphatidylinositol 3-kinase/Akt pathway on influenza A virus propagation

Phosphoinositide-3 kinase-Akt pathway controls cellular entry of Ebola virus

Phosphatidylinositol 4-kinase III␤ is required for severe acute respiratory syndrome coronavirus spike-mediated cell entry

A complex comprising phosphatidylinositol 4-kinase III␤, ACBD3, and Aichi virus proteins enhances phosphatidylinositol 4-phosphate synthesis and is critical for formation of the viral replication complex

The molecular basis of Aichi virus 3A protein activation of phosphatidylinositol 4 kinase III␤, PI4KB, through ACBD3

The phosphoprotein of rabies virus is phosphorylated by a unique cellular protein kinase and specific isomers of protein kinase C

Cell-based genomic screening: elucidating virus-host interactions

Human embryonic stem cell-derived neuronal cells form spontaneously active neuronal networks in vitro

Generation of bovine respiratory syncytial virus (BRSV) from cDNA: BRSV NS2 is not essential for virus replication in tissue culture, and the human RSV leader region acts as a functional BRSV genome promoter

A simple method for analyzing actives in random RNAi screens: introducing the "H Score" for hit nomination & gene prioritization

KEGG as a reference resource for gene and protein annotation

KEGG: Kyoto encyclopedia of genes and genomes

KEGG: new perspectives on genomes, pathways, diseases and drugs