A parallel genome-wide RNAi screening strategy to identify host proteins important for entry of Marburg virus and H5N1 influenza virus

To perform a high throughput RNAi screen to identify the host factors involved in viral entry of highly pathogenic Marburg virus and avian influenza virus H5N1, which require BSL-3 and BSL-4 facilities, respectively, we adopted a surrogate system which allows us to perform the initial screening in a BSL-2 facility [18]. This human immunodeficiency virus-1 (HIV-1) based surrogate assay has been widely used by us and others to investigate the entry mechanisms of highly pathogenic enveloped viruses such as filoviruses [11, 12, 16] (Ebola and Marburg viruses), avian influenza virus H5N1 [13], and severe acute respiratory syndrome coronavirus (SARS-CoV) [19]. It has also been used to identify and develop entry inhibitors as antivirals [14, 15].

In this study, avian influenza virus H5N1 pseudovirions (AIV [13]) and Marburg virus pseudovirions (MARV [20]) were generated and used in the genome wide RNAi screen, as described in the Materials and Methods. The basic principle of this surrogate assay is based on the following two aspects: (1) AIV and MARV viral envelope glycoproteins (HA and GP, respectively) are necessary and sufficient to mediate virus entry, and (2) these glycoproteins can be efficiently incorporated into HIV-1 viral particles. Therefore these pseudovirions retain the entry property dictated by avian influenza virus H5N1 glycoprotein (HA) or Marburg glycoprotein (GP). Thus, aspects of these entry mechanisms can be evaluated despite of the surrogate nature of the viral particles. Furthermore, the pseudovirions carry a luciferase reporter gene which can be used to measure the entry activity of AIV and MARV [18].

The overall RNAi screening protocol is outlined in Fig. 1. The key feature of the protocol is that AIV and MARV pseudovirions were used in parallel in the RNAi screen which allowed us to reduce the number of false positives and to quickly identify Marburg-specific and flu-specific host factors (which will be further discussed below). Briefly, human A549 cells (~1000 cells/well) were reverse transfected with 10nM small interfering RNAs (siRNAs) in two identical 384-well plates simultaneously. These plates were incubated for 48 h for target gene expression knockdown, followed by challenging one plate with AIV pseudovirions and the other plate with MARV pseudovirions. Virions were replaced with fresh medium 24 h post infection and the plates were incubated for additional 24 h. Luciferase assays were then performed to quantify virus infection.

Screen data quality is crucial for subsequent data analysis and hit selection. Quality control of the screen is commonly indicated by the Z’ factor [21] which assesses the separation between measured signals of the positive and negative controls in an assay plate.

In our screens, we included three controls: non-targeting siRNA as the negative control and siRNAs targeting luciferase or ATP6V0C as the positive controls. The pseudovirion carries a luciferase reporter gene, and AIV and MARV virus entry is dependent on the low-pH environment in the endosome/lysosome which is regulated by vacuolar ATPase. So siRNA targeting luciferase or ATP6V0C (a component of vacuolar ATPase) is expected to significantly reduce the final luciferase activity level. In addition, these two positive controls can serve as transfection efficiency controls, monitoring siRNA transfection. Eight control siRNAs of each type were arranged on the column 23 or 24 of a 384-well assay plate, as shown in Fig. 2a.

Figure 2b represents the normalized signal distributions of the samples and the controls. The normalized signal was obtained by dividing the measured signal in each well by the median luciferase signal of all sample wells in a plate. It is clear that both siRNAs targeting luciferase and ATP6V0C greatly reduced luciferase signal level with little variance, confirming their efficacy and excellent siRNA transfection efficiency in the screen. The samples show a broad bell-shaped signal distribution, with a little positively skewed. The signal distribution of the non-targeting control was expected to overlap with that of samples. However in our screen, this distribution was shifted to the right, showing a higher signal level. The skewed sample distribution and the shift of negative control distribution suggest that the assay plates were affected by edge effects or the positional effects (see below).

Despite of the adverse impact of the positional effects, the Z’s calculated from two control combinations (luciferase versus non-targeting and ATP6V0C versus non-targeting), are very close to 0.5, an indication of excellent assay quality (Fig. 2c). In comparison with Z’s from AIV screen plates, Z’s from MARV screen plates exhibited less favorable values and wider ranges, suggesting a larger variation in MARV screen data. This is consistent with what is shown in Fig. 2b, that is that both samples and non-targeting controls from MARV screen plates have flatter signal distributions, as compared with those from AIV screen plates.

In the aforementioned screen strategy, we were more interested in identifying virus-specific host proteins which could be revealed by a comparative analysis approach, due to the parallel nature of the screen. In this approach, the infection rates for AIV and MARV for each siRNA were obtained respectively by normalizing the measured luciferase signals with corresponding plate signal medians; the relative infection index, namely the logarithm of the ratio of their infection rates to base 2 - log₂(ratio_MARV/AIV), was then calculated for each siRNA to indicate specificity. Theoretically, an siRNA with relative infection index equaling to zero means no bias of the siRNA on the infections of both viruses; an siRNA inhibiting MARV infection or enhancing AIV infection results in a minus value, and an siRNA with an the opposite effect results in a positive value.

We used this comparative analysis as a means to correct the observed positional effects. The robust z-score, a robust version of z-score which indicates the number of standard deviations from the mean, was employed to represent data variance. The robust z-score distributions of the normalized signals for the columns and rows from both AIV and MARV screen plates are shown in Fig. 3. Consistent with what we show in Fig. 2b, the distributions from the columns and rows are skewed towards the high z-score value area, with the outer two columns and rows more severely affected. When the relative infection index was introduced, the excess variations of the peripheral positions were greatly reduced and the distributions of columns and rows were corrected to be more symmetrical. The same is true regarding the distributions of the all siRNAs. As shown in Fig. 3 (the bottom panel), the robust z-scores of the normalized signals from both screens are heavily skewed to the right, with almost no siRNAs showing robust z-score less than −2, a recommended criterion for picking inhibitory hits. The robust z-scores of the relative infection index, however, are evenly distributed, reducing the skewed distribution caused by the positional effects and making the data more suitable for the hit selection.

Since the relative infection index approach can reduce the systematic errors, we combined it with the individual virus infection rates to identify virus-specific host factors. As presented in Fig. 4a, the hit selection began with normalizing the measured luciferase signals by the plate signal medians for MARV and AIV, respectively. These normalized scores were used for calculating the relative infection index, from which a robust z-score was then computed for each siRNA. Thresholds of ±2 were adopted for filtering the hits. From the total 64,775 individual siRNAs, 3,036 siRNAs targeting 2,858 genes had robust z-scores less than −2, while 2,409 siRNAs targeting 2299 genes had robust z-scores greater than 2. Another criterion (2 siRNAs per gene rule) that a gene must have at least 2 siRNAs showing similar effects was applied to further filter the hits, with the aim of reducing false positive numbers. Accordingly, 173 genes with 351 siRNAs for MARV and 109 genes with 219 siRNAs for AIV were selected, respectively. The individual virus infection rates (the normalized signals) were then employed as a reference to distinguish an inhibitory hit from the one that enhances virus infection. The robust z-scores are plotted against normalized signals of each virus in Fig. 4b and c. A robust z-score less than −2 in Fig. 4b means that the MARV infection rate is much less than AIV infection rate for the corresponding siRNA. This may be a result of inhibited MARV infection or of enhanced AIV infection. Thus a MARV score (normalized signal) below 0.5 was used to indicate a MARV inhibitory hit, which is represented in area I in Fig. 4b. A gene with at least two siRNAs’ MARV scores < 0.5 was used to filter the hits in area I, with 106 genes being identified as host factors that inhibit MARV entry. On the contrary, MARV scores greater than 0.8, which means the siRNA has little effect on inhibiting MARV infection, were used to indicate a hit enhancing AIV entry. Area III in Fig. 4b represents these hits and 4 host factors were identified with the 2 siRNAs per gene rule. Similarly, area I in Fig. 4c represents AIV inhibitory hits and 42 genes were identified; area III represents the hits enhancing MARV entry with 12 genes finally identified.

The use of relative infection index together with individual virus infection rate and the 2 siRNAs per gene rule is a powerful means to identify virus-specific host factors and to reduce the false positive rate. The parallel nature of the screen strategy makes the results from the two viruses highly comparable and the relative infection index obtained by comparison easily establishes virus-specificity. As shown in Fig. 4b and c, only hits in areas I and III were picked for further analysis. Though many other siRNAs also showed inhibitory effects (area II), these siRNAs were ruled out for lack of virus-specificity, greatly reducing the false positive rate. Further, the 2 siRNAs per gene rule helped to reduce false positive which may result from off-target effects or systematic errors.

A selected short list of EBOV or AIV entry-related genes is shown in Table 1 to demonstrate the power of the parallel screen strategy described in this report. Four of seven subunits of the coatomer 1 (COP-I) vesicular transport complex (i.e., ARCN1, COPA, COPB2, and COPG) were reported to be critical for influenza virus replication by others [8], but only ARCN1 was identified as an AIV-specific hit in this screen (Table 1). The other three proteins were identified in our screen as “shared” host factors for both AIV and MARV (data not shown). Proton-transporting V-type ATPase was implicated as the host factor for influenza virus infection by the previous screens [8], and we show that only two subunits of ATPase, APT6V0C and ATP6AP2, are AIV-specific. A number of other subunits (i.e. ATP6AP1, ATP6V0B, ATP6V0D1, ATP6V1A and ATP6V1B2) are not specific to AIV but are “shared” host factors by both viruses. As for MARV, a few essential MARV-specific host factors were identified by the screen in this study. Cathepsin L (CTSL), which primes the filovirus glycoprotein in the endosome [22], EXT1 [23], which is involved in biosynthesis of heparan sulfate, and Niemann-Pick disease, type C1 (NPC1 [24, 25]), showed strong bias against MARV entry (Table 1). Also, HOPS complex, which mediates fusion of the endosome and lysosome, was identified as a host factor for Ebola virus infection in a gene-trap based screen [24], and we identified VPS 16, a subunit of HOPS complex, as a MARV-specific hit in the current screen.

It is interesting to note that two host proteins, FOLR1 [26] and TIM1 [27], which were previously reported as host factors important for filovirus entry, were not identified as the hits in this study (Table 1). The results from this report are consistent with a study which showed that FOLR1 was not critical for filoviral entry [28]. However, more studies are needed to evaluate the potential role of TIM1 in filoviral entry.

Human 293 T embryonic kidney cells and human lung epithelial A549cells were cultured in Dulbecco's modified Eagle's medium (DMEM, Cellgro) supplemented with 10 % fetal bovine serum (FBS, Gibco), 100 μg/mL of streptomycin and 100 units of penicillin (Invitrogen). The avian influenza virus H5N1/HIV pseudovirions (AIV) and Marburg virus/HIV pseudovirions (MARV) were generated from the following plasmids: hemagglutinin (HA), isolated from a highly pathogenic avian influenza virus, A/Goose/Qinghai/59/05 (H5N1) [13]; neuraminidase (NA) from A/Puerto Rico/8/1934 (H1N1) [13]; Marburg virus glycoprotein GP (MGP) [20]. The HIV-1 proviral vector pNL4-3.Luc.R⁻E⁻ [32, 33] was obtained through the NIH AIDS Research and Reference Reagent program.

AIV and MARV pseudovirions were produced by transient co-transfection of human 293 T cells using a polyethylenimine (PEI)-based transfection protocol [11]. Replication-defective HIV vector (pNL4-3.Luc.R⁻E⁻) together with plasmids encoding MGP or HA/NA were used for transient co-transfection into 293 T producer cells. Six hours after transfection, cells were washed with phosphate-buffered saline (PBS), and 40 mL of fresh medium was added to each plate (150 mm). Forty-eight hours post transfection, the supernatants were collected and filtered through 0.45 μm pore size filter (Nalgene). The pseudovirion stocks were stored at 4 °C prior to use.

Three siRNA libraries (Silencer® Select Human Druggable Genome siRNA Library V4, Human Druggable Genome V4 Extension and Human Genome V4 siRNA Extension) were purchased from Ambion (Austin, TX). These libraries contain 64,755 siRNAs targeting 21,585 human genes (each gene has three distinct siRNAs). Daughter plates were prepared in 384-well format at a 250nM concentration in water. Controls include siRNA targeting ATP6V0C or luciferase and non-targeting siRNA. All control siRNAs were purchased from Ambion.

All siRNAs were arrayed in 384-well plates. For each assay plate well, 2 μl siRNA was mixed with 0.1 μl Lipofectamine RNAiMAX (Invitrogen) and 9.9 μl Opti-MEM (Invitrogen). After 20 min incubation at room temperature, 38 μl cell suspension of 1000 A549 cells was added, resulting in a final siRNA concentration of 10nM. Cells were incubated at 37 °C and 5 % CO₂ for 48 h. Then the medium was removed and the cells were challenged with 30 μl MARV or AIV virions in parallel. Virions were removed 24 h later and 30 μl fresh DMEM supplemented with FBS and antibiotics was added to each well. After another 24 h incubation, 15 μl of neolite luciferase substrate (PerkinElmer, Waltham, MA) was mixed in, incubated for 5 min, and luciferase activity was measured with an Envision plate reader (PerkinElmer). All multi-well pipetting procedures were conducted by the JANUS automated liquid handling system (PerkinElmer).

The measured luminescence signal was analyzed using the statistical programming language R. The signal in each well was first normalized by the median value of all the samples from the same plate.

The Z’ factor [21] was calculated from the normalized signals of not-targeting and ATP6V0C controls or from those of not-targeting and luciferase controls on each plate with the following equation: Z ' = 1 − 3(Std_{high ‐ signal} + Std_{low ‐ signal})/(Mean_{high ‐ signal} − Mean_{low ‐ signal}).

For each siRNA, the relative infection index was calculated from the corresponding normalized signals from the MARV and AIV plates with the following equation: relative infection index = log₂ (normalized signal_MARV/normalized signal_AIV).

Robust z-score was calculated for both normalized signal and relative infection index. First, median absolute deviation (MAD [34]) was calculated as follows:and then the robust z-score was calculated for each siRNA as follows:where x represents the set of all the normalized signals or the relative infection indexes of the samples and x_ij indicates the value for a particular well at row i and column j. The constant 1.4826 is used to make MAD comparable to standard deviation.