A Genome-wide CRISPR-Cas9 Screen Identifies Host Factors Essential for Optimal Plasmodium Liver Stage Development

Prior to initiating symptomatic malaria, Plasmodium parasites infect and develop within hepatocytes. We performed a forward genetic, genome-wide CRISPR-Cas9 screen to identify host regulators of Plasmodium liver infection. Single guide RNAs targeting genes involved in vesicle trafficking, cytoskeleton organization and lipid biogenesis altered Plasmodium liver development. We observed a redistribution of Golgi-derived vesicles and fragmented Golgi stacks with the parasitophorous vacuolar membrane (PVM). The host microtubule network and non-centrosomal microtubule organizing centers (ncMTOC) also re-localized following infection, closely associating with the parasite. Knocking out the centrosomal MTOC protein CENPJ exasperated the re-localization of MTOCs to the parasite and increased infection, suggesting that the parasite relies on ncMTOC assembly. Thus, we have uncovered a mechanism by which parasites sequester host material for survival and development. Our data provide a wealth of yet untested hypotheses about the elusive biology of the liver stage parasite and serves as a foundation for future investigation.


Introduction
Malaria is transmitted to humans by the injection of Plasmodium sporozoites into the skin during the blood meal of an infectious female Anopheles mosquito. Sporozoites exit the skin by traversing blood vessels to enter the circulation and are then carried to the liver. Sporozoites leave the circulation by traversing the sinusoidal cell layer and infecting hepatocytes and forming liver stages (LS) (Mota et al., 2001;Shortt and Garnham, 1948;Vanderberg, 1981). LS parasites reside in a membrane-bound compartment in the hepatocyte termed the parasitophorous vacuole (PV), where they differentiate into exoerythrocytic merozoites. The PV membrane subsequently breaks down, and merozoites reenter the blood and infect erythrocytes (Sturm et al., 2006).
Hepatocyte infection is obligate for parasite life cycle progression, and thus is an important target for antimalarial intervention. Elimination of the parasite during this stage would block both disease symptoms and transmission. Several studies have shown that Plasmodium relies on multiple surface factors including CD81 (cluster of differentiation 81) (Silvie et al., 2003), SRB1 (scavenger receptor class B type 1)  and EphA2 (ephrin type-A receptor 2) (Kaushansky et al., 2015) for hepatocyte entry.
Multiple forward genetic screens have informed our understanding of host regulatory factors for LS malaria. A kinome-wide screen identified five kinases that play a major role in P. berghei infection . More recently, we used an orthologous approach, kinase regression, and confirmed that many of the originally described kinases play a role in a related P. yoelii system. We also showed many additional kinases regulate Plasmodium LS infection (Arang et al., 2017). Separately, an RNAi screen against 53 targets identified SRB1 as an invasion factor , and a subsequent screen against 6,951 druggable targets identified COPB2 (coatomer subunit beta) and the ADP-ribosylation factor-binding protein GGA1 (Golgiassociated, gamma adaptin ear containing, ARF binding protein 1) as development factors (Raphemot et al., 2019) for Plasmodium LS infection. While these screens have provided valuable insights into parasite-host interactions, the range of targeted genes and overlap in the identified hits has been limited, suggesting that a complete complement of factors required for Plasmodium entry and development remains to be discovered.
The recent development of genome-wide CRISPR-Cas9 technology enables generation of complete loss-of-function alleles in a variety of cell types, enabling functional genetic analyses in higher eukaryotes Shalem et al., 2014). Here, we report the first genome wide CRISPR-Cas9 screen that aims to identify novel host factors that regulate Plasmodium infection. Importantly, we identify pathways that overlap with previous screens, in addition to uncovering novel host regulators of P.

CRISPR-Cas9 screen to identify host regulators of infection
To identify host genes critical for Plasmodium LS development, we used the GeCKOv2 sgRNA pool to generate a whole-genome knockout library in HepG2-CD81 cells Shalem et al., 2014). HepG2-CD81 cells were transduced with lentivirus containing the pooled GeCKOv2 sgRNA library of 114,123 sgRNAs targeting 19,031 protein-coding genes (~6 sgRNAs/gene) and selected in puromycin for 5-7 days. To evaluate sgRNA diversity in the HepG2-CD81-GeCKOv2 library, we PCR-amplified the integrated sgRNA cassettes from genomic DNA extracted from transduced cells and subjected the amplified library to Illumina sequencing. At the gene level, sgRNAs targeting all but 15 of 19,031 (99.92%) protein-coding genes were observed. Twelve to fourteen days after post-transduction, we infected forty million puromycin-resistant cells with green fluorescent protein (GFP) expressing-Plasmodium yoelii at a multiplicity of infection (MOI) of 0.3. After 24 h of infection, cells were sorted into infected and bystander cell populations by GFP signal intensity with fluorescence-activated cell sorting (FACS) (Fig. 1A). Separately, a parallel culture of uninfected cells was also maintained to normalize the sgRNA frequency distributions. We obtained four independent biological replicates with library generation and sequencing occurring in parallel. Genes with significantly enriched sgRNAs were identified for both the bystander and infected populations, when compared to uninfected cells.
Cells that harbor genetic alterations restricting P. yoelii development (i.e., sgRNAs that target host genes important for infection) were expected to be enriched in the uninfected bin; we termed this group 'putative positive regulators of infection'. We categorized sgRNAs enriched in the infected cells as 'putative negative regulators of infection'. In this initial screen, we identified 242 sgRNAs that were statistically enriched in infected or bystander groups after accounting for multiple hypotheses. There were 67 sgRNAs significantly enriched in the infected cells compared to uninfected cells and 175 genes were significantly enriched in the bystander bin relative to uninfected bin. We reasoned that biological pathways with multiple putative regulators were more likely to be bona fide regulators of infection. We used gene ontology (GO) pathway analysis to identify significantly enriched biological processes (Fig. 1B). Genes present in both statistically enriched pathways were selected for further validation. After this stringent down-selection step, we were left with eight putative negative regulators of infection and seven putative positive regulators of infection for further study (Fig, 1C, D).

Identifying host factors that regulate Plasmodium LS invasion and development
To distinguish between genes impacting Plasmodium LS invasion versus those impacting development, we individually disrupted each of the 15 putative regulators with three sgRNAs per gene using CRISPR-Cas9 gene editing of HepG2-CD81 cells. In this system, a fluorescent reporter, GFP, is expressed only upon guide integration, enabling us to exclude any cells that did not take up and integrate the sgRNA (Fig. S1A). Knockout efficiency of CENPJ was also confirmed using western blot ( Fig. S1B and S1C). We infected each knockout line with P. yoelii sporozoites for 90 mins and assessed hepatocyte entry by flow cytometry. Among the selected 15 hits, only LRP4 (low-density lipoprotein receptor-related protein 4) exhibited significantly reduced entry of sporozoites 90 min after infection ( Fig. 2A). This is consistent with the previous finding that CSP interacts with LRP and HSPG to facilitates host cell invasion of Plasmodium (Shakibaei and Frevert, 1996).
As an orthogonal approach, we modulated HepG2-CD81 cells with small molecule inhibitors targeting positive regulators identified in the screen (  Table 1). IC50 values for each small molecule inhibitor were obtained in HepG2-CD81 cells using Live/ Dead staining (Fig. S1D). We included eltanexer, an inhibitor of exportin-1 (XPO1), a pututive negative regulator of infection (Than et al., 2020). To test the role of LRP4 in sporozoite entry of hepatocytes, we pretreated HepG2-CD81 cells with HPA-12, a ceramide transport inhibitor (Berkes et al., 2016) and an LDL-R blocking peptide which blocks LRP4, significantly inhibited sporozoite entry. Thus both genetic and peptidemediated intervention of LRP4 inhibits sporozoite entry of hepatocytes ( Fig. 2A and 2B).
We next performed an imaging-based secondary screen with the selected 15 putative regulators to assess the role of these hits on the longer-term LS development.
Individual CRISPR-Cas9 knockout lines were infected with P. yoelii sporozoites and observed for 48 hours post infection (hpi). Several of the knockout lines exhibited substantially altered LS burden (Fig. 2C). The number of LS parasites was significantly increased in CENPJ (centromere protein J) and KDELC1 (Lys-Asp-Glu-Leu containing 1) disrupted lines, illustrating that each of these factors is indeed a negative regulator of infection. In contrast, knockout of CISD1 (CDGSH iron sulfur domain 1), COL4A3BP (collagen type IV alpha-3-binding protein), CERT1 (ceramide transporter protein 1), IREB We next asked if any of the screen hits altered the growth of LS parasites.
Interestingly, knock out of VPS51 (Ang2) did not significantly alter parasite load but instead, the size of the parasite was significantly smaller suggesting it may regulate parasite growth. HepG2-CD81 cells expressing sgRNAs directed against COL4A3BP and LRP4 exhibited a significant reduction in the size of the parasite (Fig. 2E). Knockout of other putative regulators did not result in altered parasite size (Fig. 2E). Interestingly, while our screen was only set up to identify factors that altered infection rate, not LS growth, it is possible that some slow-growing parasites may have not reached the threshold of GFP levels to be included in the "infected" gate. Taken together, these studies identified several host factors influencing parasite entry, growth and development ( Fig. 2F) and illustrates the utility of genome-scale functional screening for the discovery of host factors that regulate Plasmodium LS infection.

Host Golgi and Golgi-derived vesicles interacts with Plasmodium
Interestingly CERT1, a positive regulator, and VPS51, a negative regulator, were both involved in trafficking to and from the Golgi. Specifically, CERT1 regulates ER-to-Golgi and Cis Golgi-to-trans Golgi network (TGN) traffic (Funakoshi et al., 2000). VPS51 (Ang2) regulates late endosome-to-TGN and Golgi-to-ER traffic (Perez-Victoria et al., 2010). Given this, we hypothesized that targeting vesicle trafficking to and from Golgi impacts parasite survival. Moreover, GO term enrichment analysis (Fig. 1B) suggests cytoskeleton organization, vesicular trafficking and Golgi and ER stress significantly impact LS infection. Since increases in Golgi-stress and vesicular trafficking are associated processes (Sasaki and Yoshida, 2015), we chose to visualize the Golgi and Golgi-derived vesicles in infected cells. We infected HepG2-CD81 cells with P. yoelii sporozoites and allowed infection to proceed for 24 or 36 h. Cells were stained with anti-UIS4 (upregulated in infectious sporozoites gene 4) and anti-FTCD (formiminotransferase-cyclodeaminase) antibodies to visualize the parasite parasitophorous vacuole membrane and host Golgi-derived vesicles, respectively. We observed an enrichment of Golgi-derived vesicles near the parasite (Fig. 3A), consistent with other reports (De Niz et al., 2020;Raphemot et al., 2019). To assess Golgi morphology during infection, we infected HepG2-CD81 cells with P. yoelii sporozoites and stained cells with antibodies against the Golgi peripheral cytoplasmic membrane protein, GM130 (Golgi membrane protein of 130 kDa; golgin subfamily A member 2). As a control for Golgi fragmentation, we used brefeldin A, which reversibly disrupts and fragments the Golgi, blocking assembly and transport of secretory vesicles (Sciaky et al., 1997). We observed fragmented Golgi stacks in infected cells 24 hpi, which resolved by 36 hpi (Fig.   3B). We also observed UIS4-positive membrane positioned near the Golgi stacks in nearly three quarters of the infected cells at 24 hpi (white arrowheads, Fig. 3B). This observation suggests that the PVM or the tubulo-vesicular network is positioned near the Golgi stacks and/or Golgi-derived vesicles.

Plasmodium hijacks the host microtubule network
Several studies have reported an association between LS parasites and host derived vesicles, including late endosomes (Lopes da Silva et al., 2012), autophagosomes (Real et al., 2018) and lysosomes (Niklaus et al., 2019;Risco-Castillo et al., 2015;. Microtubules (MTs) support the spatial positioning and vesicular trafficking of these and other organelles (Vale, 2003). Genes with roles in MT cytoskeletal organization were associated with a series of GO terms enriched in our screen (Fig.1B). Thus, we next asked if the localization of host vesicles, including Golgi-derived vesicles, at the Plasmodium PVM, was regulated by the host MT network. Interestingly, the Golgi acts as a MT organizing center (MTOC) by recruiting γ-tubulin with a subset of MT originating at the Golgi in mammalian cells, and across many systems (Sanders and Kaverina, 2015). To visualize changes in the MT network initiated during infection, we transfected HepG2-CD81 cells with CellLight™ RFP-α-Tubulin BacMam 2.0 and allowed the infection to proceed for 24 h (Fig.4A). Strikingly, we observed that the host MT network redistributes to, the LS parasite, appearing to wrap around the PVM (Fig.4A). In contrast, MTs in uninfected cells form a canonical network around the nucleus, radiating toward the cell periphery (Fig.4A).
Acetylated MT are the stabilized form of MTs that support kinesin-mediated trafficking of vesicles (Reed et al., 2006). We next asked if parasite associated MTs are actively trafficking by assessing the acetylation of MTs. We infected HepG2-CD81 cells with P. yoelii sporozoites for 24 hours, and then visualized acetylated alpha-tubulin and PyUIS4 by immunostaining (Fig.4A). MTs that decorate the parasite periphery were highly acetylated. In contrast, in uninfected cells, acetylated MT were scattered throughout the cell. These results are consistent with a model where the MT network drives elevated levels of vesicular traffic to the parasite periphery in infected cells.

Non-centrosomal microtubule organizing centers sequester at parasite periphery.
To understand MT redistribution in infected cells, we visualized MT organizing centers (MTOCs) (Zheng et al., 1995) and assessed their role in regulating MT nucleation and growth during infection. MT arrays originate from MTOCs that are either organized canonically at the centrosome or non-canonically at Golgi (Sanders and Kaverina, 2015;Zhu and Kaverina, 2013). Centrosome-regulated MTs are strictly radial while noncentrosomal organized MTs display asymmetrical organization (Zhu and Kaverina, 2013).
Given the non-canonical organization of MTs around the PVM and the redistribution and relocalization of Golgi-derived vesicles with the PVM, we hypothesized that MTs organized around the parasite were non-centrosomal. To test this, we evaluated the localization of calmodulin-regulated spectrin-associated protein 2 (CAMSAP2), a minusend binding protein that stabilizes non-centrosomal MTs (Yau et al., 2014). In uninfected CellLight™ GFP-α-Tubulin BacMam 2.0 transfected HepG2-CD81 cells, CAMSAP2 found distributed evenly throughout the cells. Strikingly, we observed enrichment of CAMSAP2 on the MTs around the PVM suggesting that parasite localized MTs are noncentrosomal, consistent with the hypothesis that they are stabilized at the parasite periphery (Fig.4B).
We next asked how MTs were organized in infected cells by assessing the localization of γ-tubulin with γ-TuRC (γ-tubulin ring complex). γ-TuRC is a core functional unit of MTOCs and functions as a MT nucleator (Wiese and Zheng, 2000), usually localized to the centrosome. In uninfected cells, γ-tubulin was localized primarily near nuclear periphery (88%). In infected cells, a majority of γ-tubulin foci (~60%) were found in the cytoplasm associated with the PVM. During cell maturation, γ-TuRC gradually shifts localization from the centrosome to the cytoplasm￼￼ and nucleates MTs ￼￼. Together, the association of the parasite with the Golgi, re-localization of γ-TuRC to the cytoplasm, and the observed localization of CAMSAP2 to the PVM, supports the hypothesis that MT nucleation around the parasite is of ncMTOC origin.
We next evaluated the role of CENPJ, which is involved in establishing centrosomal MTOCs (Hung et al., 2000), in assembling ncMTOCs in infected cells. We generated CENPJ disrupted lines in HepG2-CD81 cells ( Fig. S1C and D). In CENPJ disrupted cells, we observed an increase in cytoplasmic localization of γ-tubulin (~80%).
Infection in CENPJ knockout cells resulted in an increase in γ-tubulin localization (~92%) to the PVM compared to infected control cells (~60%) (Fig. 4B) and an increase in LS infection (Fig.2C). This is consistent with a model where the parasite interacts with host Golgi, and Golgi-associated ncMTOCs to traffic host vesicles to the PVM and promote LS development.

Integrating multiple forward genetic screens provides additional testable hypotheses
We report the first global screen for host factors that regulate Plasmodium LS infection.
Yet, like previous screens, our screen includes false negatives as sgRNAs are lost during the generation of the library and/or not all sgRNAs result in the disruption of the functional protein. To generate a more comprehensive picture, we systematically compared our screen, which interrogated host regulators of P. yoelii infection, with earlier forward genetic screens Raphemot et al., 2019;Rodrigues et al., 2008) that identified regulators of the closely related parasite, P. berghei (Fig. 5A, Supplementary file 1). For the purpose of analysis, we pooled results from the screens by Rodrigues et al., 2008 andPrudencio et al., 2008, as these two screens used the same methodology but had no overlapping factors evaluated. A meta ranking of the 3 screens was performed by sorting each screen separately by z score, calculating each gene's rank percentile location after sorting, and then averaging the gene rank percentile locations across the 3 screens, with no penalty for a gene being missing in a screen. This meta ranking was lastly sorted by average rank percentile location and augmented with the average z score from all screens where the function of the gene was evaluated (Supplementary file 1). Positively and negatively represented genes were sorted separately, then combined afterwards for pathway analysis. At gene level, using the z score cutoff of 2 and 1.5, our screen shared only few hits each of the other screens ( Fig.   5B and C). When we loosened the stringency of the cutoff to a Z-score of 1, there were several genes overlapping between the three screens, although false positive rates could be higher at this cut-off.
We reasoned that hits from overlapping pathways and biological processes might be present in all screens, despite little specific gene overlap. To gain insights into the biological significance of the hits from all screening efforts, we employed ClueGO to determine gene ontology (GO) (Bindea et al., 2009). Despite little or no overlap in specific gene hits, we observe significant enrichment in biological processes from the genes represented in at least two of three screens at z-score of 1.5 and 1. Specifically, we identified 19 high confident biological processes that are significantly enriched using a zscore cutoff of 1.5. This includes biological processes that have been previously described, such as scavenger receptor activity and cholesterol biosynthesis reported (Itoe et al., 2014;Labaied et al., 2011;Petersen et al., 2017;Rodrigues et al., 2008) as well as anterograde cargo transport that we have interrogated in more detail in this study ( Fig.   5D and E). Taken together, this combined resource provides a wealth of hypotheses for further investigation.

Discussion
For decades, dogma suggested that the elimination of anything short of 100% of LS parasites would result in little to no benefit in the effort towards malaria eradication. New evidence suggests this is not the case. Mathematical models suggest elimination of even a portion of hypnozoites could dramatically reduce P. vivax prevalence (White et al., 2018). Recently, it was demonstrated that targeting host aquaporin 3 leads to the elimination of P. vivax hypnozoites from field isolates (Posfai et al., 2020), suggesting that host targeted interventions may provide an opportunity to tackle even the Achilles heel of malaria control efforts. Another host factor known to be critical for stages infection, the tumor suppressor p53 (Kain et al., 2020;Kaushansky et al., 2013), has been associated with lower severity of infection in Malian children (Tran et al., 2019). Additionally, it was recently demonstrated that host targeted interventions can induce at least partial immunity to subsequent challenge (Ebert et al., 2020). This opens the door to the possibility of the use of host targeted drugs in casual prophylaxis strategies, provided the drugs have suitable toxicity profiles (reviewed in (Glennon et al., 2018)). Because not all host targets are suitable drug targets, a broad and comprehensive picture of factors that regulate LS malaria infection is urgently needed.
Previous forward-genetic screens have partially provided that picture and identified host factors involved in Plasmodium infection Raphemot et al., 2019;Rodrigues et al., 2008). However, these screens have exhibited very little overlap in identified factors (Fig. 5), presumably in part because each screen prioritized identifying a small, but bona fide list of "hits," and suffered a high false negative rate as a result. This has led to many key insights into the interactions between the malaria parasite and its host hepatocyte but has fallen short of providing a systematic view of the fundamental biological properties that regulate the development and survival of the LS parasite. Like the earlier screens, the CRISPR-Cas9 screen we report here does not exhibit substantial overlap with previous screens when individual gene hits are evaluated, suggesting that additional analysis is still needed in order to comprehensively assess factors that regulate infection. Yet, when we evaluate whether hits from our screen are present in similar pathways to those observed in other screens, the overlap is substantial (Fig. 5D, E). This suggests that while we may have, as a field, identified many central regulatory biological functions that control LS development, we have yet to saturate our understanding of the molecular players that mediate these biological necessities. Together, the CRISPR-Cas9 screen we present here, along with the previously reported siRNA screens, represent a key resource for the field moving forward, and we anticipate that merging findings from these experiments (Supplemental File 1) will provide many additional hypotheses to probe. One limitation of this work is that, since it is likely that at least a subset of canonical  (Coppens et al., 2006), suggesting that this may be a conserved mechanism by which Apicomplexan parasites exploit for nutrient uptake and survival.

Mosquito rearing and sporozoite production
For P. yoelii sporozoite production, female 6-8-week-old Swiss Webster mice (Harlan, Indianapolis, IN) were injected with blood stage P. yoelii (17XNL) parasites to begin the growth cycle. Animal handling was conducted according to the Institutional Animal Care and Use Committee-approved protocols. Briefly, Anopheles stephensi mosquitoes were allowed to feed on infected mice after gametocyte exflagellation was observed. Salivary gland sporozoites were isolated using a standard protocol at day 14 or 15 post-blood meal. The sporozoites were activated with 20% v/v FBS and pelleted by centrifugation at 1,000 × g to salivary gland detritus. Sporozoites were further enriched by a second centrifugation at 15,000 × g for 4 min at 4 °C, before resuspension in a desired volume of complete medium.

Pooled genome-wide CRISPR screen
To perform the whole-genome CRISPR screen, HepG2-CD81 cells were transduced with lentivirus containing the GeCKOv2 pooled sgRNA library of 123,411 sgRNAs targeting 19,031 protein-coding genes (~6 sgRNAs/gene), 1,864 microRNAs (4 sgRNA/microRNA) 1,000 negative controls (2 sgRNA/control), and selected in puromycin for 5-7 days. On day 12-14 post-transduction, 40 million puromycin-resistant cells were infected with GFP tagged-P. yoelii at a MOI of 0.3. After 24 h of infection, cells were sorted as infected and uninfected by FACS into different bins based on GFP signal. A non-treated, non-infected control was also collected for each experiment to assess library representation. The experiment was performed four independent times. Genomic DNA from each sample was isolated using QIAamp DNA mini kit (Qiagen, Hilden, Germany).

Next-generation sequencing
Libraries were generated using a 2-step PCR according to previously published protocol . Briefly, an initial PCR was performed using AccuPrime Pfx Supermix (Invitrogen, Waltham, MA, USA) with lentiCRISPRv2 adaptor primers to amplify the sgRNA region and add priming sites for Illumina indexing. Amplicons were purified using FlashGels (Lonza, Allendale, NJ, USA) and purified PCR products were used as templates for subsequent PCR amplification. Sufficient PCR reactions were performed to maintain library coverage. Next, a second PCR was performed in order to add Illumina P5 and P7 index sequences, as well as barcodes for multiplexing, and samples were re- Undetected guides (RPM below 0.1) were excluded from further calculations. Fold change with respect to 'before infection' was calculated by dividing RPM in 'infected' or 'bystanders' conditions by RPM in 'before infection' condition. The differential abundance of a guide is represented as the log2 ratio of fold change in 'infected' condition divided by the fold change in 'bystanders' condition. If less than two screens call a guide detected (RPM >= 0.2), a log2FC of 0 and p-value of 1 are reported for this guide. Otherwise, the final log2FC of the guide is the arithmetic mean of the log2 ratios from each detected screen, and the final p-value of the guide is calculated by one sample t-test that the log2 ratios of the detected guides was not zero. The GeCKO library contains 6 independent guides for each protein-coding gene. The log2FC and p-value at the gene level is calculated from log2FC and p-value of its 6 guides. The log2FC of a gene is equal to the log2FC of the guide with the lowest (best) p-value. The corrected p-value of a guide is set to 1 if the sign of its log2FC is opposite to the log2FC of the gene. Then the p-value of a gene is calculated as the product of corrected p-values from all guides not excluded from calculations.
Gene set enrichment analysis on all genes with positive/negative log2FC was performed based on major knowledgebases including HUGO Gene Nomenclature Committee

Immunofluorescence
For imaging experiments, HepG2-CD81 wild type or knockout cells were plated in 8 well chamber slides (Labtek) and infected with P. yoelii sporozoites. Cells were fixed with 3.7% Approximately 20-30 slices were acquired per image stack. For deconvolution, the 3D data sets were processed to remove noise and reassign blur by an iterative Classic Maximum Likelihood Estimation widefield algorithm provided by Huygens Professional Software (Scientific Volume Imaging BV, The Netherlands). For the highthroughput secondary screen, cells were plated onto 96 well plate, infected and stained as explained above Images were acquired using Keyence BZ-X800 automated microscope and infection rate were quantified using Imaris 9.5, image analysis software.

Meta-analysis of screens
Z-scores of positively and negatively represented genes in each screen were calculated separately. Meta ranking was performed by function metaRank() from the R package DuffyTools, (using arguments: mode="percentile", rank.average.FUN=mean, naDropPercent = 0.75). Positively represented genes with z-scores greater than the cutoff (1.0, 1.5, 2.0) and negatively represented genes with z-scores smaller than the cutoff (-1.0, -1.5, -2.0) were selected as hits for each screen. Hits of positively and negatively represented genes were combined for further pathway enrichment analysis. To compare datasets of uneven sizes, gene rank percentiles were assigned to positively and negatively represented genes in each screen separately. Genes were ranked by the average percentiles across all datasets where they were screened.

Gene Ontology analysis on integrated forward genetic screens
Identified hit genes from all the four screens were uploaded in ious combinations in the ClueGO plug-in (version 2.3.3), implemented in Cytoscape v3.4.0 (http://cytoscape.org/) to generate gene ontology (GO) and pathway enrichment networks. Enriched functionally annotated groups were obtained with the following setting parameters: organism was set to Homo sapiens; the total gene set used in each of the screen were used as reference; the gene ontology terms were accessed from the following ontologies/pathways: Biological Process and Reactome Pathway database evidence code was restricted to 'All_without_IEA'. The GO fusion option was also selected. The significance of each term was calculated with a two-sided hypergeometric test corrected with Benjamini-Hochberg correction for multiple testing. The kappa score was set to 0.5 and the GO tree levels were restricted at 6-16 (medium-detailed specificity). For GO term selection, a minimum of 3 genes and 3% coverage of the gene population was set. GO terms were grouped with an initial group size of 2 and 50% for group merge. The remaining parameters were set to defaults.    HepG2-CD81 knockouts with significantly different entry, development or growth rates than the scrambled control. * Statistically significant at p-value < 0.05.