Automated Genome-Wide Visual Profiling of Cellular Proteins Involved in HIV Infection

Abstract

Recent genome-wide RNAi screens have identified >842 human genes that affect the human immunodeficiency virus (HIV) cycle. The list of genes implicated in infection differs between screens, and there is minimal overlap. A reason for this variance is the interdependence of HIV infection and host cell function, producing a multitude of indirect or pleiotropic cellular effects affecting the viral infection during RNAi screening. To overcome this, the authors devised a 15-dimensional phenotypic profile to define the viral infection block induced by CD4 silencing in HeLa cells. They demonstrate that this phenotypic profile excludes nonspecific, RNAi-based side effects and viral replication defects mediated by silencing of housekeeping genes. To achieve statistical robustness, the authors used automatically annotated RNAi arrays for seven independent genome-wide RNAi screens. This identified 56 host genes, which reliably reproduced CD4-like phenotypes upon HIV infection. The factors include 11 known HIV interactors and 45 factors previously not associated with HIV infection. As proof of concept, the authors confirmed that silencing of PAK1, Ku70, and RNAseH2A impaired HIV replication in Jurkat cells. In summary, multidimensional, visual profiling can identify genes required for HIV infection.

Introduction

Human immunodeficiency virus (HIV) subcontracts a large part of its life cycle to human host factors. With only 15 proteins encoded in the HIV genome, the host cell and viral life cycles are closely intercalated: Viral entry, reverse transcription and nuclear entry, integration into the host genome, transcription and posttranscriptional processing, and assembly and budding of viral particles recruit a multiplicity of host factors to accomplish their tasks.^1,2 How closely the viral life cycle depends on host cell mechanisms has been demonstrated in four elegant studies.^3–6 These studies identified nearly 1000 host genes necessary for the HIV life cycle in genome-wide RNA interference (RNAi) screens, many of which are closely connected to host cell housekeeping functions. For instance, the authors showed that HIV infection requires intact nuclear pore complexes, with nucleoporins Nup85, Nup107, Nup133, and Nup160 featuring prominently, just as it requires nuclear transport factors, or the rab6-dependent vesicular transport machinery. The interdependence of the viral life cycle on normal cellular housekeeping functions makes it a sensitive indicator of any cellular malfunction, rendering the distinction between direct and indirect effects on the viral life cycle difficult. For example, nuclear pore complex factors and karyopherins are positioned at a central junction between cellular transcription and translation and affect innumerable cellular processes, as does the vesicular transport machinery. Tinkering with these cellular functions may therefore block the viral life cycle without directly touching upon any therapeutically important/specific host–virus interaction—for example, by generally slowing down transcription or translation, by blocking ribosomal protein import, or by impairing the nuclear export of mRNA or ribosomal subunits. This potential of indirect or pleiotropic effects on viral life cycle may contribute to the high number of host genes identified and render standard analyses refractory to identifying cellular targets for antiviral therapies.

Materials and Methods

Chemicals

Chemicals were purchased from Sigma-Aldrich (St. Louis, MO). DRAQ5 was from BioStatus (Leicestershire, UK). siRNA duplexes were from Dharmacon (Lafayette, CO). The siRNA library comprised 1.0 nM of the Dharmacon siARRAY whole human genome siRNA library (Thermofisher, West Lafayette, CO) containing 84 508 siRNAs corresponding to four unique siRNA duplexes targeting each of 21 127 unique human genes. Primary antibodies were from Santa Cruz Biotechnology (Santa Cruz, CA), and all fluorescent secondary antibodies were from Molecular Probes/Invitrogen (Carlsbad, CA). Transfection reagents were from commercial sources.

Cell lines and cell culture

Long terminal repeat (LTR)–green fluorescent protein (GFP) HeLa CD4+ cells (a gift from A. Boese, IP Korea) were produced as described in Bakal et al. ⁷ Wild-type HeLa (ATCC, Manassas, VA) and GFP-torsin expressing HeLa (a gift from R. Graihle, IP Korea) were cultivated in high glucose glutamax Dulbecco’s modified Eagle’s medium (Invitrogen) supplemented with 110 mg/mL sodium pyruvate, 10% fetal calf serum (FCS; Gibco, Carlsbad, CA), and 1% penicillin streptomycin (Invitrogen). Jurkat E6-1 clones were cultured in RPMI medium 1640 (Invitrogen) supplemented with 10% FCS (Gibco), 1% penicillin streptomycin (Invitrogen), 1 mM sodium pyruvate (Gibco), and 10 mM HEPES. Cell lines were cultivated on arrays for 12 to 72 h for quantifying reverse transfection. For HIV infection, 650 000 LTR-GFP HeLa CD4+ cells were seeded per array (24 × 60 mm) and cultivated in Opti-MEM (Invitrogen), 5% FCS (Gibco), and 1% penicillin streptomycin (Invitrogen) for 28 h. Cells were infected with HIV-1, strain IIIB virus (Daymoon Industries, Cerritos, CA), multiplicity of infection (MOI) = 0.14, and inoculated for 3 h. After the viral supernatant was removed, cells were cultivated in Opti-MEM (Invitrogen) supplemented with 5% FCS (Gibco) and 1% penicillin streptomycin (Invitrogen) for 45 h. Arrays were fixed in 4% (w/v) paraformaldehyde in Dulbecco’s phosphate-buffered saline (PBS) and stained with 2.5 µM Draq5 before imaging.

RNAi array printing

For preparing the reverse transfection solution for array printing, 5 µL of 20 µM siRNA (Thermofisher) was transferred into a 96-well plate. Then, 8 µL of Red mixture, which contains 20 µM Red siGLO (Thermofisher), 5 µL of 1.6 M sucrose dissolved in RNase-free water, and EC buffer (Qiagen, Valencia, CA), was added. Next, Effectene (Qiagen) was added and mixed thoroughly. The mixture was incubated for 20 min at room temperature (RT). Then, 5 µL of 1.6% (w/v) gelatin was added and mixed and printed as 3888 spot arrays (108 × 36 spots) on glass coverslips using SMP9 stealth pins (Telechem, Atlanta, GA) and a high-throughput microarray printer (Genomic Solutions, Ann Arbor, MI) at 22 to 25 °C, 55% to 65% relative humidity (RH), enclosed in a custom-built clean chamber providing a sterile HEPA-filtered atmosphere. Arrays were stored in a desiccator with no significant alterations in performance from 1 week to 18 months postprinting. Seven slides covered the genome and contained 16% of control siRNA spots.

RNAi array acquisition

Arrays were acquired with a point scanning confocal reader (ImageXpress Ultra; Molecular Devices) as 16-bit TIFF files written directly to an external database.

RNAi array image analysis

Images were read directly from the database for analysis using software designed for this purpose. For automated identification and reconstruction of siRNA spots, a miniature image was created from the three-channel 16-bit images acquired for one array by reducing all tiles to a composite image of 3/100 the original scale. We applied an automated grid-fitting algorithm to identify siRNA spots in the entire array. Locations of spots obtained by grid fitting were used to collect pieces of high-resolution pictures to obtain an annotated spot image. The Draq5 intensity and GFP intensity of each cell were retrieved as the integrated intensity of pixels over a disk of a fixed-size radius of 5 pixels around the maximas. We used Delaunay triangulation and the length and intensity along the links between nuclei to measure syncytia formation and cell dispersion (see Table 1 ). Finally, 15 descriptors were retrieved for analysis.

Table 1.

Detailed List of the 15 Image Descriptors

	Name	Description	Meaning
1	cellNumber	Number of cell	A low number indicates cell death.
2	linkMinGFPAvg	Average of the minimum value of green fluorescent protein (GFP) along the line going from a cell to its neighbor	A high value indicates the cell is in a syncytia formation.
3	linkMinGFPSdtdev	Standard average of the minimum value of GFP along the line going from a cell to its neighbor	A high value indicates a mix of fused cells in syncytia formation and nonfused cells.
4	linkLengthAvg	Average of the distance between two neighbor cells	A high value means cells are spread.
5	linkLengthSdtdev	Standard deviation of the distance between two neighbor cells	A high value means an odd dispersion of cells (often happens in the case of cell death).
6	inTotalGFP	Integration of GFP over the image inside the boundaries of the precisely fitted experiment	A high value indicates infection measured on the spot.
7	inIntNucleiAvg	Average intensity of cell nuclei inside the boundaries of the precisely fitted experiment	An irregular value means cell toxicity measured on the spot.
8	inIntNucleiSdtdev	Standard deviation of intensity of cell nuclei inside the boundaries of the precisely fitted experiment	A high value means cell toxicity measured on the spot.
9	inIntGFPAvg	Average intensity of GFP per cell inside the boundaries of the precisely fitted experiment	A high quantity of reporter per cell indicates infection measured on the spot.
10	inIntGFPSdtdev	Standard deviation of GFP per cell on the precisely fitted experiment	A low value indicates a well-spread infection measured on the spot.
11	outTotalGFP	Integration of GFP over the image around the boundaries of the precisely fitted experiment	A high value indicates infection on the border around the spot.
12	outIntNucleiAvg	Average intensity of cell nuclei around the boundaries of the precisely fitted experiment	An irregular value means cell toxicity measured on the border around the spot.
13	outIntNucleiSdtdev	Standard deviation of intensity of cell nuclei around the boundaries of the precisely fitted experiment	A high value means cell toxicity measured on the border around the spot.
14	outIntGFPAvg	Average intensity of GFP per cell around the boundaries of the precisely fitted experiment	A high quantity of reporter per cell indicates infection measured on the border around the spot.
15	outIntGFPSdtdev	Standard deviation of GFP per cell around the boundaries of the precisely fitted experiment	A low value indicates a well-spread infection measured on the border around the spot.

RNAi array data analysis

Once images of every siRNA spot for seven genome-wide screens were analyzed, we obtained seven sample values for each of the 15 descriptors and for each spot and more than seven sample values for all controls as CD4, SCRAMBLED, GFP, XPO1, and so on. After normalization of the arrays, we removed the low-frequency signal produced by the difference in density across the cell layer to compare the sample distributions of each siRNA response to our positive control CD4. A statistical test (described later in Fig. 2 ) was applied to select the siRNA that showed a similar profile to CD4 given the metric we defined.

Production of lentiviral vectors and transduction of target cells with shRNA

A set of five shRNA-expressing plasmids targeting PKN-1 and six plasmids for Ku70 (Open Biosystems, Huntsville, AL) were tested. Transfection of these plasmids results in the synthesis of RNA that can be packaged into lentiviral particles. Virions are then used to transduce target cells, where shRNA will be expressed. These vectors also carry the puromycin resistance gene, allowing for selection of transduced cells.

To produce lentiviral particles, each shRNA expressor plasmid was transfected in 293-T cells, together with an HIV Gag-Pol expressor (pCMVdR8.2) and a VSV-G coding plasmid (pHCMV-G). ⁸ Forty-eight hours after transfection, virions produced in the supernatant were aliquoted and stored at −80 °C.

Jurkat cells (3 × 105/mL) were incubated with shRNA lentiviral vectors (40 ng of p24) for 4 h at 37 °C and gently shaken every hour. Cells were then washed once with PBS and cultured in RPMI (10% FCS, penicillin/streptomycin). Forty-eight hours after cell transduction, puromycin (1 µg/mL) was added to the culture’s medium and maintained thereafter. Resistant cultures were obtained within 1 to 2 weeks, after which the concentration of puromycin was lowered to 0.5 µg/mL.

Western blot analysis

Down-modulation of the expression of targeted proteins was assessed by Western blot analysis. Transduced Jurkat cells were lysed, and 40 µg of lysate protein was loaded in sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and immunoblot. For PKN-1, the primary antibody was PKN C-19, and the blot was revealed with a donkey anti-goat–horseradish peroxidase (HRP) antibody. For Ku70, the primary antibody was a mouse monoclonal to Ku70 (Abcam, Cambridge, UK), and the blot was revealed with a sheep anti-mouse-HRP antibody (Amersham ECLTM, GE Healthcare, Buckinghamshire, UK). Cultures in which more than 80% down-modulation was observed were further characterized to assess their susceptibility to HIV infection.

HIV replication in shRNA-transduced Jurkat cells

shRNA-transduced Jurkat cells were plated in 96-well plates (2 × 10⁵ in 200 µL/well) and exposed to HIV (clone NL4-3). Infections were performed with 0.1 and 1 ng p24/well (corresponding to low and high MOI, respectively) for 4 h. Cells were then washed and cultured in RPMI (10% FCS, penicillin/streptomycin) supplemented with 0.5 µg puromycin. The percentage of HIV Gag-expressing cells in culture was monitored every 2 to 3 days by permeabilization and intracellular staining with anti-Gag p24 phycoerythrin mAb (KC57; Beckman Coulter, Brea, CA). An isotype-matched mAb was used as a negative control.

RNA isolation and RT-PCR

Total RNA was isolated from siRNA-transfected Jurkat cells using the Trizol method (Invitrogen). cDNA was produced using 1 µg total RNA and MMLV–reverse transcriptase (Promega, Madison, WI) in 25-µL reaction mixtures in the presence of 50 pmole oligo (dT) primer and 20 µM dNTP mixture for 60 min at 37 °C. For PCR amplification, specific oligonucleotide primer pairs (0.2 pmole each) were incubated with 200 ng cDNA, 1 unit of LA Tag polymerase (Takara, Madison, WI), 1× LA PCR buffer 2 (2.5 mM MgCl₂), and 100 µM dNTP in 25-µL reaction mixtures. The sequences of primers used in this study were as follows:

• RNAseH2A sense primer 5′-GACCCTATTGGAGAGCGAGC-3′ and RNAseH2A antisense primer 5′-GTCTCTGGCATCCCTA CGGT-3′

• GAPDH sense primer 5′-TGATGACATCAAGAAGGTGGTGAA G-3′ and GAPDH antisense primer 5′-TCCTTGGAGGCCAT GTGGGCCAT-3′

PCR conditions were 95 °C for 30 s, 54 °C for 30 s, and 72 °C for 3 min, for a total of 40 cycles. The PCR products were applied onto the 1% agarose gel and visualized by ethidium bromide staining.

Forward siRNA transfection and virus infection

Jurkat cells (40 000 cells/well) were transfected in 24-well plates with 1 µM Accell siRNA (Dharmacon) against selected individual RNAseH2A or nontargeting siRNA and incubated for 72 h. Cells were infected with HIV-1 strain IIIB virus (Daymoon Industries) at MOIs of 0.5 or 0.01 for 3 h. After the viral supernatant was removed, cells were cultivated in RPMI medium 1640 (Invitrogen) supplemented with 10% FCS (Gibco), 1% penicillin streptomycin (Invitrogen), 1 mM sodium pyruvate (Gibco), and 10 mM HEPES for 96 h. Viral release was determined by detection of p24 HIV-1 viral core antigen in cell-free supernatants by a p24 enzyme-linked immunosorbent assay (ELISA; PerkinElmer, Waltham, MA).

Results

Establishing visual profiling on array-based imaging screens

We developed a high-content, genome-wide RNAi screen ( Figs. 1 and 2 ) to identify host genes that block the HIV infection without otherwise affecting cellular growth or viability. As a reference RNAi experiment, we chose the effect of CD4 receptor silencing on HIV infection. Although essential for viral entry, the depletion of CD4 does not significantly affect cell growth or viability. To identify other human genes, we sought those that reproduce the CD4-phenotype when silenced. The approach relied on identifying sufficient visual parameters to reliably recognize the specific imaging profile of CD4-silencing inhibition of HIV infection in a genome-wide set of RNAi experiments. Visual profiling is complicated by each RNAi-induced knockdown experiment having specific functional and morphological effects on host cells. Accordingly, we selected descriptors a priori to distinguish the CD4 profile from a genome-wide, heterogeneous population of individual visual profiles. We then demonstrated that this set of descriptors formed a metric that distinguished between CD4 and SCRAMBLED profiles (see Fig. 2 ) as well as from GFP or XPO1 profiles. The choice or weights of those descriptors could not be obtained automatically because there is no pre-defined knowledge about the phenotypic variations that each of the 21 000 gene knockdowns would produce on cell phenotype.

Fig. 1.

(A) Identification and extraction of 3888 spot confocal images per array. (B) Image analysis of each spot image extracting 15 descriptors. Pictures show the analysis of syncytial networks through a Delaunay triangulation of cell nuclei for a CD4 spot and for a SCRAMBLED spot. Yellow square shows the automated computation of the precise localization of an siRNA spot defining the inside (see “in” descriptors, with a high probability of transfection over the spots) versus the outside (“out” descriptors, with a lower probability of transfection around the spots) and retrieving cell number, nuclei intensity, green fluorescent protein (GFP) intensity, minimum intensity on the links between cells, distance between cells, and their respective standard deviations along with the total GFP (15 descriptors in total; see Table 1 for a detailed list). (C) All seven genome experiments and controls (190 000 experiments) projected into a 15-dimensional space. Each point is an experiment projected from the 15 dimensions to the three most discriminant axes between CD4 and SCRAMBLED distributions (DFA). A subset of the various phenotypes preventing a simple uni- or bidimensional threshold selection of the hits is shown here, highlighting an issue in RNAi screens.

Fig. 2.

(A) Projection of CD4 and SCRAMBLED distributions from 15 dimensions to the two most discriminant factors. These two 15-variate distributions are well resolved, with a Wilks’ lambda of 0.66 (a value under 0.9 means two different distributions). (B) Selection of all experiments falling in the 15-dimensional CD4 distribution (in red). The projection is on the three most discriminant factors. (C) Calculation of the density ratio of each gene to fall into a CD4 distribution. A randomly chosen experiment has a 3% (= 3/100) chance to fall in a CD4 distribution, whereas CLDN19, for example, has a 66% (= 4/6) chance. The density ratio for CLDN19 is therefore 0.05 (= 0.3/0.66), that is, as all of the selected hits, well under 1 (= 0.03/0.03, the density of a randomly chosen experiment). An equivalent, less intuitive selection can be obtained by computing the p-value of a binomial distribution between randomly chosen and CD4-like chosen genes. (D) Examples of hit images found and their corresponding density ratio. First column (in red) shows the siRNA spot, second column (in blue) shows the nuclei stain, third column (in green) shows the green fluorescent protein (GFP) reporter, and the last column shows the merged image for MED28, Ku70, RNAseH2A, and SCRAMBLED siRNA.

We used 15 imaging parameters ⁷ to describe HeLa-CD4+ LTR-GFP cells infected with HIV-1 (see Table 1 and Suppl. Figs. S1 and S2). HIV infection enables TAT-driven transactivation of the stably integrated GFP and thus recapitulates early steps in viral infection. ⁹ These parameters included cell number, relative cell distribution in an area, syncytia formation, GFP reporter intensity, and others ( Table 1 ).

To investigate this multidimensional imaging space on a genome-wide scale, we developed a visual array screening system based on cellular microarrays^10–17: siRNAs spotted onto a glass wafer were reverse transfected into cells and then imaged with an automated confocal microscope (Suppl. Fig. S1 and Fig. 1A,B). Dedicated imaging software was developed to identify and annotate siRNA spots on the high-resolution confocal images of entire arrays (Fig. 1A,B). This enabled the rapid, reliable, and relatively inexpensive screening of multiple, independent genome-wide RNAi knockdown screens.

Multidimensional visual profiling can distinguish a CD4-dependent HIV infection block from indirect or unspecific RNAi effects

To test the robustness of multidimensional image profiling, we analyzed whether the 15 visual parameters reliably distinguished a CD4-like HIV infection phenotype from a background phenotype of cells reverse transfected with scrambled siRNA.

We reverse transfected Hela-CD4+ LTR-GFP cells with siRNAs and then infected the cells with HIV-1 (MOI 0.14) for 48 h. Under these conditions, HIV infection was significantly repressed in cells transfected with a CD4 siRNA ( Fig. 1B ). We acquired the 15 visual parameters from 215 individual CD4 depletion experiments and compared these with 3896 scrambled experiments.

To measure virally induced syncytia formation as well as cell dispersion, the Delaunay triangulation ¹⁸ was computed from the set of all nuclear centers to establish a unique neighborhood map linking cells. The minimum value of GFP for unique nucleus–nucleus links was retrieved and used to discriminate syncytial fused from unfused cell clusters (see Suppl. Fig. S2). The Delaunay triangulation also gave the Voronoi diagram, ¹⁹ which was used to separate densely packed nuclei. The algorithms produced a 15-dimensional vector for each experiment/spot. Because both the CD4 and the non-targeting siRNA populations were 15-variate Gaussian distributions, we set a classifier using the Mahalanobis distance. The results of this approach are shown in Figure 2 , with the “signature” of a CD4-dependent HIV phenotype being clearly distinct from the 15-dimensional representation of the SCRAMBLED population.

To further validate the visual profiling approach, we analyzed whether a multiparametric analysis was sufficient to distinguish effects on virus–host interactions from indirect, pleiotropic, or false-positive effects. In its most skewed representation, a housekeeping RNAi phenotype could block HIV reporter expression by generally slowing transcription efficiency, without otherwise disturbing infection. This was simulated through the GFP-RNAi-based knockdown of the LTR-driven GFP, which significantly knocks down HIV reporter levels without affecting HIV infection. Intriguingly, when comparing the visual profiles of CD4-depleted, HIV-infected cells with those transfected by GFP siRNA, 15-dimensional distributions for both populations were strikingly different (Wilks’ lambda: 0.685 << 0.9). Neither of the two populations revealed any detectable expression of the LTR-GFP reporter, but only in the case of CD4 knockdown was a lack of syncytia formation measured. Therefore, the visual profiling approach can apparently discern the additional morphological symptoms imprinted on the host cell by infection in CD4 knockdown cells. This is a crucial result, as it allowed the exclusion of false positives embedded in the transcriptional machinery of infected host cells.

To scrutinize the type of housekeeping functions identified, we analyzed the visual profile induced by knockdown of a core housekeeping gene. As the nuclear transport pathway has been represented through several karyopherin and nucleoporin hits in previous HIV screens,^3,4,20 we used the exportin XPO1 as an example housekeeping gene. Exportin1 is an essential nuclear export factor and plays a role in the export of the unspliced viral HIV mRNA during the late viral replication cycle. ²¹ Knockdown of XPO1 blocks protein transport by stopping ribosomal subunit export and stopping import factor recycling. The nuclear transport block severely impairs cell growth and, as a consequence, HIV replication. The underlying aim of this “housekeeping” experiment was to establish whether the 15-dimensional imaging measurement had sufficient parameters to identify differences between XPO1- and CD4-depleted cells caused by the nuclear transport block, such as a change in nucleus volume, cell size, or, eventually, cell density.

The image analysis did not identify XPO1 as part of the CD4 population (XPO1 vs CD4 distribution had a Wilks’ lambda of 0.540 << 0.9). The ability of the analysis to distinguish the housekeeping defect may be based on one of the 15 parameters, such as nucleo-cytoplasmic intensity ratio, or a cumulative combination of minor effects on several parameters.

A further comparison between CD4 and rab9a knockdown, known to play a role in the HIV life cycle, showed that visual profiling excluded genes involved in the late replication cycle, as would be expected in our experimental setup (rab9a didn’t show significant 15-dimensional distribution difference with scrambled, data not shown). However, there was also gene knockdown showing a distinct profile of CD4 from a scrambled siRNA background, but which nonetheless does not fit the profile of CD4. One such example given here is p65, a major transcriptional regulator. We decided not to follow up on that factor group as the goal of this study was not to identify a maximum number of human HIV cofactors but rather to be as precise as possible in identifying genes that reproduce a CD4-like replication block when depleted. The controls described above show that the principle of 15-dimensional visual profiling can effectively achieve this task.

Fifty-six genes reproducing the visual profile of a CD4-dependent replication block

Seven complete human genomes in siRNA were cultivated for 28 h with the LTR-GFP HeLa CD4+ cell line to permit host gene silencing. Arrays were then infected with HIV-1 (MOI of 0.14) for 3 h, washed, and further incubated for 45 h prior to imaging. Once the arrays were imaged, the identity of each spot was retrieved, and HIV infection was independently analyzed on each spot using the following image and data analysis strategy.

The array is a large experiment with a low-frequency spatial variation due to imperfect cell density across its surface. Given variations in cell culture, the measurements of infection were normalized across the arrays. We used a median filter with a radius size of 3 spots to filter result arrays (see Suppl. Fig. 2E). This normalization produced relative values that were comparable for all measurements across the seven genomes, irrespective of small changes in tissue culture.

We computed the 15-dimensional Mahalanobis metric relative to CD4 and scrambled classes for all 190 512 data points. We selected the points closer to CD4 than scrambled and simultaneously with a distance to the CD4 class center inferior to the square root of a given G, which is determined from a χ² (with 15 degrees of freedom) as corresponding to a gating probability chosen to be at least 0.99. This identified 1680 experiments (0.8%) as potentially similar to CD4.

Because we performed seven independent genome-wide screens, we computed a density score for each siRNA within the CD4 class. For each gene, the ratio between the percentages of experiments inside the CD4 class versus outside was computed. All genes less than a score of 1 were considered to have an abnormally high representation in the CD4 class, and the lower the score, the stronger that representation. For example, CD4 has a score value of 0.043, which means it is 23 times (1/0.043) denser inside the CD4 class than outside. A score lower than 0.1 means a density at least 10 times higher inside the CD4 class than outside ( Fig. 2 ). Fifty-six unique genes were identified ( Table 2 ), of which 49 had a ratio under 0.1, demonstrating an overrepresentation in the CD4 class that cannot be explained by chance ( Fig. 2 ). Of these genes, 45 were identified as novel in terms of their involvement in HIV infection (see Table 2 ). The remaining 11 genes (MED28, CXCR4, TKLT2, JMY, GRWD1, UBE2D3, PRPS1, PRPSAP2, SNRPD1, SLC40A1, and UNG2) have been previously shown, either directly or indirectly, to be involved in HIV infection (see Table 2 for description and Table 3 indicating which of the 56 hits are expressed in physiologically relevant HIV target cells).

Table 2.

List of Hits, Their Function, and Direct Known or Putative Interaction with HIV-1

Gene	Symbol	Function	Score	Direct Interaction	Putative Interaction	PMIDs
Mediator of RNA polymerase II transcription subunit 28	MED28	Transcription regulation	0.04		Via MERLIN/NEF	18187620
Claudin 19	CLDN19	Sensory transduction calcium-independent cell–cell adhesion	0.05
T cell surface glycoprotein CD4 precursor	CD4		0.05	HIV receptor
Netrin 4	NTN4	Regulator of neuronal and vascular growth and epithelial cell branching morphogenesis	0.06
GPR158-like 1	GPR158L1	G-protein-coupled receptor	0.06
DnaJ (Hsp40) homolog, subfamily A, member 2	DNAJA2	Modulation of G-protein signaling	0.06
Neurturin	NRTN	Member of the glial cell line–derived neurotrophic factor family	0.06
Chemokine (C-X-C motif) receptor 4	CXCR4	CXCL12 (SDF-1) chemokine receptor	0.06	HIV coreceptor
S100 calcium binding protein A7	S100A7	Member of S100 superfamily of EF-hand Ca2(+)-binding proteins	0.06
Hypothetical gene supported by AK057338	LOC401061	Hypothetical gene supported by AK057338	0.06
Hypothetical LOC401463	LOC401463	Hypothetical LOC401463	0.06
Aging-associated gene 5 protein	AGPS	Aging-associated gene 5 protein	0.06
FLJ32975	TKLT2	Transferase	0.06		Via NFIC	16189514/8628270
Junction-mediating and regulatory protein	JMY	Transcription coactivator activity, transcription regulation	0.06		Via EP300	10518217/11080476
KIAA1007 protein	CNOT1	Transcription complex, subunit	0.06
Helicase-like protein HFM1	HFM1	DNA binding	0.07
PI4KAP2 protein	PI4KAP2	Phosphatidylinositol 4-kinase, catalytic, α pseudogene 2	0.07
Ligand-dependent corepressor LCoR Mblk1-related protein 2	LCOR	Ligand-dependent corepressor	0.07
Cytosolic 5′-nucleotidase 1A	NT5C1A	Nucleoside metabolic process	0.07
Zinc finger protein 324A	ZNF324	Zinc finger protein	0.07
Aflatoxin B1 aldehyde reductase member 4	AFAR3	Aldo-keto reductase	0.08
Acyloxyacyl hydrolase (neutrophil)	AOAH	Hydrolase	0.08
Thioredoxin domain-containing protein 15 precursor	C5ORF14	Unknown	0.08
Uncharacterized protein C7orf10 dermal papilla-derived protein 13	C7ORF10	Transferase	0.08
FK506-binding protein 11 precursor	FKBP11	FK506 binding protein	0.08
Protein FAM118B	FAM118B	Unknown	0.08
CDNA: FLJ22490 fis, clone HRC10983	CSPP1	Centrosome and spindle pole–associated protein 1	0.08
Highly divergent homeobox	HDX	Unknown	0.08
Endoplasmic reticulum resident protein ERp27 precursor	ERP27	Endoplasmic reticulum	0.08
UDP-glucuronosyltransferase 3A1 precursor	UGT3A1	UDP-glucuronosyltransferase 3A1 precursor	0.08
Solute carrier family 25 member 35	SLC25A35	Transporter	0.08
Putative uncharacterized protein CXorf55	CXORF55	Unknown	0.08
Adenosine triphosphate (ATP)–dependent DNA helicase II 70-kDa subunit	Ku70	Positive regulation of transcription, DNA dependent	0.08			11406603
Glutamate-rich WD repeat-containing protein 1	GRWD1	Glutamate-rich WD repeat-containing protein 1	0.08		Via CUL4A	17620334
GSTTP1 protein	GSTTP1	Unknown	0.08
RPL32P3 protein	RPL32P3	Ribosome	0.08
Semaphorin-3G precursor	SEMA3G	Semaphorin	0.08
Coiled-coil domain-containing protein 64A	CCDC64	Unknown	0.08
Protein FAM64B	FAM64B	Unknown	0.08
Solute carrier family 25 member 12	SLC25A12	Transporter	0.08
Ubiquitin-conjugating enzyme E2 D3	UBE2D3	Ubiquitin-protein ligase activity	0.08		Via NEDD4	9990509/15013426
Zinc finger protein 143	ZNF143	Zinc finger protein	0.08
Zinc finger protein 333	ZNF333	Unknown	0.08
Nuclear fragile X mental retardation protein interacting protein 2	NUFIP2	RNA binding	0.09
Pyruvate dehydrogenase complex E2 subunit	DLAT	Mitochondrion	0.09
Protein kinase PKN-α	PKN1	Activation of JNK activity protein amino acid phosphorylation	0.09			16352537
Ribose-phosphate pyrophosphokinase 1	PRPS1	Ribose-phosphate pyrophosphokinase 1	0.09		Via Maspardin/CD4	16189514/11113139
Phosphoribosyl pyrophosphate synthetase-associated protein 2	PRPSAP2	Nucleotide biosynthesis	0.09		Via PRPS1	9545573
Ribonuclease H2 subunit A Aicardi-Goutières syndrome 4 protein	RNASEH2A	DNA replication	0.09		Aicardi-Goutières syndrome	16845400/19525956
Small nuclear ribonucleoprotein Sm D1	SNRPD1	mRNA processing and splicing	0.09	TAT		11780068
Transcription elongation regulator 1–like protein precursor	TCERG1L	Transcription elongation regulator 1–like protein precursor	0.11
Prostacyclin synthase	PTGIS	Lipid metabolic process	0.11
KIAA1979 protein	ZNF525	Unknown	0.12
Chemokine (C-C motif) ligand 2	CCL2	Ligand	0.14
Solute carrier family 40 member 1 ferroportin-1	SLC40A1	Iron ion transmembrane transporter activity	0.15	Iron homeostasis		16043695
Zinc finger protein 44	ZNF44	Transcription regulation	0.15
Uracil-DNA glycosylase 2	UNG2	DNA repair	0.16	VPR		15096517/15721252

Italicized entries = known as involved in HIV replication. Bold entries = secondarily tested in Jurkat cells in this study.

Table 3.

List of Hits Expressed in T Lymphocytes, Macrophages, or Dendritic Cells

Symbol	Expression in T Lymphocytes, Macrophages, or Dendritic Cells	Reference
MED28/MAGICIN	Yes (in Jurkat T cells after CD3 stimulation)	Lee, M. F.; Beauchamp, R. L.; Beyer, K. S.; Gusella, J. F.; Ramesh, V. Biochem. Biophys. Res. Commun. 2006, 348(3), 826–831.
CLAUDIN19	No
CD4	Yes
NETRIN 4	No	Staquicini, F.; Dias-Neto, E.; Li, J.; Snyder, E. Y.; Sidman, R. L.; Pasqualini, R.; Arap, W. Proc. Natl. Acad. Sci. 2009, 106(8), 2903–2908.
GPR158L1	Unknown
DNAJA2	Unknown
NEURTURIN	Yes (dendritic cells)	Human Protein Reference Database (www.hprd.org)/Human Proteinpedia (www.humanproteinpedia.org)
CXCR4	Yes
S100A7	Yes (dendritic cells)	Human Protein Reference Database (www.hprd.org)/Human Proteinpedia (www.humanproteinpedia.org)
LOC401061	Unknown
LOC401463	Unknown
AGPS	Yes (Jurkat; dendritic cells)	Bantscheff, M.; Eberhard, D.; Abraham, Y.; Bastuck, S.; Boesche, M.; Hobson, S.; Mathieson, T.; Perrin, J.; Raida, M.; Rau, C.; et al. Nat. Biotechnol. 2007, 25(9), 1035–1044. Human Protein Reference Database (www.hprd.org)/Human Proteinpedia (www.humanproteinpedia.org)
TKTL2	No
JMY	Yes (Jurkat)	Shikama, N.; Lee, C. W.; France, S.; Delavaine, L.; Lyon, J.; Krstic-Demonacos, M.; La Thanque, N. B. Mol. Cell. 1999, 4(3), 365–376.
CNOT1	Yes (Jurkat and primary T cells)	Albert, T. K.; Lemaire, M.; van Berkum, N. L.; Gentz, R.; Collart, M. A.; Timmers, H. T. Nucleic Acids Res. 2000, 28(3), 809–817. Bantscheff, M.; Eberhard, D.; Abraham, Y.; Bastuck, S.; Boesche, M.; Hobson, S.; Mathieson, T.; Perrin, J.; Raida, M.; Rau, C.; et al. Nat. Biotechnol. 2007, 25(9), 1035–1044.
HFM1	No
PI4KAP2	Unknown
LCOR	No
NT5C1A	No
ZNF324	Yes (PBMC)	Rue, S. W.; Kim, B. W.; Jun, D. Y.; Kim, Y. H. Biochim. Biophys. Acta 2001, 1522(3), 230–237.
AFAR3	No
AOAH	Yes (macrophages)	Hagen, F. S.; Grant, F. J.; Kuijper, J. L.; Slaughter, C. A.; Moomaw, C. R.; Orth, K.; O’Hara, P. J.; Munford, R. S. Biochemistry. 1991, 30(34), 8415–8423. Ojogun, N.; Kuang, T. Y.; Shao, B.; Greaves, D. R.; Munford, R. S.; Varley, A. W. J. Infect. Dis. 2009, 200(11), 1685–1693.
C5ORF14	Unknown
C7ORF10	No
FKBP11	No
FAM118B	Unknown
HDX	No
ERP27	No
UGT3A1	No
SLC25A35	No
CXORF55	Unknown
Ku70	Yes (primary T cells)	Shi, L.; Qiu, D.; Zhao, G.; Corthesy, B.; Lees-Miller, S.; Reeves, W. H.; Kao, P. N. Nucleic Acids Res. 2007, 35(7), 2302–2310.
GRWD1	No
GSTTP1	Unknown
RPL32P3	No
SEMA3G	Unknown	Expressed in lymph node; Uhlén, M.; Björling, E.; Agaton, C.; Al-Khalili Szigyarto, C.; Amini, B.; Andersen, E.; Andersson, A.-C.; Angelidou, P.; Asplund, A.; Asplund, C.; et al. Mol. Cell Proteomics. 2005, 4(12), 1920–1932.
CCDC64	Unknown
FAM64B	Unknown
SLC25A12	No	Also called Aralar
UBE2D3	Yes (primary T cells and Jurkat)	Also called UBC4/5; Jensen, J. P.; Bates, P. W.; Yang, M.; Vierstra, R. D.; Weissman, A. M. J. Biol. Chem. 1995, 270, 30408–30414.
ZNF143	Yes (weakly in MOLT-4)	Human Protein Atlas (www.proteinatlas.org)
ZNF333	Unknown
NUFIP2	Yes (macrophages; MOLT-4)	Human Protein Atlas (www.proteinatlas.org)
DLAT	Yes (Jurkat and PBMC)	Bantscheff, M.; Eberhard, D.; Abraham, Y.; Bastuck, S.; Boesche, M.; Hobson, S.; Mathieson, T.; Perrin, J.; Raida, M.; Rau, C.; et al. Nat. Biotechnol. 2007, 25(9), 1035–1044. Human Protein Atlas (www.proteinatlas.org)
PKN1	Yes (primary T cell)	Miscia, S.; Ciccocioppo, F.; Lanuti, P.; Velluto, L.; Bascelli, A.; Pierdomenico, L.; Genovesi, D.; Di Siena, A.; Santavenere, E.; Gambi, F.; et al. Neurobiol. Aging. 2009, 30(3), 394–406.
PRPS1	Yes (dendritic cells)	Human Protein Reference Database (www.hprd.org)/Human Proteinpedia (www.humanproteinpedia.org)
PRPSAP2	Yes (dendritic cells)	Human Protein Reference Database (www.hprd.org)/Human Proteinpedia (www.humanproteinpedia.org)
RNASEH2A	Yes (Jurkat)	This study
SNRPD1	Yes (Jurkat)	Bantscheff, M.; Eberhard, D.; Abraham, Y.; Bastuck, S.; Boesche, M.; Hobson, S.; Mathieson, T.; Perrin, J.; Raida, M.; Rau, C.; et al. Nat. Biotechnol. 2007, 25(9), 1035–1044.
TCERG1L	Unknown
PTGIS	Yes (various T cell lines and monocyte/macrophage cell lines)	Human Protein Atlas (www.proteinatlas.org)
ZNF525	Unknown
CCL2	Yes (various T cell lines and monocyte/macrophage cell lines)	Owen, J. L.; Lopez, D. M.; Grosso, J. F.; Guthrie, K. M.; Herbert, L. M.; Torroella-Kouri, M.; Iragavarapu-Charyulu, V. Cell. Immunol. 2005, 235(2), 122–135. Human Protein Atlas (www.proteinatlas.org)
SLC40A1	Unknown
ZNF44	Yes	Human Protein Atlas (www.proteinatlas.org)
UNG2	Yes	Human Protein Atlas (www.proteinatlas.org)

PKN1 Ku70 or RNAseH2A depletion in Jurkat cells impairs HIV1 replication

To assess the significance of the screening results, it is important to demonstrate for some of the identified genes that they affect HIV replication in human T lymphocytes. Our aim was not to perform a systemic analysis of all identified proteins but rather to bring a proof of concept for the validity of our screening procedure.

We selected 3 genes from the list of these 56 identified genes: PKN1, Ku70, and RNAseH2A on HIV replication. PKN1 (also termed PAK1) is a kinase involved in various signaling pathways, whose implication in HIV replication has been reported. ²² The DNA-PK is a trimeric nuclear protein kinase consisting of a catalytic subunit and the Ku70/Ku80 complex that regulates its kinase activity. DNA-PK is a component of the dsDNA break repair pathway. Ku80 facilitates retroviral integration, ²³ but the role of Ku70 in this process remains largely unknown. The implication of RNAseH2A during HIV replication has not been studied so far. Mutations in the RNAseH2A gene have been shown to result in the neurological disorder Aicardi-Goutières syndrome (AGS). ²⁴

We used two methods to examine the effect of gene knockdown on HIV replication in Jurkat cells: shRNA-mediated silencing in stably transduced Jurkat cell lines and transient transfection of siRNA. For CD4, PKN1, and Ku70, up to six shRNAs in a lentiviral vector were transduced into Jurkat cells, and stable cell lines were produced. We selected the shRNAs that were more efficient in silencing the target proteins ( Fig. 3A ). Fluorescence-activated cell sorting (FACS) analysis showed that transduction by an shRNA against CD4 (CD4-16) resulted in a 75% reduction of CD4 expression from the cell surface ( Fig. 3A , left panel). Western blot analysis showed that Ku70 was downregulated by approximately 95% and 80%, by shKu70-612 and 611 respectively, whereas PKN1 was downregulated by more than 95% by shPKN485 ( Fig. 3A , right panels). Cell growth and viability were not significantly affected in the silenced populations (not shown). CD4 and CXCR4 surface expression were normal in Ku70- and PKN1-silenced cells. The selected cells were then infected with HIV at two MOIs, and viral replication was measured by following the appearance of Gag+ cells over time. In control cells, transduced with a vector expressing an irrelevant shRNA, the percentage of Gag+ cells reached 20% to 70% at day 9 post infection, for the low and high MOI, respectively. As shown in Figure 3B , at the low MOI, silencing of CD4, PKN1, or Ku70 significantly and comparably impaired HIV replication. The inhibition was less marked at the high MOI, reflecting probably the incomplete silencing achieved by the shRNAs. An analysis of three independent experiments indicated that viral replication was decreased by 60% to 70% in Ku70- or PKN1-silenced cells ( Fig. 3C ).

Fig. 3.

Ku70 and PKN1 down-modulation in Jurkat T cells impairs HIV replication. Jurkat cells were transduced with lentiviral vectors encoding shRNA against CD4, Ku70, or PKN1. As controls, cells transduced with an irrelevant shRNA (CTRL, directed against a murine protein) or nontransduced Jurkat cells (CTRL nt) were used. (A) Down-modulation of the targeted genes was assessed by fluorescence-activated cell sorting (FACS) analysis (CD4, left panel) or by Western blot analysis, using antibodies for PKN1 (upper right panel) and Ku70 (lower right panel). Molecular mass is indicated in kilodaltons. (B) HIV replication in transduced Jurkat cells was monitored over time by measuring the percentage of Gag+ cells in culture. Transduced cells were infected with a low and a high multiplicity of infection (MOI; upper and lower panels, respectively). One representative experiment is shown. (C) Mean and SD of three independent experiments comparing HIV replication in transduced Jurkat cells. Virus replication was followed as in B. The area under the curve (AUC) for shRNA-transduced cultures was then calculated and expressed as a percentage of control-transduced cells (CTRL).

We then studied the role of RNAseH2A in HIV replication. Lentiviral-mediated shRNA silencing did not lead to efficient downregulation of RNAseH2A (not shown). Transient transfection of Jurkat cells with RNAseH2A siRNAs gave robust silencing of RNAseH2A expression by RT-PCR ( Fig. 4A ). Seventy-two hours posttransfection, the Jurkat cells were infected with HIV (at MOI 0.1 and 0.05) for 96 h and viral p24 production measured by a cell supernatant ELISA. RNAseH2A silencing with two independent duplexes depressed viral infection by 50% to 60% ( Fig. 4B ) while not affecting cell growth or viability. Altogether, these experiments show that depletion of PKN1, Ku70, and RNAseH2A impairs HIV infection in Jurkat cells, confirming the visual profile results obtained in the genome-scale, HeLa-based screen.

Fig. 4.

RNAseH2A facilitates HIV replication. Jurkat cells were transfected with individual siRNA #1,#2 against RNAseH2A. As controls, cells transfected with nontargeted siRNA (CTRL, scramble siRNA) or nontransfected Jurkat cells (CTRL nt) were used. (A) RNAseH2A mRNA reduction by individual siRNA. Jurkat cells were transfected with the indicated siRNAs, and then cDNA was prepared and RNAseH2A mRNA expression levels were measured by RT-PCR. (B) Jurkat cells were transfected with the indicated siRNAs for 72 h and then infected with HIV-1 (strain IIIB) at a multiplicity of infection (MOI) of 0.05 (upper panel: low) and MOI of 0.1 (lower panel: high). At 96 h postinfection, p24 antigen was measured in cell culture supernatants by p24 enzyme-linked immunosorbent assay.

Discussion

During the past decade, the combined worldwide research effort on HIV and host virus interactions has yielded a list of about 25 host factors shown to play a crucial role in the viral infection. In contrast, in the past 3 years, several genome-wide analyses have identified multiple additional cofactors: 273 by Brass et al., ³ 295 by König et al., ⁶ 224 by Zhou et al., ⁴ and 252 by Yeung et al. ⁵ Of note, the study by Yeung et al. ⁵ used a relevant T cell line (Jurkat), whereas the previous screens relied on epithelial/fibroblast adherent cells. Although this impressive increase in scientific output shows the potential of genome-wide, systematic approaches, it also highlights its limitations: Very little overlap was found between the results. ² Notably, this cannot be explained by differences in assay procedures, which are based on similar principles. Furthermore, all groups have performed extensive work in following up on the verification of their genes employing independent approaches, and it is reasonable to assume that the identified genes are correct: All these genes do affect the HIV infection, directly or indirectly.

The meta-analysis by Bushman et al. ² revealed that among 842 genes, only 34 were represented in any two of those assays with an average overlap of 6%. We have also cross-examined our hit list with the results of each of those three screens and also obtained a low correlation finding CD4, MED28, CXCR4, ERP27, CSPP1, and DNAJA2 in common with at least one of those screens in our selected hits. This corresponds to 6% of the HIV human host factors we identified, indicating that our selection methods correlate as poorly with previous approaches as they correlate to one another. In addition, from the 34 genes represented in any two of those previous screens and identified by the Bushman et al. ² meta-analysis, 9 appeared in our analysis as having a density score inferior to 0.5. Those genes are MED28, CD4, CXCR4, CRSP2, CTDP1, TRIM55, JAK1, RANBP2L1, and p65.

Why are the genes identified in the independent screening assays different? A closer analysis would suggest that the issue lies in a combination of a highly sensitive assay system with nonstringent selection criteria. The sensitivity of the assay system is based in the close interaction between viral replication and host cell function, where any change in cellular function will affect the elevated viral replication rates of a laboratory-optimized assay system. A broad and pleiotropic response of hits combined with a nonstringent selection system will inevitably produce a high ratio of false positives. Indeed, three factors appear to contribute to the lack of stringency in the hit selection system of previous screens:

1. Previous analyses chose arbitrary threshold levels favoring a selection of a small fraction of the experiments in a primary screen. In Brass et al., ³ this threshold is defined at two standard deviations from the mean of the microplate; in König et al., ⁶ it is 45% of infectivity; in Zhou et al., ⁴ it is two standardized mean differences; and in Yeung et al., ⁵ it is twofold mRNA enrichment in cells that survived HIV infection. Accordingly, all studies chose an arbitrary value on the axis of apparent infectivity (cell number) under which primary hits are selected. On one hand, such a selection criterion removes between 95% ⁴ and 98% ³ of the genes, many of which may nevertheless constitute further hits. On the other hand, thresholding will also select false positives from the tail of the distribution. Issues concerning this kind of analysis for the RNAi screen were underlined by Birmingham et al. ²⁵

2. The issue of thresholding is compounded by the lack of sufficient selection parameters. The screens mostly use a small number of measurements for identifying hits, such as cell number and reporter production. Systematic gene silencing clearly has numerous consequences on cell function, producing a large variety of unknown phenotypes ( Fig. 1C ). Most of those phenotypes are unpredictable, may or may not be related to HIV infection, and can have a significant effect on a single readout such as a fluorescent reporter assay. The direct consequence is that complex multiple phenotype loss-of-function screens simply cannot be resolved or sorted by thresholding one or two criteria such as infectivity and cell number because this is akin to arbitrarily creating two to four classes while thousands of classes coexist. It requires a more subtle deconvolution.

3. Finally, in those previous works, the experiments were generally only performed in duplicate, but siRNA screens are now known to show high variation. ²⁵

To tackle these problems, screen complexity was reduced by focusing on early steps of HIV infection (gene knockdowns that resemble CD4 silencing), reducing the number of host–viral interaction steps. To simultaneously increase hit selection stringency—at the primary screen—we applied a specific selection algorithm that employs 15 criteria across seven genome-wide screens comprising >190 000 experiments (including controls) to discriminate CD4 block from other phenotypes ( Fig. 2A ).

Our procedure individually profiled all infectious knockdown experiments on a genome-wide level. Fifteen parameters seem sufficient to specifically profile and recognize the morphometric state of CD4 viral infection block, as syncytia formation and GFP reporter production will respond starkly differently from infected cells (3% of experiments fall in the CD4 distribution). However, the level of specificity that a given constellation of 15 parameters can convey becomes apparent when we are discerning unspecific effects on viral replication. Not only will the knockdown of GFP deviate from the 15-dimensional CD4 profile, but housekeeping genes such as XPO1 will have cellular effects in addition to the viral replication block, which makes these profiles different from the reference. It will be worth determining further the phenotype of other housekeeping genes.

However, it is important to keep in mind that the gain in specificity is a trade-off, in that many host genes that are both directly involved in viral interaction and have an essential housekeeping function will not be selected. In contrast to the much more comprehensive list of the previous studies,^3–6 we identify a small set of genes that may be essential for viral replication, the knockdown of which does otherwise not affect cell viability.

In this study, our aim was not to precisely determine at which stage of the viral life cycle the cellular proteins are implicated. This will require further investigation. Our goal was to demonstrate that the screening method identified genes that are HIV host factors. As a proof of concept, we confirmed that three of the selected proteins—namely, PKN1, Ku70, and RNAseH2A—also regulate HIV replication in the Jurkat T cell line.

Further work on the host biology of the human host factors identified here will require use of primary lymphocytes and other physiologically relevant cellular targets of HIV to exclude factors that are not therapeutic targets. Notably, some of the identified proteins are probably expressed in epithelial-like cells and not in lymphocytes. This is the case for claudin 19, which is not expressed in T cells and whose implication in HIV replication is questionable. A caveat of using HeLa cells is that T cell–specific factors may be missed in the screen.

In summary, the list of 56 human host factors may not capture all aspects of host viral interactions during the early steps of the HIV infection yet generates a number of prime cellular targets for therapy development. This novel genome-wide screening technique may also be used for the study of microbial–host interactions and other pathological or physiological cellular processes.