Using conserved ELMs found in viral genes, Evans et al. 1 identified complementary 'counter domains' on cellular proteins that serve as binding partners for the viral proteins. Owing to the high mutation rates of RNA viruses, the preservation of these viral sequence motifs would presumably confer viral fitness. From alignments of more than 70 HIV-1 sequences, 56 ELMs were found to be conserved among the individual HIV-1 gene products. At first glance, the identification of 2,348 predicted cellular interaction partners - close to 10% of the human protein-encoding genes - seems unreasonably high. However, the dataset includes both a set of direct viral-host protein interactions (called H1) and a larger set of competitive interactions (H2). The H2 set comprises cellular proteins that share ELMs with viral proteins and that may not, therefore, necessarily bind to viral proteins. The matching of viral proteins to counter domains is likely to increase in the future, because the discovery of H1 and H2 is limited by the number of ELMs in the ELM database, currently only 133 2.

Looking at the H1 set alone, the authors 1 reported enrichments in cellular proteins that bind to the viral proteins Tat, Env, and Nef. In accordance with the apoptosis-inducing properties of gp120 (a protein formed from cleavage of Env) and Nef, the authors found (by Gene Ontology and KEGG pathway analysis) an enrichment in cellular partners of Env and Nef involved in apoptosis and protein kinase signaling in the H1 set. In line with the role of Tat as a transactivator, there was an enrichment in cellular partners along the Ras p38/mitogen activating protein kinase signaling cascade, including the small GTPase Ras itself.

As partial validation, Evans et al. 1 compared their predictions with the findings of several genome-wide RNAi screens, with varying results. In those surveys, almost the entire human genome was screened by individual gene silencing in HIV infection or transfection systems. The RNAi screens rest on the assumption that any reduction in HIV production resulting from the knockdown of an individual gene implies a role for that gene in virus replication.

Using a HeLa-CD4 cell line, Brass et al. 4 and Zhou et al. 6 both reported more than 200 host-dependency factors. Following transfection with small interfering RNAs (siRNAs), Brass et al. 4 implicated cellular factors on the basis of their effects on two indicators: the production of viral Gag p24 antigen during early viral replication; and the infectivity of culture supernatant from siRNA-transfected cells. The screen by Zhou et al. 6 was carried out in a similar manner, but with infectivity monitored over a longer period of time. A similar screen by König et al. 5 was carried out with a vesicular stomatitis virus glyco-protein (VSV-G)-pseudotyped virus in 293T cells; while substituting the HIV envelope with VSV-G allowed for the incorporation of the HIV core into the 293T cells, the experimental design precluded the identification of host factors that participate in viral entry mediated by fusion of the HIV envelope with the cellular membrane. As a result, only events from uncoating of the viral core to viral gene expression were detectable by that siRNA screen.

Despite the differences in experimental designs, the three screens 456 reported similar findings at the level of cellular functions. In the two screens performed in HeLa-CD4 cells that captured the production of infectious virions, both groups 46 identified large numbers of cellular factors involved in Tat-mediated transcription as part of the Mediator complex, and also cellular factors involved in energy metabolism regulated by the Akt kinase. Similarly, Brass et al. 4 and König et al. 5 both identified constituents of the nuclear pore complex, presumably related to the nuclear entry of the viral pre-integration complex. However, although all three screens 456 each reported between 200 and 300 cellular genes, together they reported a total of 842 unique genes as contributing to HIV replication. Only three genes were commonly reported by all three screens.

Perhaps more significantly, as revealed in a recent meta-analysis of the genome-wide screens 7, several known cellular cofactors of HIV replication were not identified by any of the siRNA screens 456. Missing were the cell-surface antigens HLA-B57 and HLA-C, both of which regulate the immune response to HIV and have known effects on viral loads and disease progression; also missing was the integration cofactor LEDGF/p75. Although components of the ubiquitylation machinery were identified, members of the Tsg101/ESCRT pathway, which is specifically co-opted for the virus to leave the cell, were also missing. In addition, proteins such as the Gag-binding protein cyclophilin A and the HIV long terminal repeat binding partner Sp1 were identified only in the 293T-based screen 5. The false negatives arising from siRNA screens can be attributed to the fact that cellular proteins that are particularly abundant and stable cannot be easily knocked down by siRNA transfection in the time available in these experiments. Furthermore, RNAi specificity requirements and siRNA off-target effects can introduce false positives and negatives into the analysis. Considering the various gaps in identification by the siRNA screens, it should come as no surprise that Evans et al. 1 reported a moderate overlap between findings from the siRNA screens 456 and those that they made on the basis of conserved ELMs and counter domains.

Advances in high-throughput transcriptomic and proteomic approaches have allowed the comprehensive characterization of the 'natural' environment of HIV replication, namely the cellular factors around the virus in the cell types it infects. Following a transcriptomic approach, Imbeault et al. 8 have recently reported a microarray study of HIV-infected primary CD4 T cells. This study revealed the induction of type-I interferons and the subsequent upregulation of p53 transcription following infection with the NL4-3 strain of HIV-1. Likewise, in a recent proteomic analysis of primary CD4 cells infected with the LAI strain of HIV-1, our group 9 reported signs of apoptosis following infection on the basis of accumulation of poly-(ADP)-ribose polymerase (PARP) and indications of compromised mitochondrial integrity. Furthermore, treatment with a reverse transcriptase inhibitor reversed many of the protein abundance changes associated with robust virus replication, thus allowing us to identify perturbations in karyopherin-mediated nuclear trafficking and mRNA-splicing factors associated with DNA repair as unique host response signatures that accompanied HIV replication 9.

More recently, Nathans et al. 10 performed a microRNA microarray analysis, revealing a specific cellular microRNA (miR-29) that targeted the HIV-1 3' untranslated region, thereby repressing virus production. Unlike the examples discussed thus far, this represents a possible mechanism by which the virus modulates immune evasion rather than facilitating viral replication directly.

The identification of cellular factors that contribute to HIV replication will pave the way for the development of antivirals that are more resilient to the evolution of viral drug resistance; Maraviroc, an analog of the chemokine receptor CCR5, is a step in this direction. To this end, one could argue we can simply focus on identifying cellular factors that promote HIV infectivity (Figure 1a). However, given the technical limitations with the coverage gaps of siRNA screens and the difficulty in adapting primary cells for screening purposes, complementary high-throughput approaches need to be adopted for the full elucidation of virus-host interactions and their contribution to HIV replication. As reported by Dampier et al. 3, the conservation of different ELMs in the HIV reverse transcriptase can predict a patient's response to therapy, pointing to the potential utility of virus-host interactions as a prognostic tool for response to drug regimens. In this regard, defining the structural basis of interaction - such as by predicting virus-host binding interactions between ELMs and counter domains as described by Evans et al. 1 - helps provide the proper context for interpreting the high-density, high-throughput data (Figure 1b).