The principal feature of retroviruses is that upon entry and release of their viral RNA genome into the cytoplasm of the host cell, it is reverse transcribed by the viral reverse transcriptase into a linear double stranded cDNA copy (Figure 1A) (for reviews, see [1-4]). This viral DNA is not naked, but is associated with viral and cellular proteins in a nucleoprotein complex called the preintegration complex (PIC). Depending on the retrovirus, the PIC is subsequently translocated into the nucleus, either actively through nuclear pores, or upon nuclear membrane disruption occurring during mitosis. There, it is, either integrated into the genome of the host cell, or remains unintegrated for a certain time, or is degraded (Figure 1) (for reviews, see [3,5]).

The insertion of the viral DNA into the host cell genome is catalyzed by the virally encoded integrase (IN) enzyme (for reviews, see [6-9]). Retroviral INs typically range between 280 and 450 amino acids (HIV-1 IN: 288 amino acids, 32 kDa), and are characterized by three functional domains: (i) the N-terminal domain, containing an HHCC zinc-binding motif, (ii) the catalytic core domain (residues 50-212 of HIV-1), containing the critical magnesium-binding D-D-35-E motif that constitutes the active site, and (iii) the C-terminal domain. The three domains of IN appear to be involved in DNA binding and multimerization. Indeed, the full concerted integration seems to require an IN tetramer, i.e., one IN dimer at each viral end [10-14].

Although unintegrated viral DNA can be used as template for viral transcription [15-17], integration is required for productive viral replication. However, the efficiency of integration is quite low, and depends on restrictions occurring during the early steps of infection. By infecting human osteosarcoma (HOS) cells with VSV-G pseudotyped HIV particles, Thomas et al. measured that only 5% of HIV viruses successfully entered the cell and initiated reverse transcription [18]. Of these, 28% (i.e., 1.5% total) completed reverse transcription, translocated to the nucleus (with an efficiency ranging around 2-3% according to data using IN-eGFP fusion proteins from Cereseto and collaborators [19,20]) and finally only ~13% of viruses that initiated reverse transcription achieved insertion in the host cell genome, which represented 0.41% of the virus input [18].

The site of viral DNA insertion is critical for the virus, as it can influence the rate of viral transcription. Indeed, integration into transcriptionally active regions may favor viral gene expression, thus facilitating productive infectious progeny particles, while integration into transcriptionally repressed chromatin may disfavor viral gene expression, thus possibly facilitating viral latency [21-24].

The ability of retroviruses to integrate has also important consequences for the host as it can affect the expression of genes surrounding the proviral DNA. Indeed, viral DNA disruptive insertion into a gene may alter its expression (reduced gene expression), thereby potentially affecting cellular physiology. More importantly, the activity of inserted viral promoters or enhancers near cellular genes may also affect their physiological expression (increased gene expression), potentially leading to tumorigenesis when these genes are proto-oncogenes [25,26]. This process, known as insertional mutagenesis, raised justified critical issues regarding the safety of retroviral-based vectors used in gene therapy (for reviews, see [27-32]). However, not all retroviruses display the same genotoxic potential, as gammaretroviruses for example appeared to be more prone to insertional mutagenesis than lentiviruses [33-37]. This phenomenon can be partly explained by their preferred genomic site for proviral DNA insertion, i.e., into promoter regions for gammaretroviruses and along transcription units for lentiviruses.

It is now clear that the chromosomal site of viral DNA integration is not random, but in contrast that retroviruses display specific preferences at distinct genomic positions. This review will summarize the current knowledge about retrovirus-specific favored integration sites, as well as the current models explaining these preferences.

In vitro, IN is sufficient to carry out the first two steps of the integration reaction, i.e., 3’ processing and strand transfer reactions, resulting in the covalent attachment of the viral DNA on virtually any DNA target (random integration) (Figure 1B) [38]. To succeed, three principal components are minimally required: (i) purified viral integrase, (ii) a donor DNA mimicking a viral DNA terminal sequence to be recognized by IN, and (iii) an acceptor DNA in which the donor DNA will be inserted (for more details, see [7]).

In the first in vitro assays, the donor DNA consisted in short oligonucleotide duplexes (21 bp minimum) containing the terminal LTR sequence (either U3 or U5), allowing to reproduce the 3’ processing efficiently, as well as the strand transfer reaction (albeit with lower efficiency). However, these were only half-site integration as only one donor DNA (e.g. one viral LTR) was inserted in the acceptor DNA (Figure 1C) [39], and not both in a concerted motion. This gave rise to the development of new assays, full-site or concerted integration assays, which use a longer donor DNA containing both terminal sequences (Figure 1D) [40,41].

 These studies showed that in vitro, HIV IN displayed only a weak preference for the primary DNA sequence [42-49], slightly favoring the palindromic TNNGT(A/T)ACNNA DNA sequence (bold nucleotides indicate the asymmetrical insertion points, resulting in the final 5 bp duplication flanking the proviral DNA, depicted in blue in Figure 1B). Furthermore, the addition of nucleosomes on the target DNA improved the in vitro efficiency of integration, and favored integration on distorted DNA and outwardly-facing major grooves sites of the nucleosomal DNA [44,46,47,50-53].

In order to investigate whether simple tethering of IN to a specific DNA site could confer integration preferences in vitro, fusions of IN to specific DNA binding proteins were engineered [54]. The fusion of HIV IN to the DNA binding domain of λ repressor (λR) lead to increased integration targeting at sites surrounding the predetermined λR DNA binding sites (λ operator sites) [54]. Fusion of IN with other DNA binding proteins such as LexA [55] or the polydactyl Zinc finger protein E2C [56] reached similar results. These studies provided proof-of-concept that integration site selection in vitro could be modified and redirected more preferentially to specific DNA sites.

The availability of the human genome sequence and other vertebrate genomes made possible to interrogate where in the host cell genome retroviruses integrated, and more precisely what were the chromosomal features (according to current genomic annotations) that were favored for retroviral integration. To achieve this, host DNA regions flanking the proviral DNAs were amplified, sequenced, and finally aligned to the host genome sequence (method overview reviewed in [32,57,58]).

Schroder et al., in 2002, revealed for the first time that HIV favored integration in transcription units and disfavored Alu repeats [59]. One year later, Wu et al. showed that murine leukemia virus (MLV) had distinct preferences, favoring integration at transcription start sites and CpG islands [60]. Since then, multiple genome-wide studies confirmed these preferences and revealed the integration site preferences for almost all retroviral genera, with the exception of epsilonretroviruses (Tables 1 and 2).

These studies demonstrated that in vivo the site of retroviral integration was not random, and that integration site preferences were retrovirus-specific (Table 1 ): lentiviruses favor integration in active transcription units, with no preference along the transcript, nor for introns or exons; gammaretroviruses, spumaviruses and endogenous retroviruses (HERV class II) integrate preferentially around transcription start sites and CpG islands, features associated with host gene promoters; alpharetroviruses and deltaretroviruses displayed only weak preferences for integrating in transcription units and CpG islands; and finally betaretroviruses show no integration site preferences, displaying a random distribution of integration sites in the host genome.

The integration site preferences are not host-specific as the same distribution of integration sites can be observed in different host vertebrate cells, including human, simian, murine, avian and canine cells (Table 1 ). Furthermore, integration targeting is independent of the route of viral entry, as HIV-based vectors using a natural CCR5-tropic HIV envelope or a VSV-G pseudotype envelope displayed the same integration site distribution [67].

Three models, which are not mutually exclusive, have been proposed to date to explain integration site selection: (i) chromatin accessibility, (ii) cell cycle effects, and (iii) tethering mechanism. However, while the first two models can globally influence integration site targeting, only the last one - integration by a tethering mechanism - can provide a logical explanation to the observed differences of integration targeting preferences among retroviruses.

LEDGF/p75 was initially identified in a complex co-immunoprecipating with positive cofactor 4 (PC4), a general coactivator of transcription [113,129]. PC4 has been involved in many transcription steps: (i) PC4, by interacting with upstream activators and the general transcriptional machinery, can enhance the efficiency of pre-initiation complex assembly, thereby promoting transcription initiation, in cooperation with TBP-associated factors (TAFs) [130]; (ii) PC4 improves transcription activation by stimulating promoter escape [131]; (iii) Sub1, the yeast PC4 homolog, facilitates transcription elongation and may also prevent premature transcription termination [132]. Thus, being associated to a PC4-containing complex, itself associated with the transcription machinery, it is reasonable to hypothesize that LEDGF/p75 might be associated with the RNA polymerase II complex during elongation.

Therefore, based on the current knowledge about HIV integration site selection, i.e., preferentially integrating into active transcription units, it is tempting to speculate a dynamic, more than a static, tethering model, in which LEDGF/p75 would be associated with PC4 and the RNA polymerase II elongation complex (Figure 2). In this model, LEDGF/p75 would recruit HIV preintegration complex while transcribing genes or at pausing sites, thereby explaining integration sites all along the transcription units.

Consistent with these results are (i) integration in active transcription units, with no preference along the transcription unit, neither for exons, nor for introns, (ii) LEDGF/p75 initially characterized as a transcriptional coactivator, associated with the transcription machinery (via PC4-containing complex), (iii) LEDGF/p75 interaction with the menin/MLL histone methyltransferase, involved in H3K4me3 histone modification, a mark associated with active transcription, and (iv) no sequence consensus for LEDGF/p75-binding DNA sites, suggesting that the N-terminal domain of LEDGF/p75 may serve as a hook to anchor the protein to the chromatin but that the location specificity is given by an additional chromosome-bound protein, yet to be identified.

Further studies on LEDGF/p75 should help refining the detailed mechanism of LEDGF/p75-mediated HIV integration.