Tubulovesicular Structures within Vesicular Stomatitis Virus G Protein-Pseudotyped Lentiviral Vector Preparations Carry DNA and Stimulate Antiviral Responses via Toll-Like Receptor 9

Abstract

Recombinant lentiviral vectors (LVs) are commonly used as research tools and are being tested in the clinic as delivery agents for gene therapy. Here, we show that Vesicular stomatitis virus G protein (VSV-G)-pseudotyped LV preparations produced by transient transfection are heavily contaminated with tubulovesicular structures (TVS) of cellular origin, which carry nucleic acids, including the DNA plasmids originally used for LV generation. The DNA carried by TVS can act as a stimulus for innate antiviral responses, triggering Toll-like receptor 9 and inducing alpha/beta interferon production by plasmacytoid dendritic cells (pDC). Removal of TVS markedly reduces the ability of VSV-G-pseudotyped LV preparations to activate pDC. Conversely, virus-free TVS are sufficient to stimulate pDC and act as potent adjuvants in vivo, eliciting T- and B-cell responses to coadministered proteins. These results highlight the role of by-products of virus production in determining the immunostimulatory properties of recombinant virus preparations and suggest possible strategies for diminishing responses to LVs in gene therapy and in research use.

Lentiviral vectors (LVs) are used extensively as research tools and are promising agents for genetic therapy of human diseases (11, 24, 41, 66). They have several advantages over other virus-based vectors, including a large gene-carrying ability coupled with the capacity to integrate into the genomes of nondividing cells (11, 24, 41, 66). Furthermore, compared with the use of adenovirus and vaccinia virus vectors, the use of recombinant LVs in the clinic is not limited by preexisting immunity to viral proteins (18, 34). Gene transfer for therapeutic purposes should, if possible, be immunologically silent or provoke immunological tolerance. However, lentiviral vectors have been shown to elicit powerful cytotoxic-T-lymphocyte (CTL) responses against transgene-encoded proteins (17, 20, 49, 52). Similarly, human immunodeficiency virus type 1 (HIV-1), from which most recombinant LVs are derived, elicits potent cell- and antibody-mediated responses in humans (45). Their immunogenicity indicates that lentiviruses and recombinant LVs must activate innate viral sensing pathways, which subsequently couple to adaptive immunity. Remarkably, little is known about innate responses to recombinant LVs. Studying the interaction between LVs and cells of the innate immune system might therefore suggest strategies to maximize the therapeutic use of LVs in the clinic, as well as avoid transgene-independent effects in the laboratory.

The pathways mediating innate sensing of virus infection are becoming increasingly understood, in part through work on dendritic cells (DC), a type of leukocyte that is central for translating innate responses into adaptive immunity (1, 36). Exposure to viruses can lead to DC activation, inducing upregulation of major histocompatibility complex and costimulatory molecules, as well as secretion of immunomodulatory cytokines (50). The latter include type I interferons (IFN-α/β), a group of cytokines that are important for viral containment and for amplifying DC activation and T-cell priming (40, 62). Thus far, three distinct mechanisms that promote DC activation and IFN-α/β synthesis upon exposure to viruses have been recognized. Direct infection can promote DC activation through cytoplasmic sensing of viral genomes or viral replication intermediates such as double-stranded DNA (dsDNA) or dsRNA (15, 33, 35, 60). DC can also become activated upon phagocytosis of some virally infected cells, whose dsRNA content acts as a trigger for Toll-like receptor 3 (TLR3) (57). Finally, DC can directly recognize the presence of genomes from single-stranded RNA (ssRNA) and DNA viruses in endosomes by means of TLR7 and TLR9, receptors for ssRNA and DNA, respectively (14, 30, 38, 39, 43, 44). The most likely stimuli for inducing innate and adaptive immunity to LVs are therefore the nucleic acids carried by the virions, including the ssRNA viral genome and/or nascent DNA transcripts (63). In this regard, it has been reported that HIV-1 induces IFN-α production by plasmacytoid dendritic cells (pDC), most probably by triggering TLR7 (4, 22, 23, 55, 67). pDC constitute a subtype of DC that is uniquely able to couple TLR7 and TLR9 signaling to IFN-α gene expression (31) and plays an important role in immunity to some viruses (1, 12, 42).

Recombinant lentiviruses are commonly produced by transient transfection of producer cells with a plasmid encoding the viral RNA genome (bearing the transgene), as well as additional plasmids that provide in trans the necessary components for generation of infectious virus (Gag, Pol, Rev, and Env proteins) (41). Vesicular stomatitis virus G protein (VSV-G) is the most commonly used Env glycoprotein because VSV-G-pseudotyped LVs are pantropic, can be concentrated by centrifugation, and are relatively resistant to freezing (41). To investigate whether recombinant LVs induce an innate immune response via TLR7 and/or other mechanisms, we monitored IFN-α production by pDC-containing murine bone marrow (BM) cells upon exposure to standard preparations of VSV-G-pseudotyped LVs generated by transient transfection. We show here that such preparations act as strong activators of pDC and induce secretion of high levels of IFN-α. However, we show that pDC stimulation is not due to the virus itself but rather is attributable to tubulovesicular structures (TVS) that are produced by VSV-G-transfected cells and copurify with virus particles during viral concentration. These TVS carry residual amounts of the DNA plasmids used in the lentivirus production process and act as agonists for TLR9. These results indicate that contaminating TVS profoundly determine the safety and immunostimulatory properties of recombinant LV preparations and suggest strategies for optimizing the use of recombinant LVs in gene therapy and research.

MATERIALS AND METHODS

Plasmids.

pLentiLox 3.7, pMDLg/pRRE, and pRSV-Rev were kindly provided by Luc van Parijs (Massachusetts Institute of Technology, Cambridge, MA). The XbaI-XhoI fragment of pLentiLox 3.7, containing the U6 promoter, was excised to generate pLLCG, a lentiviral vector coding for enhanced green fluorescent protein (eGFP) under the control of a cytomegalovirus (CMV) promoter. pVSV-G, encoding the vesicular stomatitis virus envelope protein (Indiana strain), was from Invitrogen. pCAAGS-LCMV-G, encoding the lymphocytic choriomeningitis virus (LCMV) glycoprotein (Armstrong strain), was kindly provided by Daniel Pinschewer, University Hospital Zürich. pCAAGS-THOV-G, encoding the thogotovirus (THOV) glycoprotein, was a kind gift from Georg Kochs, University of Freiburg. pMLV-A, encoding the murine leukemia virus (MLV) amphotropic envelope protein, has been described previously (53). pLNC-VSV-G, a retroviral vector expressing VSV-G, was generated by replacing the eGFP gene of pLeGFP-N1 (Clontech) with VSV-G. pMSCVneo was from Clontech.

Cells.

293FT, 293T, 293GP, and STAR-HV cells were grown in Dulbecco modified Eagle complete medium (containing 10% fetal calf serum [FCS], 200 μM l-glutamine [Gibco×, 10 U/ml penicillin [Gibco], and 10 μg/ml glutamine [Gibco]). Bone marrow cells were isolated from C57BL/6 (Charles River), MyD88^−/−, TLR9^−/−, or TLR7^−/− (a kind gift from S. Akira, Osaka, Japan) mice or from Rag-2^−/− mice (all bred at Cancer Research UK). After red blood cells lysis, cells were cultured in RPMI 1640 complete medium at a density of 10⁶ cells/well in 96-well U-bottom plates with various stimuli, and IFN-α accumulation in supernatants was measured by enzyme-linked immunosorbent assay (ELISA) 18 to 24 h later as described previously (15). In some experiments, BM cells from Rag2^−/− mice were fractionated into B220⁺ and B220⁻ fractions by using anti-CD45R (B220) magnetic MicroBeads and AutoMacs separation (Miltenyi Biotec) and seeded at 2 × 10⁵ cells/well. Intracellular staining for IFN-α was performed as described previously (15). BM-DC were grown in the presence of Flt3L as described previously (14) and stimulated at 2 × 10⁵ cells in flat-bottom 96-well plates.

Viruses and stimuli.

LVs expressing eGFP were produced as follows. 293FT or 293T cells (6 × 10⁶) were seeded in a 10-cm dish and transfected using Lipofectamine 2000 (Invitrogen) as per the manufacturer's instructions with pMDLg/pRRE (5 μg), pRSV-REV (5 μg), pLLCG (10 μg), and the plasmid encoding the desired envelope protein (5 μg; VSV-G unless stated otherwise). For LVs lacking the envelope protein (LVΔenv), the Env plasmid was omitted. For LVs without viral RNA (LVΔvRNA), pLLCG was omitted. For control supernatant, all plasmids were omitted. At 16 to 24 h after transfection, cells were washed and fresh medium added. Supernatant was collected at 48, 60, and 72 h after transfection; filtered though a 0.45-μm Millex-HV filter (Millipore); and stored at −80°C. Before use, supernatants were pooled and ultracentrifuged through a 20% sucrose cushion for virus concentration. The virus titer was determined by transduction of 293FT cells. One infectious unit corresponds to one GFP-forming unit (GFU) on 293FT cells. Noninfectious virus (LVΔenv and LVΔvRNA) was quantified by ELISA for HIV p24 (Aalto), using infectious virus of known GFU as a standard.

For LV fractionation, concentrated supernatant containing to 4 × 10⁸ GFU of virus was loaded on a 20 to 60% continuous sucrose gradient and ultracentrifuged for 16 h at 85,000 × g in a Beckman SW40Ti rotor. Fractions of 1.5 ml were sequentially collected, diluted in phosphate-buffered saline (PBS), and ultracentrifuged for 1.5 h. Pellets were resuspended in complete culture medium.

Retrovirus encoding VSV-G was generated by transfection of 6 × 10⁶ 293GP cells with pLNC-VSV-G (10 μg) and pVSV-G (5 μg). Supernatant was harvested, concentrated 10-fold by ultracentrifugation, and used for infection of STAR-HV cells (32) to produce infectious LVs.

CpG 1668 (Sigma), poly(I:C) (Sigma), loxoribine (Invivogen), and R848 (Invivogen) were pretitrated and used at concentrations that induce maximal IFN-α production from pDC. Neutralization of VSV-G was accomplished using the V17 MAb (2).

Western blotting.

Concentrated supernatant was mixed 1:1 with Laemmli buffer and heat inactivated at 94°C for 5 min. A 10-μl portion was run on precast 10% sodium dodecyl sulfate-polyacrylamide gels (Invitrogen) and blotted onto an Immobilon P membrane by using a Bio-Rad semidry transfer apparatus. After blocking in 5% bovine serum albumin, the membrane was stained with mouse anti-VSV-G (MAb P5D4 (37) (Cancer Research UK), mouse anti-GFP (MAb GFP 3E1; Cancer Research UK), mouse anti-β-actin (MAb clone AC-15; Sigma), or biotinylated mouse anti-HIVp24 (Aalto). Secondary reagents were horseradish peroxidase (HRP)-conjugated goat anti-mouse antibody (Santa Cruz Biotechnology) or HRP-streptavidin (Sigma). The HRP signal was detected using the West Pico kit (Pierce).

PCR and reverse transcription-PCR (RT-PCR).

DNA and RNA were extracted from 35 μl concentrated supernatant. RNA was prepared by DNase digestion (DNA-free; Ambion) and reverse transcribed using Superscript II reverse transcriptase (Gibco), 1 μM deoxynucleoside triphosphates, and 10 μM random hexanucleotides (Gibco). DNA and cDNA were amplified using 30 PCR cycles, each consisting of 30 s at 94°C, 30 s at 55°C, and 60 s at 72°C. Sequences of primers were as follows: 5′ VSV-G, GAAGTGCCTTTTGTACTTAG; 3′ VSV-G, GATCGGATGGAATGTGTTAT; 5′ GFP, GGCCACAAGTTCAGCGTGTC; 3′ GFP, TTGCCGTCCTCCTTGAAGTC; 5′ Pol, TCAGAAGCAGGAGCCGATAG; 3′ Pol, TGCAGCCAATCTGAGTCAAC; 5′ β-actin, GTTTGAGACCTTCAACACCCC; and 3′ β-actin, GTGGCCATCTCCTGCTCGAAGTC.

Electron microscopy.

Cells were fixed in 4% glutaraldehyde for 1 h before being processed for routine araldite processing and sectioning on a Leica ultramicrotome. Sections were placed on 200-mesh nickel grids and poststained in lead citrate and uranyl acetate. For negative staining of virus-containing supernatants, glutaraldehyde-fixed, concentrated supernatants were spread on glow-discharged carbon-Formvar 400 mesh nickel grids and allowed to settle in a moist chamber for a few minutes. The grids were washed twice in distilled water and stained with aqueous 1% uranyl acetate. Specimens were examined using a JEOL 1010 (JEOL Ltd., Tokyo, Japan) electron microscope.

Immunization.

Mice were immunized intraperitoneally with egg white (7) equivalent to 250 μg ovalbumin (OVA), in either PBS alone, mixed with 320 μl VSV-G-TVS, or mixed with poly(I:C) (50 μg). At 1 week after immunization, blood was analyzed for the presence of OVA/H2-K^b tetramer-positive T cells (57). At 10 to 12 days after immunization, CTL activity was assessed by in vivo target killing assay as described previously (57). For CD4⁺ T-cell responses, splenocytes were incubated overnight in medium in the presence or absence of FCS. Brefeldin A was added for the last 3 h and cells stained for intracellular IFN-γ. For detection of serum antibodies, ELISA plates were coated overnight with OVA (Calbiochem) or FCS before blocking with fish skin gelatin (1% in PBS) and addition of serially diluted serum samples. Biotinylated anti-mouse immunoglobulin G antibody (Jackson ImmunoResearch) and streptavidin-alkaline phosphatase diluted in PBS containing 1% fish skin gelatin were used as detection reagents.

RESULTS

VSV-G-pseudotyped LV preparations elicit IFN-α production by plasmacytoid dendritic cells.

To determine the potential of LVs to stimulate innate immunity, we asked whether they could trigger leukocytes to produce IFN-α/β, one of the earliest hallmarks of innate antiviral defense. Upon exposure to graded doses of a standard LV preparation, BM cells produced IFN-α at levels comparable to those maximally elicited by a control stimulus, CpG-containing DNA oligonucleotide (CpG) (Fig. 1A). Although all nucleated cells have the potential to secrete interferons (40), intracellular staining revealed that only a small subset of BM cells made high levels of IFN-α in response to LVs. These cells were CD11c^low, B220⁺, and Ly6C⁺ (Fig. 1B), a phenotype indicative of pDC. To confirm their identity, BM cells from lymphocyte-deficient RAG-2^−/− mice were fractionated on the basis of B220 expression. Only the B220-enriched population produced IFN-α in response to CpG, indicating that the majority of pDC were contained in this fraction (Fig. 1C). Similarly, only the B220⁺ fraction secreted significant levels of IFN-α in response to LV preparations (Fig. 1C). We conclude that LV preparations can stimulate pDC, leading to the production of high levels of IFN-α.

FIG. 1.

LV preparations induce IFN-α production by pDC. (A) IFN-α secretion by BM cells after treatment with preparations of VSV-G-pseudotyped LVs at the indicated multiplicity of infections (MOIs) or with CpG (0.5 μg/ml; positive control). Supernatant (SN) of untransfected cells was concentrated as for the LV preparation and used as a negative control. (B) Intracellular staining for IFN-α in BM cells stimulated with LVs at an MOI of 0.2 for 6 h. Left panel, IFN-α versus CD11c. Right panel, Ly6C and B220 expression on IFN-α-positive cells gated as indicated in the left panel. Numbers represent the percentages of cells in each gate or quadrant. (C) IFN-α secretion by B220⁺ and B220⁻ fractions of BM cells taken from RAG-2^−/− mice after treatment with VSV-G-pseudotyped LVs (MOI of 0.2) or CpG. The B220⁺ fraction consisted of >90% pDC; the B220⁻ fraction contained around 7% pDC (not shown). n.d., not detected. Error bars indicate standard deviations of triplicates.

The IFN-α response of pDC to LV preparations is independent of viral RNA but dependent on VSV-G.

We produced LV preparations pseudotyped with four distinct Env glycoproteins, VSV-G, MLV-A, LCMV-G, or THOV-G, and assessed their innate stimulatory ability. Notably, only VSV-G-pseudotyped LV preparations induced IFN-α production by pDC (Fig. 2A). We hypothesized that the viral genome acted as the innate trigger and that VSV-G glycoprotein was uniquely able to deliver the virus to TLR7-accessible endocytic compartments. To test this hypothesis, we omitted the plasmid encoding the viral RNA and produced VSV-G-pseudotyped virus preparations lacking viral genomes (LVΔvRNA). Surprisingly, these preparations were almost as potent as preparations of conventional LVs in stimulating an IFN-α response (Fig. 2B). In contrast, preparations of LVs lacking the VSV-G glycoprotein (LVΔenv) were unable to stimulate IFN-α production, as were preparations obtained by transfecting only the plasmid encoding the viral RNA (Fig. 2B). Unexpectedly, preparations from cells transfected only with the Rev- and VSV-G-encoding plasmids were comparable to full LV preparations at stimulating IFN-α production (Fig. 2B). We asked whether expression of VSV-G was necessary for this response. Preincubation of VSV-G-transfected cell supernatants with a neutralizing anti-VSV-G monoclonal antibody completely blocked their ability to induce IFN-α from BM cells, whereas the same antibody had no effect on the control stimulus, CpG (Fig. 2C). Similarly, exposure of the supernatants to low pH, which inactivates VSV-G, abrogated IFN-α induction (data not shown). We conclude that viral RNA is not required for the interferon-inducing properties of LV preparations and that supernatants from cells transiently transfected with VSV-G are sufficient to induce IFN-α via a mechanism dependent on VSV-G expression.

FIG. 2.

VSV-G expression is required for the IFN-α response to LVs. (A) BM cells were stimulated with LV preparations (multiplicity of infection of 1) pseudotyped with the indicated glycoproteins; IFN-α was measured after overnight culture. (B) Stimulation of BM with concentrated supernatants from 293FT cells transfected with the indicated combinations of plasmids to generate preparations of VSV-G-pseudotyped LV, LV lacking viral RNA (LVΔvRNA), or LV lacking the envelope protein (LVΔenv). Concentrated supernatants from untransfected cells (SN) or from cells transfected with plasmids encoding Rev plus VSV-G (VSV-G) or Rev plus viral RNA (vRNA) were tested as controls. Where applicable, stimuli were normalized for HIV p24 content (data not shown). (C) Concentrated supernatant from VSV-G-transfected cells (VSV-G-SN) or CpG was preincubated with a neutralizing anti-VSV-G or control MAb before being added to BM cells. IFN-α in supernatants was measured 16 h later. n.d., not detected. Error bars indicate standard deviations of triplicates.

VSV-G transfection leads to accumulation of VSV-G-bearing tubulovesicular structures in the extracellular milieu.

The IFN-α-inducing activity of VSV-G supernatants could be concentrated by ultracentrifugation, suggesting the presence of a particulate stimulus (data not shown). We therefore examined the transfected cells and supernatants by electron microscopy for the presence of potential VSV-G-bearing structures. Cells transfected with the four plasmids used for virus production showed abundant tubulovesicular structures within intercellular spaces (Fig. 3A, panel iv). These TVS were also found in purified LV preparations (Fig. 3A, panel i) and were far more abundant than actual virus particles. TVS were not found upon transfection with the plasmids encoding viral RNA, Gag-Pol, and Rev (Fig. 3A, panels iii and vi) but were generated after VSV-G transfection (Fig. 3A, panels ii and v). TVS were also detectable by flow cytometry as a population with distinct scatter properties (data not shown). Thus, VSV-G-pseudotyped LV preparations, as well as supernatants from VSV-G-transfected cells, contain large amounts of TVS.

FIG. 3.

VSV-G transfection induces formation of TVS that carry cellular proteins and constitute the main pDC stimulus in VSV-G-pseudotyped LV preparations. (A) Electron micrographs of supernatants (left column) and transfected cells (right column) producing VSV-G-pseudotyped LV, VSV-G only, or LV without envelope protein (LVΔenv). Bars, 200 nm (left column) and 1 μm (right column). Black arrows indicate TVS; white arrows show virus particles. *, cell cytoplasm. (B) Aliquots of preparations of LVs, TVS, and LVΔenv were blotted for the indicated proteins. (C) VSV-G-LV preparations were fractionated on a continuous sucrose gradient. Each fraction was assessed for IFN-α induction in BM cells and for virus titer (GFU) on 293T cells. Western blots show the amounts of VSV-G and HIV p24 in each fraction. Fraction 8 represents the bottom of the gradient (dense fraction).

We asked whether VSV-G was associated with TVS. Western blots showed the presence of VSV-G in preparations of LVs and TVS but not in control preparations of LVΔenv (Fig. 3B). Notably, other proteins derived from the producer cells, including GFP and β-actin, were also detectable in LV and TVS preparations but not in preparations of LVΔenv (Fig. 3B and data not shown). We conclude that TVS contain both VSV-G and proteins of cellular origin.

We next tried to determine whether virus particles and/or TVS were responsible for the interferon-inducing activity of LV preparations. We fractionated these preparations by continuous sucrose gradient ultracentrifugation and tested each fraction for the presence of virus, as well as for interferon-inducing activity. As can be seen in Fig. 3C, the interferonegenicity of LV preparations was found within a limited number of fractions, all of which contained both TVS and infectious virus (Fig. 3C, fractions 4 through 7). These fractions also contained the bulk of the VSV-G and p24 protein (Fig. 3C). In contrast, fraction 8 did not induce interferon and contained little VSV-G despite having comparable levels of infectious virus. The latter observation suggests that actual virus particles do not act as the major stimulus for IFN-α production by BM cells. Furthermore, the data indicate that virions contribute only a small fraction of the VSV-G within LV preparations, even though this small amount of protein is clearly sufficient to mediate virus entry into cells. We conclude that VSV-G-pseudotyped LV preparations contain large amounts of TVS produced as a by-product of VSV-G transfection and that these TVS constitute the primary stimulus for IFN-α production by pDC.

TVS carry plasmid DNA and are recognized via TLR9.

MyD88-dependent signaling is a major pathway for eliciting interferon production by pDC. Notably, MyD88^−/− BM cells were completely unable to produce IFN-α in response to LV preparations but responded normally to electroporation with the control stimulus poly(I:C), which stimulates IFN-α synthesis via a TLR-independent cytosolic pathway (15) (Fig. 4A). Because the main IFN-α-inducing MyD88-coupled receptors in pDC are TLR7 and TLR9, we next examined their possible involvement. The response to LV preparations was preserved in TLR7^−/− BM cells, whereas that to the control TLR7-dependent stimulus R848 was totally abrogated (Fig. 4B). In contrast, TLR9-deficient BM cells had a markedly diminished ability to produce IFN-α in response to LV preparations or to the TLR9-dependent control stimulus CpG but responded normally to R848 (not shown) or to another TLR7 agonist, loxoribine (Fig. 4C). Similarly, the response to supernatants of cells transfected solely with VSV-G was reduced >100-fold in TLR9^−/− BM cells (Fig. 4C). Therefore, the innate stimulatory properties of TVS are largely dependent on signaling via TLR9.

FIG. 4.

IFN-α induction by LV preparations and TVS is dependent on TLR9 signaling. (A) BM cells from C57BL/6 or MyD88^−/− mice were cultured in medium alone (no stim), with LVs (multiplicity of infection of 0.1) or were electroporated with poly(I:C) (0.5 μg) as described previously (15). IFN-α was measured after overnight incubation. (B) IFN-α from TLR7^−/− or wild-type BM cells after stimulation with LVs (multiplicity of infection of 0.1), R848 (1 μg/ml), or CpG (0.5 μg/ml). (C) BM cells from wild type or TLR9^−/− mice were stimulated with LVs (multiplicities of infection of 0.1 and 0.01), VSV-G TVS (10× and 1× concentrated), Loxoribine (20 mM), or CpG. IFN-α was measured after overnight culture. n.d., not detected. Error bars indicate standard deviations of triplicates.

Because TLR9 is a receptor for DNA, we examined the extent to which this nucleic acid might be present within LV preparations or in supernatants from cells transfected with VSV-G. In both cases, fluorescence-activated cell sorter staining with an appropriate dye revealed the presence of nucleic acids within a population having the scatter properties of TVS (data not shown). RT-PCR and PCR analyses of nucleic acids extracted from LV preparations confirmed the presence of RNA and DNA, respectively, corresponding to each of the plasmids that had been transfected into the producer cells (Fig. 5A and data not shown). DNA for β-actin, presumably derived from the genome of the producer cells, was also found in both preparations (Fig. 5A). In contrast, control LV preparations lacking TVS (LVΔenv) did not generate amplicons for plasmid or β-actin DNA by PCR analysis (Fig. 5A). To confirm the presence of intact plasmids, DNA extracted from LV preparations was used to transform bacteria. DNA corresponding to 1 ml of LV preparation generated five times as many antibiotic-resistant bacterial colonies as 10 pg of the control plasmid pUC19, and each of the plasmids used for virus generation could be recovered from the transformants (Fig. 5B and data not shown). In contrast, no plasmids could be recovered from preparations of LVΔenv (data not shown). Taken together, these experiments indicate that VSV-G-pseudotyped LV preparations generated by transient transfection contain TVS that carry residual plasmid DNA, which may provide the major stimulus for triggering TLR9.

FIG. 5.

TVS and LV preparations generated by transient transfection contain plasmid DNA. (A) RT-PCR and PCR on RNA and DNA isolated from preparations of LVs, LVs lacking the envelope protein (LVΔenv), or TVS (concentrated supernatant from VSV-G-transfected cells). (B) DNA extracted from LV preparations was used to transform bacteria. Ampicillin-resistant individual colonies were picked, and the plasmid DNA extracted and analyzed by digestion with restriction enzymes. The graph shows the frequency of bacterial colonies containing each plasmid (average ± standard deviation from three experiments; n > 150 colonies). (C) LV preparations generated by retroviral transduction do not stimulate IFN-α production by BM cells. VSV-G was introduced into STAR-HV cells by transient transfection or by retroviral transduction. Supernatant containing VSV-G-pseudotyped LV (multiplicity of infection of 0.1) was added to BM cells. IFN-α was measured by ELISA 16 h after treatment. CpG (0.5 μg/ml) was used as a positive control. n.d., not detected.

Avoiding transient transfection of producer cells decreases the innate stimulatory potential of LV preparations.

To confirm that the interferon-inducing potential of LV preparations is largely determined by plasmid contamination itself, we resorted to a method for LV production that avoids exposure of the producer cells to DNA. We made use of STAR-HV cells, which contain an HIV-based packaging vector encoding GFP and stably express HIV Gag, Pol, and Rev proteins, needing only an exogenously provided Env gene in order to secrete infectious LVs encoding GFP (32).

VSV-G gene introduction by either plasmid transfection or retroviral transduction allowed LV production by STAR-HV cells as determined by titration on 293T cells (data not shown). LV preparations produced by regular plasmid transfection induced IFN-α production, as expected (Fig. 5C). In contrast, no IFN-α was induced by LV preparations containing the same viral titer but generated by retroviral transduction cells and not exposed to plasmid DNA (Fig. 5C). We conclude that the interferon-inducing ability of LV preparations clearly depends on the method of virus generation and not on the viral particles themselves.

TVS act as adjuvants in vivo.

Given that LV preparations can act as immunogens but their innate stimulatory properties in vitro appear highly dependent on the presence of TVS contaminants, we wondered whether TVS were capable of acting as adjuvants in vivo. Mice immunized with egg white, a source of adjuvant-free ovalbumin (7), did not generate a CD8⁺ T-cell response, as noted by background levels of expansion of tetramer-positive cells and a lack of killing of OVA peptide-pulsed targets in vivo (Fig. 6A and B). In contrast, egg white mixed with TVS induced marked expansion of tetramer-positive cells, which was associated with OVA-specific cytotoxic activity and nearly comparable to the response induced by a control adjuvant, poly(I:C) (Fig. 6A and B). When spleen cells were restimulated in vitro, a strong CD4⁺ Th1 response to FCS components was observed in mice that had received TVS (Fig. 6C). This correlated with the presence of anti-FCS antibodies in the mice (Fig. 6D), indicating in vivo priming of a Th- and B-cell response against TVS-associated FCS proteins. Although anti-OVA CD4⁺ T cells were not detectable by ex vivo restimulation (data not shown), anti-OVA immunoglobulin G was found in mice receiving egg white plus TVS or egg white + poly(I:C), showing that either adjuvant had also led to induction of an anti-OVA CD4⁺ T-cell and B-cell response (Fig. 6D). We therefore conclude that TVS generated by transient transfection with VSV-G can act as adjuvants to promote adaptive immune responses against associated antigens.

FIG. 6.

TVS act as adjuvants to induce adaptive immune responses. C57BL/6 mice were immunized intraperitoneally with egg white in PBS (n = 2) or with added TVS (n = 4) or poly(I:C) (PIC) (n = 2). (A) Contour plots show OVA/H-2K^b tetramer-positive Thy1.2⁺ cells in blood of representative mice 1 week after immunization. The graph shows the average (±standard deviation) frequency of OVA/H2-K^b tetramer-positive Thy1.2⁺ cells for all mice. *, P < 0.05; (as determined by Student's t test, compared to PBS-OVA-immunized mice). (B) At 10 days after immunization, mice were challenged with congenic CD45.1 splenocytes loaded with 20 nM (carboxyfluorescein [CFSE] low), 200 nM (CFSE intermediate), or 0 nM (CFSE high) of OVA peptide (SIINFEKL). Histograms show target cells (gated on CD45.1) from representative mice at 48 h after injection. The graph shows the amount of specific killing (average ± standard deviation) in all groups from one representative experiment. *, P < 0.05; **, P < 0.001 (as determined by Student's t test, compared to PBS-OVA-immunized mice). n.d., not detected. (C) At 12 days after immunization, splenocytes were isolated, cultured overnight in the absence or presence of FCS, and stained for intracellular IFN-γ. Contour plots show gated Thy1.2⁺ cells from representative mice. Values in the bottom right panel ranged from 0.16 to 2.2% IFN-γ⁺ Thy1.2⁺ CD4⁺ cells. (D) At 12 days after immunization, sera of mice were tested for the presence of specific antibodies against OVA and FCS. Data are displayed as titration curves from individual representative mice. IgG, immunoglobulin G; OD, optical density.

DISCUSSION

Recombinant LVs are used extensively to transduce nondividing cells, and they constitute attractive vehicles for gene therapy. However, although less immunogenic than other viral vectors, LVs can elicit immune responses in vivo, limiting their clinical application in gene delivery (17, 20, 49, 52). The immunogenicity of LVs has been attributed to their ability to transduce dendritic cells (20, 49), although it is not clear why this factor alone should be sufficient, as antigen targeting to DC in the absence of exogenous stimuli promotes tolerance rather than immunity (58, 59). Therefore, it seems likely that the ability of LVs to prime immune responses reflects their ability to deliver both antigen and innate stimuli for DC activation (52). The ability of recombinant LVs to activate conventional DC is controversial (8, 10, 16, 19, 20, 28, 29, 56, 61). In contrast, their capacity to stimulate pDC has not been assessed, despite the potential importance of this cell type in priming antiviral responses. Here, we report that standard VSV-G-pseudotyped LVs prepared by transient-transfection methods activate pDC, resulting in high levels of IFN-α production. However, most of this activity is attributable not to the viral particles themselves but to contaminating tubulovesicular structures, which largely outnumber virions and carry nucleic acids, including the plasmids used in transfection. The observation that TVS carry DNA may account for previous reports of plasmid DNA presence within LV preparations (54) and implies that TVS can activate pathways of innate defense against DNA viruses. Consistent with this notion, we show that TVS induce strong innate responses via a TLR9-dependent pathway and act as adjuvants in vivo for coadministered antigens.

TVS may be similar to microvesicles found within HIV preparations, which appear to be the actual source of many cellular proteins previously thought to be associated with the virus envelope (5, 26, 64). Like those microvesicles, TVS carry proteins of producer cell origin, including plasmid-encoded ones (VSV-G and GFP) (Fig. 3B and data not shown), proteins of cellular origin (β-actin) (Fig, 3B), and proteins derived from the culture medium (FCS components) (Fig. 6C and D). Notably, TVS can mediate transfer of some of these proteins (e.g., GFP) into target cells (A. Pichlmair and C. Reis e Sousa, unpublished observations), likely explaining the phenomenon of “pseudotransduction” observed with LV preparations (46). The actual origin of TVS and microvesicles is unclear, although the presence of cell-derived vesicular structures within supernatants of cultured cells is a long-established phenomenon (13). TVS are unlikely to be apoptotic bodies, because supernatants harvested early after transfection were more potent IFN-α inducers yet cell death was more prominent at later time points (data not shown). Furthermore, deliberate induction of apoptosis in cells transfected with MLV-A, THOV-G, or LCMV-G Env genes does not generate supernatants capable of stimulating interferon production from BM cells (data not shown). Similarly, TVS are unlikely to be exosomes, which have a different appearance by electron microscopy (21). We favor the possibility that TVS are distinct structures, which are actively induced by VSV-G overexpression. Consistent with this notion, VSV-G is known to promote budding of vesicular stomatitis virus (9), and an alphavirus replicon encoding VSV-G but not viral capsid proteins spontaneously generates infectious virus-like particles (51).

VSV-G not only is involved in TVS induction but also is necessary for TVS to act as a pDC stimulus, as demonstrated by monoclonal antibody blocking studies (Fig. 2C). This, together with the fact that VSV-G was unique among Env gene products (Fig. 2A), as well as sufficient (Fig. 2B), for generating interferon-stimulating activity in concentrated supernatants, initially led us to focus on the notion that the VSV-G itself was responsible for stimulating pDC. We now think that this is unlikely, because fractionation studies show that the presence of VSV-G does not absolutely correlate with interferon induction potential. This can be seen in Fig. 3C, in which fraction 3 contains immunodetectable VSV-G but does not stimulate IFN-α production, likely because it contains little DNA (data not shown). The TLR9 dependence of the innate stimulatory activity further suggests that TVS-associated DNA rather than VSV-G constitutes the actual stimulus. Therefore, the most likely explanation for our results is that VSV-G acts to promote binding and uptake of TVS by pDC, effectively concentrating and delivering these structures to endocytic compartments where their DNA content is sensed by TLR9 and initiates signaling via MyD88. In this regard, TVS mimic DNA viruses such as herpes simplex virus types 1 and 2, which trigger pDC via a similar mechanism (39, 43). Notably, preparations of LV pseudotyped with other viral glycoproteins (THOV-G, LCMV-G, or MLV-A) do not act as strong pDC stimuli (Fig. 2A). These glycoproteins may not induce TVS formation or may not permit TVS enrichment during viral concentration. Alternatively, THOV-G-, LCMV-G-, or MLV-A-coated TVS may be excluded from TLR9-accessible cellular compartments.

Human pDC exposed to HIV-1 produce IFN-α and display signs of maturation (4, 22, 23, 55, 67). Therefore, it is surprising that, in the absence of TVS, the recombinant LVs tested here elicit only very weak IFN-α responses from murine pDC. Although this may reflect intrinsic differences between human and mouse pDC in their responsiveness to lentiviruses, it also shows that lentiviral particles are only poorly stimulatory compared to other ssRNA viruses such as influenza virus, Sendai virus, or VSV, all of which trigger strong murine pDC activation via TLR7 (14, 44). However, recombinant LV particles are not totally inert and promote low levels of cytokine secretion by pDC, probably via a TLR7-dependent pathway (data not shown). This may especially be relevant for responses in vivo, where opsonization of the virus could increase pDC stimulation (48). Indeed, antibodies present in sera of HIV-infected patients can enhance IFN-α production if added together with HIV-1 to peripheral blood mononuclear cells (27), an effect likely explained by increased FcR-mediated internalization of HIV particles into pDC endosomes and recognition via TLR7. Nevertheless, we have recently found that LV preparations can still induce CTL priming in MyD88-deficient mice (Pichlmair and Reis e Sousa, unpublished observations). Thus, there are immunostimulatory properties of LVs in vivo that cannot be attributed to pDC or TLR7/9 stimulation, and, indeed, we do not exclude the possibility that LV and TVS may act in a synergistic manner in some instances. The full elucidation of the mechanisms regulating LV immunogenicity will have to wait further developments in the rapidly evolving field of antiviral innate immunity (36).

Our results on the innate stimulatory potential of LV preparations have important practical implications for the use of recombinant LVs. There are many examples in which innate responses to vectors have adverse effects. For adenovirus vectors it has been noted that induction of cytokines limits high expression of delivered genes (6). Similarly, interferons reduce gene expression from retroviral vectors, which might contribute to the relatively poor performance of such vectors in vivo (25). Delivery of VSV-G-pseudotyped LVs to mouse brain induces local inflammation and a systemic immune response which can limit transduction efficiency (3). Therefore, gene therapy approaches, as well as in vitro transduction of immune cells, would likely benefit from having LVs with low innate stimulatory potential. In this regard, our results suggest that pseudotyping with THOV-G, LCMV-G, or MLV-A may be preferable to that with the more widely used VSV-G. This needs to be weighed against the fact that THOV-G, LCMV-G, or MLV-A also results in production of LV preparations containing lower virus titers than those obtained by VSV-G-pseudotyping (references 11 and 65 and data not shown). Titration of the VSV-G plasmid proved not to be useful, as it resulted in a dramatic drop of virus titer without avoiding IFN-α induction from BM (data not shown). If VSV-G pseudotyping is to be used, the immunogenic potential of virus preparations may be decreased by generation of virus in stable producer cell lines by using retroviral transduction (Fig. 5C) or inducible VSV-G expression (47) or by careful purification of virus particles. We have tried to formally correlate immunogenicity with virus production methods but have been unable to purify large enough quantities of virus by gradient fractionation to set up animal experiments, and we have yet to obtain a stable cell line producing LVs encoding a suitable model antigen (data not shown). However, it has been reported that sucrose gradient fractionation decreases LV immunogenicity (3). More work will be needed to see how different virus generation and purification methods affect contamination by TVS and immune responses.