Toll-like receptor 2–mediated innate immune response in human nonparenchymal liver cells toward adeno-associated viral vectors


  • Potential conflict of interest: Nothing to report.


Adeno-associated viral vectors (rAAV) are frequently used in gene therapy trials. Although rAAV vectors are of low immunogenicity, humoral as well as T cell responses may be induced. While the former limits vector reapplication, the expansion of cytotoxic T cells correlates with liver inflammation and loss of transduced hepatocytes. Because adaptive immune responses are a consequence of recognition by the innate immune system, we aimed to characterize cell autonomous immune responses elicited by rAAV in primary human hepatocytes and nonparenchymal liver cells. Surprisingly, Kupffer cells, but also liver sinusoidal endothelial cells, mounted responses to rAAV, whereas neither rAAV2 nor rAAV8 were recognized by hepatocytes. Viral capsids were sensed at the cell surface as pathogen-associated molecular patterns by Toll-like receptor 2. In contrast to the Toll-like receptor 9–mediated recognition observed in plasmacytoid dendritic cells, immune recognition of rAAV in primary human liver cells did not induce a type I interferon response, but up-regulated inflammatory cytokines through activation of nuclear factor κB. Conclusion: Using primary human liver cells, we identified a novel mechanism of rAAV recognition in the liver, demonstrating that alternative means of sensing rAAV particles have evolved. Minimizing this recognition will be key to improving rAAV-mediated gene transfer and reducing side effects in clinical trials due to immune responses against rAAV. (Hepatology 2012;55:287–297)

Vectors based on the adeno-associated virus (AAV) have emerged as one of the leading delivery systems in gene therapy.1 One key advantage of rAAV vectors is their low immunogenicity. Nevertheless, upon vector application humoral immune responses are induced which pose a challenge for reapplication. Furthermore, in patients receiving intrahepatic vector injection, expansion of capsid-specific CD8+ T cells was observed that was likely responsible for destruction of rAAV-transduced cells and the loss of therapeutic efficacy.2 Adaptive immune responses depend on the detection of microorganisms via pathogen-associated molecular patterns (PAMPs) recognized by germline-encoded pattern-recognition receptors (PRRs). Among the critical sensors of viruses are Toll-like receptors (TLRs), retinoic acid-inducible gene I–like receptors and members of the nucleotide-binding and oligomerization domain like receptor family.3 Upon ligand binding, signal transduction pathways are activated leading to the induction of inflammatory cytokines, chemokines and/or interferons (IFNs).3 For AAV, the only PRR identified so far is TLR9, a PRR that senses DNA in endosomes.4 This sensor is highly expressed on plasmacytoid dendritic cells (pDCs) that produce high levels of type I IFN in response to rAAV infection.4

Approximately 70% of the liver is composed of hepatocytes, whereas the remaining nonparenchymal liver cells (NPCs) are mainly represented by liver sinusoidal endothelial cells (LSECs, ∼15%), lining the hepatic sinusoids, and Kupffer cells (KCs, ∼10%), the liver resident macrophages. Both hepatocytes as well as NPCs mount innate immune responses upon challenge with microorganisms, but differ in the repertoire of PRRs and consequently in the PAMPs that can be sensed.3, 5 To date, neither PRR(s) nor PAMP(s) have been identified recognizing rAAV in the liver, although this tissue is one of the main targets for human gene therapy.

Because rodents and humans differ in their PRR composition, murine models may not adequately reflect the situation in humans. We therefore investigated cell autonomous immune responses in primary human liver cells using mixed liver cell cultures and monocultures of hepatocytes, KCs and LSECs, respectively, as model systems. We found that (1) KCs as well as LSECs, but not hepatocytes are capable of sensing rAAV, (2) AAV capsids represent PAMPs and are sensed by TLR2 on KCs and LSECs, and (3) AAV-mediated innate immune activation in NPCs triggers the production of inflammatory cytokines and chemokines, but—in contrast to pDCs—no type I IFN response. In addition, we demonstrated for the first time that macrovascular endothelial cells are also capable of sensing AAV capsids following stimulation by proinflammatory cytokines that cause TLR2 up-regulation. These novel insights on the mechanism and consequence of innate immune recognition of AAV are crucial for overcoming the current limitations of AAV-mediated gene therapy.


AAV, adeno-associated virus; ELISA, enzyme-linked immunosorbent assay; FBS, fetal bovine serum; GOI, genomic particle per cell; HEK, human embryonic kidney; HKLM, heat-killed Listeria monocytogenes; HUVEC, human umbilical vein endothelial cell; IFN, interferon; IL, interleukin; KC, Kupffer cell; LSEC, liver sinusoidal endothelial cell; mRNA, messenger RNA; NFκB, nuclear factor κB; PAMP, pathogen-associated molecular pattern; pDC, plasmacytoid dendritic cell; PHLC, primary human liver cell; PRR, pattern-recognition receptor; SD, standard deviation; TNFα, tumor necrosis factor α; TLR, Toll-like receptor.

Patients and Methods

Reagents and Cell Lines.

Alexa Fluor 488–conjugated Escherichia coli (E. coli) were purchased from Invitrogen (Paisley, UK), ODN2395-CpG oligonucleotides (ODN2395), heat-inactivated (“killed”) Listeria monocytogenes (HKLM), and the cell lines human embryonic kidney 293 (HEK293XL)/hTLR9 and HEK293/hTLR2-CD14 were obtained from Invivogen (San Diego, CA). Cell lines were cultivated according to the manufacturer's instructions.

Primary Cells.

Primary human liver cell (PHLC) cultures were prepared from surgical noncancerous human liver biopsies according to the guidelines of the ethics committee of the Technical University of Munich (Germany) and after written informed consent of patients.6 PHLC cultures contained ≥ 85% hepatocytes and 3%-15% NPCs, mainly KCs and LSECs and were cultivated as described.6 For isolation of KCs and LSECs, a cell suspension obtained from a 50g-differential centrifugation step during PHLC preparation was used (Broxtermann and Protzer, unpublished data). KCs and LSECs were separated as one fraction by OptiPrep density gradient centrifugation (Axis-Shield, Oslo, Norway) and transferred to cell culture plates. Adherent KCs were carefully washed in phosphate-buffered saline. Nonadherent LSECs were further purified using the CD31-MicroBead-Kit (Miltenyi Biotec, Bergisch Gladbach, Germany). KCs and LSECs were maintained in PHLC medium with 10% fetal bovine serum (FBS), after 24 hours with 5% FBS, and after 72 hours without FBS. The purity of KCs was examined using primary mouse monoclonal anti-CD68 antibodies (Dako, Stockport, UK) and goat anti-mouse secondary antibodies conjugated to Alexa Fluor 594 (Invitrogen), whereas LSECs were examined for uptake of acetylated low-density lipoprotein conjugated to Alexa Fluor 488 (Invitrogen).6

Primary human hepatocytes (Lonza, Basel, Switzerland) and primary human umbilical vein endothelial cells (HUVECs; Promocell, Heidelberg, Germany) were cultivated according to the manufacturer's instructions. For prestimulation treatment, HUVECs were incubated with 100 ng/mL of recombinant HumaXpress tumor necrosis factor α (TNFα; Humanzyme, Chicago, IL) as described.7

Production of rAAV, Empty Capsids, and Mock-Inoculum.

rAAV, empty capsids and mock-inoculum were produced in HEK293 cells (ATCC number: CRL-1573).8 For production of rAAV, the plasmids pXX6-80,9 pscAAV/EGFP8 or pscLuci, and pRC8 or pXR810 were used. All rAAV vectors contained a self-complementary vector genome. Empty capsids were produced by transfecting pXX6-80 and pRC or pXR8, whereas for production of mock-inoculum, pXX6-80 and pscLuci were used.

All preparations were negative for Mycoplasma contamination (Venor-GeM Mycoplasma detection kit; Minerva Biolabs, Berlin, Germany) and contained less than 0.005 EU of endotoxin/mL (Limulus amoebocyte lysate kinetic chromogenic assay; Lonza). Furthermore, preparations were assayed for DNA contamination (Supporting Information).

Genomic particle titers were determined by real time LightCycler (LC) polymerase chain reaction (PCR) (Roche Diagnostics, Mannheim, Germany) using transgene-specific primers (Supporting Table 1). Cells were incubated with rAAV using the indicated genomic-particle-per-cell (genomic particle of infection [GOI]) ratios. As a negative control, cells were incubated with mock-inoculum using a volume equal to the highest volume applied for rAAV.

Quantification of Vector Uptake.

Total DNA was extracted by the QIAamp MinElute Kit (Qiagen, Hilden, Germany) after a 5-minute-treatment of cells with 0.05% trypsin followed by three washing steps with phosphate-buffered saline. Vector genomes and single-copy reference gene (plasminogen activator; PLAT) were quantified (LC System and Quantification Software, Roche Diagnostics).

Protein Analyses.

Interleukin-8 (IL-8) in cell culture supernatants was quantified by enzyme-linked immunosorbent assay (ELISA) (BD Biosciences, San Diego, CA). To detect nuclear factor κB (NFκB) activation, 2 μg of nuclear proteins prepared with NE-PER reagent (Pierce, Rockford, IL) were analyzed using the EZ-Detect kit for NFκB p50 (Pierce).5 Luciferase activity was measured using the Renilla Luciferase Assay (Promega, Madison, WI). The levels were normalized to the total protein concentrations determined by BCA protein assay (Pierce).

Quantification of Gene Expression.

Total RNA was isolated, deoxyribonuclease I–digested, and reverse-transcribed using First-Strand Synthesis Supermix (Invitrogen). Gene expression levels were determined (LC System and Quantification Software, Roche Diagnostics) using the primers listed in Supporting Table 1.

The expression of genes involved in human innate and adaptive immune responses was examined using the RT2Profiler PCR Array (SA Biosciences, Frederick, MD) on an LC480 System (Roche Diagnostics). Data analysis was performed by ΔΔCt method. Transcripts were considered to be not detectable when the threshold cycle (Ct) value was ≥35.

Statistical Analysis.

Statistical calculation was performed by Student t test. A P < 0.05 was considered statistically significant.


NPCs, but Not Hepatocytes Respond to rAAV by Activation of NFκB and Induction of Inflammatory Cytokines.

Recombinant AAV2 and rAAV8 are widely employed in clinical trials to overexpress therapeutic genes in hepatocytes. When applying both serotypes to highly purified primary human hepatocytes, long-term transgene expression could be achieved albeit with significant differences in entry efficiency and level of transgene expression (Fig. 1).

Figure 1.

rAAV transduction of primary human hepatocytes. Hepatocytes (5 × 105) were transduced with rAAV2 or rAAV8. (A) At 3 hours post infection (p.i.), total DNA was isolated from hepatocytes transduced at a GOI of 1 × 103. The number of intracellular vector genomes was quantified by quantitative PCR and normalized to PLAT. Values represent copy number per cell ± standard deviation (SD) (n = 3). (B) Luciferase activity (in relative light units [RLU] ± SD) was determined in triplicates at the indicated time points and normalized to the total amount of protein.

Although hepatocytes are not classical immune cells, they respond to microbial products including lipopolysaccharide.3, 11 When primary human hepatocytes were challenged with E. coli, a 28-fold increase in IL-8 and a 17-fold increase in TNFα gene expression, as well as IL-8 secretion were observed. In contrast, neither IL-8 nor TNFα expression was found to be up-regulated in rAAV-transduced hepatocytes and no IL-8 was detected in cell culture supernatants (data not shown) indicating that rAAV does not induce innate immune responses in hepatocytes.

Hepatocytes represent the most abundant, but not the only cell type in the liver. We therefore investigated mixed primary human liver cell (PHLC) cultures, consisting of hepatocytes and NPCs (Fig. 2A), for their ability to sense rAAV. Although gene expression was readily detectable for rAAV2 (but not rAAV8) at a GOI of 1 × 103 (Fig. 2B), neither rAAV2 nor rAAV8 induced an innate immune response (Fig. 2C,D). Increasing the GOI to 1 × 105, however, caused activation of NFκB (Fig. 2C) and induced the expression of proinflammatory cytokines (Table 1 and Fig. 2D). Compared to highly immunogenic E. coli, the induced immune response toward rAAV was lower and declined faster (Table 1). Notably, expression of the type I-IFN–inducible 2′,5′-oligoadenylate synthetase (2′5′-OAS) gene was not influenced by rAAV, whereas this gene was up-regulated seven-fold in E. coli-challenged cells (data not shown).

Figure 2.

rAAV transduction of PHLC cultures. (A) Phase contrast image of PHLC cultures (400-fold magnification): hepatocytes (Hep), Kupffer cells (KCs), liver sinusoidal endothelial cells (LSECs). Cells (1 × 106) were transduced with rAAV2 or rAAV8 or incubated with the mock-inoculum. (B) Relative transgene expression levels were determined by RT-qPCR at 24 hours p.i. and normalized to the aminolevulinate synthase 1 (ALAS1) reference gene. Data from three independent PHLC culture preparations ± SD are shown. For better comparison, transgene expression levels of “rAAV2-103-sample” in each experiment were set to one. (C) Activation of NFκB in PHLC cultures transduced with rAAV2 and rAAV8 at 3 hours p.i. Nuclear extracts from E. coli-treated cultures (100 particles per cell) were used as positive controls. The results are expressed as fold-increase (median ± SD of two independent experiments, each performed in triplicate) over mock-treated cells; **P < 0.005. (D) IL-8 was measured of supernatants from rAAV-transduced PHLC cultures at the indicated time points. Median values ± SD of two independent experiments (each performed in triplicate) are given. rAAVs encoded for enhanced green fluorescent protein (GFP).

Table 1. Up-Regulation of Cytokine Genes in AAV-Infected PHLC Mixed Cultures
GeneFold Increase in Gene Expression*
rAAV2rAAV8E. coli
6 hours p.i.24 hours p.i.6 hours p.i.24 hours p.i.6 hours p.i.24 hours p.i.
  • *

    Fold increase: transduced versus mock-treated cells.

  • rAAV: GOI of 1 × 105, E. coli: 100 particles per cell.

  • nc, no change in gene expression.


In order to better characterize host responses toward rAAV, we performed microarray analyses of genes critically involved in innate and adaptive immune responses. Table 2 shows the gene clusters for which altered expression patterns were observed. Overall, these analyses revealed that both, rAAV2 and rAAV8, applied at a GOI of 1 × 105, caused limited but clearly detectable changes in gene expression. Considering a two-fold change in gene expression as significant, TLR2 was up-regulated, whereas TLR3, TLR6 (for rAAV2) and in particular TLR9 were down-regulated at the messenger RNA (mRNA) level. Furthermore, we observed activation of genes involved in NFκB signaling and inflammatory responses including members of the IL-1 pathway (IL-1α, IL-1β, IL-1-like receptor 2, IL-1 receptor antagonist and IL-1 receptor associated kinase 2), the proinflammatory cytokines TNFα and IL-6, the anti-inflammatory cytokine IL-10 (rAAV8), chemokine (C-C motif) ligand 2 (CCL2 = MCP-1), chemokine (C-C motif) receptor 3 (CCR3), inducible nitric oxide synthase 2 (NOS2A), C-reactive protein (CRP), inflammatory caspase-4 (CASP4), adenosine A2a receptor (ADORA2A), and defensin beta 4 (DEFB 4). Consistent with the unaltered expression of the 2′5′-OAS gene observed in our real-time quantitative PCR (reverse transcription (RT)-qPCR) analysis, expression of type I IFN and IFN-responsive genes was unchanged (Table 2). This leads to the conclusion that in contrast to pDCs,4 this pathway is not activated by rAAV in PHLC cultures.

Table 2. Expression of Selected Genes Involved in the Human Innate and Adaptive Immune Response
Functional Gene GroupingsGeneDescriptionPublic IDFold Change in Gene Expression*
  • *

    All fold change values were calculated relative to mock-treated control. The ≥10.0-fold change values can be roughly estimated, because the Ct-values of the mock-control were below the detection limit.

  • nd, gene expression not detectable.  Significant changes in gene expression are indicated in bold.

TLR members
 TLR1Toll-like receptor 1NM_0032631.10.8
 TLR2Toll-like receptor 2NM_0032642.52.0
 TLR3Toll-like receptor 3NM_0032650.50.5
 TLR4Toll-like receptor 4NM_1385540.80.7
 TLR6Toll-like receptor 6NM_0060680.51.1
 TLR8Toll-like receptor 8NM_138636ndnd
 TLR9Toll-like receptor 9NM_0174420.20.1
 TLR10Toll-like receptor 10NM_030956ndnd
Cytokines, chemokines and their receptors
 IL1AInterleukin 1, alphaNM_00057510.010.0
 IL1BInterleukin 1, betaNM_00057610.010.0
 IL1F5Interleukin 1 family, member 5 (delta)NM_012275ndnd
 IL1F6Interleukin 1 family, member 6 (epsilon)NM_014440ndnd
 IL1F7Interleukin 1 family, member 7 (zeta)NM_173205ndnd
 IL1F8Interleukin 1 family, member 8 (eta)NM_173178ndnd
 IL1F9Interleukin 1 family, member 9NM_019618ndnd
 IL1F10Interleukin 1 family, member 10 (theta)NM_173161ndnd
 IL1R1Interleukin 1 receptor, type INM_0008770.80.9
 IL1R2Interleukin 1 receptor, type IINM_0046330.70.8
 IL1RL2Interleukin 1 receptor-like 2NM_0038544.22.7
 IL1RNInterleukin 1 receptor antagonistNM_0005772.11.9
 TNFTumor necrosis factor (TNF superfamily, member 2)NM_00059416.922.4
 TGFB1Transforming growth factor, beta 1NM_0006601.01.0
 IL6Interleukin 6 (interferon, beta 2)NM_00060010.010.0
 IL10Interleukin 10NM_0005721.52.9
 IL12RB2Interleukin 12 receptor, beta 2NM_0015591.70.7
 IFNA1Interferon, alpha 1NM_024013ndnd
 IFNB1Interferon, beta 1, fibroblastNM_002176ndnd
 CCL2Chemokine (C-C motif) ligand 2NM_00298217.314
 CCR3Chemokine (C-C motif) receptor 3NM_0018373.23.1
 CXCR4Chemokine (C-X-C motif) receptor 4NM_0034670.70.7
 TREMTriggering receptor expressed on myeloid cells 1NM_018643ndnd
 PPBPPro-platelet basic protein (chemokine [C-X-C motif] ligand 7)NM_002704ndnd
Other genes involved in the inflammatory response and NFκB signaling
 IRAK1Interleukin-1 receptor-associated kinase 1NM_0015690.91.0
 IRAK2Interleukin-1 receptor-associated kinase 2NM_0015702.02.0
 MYD88Myeloid differentiation primary response gene (88)NM_0024680.70.7
 NFKB1Nuclear factor of kappa light polypeptide gene enhancer in B-cells 1NM_0039981.81.6
 NFKBIANuclear factor of kappa light polypeptide gene enhancer in B-cells inhibitor, alphaNM_0205291.21.1
 IKBKBInhibitor of kappa light polypeptide gene enhancer in B-cells, kinase betaNM_0015560.70.6
 MAPK14Mitogen-activated protein kinase 14NM_0013151.00.7
 MAPK8Mitogen-activated protein kinase 8NM_0027501.81.6
 NOS2ANitric oxide synthase 2, inducibleNM_00062524.421.2
 DEFB4Defensin, beta 4NM_00494210.010.0
 CRPC-reactive protein, pentraxin-relatedNM_00056716.125
 C5Complement component 5NM_0017350.60.6
 C8AComplement component 8, alpha polypeptideNM_0005620.60.6
 CASP4Caspase 4, apoptosis-related cysteine peptidaseNM_0012252.73.1
 CD14CD14 moleculeNM_0005910.91.0
 CYBBCytochrome b-245, beta polypeptideNM_0003970.70.7
 IRF1Interferon regulatory factor 1NM_0021981.41.1
 IFNGR1Interferon gamma receptor 1NM_0004161.31.5
 IFNGR2Interferon gamma receptor 2 (interferon gamma transducer 1)NM_0055341.11.0
 ADORA2AAdenosine A2a receptorNM_0006752.21.7
Innate immune response
 COLEC12Collectin subfamily member 12NM_1303860.60.5
 PGLYRP1Peptidoglycan recognition protein 1NM_0050910.81.0
 PGLYRP2Peptidoglycan recognition protein 2NM_0528900.50.8
 PGLYRP3Peptidoglycan recognition protein 3NM_052891ndnd
 SFTPDSurfactant protein DNM_0030190.40.4
 CD1DCD1d moleculeNM_001766nd0.7

Together, these results demonstrate that rAAV2 and rAAV8 induce innate immune responses in PHLC cultures upon reaching a certain threshold concentration. Sensing likely occurs through NPCs because pure hepatocytes did not respond to challenge by rAAV.

AAV Capsids Function as PAMPs.

The comparable level and profile of immune activation (Tables 1 and 2) induced by both serotypes despite their significant differences in the number of intracellular particles (Supporting Fig. 1) argued against recognition of viral vector genomes by TLR9 as the main cause of the observed innate immune responses. In keeping with this hypothesis, we aimed to determine whether the AAV capsid could function as PAMP for NPCs. We therefore generated empty capsids of rAAV2 and rAAV8. In the absence of a suitable capsid ELISA for rAAV8, we estimated the amount of capsids by western blotting (Supporting Fig. 2). Furthermore, we carefully assayed all preparations for the presence of endotoxin, Mycoplasma, or DNA contaminations: Although neither endotoxin nor Mycoplasma or cellular DNA was detected, we detected trace amounts of plasmid DNA in all preparations including the mock-inoculum. Based on our assay conditions, this contamination would result in the transfer of <0.24 plasmid copies per cell (Supporting Table 2).

In order to determine whether this DNA contamination could activate TLR9, we tested our preparations on HEK293XL/hTLR9 cells, stably expressing TLR9, using up-regulation of IL-8 gene expression as read-out. In contrast to the TLR9-agonist ODN2395 and to the vector preparations (rAAV2 and rAAV8), neither the AAV2 nor the AAV8 capsid preparation induced IL-8 expression (Fig. 3).

Figure 3.

Challenge of HEK293XL-hTLR9 cells with empty capsid preparations. HEK293XL-hTLR9 cells (1 × 106) were incubated with empty capsid preparations using an amount of capsids equivalent to rAAV transductions at a GOI of 1 × 105. In addition, cells were transduced with rAAV2 and rAAV8 (GOI of 1 × 105) or challenged with 5 μM ODN2395. Relative IL-8 gene expression was determined by RT-qPCR at 3 hours p.i. The results of three independent experiments ± SD are presented as percentage of control (mock-transduced cells) after normalization to the glyceraldehyde-3-phosphate dehydrogenase (GAPDH) reference gene. rAAVs encoded for luciferase. *P < 0.05, **P < 0.005.

NPCs such as KCs and LSECs are copurified with hepatocytes in PHLC cultures (Fig. 2A) and are capable of pathogen recognition.3 To specifically investigate the impact of these NPCs in AAV-induced immune responses, we isolated KCs (Fig. 4A,B) and LSECs (Fig. 4D,E) and transduced them as monocultures with viral vectors or with empty capsids. In support of our hypothesis, empty capsids and vectors induced a comparable release of IL-8 from KCs and LSECs (Fig. 4C,F), whereas negligible (KC) or no (LSEC) IL-8 secretion was detected in cells treated with ODN2395.

Figure 4.

Recognition of rAAV by human NPCs. Purified human KCs (A) and human LSECs (D) are shown in phase contrast images (100-fold magnification). (B) The purity of KCs was evaluated by immunostaining (Alexa Fluor 594). Nuclei were stained with DAPI. (E) LSECs were identified by uptake of acetylated low-density lipoprotein. A total of 3 × 105 purified KCs (C) or 7 × 105 LSECs (F) were transduced with rAAV at a GOI of 105 or with the appropriate amounts of empty capsids. ODN2395 (5 μM), E. coli (100 particles per cell) and mock-inoculum were used as controls. IL-8 was measured in triplicates by ELISA at 6 hours p.i. rAAVs encoded for luciferase.

To further confirm these results, we performed quantitative RT-qPCR analyses for IL-1β, TNFα, IL-6, and—as a control—for IL-8 and found that all four cytokines were up-regulated (Table 3). Notably, with exception of the IL-6 gene in KCs, challenge of KCs and LSECs with empty capsids and rAAV vectors resulted in up-regulation of cytokines to similar levels.

Table 3. Up-Regulation of Cytokine Genes in Human KCs and LSECs Following Incubation with AAV Empty Capsids or rAAV
GeneFold Increase in Gene Expression*
AAV2AAV8E. coli
VectorEmpty CapsidVectorEmpty Capsid
  • *

    Fold increase: transduced versus mock-treated cells, 6 hours p.i.

  • rAAV: GOI of 1 × 105, E. coli: 100 particles per cell.

Kupffer cells     

Collectively, these findings indicate that AAV2 and AAV8 capsids are recognized as PAMPs by human NPCs, and that besides KCs, LSECs are capable of sensing rAAV followed by induction of cell autonomous immune responses.

AAV Capsids Are Recognized by Human TLR2.

TLR2 and TLR4 recognize viral surface structures.12 Although TLR4 is expressed11 and is functional in primary human hepatocytes as demonstrated by their responsiveness to E. coli, we did not observe innate immune activation upon challenge with rAAV strongly arguing against an involvement of this PRR. In contrast, we found that TLR2 expression was up-regulated following challenge of PHLC with rAAV (Table 2). Because up-regulation of TLR2 gene expression is described in response to TLR2-mediated sensing of influenza A viruses,13 we examined whether AAV capsids are recognized by this PRR on HEK293 cells stably transfected with TLR2 and its accessory protein CD14.

In contrast to the parental cell line HEK293, which is devoid of TLR2,14 HEK293/hTLR2-CD14 cells respond to AAV2 as well as AAV8 capsids with a modest but significant up-regulation of IL-8 expression (Fig. 5A) and secretion (Fig. 5B) compared to mock-inoculum-treated cells.

Figure 5.

TLR2-dependent recognition of AAV capsids. (A, B) 3 × 105 HEK293/hTLR2-CD14 cells were treated with empty capsids (amounts equivalent to a GOI of 1 × 105). Cells incubated with HKLM (100 particles per cell) or treated with the mock-inoculum served as positive and negative control, respectively. (A) IL-8 gene expression was determined by RT-qPCR at 3 hours p.i. Values are expressed as percentage of mock-transduced cells ± SD after normalization to the GAPDH reference gene. (B) Secreted IL-8 was quantified by ELISA at 24 hours p.i. Mean ± SD from three experiments is shown. (C, D) 5 × 105 primary human KCs or 5 × 105 LSECs were pretreated with 10 μg/mL of either anti-TLR2 antibody (Ab) or a control antibody (C) for 1 hour or left untreated (–) before addition of capsids (amounts equivalent to a GOI of 1 × 105). TNFα (C) and IL-1β (D) gene expression levels were determined by RT-qPCR at 6 hours p.i. Data representing three independent measurements are shown as percentage of the negative control after normalization to the GAPDH reference gene. (E,F) 3 × 104 HUVEC were left untreated (–) or incubated for 24 hours with 100 ng/mL TNFα (+), followed by a 14-hour TNFα washout with regular medium. Cells were further incubated for 6 hours with a mock-inoculum, HKLM (500 particles/cell) or AAV capsids (amounts equivalent to a GOI of 1 × 105). (E) Total RNAs was isolated and transcribed into complementary DNA. The levels of TLR2, TLR4, TNFα, and GAPDH (internal control) were determined using specific primers and PCR (45 cycles), PCR products were separated on a 2% agarose gel; a representative gel is shown. (F) mRNA levels of E-selectin were determined by RT-qPCR after normalization to the GAPDH reference gene and shown as percentage of untreated mock-infected control (100%). Samples were analyzed in triplicates; a representative experiment out of two is shown. For panels (A) through (D), and (F): *P < 0.05 or **P < 0.005 versus controls are indicated.

To examine a potential involvement of TLR2 in recognition of AAV capsids in human liver cells, we first determined TLR2 expression in hepatocytes and NPCs. The strongest TLR2 expression was measured in KCs (set as 100%), followed by LSECs (13%) and hepatocytes (0.3%). Furthermore, we challenged KCs and LSECs with HKLM, which led to the up-regulation of proinflammatory cytokine genes (Supporting Table 3), demonstrating that the TLR2 signaling pathway is functional in KCs as well as in LSECs.

If AAV capsids are sensed by TLR2, release of proinflammatory cytokines should be attenuated in the presence of inhibitory antibodies. We therefore challenged KCs and LSECs with AAV capsids in the absence and presence of antibodies against TLR2 that inhibited HKLM-induced TNFα production in human monocytes.15 While incubation of KCs with AAV capsids increased TNFα expression 2.3- (AAV2) and 2.6-fold (AAV8), and IL-1β expression 15.1- (AAV2) and 14.3-fold (AAV8) compared to the mock control, preincubation of KCs with anti-TLR2 antibodies significantly hindered this response (Fig. 5C,D). Specifically, TNFα expression was reduced 3.9- (AAV2) and 3.4-fold (AAV8), whereas IL-1β expression dropped 4.2- (AAV2) and 3.1-fold (AAV8).

Similarly, incubation of LSECs with AAV capsids resulted in 2.1- (AAV2) and 2.5-fold (AAV8) up-regulation of TNFα and a 2.8- (AAV2) and 3.6-fold (AAV8) increase in IL-1β gene expression. Again, preincubation with anti-TLR2 antibodies significantly reduced TNFα expression 2.4- (AAV2) and 2.9-fold (AAV8) and IL-1β expression 2.4- (AAV2) and 3.5-fold (AAV8), whereas the isotype control showed no significant inhibitory effect (Fig. 5C,D). Together, these findings strongly argue for TLR2-dependent recognition of AAV2 and AAV8 capsids by NPCs.

Although KCs have been proposed to sense rAAV in a mouse model,16 endothelial cells and in particular microvascular LSECs have not been previously described as sentinel cells for rAAV. In order to confirm that endothelial cells are sentinel cells for rAAV and that sensing occurs through TLR2, but not through TLR4, we expanded our analysis to HUVEC. HUVECs are macrovascular endothelial cells and constitutively express TLR4, whereas TLR2 mRNA and protein are barely detectable.17 Consequently, HUVECs do not respond to TLR2 ligands unless being prechallenged by inflammatory stimuli such as IL-1β or TNFα, which up-regulate TLR2 expression up to 300-fold.17 After removal of inflammatory stimuli, HUVECs return to a basal level of inflammatory responsiveness within 8 hours, but retain their ability to sense pathogens through TLR2.7, 17

Hence, we challenged untreated and TNFα-pretreated HUVECs, the latter following a 14-hour TNFα washout, with AAV2 and AAV8 capsids. Mock-inoculum and HKLM were used as controls. In accordance with the literature, TLR4, but no TLR2 expression was detected in untreated HUVECs, whereas TNFα-pretreated HUVECs expressed TLR4 and TLR2 (Fig. 5E). Neither HKLM nor AAV capsids were recognized in untreated HUVECs. In contrast, expression of TNFα was induced (Fig. 5E) when pretreated HUVECs were challenged with AAV capsids or HKLM.

To confirm and to quantify the innate immune response of TLR2-expressing HUVEC to AAV capsids, we quantified E-selectin expression, a sensitive marker for HUVEC activation.17 RT-qPCR revealed that neither capsids nor HKLM induced a change in the level of E-selectin when incubated with untreated HUVECs. In contrast, TNFα pretreated HUVECs showed an elevated E-selectin level, which was further up-regulated upon challenge with either AAV2 capsids or AAV8 capsids or HKLM (Fig. 5F).

Together, these data confirm our observation, that AAV capsids are recognized as PAMP by TLR2. Furthermore, our findings show for the first time, that micro- and macrovascular endothelial cells are capable of sensing AAV.


rAAV are frequently used in human gene therapy.1 Despite their expected low immunogenicity, induction of adaptive immune responses was held responsible for the unexpected failure of long-term gene expression in a liver-directed clinical trial.2 Because innate responses are key to induce adaptive immunity, we rigorously investigated innate pattern recognition of rAAV in primary human liver cells.

In this first study on innate immune responses toward AAV in the human liver, we demonstrated that rAAV2 and rAAV8 activate NFκB-mediated immune responses upon interaction with NPCs. We observed up-regulation of TNFα, IL-6, IL-8, and several components of the IL-1 pathway, but notably no type I IFN response (Tables 1 to 3).

Pattern recognition of rAAV by NPC (Tables 1 and 2) resembles the activation profile induced by rAAV1, rAAV2 and rAAV8 in the human macrophage cell line THP-1,18 and agrees with results obtained by microarray analysis of rAAV2-treated mouse livers arguing for the existence of an additional recognition mechanism for AAV that substantially differs from the TLR9-mediated sensing of vector genomes by pDCs.

We clearly identified the viral capsid structure as the component that induces cell autonomous immune responses in human KCs and LSECs because empty capsids and rAAVs induced a comparable response (Fig. 4 and Table 3). Recognition of this newly identified PAMP is mediated by the membrane-located PRR TLR2. As evidence for TLR2 function in rAAV sensing we demonstrated up-regulation of TLR2 gene expression (Table 2) upon challenge with rAAV, and immune activation of HEK293/hTLR2-CD14 (Fig. 5A,B), but not of HEK293 (data not shown) or HEK293/hTLR9 by AAV capsids (Fig. 3). Human liver NPCs that recognized rAAV were able to sense pathogens through TLR2, whereas human LSECs showed no and KCs showed only minimal response to TLR9 agonists (Fig. 4). We furthermore showed that anti-TLR2 antibodies inhibit capsid sensing in human KCs and LSECs (Fig. 5C,D), and that upon activation of TLR2 expression, HUVECs become responsive to AAV capsids (Fig. 5E,F).

TLR2 was originally identified as PRR for lipoproteins and peptidoglycans.19 Recently, however, its function in sensing viral pathogens has become increasingly evident.12, 13 Among all TLRs, TLR2 recognizes the structurally broadest range of PAMPs. Because these molecules are diverse, TLR2 will interact with different agonists at different affinities. This explains why AAV capsids elicited a significantly lower, albeit clearly detectable level of innate immune activation compared to the strong TLR2 agonist HKLM (Table 3 and Fig. 5). Supporting this idea doubling of the GOI from 1 × 105 (used throughout this study) to 2 × 105 resulted in a 10-fold increase in IL-8 gene expression in HEK293/hTLR2-CD14 (data not shown).

Among liver NPCs, we focused on KCs and LSECs as the most prominent cell types. In agreement with earlier studies in mice,16 primary human KCs, similar to murine KCs, exhibit innate immune responses to rAAV (Fig. 4). LSECs, although capable of pathogen recognition and of antigen presentation to CD4+ and CD8+ T cells,20 have not previously been investigated for their potential involvement in sensing rAAV. In terms of cytokine induction, LSECs had a lower level of up-regulation than KCs, but clearly responded to AAV capsids. Notably, besides LSECs, macrovascular endothelial cells from human umbilical veins were also capable of sensing AAV capsids via TLR2, revealing that endothelial cells in general have to be considered as sentinel cells for rAAV. Hence, at least three human cell types (DCs, monocytes/macrophages [KCs], and endothelial cells [LSECs and TLR2+-HUVECs]) are involved in sensing rAAV infection through at least two different mechanisms, namely TLR9-mediated recognition of vector genomes and TLR2-mediated detection of structural components of the viral capsid.

Sensing pathogens via different PRRs each activating distinct signaling pathways and innate immune responses increases the likelihood that invading pathogens can be efficiently destroyed by the host. However, this poses a challenge for the clinical use of viral vectors that are sensed by the host as pathogens based on PAMPs shared with the parental virus. After identification of the recognition pathways, strategies that reduce or avoid (innate) immune activation can now be developed. Activation of NFκB following TLR2 signaling could, for example, be prevented by NFκB inhibitors. The modification of viral vector tropism can further help to reduce the required vector dose and exposure of associated PAMPs by limiting off-target transduction.

In summary, in this first study on innate immune responses triggered by AAV in primary human cells, we demonstrated that TLR2 is an innate immune sensor of AAV capsids on KCs, LSECs, and activated HUVECs. This newly identified mechanism complements the current understanding of host immune responses against rAAV and supports efforts to increase the efficacy and safety of rAAV-mediated gene therapy.


We thank Jude Samulski (University of North Carolina), Jim Wilson (University of Pennsylvania), and Sibille Quadt-Humme (University of Cologne) for pXX6-80, pXR8, and pscLuci, respectively, Jürgen Kleinschmidt (DKFZ, Germany) for the antibody B1, and Thomas Plum for his assistance in performing the experiment in HUVECs. We thank Martin Krönke (University of Cologne) for continuous support and Oliver Coutelle (University of Cologne) for critical reading of this manuscript. This work was supported by the German Research Foundation (SFB670 to H.B. and SFB576 to U.P.) and the Center for Molecular Medicine Cologne (ZMMK) to H.B.