Research ArticleVIRAL TOXIN

Dengue virus NS1 protein activates cells via Toll-like receptor 4 and disrupts endothelial cell monolayer integrity

See allHide authors and affiliations

Science Translational Medicine  09 Sep 2015:
Vol. 7, Issue 304, pp. 304ra142
DOI: 10.1126/scitranslmed.aaa3863

A leak in the dike

Everyone knows how mosquitos can wreck an end-of-summer picnic. But in some climates, these pesky intruders persist and carry a variety of detrimental diseases—some with no preventative vaccines or targeted therapies. One such passenger is dengue virus (DENV), which infects up to 400 million people each year and comes in several serotypes (1 to 4) and disease presentations—from mild infection to severe disease and sometimes death. But to treat or prevent dengue requires that we have a more complete picture of the disease pathology. Now, Modhiran et al. and Beatty et al. describe the results of in vitro and in vivo experiments that point to circulating dengue virus non-structural protein 1 (NS1) and the innate immune Toll-like receptor 4 (TLR4) as a focus for basic scientists as well as vaccine and drug developers.

DENV infection protects a patient from future reinfection with the same DENV serotype as well as producing temporary immune protection from severe dengue disease caused by a different DENV serotype. But unlike diamonds, this immune protection doesn’t last forever, and when the protected period passes, the patient becomes at increased risk of enhanced infection and progression to severe disease if he or she is infected with a second DENV serotype. This severe form of dengue infection is believed to result from immunopathogenic processes that induce cytokine storm and cause vascular leakage that leads to shock. Until now, no dengue viral proteins have been linked to vascular endothelium permeability (that is, vascular leakage).

Beatty et al. show that inoculation of mice with DENV NS1 protein alone induces both vascular leak and secretion of inflammatory cytokines and that administration of NS1 with a sublethal dose of DENV2 leads to lethal vascular leak syndrome. In human endothelial cell monolayers in culture, NS1 from any of the four DENV serotypes triggered endothelial barrier permeability. NS1’s pathogenic effects were blocked by NS1-immune polyclonal mouse serum or monoclonal antibodies to NS1 (in vivo and in vitro), and immunization of mice with NS1 protected against lethal DENV2 challenge.

In an independent study, Mondrian et al. explore the underlying mechanism of NS1’s effects. They show that highly purified NS1 acts as a pathogen-associated molecular pattern (PAMP) that activates mouse macrophages and human peripheral blood mononuclear cells (PBMCs) in culture via TLR4, resulting in release of inflammatory cytokines—an effect that was blocked by either a TLR4 antagonist or an anti-TLR4 antibody. Then, in an in vitro model of vascular leak, the authors found that NS1 fractured the integrity of endothelial cell monolayers through a TLR4-dependent pathway, a finding that was supported by the observation that a TLR4 antagonist quelled capillary leak in a mouse model of dengue virus infection.

Together, these new findings highlight NS1 as an instigator of dengue-associated vascular leak and thus pinpoint a potential target for dengue drugs and component for dengue vaccines.


Complications arising from dengue virus infection include potentially fatal vascular leak, and severe disease has been linked with excessive immune cell activation. An understanding of the triggers of this activation is critical for the development of appropriately targeted disease control strategies. We show here that the secreted form of the dengue virus nonstructural protein 1 (NS1) is a pathogen-associated molecular pattern (PAMP). Highly purified NS1 devoid of bacterial endotoxin activity directly activated mouse macrophages and human peripheral blood mononuclear cells (PBMCs) via Toll-like receptor 4 (TLR4), leading to the induction and release of proinflammatory cytokines and chemokines. In an in vitro model of vascular leak, treatment with NS1 alone resulted in the disruption of endothelial cell monolayer integrity. Both NS1-mediated activation of PBMCs and NS1-induced vascular leak in vitro were inhibited by a TLR4 antagonist and by anti-TLR4 antibody treatment. The importance of TLR4 activation in vivo was confirmed by the reduction in capillary leak by a TLR4 antagonist in a mouse model of dengue virus infection. These results pinpoint NS1 as a viral toxin counterpart of the bacterial endotoxin lipopolysaccharide (LPS). Similar to the role of LPS in septic shock, NS1 might contribute to vascular leak in dengue patients, which highlights TLR4 antagonists as a possible therapeutic option.


Dengue virus (DENV) infection is an increasing problem in tropical and subtropical areas and is endemic in more than 100 countries (1). As climate change extends the range of the mosquito vector and travel increases, more populations will become at risk for contracting dengue infection. An estimated 390 million dengue infections occur globally each year, with 96 million of these being symptomatic and giving rise to many thousands of deaths (1). Despite this global health burden, no vaccine or therapeutic intervention has yet been licensed. Dengue infection causes a spectrum of diseases ranging from undifferentiated fever and classical dengue fever to severe and potentially fatal dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS). Severe disease is characterized by the rapid onset of capillary leak accompanied by thrombocytopenia and altered hemostasis. Although the mechanistic basis of dengue-mediated capillary leak remains the subject of considerable conjecture, the host immune response appears to be an important determinant of disease outcome. In comparison with less severe cases, DHF/DSS patients have higher levels of circulating proinflammatory cytokines and chemokines, referred to as a cytokine storm (2). These vasoactive mediators are thought to play a major role in the collapse of vascular integrity (3, 4), and both activated T cells and monocytes have been proposed as critical in their production (5, 6). Although DHF and DSS can accompany primary infection in infants (7), they more frequently occur upon secondary infection in older children and adults with a different DENV serotype. The anamnestic induction of high levels of cross-reactive but subneutralizing antibodies during the acute stage of secondary infection has been hypothesized to contribute to antibody-dependent enhancement (ADE) of infection in Fc receptor–bearing cells, which results in both higher virus levels and elevated cytokine production (8, 9). ADE in primary infection in infants in the absence of a secondary immune response is thought to result from the presence of maternally acquired antibodies that have declined to subneutralizing levels (7). It has also been proposed that, in secondary infections, activation of cross-reactive, and consequently, low-affinity memory T cell responses to the primary infecting virus results in delayed viral clearance (5) and elevation of inflammatory cytokines that contribute to the pathology of severe disease (2). However, severe disease in primary infections of infants (7) demonstrates that memory T cell responses are not essential for severe pathology. Viral strain virulence, host genetics, and nutritional status are all likely additional risk factors that contribute to the progression to severe disease (10).

The DENV nonstructural protein NS1 is a multifunctional, 48- to 55-kD glycoprotein that is initially synthesized as a soluble monomer and becomes membrane-associated after dimerization in the lumen of the endoplasmic reticulum (11). The recent crystal structure determination of NS1 has revealed exposed hydrophobic domains on the dimeric form that are likely to mediate this membrane association (12). Intracellular NS1 participates in early viral RNA replication and is found in virus-induced vesicular compartments that house the viral replication complexes (13). NS1 is also transported to the cell surface, where it either remains associated with the cell membrane or is secreted as a soluble, lipid-associated hexameric species. A small subset of cell surface-expressed NS1 has been shown to be membrane-associated via a glycosylphosphatidylinositol anchor, which can mediate signal transduction upon specific antibody engagement, thereby facilitating enhanced production of viral progeny (14). Secreted NS1 (sNS1) can be detected in the bloodstream of infected patients from the first day of symptoms and circulates at levels in the ng/ml to μg/ml range during the acute phase of infection (15, 16). This circulating viral biomarker has now become a target for diagnosis of early acute infection (17, 18), with levels correlating with viremia and disease severity in secondary dengue infection (16, 19). NS1 is thought by some to play a number of roles in the pathology of infection, including both stimulation and inhibition of complement pathways as well as damage to platelets and endothelial cells by cross-reactive anti-NS1 antibodies (11, 19, 20).

Here, we examined whether sNS1 directly stimulates innate immune responses. Recognition of pathogen-associated molecular patterns (PAMPs) by immune-stimulating receptors such as Toll-like receptors (TLRs) on innate immune cells can mediate induction of cytokines and promote antigen-presenting cell function. For example, components of the Gram-negative bacterial outer membrane, lipopolysaccharide (LPS) and lipoproteins, are recognized by TLR4 and TLR2, respectively (21). Recognition of viral nucleic acids by endosomal TLR3, TLR7, and TLR9 as well as other cytosolic receptors provides a versatile means of response to a wide range of viruses. Several viral proteins are also reported to be direct TLR agonists; the measles virus hemagglutinin protein, hepatitis C virus core and NS3 proteins, and NSP4 of rotavirus all signal through TLR2 (2224), whereas the respiratory syncytial virus (RSV) fusion (F) protein (25), Ebola virus glycoprotein (26), and mouse mammary tumor virus envelope protein (27) were reported to induce inflammatory cytokines via TLR4 activation. Specific TLR activation by viral proteins seems counterintuitive. It would be reasonable to expect that RNA viruses would rapidly evolve to evade TLR detection, much as viruses have evolved numerous strategies for avoiding the inhibitory effects of interferon (IFN). However, in some cases, TLR activation might induce a cellular state that favors viral replication or transmission. Innate immune responses are not always protective but may contribute to pathology, particularly when uncontrolled.

Here, we show that DENV NS1 elicits inflammatory cytokine production and endothelial cell monolayer leak via TLR4 activation, and TLR4 antagonism decreases vascular leak in a mouse model of infection. By analogy to LPS-induced septic shock, NS1-induced cytokines and direct effects on endothelial cell function might play an important role in the vascular leak and shock that is the hallmark of DHF/DSS.


NS1 activates human and mouse immune cells

We hypothesized that circulating NS1 directly activates innate immune cells. Recombinant, hexameric sNS1 was isolated from media harvests of S2 cells stably expressing DENV-2 NS1 by immunoaffinity chromatography, and its purity was assessed by SDS–polyacrylamide gel electrophoresis and size exclusion chromatography (fig. S1). Purified recombinant NS1 was shown to be LPS-free by the limulus amebocyte lysate (LAL) assay; readings for NS1 at 40 μg/ml indicated <0.1 endotoxin units (EU)/ml or <10 pg of LPS/ml equivalent. Throughout this study, LPS was used as a positive control for TLR4 activation. Purified sNS1 and LPS both strongly induced the secretion of interleukin-6 (IL-6) from peripheral blood mononuclear cells (PBMCs) (Fig. 1A), suggesting that NS1 behaves as a PAMP. NS1 immunodepletion prevented IL-6 induction, whereas a control antibody neither depleted NS1 nor abrogated activity (Fig. 1A). A nuclear factor κB (NF-κB)–responsive mouse macrophage reporter cell line was also activated by NS1 (fig. S2).

Fig. 1. Dengue NS1 activates human and mouse innate immune responses, and is not inhibited by LPS-binding antibiotic polymyxin B.

(A) NS1 induces IL-6 secretion by human PBMCs. Cells were treated for 24 hours with LPS (100 ng/ml), NS1 (40 μg/ml), or an analogous NS1 sample immunodepleted with anti-NS1 or anti-E antibody. IL-6 was measured by enzyme-linked immunosorbent assay (ELISA). Data are means ± SD of three donors. The inset immunoblot is of the supernatant fraction after immunodepletion using anti-NS1 antibody or isotype-matched control antibody. (B) Polymyxin B does not block NS1-induced IL-6 in human PBMCs. NS1 and LPS were preincubated for 30 min with or without polymyxin B, then added to PBMCs, and incubated for 24 hours. The final concentration of polymyxin B was 25 μg/ml, with varying concentrations of LPS (0.1, 1, and 10 ng/ml) and NS1 (5, 10, and 20 μg/ml). The experiment was run in parallel with (A) and separated for clarity (mean ± SD of three donors). (C) Polymyxin B does not block NS1-induced loss of surface CSF1R on BMMs. NS1 and LPS were preincubated for 30 min with or without polymyxin B and then added to BMMs that had been starved of CSF1 overnight. The final concentration of polymyxin B was 25 μg/ml (19.2 μM), with LPS (10-fold dilutions from 100 ng/ml, or 10 nM assuming average Mr = 10 kD) and NS1 (2-fold dilutions from 10 μg/ml, or 33.3 nM). After 1 hour of treatment, CSF1R levels were determined by flow cytometry, with the % of cells with high levels of receptor plotted. Data are means ± SD from three independent experiments.

In defining new PAMPs, the elimination of bacterial contaminants as possible mediators is critical. The immunodepletion results above demonstrate that either NS1 or a factor bound to NS1 activates both mouse and human cells. Because hexameric NS1 is a lipid-binding protein (28), it is conceivable that the insect (S2) cell–derived NS1 could have bound trace amounts of bacterial lipids such as LPS during the purification process. Although the LAL assay did not detect any endotoxin, to further allay concerns about contaminating LPS, we examined the induction of IL-6 in human PBMCs using NS1 pretreated with the LPS-binding antibiotic polymyxin B. Polymyxin B inhibited the response to concentrations of LPS ≤10 ng/ml, whereas the NS1-mediated response was unaffected (Fig. 1B). Polymyxin B was also able to inhibit low concentrations of LPS spiked into NS1 samples (fig. S3), demonstrating that NS1 was unlikely to sequester contaminating endotoxin, making it unavailable for polymyxin B inhibition. To facilitate characterization of the mechanism of activation, we moved to mouse cells and measured down-modulation of the receptor for the macrophage growth factor, colony-stimulating factor 1 (CSF1), in bone marrow–derived macrophages (BMMs). Cell surface CSF1 receptor (CSF1R) is increased on CSF1-starved BMMs and rapidly lost in response to TLR agonists including LPS, making a convenient and sensitive assay for TLR agonists (29). Both LPS and NS1 down-modulated surface CSF1R (fig. S4). Only the LPS response was sensitive to polymyxin B (Fig. 1C), and thus, LPS contamination is not a likely explanation for the NS1 response of mouse or human cells.

NS1 activity is not due to a contaminant

The possibility remained that the active component of the NS1 preparation was a NS1-binding contaminant distinct from LPS. To eliminate any concerns about introduction of bacterial contaminants during the purification process, we assessed the activity of NS1 expressed by mammalian cells without purification. Medium, transfection reagent [polyethylenimine (PEI)], plasmids, and transient transfection harvests were all tested by the LAL assay and shown to be LPS-free (<0.1 EU/ml or <10 pg of LPS/ml equivalent). As expected, the S2-derived NS1 caused loss of CSF1R, and this was prevented by NS1 immunodepletion (Fig. 2A). Medium derived from NS1 expression vector–transfected Chinese hamster ovary (CHO) cells also reproducibly reduced surface CSF1R levels, whereas immunodepletion of NS1 prevented this activity (Fig. 2B). Media from untransfected or empty vector–transfected CHO cells had no effect on CSF1R levels. CHO cell–derived NS1 also induced IL-6 from human PBMCs, and this activity was specifically removed by immunodepletion of NS1 (Fig. 2C). Together, the data show that the NS1 lipoprotein, rather than any contaminant in the NS1 preparations, has direct immunostimulatory activity.

Fig. 2. NS1 from several sources has immunostimulatory activity.

(A) NS1 immunodepletion prevents the down-modulation of surface CSF1R by S2 cell–expressed NS1. BMMs starved of CSF1 overnight were treated with either phosphate-buffered saline (PBS), CSF1 (104 U/ml), LPS (100 ng/ml), NS1 (10 μg/ml), or NS1 immunodepleted with anti-NS1 antibody or control anti-E antibody. Data are means ± SD from three independent experiments. (B) Minimally processed CHO cell–expressed NS1 has similar activity to S2 cell–derived NS1. CHO cells were transfected with either empty expression vector or a plasmid encoding NS1 under the control of the cytomegalovirus promoter. Culture medium, concentrated with 100-kD cutoff filters from untransfected cells (control SN), empty vector (pcDNA SN), or NS1 expression vector–transfected cells (NS1 SN), was applied to CSF1-starved BMMs. The final concentration of NS1 (5 μg/ml) (as quantified by capture ELISA) represents about fourfold higher levels than produced in the CHO cell medium. Cells were also exposed to NS1 SN after immunodepletion with anti-NS1 antibody or control anti-E antibody and to CSF1. Data are means ± range from two independent experiments. Immunoblot analysis for NS1 confirms immunodepletion. (C) CHO cell–derived NS1 induces IL-6 in human PBMCs. Culture medium from transient expression concentrated as above was applied to human PBMCs. Treatments are as per (B), with the inclusion of medium from CHO cells exposed to PEI transfection reagent (PEI SN) as an extra control. Released IL-6 was measured by ELISA, and data are means ± SD from four donors, except for immunodepleted samples, which represent three donors.

All flaviviruses encode a homologous NS1 protein, with remarkable structural similarity observed between DENV NS1 and West Nile virus (WNV) NS1 (12). Therefore, it was of interest to determine whether other flavivirus NS1 species were also stimulatory. We transiently expressed WNV (NY99) and yellow fever virus (YFV) NS1 in CHO cells and found that both induced detectable secretion of IL-6 from cultured PBMCs (fig. S5). This suggests a possible common strategy among the flaviviruses of inducing innate immune activation. However, the accompanying paper (30) found that WNV NS1 did not disrupt endothelial cell monolayer integrity, consistent with lack of peripheral vascular leak in WNV disease. Ultimately, the contribution of different species of NS1 to TLR4-mediated responses in flavivirus infections will depend on their levels of release, distribution in the body, and individual potencies, which remain to be fully determined. Of interest, a short peptide motif in the N-terminal region of NS1 that is divergent between WNV and DENV targets most of the WNV NS1 to the cell surface, whereas DENV NS1 is predominantly secreted (31).

NS1 is recognized via TLR4 and induces proinflammatory cytokines

Proinflammatory cytokines such as tumor necrosis factor–α (TNF-α), IFN-γ, IL-6, IL-1β, and MIF (macrophage migration inhibitory factor) and chemokines such as IL-8, IP-10, and MCP-1 are induced during the acute phase of dengue infection, with levels correlating with disease severity (3, 7, 32, 33). NS1 induced a dose-dependent increase in the levels of mRNA for TNF-α, IL-6, IFN-β, IL-1β, and IL-12 in mouse BMMs, although this had not reached a maximal response at NS1 of 40 μg/ml (Fig. 3A and fig. S6A). NS1 induced these genes in a similar time course to LPS (Fig. 3B and fig. S6B). Whereas the response of mouse BMMs to relevant concentrations of NS1 was much lower than the response to LPS, both agents produced strong induction of cytokine mRNAs in human PBMCs, with NS1 being a particularly potent inducer of IL-6 and IL-1β gene expression (Fig. 3C and fig. S6C). This suggests a measure of species specificity in the recognition of the NS1 protein. IL-12 was induced at a later time than other genes, and IP-10 and MCP-1 showed modest induction by both agents (fig. S6C).

Fig. 3. Induction of TNF-α and IL-6 mRNAs in mouse macrophages and human PBMCs in response to S2-derived NS1 protein.

(A) Dose-dependent production of cytokine mRNAs from treated murine BMMs. BMMs were incubated with purified NS1 (1.25, 2.5, 5, 10, 20, and 40 μg/ml) or LPS (0.1, 1, 10, and 100 ng/ml) and harvested after 3 hours. Expression of individual mRNAs was measured by real-time polymerase chain reaction (PCR) and expressed relative to Hprt mRNA. (B) Time course of BMM response to LPS and NS1. BMMs were incubated with either no additions (control), NS1 (40 μg/ml), or LPS (1 ng/ml) for the indicated times. Cytokine mRNAs were measured as described in (A). A control sample is included for each time point, but generally cannot be seen on this scale. (C) Time course of PBMC response to LPS and NS1. PBMCs were treated with no additions (control), NS1 (10 μg/ml), or LPS (10 ng/ml) for the indicated times. Cytokine mRNAs were measured by real-time PCR and expressed relative to Hprt mRNA. Data are means ± range from two independent experiments (A and B) or means ± SD from four donors (C).

TLRs are the most likely candidates for cell surface receptors that recognize extracellular NS1 and mediate activation of NF-κB and cytokine induction. TLR2 and TLR4 have been observed to respond to several viral proteins (2227). We therefore examined down-modulation of CSF1R on BMMs from mice lacking either individual TLRs or TLR signaling adapters MyD88 and TRIF. The responses to both LPS and NS1 were lost in Tlr4−/− and MyD88−/−/Trif−/− macrophages but were intact in Tlr2−/− cells. Pam3CSK4 was used as a control stimulus for TLR2 (Fig. 4, A and B). Tlr4−/− BMM also failed to respond to NS1 with increased TNF-α, IL-1β, or other cytokine mRNAs (Fig. 4, C and D, and fig. S7A), but these responses were intact in Tlr2−/− cells (Fig. 4, E and F, and fig. S7B). To confirm that human TLR4 recognizes NS1, IL-8 mRNA was measured in HEK293 cells ectopically expressing TLR4/MD-2 or TLR2, stimulated with NS1, LPS, or Pam3CSK4. A response to NS1 and LPS was seen in cells expressing TLR4/MD-2 (Fig. 4G) but not TLR2 (Fig. 4H), consistent with the results from BMMs of gene-targeted mice.

Fig. 4. NS1 activates cells via TLR4.

(A) Down-modulation of BMM cell surface CSF1R by NS1 requires TLR4. CSF1R on wild-type (WT) C57BL/6 BMMs, Tlr4−/−, and MyD88−/−/Trif−/− BMMs was measured after 1 hour of treatment with CSF1 (104 U/ml), LPS (100 ng/ml), and NS1 (10 μg/ml). (B) Down-modulation of BMM cell surface CSF1R by NS1 does not require TLR2. BALB/c WT BMMs and Tlr2−/− BMMs were treated as in (A) or with Pam3CSK4 (100 ng/ml) as a TLR2 stimulus. (C and D) NS1 induction of TNF-α and IL-1β mRNAs requires TLR4. WT (C57BL/6) and Tlr4−/− BMMs were incubated with NS1 (10 μg/ml), LPS (1 ng/ml), or Pam3CSK4 (100 ng/ml) for 3 hours, and mRNA expression was determined using real-time PCR relative to Hprt mRNA. (E and F) NS1 induction of cytokine mRNAs does not require TLR2. WT BALB/c and Tlr2−/− BMMs were treated as in (C). (G and H) Human embryonic kidney (HEK) 293 cells expressing human TLR4 and MD-2 (G), but not TLR2 (H), respond to 3 hours of treatment with NS1 (10 μg/ml), with induction of IL-8 mRNA, measured by real-time PCR relative to Hprt mRNA, and normalized to the control sample. LPS (10 ng/ml) and Pam3CSK4 (100 ng/ml) provide control stimuli for TLR4 and TLR2, respectively. Data are means ± SD of three independent experiments (A) or means ± range from two independent experiments (B to H).

To determine whether cell surface TLR4 and NS1 colocalize, adherent PBMCs were exposed to NS1, then washed and fixed, and stained with antibodies for confocal microscopy. Although colocalization was not complete, a large number of foci displayed overlapping signal for TLR4 and NS1 on the surface of cells (Fig. 5A; controls in fig. S8). Z stack analysis showed clear association in isolated regions of the membrane (Fig. 5A and movie S1). Among PBMCs, monocytes are the most abundant cells with readily detectable TLR4 expression, with lymphocytes expressing very little TLR4 mRNA (34). The adherent PBMCs were a mixture of cell types, and notably, the smaller cells without observable surface TLR4 were also devoid of detectable NS1 binding (Fig. 5B).

Fig. 5. Colocalization of NS1 and TLR4 on adherent PBMCs.

PBMCs were allowed to adhere to coverslips for 2 hours and were then treated with NS1 (10 μg/ml) for 45 min before fixation. Cells were stained for cell surface TLR4 (red) and NS1 (green) without permeabilization. (A) Orthographic and planar projections were obtained by confocal microscopy. In the right hand panel, regions of colocalization appear as yellow (selected regions arrowed). Single-color controls showed no bleed between fluorescence channels. There was no background staining with secondary antibodies alone, and the secondary antibodies were demonstrated to be specific for the species of primary antibody (fig. S5). (B) NS1 binding correlates with TLR4 expression in the mixed PBMC population. Cells were stained as per (A).

To demonstrate the involvement of TLR4 in human PBMC responses to NS1, we took advantage of a naturally occurring TLR4 antagonist. The pattern of lipid A acylation of Rhodobacter sphaeroides LPS (LPS-RS) renders it nonstimulatory and inhibitory to active LPS by competitive binding to the TLR4/MD-2 complex (35, 36). Pretreatment of PBMCs with LPS-RS completely inhibited induction of IL-6 and IL-8 by LPS and NS1 but not Pam3CSK4 (Fig. 6A and fig. S9). A TLR4 blocking antibody also reduced the NS1 response. More complete inhibition of NS1 than LPS activity was seen, which may reflect a greater ability of the antibody to sterically block the larger NS1 macromolecule (Fig. 6B and fig. S9).

Fig. 6. TLR4 antagonism blocks NS1 activity in vitro and reduces vascular leak in a murine DENV infection model.

(A) NS1-induced IL-6 production by PBMCs was inhibited by LPS-RS. Cells were preincubated with LPS-RS (10 μg/ml) for 30 min and subsequently stimulated with LPS (100 ng/ml), NS1 (10 μg/ml), or Pam3CSK4 (500 ng/ml) for 24 hours. IL-6 was quantified by ELISA. (B) Anti-TLR4 antibody reduces the response to NS1. PBMCs were preincubated with anti-TLR4 (2 and 10 μg/ml) for 1 hour and then stimulated with NS1, LPS, or Pam3CSK4 for 24 hours as above. IL-6 was quantified by ELISA. Data are means ± range for two donors (A and B). (C) NS1 induces permeability of HMEC-1 monolayers, and its action is inhibited by LPS-RS and anti-TLR4 antibody. Confluent cells in Transwells were pretreated with or without LPS-RS (100 μg/ml) or anti-TLR4 antibody (TLR4 Ab) (20 μg/ml) for 1 hour, and then treated with LPS (100 ng/ml) or NS1 (10 μg/ml). Monolayer integrity was assessed by measurement of transendothelial electrical resistance (TEER) at the indicated times. Data are means ± SD over four replicate wells. (D and E) AG129 mice were injected intraperitoneally with 50 μg of LPS-RS or saline, 1 hour before infection with DENV-2 strain D220 [104 plaque-forming units (PFU), intraperitoneally]. Mice were treated daily with the same doses of LPS-RS or saline. Evans blue was administered intravenously on day 4 after infection and allowed to circulate for 2 hours before mice were anesthetized and perfused, and organs were harvested. Dye content from intestines (D) and liver (E) was assessed colorimetrically after extraction into dimethylformamide to assess relative vascular integrity. Means ± SD are indicated for n = 5 to 7. Statistical significance was assessed by one-way analysis of variance (ANOVA) (Sidak multiple comparisons test; *P < 0.05, **P < 0.01, ****P < 0.0001 for comparisons indicated on the graphs).

A TLR4 antagonist protects against vascular leak in a mouse model of infection

Vascular leakage leading to shock is one of the hallmarks of severe dengue disease. NS1 could contribute to vascular leak through direct activation of endothelial cells, which can also express TLR4 (37). We analyzed the effect of exposure to NS1 or LPS on the integrity of human microvasculature endothelial cell (HMEC-1) monolayers grown in Transwell cultures by measurement of transendothelial electrical resistance (TEER). Compared to the mock-treated monolayers, leak was observed by the first hour for NS1-treated cells and 5 hours for LPS-treated cells with both peaking between 9 and 11 hours (Fig. 6C). Pretreatment with LPS-RS completely blocked leak, and an anti-TLR4 antibody inhibited leak induced by both NS1 and LPS (Fig. 6C). This result implies that in addition to mediating cytokine induction from monocytes/macrophages, NS1 recognition by TLR4 on endothelial cells directly contributes to vascular leak. To extend these in vitro findings into an in vivo model, we examined the effects of LPS-RS treatment on capillary leak induced in DENV-2 (strain D220)–infected (38) AG129 mice. The virus dose was nonlethal, but infected mice examined 4 days after infection exhibited significant vascular leak in several tissues as measured by tissue content of intravenously delivered Evans blue dye (Fig. 6, D and E). Daily intraperitoneal injection of LPS-RS (50 μg) greatly reduced capillary leak induced by DENV infection (Fig. 6, D and E), indicating that inhibition of NS1-mediated TLR4 activation may offer a new disease intervention strategy.


Here, we showed that the hexameric, secreted form of DENV NS1 directly activated mouse macrophages and human PBMCs via TLR4 with the consequent release of proinflammatory cytokines. Thus, NS1 might contribute to triggering of the cytokine storm proposed as being responsible for the vascular leak and shock in severe dengue disease (3). In addition, NS1 directly impaired endothelial cell monolayer integrity in an in vitro model of leak, and a TLR4 antagonist inhibited vascular leak in vivo. These findings provide a mechanistic framework for the in vivo activity of sNS1 reported in the accompanying paper (30), and together, these studies provide a new paradigm for NS1 in dengue disease pathogenesis.

TLR4 is expressed on several cell types including monocytes, macrophages, and endothelial cells. TLR4 and co-receptor MD-2 together mediate responses to bacterial LPS, with the crystal structure of the complex revealing that the predominant interaction involves the acyl chains of the central lipid A core of LPS being buried in a hydrophobic pocket in MD-2 (39). Antagonists of human TLR4 signaling, such as LPS-RS, and its synthetic counterpart eritoran bind within the MD-2 hydrophobic pocket but lack a critical acyl chain required for TLR4–MD-2 complex dimerization and signaling (40). We found that LPS-RS was also a potent antagonist of NS1-mediated cytokine induction and vascular leak, suggesting that NS1 also interacts with MD-2.

In addition to LPS, TLR4 is reported to mediate responses to many other structurally and chemically diverse ligands of bacterial, viral, and host origin. It has been proposed that rather than being true TLR4 ligands, many of these actually either bind and present LPS to cells or sensitize cells to respond to LPS (41). We used several approaches to confirm that NS1 lipoprotein alone mediates the observed TLR4 activation, although the relative contributions of NS1 protein and its associated lipid moieties remain to be established. There is precedent for direct TLR4 recognition of a viral protein in the reported response to the RSV F protein (25). RSV F protein responses, like those of NS1, are inhibited by LPS-RS, with the hydrophobic N-terminal region of the F protein shown to be responsible for interaction with the TLR4–MD-2 complex (42). The crystal structure of NS1 dimer and hexamer assemblies has revealed exposed hydrophobic domains (12), which are candidate regions for TLR4 interaction.

Increased vascular permeability is the primary manifestation of severe dengue, and so, the mechanism of endothelial cell responses to dengue infection is of great interest. In dengue patients, levels of TNF-α, IL-1β, IL-6, IFN-γ, IL-8, and MCP-1 all correlate with disease severity (3, 32, 33). NS1 was shown in this study to directly induce a similar array of proinflammatory cytokine and chemokine mRNAs, many of which are implicated in the perturbation of endothelial cell barrier function. For example, the combination of IL-1β, TNF-α, and IFN-γ affects the barrier function of endothelial cells (43), and IL-8 and MCP-1 have been implicated in alteration of endothelial tight junctions in dengue infection (32, 44). In the context of dengue infection, TNF-α has been suggested to be a mediator of dengue-associated hemorrhage via the induction of endothelial cell apoptosis (45). In addition to its role in the induction of cytokines, NS1 might also have a direct effect on endothelial cells. The loss of endothelial monolayer integrity in vitro that we observed in this study after only 1 to 2 hours of exposure to NS1 (Fig. 6C) is most likely a direct effect of NS1-induced signaling pathways on cellular tight junctions. LPS directly induces endothelial leak via TLR4, with the involvement of calcium influx (46) and tyrosine kinase pathways (47). A similar role for NS1 is supported by the inhibition of NS1-induced leak by LPS-RS in vitro (Fig. 6C) and is suggested by our finding in a mouse model that vascular leak associated with DENV infection can be substantially blocked by pretreatment with LPS-RS (Fig. 6, D and E).

The role of both virus and host cell factors in the kinetics and severity of dengue disease progression, particularly in secondary infections, is complex (35, 9), and sNS1 is likely to be only one of a number of risk factors that determine disease severity. NS1 itself might have multiple effects on pathophysiology (11) via mechanisms that include direct engagement with complement components (19) and the induction of anti-NS1 antibodies, which cross-react with platelets (20). However, the decrease in vascular leak in a mouse model of infection seen herein upon treatment with a TLR4 antagonist suggests that NS1 activation of TLR4 has a causal role in pathology. If so, vascular permeability and severe pathology might be expected to coincide with peak levels of circulating NS1. However, the observation in the accompanying paper (30), that maximal morbidity was not observed until 3 days after administration of NS1 to mice, suggests a more complex scenario. In human disease, severe complications occur at a time when both viremia (16) and detectable circulating NS1 are declining, although these follow different kinetics, with NS1 levels dropping more than a day later than virus (16). This apparent clearance of NS1 from circulation might reflect sequestration of NS1 within tissues, not necessarily the cessation of NS1 expression or its related effects. Indeed, in vitro work has shown that NS1 binds avidly to endothelial cells within tissue sections (48). Hence, the concentrations of NS1 at its site of action in vivo might not be reflected in detectable circulating levels. In summary, the dynamics of NS1 are currently difficult to assess, and the effects of NS1 might not be immediate.

A role for TLR4 in dengue pathology offers a route to therapeutic intervention with the possible repurposing of existing sepsis drug candidates. Although the LPS antagonist eritoran was not successful in phase 3 clinical trials for sepsis (49), treatment was only initiated after onset of organ dysfunction. It remains to be seen whether there is a therapeutic window in which the blocking of NS1 activity early in acute DENV infection with compounds such as eritoran can ameliorate subsequent progression to pathology. Given that the Ebola glycoprotein has also been reported to activate TLR4 (26) and that there is a role for TLR4 in the pathology induced by infection with another flavivirus, Japanese encephalitis virus (JEV) (50), there may be a broader scope for TLR4 antagonists in the treatment of viral disease.

NS1 is also considered to be a target for vaccine development (51), and immunization with NS1 protects mice from lethal DENV challenge (52, 53). Our study and that of the accompanying paper (30) suggest that such vaccines could have additional efficacy by clearing NS1 early during acute infection and thereby reduce NS1-induced pathology. Overall, the parallels between NS1 and LPS are compelling; LPS in bacterial sepsis contributes to a cytokine storm, vascular permeability, and septic shock, and NS1 might generate a similar viral toxic shock in DENV-infected patients.

Note added in proof: While this paper was in press, another publication appeared asserting that dengue NS1 activates mouse macrophages and human PBMC via TLR2 and TLR6 (54). That work predominantly used commercially available E. coli–derived DENV-1 NS1 with no assessment of the level of E. coli PAMP contamination. Their conclusion that TLR2 is involved in NS1 recognition is not consistent with knockout data we show in Fig. 4, and our demonstration of the inhibition of responses by LPS-RS strongly implicates TLR4, and not TLR2 or TLR6.


Additional detailed materials and methods can be found in the Supplementary Materials.

Experimental design

The aim of this study was to investigate a direct role for the DENV secreted glycoprotein NS1 in disease pathogenesis. Using in vitro models, we assessed the ability of recombinant NS1 to activate human PBMCs and induce permeability in a model vascular cell layer. Cellular activation was assessed through the measurement of both mRNA and expressed protein. Using cells from knockout mice, we attributed this activity to a TLR4-dependent pathway. We sought to exclude the possibility of LPS contamination by using polymyxin B LPS sequestration, sensitive LPS detection, and NS1 derived from an alternate source and in the absence of purification. We investigated the ability of TLR4 antagonists to block NS1-induced cellular activation and leak using in vitro models and then translated these findings to an in vivo dengue disease model. AG129 mice (type I and II IFN receptor–deficient) were chosen because they exhibit typical signs of severe dengue disease. Vascular leak was assessed using Evans blue quantification from tissue samples. Group sizes were selected on the basis of our experience with these systems. Mice were age- and sex-matched between groups. Investigators were not blinded when conducting or evaluating the experiments, and no randomization was necessary. No data were excluded from this study.


Ultrapure LPS from Escherichia coli 0111:B4 strain, LPS-RS, and polymyxin B were obtained from InvivoGen. Pam3CSK4 provided by K. Smith (Department of Pathology, University of Washington, Seattle, WA) was originally purchased from Roche. The anti-TLR4 antibody (HTA125) was purchased from Abcam.

Cell culture

BMMs were obtained by differentiation of mouse bone marrow progenitors in the presence of CSF1 (104 U/ml) for 7 days (29). The use of mice was approved by the University of Queensland Animal Ethics Committee. Blood was obtained from healthy volunteers under approval from the University of Queensland Medical Research Ethics Committee, and PBMCs were isolated by Ficoll gradient. HEK293 cells, stably transfected with hTLR2 or hTLR4/MD-2, were obtained from D. Golenbock (University of Massachusetts). Human microvasculature endothelial cells (HMEC-1) were obtained from P. Hogg (Lowy Cancer Research Centre, University of New South Wales).

Generation and purification of DENV-2 NS1 and transient expression of WNV and YFV NS1

Recombinant DENV-2 NS1 was expressed by stably transfected Drosophila S2 cells. The protein was affinity-purified from culture medium using a column coupled with the 2A5.1 anti-NS1 monoclonal antibody. NS1 proteins from DENV, WNV (strain NY99), and YFV (stain 17D) were also transiently expressed in CHO cells. Transfection complex was washed away after 4 hours. The medium was harvested at day 2 after transfection and concentrated using a 100-kD cutoff spin column (Millipore). Recombinant NS1 preparations were tested for LPS contamination by the LAL assay using the LAL Chromogenic Endotoxin Quantitation Kit (Pierce) and shown to be endotoxin-free (background readings representing <0.1 EU/ml or <10 pg of endotoxin/ml equivalent were found for purified DENV-2 NS1 at 40 μg/ml and for the working stocks of transiently expressed and concentrated NS1 preparations). Immunodepletions of NS1 were performed using mouse monoclonal anti-NS1 antibody (1H7 or 2A5) and control anti-E antibody (3H5), purified from ascites, and bound to protein G Dynabeads (Life Technologies).

CSF1R down-modulation assay

BMMs were cultured overnight in complete RPMI 1640 medium without CSF1. Cells were subsequently treated with NS1, LPS, Pam3CSK4, or immunodepleted NS1 for 1 hour in a microcentrifuge tube, then stained for CSF1R, and analyzed by flow cytometry as described (29).

Analysis of mRNA by quantitative real-time reverse transcription PCR

Mouse BMMs (with CSF1) or human PBMCs were treated with NS1 or TLR ligands for indicated times and HEK293 cells and HEK293-TLR cell lines for 3 hours before harvest for RNA extraction. Complementary DNA was prepared using random hexamer primers, and quantitative PCR was carried out using gene-specific primers (tables S1 and S2) and SYBR Green PCR MasterMix (Life Technologies). Hprt gene expression was used as the reference for both mouse and human genes.

Detection of cytokine productions by ELISA

Human PBMCs were incubated for 24 hours with NS1 or LPS. In some experiments, LPS-RS was preincubated with cells for 30 min. The media of treated cells were harvested and centrifuged at 1000g for 5 min. The levels of IL-6 and IL-8 production were quantified by ELISA (R&D Systems).

Indirect immunofluorescence and confocal microscopy

PBMCs adherent to coverslips were incubated with NS1 (10 μg/ml) for 45 min at 37°C. Cells were washed, fixed with 4% paraformaldehyde in PBS, blocked with PBS containing 5% nonfat dry milk and 0.05% Tween 20, and stained with rabbit anti-NS1 polyclonal antibody and mouse anti-TLR4 monoclonal antibody (HTA125), followed by goat anti-rabbit Alexa Fluor 488–conjugated and goat anti-mouse Alexa Fluor 555–conjugated secondary antibodies (Life Technologies). The cells were examined using Zeiss 510 and 710 META microscopy at ×100 magnification. The images were analyzed using ImageJ version 10.

TEER measurement of HMEC-1 monolayer integrity

Transwell polycarbonate membranes (Corning) were coated with 100 μl of fibronectin (Sigma) at 10 μg/ml for 30 min at 37°C and 5% CO2, the liquid was subsequently removed, and the membranes were dried for 45 min at 37°C and 5% CO2. HMEC-1 cells were seeded on Transwells, and optimal monolayer formation was established 4 days after seeding. Before agonist stimulation, the cells were preincubated with LPS-RS (100 μg/ml) or anti-TLR4 antibody (20 μg/ml) for 1 hour before the addition of LPS or NS1. Endothelial permeability was evaluated over 48 hours by measuring TEER using the EVOM2 epithelial voltohmmeter (World Precision Instruments).

Efficacy of LPS-RS in inhibiting DENV-induced vascular leak in vivo

Mouse experiments were approved by the University of Queensland Animal Ethics Committee. AG129 mice, which lack receptors for IFN-α/β as well as IFN-γ (55), were bred at the University of Queensland. Mouse-adapted DENV-2 strain D220 (38) was provided by E. Harris (School of Public Health at the University of California, Berkeley) and stocks were prepared in C6/36 cells. Mice were administered with LPS-RS (50 μg per mouse, intraperitoneally) 1 hour before infection with DENV (104 PFU per mouse, intraperitoneally) or mock infection. LPS-RS or PBS was administered daily (50 μg per mouse, intraperitoneally) for the duration of the experiment. Mice were observed every 12 hours for morbidity scoring. Vascular leakage was quantified with Evans blue dye as previously described (56). Briefly, 200 μl of Evans blue dye (5 mg/ml) was injected intravenously 4 days after infection and allowed to circulate for 2 hours before mice were anesthetized and extensively perfused with PBS. Tissues were collected in preweighed tubes containing 1 ml of N,N-dimethylformamide and incubated at 37°C for 24 hours. Extracted Evans blue dye was assessed by measuring OD610 in samples, in comparison with a standard curve. Data were expressed as ng Evans blue dye/mg tissue weight.

Statistical analysis

Statistical analyses were performed in Prism 6.0 (GraphPad Software Inc.). Data are shown as means ± SD or means ± range. Significance testing of vascular leak levels within mice was performed using ordinary one-way ANOVA (Sidak’s multiple comparison test).


Materials and Methods

Fig. S1. Purification and characterization of DENV-2 NS1 from S2 cells.

Fig. S2. Dengue NS1 activates a mouse macrophage NF-κB reporter cell line.

Fig. S3. Polymyxin B can efficiently inhibit LPS in the presence of NS1.

Fig. S4. NS1 protein causes the loss of surface CSF1R on murine BMMs.

Fig. S5. Recombinant NS1 from other flaviviruses has immunostimulatory activity.

Fig. S6. Induction of cytokine mRNAs in C57BL/6 BMMs and human PBMCs in response to S2-derived NS1 protein.

Fig. S7. NS1 activates cells via TLR4.

Fig. S8. Controls for colocalization of NS1 and TLR4 on adherent PBMCs.

Fig. S9. Inhibition of PBMC production of IL-8 in response to NS1 by anti-TLR4 antibody and TLR4 antagonist.

Table S1. List of mouse primers for RT-PCR.

Table S2. List of human primers for RT-PCR.

Movie S1. TLR4 and NS1 colocalization on PBMCs; video of immunofluorescence confocal microscopy Z stack.

Primary data tables (Excel file)

References (57, 58)


  1. Acknowledgments: We thank M. Sweet, A. Blumenthal, K. MacDonald, and M. Wykes for reagents, discussion, and cells from knockout mice; E. Harris for providing the mouse-adapted DENV-2 strain D220; and S. Akira for his generosity in providing knockout mice. Funding: This work was funded by National Health and Medical Research Council (NHMRC) grant 1067226 to P.R.Y. and K.J.S. K.J.S. was also supported by Australian Research Council and NHMRC Fellowships FT0991576 and 1059729. Author contributions: P.R.Y., K.J.S., N.M., D.W., and D.P.S. designed the research. N.M., D.W., A.K.P., D.P.S., L.L., and D.A.M. performed the research. N.M. and A.K.P. performed in vitro activation studies. N.M. and D.W. performed in vitro monolayer leak experiments. N.M., D.W., and D.A.M. performed the mouse experiments. L.L., A.K.P., D.A.M., and P.R.Y. developed recombinant NS1 protein expression and purification procedures. D.W. performed the negative stain electron microscopy and N.M. performed the confocal microscopy. D.P.S. and D.A.H. provided intellectual input in project direction. N.M., D.W., D.A.H., K.J.S., and P.R.Y. wrote the paper. Competing interests: A provisional patent, PCT/AU2014/050403, has been filed in Australia, titled “Methods and compositions for treating or preventing flavivirus infections.”
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article