Both Virus and Tumor Necrosis Factor Alpha Are Critical for Endothelium Damage in a Mouse Model of Dengue Virus-Induced Hemorrhage

Abstract

Hemorrhage is a common clinical manifestation in dengue patients. However, the pathogenic mechanism of dengue virus (DV)-induced hemorrhage awaits clarification. We established a mouse model of DV hemorrhage using immunocompetent C57BL/6 mice by injecting DV serotype 2 strain 16681 intradermally. While inoculation of 3 × 10⁹ PFU of DV induced systemic hemorrhage in all of the mice by day 3 of infection, one out of three of those injected with 4 × 10⁷ to 8 × 10⁷ PFU developed hemorrhage in the subcutaneous tissues. The mice that were inoculated with 4 × 10⁷ to 8 × 10⁷ PFU but that did not develop hemorrhage were used as a basis for comparison to explore the pathogenic mechanism of dengue hemorrhage. The results showed that mice with severe thrombocytopenia manifested signs of vascular leakage and hemorrhage. We observed that high viral titer, macrophage infiltration, and tumor necrosis factor alpha (TNF-α) production in the local tissues are three important events that lead to hemorrhage. Immunofluorescence staining revealed that DV targeted both endothelial cells and macrophages. In addition, the production of high levels of TNF-α in tissues correlated with endothelial cell apoptosis and hemorrhage. By comparing TNF-α^−/− to IgH^−/−, C5^−/−, and wild-type mice, we found that TNF-α was important for the development of hemorrhage. In vitro studies showed that mouse primary microvascular endothelial cells were susceptible to DV but that TNF-α enhanced DV-induced apoptosis. Our mouse model illustrated that intradermal inoculation of high titers of DV predisposes endothelial cells to be susceptible to TNF-α-induced cell death, which leads to endothelium damage and hemorrhage development. This finding highlights the contribution of the innate immune response to dengue hemorrhage.

Dengue virus (DV) is a mosquito-borne viral disease that affects humans in both the tropics and subtropics. The mild form of dengue is dengue fever (DF). Approximately one-third of DF patients may have hemorrhage manifestations ranging from mild skin hemorrhage, gingival or nasal bleeding, and gastrointestinal bleeding to severe hemorrhage (9). Severe hemorrhage with low platelet counts, plasma leakage, and pleural or other effusions are the characteristics of dengue hemorrhagic fever (DHF) (9). Therefore, hemorrhage, mild or severe, is clinically significant for DV infection.

Virulent virus strains and preexisting nonneutralizing heterologous anti-DV antibodies are reported to be risk factors for DHF (10, 22). Immune responses involving the activation of T lymphocytes, the production of cytokines or chemokines, and complement activation are also considered to be important for the pathogenesis of DHF (8, 19, 23). Higher levels of tumor necrosis factor alpha (TNF-α) also correlate with severe disease (2, 8). The production of TNF-α instead of gamma interferon (IFN-γ) by activated T cells was suggested to contribute to the pathogenesis of dengue disease (20, 21, 23). Patients with DHF have subnormal levels of C3, C4, and C5 in the serum and increased complement metabolism (26). From those studies, it appears that both innate as well as adaptive immunity are implicated in severe disease. However, the key factor(s) that induces dengue hemorrhage has not been revealed.

Over the past decade, the study of direct and indirect interactions between endothelial cells (ECs) and DV has become a central focus in the understanding of the pathogenesis of DV hemorrhage. DV antigens (Ags) are detected in vascular endothelium in biopsy tissues from DHF/dengue shock syndrome (DSS) patients, indicating that DV targets ECs in vivo (15). It remains to be clarified, however, whether DV, the host immune response, or the complex interplay between the virus and the host causes EC damage in the infected host. An animal model will be best for the investigation of the complex relationship between DV and the host that results in hemorrhage.

Small-animal models that have been used to study dengue disease include hu-PBL-SCID mice, SCID-K562 mice, SCID-HepG2 mice, STAT-1^−/− mice, AG129 mice with IFN-α/β and -γ receptor deficiencies, and immunocompetent BALB/c, A/J, and C57BL/6 mice (4, 14, 16, 18, 28, 32). By intravenous, intraperitoneal, or intracerebral inoculation of DV, the animals had liver pathology, thrombocytopenia, or neurological symptoms. Hemorrhage manifestation was not reported as being a prominent feature in any of these animal models.

In this study, we successfully developed a mouse model of dengue hemorrhage. Immunocompetent mice receiving intradermal inoculations of DV developed hemorrhage. High viral titers, macrophage infiltration, and the production of TNF-α correlated with EC apoptosis and hemorrhage. Moreover, results of gene knockout experiments clearly demonstrated that TNF-α was key to the development of hemorrhage. In vitro studies employing primary mouse microvascular ECs demonstrated that TNF-α enhanced DV-induced apoptosis. Together, these results demonstrate that DV predisposes EC to TNF-α-induced apoptosis, which leads to hemorrhage.

MATERIALS AND METHODS

Mice.

C57BL/6, IgH^−/− (Igh-J^fm/Cgn), TNF-α^−/− (Tnf^tm/Gk1), A/HeJ (Hc⁰), and A/J mice were originally obtained from the Jackson Laboratory (Bar Harbor, ME) and bred at the Laboratory Animal Center of the National Taiwan University College of Medicine. All mice were housed in sterile cages fitted with filtered cage tops and fed with sterilized food and water. Four- to five-week-old mice were used for all experiments.

Virus.

DV serotype 2 (DV-2) strain 16681 was used throughout this study. DV-2 16681 was originally isolated from a Thai patient who suffered from DHF (24). The virus was propagated in the C6/36 insect cell line cultured in Dulbecco's modified Eagle's medium (Invitrogen, Grand Island, NY) containing 2% heat-inactivated fetal calf serum (Biological Industries, Kibbutz Beit Haemek, Israel) at 28°C. Culture supernatants were collected 5 days after virus infection. The virus titer was then determined by plaque assay on a BHK cell line. To prepare high titers of DV, virus supernatant was concentrated on a Centriplus device (10-kDa cutoff) (Amicon; Millipore) by centrifugation before the plaque assay. The virus titer could reach 10¹⁰ PFU/ml after concentration. The inoculum was prepared by diluting virus stocks in phosphate-buffered saline (PBS) immediately before inoculation. To inactivate the virus, virus stocks were treated with UV at 50 mJ/cm² for 3 min in the SPECTROLINKER XL-1000 CROSSLINKER (Spectronics Corporation, Westbury, NY). Loss of infectivity was confirmed by plaque assay. Japanese encephalitis virus (JEV) (strain B1PS3) was also propagated in the C6/36 cell line by following the same protocol as that used for DV.

Intradermal infection.

Mice at 4 to 5 weeks of age were inoculated with DV (in 0.4 ml) intradermally at four sites on the upper back. Mice given PBS, mock C6/36 culture supernatant, an equivalent titer of UV-inactivated DV (UV-DV), or viable JEV through the same route were used as controls. Western blot analysis for E and PreM proteins showed that at the same PFU (4 × 10⁷), JEV expressed slightly higher levels of Ags than DV (data not shown). To observe hemorrhage development, mice were killed at day 3 after inoculation, and the subcutaneous tissues in the back, the abdomen, and the axillary areas and thorax were exposed.

Platelet counts.

Blood was collected from mice at different time points after virus inoculation, placed into an EDTA-coated Microtainer tube, and mixed immediately (Vacutainer system; Becton Dickinson, Franklin Lakes, NJ). A blood smear was made and read to ensure the absence of star-shaped fibrin. Platelets were enumerated using an Abbott Cell-Dyn 3700 apparatus (Abbott Laboratories, Abbott Park, IL) through service provided by the National Mouse Mutagenesis Program Core Facility (Academia Sinica, Taiwan, Republic of China).

RT real-time PCR and real-time RT-PCR to quantify DV in mice.

To perform reverse transcriptase (RT) real-time PCR, tissues were collected from mice intradermally inoculated with PBS or DV at different time points after inoculation. Tissues from skin, spleen, liver, and brain and subcutaneous tissues were collected, with the latter two taken from the site at the diagonal crossing of the four intradermal injection sites. Cellular RNA was extracted from all the tissues with TRIzol reagent (Invitrogen), and 3 μg of RNA was reverse transcribed to cDNA by using Superscript II Reverse Transcriptase (Invitrogen). cDNA was amplified in a 25-μl reaction mixture containing primers, SYBR green I (Invitrogen), Flu A (Bio-Rad), and Hot-Start Taq DNA polymerase (Invitrogen) by using the iCycler Real-Time PCR detection system (Bio-Rad, Hercules, CA). The condition of amplification was a reaction at 95°C for 3 min, which was followed by 40 cycles of 95°C for 30 s, 56°C for 30 s, 72°C for 10 s, and 83°C for 10 s and then 1 cycle of 95°C for 1 min, 57°C for 1 min, and 55°C for 10 s. The reaction was stopped at 10°C. The viral capsid gene was normalized against β-actin. The primers for viral capsid gene amplification were C22A and C63B, as described previously (4).

A real-time RT-PCR assay was performed to quantify the DV copy number in sera (30). Briefly, RNA was extracted from mouse serum. Serial 10-fold dilutions of 1 × 10⁸ copies of positive-sense viral capsid gene were used as standards. Primers (d2C16A and d2C46B), the TaqMan TARMA probe, and a One-Step RT-PCR Master Mix reagent kit (PE Biosystems, Foster City, CA) were used. The amplification conditions were described previously (30). The ABI Prism 7700 sequence detector (Applied Biosystems, Foster City, CA) was used to analyze the emitted fluorescence during amplification. The sensitivity of the assay is 2.5 copies of RNA (corresponding to 357 copies per ml of serum) per reaction.

RT real-time PCR to quantify TNF-α mRNA.

Total cellular RNA was extracted from subcutaneous tissues and reverse transcribed to cDNA. The TNF-α gene was amplified using the iCycler Real-Time PCR detection system as described above. The level of TNF-α mRNA expression was normalized against β-actin. The sequences of TNF-α primers were 5′-ATCCGCGACCTCGCCCTG-3′ and 5′-ACCGCCTGGAGTTCTGGAA-3′, and the sequences of β-actin primers were 5′-AGGTGTGCACCTTTTATTGGTCTCAA-3′ and 5′-TGTATGAAGGTTTGGTCTCCCT-3′.

TNF-α ELISA.

Subcutaneous tissues (110 to 114 mg) from DV-infected or control mice were harvested and homogenized in 1 ml PBS buffer by POLYTRON (KINEMATICA AG, Lucerne, Switzerland). Supernatants were collected from homogenized tissues after centrifugation and stored at −70°C. Culture supernatants collected from viable DV-infected, UV-DV-treated peritoneal macrophages, or macrophage cultures without any treatment were also stored at −70°C. The concentration of TNF-α in tissue homogenates and culture supernatants was quantified by TNF-α enzyme-linked immunosorbent assay (ELISA) (eBioscience, La Jolla, CA). Known concentrations of recombinant murine TNF-α were used as standards.

Hematoxylin and eosin staining.

Subcutaneous tissues were fixed in 4% neutral formalin solution, embedded in paraffin, and sectioned at a thickness of 3 μm. After deparaffinization and rehydration, the sections were stained with hematoxylin and eosin. The sections were then dehydrated before mounting.

Immunofluorescence staining.

Subcutaneous tissues removed from control mice as well as mice inoculated intradermally with viable DV or UV-DV were embedded in optimal cutting temperature medium in a dry-ice bath. Frozen tissues were then cryosectioned (CRYOTOME SME; Shandon, Pittsburgh, PA) at a thickness of 10 to 15 μm. The sections were fixed in acetone for 5 min and then blocked by treatment with PBS containing 5% goat serum. Phycoerythrin (PE)-conjugated rat anti-mouse Mac-1 antibody (Ab) (clone M1/70), PE-conjugated rat anti-mouse TNF-α Ab (clone MP6-XT22), PE-conjugated rat anti-mouse CD31 Ab (clone PECAM-1), or fluorescein isothiocyanate (FITC)-conjugated rat anti-mouse F4/80 Ab (clone BM8) was added. It was confirmed that Mac-1⁺ cells in subcutaneous tissues were F4/80⁺, CD14⁺ (clone Sa2-8), and CD207⁻ (clone eBio RMUL.2) macrophages. To double stain macrophage and TNF-α, FITC-conjugated anti-F4/80 and PE-conjugated anti-TNF-α Abs were added simultaneously on the same cryosection. Hoechst 33258 dye (Sigma-Aldrich) was used to stain nuclei. All antibodies for immunofluorescence staining were obtained from eBioscience.

In situ detection of DNA fragmentation.

Tissue cryosections or cell monolayers were fixed in 4% paraformaldehyde at room temperature for 10 min and then treated with 3% H₂O₂ in methanol. Cryosections or monolayers were permeabilized with 0.1% Triton X-100 in 0.1% sodium citrate on ice for 2 min. To detect apoptotic cells in tissues, an FITC-conjugated terminal deoxynucleotidyltransferase-mediated dUTP-biotin nick end-labeling (TUNEL) mixture (In Situ Cell Death Detection kit; Roche Applied Science, Indianapolis, IN) was added to tissue sections. PE-conjugated anti-Mac-1 or PE-conjugated anti-CD31 Ab was then used to further determine the types of cells undergoing apoptosis. To detect cell death in primary EC monolayers, Converter-POD (peroxidase-conjugated anti-fluorescein Ab) was added, and 3,3′-diaminobenzidine was used as a substrate for color development. Hoechst 33258 dye was used to stain nuclei.

Staining for DV Ag.

Cryosections were fixed with 4% paraformaldehyde at room temperature for 10 min. To detect viral Ag on cryosections, polyclonal rabbit anti-DV antiserum was added, and the sections were left at 4°C overnight. PE-conjugated anti-rabbit Ab (Invitrogen) was then added to the sections and left at 4°C for 1 h. To detect DV target cells, PE-conjugated anti-Mac-1 or PE-conjugated anti-CD31 Ab and rabbit anti-DV Ab were added to the same section simultaneously. After incubation at 4°C overnight, FITC-conjugated anti-rabbit Ab was added. To determine whether DV Ag-expressing cells were apoptotic, tissue sections were double stained with FITC-conjugated TUNEL mixture and rabbit anti-DV Ab by following the same protocol.

Mouse microvascular EC isolation.

The method of isolation of mouse microvascular ECs was adopted from a previously described method (33). Briefly, brains from mice at 8 to 12 weeks of age were collected. Cerebral cortices were harvested, and the solid tissues were homogenized in RPMI medium containing 2% inactivated fetal calf serum by use of a glass tissue grinder. The homogenates were separated by 15% dextran (average molecular mass of 68,800 Da; Sigma-Aldrich) and treated with a collagenase-dispase mixture (Boehringer Mannheim, Indianapolis, IN) at 0.05% (wt/vol) at 37°C for 6 h with occasional agitation. Cells were washed with buffer and resuspended in RPMI medium containing 30 μg/ml ENDO-GRO (Sigma-Aldrich). ECs were grown in flasks precoated with 1% gelatin. Cells at passages 3 to 4 containing 90 to 95% CD31⁺ cells were used in all experiments.

DV infection of cells.

Thioglycolate-elicited peritoneal macrophages or ECs were seeded onto 12-mm round glass coverslips (Assistant, Sondheim, Germany). The cells were cultured in RPMI medium containing 1% inactivated fetal calf serum at 37°C for 2 h. Viable DV or UV-DV was then added to the monolayers at a multiplicity of infection (MOI) of 5 or 10 and incubated at 37°C for 2 h with gentle shaking every 15 min. The monolayer was washed with Hanks’ balanced salt solution and cultured in the respective fresh growth medium. Culture supernatants from infected macrophage monolayers were harvested after infection and stored at −70°C before the assay. Undiluted supernatants were used to assess the effect of DV-infected macrophage culture supernatants on ECs. The percentage of apoptotic cells was determined by using the In Situ Cell Detection system as described above. To neutralize the effect of TNF-α, supernatants from DV-infected macrophages were preincubated with neutralizing anti-mouse TNF-α Ab (clone MP6-XT3; eBioscience) before addition to the EC culture. To assay the effect of TNF-α, ECs with or without DV infection were cultured in medium containing human recombinant TNF-α (Biosource, Camarillo, CA) at 30, 300, or 3,000 pg/ml.

Statistical analysis.

The Statistical Package for the Social Sciences (SPSS Inc.) program was employed for statistical analysis. The difference between the means of experimental groups was analyzed using analysis of variance and the Tukey post hoc test. Statistical significance was determined at P values of <0.01 and <0.05.

RESULTS

DV induces hemorrhage in C57BL/6 mice.

To mimic natural infection in humans, immunocompetent C57BL/6 mice were injected intradermally with DV. Hemorrhage developed at different sites including the subcutaneous tissue, abdomen, the intestine (Fig. 1A), the skin, and lymph nodes (data not shown) at day 3 after virus inoculation. The incidence, severity, and site of hemorrhage development correlated with the size of virus inoculum. While 100% and 73% of the mice receiving 3 × 10⁹ and 1 × 10⁹ PFU, respectively, developed systemic hemorrhage, 33% of those given 8 × 10⁷ PFU exhibited hemorrhage at the subcutaneous tissue (Table 1 and Fig. 1A and B). Notably, subcutaneous hemorrhage developed at approximately the diagonal crossing of the lines connecting the four injection sites, and the degree of hemorrhage ranged from mild to severe (Fig. 1A and B). Interestingly, although mice exhibited severe hemorrhage after receiving a high viral inoculum (Fig. 1A), they appeared to be healthy, and no sign of paralysis was observed. To confirm that hemorrhage is DV specific, mice inoculated with PBS, UV-DV (3 × 10⁹ and 4 × 10⁷ PFU), mock C6/36 supernatant, and viable JEV (4 × 10⁷ PFU) were examined for hemorrhage development. Data showed that only the mice that were inoculated with viable DV but not any other inoculum developed hemorrhage, indicating that hemorrhage development is viable-DV specific. Moreover, microscopic examination of the subcutaneous hemorrhage tissues revealed red blood cell extravasation (Fig. 1C). These results demonstrate that intradermal inoculation of DV induces hemorrhage in mice and that hemorrhage is accompanied by vascular leakage.

FIG. 1.

Intradermal inoculation of DV induces hemorrhage in immunocompetent mice. (A) C57BL/6 mice were injected intradermally with 3 × 10⁹ PFU of viable DV at four different sites in the back as indicated by arrows. Subcutaneous tissue, abdomen, and intestine were exposed at day 3 after inoculation. (B) Mice were inoculated intradermally at four sites, as indicated by arrows, with PBS, 4 × 10⁷ to 8 × 10⁷ PFU of viable DV (Live DV), or UV-DV. Hemorrhage was observed in subcutaneous tissues at day 3 of infection. Hemorrhage sites in the tissues are circled. (C) Sections of subcutaneous tissues from mice inoculated with PBS as well as viable DV as in B were stained with hematoxylin and eosin. The magnification was ×100 (left) and ×400 (right).

TABLE 1.

Correlation between virus inoculum and hemorrhage^a

Virus titer (PFU)	No. of mice with hemorrhage/ no. of mice inoculated (% of mice with hemorrhage)
3 × 109	16/16 (100)b
1 × 109	11/15 (73)b
8 × 107	11/33 (33)c

^aC57BL/6 mice were inoculated intradermally with the indicated titers of viable DV at four sites on the upper back. Hemorrhage was observed at day 3 after inoculation. The data from three to four experiments are pooled.

^bMice developed systemic hemorrhage as shown in Fig. 1A.

^cMice developed hemorrhage in the subcutaneous tissue as shown in Fig. 1B.

Mice with severe thrombocytopenia manifest hemorrhage.

Next, we collected peripheral blood from mice receiving 4 × 10⁷ PFU of DV and tested whether thrombocytopenia is associated with hemorrhage. Since about two-thirds of mice receiving 4 × 10⁷ PFU did not develop hemorrhage, they were used as nonhemorrhage (nH) controls. At day 3 after inoculation, the platelet counts in mice injected with DV were significantly lower (P < 0.01) in both hemorrhagic mice (H mice) (732.2 ± 109.5 × 10⁻³ platelets/μl) and nH mice (772.7 ± 49.9 × 10⁻³ plates/μl) than in control mice (923.3 ± 73.35 × 10⁻³ platelets/μl). At day 7, the counts in nH mice (772.5 ± 36.5 × 10⁻³ platelets/μl) remained at the same level as that on day 3, but that in H mice (578.0 ± 71.9 × 10⁻³ platelets/μl) was further reduced (P < 0.05). These data demonstrate that DV infection induces thrombocytopenia, and mice with severe thrombocytopenia manifest hemorrhage.

DV distribution in mice after intradermal inoculation.

To study the relationship between virus titer and hemorrhage, we compared virus capsid gene expression in tissues from H and nH mice. At day 3 after virus inoculation, the viral capsid gene was detectable in spleen, liver, brain, skin, and serum in both H and nH mice but the level of expression was significantly higher in H than in nH mice (Fig. 2A and B). Although viral capsid RNA was no longer detectable in the subcutaneous tissues at day 3 after inoculation, it was expressed at day 1 (Fig. 2A), showing that DV was present in subcutaneous tissues soon after intradermal inoculation but was cleared thereafter. Furthermore, while viral capsid gene copy number increased from day 1 to day 3 in the sera of H and nH mice, it was significantly higher in H than in nH mice (Fig. 2B). Immunofluorescence staining also showed the presence of viral antigen in the skin (increased from day 1 to day 3) (Fig. 2C) and subcutaneous tissues (at day 1) (Fig. 2D) after inoculation. Together, these data indicate a strong correlation between high viral titers and hemorrhage development.

FIG. 2.

DV capsid gene and antigen expression in tissues. Mice were inoculated with 4 × 10⁷ PFU viable DV (DV). (A) RNA was extracted from tissues collected at days 1 and 3 after virus inoculation. The viral capsid gene was amplified by RT real-time PCR. The relative levels of capsid RNA were determined as described in Materials and Methods. ** indicates a P value of <0.05 when the mean value in H mice was compared to that in nH mice. Each time point is the mean value of tissues from four different mice. ND, not detected. (B) Mouse serum was collected at days 1, 2, and 3 after DV inoculation. The viral capsid gene was amplified by real-time RT-PCR. The capsid gene copy number per microliter of serum was determined against the standards. Hemorrhage manifestation in each mouse was determined at day 3 after inoculation. ** indicates a P value of <0.05 by comparing the mean of viral copy numbers in H mice (solid bars) to that in nH mice (open bars). Skin (C) and subcutaneous tissues (D) from mice inoculated with viable virus (Live-DV) or UV-DV were collected at days 1 and 3 after inoculation. DV Ag was detected by rabbit polyclonal anti-DV and PE-conjugated goat anti-rabbit Ig Abs. The cryosections of tissues collected from H mice were stained with naïve rabbit serum as a control to examine the specificity of rabbit anti-DV serum. Hoechst 33258 stain revealed the nuclei. Data presented are representative of three repeated experiments. The magnification was ×200 (C) and ×400 (D).

Macrophages infiltrate and produce TNF-α in the hemorrhage tissue.

To investigate the contribution of immune cells to hemorrhage development, subcutaneous tissues from H and nH mice were examined for the presence of NK cells, neutrophils, macrophages, dendritic cells, and T cells. The results revealed that no appreciable numbers of NK cells (NK1.1⁺), T cells (CD3⁺), dendritic cells (CD11c⁺), Langerhans cells (CD207⁺), or neutrophils (Gr.1⁺) were present (data not shown). Only macrophages (Mac-1⁺/F4/80⁺/CD14⁺) were detected in the subcutaneous tissues of both H and nH mice, but the level was twofold higher (P < 0.05) in H mice (21.5 ± 4.2%) than in nH mice (10.6% ± 2.0%) (Fig. 3).

FIG. 3.

Macrophage infiltration in hemorrhage tissue. The cryosections of subcutaneous tissues collected from mice injected with PBS, 4 × 10⁷ PFU of viable DV (Live-DV), or an equivalent titer of UV-DV at days 1 and 3 after inoculation were stained with PE-conjugated rat anti-mouse Mac-1 Ab. Cryosections of tissues from H mice were stained with PE-conjugated anti-mouse isotype IgG2b Ab as a staining control. H mice or nH mice were determined at day 3 after inoculation. The number ± standard deviation on the top of each picture indicates the mean percentage of Mac-1⁺ cells in each group. The percentage of Mac-1⁺ cells was determined by dividing the count of Mac-1⁺ cells by the total number of Hoechst-positive cells in the same field. The data were pooled from counting cells in four sections obtained from four different mice. Three to five fields in each section and a total of 750 to 800 cells were counted. **, P < 0.05, by comparing the number of infiltrating macrophages in tissue from H mice to that in tissue from nH mice at day 3 after infection.

Since it was reported previously that high TNF-α expression correlates with hemorrhage (12), in situ TNF-α expression in subcutaneous tissues was then quantified. RT real-time PCR results showed a significant increase in TNF-α mRNA in H mice over the PBS control at day 3 after DV infection (Table 2). It was further increased in H but not in nH mice at day 3 (Table 2). Quantitation of in situ TNF-α protein in tissue homogenates showed that while TNF-α levels increased from day 1 on, the level in H mice (155.2 ± 65.7 pg/ml) was about eight times higher than that in nH mice (19.2 ± 1.1 pg/ml) at day 3 after infection (Table 2). These results indicate a strong association between macrophage infiltration, TNF-α production, and hemorrhage.

TABLE 2.

High TNF-α expression correlates with hemorrhage

TNF-α	Inoculumc	Mean TNF-α level± SD at day after inoculationd:
		1	2	3
mRNA (TNF-α/β-actin)a	Live DV	0.43 ± 0.09	0.63 ± 0.05	1.10 ± 0.22**† (H), 0.47 ± 0.14 (nH)
	UV-DV	ND	ND	0.46 ± 0.09e
	PBS	ND	ND	0.50 ± 0.01e
Protein (pg/ml)b	Live DV	28.4 ± 4.5*	65.5 ± 15.2*	155.2 ± 65.7*† (H) 19.2 ± 1.1*(nH)
	UV-DV	ND	ND	15.6 ± 1.0e
	PBS	ND	ND	14.1 ± 0.2e

^aThe relative levels of TNF-α mRNA expression were determined by RT real-time PCR.

^bThe concentration of TNF-α protein in the supernatants of tissue homogenates was quantified by ELISA.

^cMice were inoculated intradermally with 4 × 10⁷ PFU of viable DV (Live DV), UV-inactivated DV (UV-DV), or PBS on the upper back.

^dSubcutaneous tissues were collected from mice at days 1, 2, and 3 after live DV inoculation and at day 3 after PBS or UV-DV inoculation. Data at each time point were pooled from six mice. The averages of values are expressed as means ± standard deviations. * and ** indicate P values of <0.01 and <0.05, respectively, compared to PBS-inoculated mice. †, P < 0.01 compared to nH mice at day 3. ND, not done.

^eNo hemorrhage was observed in UV-DV- or PBS-inoculated mice.

Immunofluorescence staining of subcutaneous hemorrhage tissues showed that TNF-α colocalized with macrophages (Fig. 4A), indicating that macrophages are a likely source of TNF-α. In vitro experiments were performed to determine whether DV-infected macrophages produce TNF-α. Figure 4B shows that DV induced macrophage production of TNF-α at as early as 6 h (122.1 ± 32.8 pg/ml) after infection. The level of TNF-α reached the peak at 12 h (329.0 ± 86.2 pg/ml) and declined thereafter. These results together revealed that macrophage infiltration and production of TNF-α in the tissues after DV infection are two important events that lead to hemorrhage.

FIG. 4.

DV induces macrophage TNF-α production. (A) Mice were inoculated with 4 × 10⁷ PFU of viable DV. The cryosections of subcutaneous tissues from hemorrhagic mice were stained with FITC-conjugated anti-F4/80 and PE-conjugated anti-TNF-α Abs. Isotype Ab staining with the controls FITC-conjugated anti-mouse IgG2a and PE-conjugated anti-mouse IgG1 were negative. Data presented are representative of three repeated experiments. The magnification was ×400. (B) Supernatants from DV-infected macrophages (MOI of 10) were collected at 0, 6, 12, 21, and 48 h postinfection (h.p.i.). The TNF-α concentrations in the supernatants were determined by ELISA. ** indicates a P value of <0.05 when the mean TNF-α concentration in the supernatants from DV-infected macrophages was compared to that from macrophages cultured in medium without virus. The levels of TNF-α in supernatants from macrophages cultured in medium or medium containing UV-DV were below the level of detection (data not shown). Data presented are pooled from three experiments.

TNF-α is key to the development of dengue hemorrhage in mice.

TNF-α^−/− mice were employed to ascertain the role of TNF-α in hemorrhage development. Since Ab and complement have been implicated in the protection as well as the pathogenesis of hemorrhage in dengue (23) and DV infection-induced DV-specific immunoglobulin M (IgM) and IgG production and C3 deposition were observed in hemorrhage tissues (data not shown), mice with IgH and complement defects were used as controls. Table 3 shows that the percentages of TNF-α^−/− mice (5%) that developed hemorrhage were significantly lower (P < 0.05) than those of wild-type, IgH^−/−, and A/HeJ (C5^−/−) mice. The percentages of hemorrhage development in IgH^−/− (29%) and C5^−/− (36%) mice were not much different from those of their respective wild-type mice (35% and 33%, respectively). These data clearly indicate that TNF-α, but not Ab production or complement deposition, is key to the development of hemorrhage in DV-infected mice.

TABLE 3.

TNF deficiency diminishes DV-induced hemorrhage

Mouse straina	No. of mice with hemorrhage/ no. of mice inoculated (% of mice with hemorrhage)b
C57BL/6	7/20 (35)
IgH−/−	4/14 (29)
TNF-α−/−	1/20 (5)**
A/J	6/18 (33)
A/HeJ	8/22 (36)

^aIgH^−/− mice, TNF-α^−/− mice, and their wild-type C57BL/6 counterparts as well as A/HeJ mice and their wild-type A/J counterparts were injected intradermally with 4 × 10⁷ PFU of viable DV. The data from three experiments are pooled.

^bMice developed hemorrhage in the subcutaneous tissue as shown in Fig. 1B. ** indicates that the P value when TNF-α^−/− mice were compared to wild-type, IgH^−/−, A/J, or A/HeJ mice each was <0.05.

ECs are targets of DV and become apoptotic in the hemorrhage tissue.

To determine whether hemorrhage is accompanied by cell death, subcutaneous tissues were stained with TUNEL reagents. The results show an overwhelming presence of apoptotic cells in tissues from H mice (37.2% ± 18.0%) but not nH mice (1.0% ± 0.4%) at day 3 after infection (P < 0.05). Double staining revealed that apoptotic cells were DV Ag positive (Fig. 5). Further analysis of the types of cells undergoing apoptosis revealed that as many as 94.0% of CD31⁺ ECs (1) and only 23% of macrophages (F4/80⁺/Mac-1⁺) in the hemorrhage tissues were apoptotic, indicating a strong link between EC apoptosis and hemorrhage. In addition, double staining unveiled that both ECs and macrophages are targets for DV at the early phase of infection (Fig. 6).

FIG. 5.

DV Ag-expressing cells are apoptotic. The cryosection of subcutaneous tissues collected from mice injected with 4 × 10⁷ PFU of viable DV (Live-DV) or UV-inactivated DV (UV-DV) at day 2 after inoculation was stained with rabbit anti-DV serum plus PE-conjugated goat anti-rabbit IgG, FITC-conjugated TUNEL mixture, and Hoechst 33258 stain.

FIG. 6.

DV targets ECs and macrophages in the subcutaneous tissue after intradermal inoculation. The cryosections of subcutaneous tissues collected from mice receiving 4 × 10⁷ PFU of viable DV at day 1 after inoculation were stained with (A) rabbit anti-DV antiserum plus FITC-conjugated goat anti-rabbit IgG and PE-conjugated anti-Mac-1 or (B) rabbit anti-DV antiserum plus FITC-conjugated goat anti-rabbit IgG, PE-conjugated anti-CD31 Ab, and Hoechst 33258 stain.

In vitro experiments demonstrate that TNF-α in the DV-infected macrophage culture supernatant induces EC apoptosis.

Primary microvascular ECs were isolated to study their susceptibility to DV- and TNF-α-induced cell death. The results demonstrate that DV alone at MOIs of 5 and 10 induced 11.0% ± 1.1% and 29.9% ± 5.6% EC apoptosis, respectively, at 24 h of infection (Fig. 7A). Recombinant TNF-α at 3,000 pg/ml but not at 30 or 300 pg/ml induced EC death (Fig. 7A). Interestingly, the addition of 300 pg/ml of TNF-α to DV-infected cells increased cell death from 11.0% ± 1.1% to 21.0% ± 3.8% (P < 0.01) at an MOI of 5 and from 29.9% ± 5.6% to almost threefold to as high as 85.4% ± 8.3% (P < 0.05) at an MOI of 10, and the addition of 3,000 pg/ml of TNF-α increased cell death to 92.0% ± 0.3% at an MOI of 10. It is worth noting that the addition of 300 pg/ml of TNF-α to UV-DV-treated cells did not increase the percentage of cell death compared to that of TNF-α-treated uninfected cells, indicating that EC death is not a result of overstimulation by high concentrations of Ag. Moreover, treatment with 12-h supernatants from DV-infected macrophage cultures completely destroyed EC monolayers preinfected with DV (Fig. 7B). Over 95% of DV-infected ECs at an MOI of 10 were apoptotic after treatment with 12-h supernatants (Fig. 7C). Importantly, the effect of macrophage culture supernatants on ECs was completely neutralized by anti-TNF-α Ab pretreatment, regardless of the virus titer with which the ECs were infected (Fig. 7B and C). These results together demonstrate that DV induces macrophage production of TNF-α and that TNF-α enhances DV-induced EC death.

FIG. 7.

Macrophage produces TNF-α to enhance DV-induced EC death. (A) Mouse microvascular ECs were infected with viable DV at an MOI of 5 or 10 or cultured in medium alone. Recombinant TNF-α was added to the EC culture with (black bar) or without (open bar) DV infection. ECs treated with UV-DV and recombinant TNF-α were used as the controls (hatched bar). Cell death was determined by a TUNEL reaction. The percentage of apoptotic cells in each condition was determined by dividing the number of TUNEL-positive cells by the total number of Hoechst-positive nuclei counted. Data from three repeated experiments were pooled. (B) DV-infected ECs were treated with culture supernatants from DV-infected macrophage cultures collected at 6 or 12 h after infection. Half of the macrophage culture supernatants were pretreated with neutralizing anti-TNF-α Ab at 1 μg/ml before they were added to the EC monolayer. The EC monolayer morphology was observed by phase-contrast light microscopy. The magnification is ×200. (C) Microvascular ECs were cultured in medium alone or infected with DV at an MOI of 5 or 10 without other treatments. Other infected ECs were treated with 12-h supernatants collected from DV-infected macrophage cultures with or without anti-TNF-α Ab pretreatment. Infected ECs treated with anti-TNF-α Ab were used as controls. Isotype rat IgG1 was used to demonstrate the specificity of anti-TNF-α Ab-mediated block. Percent EC apoptosis was calculated as described above (A). Data were pooled from three separate experiments. h.p.i., hours postinfection. * and ** indicate P values of <0.05 and <0.01, respectively.

DISCUSSION

It has been documented that high levels of TNF-α are associated with severe dengue disease (2, 13, 34). There is also a positive relationship between soluble TNF receptor (sTNFR)/sTNFR-II levels and the severity of DHF (2, 12). Single-nucleotide polymorphism analysis identified TNF-α polymorphism at TNF allele 308A to be a possible risk factor for hemorrhagic manifestations in DF patients (6). The TNF-α promoter with a polymorphism at position −308 is a stronger transcriptional activator, which is responsible for the production of higher levels of TNF-α (31). The TNF 308A-positive patients are more likely to develop DHF than TNF 308A-negative patients (6). These observations together strongly indicate that high levels of TNF-α are critical for the pathogenesis of severe dengue illness. However, the direct causal relationship between TNF-α and DV hemorrhage has not been clearly established. We showed in this study the direct relationship between high in situ levels of TNF-α and hemorrhage in a DV-infected host (Table 2) and that without TNF-α, the chances to develop hemorrhage are greatly diminished (Table 3). In addition, we observed that tissue viral titers in TNF^−/− mice were higher than those in wild-type mice, yet mice did not develop hemorrhage (data not shown). These data together provide direct evidence for the critical role of TNF-α in dengue hemorrhage.

Our results demonstrated that DV infection induces mouse macrophages to produce TNF-α (Fig. 4A and B). It has been reported that human monocytes infected by DV produce TNF-α (3, 5). Sensitized T cells from dengue patients also produce TNF-α in culture upon restimulation (7, 20). A recent publication reported the cytokine responses of DV-specific memory CD4⁺ T cells in peripheral blood mononuclear cells of volunteers who received experimental live attenuated monovalent DV vaccines (21). While Ags from homotypic DV elicit the highest IFN-γ response, peptide sequences from heterotypic DV elicits CD4⁺ T-cell responses with higher TNF-α⁺-to-IFN-γ⁺ T-cell ratios. In our mouse model, hemorrhage develops at as early as day 3 after virus inoculation, a time too early for T cells to mount a significant response. Thus, our model illustrates the importance of innate immune responses to DV infection in the pathogenesis of hemorrhage. However, it is entirely possible that inappropriate T-cell activation resulting in the production of TNF-α in later stages of primary infection or in secondary infection may worsen the disease.

Vascular ECs line the inner surface of blood vessels and play an important role in vascular functions (17). It is thus a logical assumption that hemorrhage or altered vascular permeability is a result of EC damage. Whether DV targets ECs in dengue patients has been a point of debate for decades until recently, when Jessie et al. reported their immunohistochemical staining and in situ hybridization of biopsy and autopsy specimens from DF/DHF/DSS patients (15). Their report clearly showed the presence of DV Ags in sinusoidal ECs of the liver as well as in the vascular endothelium of the lung (15). Our study of the hemorrhagic mouse tissue also showed that DV targets ECs (Fig. 6). Double staining of hemorrhagic tissues with anti-NS1 and anti-CD31 or anti-CD14 antibodies revealed that viral NS1 Ag was detectable in CD31⁺ cells at 6 h and the intensity increased at 12 h and 24 h after inoculation, but it was not in CD14⁺ cells at any of these early time points, confirming that DV replicates in ECs soon after infection. In addition, unlike what was observed for human skin, neither CD11c⁺ nor CD207⁺ cells stained for DV Ag at these early time points (data not shown). Thus, it appears that DC does not make a major contribution to dengue hemorrhage in our mouse model and that ECs, as an early target of DV, are predisposed by DV to the deleterious events that occur subsequently.

The loss of endothelium integrity in DHF/DSS has long been attributed to the inflammatory mediators released by mononuclear phagocytes, of which TNF-α is the primary candidate (10, 11). However, macrophage production of TNF-α is common to a variety of infections (29) in which a profound alteration of endothelium permeability may not be a prominent feature. It remains a question why TNF-α is specifically of importance to dengue hemorrhage. In their report showing a positive relationship between high sTNFR levels and disease severity, Bethell et al. proposed that the effect of systemic cytokines on the endothelium is determined by the local events that occur, such as viral infection (2). It is interesting that no viral RNA was detectable at day 3 when hemorrhage develops at the subcutaneous tissues, although the presence of virus was documented at day 1 (Fig. 2A). Thus, it appears that the presence of DV predisposes the endothelium to vascular damage triggered by TNF-α, yet its concurrent presence is not required for hemorrhage development.

Nonhuman primates have been used to study dengue (9). The animals showed no signs of disease but developed viremia. The mouse model of DV hemorrhage that we reported in this study does not show all the characteristics of a natural DV infection in humans, but the mice develop hemorrhage and thrombocytopenia. After a person is bitten by an infective mosquito, the virus undergoes replication and circulates in the peripheral blood during the incubation period of 3 to 14 days (average, 4 to 7 days) before the onset of fever (25). Introducing high titers of DV through four intradermal inoculations was critical to the success of our hemorrhage model. The frequency of mice that developed hemorrhage after subcutaneous inoculations was similar to that after intradermal inoculations, but hemorrhage was less severe (data not shown). The virus introduced intradermally, mimicking the natural route of infection, was efficiently spread to other tissues and the circulation (Fig. 2A and B), and hemorrhage developed within only 3 days. It appears that introducing a high DV inoculum intradermally quickly meets the threshold of virus titer that is required to stage a pathogenic event in the murine host. Therefore, our mouse model models only the immediate events that lead to hemorrhage but not the entirety of the pathogenesis of human dengue hemorrhage.

It is noteworthy that hemorrhage development is DV specific, and virus viability is essential. Introducing neither UV-DV nor viable JEV, another mosquito-borne flavivirus, through the same route at an otherwise equivalent titer did not result in hemorrhage. In addition, injecting recombinant TNF-α (8 μg) along with mock C6/36 supernatants through the same route did not induce hemorrhage (data not shown), excluding the possibility that injury (as might be induced through needle sticks) is involved in the process. Based on these observations, we speculate that a specific interaction between DV and cells, especially at high DV titers, perhaps through the activity of a specific viral component(s), is critical for the development of hemorrhage.

In the present study, we established a mouse model of DV hemorrhage through the intradermal inoculation of high titers of virus. By employing this mouse model, we identified TNF-α as being one of the key host factors causing dengue hemorrhage, which is consistent with the recent observation reported by Shresta et al., who found that TNF-α is the key mediator of severe DV-induced disease in mice (27). In our model, hemorrhage develops within 3 days after virus inoculation, a time that is too early for the adaptive immune response to take effect. Although our finding emphasizes the importance of the innate immune response in hemorrhage development, it does not rule out the contribution of nonneutralizing Ab-dependent enhancement in increasing infection or the involvement of TNF-α-producing activated T cells in the pathogenesis of DV hemorrhage in secondary infection. Experiments to address these questions are currently being undertaken.