ABSTRACT

   Dengue virus (DENV) infection remains a critical global health threat,
   with DENV-2 being the most virulent serotype capable of causing lethal
   dengue hemorrhagic fever (DHF), a severe complication characterized by
   plasma leakage and hemorrhagic manifestations. While the search for
   viral receptors and immunocompetent animal models has persisted since
   the first recorded outbreak in 1779, significant gaps remain. Here, we
   establish the first immunocompetent murine model of DHF with intact
   innate/adaptive immunity by generating hTim4-transgenic C57BL/6J mice.
   This model recapitulates fatal DHF complications seen in humans,
   including systemic hemorrhage, dengue encephalitis and intestinal
   ischemia/gangrene. Integrated single-cell RNA sequencing and spatial
   transcriptomics analysis of hemorrhagic gut lesions demonstrated that
   DENV-2 infection induces Syk protein overexpression, leading to
   enhanced Th2 cytokine secretion and impaired hemostatic regulation.
   This cascade enhances vascular permeability, promotes plasma leakage,
   and drives multiorgan hemorrhage, a mechanism corroborated by parallel
   analyses of human DHF tissues. Critically, Th2-biased cytokine storm
   mirrors clinical findings in severe human dengue cases. Our work not
   only identifies hTim4 as a functional DENV-2 receptor but also provides
   a mechanistically grounded platform for DHF pathogenesis studies,
   bridging critical gaps between preclinical models and human
   immunopathology.

   KEYWORDS: Dengue virus, Htim4 receptor, dengue hemorrhagic fever,
   hemorrhage, cytokine storm, th2-bias

Introduction

   Approximately half of the global population is at risk of dengue virus
   (DENV) infection, posing a significant worldwide public health concern
   [[74]1–4] that surpasses HIV as one of the top 10 health threats for
   2019 according to WHO
   ([75]https://www.who.int/health-topics/dengue-and-severe-dengue#tab=tab
   _1). The dengue epidemic is on the rise due to factors such as global
   warming and unplanned urbanization since dengue outbreak in 1779
   [[76]5]. Particularly in 2023, there have been over 5 million reported
   infections globally, including 5000 deaths, nearing a historic peak.
   New regions across the world, including southern Europe, are being
   affected [[77]6]. Notably, in the updated pathogens that could spark
   the next pandemic issued by WHO in August 2024, dengue virus was back
   on the list [[78]7].

   DENV infection can lead to classical dengue fever (DF), severe dengue
   hemorrhagic fever (DHF), and even dengue shock syndrome (DSS),
   characterized by high fever, intense joint and muscle pain (hence DHF's
   nickname “broken bone fever”), systemic hemorrhagic tendency,
   thrombocytopenia, capillary leakage [[79]8,[80]9] and neurological
   complications [[81]10]. The main challenges in combating the dengue
   epidemic include an unclear understanding of its pathogenesis and
   antibody-dependent enhancement (ADE), along with a lack of specific
   treatment options and ineffective vaccines for prevention [[82]11].
   Currently only two dengue vaccines have received approval; one being
   Dengvaxia vaccine which has been associated with severe adverse events
   following vaccination [[83]12–14] and the other being Qdenga vaccine
   which was licensed in Indonesia in August 2022 before later gaining
   approval in Europe [[84]15–17]. However, the long-term data regarding
   their preventive efficacy and safety remains necessary. For decades,
   there has been a persistent need to establish an ideal animal model for
   dengue virus [[85]18], particularly a murine model. Despite significant
   progress made in using various animal species to develop DENV
   infectious models, certain limitations remain inherent to these models
   [[86]18–23]. Rhesus macaques and tree shrews exhibit low viremia
   levels, while bonnet macaques, marmosets, chimpanzees, and Yucatan
   miniature pigs fail to display overt clinical manifestations. Adult
   wild-type mice are resistant to DENV infection and hard to simulate
   hallmark symptoms of DHF or DSS, even at high viral doses. The AG129
   remains the exclusive murine platform selective recapitulating
   hemorrhagic pathology, however, its congenital immunodeficiency leads
   to dysregulated immune responses, rendering it unsuitable for
   investigating DENV immune pathogenesis and restricting its application
   to the evaluation of vaccine-induced neutralizing antibodies, which
   necessitates passive transfer of antibodies from immunized donors
   [[87]18–23]. This has resulted in a critical bottleneck in hemorrhagic
   fever preventive and therapeutic development.

   Notably, host receptors serve as the molecular portals for viral
   invasion, and the functional validation of these receptors is a key
   entry point for constructing susceptible animal models and elucidating
   infection mechanisms. Prior studies have identified numerous candidate
   molecules for DENV receptors [[88]20]. However, these candidates have
   no functional validation evidence in vivo, lacking genetically modified
   murine models harboring these receptors [[89]20,[90]24], which
   substantially impedes the credibility of these molecules as dengue
   virus receptors and their applications in pathogenic mechanisms, the
   development of vaccines and antiviral drugs. Through a systematic
   literature review and analysis, it was demonstrated that within the Tim
   receptor family, only Tim1 and Tim4 mediate DENV invasion via the
   phosphatidylserine-dependent phagocytosis pathway [[91]24]. Candidate
   receptors DC-SIGN serves as a critical mediator in antigen presentation
   and the initiation of immune activation cascades [[92]21]. In contrast,
   Tim4 may contribute to disease progression by modulating antiviral
   immunity through immune responses and viral internalization [[93]24].
   Therefore, Tim1 and Tim4 receptor were selected to develop genetically
   modified murine models. Animal infection experiments confirmed that
   Tim1 transgenic mice were not susceptible to DENV, whereas the Tim4
   receptor molecule significantly enhanced host susceptibility.
   Consequently, this study successfully established a stable and
   heritable hTim4 overexpression immunocompetent mouse model, achieving
   for the precise simulation of DENV infection pathology in a mammalian
   system. This model presents three breakthrough features: (1) It shows
   higher susceptibility to DENV-2 (50 PFU) infection than wild-type adult
   mice [[94]18,[95]20], with viral loads comparable to those in the
   plasma of human severe patients; (2) It systematically reproduces the
   core pathological phenotypes of DHF/DSS, including multi-organ
   hemorrhage (with fatal hemorrhage in the digestive tract as a typical
   manifestation), disseminated intravascular coagulation (DIC), and
   dengue encephalitis-like injury; (3) It detects a Th2-type cytokine
   storm (the phenomenon impaired in AG129 mice) highly consistent with
   that in human severe cases [[96]25–28], providing a quantifiable window
   for the study of immunopathological mechanisms. These findings not only
   provide direct evidence for the receptor function of Tim4 through in
   vivo validation but more importantly establish an animal research
   platform capable of parallel analysis of the entire chain of DENV
   infection-immunity-pathology.

Results

Generate and validate immunocompetent htim4-transgenic murine models

   It has been reported that hTim4 may be a potential receptor of DENV
   virus [[97]20,[98]24]. To address the lack of immunocompetent
   dengue-susceptible murine models, a linearized plasmid containing the
   cDNA of hTim4 gene and EGFP was employed to inject C57BL/6 embryos and
   three founders 17#, 31#, and 69#, were obtained, ([99]Figure 1A and
   Table S1). Only founder 69# exhibited stable germline transmission and
   high hTim4/EGFP expression ([100]Figure 1B,C). To establish a
   dengue-susceptible model that accurately recapitulates clinical
   manifestations while preserving intact immune and coagulation
   functions, we rigourously evaluated the immune competence and
   hemostatic profiles of hTim4-transgenic C57BL/6J mice (termed as hTim4
   mice). In the immune quiescent state, hTim4 mice exhibited immune cell
   levels similar to those of WT mice, while much higher than that of the
   Rag2^-/- deficient mice (positive controls) ([101]Figure 1D and Figure
   S1A). RNA-seq analysis revealed minimal transcriptional alterations
   (185 upregulated and 179 downregulated genes out of a total of 29,253
   genes, <1.2% affected) with no significant disruption of the IFN
   pathway, as determined by Reactome function enrichment analysis
   comparing wild-type C57BL/6J mice (WT mice) with hTim4 mice
   ([102]Figure 1E and Figure S1B,C). In Immunologically activated state,
   the dsRNA analogue poly(I:C) significantly induced IFN-β and CD69+ NK
   (CD3-CD49+) cells increase [[103]29,[104]30] in both WT (P<0.01) and
   hTim4 (P<0.001) mice ([105]Figure 1F-J), while no significant changes
   were observed in CD4 T (CD4+CD8-), CD8 T (CD4-CD8+) and NKT (CD3+CD49+)
   cells ([106]Figure 1H-K); VSV infection has been demonstrated to
   effectively evaluate adaptive immunity [[107]31,[108]32] ([109]Figure
   1L). Viral dynamics showed peak viremia at 3 dpi and clearance by 7 dpi
   in WT and hTim4 mice infected with VSV ([110]Figure 1M and Table S2).
   Significant upregulation of CD8 T cells (7 dpi) and CD69 activation
   marker (3 dpi) was observed ([111]Figure 1O, P). hTim4 mice exhibited
   increased B cell counts at 14 dpi (P<0.05 vs WT, P = 0.119)
   ([112]Figure 1Q) and higher VSV-neutralizing titers [[113]33] (NT50:
   1/8,080-1/19,300) compared to WT (NT50:1/20,200-1/50,800) ([114]Figure
   1R). Collectively, comprehensive immune profiling demonstrated
   preserved immune competence. Coagulation parameters, including platelet
   (PLT) counts, activated partial thromboplastin time (APTT), prothrombin
   time (PT), MPV, PDW, and PCT confirmed intact hemostasis in hTim4 mice
   ([115]Figure 1S-U and Figure S1D, E). In summary, this novel hTim4
   transgenic mouse was immunocompetent amd enables compromising immune or
   coagulation functions that may be suitable for dengue research.

Figure 1.

   [116]Figure 1.
   [117]Open in a new tab

   hTim4 transgenic mouse model with intact immune and coagulation
   systems. (A) the strategy used to generate hTim4 transgenic mice (hTim4
   mice). Zygotes from C57BL/6J mice (WT) were injected with hTim4 Tol2
   constructs. (B, C) verification of hTim4 expression. EGFP fluorescence
   confirmed systemic expression of hTim4 in newborn wild-type WT (gray)
   mice (n = 3) and hTim4 (green) mice (n = 3), visualized by radiant
   efficiency on IVIS Spectrum CT Imaging System (B). The RNA of heart,
   liver, spleen, lung, kidney, intestine, brain and reproductive organs
   collected from 6-week-old WT (n = 3) and hTim4 mice (n = 3) were
   quantified by RT-qPCR (C). (D) the proportions of immune cells. The
   proportions of B cells, T cells and NK cells by flow cytometry between
   the hTim4 group (n = 4) and WT group (n = 6), with Rag2^-/- group
   (n = 8) serving as positive control. (E) genes expression. No
   differential expression was observed among genes related to the IFN
   pathway. (F-K) the innate immune reactions simulated by poly(I:C).
   Experimental design (F). 100 μl of poly(I:C) were injected into each
   mouse on days 1, 3, and 5 between the poly(I:C) WT (n = 4) and
   poly(I:C) hTim4 (n = 4) mice. Equal volumes PBS were done as control.
   The blood and spleen were collected for further studies. IFN-β level by
   RT-qPCR on day 7. (G). The CD4 T cells (CD4+CD8-), CD8 T cells
   (CD4-CD8+), NK cells (CD3-CD49+) and NKT cells (CD3+CD49+) in the
   spleen tissue were counted by flow cytometry on day 7 (H-K). (L-R) the
   adaptive immune reactions infected with VSV. Experimental design (L).
   100 μl VSV (1×10^8 PFU) were nasal inhabited into each mouse (VSV WT,
   n = 12; VSV hTim4, n = 12). Equal volumes PBS were done as control. The
   blood and spleen were then collected for following assays. Blood
   viremia (M). The CD4 T cells (CD4+CD8-), CD8 T cells (CD4-CD8+), CD69
   cells (CD69+) and B cells (CD3-CD19+) in the spleen tissue were counted
   by flow cytometry on 3, 7, 14 dpi (N-Q). The neutralizing antibody
   titer by NT50 (R). (S-U) the intact coagulation function in hTim4 mice.
   No significant differences were found in PLT, PDW, PCT (H), APTT (I),
   and PT (J). data: mean ± SD. Statistical analysis: C, D, G-K, M-U:
   Two-sided unpaired Student's t-test. *P<0.05; **P<0.01; ***P<0.001. The
   exact P values were numerically documented in Table S4

Htim4 mediates DENV-2 infectivity via receptor function in transgenic mice

   To elucidate the function of hTim4, a series of assays were carried
   out. The dose-dependent binding of hTim4-Fc protein to DENV-2 (Tr1751
   strain) ([118]Figure 2A, B) indicates its direct involvement in viral
   recognition, corroborating previous findings [[119]24]. Further
   validation at the cellular level revealed that primary spleen and liver
   cells from hTim4 mice ([120]Figure 2C) exhibited a 4-fold increase in
   viral attachment efficiency at 4℃ ([121]Figure 2D), a 2-fold increase
   in viral entry ([122]Figure 2E), and significantly enhanced viral
   replication ([123]Figure 2F, G), confirming that hTim4 drives the
   infection cascade by promoting viral adsorption, endocytosis, and
   replication [[124]24]. To validate these findings in vivo, we
   established an infectious model. And to circumvent the limitations of
   high-dose infection protocols (e.g. conventional AG129 models) that may
   mask authentic pathogenesis, this study attempted to develop low-dose
   infection research system. Following preliminary optimization,
   subcutaneous and intraperitoneal inoculation routes demonstrated
   suboptimal efficacy, thus the intracranial (I.C.) challenge strategy
   could produce expected results. We identified 6-week-old hTim4 mice as
   highly susceptible to 50 PFU DENV-2, after systematically screening
   3–11-week-old hTim4 mice and testing intracranial (I.C.) inoculation
   doses (50–500 PFU) (Figure S2), manifesting rapid lethality (50%
   mortality at 12 dpi and 100% at 18 dpi, [125]Figure 2H), severe weight
   loss (>25% by day 6, [126]Figure 2I), and progressive clinical
   deterioration (significantly elevated symptom scores from 8 dpi,
   [127]Figure 2J). Crucially, this model recapitulated key features of
   severe human dengue, notably sustained viremia: viral RNA was detected
   in blood from 2 to 9 dpi, peaking at 10^4.5–10^7 copies/mL on day 4,
   with 25% of individuals reaching 10^6 copies/mL ([128]Figure 2K),
   dynamically aligning with clinical severe cases [[129]34]. In summary,
   this study not only supports hTim4 as a critical functional receptor
   for DENV-2 infection but also successfully develops an intracranial
   infection animal model with immunocompetence.

Figure 2.

   [130]Figure 2.
   [131]Open in a new tab

   hTim4 mediates DENV-2 infectivity through its receptor function in
   hTim4-transgenic mice. (A) The procedures for in vitro capturing
   assays. (B) In vitro assays. the relationship between hTim4-Fc content
   and the amount of virus captured. DENV-2 (500 PFU) was mixed with 0,
   0.2, 1 and 5 (μg) hTim4-Fc soluble proteins to form the DENV-2/hTim4
   complex which was then isolated using protein G agarose beads. The
   extracted viral RNA was used to detect virus capture efficiency. (C-E)
   ex vivo assays. Primary spleen and liver cells were infected by DENV-2
   via hTim4 molecular interaction. Primary cells were inoculated for 48 h
   (10^5 cells), followed by infection with DENV-2 (500 PFU) for
   attachment assay at both 4°C and 37°C for 2 h each. (F, G) Viral
   proliferation. Viral proliferation in primary spleen or liver cells
   infected with DENV-2 over time periods ranging from 0 to 48 h
   post-infection respectively. (H-J) Dose optimization (500, 100, 50 PFU)
   for DENV-2 I.C. infection. Survival curves at different infection dose.
   (H). Weight growth rates. Weight loss exceeding 25% was considered
   fatal (I). Clinical scores (J). (K-N) hTim4 mice are highly susceptible
   to dengue virus infection. Both WT and hTim4 mice were challenged with
   50 PFU (50 PFU, I.C.) of DENV-2 Tr.1751. An equal volume of DMEM was
   inoculated as the control. All infected mice died within 18 dpi. (K).
   Weight growth rates decreased since 6 dpi. (>25% weight loss is
   considered fatal (L)). Clinical scores of these hTim4 mice were higher
   than those of WT mice (M). Blood viremia levels peaked at 4 dpi.,
   lasted beyond 9 dpi. (N). Data: mean ± SD. Statistical analysis: B, D,
   E, K, N: Two-sided unpaired Student's t-test. F, G, I, L, M: Two-Way
   ANOVA. H, K: Mantel-Cox. *P<0.05; **P<0.01; ***P<0.001. The exact P
   values were numerically documented in Table S4.

DENV-2 intracranial infection in htim4 mice simulates dengue-like
encephalitis

   Clinically, some patients have been observed to develop dengue
   encephalitis. Therefore, we aim to investigate whether hTim4 mice
   exhibit similar symptoms. We observed that DENV-2 infected mice
   exhibited clinical signs, including hunching, emaciation, tearing
   ([132]Figure 3A,B), and characteristic neurological symptoms such as
   slow action ([133]Figure 3C and Table S6) and limb paralysis
   accompanied by ataxia ([134]Figure 3D,E and Table S6), phenotypes
   closely resembling the “breakbone fever” manifestations of dengue
   hemorrhagic fever [[135]8,[136]9]. Pathological evaluation of brain
   tissues from six infected hTim4 mice demonstrated the following
   lesions: neuronal degeneration or necrosis (mild, +) in two cases,
   perivascular infiltration cell infiltration (mild, +) in six cases,
   microglial nodule formation (mild, +) in three cases, astrocytic
   hyperplasia (mild, +) in two cases, and hemorrhage (mild, +) in three
   instances ([137]Figure 3F-I). Hemorrhagic features were confirmed
   through triple-validation methods, including bright-field microscopy,
   Evans blue permeability assay, and H&E staining ([138]Figure 3F-I),
   with lesions localized to cortical regions (+), sporadic neuronal
   necrosis (→), and perivascular infiltration (*). These patterns
   precisely mirrored the spontaneous parenchymal microhemorrhage profile
   of human dengue encephalitis [[139]10]. Despite pathological
   involvement in multiple brain regions (e.g. thalamus and basal ganglia)
   [[140]10] the observed injury severity (mild) suggested that direct
   DENV-2-induced brain damage was not the primary cause of mortality. In
   conclusion, the hTim4 mouse model faithfully recapitulates the
   neuroinvasive phenotypes and encephalitis-specific pathological
   progression of intracranial DENV-2 infection, providing a robust
   disease model for investigating dengue neuropathogenesis. The lethal
   outcome in this model may stem from systemic pathological cascades
   triggered by viremia rather than direct central nervous system injury.

Figure 3.

   [141]Figure 3.
   [142]Open in a new tab

   The hTim4-transgenic murine model of DENV-2 intracerebral challenge
   recapitulates key clinical features of dengue-associated encephalitis.
   Six-week-old hTim4 mice were intracranially inoculated with DENV-2 at a
   dose of 50 PFU (DENV-2 group, n = 44), while the control group received
   DMEM infection (DMEM group, n = 22). (A-E) Visible symptoms and
   clinical score. Visible symptoms in hTim4 mice infected with DENV-2
   included viral load of brain (A), hunchback, tearing (B), slow action
   (C), limb paralysis (D) and ataxia (E). (F) Brain hemorrhage was
   observed through naked eye observation and Evan's blue staining. (G)
   The injury grade among 6 randomly selected DENV-2 infected mice
   subjected to pathological examination. (H,I) Pathological features in
   brain with DENV-2 infected mice. (H) perivascular infiltration (*),
   neuronal necrosis (→), and bleeding (+). (E) Pathological features of
   lung hemorrhage. bleeding (+), pulmonary hemosiderosis (▴), and edema
   (★). (I) Neuronal degeneration/necrosis (+), perivascular infiltration
   (+), microglial nodules (+), astrocytic hyperplasia (+) and hemorrhage
   (+) were graded mild. (I) Data are presented as mean ± SD. Statistical
   analysis: A, Two-sided unpaired Student's t-test. C-E, Two-way ANOVA.
   *P<0.05; **P<0.01; ***P<0.001. The exact P values were numerically
   documented in Table S4.

DENV-2 intracranial infection in htim4 mice recapitulates DHF manifestations

   Systemic hemorrhage, as a core clinical manifestation of DHF,
   frequently progressed to disseminated intravascular coagulation (DIC),
   also named DSS [[143]8]. However, recapitulating this complex
   pathological cascade in animal models has proven challenging
   [[144]20,[145]21]. Our study demonstrates that DENV-2-infected hTim4
   transgenic mice precisely mimic these clinical features, with
   characteristic hemorrhagic foci observed in multiple organs including
   brain ([146]Figure 3), intestine, lung, heart, liver, and spleen
   ([147]Figure 4A, B and Figure S3A-C). Systematic coagulation analysis
   revealed prolonged clotting time from 4 days post-infection (dpi),
   significantly increased tail bleeding points by 8 dpi (P < 0.05,
   FigureS3D,E), accompanied by thrombocytopenia ([148]Figure 4C), reduced
   mean corpuscular volume (MCV) (P < 0.05, FigureS3F) and over 50%
   prolongation of APTT and PT ([149]Figure 4D,E). The observed APTT
   prolongation (>1.5× reference range) and PT extension (> 3s) closely
   mirror diagnostic criteria for human DIC, validating the clinical
   relevance of this model. Notably, concurrent reductions in WBC and NEU
   counts (Figure S3G,H) suggest immune dysregulation may exacerbate
   coagulation disorders. To further elucidate the hemorrhage dynamic
   changes, taken intestinal hemorrhage for example, we divided the DENV-2
   mice into mild, moderate and severe state/condition based on
   pathological intestinal scoring (Table S7) and intestinal perfusion
   pressure drop rate (Table S3). Using high resolution laser speckle
   contrast imaging (HR-LSCI) [[150]35], we quantitatively characterized
   DSS hemodynamics: intestinal blood flow velocity progressively declined
   (Table S3, [151]Figure 4F, S3I) with disease severity, accompanied by
   28.48% (moderate) and 42.23% (severe) reductions in perfusion pressure
   respectively ([152]Figure 4G,H) – consistent with clinical DSS staging
   criteria (early phase: 10–20% decline; late phase: >30% decline)
   [[153]36–38]. Immunohistochemical staining verified that platelet
   endothelial cell adhesion molecule-1 (PECAM-1/CD31) labelling showed no
   significant differences in intestinal microvascular density or
   structural integrity between infected and control groups ([154]Figure
   4I-K), suggesting preserved vascular architecture. HE staining analysis
   revealed severe disruption of intestinal villi architecture in critical
   cases, manifested by blood cell extravasation, nuclear atypia, and
   cytoplasmic atrophy ([155]Figure 4L). Immunofluorescence staining
   assays discovered DENV-2 particles (brown, ➫) co-localizing with host
   receptor hTim4 (green) at lesion sites in infected group, directly
   demonstrating that those pathological changes are caused by DENV-2
   infection ([156]Figure 4M,N): villous layer shrinkage (purple) and
   epithelial disorganization (irregular nuclear arrangement, blue,
   [157]Figure 4M), albumin (ALB, yellow) penetrates beyond the vessel
   wall ( Inline graphic , CD31, purple) ([158]Figure 4N). Intriguingly,
   VE-cadherin (purple, ➫) expression showed no statistical difference
   between groups ([159]Figure 4O), implying a VE-cadherin independent
   mechanism underlying vascular leakage. In essence, we found that hTim4
   mice recapitulates faithfully the clinic-pathological progression of
   coagulation dysfunction, microcirculatory failure, and multi-organ
   hemorrhage induced by DENV-2 infection, demonstrating that hTim4 mice
   might serve as a pivotal experimental model for investigating the
   hemorrhagic pathogenesis of DHF/DSS.

Figure 4.

   [160]Figure 4.
   [161]Open in a new tab

   The hTim4-transgenic murine model of DENV-2 intracerebral challenge
   recapitulates pathophysiological hallmarks of DHF/DSS. (A) Viral load
   of intestine. (B) Intestine hemorrhage in hTim4 mice with DENV-2
   infection. (C-E) Blood coagulation function characteristics after DHF
   induced by viral infection showed PLT decrease (C), APTT (D) and PT (E)
   prolongation. (F-H) Blood flow features and blood perfusion pressure
   were assessed using HR-LSCI laser speckle blood flow imaging system. In
   the DMEM group, all mice had healthy intestines (F). In the DENV-2
   group, hTim4 mice infected with DHF were divided into moderate and
   severe groups based on disease severity. The color signal indicated
   decreased blood flow in the moderate group and stasis in the severe
   group; a stronger red signal did not indicate an increased number of
   blood vessels but rather faster blood flow velocity (F). Blood
   perfusion pressure was measured during systole and diastole stages (G,
   H). (I-K) No significant pathophysiological changes were observed in
   blood vessels between the DENV-2 infected group and DMEM group.
   Immunohistochemical staining with CD31 was used to label the intestinal
   vessels (I). The vascular density (J) and diameter (K) of both cross
   and longitudinal sections of the intestine were measured. (L)
   Pathological features of intestinal bleeding (3DHISTECH P250 FLASH
   system). Intestinal villi injury, crypt collapse, disorderly
   arrangement of intestinal columnar epithelial cells, and cytoplasm
   atrophy (↔) were observed in the DENV-2 group. (M-O) Molecular
   characteristics related to changes in vascular permeability were
   investigated using Zeiss LSM900 with Airyscan2 system. Nuclear
   disarrangement (blue), hTim4 (green), DENV-2 (brown, ➫) cytoplasm
   atrophy and intestinal villi injury (purple, M); albumin (yellow)
   leakage from vessels ( Inline graphic , N); VE-cadherin (purple, ➫)
   expression remained unchanged (O). Data are presented as mean ± SD.
   Statistical analysis: A, C-E, H: Two-sided unpaired Student's t-test.
   J, K: Two-sided unpaired Student's t-test (welch).*P<0.05; **P<0.01;
   ***P<0.001. The exact P values were numerically documented in Table S4.

Single-cell and spatial omics uncover syk-driven immune-mediated hemorrhage

   Currently, the core regulatory genes mediating intestinal ischemia and
   hemorrhagic phenotypes, along with their spatial distribution
   characteristics, remain unclear. To systematically investigate the
   molecular association between intestinal gene expression profiles and
   vascular permeability abnormalities, this study integrated single-cell
   RNA sequencing (scRNA-seq) and spatial transcriptomics – the latter
   enabling in situ capture of spatial localization information for gene
   expression within tissue microenvironments to establish molecular
   mapping. Experiments involved parallel analysis of uninfected (DMEM
   control) and DENV-2-infected mouse intestinal tissue samples. Raw
   sequencing data underwent standardized processing, dimensionality
   reduction (UMAP/tSNE), cell clustering, and differential expression
   gene (DEG) screening through the Seurat pipeline. The scRNA-seq
   successfully identified 14 intestinal cell subtypes: B cells, hemogenic
   endothelial cells, endothelial cells, enterocytes, epithelial cells,
   goblet cells, immune cells, macrophages, mesenchymal cells, NK cells,
   Paneth cells, smooth muscle cells, T cells, and Tuft cells ([162]Figure
   5A). Classification accuracy was validated by subtype-specific
   molecular markers (top 20 DEGs, log[2]|FC|≥0.25) ([163]Figure 5B).
   Quantitative analysis revealed significant immune cell expansion in
   DENV-2-infected group versus control: absolute counts of B cells (3022
   vs 281), T cells (1786 vs 617), NK cells (1170 vs 794), macrophages
   (283 vs 181), and unclassified immune cells (502 vs 128) were markedly
   elevated ([164]Figure 5C). Notably, however, relative proportions of
   macrophages (2.9% vs 3.03%) and NK cells (12.01% vs 13.3%) showed
   slight decreases in infected group ([165]Figure 5C), suggesting dynamic
   equilibrium regulation among immune subsets. To further validate these
   findings, spatial transcriptomics data were integrated to construct
   intestinal tissue molecular maps. tSNE visualization demonstrated
   significant spatial distribution heterogeneity between DENV-2 and DMEM
   control groups across 14 cell subtypes ([166]Figure 5D-F). Key immune
   cells (NK, T, B, macrophages, and cluster cells) exhibited pronounced
   spatial co-localization in infected group ([167]Figure 5G and Figure
   S4A-N), aligning with scRNA-seq identified immune expansion. This
   finding is consistent with clinical observations in patients,
   demonstrating an increase in immune cell count without substantial
   immune cell aggregation and infiltration.

Figure 5.

   [168]Figure 5.
   [169]Open in a new tab

   Molecular signatures of intestinal bleeding in mice with DHF. (A-C)
   Molecular signatures of scRNA-seq analysis. We utilized the Cell Ranger
   software's (V6.0) cellranger count module for alignment, filtering,
   barcode counting, and UMI counting to generate feature barcode matrices
   and identify clusters. Furthermore, cells with fewer than 200 genes or
   lower than 99 UMI, as well as those with mitochondrial gene proportions
   exceeding 25%, are filtered using Seurat (v4.0.0). Following the
   standardization and normalization of sample information, the samples
   were categorized into 14 distinct cell types via UMAP mapping
   visualization. UMAP revealed the differences in expression and
   distribution patterns among cell types between the DMEM and DENV-2
   groups (A). The heatmap of the top 20 genes for each cell type (B). The
   cell count and percentage of different cell types were compared between
   the DMEM and DENV-2 groups (C). (D-G) Molecular signatures of spatial
   sequencing analysis. Each 10 μm frozen intestine tissue section from
   the DMEM group and DENV-2 group was placed on one visium gene
   expression slide capture area per slide. Bright-field images were
   acquired following the Spatial Transcriptomics procedure, and
   sequencing libraries were constructed using the visium Spatial Gene
   Expression Slide & Reagent kit (10 × Genomics) for Novaseq PE150
   sequencing. The expressions of sequencing libraries underwent
   normalization, dimensionality reduction, spot clustering, and
   differential expression analysis using Seurat software (D). tSNE
   mapping visualized the expression and distribution of all gene spots
   across the samples (E). UMAP mapping demonstrated the expression and
   distribution differences between DMEM and DENV-2 groups (F). Cell
   types’ spatial location distribution. NK cells, paneth cells, tuft
   cells, T cells, macrophages, and B cells being located together showed
   the evidence of immune response presence (G).

   Considering these subsets’ capacity to secrete inflammatory mediators,
   their aggregation may exacerbate immune-mediated intestinal hemorrhage
   through cytokine storm induction [[170]5,[171]39]. For mechanistic
   exploration, multi-dimensional functional annotation of DEGs was
   performed: GO analysis of scRNA-seq DEGs revealed three core functional
   modules – extracellular matrix (ECM) remodeling, hemostasis regulation,
   and cytokine/chemokine network modulation ([172]Figure 6A and Figure
   S5). Cytokine modules were primarily driven by B cells, immune-related
   cells, macrophages, and T/NK cells; hemostasis functions correlated
   with B cells, endothelial cells, and smooth muscle cells; ECM
   remodeling involved coordinated actions of hemogenic endothelial,
   mesenchymal, and smooth muscle cells; Reactome pathway analysis of
   spatial transcriptomic DEGs from cluster cells further confirmed
   central pathways in immune dysregulation, coagulation abnormalities,
   and cell junction remodeling ([173]Figure 6B,C and Figure S4O,P),
   corroborating GO findings. Cross-integration of spatial and single-cell
   DEGs identified 10 hub genes: Col1a2, Col1a1, Col3a1, Col4a1, Serpinh1,
   Cd9 (downregulated), and Syk, Cd74, Ctss, Cd36 (upregulated)
   ([174]Figure 6D). The qPCR validation confirmed significant
   downregulation of Col1a2, Col1a1, Col3a1, and Serpinh1, with concurrent
   upregulation of Syk, Cd74, and Cd36. Notably, Cd9, Ctss and Col4a1
   exhibited no statistically significant differences ([175]Figure 6E).
   Previous studies indicate that Syk promotes Th2-type inflammatory
   responses [[176]40,[177]41], while the roles of Cd74 and Cd36 in
   Th1/Th2 polarization bias remain controversial [[178]42–45]. These
   genes are functionally associated with coagulation regulators
   [[179]46]. In this study, we identified of Syk as a core node bridging
   immune responses (cytokine secretion) and coagulation/hemorrhage
   processes ([180]Figure 6F), as well as the regulation role of Syk to
   cytokine storm network by protein–protein interaction (PPI) analysis
   ([181]Figure 6G), substantiate the existing pathogenic hypothesis of
   DHF: Cytokine storms can directly disrupt endothelial tight junctions,
   significantly increasing vascular permeability and triggering
   hemorrhage with multi-organ dysfunction [[182]35, [183]40–44].
   Collectively, these data indicated that Syk drove the occurrence of DHF
   via regulating cytokine secretion and vascular endothelial dysfunction.

Figure 6.

   [184]Figure 6.
   [185]Open in a new tab

   DENV-2 infection induces a Th2-type cytokine storm, leading to
   increased vascular permeability. (A) GO function enrichment analysis of
   DEGs among all cell types between the DMEM and DENV-2 groups. (B) Gene
   spots were divided into nine clusters ranging from 0 to 8 by tSNE map.
   (C) Reactome function enrichment analysis. Reactome function enrichment
   analysis of DEGs in clusters 0–8 were showed in heatmap. (D) The
   overlap analysis. The overlap analysis among the DEGs from the pathways
   co-related to platelet activation/aggregation, hemostasis,
   extracellular matrix organization between GO and reactome function
   enrichment analysis. (E) The mRNA level of target genes. (F) The key
   genes CD36, CD74 and Syk. CD36, CD74 and Syk linked the network of
   coagulation dysfunction and immune response. (G) The relationship
   between Syk and the production and release of cytokines showed by
   protein-protein interaction (PPI) network analysis.

Syk-mediated th2-polarized cytokine storm

   Intensive studies have elucidated that cytokine storms are a key factor
   causing DHF[[186]1,[187]3,[188]8], and we revealed that Sky might
   regulate the secretion of cytokines ([189]Figure 6G), and then we
   wanted to know the bias of cytokines. Initially, flow cytometry was
   employed to analyse serum cytokine profiles in hTim4 mice with/without
   DENV-2-infection. To eliminate potential interference from
   hTim4-induced immune responses [[190]47], a parallel VSV infection
   control experiment was specifically designed (Figure S6A). Results
   demonstrated significant elevation of Th1 cytokines in both groups
   post-VSV infection, confirming viral factors rather than hTim4 as the
   primary driver of Th1 responses, thereby effectively excluding
   hTim4-related confounding effects on Th2 detection. The results showed
   that hTim4 mice exhibited marked Th2-polarized cytokine storm
   characteristics compared to DMEM controls: Il6 (5.3-fold increase, P <
   0.01), Il10 (1.4-fold increase, P < 0.05), Ccl2 (1.48-fold increase, P
   < 0.05), and Ccl4 (1.44-fold increase, P < 0.05) were significantly
   upregulated ([191]Figure 7A), while Th1 cytokines (I12–2, Il12p40/p70,
   Ifng) showed no statistically significant alterations [[192]5,[193]27].
   To reveal the relationship between Sky gene expression and cytokine
   secretion, qRT-PCR validation revealed significant Syk upregulation in
   intestinal tissues of DENV-2-infected hTim4 mice (P < 0.05, [194]Figure
   7B, C and Table S2), accompanied by markedly elevated Il10 (P < 0.001,
   [195]Figure 7E) and a trend towards Il6 increase (P = 0.4541,
   [196]Figure 7D). In contrast, WT mice showed only marginal trends in
   Syk and Il6 expression, with significant Il10 downregulation (P <
   0.05). scRNA-seq further delineated a Th2-characteristic molecular
   signature: significant upregulation of Il10, Ccl2, and Ccl5 (RANTES)
   alongside pronounced Ifng reduction (Figure S6B, C). Multi-omics data
   collectively confirmed Th2-type immune bias in hTim4 mice post-DENV-2
   infection. To validate the clinical relevance of this Th2 bias, serum
   samples from a human cohort during the 2014–2015 DENV-2 epidemic in
   Southern China (including DF/DHF patients and healthy controls) were
   analysed. The human infection group displayed a Th2-dominant cytokine
   storm pattern highly consistent with the murine model ([197]Figure 7F),
   establishing cross-species conservation of this phenomenon. Evidences
   from clinical observations also indicated that Syk activation promotes
   the secretion of proinflammatory mediators such as Tnfα and Il6,
   potentially triggering systemic cytokine storms [[198]48].
   Additionally, patients with DH demonstrate a shift towards Th2-dominant
   cytokine responses [[199]28]. In conclusion, this study systematically
   elucidates that DENV-2 infection induces Syk-mediated Th2-polarized
   cytokine storms, driving further the occurrence of DHF, providing
   critical insights into its pathological association with vascular
   hyperpermeability.

Figure 7.

   [200]Figure 7.
   [201]Open in a new tab

   The validation of conserved Th2 cytokine storm signatures across human
   and murine systems. (A) Elevated levels of Th2-type cytokines and
   pro-Th2 chemokines were observed. Il2, Il5, Il6, Il9, Il10, Il12, Ifng,
   Ccl2, Ccl3, Ccl4 and Ccl5 were examined in DENV-2 infected hTim4 mice
   (n = 8), DMEM mice (n = 6) and wild-type C57BL/6 mice (WT, n = 8).
   (B-E) Validation of Syk driven Th2 bias by RT-qPCR. In the intestinal
   tissue form DENV-2 infected WT and hTim4 mice, the mRNA level of
   DENV-2, Syk, Il6 and Il10. (F) The cytokines bias in DENV-infected
   patients (n = 10) and healthy humans (n = 10). (G) The graphical
   abstract of our study. Data: mean ± SD. Statistical analysis: A-F:
   Two-sided unpaired Student's t-test. * P<0.05; **P<0.01; ***P<0.001.
   The exact P values were numerically documented in Table S4.

Discussion

   This study systematically confirmed through in vitro, ex vivo and in
   vivo experiments that the hTim4 molecule is a key functional receptor
   for DENV-2 infection. The hTim4 transgenic mouse model demonstrated
   similar disease phenotype with DHF in terms of immune response and
   molecular mechanism ([202]Figures 2–7), providing a novel animal model
   that is both scientifically rigourous and clinically relevant for the
   analysis of dengue fever pathogenesis, and may further be developed for
   antiviral drug screening or vaccine evaluation. This model mediates
   DENV-2 infection through the natural receptor hTim4, precisely
   reproducing the hallmark features of dengue infection observed in
   humans, including high viremia ([203]Figure 2), atypical symptoms or
   microhemorrhage and inflammatory response in the brain parenchyma with
   clinical dengue encephalitis features ([204]Figure 3), as well as
   systemic hemorrhage and DSS induced by disseminated intravascular
   coagulation (DIC) ([205]Figure 4). The infection-induced plasma leakage
   (ALB exudation, [206]Figure 4M) and decreased mesenteric blood
   perfusion pressure (decrease > 40%, [207]Figure 4E-G and Table S3) are
   highly consistent with clinical DHF/DSS symptoms; the pathological
   features in the brain are highly consistent with the thalamus/basal
   ganglia injury pattern in human dengue-related
   encephalitis[[208]10,[209]49]. Compared with the currently recognized
   AG129 mouse model (deficient in interferon-α/β/γ receptors), hTim4 mice
   have several advantages: (1) They Retaining the complete type I/II
   interferon signalling pathway, overcoming the abnormal function of
   immune deficiency [[210]23]; (2) Mediating DENV infection through a
   natural receptor precisely simulates the Th2-type cytokine response
   ([211]Figures 5–7), consistent with the immune characteristics of
   severe clinical patients (IL-10↑/IFN-γ↓) [[212]25–28, [213]50]; (3)
   Overcomes the Th1 immune bias of C57BL/6 wild-type mice [[214]51] and
   the Th1/Th2 response damage caused by immune deficiency in AG129
   [[215]21,[216]23], more closely resembling the immune response pattern
   in the human population.

   Dengue fever is naturally transmitted through arthropod-borne routes
   [[217]52], whereas this study employs I.C. inoculation ([218]Figures
   2–4), which has certain limitations. The I.C. route cannot fully
   replicate the entire complexity of mosquito-borne transmission
   infection, especially the potential impact of mosquito salivary factors
   mediating immune regulation on infection and the complex interaction
   among vector-host–pathogen. The core objective of this study is to
   reveal the pathological changes induced by DENV infection leading to
   DHF/DSS, rather than to simulate the entire natural infection history.
   It is worth noting that existing studies have shown that despite
   different routes (subcutaneous, intraperitoneal, intracranial), key
   pathogenic phenotypes (such as plasma leakage) can be induced in the
   AG129 mouse model [[219]20, [220]22, [221]53–55]. This indicates that
   certain core pathogenic mechanisms (such as vascular leakage) may be
   intrinsic responses of tissues/hosts rather than absolutely dependent
   on specific initial infection routes, which partly supports the
   rationality of our use of the intracranial route to study the dengue
   hemorrhagic fever disease model. Secondly, I.C. inoculation effectively
   bypasses the peripheral immune surveillance mechanism, facilitating the
   study of early inflammatory cascade reactions; the induced neurological
   symptoms can systematically analyse the pathophysiology of the
   neurovascular unit, which would be masked in the multi-system dynamics
   of natural infection; it extends the window period of the disease
   process ([222]Figure 2), which is conducive to studying the
   pathological mechanism of immune disorder-plasma leakage.

   Clinically, dengue encephalitis is predominantly mild or asymptomatic
   [[223]56,[224]57], while severe dengue (e.g. DHF/DSS) is key fatal due
   to antibody-dependent enhancement (ADE) and cytokine storm-induced
   vascular leakage syndrome (shock, multi-organ failure) within 3–7 days
   post-infection ([225]https://apps.who.int/iris/handle/10665/44188)
   [[226]49]. Consistent with this, AG129 mice exhibit rapid mortality
   within 7 days due to immunodeficiency
   [[227]23,[228]53,[229]54,[230]58]. However, their pathological features
   (e.g. multi-organ hemorrhage) likely result from non-specific immune
   exhaustion triggered by ultra-high viral doses (10^4∼8 PFU), which
   fundamentally differ from human immunopathology
   [[231]23,[232]53,[233]54,[234]58]. Such high doses may breach
   physiological immune barriers, causing non-target tissue damage and
   obscuring authentic virus-host dynamics. In addition, wild-type BALB/c
   (DENV-2) and C57BL/6 (DENV-3) mice develop severe encephalitis and
   brain tissue damage following I.C. inoculation
   [[235]51,[236]59,[237]60]. whereas hTim4 mice display only mild
   encephalitis ([238]Figure 3), suggesting their utility in modeling
   asymptomatic/mild clinical encephalitis.

   Notably, while severe dengue (DHF/DSS) is more common in secondary
   infections, primary infections can also induce critical illness
   [[239]49]. A clinical study of 344 pediatric cases showed comparable
   severe disease rates between primary (32.5%, 112 cases) and secondary
   infections (32.7%). Infants (≤1 year) exhibited the highest severe rate
   (64.7%), exclusively associated with primary infections; among fatal
   cases, 5 originated from primary and 2 from secondary infections
   [[240]49]. These primary infection cases met the WHO 2009 criteria for
   severe dengue, including shock (pulse pressure ≤20 mmHg), severe plasma
   leakage, hemorrhage (gastrointestinal/intracranial), and organ failure,
   with mortality driven by circulatory collapse and multi-organ
   dysfunction rather than encephalitis
   ([241]https://apps.who.int/iris/handle/10665/44188) [[242]49]. These
   clinical findings converge with our research conclusions to
   substantiate the need for critical reappraisal of ADE in secondary
   infections as the predominant pathogenic mechanism underlying severe
   dengue manifestations. Notably, multifactorial determinants – including
   serotype-specific variations, virulence determinants of dengue strains,
   direct viral endothelial cytotoxicity, and host immune dysregulation –
   may collectively orchestrate the pathological progression towards
   severe disease outcomes in primary infections. Compared to the dengue
   fever disease model of AG129 model, the hTim4 mouse model is low-dose
   specificity, requiring only 50 PFU to trigger controlled inflammatory
   responses (e.g. IFN-λ-mediated antiviral signalling), avoiding
   high-dose-induced non-target tissue damage ([243]Figure 2–4). And it
   Preserved natural immune pathways, faithfully reflecting virus-host
   equilibria ([244]Figure 1). Extended disease progression window (50%
   survival: 12 days; 100% mortality: 18 days) enables phased analysis of
   Th2 polarization and vascular leakage mechanisms ([245]Figures 2 and
   [246]7). Although AG129 mice rapidly replicate the 7-day lethal window,
   their immunodeficiency-driven multi-organ hemorrhage (a direct
   consequence of innate immune deficiency) diverges from clinically
   relevant immune-driven mechanisms (e.g. endothelial injury)
   [[247]21,[248]23]. As to adult wild-type mice, they were not
   susceptible to DENV [[249]61,[250]62], failed to develop hallmark
   symptoms of DHF/DSS even at high viral doses [[251]51] and maintain an
   prolonged infection window ([252]Figure 2N). In contrast, hTim4 mice
   recapitulate vascular leakage and organ damage at ultra-low doses,
   closely mirroring human pathology and providing a robust tool to
   transcend ADE limitations and explore novel strain-host interaction
   mechanisms. Future optimizations, including viral serotype/strain
   selection and alternative infection routes will enhance the clinical
   translatability of vascular leakage phenotypes. In summary, while hTim4
   mice cannot fully replicate natural vector-borne transmission, they
   offer critical insights into primary infection-driven severe mechanisms
   of DHF/DSS.

   Our study employed integrated scRNA sequencing and spatial
   transcriptomics to elucidate the molecular mechanisms driving immune
   dysregulation and hemorrhagic manifestations during DENV-2 infection.
   Our findings demonstrate that intestinal B cells, vascular endothelial
   cells, and immune-associated cells collectively orchestrate
   dysregulation of cytokine-coagulation crosstalk through coordinated
   modulation of prothrombotic factors (Col1a1, Col3a1) [[253]46] and the
   Syk signalling network ([254]Figure 5,[255]6 and Figure S4–5). It is
   worth noting, the Syk molecular may plays key roles. Our results
   indicated that three core nodes emerged at the immune-hemostatic
   interface ([256]Figure 6). Syk potentially mediates Th2-polarized
   cytokine secretion (Il-6/Il-10) [[257]40,[258]41], Cd36 enhances
   hematoma resolution via Il-10/Stat3 signalling [[259]42,[260]43] and
   Cd74 synergizes with MIF to amplify inflammatory cascades
   [[261]44,[262]45]. Experimental validation revealed Syk upregulation in
   DENV-2-infected murine tissues, correlating with Th2-skewed cytokine
   profiles mirroring clinical dengue cases ([263]Figure 7 and Figure S6).
   The observed Syk-cytokine network interactions ([264]Figure 6) suggest
   Syk could coordinately drive hemorrhagic pathogenesis through dual
   modulation of immune and coagulation systems. Notably, primary
   infection cohorts exhibited severe phenotypes pathologically analogous
   to ADE-associated secondary infections, with comparable epidemiological
   prevalence [[265]49]. While our multi-omics data predominantly support
   direct cytokine storm-mediated endothelial injury – particularly
   through IL-10 overproduction – partial immunological overlap with ADE
   mechanisms was evident, including monocyte/macrophage activation and
   Il-10 dysregulation. Although classical ADE typically involves
   heterotypic reinfection via Fcγ receptor-mediated viral internalization
   and innate immunity suppression [[266]63], we identified ADE-like
   immunological signatures in severe primary infections, implying
   alternative pathway being activated. In all, we propose hypotheses to
   explain the pathogenesis model: both primary infection-induced Th2
   polarization and secondary infection-mediated ADE ultimately disrupt
   Syk-regulated cytokine-coagulation homeostasis, culminating in vascular
   leakage. These insights inform further a precision therapeutic
   framework: strategic Syk inhibition may rebalance Th1/Th2 responses to
   mitigate cytokine storms, while concurrent Cd36/Cd74 targeting could
   address coagulation abnormalities. Targeting Syk signalling pathway
   represents a possibly therapeutic strategy for dengue fever. Activation
   of Syk contributes to severe pathological processes, including vascular
   leakage and cytokine storm, by promoting the release of
   pro-inflammatory cytokines and excessive activation of immune cells.
   Inhibition of Syk may mitigate dengue-related tissue damage through
   regulation of the immune pathological response; however, its direct
   effects on viral replication require further verification. Currently,
   the clinically approved Syk inhibitor Fostamatinib is predominantly
   utilized for treating adult chronic immune thrombocytopenia (ITP)
   [[267]64], with no studies exploring its application in dengue fever.
   Other investigational Syk inhibitors, such as Entospletinib and
   PRT062607, are primarily being evaluated in clinical trials for
   autoimmune diseases [[268]65,[269]66] and hematological malignancies
   [[270]67,[271]68]. Notably, PRT062607 has demonstrated potential for
   intervening in sepsis-induced heart dysfunction and multiple organ
   failure [[272]69]. Based on existing research evidence, repurposing
   evaluations of Syk inhibitors or the development of next-generation
   compounds with enhanced pharmacokinetic properties and safety profiles
   could provide a breakthrough direction for this field. Given limited
   antiviral efficacy of single-pathway interventions, combinatorial
   regimens incorporating neutralizing monoclonal antibodies are
   warranted. Vaccine development must prioritize Th1-biased immunity to
   counter pathogenic Th2 polarization while circumventing ADE risks. By
   delineating the immune-hemostatic interactome, this work establishes a
   multi-target intervention paradigm for DHF management.

   This study demonstrated that the Th2-biased cytokine storm pattern
   observed in hTim4 mice with DENV-2 infection, characterized by elevated
   levels of Il6 (**P = 0.0082), Il9 (P = 0.3965), Il10 (*P = 0.0168), was
   highly consistent with the serum profiles of clinical dengue fever
   patients ([273]Figure 7). Notably, the Il10 levels and liver injury
   markers AST/ALT revealed a strong positive correlation [[274]70]. Given
   that liver injury was associated with coagulation dysfunction, high
   Il10 expression was verified to closely link to adverse outcomes in
   both dengue fever and COVID-19 [[275]71,[276]72]. As reported, primary
   infection predominantly induced a Th1 immune response
   [[277]73,[278]74], whereas secondary infection markedly enhanced Th2
   cytokine production [[279]75,[280]76]. This shift from Th1 to Th2
   immunity might exacerbate plasma leakage and organ damage through
   endothelial dysfunction mediated by Il6 and Il8 [[281]77,[282]78],
   thereby promoting the progression of severe dengue manifestations, such
   as DHF and DSS [[283]79,[284]80]. Previous studies exhibited divergent
   perspectives on the temporal dynamics of Th2 responses: some indicated
   that these responses predominantly occurred during the recovery phase
   [[285]81], whereas others highlighted their persistence throughout the
   acute and defervescence phases [[286]77]. Our findings not only support
   the association between Th2 polarization and the progression of severe
   dengue [[287]77,[288]82], but also offer a novel framework for
   validating traditional disease-stage immune models. This research
   innovatively developed an hTim4 mouse infection model to
   comprehensively investigate the molecular mechanisms underlying
   DENV-2-induced hemorrhagic manifestations resembling DHF/DSS. Utilizing
   a robust experimental framework, the study integrated intracranial
   viral challenge, stringent virus control experiments, cross-species
   clinical validation, and multi-omics analysis. These approaches
   collectively revealed that DENV-2 infection triggers Syk-mediated
   Th2-polarized cytokine storms, which intensify vascular
   hyperpermeability and plasma leakage ([289]Figure 7G). The experimental
   design adhered strictly to control principles, while data analysis
   incorporated multiple corrections to ensure the scientific rigour of
   the findings. The strong correlation between the model system and
   clinical observations underscores the reliability of hTim4 mice in
   mimicking the immune response to DENV infection, providing a solid
   theoretical basis for future mechanistic studies.

Materials and methods

Mice, ethics and serum samples

   Wild-type C57BL/6J mice and hTim4 genetically modified mice were bred
   by the Institute for Laboratory Animal Resources, NIFDC, and maintained
   in a SPF facility. These mice, aged between 3–11 weeks, were randomly
   assigned to experimental groups. The age of all mice remained
   consistent throughout each individual experiment. Virus inoculation and
   dissection procedures were performed in an ABSL-2 facility under
   anesthesia induced and maintained with Zoletil 50 and xylazine. All
   studies received approval from the Institutional Animal Care and Use
   Committee at the Institute for Laboratory Animal Resources, NIFDC
   (2019-B-16, 2024(B)027). The serum assay of dengue patients was
   reviewed by the Institutional Ethics Committee, Yunan Institute of
   Parasitic Diseases (No.03 - “Technical Cooperation Research on
   Prevention and Control of Emerging Vector-borne Infectious Diseases and
   Rapid Response Treatment in Belt Road Area”).

Establishment of htim4 genetic modified mice

   The human Tim4 (hTim4) gene was precisely inserted into the Tol2
   vector. Briefly, a CAG promoter and hTim4 gene cDNA
   ([290]NM_001146726), followed by an EGFP sequence, were synthesized by
   Sino GenoMax and cloned into Tol2 vector (donated by BIOCYTOGEN). Tol2
   transposase mRNA was transcribed in vitro from the plasmid pT7-Tol2
   using mMESSAGE mMACHINETM T7 Transcription Kit (Ambion, AM1344), then
   add A using Poly(A) Tailing Kit (Ambion, AM1350) and purified by RNeasy
   Mini Kit (QIAGEN, 74104) according to manufacturer’s instructions.
   Subsequently, both Tol2 transposon DNA (30 ng) and mRNA (50 ng) were
   micro-injected together into C57BL/6J mouse zygotes which were further
   transplanted into pseudo-pregnant mice to generate founder mice (Table
   S1). Genotyping of offspring was conducted to screen hTim4 gene
   positive founder mice (Table S2).

Fluorescence detection of htim4 mice in vivo

   The neonatal mice aged 3–5 days, genotyped as hTim4 positive, were
   utilized for in vivo fluorescence detection using the IVIS Spectrum CT
   imaging system. The expression of hTim4 was quantified by measuring
   EGFP immunofluorescence through radiant efficiency analysis.

RNA-Seq

   The total RNAs from lung tissues were isolated using the RNeasy Plus
   Universal Mini Kit (Qiagen, 73404) and assessed for quality
   (OD260/280 = 1.8∼2.2, OD260/230 ≥ 2.0, RIN ≥ 8, 28S:18S ≥ 1.0, > 10
   μg). Subsequently, they were subjected to treatment with the TruSeqTM
   stranded total RNA Kit (Illumina, 20020596) for transcriptome strand
   library preparation. Initially, ribosomal RNA was depleted using the
   Ribo-Zero Magnetic kit (Illumina, 20040526), followed by fragmentation
   of the remaining RNAs. First-stranded cDNAs and double-stranded cDNAs
   were then synthesized and purified accordingly. Each blunt fragment of
   the resulting double-stranded cDNAs had a single “A” nucleotide added
   at its 3′ end as a handle for ligation with multiple indexing adapters
   at both ends of the ds cDNAs. Target fragments ranging from 300 to
   420 bp in size were selected and amplified before quantification using
   a TBS380 mini fluorometer (Tubner Biosystem). Finally, the constructed
   RNA-seq library was sequenced on an Illumina HiSeq xten//NovaSeq6000
   platform by Novagen (TianJin, China). All procedures strictly adhered
   to the instructions provided by Illumina.

Identification of differential expressed genes

   The differential expression of genes (DEGs) was analysed between
   wild-type (WT) mice (used as a reference group) and hTim4 C57BL/6J
   mice. Subsequently, the clean reads were aligned to the murine
   reference genome using HIASAT software in orientation mode. The TPM
   method was employed to calculate the expression level of each
   transcript, while RSEM was used for gene abundance quantification.
   DESeq2/DEGseq/EdgeR with adjusted P-values were utilized to determine
   whether a gene exhibited differential expression. Genes with an
   adjusted P-value ≤ 0.05 (as determined by DESeq2 or EdgeR), along with
   fold changes > 1 or < −1 for differentially expressed mRNAs (DEmRNAs),
   were considered as exhibiting significant differential expression
   between lung tissues of WT and hTim4 mice.

Reactome functional enrichment analysis

   To elucidate the functional profile of differentially expressed genes
   (DEGs), we conducted Reactome pathway enrichment analysis for gene
   lists. The top five pathways enriched with DEGs were selected for
   further investigation. Violin diagrams were employed to present the top
   ten genes in each pathway.

Determination of tissue distribution of htim4 mRNA expression

   The hTim4 mice were anesthetized using Zoletil 50 plus xylazine.
   Subsequently, the heart, liver, spleen, lung, kidney, intestine, brain
   and reproductive organs were harvested and weighed before being
   homogenized through bead dissociation with a tissue grinder (Qiagen).
   Total RNA levels were extracted using canonical precipitation with
   trizol, chloroform, iso-propanol and quantified via quantitative
   reverse transcriptase PCR (qRT-PCR) following the instructions provided
   in the RT reagent Kit (Takara, RR047) and qPCR kit (Takara, RR820).
   Validation of dengue RNA was performed by qRT-PCR utilizing DENV-2
   primers (Table S2).

Flow cytometry

   Blood samples were collected from the inner canthus vein of mice in
   tubes containing 1% EDTA (0.5M/L). T, B, and NK cells in the peripheral
   blood were labelled using corresponding antibodies (BioLegend),
   including anti-Mouse CD45 antibody, anti-Mouse CD19 antibody, and
   anti-Mouse CD49b antibody. The BD LSRII flow cytometer and FACS Diva
   6.0 Software were employed for detection and quantification.

Blood routine examination

   Blood samples were collected from the inner canthus vein of mice using
   EDTA-containing tubes, and blood routine examinations were performed
   with a Hematology analyer (Mindary, BC-5000-VET) according to the
   manufacturer's instructions.

Cells and viruses

   Vero cells were cultivated at 37°C in the minimum essential medium
   (MEM) complemented with 5% fetal bovine serum (FBS). Aedes albopictus
   C6/36 cells were grown at 28°C in RPMI 1640 medium enriched with 10%
   FBS. The DENV-2 (Tr1751 strain) utilized for the challenge model was
   offered by Professor An Jing from Capital Medical University. The virus
   propagated in C6/36 cells and was stored in a −80°C freezer. The titer
   of the virus (10^4 PFU/ml)was ascertained by plaque assays on Vero
   cells.

Absolute quantification of dengue genome copies

   In our study, absolute RT-qPCR was conducted to quantify the viral
   load. Initially, a standardized plasmid containing a chemically
   synthesized 156 bp fragment from protein E of Tr.1751 genomic DNA
   oligonucleotide inserted into the pET-32a+ plasmid vector was
   generated. The standard quality control sample was diluted in a 10-fold
   gradient, and after DNA extraction, real-time qPCR was performed.
   Subsequently, the concentration of viral genome copies was calculated
   using the formula (genomic copies/μl = (6.02 × 10^23) × (ng/μl × 10^−9)
   ÷ (RNA/DNA length × 340/660)). “/”represents “either”. A standard curve
   based on the copy number concentration of each sample and its
   corresponding CT value was plotted, yielding a linear equation:
   Y = −3.141 lgX + 36.077, R^2 = 0.9971. Prior to RT-qPCR detection, the
   test sample underwent treatment with RNase A at a concentration of 100
   μg/mL to eliminate naked viral RNAs. The CT value obtained for the test
   sample was then used in equation Y to determine the copy number
   concentration and subsequently assess the viral load.

Prothrombin and activated partial thromboplastin time test

   The blood of anesthetized mice was collected in tubes containing 1/10
   sodium citrate (10^9 mM). Anticoagulation was achieved by centrifuging
   at 3000 rpm for 10∼15 min, followed by removal of platelets and
   collection of the resulting supernatant. Throughout the experiment, all
   reagents were halved in quantity. The assays were performed according
   to the protocols provided with the prothrombin (PT) test kit (SolarBio,
   BC8081) and activated partial thromboplastin time (APTT) test kit
   (SolarBio, BC8080).

Primary liver and spleen cells culture

   Primary hepatocytes were isolated using the two-step collagenase
   perfusion technique. Briefly, mice aged 6–13 weeks were anesthetized,
   and a combination of D-HBSS medium and collagenase IV enzyme was
   introduced into the circulatory system via the portal vein to perfuse
   the liver. The resulting hepatocytes were then purified through
   filtration with a 70 μm filter followed by centrifugation at 50 g for
   2 min. Subsequently, they were cultured in William's medium
   supplemented with FBS, sodium pyruvate, HEPES,
   Insulin-Transferrin-Selenium, Dexamethasone, and Glutamax at a
   temperature of 37°C under a CO[2] concentration of 5%. Primary spleen
   cells were obtained through grinding followed by purification using a
   40 μm filter and centrifugation at 500 g for 5 min. These cells were
   also cultured in William's medium supplemented with FBS at a
   temperature of 37°C under a CO[2] concentration of 5%.

Animal experiment

   The mouse infection experiments with dengue virus were conducted under
   ABSL-2 conditions. hTim4 mice were anesthetized using Zoletil 50 plus
   xylazine. The mice in the experimental group were intracerebrally
   inoculated with DENV-2 (strain Tr.1751) at dose of 50 PFU (corresponds
   to 10^3 TCID[50]/ml 30 μl), 100 PFU (corresponds to 10^4 TCID[50]/ml 30
   μl) and 500 PFU (corresponds to 10^5 TCID[50]/ml 30 μl), while the
   control group received intracerebral inoculation with DMEM. The viral
   volume administered was approximately 30 μL (1.67×10^3 PFU/ml). The
   survival time, weight, and clinical score of the mice were monitored
   for a period of 21–28 days.

Viral load of tissues

   DENV-2-infected hTim4 mice were anesthetized using a combination of
   Zoletil 50 and xylazine. Blood samples (100 μl) were collected via
   orbital blood extraction at 2, 4, 6, and 8 dpi. Subsequently, RNA was
   extracted from the blood using RNA iso-Plus (Takara 9109). The brains
   and intestines were harvested, weighed, and homogenized with bead
   dissociation using a tissue grinder (Thermofisher). The RNAs were
   quantified by qRT-PCR for absolute quantification of dengue genome
   copies.

Evans blue staining

   The quantification of vascular leakage was performed using Evans blue
   staining. Briefly, each mouse received an intravenous injection of 150
   μl of Evans blue dye (0.5% in PBS) at 8–11 dpi. For hTim4 mice,
   anesthesia was induced with Zoletil 50 plus xylazine and perfusion with
   PBS was conducted after a 2-hour interval. Subsequently, the brain,
   lung, and intestine were collected for extraction and dying using
   formamide (F7503, Sigma).

Clotting time/blood drops

   The hTim4 mice were anesthetized using a combination of Zoletil 50 and
   xylazine. Subsequently, the distal portion of their tails was
   amputated, and the subsequent hemostatic response was evaluated by
   quantifying the number of blood droplets and measuring clotting time.

High resolution laser speckle contrast imaging (HR-LSCI)

   After administering abdominal anesthesia, mice were positioned in a
   prone position and secured to the mouse fixator. Following alcohol
   disinfection, a longitudinal incision was made on the scalp to fully
   expose the skull. The periosteum was then separated, while keeping the
   skull moist with normal saline solution. Similarly, after alcohol
   disinfection, a longitudinal incision was made on the abdomen skin to
   fully expose the mesentery and intestinal tissue. Normal saline
   solution was used to maintain moisture during this process. Vessels and
   blood flow in mice were captured using an HR-LSCI laser speckle blood
   flow imaging system (LSI BFI PLUS, LogiScience Technology Co.,
   Ltd，China). Parameters were set according to those specified in the
   HR-LSCI system settings section below. The system utilized a 780 nm
   semiconductor sweep laser as its light source with an exposure time of
   5 min. Reflected light was collected by an sCMOS camera at a recording
   rate of 1 s/frame for 3 min. Perfusion information was converted into
   two-dimensional perfusion images using HR-LSCI imaging algorithm and
   LSI HMC motion correction algorithm, followed by image processing and
   analysis conducted through the built-in LSI BFI PLUS image software.

Hematoxylin–Eosin (H&E) staining and examination

   The brain, lung, and intestine tissues from hTim4 mice and DMEM mice at
   10–18 dpi were fixed in formalin, embedded in paraffin, and
   subsequently transferred to 70% ethanol for preservation. Each tissue
   sample was placed in processing cassettes, dehydrated through a
   sequential alcohol gradient, and finally embedded in paraffin wax
   blocks. Prior to immunostaining, 3 μm-thick tissue sections were
   dewaxed using xylene, rehydrated with decreasing concentrations of
   ethanol, and washed with PBS. Subsequently, the sections were stained
   using an H&E staining kit (Leagene; DH0003). The stained sections were
   scanned using a NanoZoomer scanner (HAMAMATSU; C13210-01) and previewed
   utilizing the Nano Zoomer Digital Pathology software (NDP.viewer).

Immunohistochemistry examination

   The hTim4 mice and DMEM mice were selected to assess the expression of
   specific markers. The intestinal tissue paraffin sections were dewaxed
   using xylene and a serial alcohol gradient, followed by incubation in
   3% H[2]O[2] for 15 min. Antigen retrieval was performed by boiling the
   sections in citrate buffer (pH 6.0). Subsequently, the sections were
   blocked with fetal bovine serum, and primary antibodies including CD31
   (1:2000, rabbit, abcam, ab182981) were incubated overnight at 4 °C.
   After washing with PBS, secondary antibodies (1:250, servicebio,
   GB23303) were incubated for 1 h at 37 °C before sealing with glycerin.
   Panoramic scanning of the sections was conducted using a 3DHISTECH
   scanner (P250, FLASH). Capillary and microvessel density and diameter
   measurements were obtained from electron microscope images.

Confocal assays

   The intestinal tissue paraffin section was dewaxed using xylene and a
   serial alcohol gradient, followed by incubation in 3% H[2]O[2] for
   15 min. Antigen retrieval was performed by boiling with citrate buffer
   (pH 6.0). After blocking with fetal bovine serum, the following primary
   antibodies were incubated overnight at 4 °C (Table S5): DENV-2 NS1
   (ab41616, abcam), CD31 (ab182981, abcam), villin (ab97512, abcam),
   murine albumin (ALB, SAB3500217, sigma) and VE-cadherin (ab205336,
   abcam). The secondary antibodies Donkey Anti-Rabbit IgG H&L-Alexa Fluor
   647 (ab150075) and Goat Anti-Chicken IgY H&L-Alexa Fluor 555 (ab150170,
   abcam), Goat Anti-Mouse IgG H&L-Alexa Fluor 568 (ab175473, abcam) and
   Goat Anti-Mouse IgG H&L-Alexa Fluor 594 (ab150160, AbCam) were then
   incubated for one hour at 37 °C before examination under a confocal
   fluorescent microscope (Zeiss LSM900).

Single-Cell RNA sequencing (ScRNA-seq) and data preprocessing

   The fresh intestines of hTim4 and DMEM mice were applied for ScRNA-seq.
   The ScRNA-seq and data preprocessing were conducted by Beijing Capital
   Biotechnology Co., Ltd. The single-cell 3′ Library and Gel Bead Kit
   V3.1 provided by 10x Genomics (10x Genomics, 1000121) were used, along
   with the Chromium single-cell G-Code Kit (10x Genomics, 1000120), to
   process a cell suspension containing 300–600 live cells/mL, cell
   viability >85%, cell diameter: 5–40μm, cell clumping rate <20% and
   nuclear rate >70% as determined by Count Star. The suspension was
   loaded onto the 10x Genomics Chromium single-cell controller following
   the manufacturer's protocol in order to generate single-cell bead
   suspensions. Specifically, individual cells were suspended in PBS with
   0.04% BSA, with approximately 11,000 cells being added to each channel
   with the goal of about 10,000 active cells. These active cells were
   captured and lysed to release RNA which was then barcoded and reverse
   transcribed in each GEM. The reverse transcription process was carried
   out in the Bio-Rad S1000TM Touch thermal cycler at 53°C for 45 min
   followed by incubation at 85°C for an additional five minutes before
   storage at 4°C. Subsequently, the generated cDNA underwent
   amplification and quality assessment using the Qubit3.0 Fluorometer and
   Agilent 2100 Bioanalyzer Tape Statio system (Agilent Technologies). The
   ScRNA-seq library was prepared using the Single Cell 3′ Library and Gel
   Beads Kit V3.1, and subjected to sequencing on the Illumina Novaseq
   6000 platform with paired-end (PE) reads of 150 bp in length. The
   sequencing depth reached a minimum of 50,000 reads per cell. For the
   quality-controlled data by FastQC (v0.11.2), we utilized the Cell
   Ranger software's (V6.0) cellranger count module for alignment,
   filtering, barcode counting, and UMI counting to generate feature
   barcode matrices and identify clusters. Furthermore, cells with fewer
   than 200 genes or lower than 99 UMI, as well as those with
   mitochondrial gene proportions exceeding 25%, are filtered using Seurat
   (v4.0.0). Principal component analysis (PCA) is then performed for
   dimensionality reduction, followed by visualization using TSNE and
   UMAP. Next, the Single R uses reference transcriptome data for pure
   cell types to infer the origin of each individual cell, thereby
   performing unbiased cell type identification on single cell RNA
   sequencing data. For humans, use Blueprint_Encode or HPCA was used, and
   for mice, ImmGen or Mouse.RNAseq. The cell types annotated by Single R
   ([291]https://bioconductor.org/packages/devel/bioc/html/SingleR.html)
   are manually corrected using literature-reported cell-specific Marker
   genes. Differential genes (DEGs) are calculated for each cell type, and
   then enrichment analysis is performed using GO (Gene Ontology,
   [292]http://geneontology.org/), KEGG (Kyoto Encyclopedia of Genes and
   Genomes), and Reactome ([293]http://www.reactome.org/) databases, and
   the significance level of pathways is calculated using Fisher's test,
   with the goal of identifying gene sets that are enriched in the cell
   type of interest. The results were visualized using R package.

Spatial sequencing

   Briefly, the intestines of hTim4 mice were embedded using SAKURA (4583)
   and subsequently frozen at −80 °C. The embedded tissue blocks underwent
   RNA integrity assessment by Agilent 2100, ensuring a RIN value greater
   than 7. For construction of the sequencing library, the Visium Spatial
   Gene Expression Slide & Reagent kit from 10 × Genomics (PN-1000184) was
   employed. Each tissue slide was meticulously positioned on a Capture
   Area within a Visium gene expression slide, followed by staining with
   Hematoxylin and Eosin (H&E), imaging, and permeabilization. During
   permeabilization, mRNA molecules were released and specifically bound
   to oligonucleotides that had been pre-coated on the Capture Area.
   Subsequently, reverse transcription was performed and sequencing
   libraries were prepared according to the manufacturer's protocol. The
   resulting raw FASTQ files and pre-existing histology images were
   processed using Space Ranger (v1.2.0, 10 × Genomics) with default
   parameters. The filtered gene-spot matrix and fiducial-aligned
   low-resolution image were utilized for downstream data analyses. Gene
   expression normalization, dimensionality reduction, spot clustering,
   and differential expression analysis were conducted using the Seurat
   package. Spots with less than 100 detected genes were excluded from
   further analysis. Normalization across spots was achieved using the
   built-in SCTransform tool, followed by selection of 3000 highly
   variable genes for principal component analysis (PCA). Spot clustering
   involved building a graph based on the first 20 principal components
   (PCs), which was then segmented at a resolution of 0.5 units.
   Differential gene expression analysis for each cluster was performed
   via Find All Markers function utilizing non-parametric Wilcoxon rank
   sum test criteria. Genes exhibiting fold change >2 and adjusted P-value
   <0.05 were considered significantly differentially expressed genes
   (DEGs). Enrichment testing for candidate gene sets employed the cluster
   Profiler R package based on hypergeometric distribution principles.
   Pathways with corrected P-values below 0.05 were deemed significantly
   enriched, while reactome enrichment analysis was employed to profile
   DEG functions.

Cytokine testing and analysis

   The serum was obtained by centrifuging the collected blood samples from
   hTim4 mice and DENV-2 infected patients at 10,000 rpm. Cytokine
   concentrations in the serum were quantified as pg/mL using the MSD
   (Meso Scale Discovery) method with the Bio-Plex 200 system and Bio-Plex
   Pro Mouse Cytokine 23-plex Assay (M60009RDPD, Bio-Rad, USA), and
   Bio-Plex Pro Human Cytokine 27-plex Assay (#M500KCAF0Y, Bio-Rad, USA)
   following the instruction manual (10014905, Bio-Rad, CA, USA).

Statistical analysis

   The statistical analyses were conducted using Prism V.8.0 and V.12.0
   (GraphPad, La Jolla, California, USA). All data are presented as means
   ± standard deviation (SD). Statistical significance between two groups
   was determined using two side unpaired Student's t-test, One-Way ANOVA,
   and Two-Way ANOVA. Significance levels were denoted as follows: *P <
   0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; ns: not statistically
   significant). Most experiments were performed at least three times to
   ensure reproducibility of the key findings.

Author contributions

   Conceptualization, F.C.F. and W.Y.C.; Writing – Original Draft, F.C.F.,
   W.Y.Y. and Y.Y.S.; All experiments, Y.Y.S., W.Y.Y, Y.W., C.H.H. and
   X.R; Mouse modeling construction, L.S.S, Z.H.Y., G.W.D, W.X. and C.Y.;
   Mouse modeling experiments, Y.Y.S., X.R. and W.Y.Y.; Viral experiments,
   Y.W., L.S.N. W.Y.N. and L.C.; Functional experiments, Y.Y.S. and W.Y.Y;
   Pathological diagnose, Q.Z., Y.Y.W, L.Z. and W.S.L; Sequencing
   analysis, C.H.H.;Other supports, H.W.J., F.R., and Z.H.N.; Statistical
   analysis, Y.Y.S.; Writing-Review & Editing, W.Y.Y, Y.Y.S., F.C.F. and
   W.Y.C.; Resources, A.J. and W.W; Supervision, F.C.F. and W.Y.C. All
   authors have read and approved the article.

Supplementary Material

   Revised_supplemental_Table_and_Figure-clean.doc
   [294]TEMI_A_2531178_SM7489.doc^ (81.2MB, doc)

Acknowledgements