Longitudinal proteomic profiling of the inflammatory response in dengue patients

Background The immunopathogenesis of dengue virus (DENV) infection remains incompletely understood. To increase our understanding of inflammatory response in non-severe dengue, we assessed longitudinal changes in the inflammatory proteome in patients with an acute DENV infection. Methods Using a multiplex proximity extension assay (PEA), we measured relative levels of 368 inflammatory markers in plasma samples from hospitalized patients with non-severe DENV infection in the acute (n = 43) and convalescence (n = 35) phase of the infection and samples of healthy controls (n = 10). Results We identified 203 upregulated and 39 downregulated proteins in acute versus convalescent plasma samples. The upregulated proteins had a strong representation of interferon (IFN) and IFN-inducible effector proteins, cytokines (e.g. IL-10, IL-33) and cytokine receptors, chemokines, pro-apoptotic proteins (e.g. granzymes) and endothelial markers. A number of differentially expressed proteins (DEPs) have not been reported in previous studies. Functional network analysis highlighted a central role for IFNγ, IL-10, IL-33 and chemokines. We identified different novel associations between inflammatory proteins and circulating concentrations of the endothelial glycocalyx disruption surrogate marker syndecan-1. Conclusion: This unbiased proteome analysis provides a comprehensive insight in the inflammatory response in DENV infection and its association with glycocalyx disruption.

Conclusion: This unbiased proteome analysis provides a comprehensive insight in the inflammatory response in DENV infection and its association with glycocalyx disruption.

Author summary
Dengue virus (DENV) infection is the most common arboviral infection worldwide. Significant gaps remain in our understanding of the pathogenesis of dengue, including the changes that occur in the immune and inflammatory responses. In this present study, we used a multiplex proximity extension assay (PEA) from Olink Proteomics AB to assess levels of 368 inflammation-related proteins in plasma samples from the acute and convalescence phase of the infection in hospitalized adult patients with non-severe dengue. We show a pronounced increase in interferon (IFN) and IFN-related antiviral proteins, a selection of cytokines and cytokine receptors, members of the CCL and CXCL chemokine family, pro-apoptotic proteins and proteins involved in endothelial integrity in the acute phase of the illness. In addition, we found a large number of significant associations between these proteins and concentrations of the endothelial glycocalyx degradation marker syndecan-1. In summary, our proteome analysis provides a comprehensive insight in longitudinal changes in inflammation-related proteins in non-severe dengue and the association with endothelial glycocalyx perturbation.

Background
Dengue is one of the most important arboviral infections, causing an estimated 100 million symptomatic infections annually worldwide [1]. The clinical manifestations of dengue infection range from a mild febrile illness to a life-threatening disease [2]. Dengue is classified into dengue with or without warning signs and severe dengue [3]. Patients with warning signs require strict observation and medical intervention. Only a small proportion of patients progress to severe disease, mostly characterized by a transient vascular hyperpermeability syndrome that may lead to fluid accumulation and shock, severe bleeding and organ impairment [3].
The immunopathogenesis of dengue virus (DENV) infection is only partially elucidated [4]. It is generally acknowledged that a dysregulated innate and adaptive immune response is a key factor in the complications of dengue [5]. Earlier studies have highlighted the upregulation of different cytokines, chemokines and lipid mediators in hospitalized dengue patients [6,7]. However, our understanding of the immunological pathways in dengue remains incomplete as most studies measured only a limited number of inflammatory markers.
A comprehensive description of the changes in inflammatory pathways may help to better map the pattern of host defense responses during an acute DENV infection. In this study, we studied the expression of 368 inflammatory proteins using a Proximity Extension Analysis (PEA) technology (Olink Proteomics AB) in plasma of hospitalized patients with DENV infection in the acute and in the convalescent phase and in a small group of healthy volunteers. We also constructed a protein interaction network and explored associations of the differentially expressed proteins (DEPs) with participant characteristics, including signs of plasma leakage. Finally, disruption of the endothelial glycocalyx integrity is increasingly recognized as a key process in plasma leakage and, therefore, we determined associations of DEPs with plasma syndecan-1 concentrations, which is a surrogate marker of endothelial glycocalyx degradation.

Proteomic analysis
Plasma proteins were measured with the Olink Explore 384 Inflammation panel using proximity extension technology (Olink Proteomics AB, Uppsala, Sweden) [9]. This panel includes 368 inflammation-related proteins. To enhance protein detectability and specificity, this assay utilizes the binding of target proteins by paired oligonucleotide antibody probes, followed by hybridization and amplification using a standard real-time PCR. Protein concentration are reported as normalized protein expression values (NPX), which is an arbitrary unit in a Log2 scale that is calculated from normalized Ct values. Validation data of the assay are available on the Olink website (www.olink.com). All samples were measured in the same batch in October 2020. Proteins were excluded from analysis when values were both below the detection limit in more than 25% of all samples and the difference in undetectable values between the acute and convalescence samples was less than 20%.
A functional network analysis was performed using the publicly available STRING database (version 11.5) [10] and visualized using Cytoscape version 3.9.0 [11]. We limited the analysis to the top 50 DEPs between acute and convalescent samples. The minimum required interaction score was set to a medium confidence (0.4). Finally, to identify relevant pathways, we performed a pathway analysis of the 50 DEPs using the gene ontology database.

Clinical and laboratory parameter measurements
The presence of ascites, pleural fluid and gall-bladder wall thickening was assessed daily using handheld ultrasonography (Phillips Lumify, Amsterdam, The Netherlands) with 6 hours fasting prior to the procedure. A complete blood count was performed twice daily using a standardized hematology analyzer in each study center. Circulating albumin and syndecan-1 concentrations were determined as previously described [8].

Statistical analysis
Differential expression analysis was performed using a moderated paired t-test (acute versus convalescent samples (n = 35)) or a linear model (acute (n = 43) versus control (n = 10) samples) of NPX values using Limma R package [12], which employs an empirical Bayes method to moderate the standard errors of the fold change. Age, sex and BMI were statistically insignificant between patients and controls and were therefore not included in the linear model as covariates. Associations between variables were analyzed using Spearman's correlation analysis. The Benjamini-Hochberg multiple testing correction was performed in differential expression and correlational analyses with an FDR p-value < 0.05 considered to be statistically significant. All analyses and visualizations were performed using RStudio version 1.3 for Mac (RStudio, Boston, MA, USA). To further validate our findings, we intersected the DEPs identified in our study with differentially expressed transcripts (DETs) reported in earlier studies that compared gene profiles between acute versus convalescent dengue patients or controls or between patients with severe versus non-severe illness [13][14][15][16][17][18][19][20][21][22][23][24].

Clinical Characteristics
Clinical characteristics of the patients and controls were summarized in Table 1. Plasma samples from 43 dengue patients were included in the analysis; for 35 patients, paired convalescent samples were available. In addition, plasma samples of 10 adult healthy Indonesian volunteers were used. The dengue patients were enrolled at a median of 5 days (IQR, 4-5) after symptom onset; 25 (58.1%) patients were classified as probable primary dengue infection and 18 (41.9%) as probable secondary infection. At enrollment, 26 (62%) patients had a thickened gallbladder wall (>3mm) and 15 (35.7%) patients had ascites and/or pleural fluid detected by ultrasonography. None of the participants fulfilled the World Health Organization (WHO) 2009 criteria for severe dengue and none died during follow-up [3]. In addition, using the WHO 2011 criteria [25], 25 (58%) of the dengue patients were classified as having dengue fever (DF) and 18 (42%) as dengue hemorrhagic fever (DHF) at study enrollment. At hospital discharge, 17 (40%) were classified as having had DF and 26 (60%) as DHF.

Architecture of the plasma inflammatory proteome
Principal component analysis (PCA) of the inflammatory proteins (n = 337) revealed a clear separation between the acute dengue samples with convalescent dengue and control samples (all p < 0.0001) (Fig 1A). An unsupervised hierarchical clustering also suggested a similar finding (S2  (Figs 1B and  S3B). Overall, more inflammatory proteins were up-than downregulated and with a higher log fold-change. Ten convalescent samples were from participants allocated to the oseltamivir group, but there were no significant differences in protein concentrations in convalescent plasma from those allocated to the oseltamivir or placebo group (S6 Fig).
[ Cathepsin C; Activator of serine proteases in immune cells. [54] (Continued ) hepatocyte growth factor (HGF) and colony stimulating factor 1 (CSF1). Endothelial dysfunction is a key factor in dengue pathology and different endothelial markers were upregulated, including endothelial cell-specific molecule 1 (ESM1; also known as endocan), cysteine rich transmembrane BMP regulator 1 (CRIM1), which is involved in endothelial maintenance and integrity [29], and vascular endothelial growth factor D (VEGFD). Finally, different proteases that are known to play a key role in the antiviral immune response were upregulated. These  were corticotropin-releasing hormone binding protein (CRHB), which modulates corticotropin releasing factor (CRH), the neuropeptide and natural killer cell modulator galanin (GAL) [35] and endothelial C-type natriuretic peptide precursor (NPPC), which is important in fluid homeostasis [36].

Functional network analysis of the inflammatory proteome
To further increase our insight in the functional interactions between the top DEPs, we constructed a protein-protein association network, using the STRING database [10]. We limited the analysis to the top 50 DEPs with a cutoff Log2 fold-change of > 1.5 for upregulated proteins (n = 42) and >1 for downregulated proteins (n = 8). This analysis yielded an enriched network with IFNγ as a central protein, along with other cytokines (IL10 and IL33) and members of the CXC chemokine ligand (CXCL8/10) and CC chemokine ligand (CCL3/7) family (Fig 3). Next, using the gene ontology biological processes term, we highlighted several important pathways involved in our protein-protein interaction network including response to IFNγ, regulation of viral life cycle, regulation of cell death and chemokine-mediated signaling pathway.

Associations of the inflammatory proteins with clinical and laboratory profiles
Next, to understand the possible relationship between the inflammatory proteins and clinical outcomes, we assessed the associations of the DEPs with markers of dengue severity. Overall, there were no differences in levels of inflammation-related proteins between patients classified as DF and those with DHF (S4A and S4B Fig). Age, sex and BMI are associated with a higher risk for dengue complications [2,57,58]. In our cohort, especially age was associated with different inflammatory markers including TNF, IL1RN, chemokines and urokinase receptor (PLAUR) (Fig 4). Participants classified as having a probable secondary dengue infection had higher concentrations of different molecules involved in the host immune response, including lymphocyte function (CEACAM21, FOXO1, MZB1, PARP1, PPP1R9B, PSIP1, SH2D1A), glucocorticoid hormone regulation (CRHBP) and megakaryocyte maturation (MPIG6B). In contrast, no differences were observed in concentrations of pro-inflammatory cytokines, interferon-inducible effector proteins or granzymes. Overall, there were only few significant associations between DEPs and the following markers of dengue severity: platelet number, hematocrit, plasma albumin, gallbladder wall thickness and the presence of ascites or pleural fluid. Participants with ascites and/or pleural effusion during hospitalization had a trend for a higher ESM1 value (S5A Fig). In contrast, plasma concentrations of syndecan-1 (SDC1), a surrogate marker of endothelial glycocalyx degradation [59], correlated positively with a large number of inflammatory proteins (Fig 4) and fluid accumulation during hospitalization (S5B Fig). The strongest associations were observed for HGF (r s = 0.74; FDR 1.18x10 -6 ) and IL18R1 (r s = 0.71; FDR 6.59x10 -6 ). The heparanase/SDC1 axis is known to regulate HGF expression and function [60,61]. SDC1 concentrations were also positively correlated with a number of other endothelium-related proteins, including glycocalyx-associated proteoglycans (agrin [AGRN] and ESM1), CRIM1, and osteoprotegerin/TNFRSF11B). Leukocyte number was negatively associated with a number of inflammatory proteins, including cytokines (IL10, IL15, IFNG, LTA, IL33, CSF1), IFN-stimulated genes (TRIM21), immune receptors (LIFR, IL18R1, IL22RA1, SIGLEC1) and chemokines (CCL4, CXCL8). Finally, plasma concentrations of the Depicted is the protein-protein Interaction (PPI) network from the STRING database. The analysis was limited to the top 50 DEPs. The pink (upregulated) and blue (downregulated) nodes denote the proteins while the lines denote the relationships. PPI enrichment p-value was < 1x10 -16 , indicating that the network has significantly more interactions than expected and that the proteins likely engage in biologically relevant interactions. Additional nodes of relevant pathways (red nodes) derived from gene ontology biological process term were included in the network.

Discussion
Our unbiased proteomics analysis in plasma samples from DENV infection patients demonstrates activation of important components of the antiviral immune responses in acute DENV infection and a resolution thereof in the convalescent phase. A total of 203 of 337 proteins were upregulated with a strong representation of IFN and IFN-related antiviral proteins (e.g. GBP2, TRIM21 and SAMD9L), cytokines and cytokine receptors (e.g. IL10, IL33, IL1RN and IL18R1), members of the CCL and CXCL chemokine family (e.g. CXCL10, CCL7, CCL3, CXCL8, CCL25 and CXCL9), pro-apoptotic proteins (e.g. GZMA, GZMB, CTSC, TPP1) and proteins involved in endothelial integrity (e.g. ESM1, CRIM1 and VEGFD). Disruption of the endothelial glycocalyx is increasingly recognized to play a central role in the vascular complications of DENV infection [62]. The pathways that are involved in this disruption are incompletely understood. Our study and validation analysis revealed a large number of associations between the glycocalyx disruption marker SDC1 with inflammatory and endothelial proteins. Our results therefore suggest that the inflammatory environment contributes to glycocalyx dysfunction in DENV infection.
IFN-mediated innate immune responses are a key process in the first line host defense against viral infections. Our study shows the upregulation of different IFN-stimulated genes (ISGs) that are likely important in orchestrating the host response against DENV replication. Guanylate-binding protein-2 (GBP2) is a member of the IFN-inducible guanosine triphosphatase (GTPase) family. GBP2 was shown to exert broad antiviral activity, including against Zika virus [63]. GBP2 was also part of a 20-gene set that was strongly associated with the progression to severe dengue [42]. In addition, the cytosolic ubiquitin ligase and antibody receptor TRIM21 was among the top upregulated DEPs and validated proteins. TRIM21 rapidly neutralizes antibody-bound viruses [64,65], suggesting that it may play a role in neutralization and possibly antibody-dependent enhancement of DENV. In addition, TRIM21 is a negative regulator of proinflammatory cytokines and type I IFN production [66]. Our findings are consistent with a recent study that showed that the expression of TRIM21 was upregulated in helper innate lymphoid cells of patients with dengue hemorrhagic fever [38].
Our findings further reinforce the central roles for IL-10, CXCL10 and IL-33 in the host response to DENV infection. IL-10 and CXCL10 were previously shown to be markers of dengue severity and plasma leakage [39,40,67]. IL-33 is a member of the IL-1 family and, unlike most other cytokines, is normally located in the nucleus. It is released by various cells, including fibroblasts and epithelial, endothelial and immune cells as an alarmin following cell injury. IL-33 has been implicated in allergy, asthma and parasitic diseases, but also increasingly in chronic inflammatory conditions such as arthritis [68]. In a mouse model of DENV infection, IL-33 production was increased and shown to exert a disease-exacerbating role [43]. These findings, as well as the findings reported here, are in seeming contrast to previous studies that reported similar or even decreased plasma concentrations of IL-33 concentrations in patients with DENV infection compared to controls [69,70], which is most-likely due to sensitivity of the currently available assays [71].
Another important finding is the upregulation of different proteases, including LAP3 and cathepsin C (CTSC). The latter is a lysosomal cysteine protease that activates cytotoxic lymphocyte-derived serine proteases, including granzyme A and B, and as such plays a role in the clearance of virus-infected cells [72]. In fact, granzyme B was the most strongly upregulated protein in our study. Granzyme B is stored in granules of cytotoxic T-cells and Natural Killer cells, both known to kill virally infected cells [73]. Cathepsins are also among the potent activators of heparan-degrading enzymes (e.g. heparanase and sialidase) and therefore may contribute to degradation of endothelial glycocalyx in dengue [54].
A transient vascular hyperpermeability syndrome is a key feature in dengue pathophysiology. The targeted proteomics panel also included different endothelial proteins. One of the most upregulated proteins in our study, ESM1 or endocan, was recently shown to be induced by DENV non-structural protein-1 (NS1) and suggested to be a promising biomarker for endothelial activation [41]. In addition, the upregulation of agrin proteoglycan and CRIM1 [74,75] and the downregulation of CDON are counterregulatory mechanisms that limit plasma leakage [34]. TNFSF11/RANKL was the most pronounced downregulated protein, whereas its physiological inhibitor osteoprotegerin (TNFRSF11B/OPG) was upregulated, as previously shown by our group [49]. Besides bone homeostasis, the RANK-RANKL-OPG axis is increasingly recognized to play a key role in both the adaptive and innate immune system [76]. A number of inflammation-related proteins were associated with the endothelial glycocalyx degradation marker SDC1. However, the lack of associations between inflammatory proteins and clinical signs of dengue disease severity (e.g. HCT, fluid accumulation) may suggest that the differentially expressed proteins reflect their role in infection control rather than severity.
A specific strength of our study is the high number of inflammatory proteins in the targeted proteomics platform, which allows a comprehensive analysis of the host immune response in DENV infection and discovery of new biomarkers. The proximity extension assay, which combines both antibody-binding and next-generation sequencing, is considered to be more sensitive and specific in comparison to mass-spectrometry proteomics analysis or other immunoassays such as ELISA [77]. In addition, we analyzed longitudinal samples of the participants, which limits interindividual variation, and participants were well characterized, including daily evaluation of fluid accumulation by ultrasonography. Limitations of our study are the fact that the number of participants was limited, and the participants were relatively homogenous in terms of severity of illness. This precludes the evaluation of the performance of the proteins identified here in the prediction of severe illness. Second, high viral loads have been reported to be associated with disease severity in some studies, but viral loads were unavailable in our study. Third, ten convalescence samples were from participants allocated to the oseltamivir group. Even though we cannot rule out potential secondary effects of the drug, there were at least two weeks between the last dose of oseltamivir and collection of convalescent plasma. In addition, DEP analysis did not reveal significant differences between samples from participants allocated to the oseltamivir or placebo group. Finally, despite the large number of target proteins in the panel, it inevitably lacks proteins that may be relevant in DENV infection, including type I interferons.
In summary, this proteomics study provides a comprehensive insight and linkage of the host inflammatory response and endothelial glycocalyx degradation in hospitalized patients with DENV infection. A number of up-and downregulated proteins have not been reported in previous studies, which shows the potential for the unbiased discovery of candidate biomarkers and predictors of severe disease using these novel technologies.