Authors: Ayesa Syenina, Christine Y.L. Tham, Danny J.H. Tng, Valerie S.Y. Chew, Jia Xin Yee, Yan Shan Leong, Noor Zayanah Hamis, Hwee Cheng Tan, Han Yi Joon, Jing Ti Chan, Yuchen Yang, Yin Bun Cheung, Eugenia Z. Ong, Jenny G. Low, Eng Eong Ooi
Categories: Articles, Orthoflavivirus, Obesity, Inflammation, Host-response
Source: eBioMedicine
Authors: Ayesa Syenina, Christine Y.L. Tham, Danny J.H. Tng, Valerie S.Y. Chew, Jia Xin Yee, Yan Shan Leong, Noor Zayanah Hamis, Hwee Cheng Tan, Han Yi Joon, Jing Ti Chan, Yuchen Yang, Yin Bun Cheung, Eugenia Z. Ong, Jenny G. Low, Eng Eong Ooi
Orthoflaviviruses, as exemplified by the prototypic yellow fever virus (YFV), have been recently identified by the World Health Organisation as major pandemic threats. Although factors such as globalisation and climate change have been identified to increase the prevalence of orthoflaviviral diseases, the ongoing obesity pandemic is a neglected but important contributory factor to exacerbating the risk of severe orthoflaviviral diseases. Indeed, how obesity affects the outcome of orthoflaviviral infection remains undefined.
We used the live attenuated yellow fever vaccine (YF17D-204 strain and hereon referred to as YF17D) to simulate an acute orthoflaviviral infection. We enrolled, using the WHO Asian-BMI classification, 34 participants without obesity (18 kg/m^2^≤BMI<25 kg/m^2^) and 35 participants with obesity (BMI≥25 kg/m^2^) to receive YF17D subcutaneously. We subsequently assessed and compared the clinical and molecular outcomes of vaccination between the two groups.
We found increased prevalence of systemic symptoms; myalgia and axillary soreness from lymphadenopathy were more common in participants with obesity compared to those without despite comparable YF17D RNAaemia levels. Moreover, transcriptomic and cytokine profiling revealed BMI-dependent differences in pro-inflammatory states at both pre- and post-infection.
Our findings indicate obesity as a susceptibility factor of symptomatic orthoflaviviral infection and suggest anti-inflammatory approaches to mitigate risks of severe orthoflaviviral diseases.
National Medical Research Council of Singapore.
Research in contextEvidence before this studyA systematic search was performed in PubMed on 3rd September 2025, to identify studies evaluating the impact of obesity on clinical outcome of orthoflaviviral infections. Search terms related to “dengue”, “yellow fever”, “Zika”, “flavivirus”, “clinical” and “obesity” were used. This search produced 59 hits, most of which were descriptive reports on the prevalence of patients with dengue and obesity and retrospective case–control studies that suggested obesity as a risk factor for severe dengue. We found one prospective study, reported in two papers—one describing the trial method^1^ and the other the trial outcome^2^—that tested the efficacy of metformin at a dose of 1–1.5 g/day for antiviral properties in patients with dengue and obesity. No study examined obesity as a susceptibility factor for symptomatic orthoflaviviral infection or explored the pre-infection and pre-symptomatic host response to infection.Added value of this studyIn this study, we prospectively enrolled healthy volunteers with or without obesity based on the WHO Asian-BMI classification and inoculated them with YF17D to simulate, both safely and ethically, an acute orthoflaviviral infection. We found increased rates of systemic symptom manifestation among participants with obesity compared to those without, despite comparable YF17D RNAaemia levels. We also found elevated levels of pro-inflammatory cytokines in participants with obesity compared to those without, both at baseline as well as post-infection even before symptom onset.Implications of all the available evidenceOur findings indicate that without increasing viral RNAaemia, individuals with obesity are at greater risk of symptomatic infection. Heightened pro-inflammatory cytokine expression at pre-infection baseline as well as immediately post-infection could be contributors to the observed increased risk of severe orthoflaviviral diseases in individuals with obesity.
The combination of globalisation, concentration of human population in urban centres and climate change has placed the world more vulnerable to infectious disease pandemics. Indeed, the World Health Organisation recently updated the list of pathogens that pose pandemic threats.^3^ One group of such pandemic threat pathogens is the Orthoflavivirus genus of viruses.^4^ Orthoflaviviruses are transmitted by arthropods such as ticks and mosquitoes, including the highly domesticated Aedes aegypti that is prevalent throughout the tropics^5^^,^^6^ and which transmits dengue, Zika and yellow fever viruses (DENV, ZIKV and YFV respectively).^7^ A. aegypti-transmitted orthoflaviviruses thus afflict millions of people each year with potentially life-threatening diseases^7^; DENV alone causes an estimated 100 million episodes of acute illness each year.^8^ Orthoflaviviruses are thus not just future pandemic threats, they are present public health challenges.
The public health challenge of orthoflaviviruses has converged with another health Obesity.^9^ Although obesity is associated with other metabolic conditions, such as diabetes that can complicate dengue, it is an independent risk factor for severe orthoflaviviral diseases.10, 11, 12 Indeed, patients with obesity and dengue show 2- to 4-fold greater risk of severe dengue than dengue patients without obesity.^11^^,^13, 14, 15, 16 Obesity also impairs placental response to ZIKV infection, potentially contributing to increased risk of vertical ZIKV transmission and congenital Zika syndrome.^17^ Despite the association between obesity and severe orthoflaviviral diseases, the host factors that influence infection outcome in individuals with obesity remain ill defined.
We have used the licenced live attenuated YF17D vaccine as a safe approach to examine acute orthoflaviviral infection. We had found that pre-infection (baseline) adaptive endoplasmic reticulum (ER) stress and reduced tricarboxylic acid (TCA) cycle activity were susceptibility factors for symptomatic infection outcome.^18^ Such baseline states predisposed early innate immune activation that correlated with systemic symptom manifestations.^19^ As obesity is associated with ER stress^20^^,^^21^ and altered immune cell metabolism,^22^^,^^23^ we postulated that individuals with obesity are more susceptible than individuals with normal body mass index (BMI) to symptomatic infection. Moreover, obesity may also modify the host response to infection, which could elevate the risk of severe orthoflaviviral diseases, including dengue.^24^ Indeed, obesity is associated with a heightened innate immune activity at steady state.^25^^,^^26^ It is thus plausible that individuals with obesity respond differently to orthoflaviviral infection,^10^ potentially with heightened pro-inflammatory response that elevates risk of severe disease. Understanding how obesity modifies baseline factors and host response to an acute orthoflaviviral infection could thus contribute to a clearer picture of disease pathogenesis.
In this study, we used YF17D to test the hypothesis that obesity (25 kg/m^2^<BMI≤35 kg/m^2^, as recommended for Asian populations)^27^^,^^28^ increases susceptibility to symptomatic infection. We found that, despite no difference in YF17D RNAaemia, individuals with obesity showed increased rate of systemic symptoms, compared to individuals without obesity. Moreover, transcriptomic profiling and multiplex cytokine measurements revealed obesity associated pro-inflammatory responses to infection, which could potentially be targeted with licenced therapeutics to reduce the risk of severe disease.
This study was approved by the SingHealth Centralised Institutional Review Board (ID: 2021/2730) and the National University of Singapore Institutional Review Board (ID: 2022-130), in accordance with the Declaration of Helsinki. Written informed consent forms were obtained from all participants.
A total of 34 participants without obesity (18 kg/m^2^≤BMI≤25 kg/m^2^) and 35 participants with obesity (25 kg/m^2^<BMI≤35 kg/m^2^) were recruited and consented to participate in the study; classification of participants with and without obesity was based on the WHO Asian-BMI classification.^27^ Inclusion and exclusion criteria for recruitment of participants are listed in Supplementary Table S1. Demographic data, including sex, were self-reported by study participants. Study participants were inoculated with the YF17D subcutaneously over the deltoid muscle. Participants were trained to recognise acute YF17D-associated adverse event (AE) symptoms and were provided with a standardised symptoms diary to record occurrence of any systemic AEs for 2 weeks post vaccination. Plasma samples for RNAaemia measurement were collected on D1, D4, D6, and D14 and for cytokines analysis on D0, D1, D4, D6, D14, and D28. Serum samples were collected on D0 and D28 for neutralising antibody measurements. Whole blood samples were collected for RNA-sequencing analysis at D0, D1, D4, D6, and D28 and for T-cell studies at D0, D14, and D28.
Total RNA was extracted from plasma samples using the MagNA Pure 24 Total NA Isolation Kit (Roche Cat# 07658036001) according to the manufacturer's instructions. To measure YF17D RNAaemia, quantitative qRT-PCR was performed using SuperScript® III One-Step Quantitative RT-PCR kit (Invitrogen Cat# 11732088) according to the manufacturer's protocol. Primers and probes to detect YF17D are as YF17D forward (5′ GAA CAG TGA TCA GGA ACC CTC TCT 3′); YF17D reverse (5′ GGA TGT TTG GTT CAC AGT AAA TGT G 3′); and YF17D probe (5’/HEX/CTA CGT GTC/ZEN/TGG AGC CCG CAG CAA T/3IABkFQ/3′). YF17D RNA standards were generated by first synthesising cDNA from YF17D viral RNA using SuperScript® First-Strand Synthesis System (Invitrogen Cat # 18080051), subsequently amplifying the YF17D amplicon using Q5 Polymerase (New England Biolabs Cat# M0491) according to manufacturer's instructions. Primers to generate YF17D amplicons are as T7 YF17D forward (TAATACGACTCACTATAGACTCTTGGA AGAGACGGCC) and T7 YF17D reverse (TTGGTGCAGCTTCCTTTCTT). Finally, amplicon was in vitro transcribed using MEGAScript™ T7 Transcription Kit (Life Technologies Cat# AMB13345) according to manufacturer's instructions. RNA copy numbers were calculated as previously described.^29^ The limit of quantification (LOQ) of RNAaemia was 12.5 copies/ul; any measurements below the LOQ were considered negative.
Plaque reduction neutralisation test (PRNT) using BHK-21 (ATCC Cat# CCL-10, RRID:CVCL_1915) cells was used to measure the titre of neutralising antibodies against YF17D. Serum samples were serial twofold diluted in RPMI maintenance media (MM) and incubated with 40 plaque forming units (pfu) of YF17D virus, in equal volumes for 1 h. The mixture was then inoculated onto BHK-21 cells and incubated for 1 h at 37 °C, after which the mixture was aspirated and replaced with an overlay consisting of 1% carboxymethyl cellulose in MM. After 5 days at 37 °C, cells were fixed with 20% formaldehyde and stained with 1% crystal violet. PRNT50 values were determined using a sigmoid dose–response curve fit and reported as reciprocal values.
Peripheral blood mononuclear cells (PBMCs) were collected from subjects in EDTA containing tubes, isolated by Ficoll–Paque density gradient centrifugation and cryopreserved until analysis. The frequency of YF17D specific T cells was quantified using PBMCs stimulated with peptides spanning the entire proteome of the YF17D vaccine strain in an IFN-γ ELISPOT assay. ELISPOT plates (Merck Cat# MSIPS4W10) were coated with 5 μg ml−1 human IFN-γ antibody (MABTECH Cat# 3420-2A, RRID:AB_3712871) overnight at 4 °C. A total of 200,000 PBMCs were seeded per well and stimulated for 18 h with the different YF17D peptide pools (Genscript Cat# SC1487) at 1 μg/ml. The plates were then incubated with 2000 human biotinylated IFN-γ detection antibody (MABTECH Cat# 3420-6-250, RRID:AB_907273), followed by streptavidin–alkaline phosphatase (streptavidin-AP) (MABTECH Cat# 3310-10-1000) and developed using the KPL BCIP/NBT phosphatase substrate (SeraCare Cat# 5420-0030). To quantify positive peptide-specific responses, average spots of the unstimulated wells were subtracted from the peptide-stimulated wells, and the results of the different peptides were pooled and expressed as spot-forming cells (SFC) per 10^6^ PBMCs. Results were excluded if negative control wells had more than 30 SFC/10^6^ PBMCs or if positive control wells, PHA (Sigma Cat# L2769), were negative.
Whole blood samples were collected in Tempus blood RNA tubes (Thermo Fisher Scientific Cat# 4342792) at the stipulated time point. RNA isolation was performed according to the manufacturer's protocol (Tempus Spin RNA Isolation Kit, Thermo Fisher Scientific Cat# 4380204). RNA was sent to Azenta Life Sciences for RNA-sequencing using the Illumina Novaseq. Pre-processing of data included removal of adaptor sequences and filtering of low-quality reads (5′ or 3′ end bases that contains N's or of quality values below 20) and contaminations (reads that are less than 75 bp long after trimming) by Cutadapt (v1.9.1). Filtered data were subsequently aligned to reference genome using Hisat2 (v2.2.1)^30^ with default parameters. Differential gene expression analysis was subsequently conducted using DESeq2 package (v1.40.2).^31^ Raw counts were imported into DESeq2 and genes with counts of less than 10 were filtered out prior to normalisation. To identify obesity-associated genes at all observed timepoints, the analysis model used accounted for sex, timepoint, and BMI group, including the interaction between BMI group and timepoint. Wald tests were used for pairwise comparisons between obese vs normal. P-values were adjusted to control the false discovery rate (FDR) at 5% using the Benjamini-Hochberg method. Gene set enrichment analysis (GSEA) was conducted on pre-ranked genes based on log2fold-change between obese vs non-obese participants with the previously established blood transcription modules (BTM)-plus.^32^
Variance stabilising transformation (VST),^31^ a DESeq2 data transformation for count data that stabilise variance across genes, was applied to normalised count data. VST counts were used for visualisation and clustering analysis.
To estimate immune cell populations, immune deconvolution was conducted on VST count data using the Microenvironment Cell Populations (MCP)-counter method^33^ with the R package immunedeconv v.2.1.0.
Plasma samples were collected at D0, D1, D4, D6 and D14 for proteomics analysis with Olink Target 48 Cytokine panel. Briefly, 45 oligonucleotide-labelled antibody probe pairs were hybridised with plasma samples at 4 °C overnight to form 45 unique DNA reporter sequences. Subsequently, the DNA reporter sequences were amplified by PCR before being detected and quantified with the Olink Signature Q100, with the readout being in pg/mL. To identify obesity-associated differentially expressed proteins, linear mixed effects models (LMMs) were fit independently to each protein using the lme4^34^ package in R with protein log2concentration as the dependent variable. The model included main effects of BMI group and timepoint, the interaction of the two terms, and random intercept for subject to account for within-individual repeated measurements. Sex was included as a covariate to adjust for imbalance. Model-based estimated marginal means were computed for each BMI group × timepoint combination using the emmeans package in R. Pairwise contrasts between BMI groups within each timepoint were tested using t-statistics with Satterthwaite-adjusted degrees of freedom. P-values were adjusted to control the false discovery rate (FDR) at 5% using the Benjamini-Hochberg method. Proteins with adjusted p-value <0.05 were considered differentially expressed.
Sample size was determined a priori on the expectation that participants with obesity have at least three times the risk of developing symptomatic outcome from YF17D inoculation compared to participants without obesity. This expectation was based on post hoc analysis of previously published data,^35^ which showed ∼20% of individuals without obesity vs ∼60% of individuals with obesity having symptomatic outcomes. Comparing 32 participants with obesity with 32 participants without obesity would thus provide >80% power to observe a statistically significant difference in the proportion of symptomatic outcomes at 2-sided 5% level. To account for possible dropouts in the study, 70 participants, with a 1 ratio of non-obese to obese, were enrolled into the study. A total of 69 participants–34 without obesity and 35 with obesity–were included in the final analysis. One subject was excluded due to non-compliance with follow up visits.
Linear models (LMs) were fit independently to RNAaemia AUC data with RNAaemia AUC as the dependent variable. The model tested BMI group as the main effect and sex was included as a covariate to adjust for imbalance. Estimated marginal means for each group were calculated using the emmeans package in R. Pairwise group differences were tested using model-based t-contrasts.
LMMs were fit to log10RNAaemia, with BMI group, timepoint, and the interaction of the two terms as the main predictors, as was done for Olink data. Sex was included as a covariate to adjust for imbalance. Model-based estimated marginal means were computed for each BMI group × timepoint combination using the emmeans package in R. Pairwise contrasts between BMI groups within each timepoint were tested using t-statistics with Satterthwaite-adjusted degrees of freedom.
To account for small sample size and separation of data, Firth logistic regression was performed independently on AE symptoms, leucocytosis, and monocytosis data using the logistf package in R. Prevalence of AE, leucocytosis, and monocytosis, respectively, were the dependent variables and BMI group was the main predictor, adjusting for sex as a covariate to adjust for imbalance.
Statistical analysis was performed on RStudio v.2024.04.2. Two-sided Fisher's exact test and Chi-square test were conducted for categorical datasets.
The funders of the study had no role in study outcomes, data analysis, and report writing.
We used the WHO Asian-BMI classification^27^ and enrolled 34 adults without obesity and 35 adults with obesity aged 21–45 years old, with no known pre-existing medical illness following written informed consent. Demographics of these volunteers are shown in Table 1. Of the participants with and without obesity, 10 of 35 (29%) and 8 of 34 (24%), respectively, were dengue IgG seropositive at the time of recruitment; this difference was not statistically significant. There were more males with obesity than females compared to the group without obesity. To address this imbalance, all our analyses were adjusted for this difference in sex.Table 1Participants demographics.Subjects (n = 69)Without obesity (18≤BMI≤25)With obesity (25<BMI≤35)P-valueSubjects n (%)34 (49.28)35 (50.79)Age (in years) Median (range)27 (23–45)29 (21–44)0.26aSex n (%) Male11 (32.35)21 (60)0.02bBMI kg/m^2^ Mean (±SD)21.56 (1.96)28.25 (2.41)<0.001aDENV IgG serostatus at baseline n (%) Positive8 (23.53)10 (28.57)0.63baStudent t-test, two sided.bChi-square, two sided.
All participants received YF17D subcutaneously over the deltoid region of the left or right arm and blood samples were collected at the pre-specified timepoints for analysis of RNAaemia, gene expression, cytokines/chemokines secretion, T cell response, and neutralising antibody response, as summarised in Fig. 1A.Fig. 1Overview of study and lack of difference in YF17D RNAaemia levels in participants with obesity compared to those without obesity. (A) Timeline of YF17D vaccination and sample collection for measurement of RNAaemia, gene expression, cytokines and chemokines, T cell response, and neutralising antibody response. (B) Linear mixed model (LMM) fitting log10RNAaemia, with BMI group, timepoint, and the interaction of the two terms as the main predictors. Model was adjusted for sex. Points represent individual samples. Bars represent model-adjusted means with 95% confidence intervals. Pairwise contrasts between BMI groups within each timepoint were tested using t-statistics with Satterthwaite-adjusted degrees of freedom. n= 34 (participants without obesity); n= 35 (participants with obesity). (C) Linear model (LM) fitting RNAaemia AUC, with BMI group as the main predictor, adjusted for sex. Points represent individual samples. Bars represent model-adjusted means with 95% confidence intervals. Pairwise contrast was tested using model-based t-statistics. n= 34 (participants without obesity); n= 35 (participants with obesity).
Symptoms experienced after YF17D inoculation were recorded daily by the participants using a symptom diary. Symptomatic participants were followed up until resolution of symptoms. As hypothesised, more individuals with obesity (19 of 35, or 54%) reported symptoms after inoculation of YF17D compared to participants without obesity (10 of 34, or 29%) (Table 2). Statistically, obesity was significantly associated with increased odds, adjusted for sex, of reported symptoms (ORadj: 3.1, 95% CIadj: 1.2–9, pvaladj: 0.027) (Table 3). We further classified reported symptoms by system organ class according to the Common Terminology Criteria for Adverse Events (CTCAE) version 4.0.^36^ We found that obesity specifically increased the odds, adjusted for sex, of myalgia (ORadj: 13.7, 95% CIadj: 1.4–1841.6, pvaladj: 0.018) and axillary soreness, with lymphadenopathy on clinical examination (ORadj: 27.9, 95% CIadj: 2.81–3784.51, pvaladj: 0.002) (Table 3). The median time from inoculation to symptom onset for myalgia (classified under musculo-skeletal in Supplementary Figure S1) and axillary soreness (classified under systemic in Supplementary Figure S1) was 5 and 4 days, respectively (Supplementary Figure S1).Table 2Symptomatic outcomes after YF17D inoculation.Symptomatic outcomeParticipants without obesity, n = 34 (18≤BMI≤25) n (%)Participants with obesity, n = 35 (25<BMI≤35) n (%)All symptoms10 (29.41)19 (54.29)Symptoms classified by system organ class Local Bruising1 (2.94)0 (0) Rash0 (0)1 (2.86) Urticaria0 (0)1 (2.86) Systemic Fever3 (8.82)4 (11.43) Axillary Lymphadenopathy0 (0)6 (17.14) Malaise5 (14.71)5 (14.29) Central nervous system Headache5 (14.71)5 (14.29) Musculo-skeletal Neck stiffness1 (2.94)0 (0) Myalgia0 (0)6 (17.14) Respiratory Cough2 (5.88)2 (5.71) Runny nose/nasal congestion4 (11.76)2 (5.71) Sore throat1 (2.94)3 (8.57) Gastrointestinal Abdominal pain0 (0)2 (5.71) Nausea1 (2.94)1 (2.86) Diarrhoea1 (2.94)0 (0) Eye Sore eyes1 (2.94)1 (2.86)Symptomatic outcomes are further classified by system organ class according to CTCAE version 4.0.Table 3Logistic regression analysis of symptomatic outcomes by BMI group.OR (95% CI) unadjOR (95% CI) adjp-value (adj)All symptoms2.8 (1.1–7.5)3.1 (1.1–9)0.027Axillary lymphadenopathy15.2 (1.7–2013.9)27.9 (2.8–3784.5)0.002Myalgia15.2 (1.7–2013.9)13.7 (1.4–1841.6)0.018Odds ratios (OR) and 95% confidence intervals (CI) are shown for the association between BMI group (with vs without obesity) and symptomatic outcomes. Results are presented from both unadjusted (unadj) models and models adjusted (adj) for sex.
Haematological analysis in our study participants found 3 (8.57%) participants with obesity with leucocytosis at baseline, and 7 (20%) and 8 (22.86%) with monocytosis at D4 and D6, respectively, post-infection (Supplementary Figure S2). Monocytosis was also detected in 3 (8.82%) participants without obesity but only at D6 post-infection (Supplementary Figure S2). Statistical analysis found no association between obesity and leucocytosis (Supplementary Table S2). However, obesity significantly increased the odds, adjusted for sex, of monocytosis at D4 post-infection (ORadj: 11.8, 95% CIadj: 1.2–1585, pvaladj: 0.030) (Supplementary Table S2).
Despite increased rate of reported symptoms and monocytosis in obese compared to participants without obesity, YF17D RNAaemia AUC, based on a priori defined sampling timepoints were comparable between the 2 groups (Fig. 1B–C). Additionally, all participants developed neutralising antibody and T cells against YF17D, further supporting the notion that there was no difference in infection rates (Supplementary Figure S3A and B). Adaptive immune responses were not analysed further, given the focus of this study on acute response to YF17D infection. These findings recapitulated our earlier findings that host factors rather than RNAaemia determined susceptibility to symptomatic infection.^18^
We have used the attenuated YF17D virus as a safe and ethical approach to simulate acute infection. The clinical outcome has thus been mild. Nonetheless, obesity has been found to be a risk factor of severe dengue, likely but not exclusively through post-symptomatic dysfunctional NK cell response to DENV infection.^37^ As immune response to infection precedes symptom onset, our controlled YF17D infection study could thus glimpse the pre-symptomatic host response to infection in participant with obesity compared to those without.
For such exploratory analyses, we first examined bulk RNA sequencing on whole blood before and at pre-defined timepoints after YF17D infection. Gene set enrichment analysis (GSEA) on pre-ranked gene lists was conducted using previously established (BTM)-plus^32^ geneset (Fig. 2A and Supplementary Figure S4). Genes were ranked by fold-change (FC) in participants with obesity relative to participants without at each timepoint. Genes associated with cell cycle and cell division of CD4 T cells and those enriched in NK cells were negatively enriched at all timepoints observed (Fig. 2A–C, Supplementary Figure S5A and B), concordant with the notion that dysfunctional NK cell response contributes to severe dengue in obese patients.^37^ In contrast, genes that were enriched in B cells were positively enriched before and after YF17D inoculation (Fig. 2A and D, Supplementary Figure S6A). Interestingly, genes associated with the antiviral interferon (IFN) signatures were positively enriched in participants with obesity at D4 and D6 after YF17D inoculation (Fig. 2A and E, Supplementary Figure S6B).Fig. 2Positive enrichment of B cell associated genes and negative enrichment of CD4+ T cells and NK cells associated genes in participants with obesity compared to those without obesity. Gene set enrichment analysis (GSEA) on pre-ranked gene list was conducted using previously established BTM-plus. Genes were ranked by fold-change in participants with obesity relative to participants without obesity. (A) Top 10 positively and negatively enriched pathways at the observed timepoints. (B–E) GSEA plot for enriched pathways; top panel depicts the running enrichment score for each timepoint with peak of curve representing maximum enrichment score (NES); middle panel shows the position of individual genes in the pathway within the ranked dataset for each timepoint; bottom panel represents the ranked list of all genes. (B) GSEA plot for P051_mitotic cell cycle in stimulated CD4 T cells. (C) GSEA plot for P067_enriched in NK cells (I). (D) GSEA plot for P122_enriched in B cells (I). (E) GSEA plot for P152_antiviral IFN signature.
The GSEA findings could reflect either differential gene expression or abundance of the different subsets of white blood cells in the peripheral blood. To determine the nature of the difference in gene enrichment, we applied the microenvironment cell populations (MCP)-counter method^33^ to estimate the abundance of different immune cells present in the peripheral blood; our clinical haematological data (Supplementary Figure S2) lacked granularity on the different subsets of immune cells. MCP analysis revealed that the increased expression levels of LEGs associated with B cells was likely due to increased abundance of B cells at all timepoints except D6 (Supplementary Figure S7). Likewise, the reduced enrichment of LEGs of NK cells reflects reduced NK cell abundance (Supplementary Figure S7). However, T cell abundance was unchanged, suggesting that the differential enrichment of genes was likely due to reduced expression of these genes in participants with obesity (Supplementary Figure S4). Taken collectively, the gene enrichment findings suggest reduced NK and T cell response to YF17D infection that could have been balanced by increased ISG expression to result in comparable RNAaemia between participants with and without obesity.
To complement the gene expression data, we measured the absolute concentration of 45 cytokines and chemokines using the Olink Target 48 Cytokine panel. Of the 45 proteins, eight proteins showed plasma concentration below the lower limit of quantification (LLOQ) in at least 311 of the total 414 samples (>75%), derived from 6 sampling timepoints from the 69 study participants (Supplementary Figure S8); these eight proteins were excluded from further analysis. We thus compared plasma concentrations of 37 proteins across the period of D0 (pre-infection), D1, D4, D6, D14, and D28 after YF17D inoculation, to identify obesity-associated differentially expressed proteins. We fitted linear mixed models (LMMs) to log2 concentration of each of the 37 proteins, which accounts for the non-independence of longitudinal data. BMI group, timepoint, and interaction of the two terms were the main effects and sex as a covariate. Proteins that were differentially upregulated in participants with obesity in at least one timepoint included angiogenic factors hepatocyte growth factor (HGF) and vascular endothelial growth factor A (VEGFA); the scavenger receptor oxidised low-density lipoprotein receptor 1 (OLR1); oncostatin-M (OSM); chemokines C–C motif chemokine ligand 13 (CCL13), CCL7, CCL19, C-X-C motif chemokine ligand 12 (CXCL12), and CXCL10; and pro-inflammatory cytokines tumour necrosis factor (TNF), interleukin-6 (IL6), TNF superfamily member 10 (TNFSF10), and interferon gamma (IFNγ) (Fig. 3A).Fig. 3Plasma pro-inflammatory proteins are differentially expressed in participants with obesity compared to those without obesity. (A) Linear mixed model (LMM) fitting each Olink protein, with BMI group, timepoint, and the interaction of the two terms as the main predictors. Model was adjusted for sex. Volcano plots of differentially expressed proteins (DEPs) between participants with obesity and participants without obesity by timepoint. Red and blues points are proteins that are significantly up- and down-regulated, respectively, in participants with obesity. All proteins are shown. n= 34 (participants without obesity); n= 35 (participants with obesity). (B) Model-adjusted mean log2concentration of significant DEPs between participants with obesity and participants without obesity across all timepoints observed. Pairwise contrasts between BMI groups within each timepoint were tested using t-statistics with Satterthwaite-adjusted degrees of freedom; p-values were adjusted with the Benjamini-Hochberg method. n= 34 (participants without obesity); n= 35 (participants with obesity). ∗P < 0.05, ∗∗P < 0.01, ∗∗∗P < 0.001.
Longitudinally, these proteins can be broadly divided into two groups. The first group consisted of proteins that were elevated in participants with obesity compared to those without at baseline and which remained relatively unchanged throughout the period of observation (Fig. 3B). These included angiogenic factors HGF, VEGFA and pro-inflammatory cytokines OLR1, and OSM (Fig. 3B). Amongst the chemokines in this group were CCL13 and CXCL12 concentration, which remained higher and lower, respectively, in participants with obesity compared to those without throughout the period of observation (Fig. 3B).
The second group consisted of proteins that showed increased concentration after YF17D infection. These included cytokines that are involved with the pro-inflammatory response, such as IL6, TNF, TNFSF10 (Fig. 3B). Remarkably, IL6 and TNF have been suggested to be a part of the pro-inflammatory response that leads to vascular leakage in severe dengue. Similarly, the concentrations of chemokines CCL3, CCL7, CCL19, and CXCL10 in participants with obesity peaked at D6 after YF17D inoculation at levels significantly higher than the non-obese (Fig. 3B). Increased concentrations of these chemokines could suggest mobilisation and migration of immune cells such as lymphocytes and macrophages, which may explain the haematological changes shown in Supplementary Figure S2. Finally, IFNγ concentrations peaked at D6 but were reduced to lower levels in participants with obesity compared to those without at D28 (Fig. 3B). This trend raises the possibility that the reduced enrichment of genes in T cells in participants with obesity may affect the memory T cell pool.
These findings collectively suggest that, besides being in a pro-inflammatory state, the host response of individuals with obesity to YF17D heighten inflammation.
The convergence of the obesity pandemic with A. aegypti-transmitted orthoflaviviral diseases, such as dengue makes the understanding of how obesity affects the host response to infection a priority. Our findings show that, as hypothesised, participants with obesity displayed increased rates of systemic events, despite no difference in either the rate or level of YF17D RNAaemia, compared to participants without obesity.
Besides increased rates of symptomatic infection, our findings also showed increased abundance of B cell associated genes and reduced abundance of CD4+ T cells and NK cells associated genes. Increased B cells associated genes could be a part of the pro-inflammatory state in individuals with obesity. Previous studies have highlighted that B cells were found to drive adipose tissue inflammation and that B cells in the periphery of individuals with obesity were able to produce more pro-inflammatory cytokines, such as IL-6 and TNF.^38^^,^^39^ The reduced NK cell abundance and reduced T cell related gene expression may, along with the baseline pro-inflammatory state, collectively contribute to increasing the risk of dysfunctional NK and T cell response in patients with severe disease.^37^
Besides post-infection differences, our findings also showed altered immune homoeostasis, or allostasis^40^ in a BMI-dependent manner with differential expression of cytokines at pre-infection baseline. It is known that pro-angiogenic factors are required for formation of new blood vessels to supply oxygen and other nutrients to adipose tissues in obesity.^41^ They are also associated with endothelial cell (EC) dysfunction and reduced vascular density.^42^ Endothelial dysfunction is characteristic of both severe yellow fever^43^ and dengue.^44^ Indeed, previous reports have not only shown that VEGFA increases the permeability of endothelial cells,^45^ they also demonstrated that increased plasma VEGFA concentration was associated with severe dengue.^46^^,^^47^ Besides pro-angiogenic factors, other pro-inflammatory cytokines, such as IL-6 and TNF that showed increased plasma concentration with YF17D infection, are also associated with severe dengue.^48^ Thus, our findings suggest individuals with obesity could potentially be at greater risk of vascular leakage during orthoflaviviral infections.
Elevated plasma concentrations of OLR1 and OSM were also found in participants with obesity compared to those without. OLR1 is involved in several different immune pathways including cytokine production and has been associated with a higher risk of severe COVID-19 and thrombotic complications.^49^ OSM is primarily secreted by activated macrophages, which along with elevated CCL3 and CCL7, chemokines that are secreted by macrophages and attract macrophages, respectively, suggest exaggerated macrophage-driven responses to orthoflaviviral infection.50, 51, 52, 53
The lack of difference in RNAaemia between participants with and without obesity despite increased expression of genes associated with type-I IFN as well as pro-inflammatory cytokines in participants with obesity is curious. In vitro, type-I IFN pre-infection but not post-infection treatment of cells prevented infection.^54^ It is thus possible that the magnitude of increased expression of these genes in the type-I IFN pathway on participants with obesity were insufficient to prevent infection. Alternatively, YF17D RNAaemia levels would have been higher than what was found in this study except for the increased pre-infection baseline type-I IFN activity. Unlike type-I IFN, however, there is no clinical evidence to suggest that inflammation impairs infection. Firstly, treatment with high dose prednisolone, an anti-inflammatory corticosteroid, did not alter DENV infection clearance.^55^ Secondly, patients with severe dengue have slower viral clearance kinetics despite pro-inflammatory responses that include high IL-6 plasma cytokine levels.^37^^,^^56^ Moreover, the pro-inflammatory response may also lead to increased expression of negative feedback response in immune cells, such as increased expression of PD-1 in T cells, that lead to dysfunctional cell-mediated immune response that delays viral clearance.^57^^,^^58^
There are several limitations in our study. Firstly, YF17D is an attenuated virus. The response to YF17D infection that we have observed may under-represent the breadth and magnitude of the host response to wild-type orthoflaviviral infection. Secondly, although we found statistically significant difference in the rate of systemic symptoms, our expectation that ∼20% of individuals without obesity vs ∼60% of individuals with obesity would be symptomatic were overly optimistic. A significantly greater number of different symptoms could yet be experienced by individuals with obesity compared to those without had we been more conservative with our a priori sample size estimation.
Finally, our findings suggest testing the use of anti-inflammatory drugs as a treatment to reduce the risk of severe dengue in patients with obesity. Although a phase II clinical trial had not found high dose corticosteroid to be beneficial in reducing dengue severity, the trial did not specifically target patients with obesity and dengue. Furthermore, the elevated expression of IL-6, even possibly at baseline and significantly following infection also suggest the use of tocilizumab, an IL-6 inhibitor that showed benefit in reducing pulmonary disease severity in patients with COVID-19 could also be tested.^59^^,^^60^
In conclusion, obesity may increase the likelihood of symptomatic infection despite comparable RNAaemia levels in individuals without obesity. The pro-inflammatory baseline state and response to infection may put individuals with obesity at risk of severe systemic orthoflaviviral diseases.
AS, CYLT, and HYJ analysed the data. YBC determined the sample size and oversaw the statistical analyses. AS and EEO wrote the manuscript. CYLT, VSYC, JXY, YSL, NZBH, and HCT performed the experiments. DJHT and JGL recruited study participants. EOZ, AS, VSYC, JTC, and YY oversaw study logistics. AS, EOZ, and EEO have accessed and verified all underlying data. JGL and EEO designed and supervised the study. EEO obtained the funding. All authors have read and approved the final version of the manuscript.
Request for data sharing can be addressed to engeong.ooi@duke-nus.edu.sg. Raw Olink data is available in the Supplementary Materials. Raw and processed bulk RNA-sequencing data have been deposited in ArrayExpress under accession E-MTAB-16701.
EEO has served in advisory capacities for Sanofi Pasteur, Takeda Pharmaceuticals, MSD (Merck), Johnson and Johnson, and Novartis on dengue. JGHL has served in advisory capacities for Takeda Pharmaceuticals, Johnson and Johnson, and Serum Institute of India on dengue. All other authors make no further declarations of interest.