Plasma proteomics identifies early markers of endothelial and inflammatory activation associated with dengue disease severity in children

This study utilizes plasma proteomics and machine learning to identify early biomarkers, such as PTX3 and CLEC11A, that distinguish subclinical from severe dengue cases in children by revealing early endothelial dysfunction and inflammatory activation.

The Gist

The Dengue Detective: How a Blood Test Could Predict a Child's Fever Before It Gets Dangerous

Imagine your body is a bustling city. When the Dengue virus invades, it's like a chaotic riot breaking out in the streets. In some people, the city's police force (the immune system) handles the riot quickly, and life goes back to normal with just a few days of fever and aches. This is a mild case.

But in other people, the riot spirals out of control. The police overreact, the city's infrastructure (blood vessels) starts to crumble, and the whole city faces a crisis. This is severe dengue, which can be fatal.

The big problem? Doctors often can't tell who will stay safe and who will crash until it's too late. By the time the "city" starts collapsing, it's an emergency.

This paper is about a team of scientists who decided to act like detectives. They looked at the "smoke signals" (proteins) floating in the blood of children with Dengue to see if they could predict the disaster before it happened.

---

1. The Investigation: Looking at the "Smoke Signals"

The researchers took blood samples from 92 children in Cambodia. Some were sick enough to be in the hospital; others had the virus but felt fine (subclinical). They also had a group of healthy kids to compare against.

Instead of just looking at the virus, they used a high-tech microscope (mass spectrometry) to take a snapshot of 500 different proteins in the blood. Think of these proteins as the city's messengers. When something goes wrong, these messengers shout out specific messages.

2. The Big Discovery: The "Leaky Pipe" Warning

The detectives found a clear pattern.

• The Healthy City vs. The Riot: When they compared the sick kids to the healthy ones, the blood of the sick kids was full of "emergency alarms." These were proteins related to inflammation (the fire) and blood clotting (the traffic jams).
• The Early Warning System: The most exciting part? They found that even before a child got really sick, their blood was already showing signs of trouble.
  • The "Leaky Pipe" Marker: One specific protein, called PTX3, was like a siren screaming "The pipes are leaking!" It showed up early and got louder in kids who ended up in the hospital. This protein is linked to the blood vessels losing their seal, which causes the dangerous fluid leakage seen in severe dengue.
  • The "Riot Control" Marker: Another protein, CLEC11A, was the loudest siren for the most severe cases. It was like a signal saying, "The riot is getting out of hand!"

3. The "Subclinical" Surprise

The researchers also looked at the kids who had the virus but didn't feel very sick. They found that even these kids had some of the "leaky pipe" signals, but they were much quieter. It's like a small crack in a pipe versus a burst pipe. The difference in the volume of these signals was the key to predicting who would get worse.

4. The Crystal Ball: A Computer Prediction

The team didn't just stop at finding the signals; they built a computer crystal ball (a machine learning model).

They fed the computer the protein data from the first few days of a child's illness. The computer learned to recognize the patterns:

• "This pattern means the child will likely recover fine."
• "This pattern means the child is heading for a severe crisis."

The computer was surprisingly good at it. It could tell the difference between a mild case and a severe one with high accuracy, just by looking at the blood proteins on day one or two.

5. Why This Matters

Right now, if a child comes into a clinic with a fever in a tropical country, doctors often have to wait and see. They watch for signs of shock or bleeding, which usually happens a few days later.

This study suggests that in the future, a simple blood test could act like a weather forecast for the body.

• If the forecast says "Stormy": The doctor can admit the child to the hospital immediately, give them fluids, and watch them closely.
• If the forecast says "Sunny": The child can go home, rest, and be monitored without clogging up the hospital.

The Bottom Line

This paper is a breakthrough because it moves us from reacting to dengue (treating it after it gets bad) to predicting it. By listening to the body's early "smoke signals," we might be able to save lives by catching the crisis before the city collapses. It turns a scary, unpredictable fever into a manageable situation with a clear plan.

Technical Summary

1. Problem Statement

Dengue virus (DENV) infection presents a spectrum of severity ranging from asymptomatic/subclinical cases to Dengue Fever (DF) and life-threatening Dengue Hemorrhagic Fever/Shock Syndrome (DHF/DSS).

• Clinical Gap: The biological mechanisms distinguishing subclinical infection from severe disease progression are poorly understood. Currently, there is no antiviral therapy, and clinical management relies on fluid resuscitation.
• Prediction Challenge: Early risk stratification is critical to prevent hypovolemic shock and death, particularly in low-resource settings. However, existing biomarkers often fail to predict severity before clinical deterioration (plasma leakage) occurs.
• Research Limitations: Previous proteomic studies have been limited by small sample sizes, reliance on adult cohorts, single time-point sampling, or comparisons only against healthy donors, lacking the granularity to distinguish subclinical from severe pediatric cases longitudinally.

2. Methodology

The study employed a longitudinal, high-resolution plasma proteomics approach combined with machine learning.

• Cohort:
  • Participants: 92 children (mostly pediatric) with laboratory-confirmed post-primary DENV-2 infections from Cambodia, plus 10 age-matched healthy donors.
  • Groups: 23 subclinical cases (asymptomatic/mild) and 69 hospitalized patients (classified as DF, DHF, or DSS per WHO 1997 criteria).
  • Sampling: Longitudinal collection across four disease phases: Acute (AP), Critical (CP), Early Recovery (ERP), and Recovery (RP), totaling 296 plasma samples.
• Proteomics Workflow:
  • Technique: Reverse-phase liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) using DIA-PASEF (Data-Independent Acquisition Parallel Accumulation–Serial Fragmentation) on a timsTOF HT instrument.
  • Quantification: Label-free quantification (LFQ) of 500 proteins after excluding red blood cell contaminants and immunoglobulin variable domains.
  • Viral Detection: The search database included the DENV-2 polyprotein to detect viral proteins (specifically NS1).
• Statistical & Machine Learning Analysis:
  • Differential Analysis: Linear mixed-effects models and two-sample t-tests (with Benjamini-Hochberg correction) to identify differentially abundant proteins between groups (Subclinical vs. Hospitalized; DF vs. DHF/DSS).
  • Classification: Development of Random Forest and Elastic Net Logistic Regression classifiers to predict disease severity.
  • Validation: Performance evaluated using Leave-One-Patient-Out Cross-Validation (LOPO-CV) and Area Under the Receiver Operating Characteristic Curve (AUROC).

3. Key Contributions

• First Pediatric Subclinical Proteome: This is the first study to provide a longitudinal plasma proteomic characterization of subclinical DENV-2 infections in children, directly comparing them to hospitalized cases at matched time points.
• Early Endothelial Dysregulation: Identification of endothelial dysfunction as an early driver of severity, detectable at the onset of symptoms (Acute Phase), prior to clinical signs of shock.
• Novel Biomarker Discovery: Identification of specific proteins (e.g., PTX3, CLEC11A, HRG) that distinguish severity levels earlier than traditional clinical markers.
• Predictive Modeling: Successful construction of machine learning models that can stratify patients into subclinical, DF, and DHF/DSS categories using a small panel of plasma proteins sampled during the acute phase.

4. Key Results

A. Longitudinal Proteome Dynamics

• Phase Separation: PCA and clustering clearly separated acute/critical phases from recovery phases. Healthy donor profiles overlapped significantly with the recovery phase, indicating homeostasis restoration by day 60.
• Acute Phase Signatures: The acute phase was dominated by upregulation of acute-phase reactants (CRP, SAA1, SERPINA3), innate immune markers (C2, CTSD, B2M), and coagulation factors.

B. Subclinical vs. Hospitalized (Severity Stratification)

• Endothelial Activation: Hospitalized patients showed significantly elevated levels of endothelial activation markers (VCAM-1, PTX3, VCAN, CD14, LBP) starting in the Acute Phase.
• PTX3 as a Key Marker: Pentraxin 3 (PTX3) exhibited the strongest and most rapid upregulation in hospitalized patients compared to subclinical cases, suggesting it is a marker of imminent vascular involvement.
• Complement Dysregulation: Hospitalized patients showed a shift from classical to alternative/lectin pathway activation (elevated C2, COLEC11, CFB).
• Viral Load: DENV NS1 protein was detected in 46% of hospitalized severe cases vs. only 13% of subclinical cases, correlating with disease severity.

C. DF vs. DHF/DSS (Severe Disease Stratification)

• Differentiation at Admission: At hospital admission, 18 proteins significantly distinguished DF from DHF/DSS.
• Top Marker: CLEC11A (Osteolectin) displayed the highest fold change, serving as the strongest discriminator for severe disease.
• Complement & Stress: Severe cases showed dysregulation in the lectin pathway (FCN2, COLEC11) and elevated cytosolic stress proteins (CYCS, CCT2, CCT3), indicating cellular damage.
• Antibody Response: Total IgA1 levels were significantly lower in DHF/DSS patients compared to DF, suggesting an insufficient IgA response may contribute to severe progression.

D. Machine Learning Performance

• Classification Accuracy:
  • Acute vs. Recovery: AUROC = 0.992.
  • Hospitalized vs. Subclinical: AUROC = 0.887.
  • DF vs. DHF/DSS: AUROC = 0.762.
• Feature Importance:
  • Acute vs. Recovery: Driven by VCAM1 and FGL1.
  • Hospitalized vs. Subclinical: Driven by HRG and SERPING1 (complement regulator).
  • DF vs. DHF/DSS: Driven by CLEC11A and CYCS.

5. Significance and Implications

• Pathophysiological Insight: The study confirms that endothelial dysfunction and dysregulated inflammation are the earliest molecular events distinguishing severe dengue, occurring before clinical plasma leakage.
• Clinical Utility: The identified protein panel (including PTX3, CLEC11A, HRG, and complement regulators) offers a potential tool for early risk stratification. This could allow clinicians to identify children at risk of progressing to DHF/DSS at hospital admission, enabling proactive fluid management and monitoring.
• Therapeutic Targets: The findings highlight specific pathways (complement activation, endothelial glycocalyx disruption) that could be targeted by host-directed therapies.
• Methodological Advance: Demonstrates the feasibility of using untargeted plasma proteomics in resource-limited settings to capture viral proteins (NS1) and host responses simultaneously, paving the way for targeted mass spectrometry diagnostics.

Limitations: The study focused on a single geographic region (Cambodia) and a single serotype (DENV-2) in a pediatric cohort. The number of subclinical samples was smaller than the hospitalized cohort, and complete longitudinal profiling was available for only a subset of participants. Validation in diverse populations and serotypes is required.