Early immune markers of clinical, virological, and immunological outcomes in patients with COVID-19: a multi-omics study

Zicheng Hu, Kattria van der Ploeg, Saborni Chakraborty, Prabhu S. Arunachalam, Diego Am Mori, Karen B. Jacobson, Hector Bonilla, Julie Parsonnet, Jason R. Andrews, Marisa Holubar, Aruna Subramanian, Chaitan Khosla, Yvonne Maldonado, Haley Hedlin, Lauren de la Parte, Kathleen D. Press, Maureen Ty, Gene S. Tan, Catherine A. Blish, Saki Takahashi, Isabel Rodríguez-Barraquer, Bryan Greenhouse, Atul J. Butte, Upinder Singh, Bali Pulendran, Taia T. Wang, Prasanna JagannathaneLife 2022 · DOI

TL;DR

This study identifies early immune system signatures in COVID-19 patients that can predict disease progression, viral load, and long-term immune responses. By combining proteomics and transcriptomics, they developed machine-learning models that accurately predict these outcomes, offering a potential tool for early risk stratification and personalized treatment.

What Problem Does This Paper Address?

While most COVID-19 infections are mild, a subset of individuals progress to severe disease, and there's significant variability in how the immune system responds. Identifying early indicators of who will develop severe illness or have a robust, lasting immune response is crucial for guiding treatment decisions and developing better vaccines. Current methods often rely on later-stage assessments, leaving a gap in early predictive capabilities.

Methods

The researchers employed a multi-omics approach, analyzing both plasma protein levels (proteomics using Olink panels) and gene expression (transcriptomics via RNA sequencing) from blood samples collected early in the course of COVID-19 infection. They used longitudinal data from a clinical trial to track immune responses over time. Key analytical techniques included:

  • Principal Component Analysis (PCA): To reduce the dimensionality of the transcriptomic and proteomic data and identify major patterns.
  • Trajectory Analysis: To understand how immune modules and proteins change over time since symptom onset.
  • Regression Models: To associate early immune markers with clinical outcomes (disease progression, viral shedding) and immunological outcomes (T cell and antibody responses).
  • Machine Learning (Random Forest): To build predictive models using a small set of protein markers to forecast future outcomes.
  • Cross-validation and Independent Validation: To assess the robustness and generalizability of the predictive models.

Datasets

  • Primary Cohort (Lambda Trial): 108 outpatients with PCR-confirmed SARS-CoV-2 infection, enrolled within 72 hours of diagnosis and followed for up to 7 months. Samples were collected at enrollment (day 0) and day 5.
  • Validation Cohort 1 (Favipiravir Trial): 54 participants from a similar clinical trial, with plasma proteomics and immunological data collected at enrollment and day 28.
  • Validation Cohort 2 (Patel et al., 2021): A published dataset of plasma proteins from 58 COVID-19 patients (26 moderate, 34 severe cases) used to test the model's ability to predict disease severity.

Key Insights and Ideas

  • Sequential Immune Activation: The study reveals a predictable sequence of immune responses in mild-to-moderate COVID-19, starting with interferon (IFN) signaling and innate immune activation, followed by T cell and B cell responses, and finally antibody production.
  • Early Predictors of Severity: Specific early immune markers, particularly those related to IFN signaling and RIG-I (a cytosolic RNA sensor), are strongly associated with increased risk of disease progression.
  • Shared Biomarkers with Vaccination: Interestingly, several immune markers identified as predictive in COVID-19 patients were also found to be associated with immune responses following mRNA vaccination, suggesting common pathways of immune memory formation.
  • RIG-I as a Master Biomarker: Plasma levels of RIG-I (DDX58) emerged as a particularly strong and stable predictor, correlating with disease progression, viral load, T cell responses, and antibody levels, independent of symptom onset time.
  • Machine Learning for Prediction: A small panel of 2-7 plasma proteins, measured early, can accurately predict future clinical and immunological outcomes, outperforming demographic factors alone.

Key Results

  • Disease Progression Prediction: Machine learning models using 2-7 plasma proteins achieved an AUC of 0.90 for predicting disease progression in the primary cohort and 0.85 in an independent dataset.
  • Immune Response Prediction: Models accurately predicted SARS-CoV-2-specific CD4+ T cell responses (AUC 0.82) and S protein-binding IgG levels (AUC 0.76 and 0.77 for day 28 and month 7, respectively).
  • IFN Signaling and Progression: Early enrichment of IFN signaling and RIG-I pathways was significantly associated with a higher risk of disease progression.
  • Peginterferon Lambda Ineffectiveness: Treatment with Peginterferon lambda did not show significant effects on immune profiles or clinical outcomes in this cohort.

Limitations and Future Work

  • Causality: While associations are strong, the study cannot definitively establish causal relationships. Future work could involve experimental manipulation of identified immune modules.
  • Early Infection Window: The study focused on the first two weeks post-symptom onset. Characterizing immune responses even earlier, pre-symptomatically, could provide further insights.
  • Individual Trajectories: The trajectory analysis was primarily population-level. More frequent sampling in future studies could better capture individual immune dynamics.
  • Variant Specificity: The study was conducted before the widespread emergence of Omicron and Delta variants. Validation with newer variants is needed.
  • Thresholds for Immunity: The use of median values as cutoffs for defining high/low responses might not reflect true protective thresholds. Further research is needed to identify clinically relevant thresholds.

The authors suggest that these predictive models could be valuable for identifying patients at high risk of severe COVID-19, guiding therapeutic interventions, and potentially predicting the degree of immune memory following infection or vaccination.

Summary updated

Abstract

Support for the study was provided from National Institute of Health/National Institute of Allergy and Infectious Diseases (NIH/NIAID) (U01 AI150741-01S1 and T32-AI052073), the Stanford's Innovative Medicines Accelerator, National Institutes of Health/National Institute on Drug Abuse (NIH/NIDA) DP1DA046089, and anonymous donors to Stanford University. Peginterferon lambda provided by Eiger BioPharmaceuticals.
BibTeX
@inproceedings{hu2022early,
  author = {Zicheng Hu and Kattria van der Ploeg and Saborni Chakraborty and Prabhu S. Arunachalam and Diego AM Mori and Karen B. Jacobson and Hector Bonilla and Julie Parsonnet and Jason R. Andrews and Marisa Holubar and Aruna Subramanian and Chaitan Khosla and Yvonne Maldonado and Haley Hedlin and Lauren de la Parte and Kathleen D. Press and Maureen Ty and Gene S. Tan and Catherine A. Blish and Saki Takahashi and Isabel Rodríguez-Barraquer and Bryan Greenhouse and Atul J. Butte and Upinder Singh and Bali Pulendran and Taia T. Wang and Prasanna Jagannathan},
  title = {{Early immune markers of clinical, virological, and immunological outcomes in patients with COVID-19: a multi-omics study}},
  booktitle = {eLife},
  year = {2022},
  doi = {10.7554/elife.77943},
}
Figure