Expression profile of HERVs and inflammatory mediators detected in nasal mucosa as a predictive biomarker of COVID-19 severity

Vita Petrone, Marialaura Fanelli, Martina Giudice, Nicola Toschi, Allegra Conti, Christian Maracchioni, Marco Iannetta, Claudia Resta, Chiara Cipriani, Martino Tony Miele, Francesca Amati, Massimo Andreoni, Loredana Sarmati, Paola Rogliani, Giuseppe Novelli, Enrico Garaci, Guido Rasi, P Sinibaldi Vallebona, Antonella Minutolo, Claudia Matteucci, Emanuela Balestrieri, Sandro GrelliFrontiers in Microbiology 2023 · DOI

This study, published in Frontiers in Microbiology, explores how the activation of ancient viral sequences in our own DNA—Human Endogenous Retroviruses (HERVs)—can serve as an early warning system for COVID-19 severity. By analyzing nasal swabs with machine learning, researchers identified a genomic "signature" that predicts hospitalization and oxygen needs with high accuracy at the very onset of infection.

TL;DR

Researchers discovered that SARS-CoV-2 triggers the expression of "fossil" retroviruses (HERVs) in the nasal mucosa, which correlates strongly with disease progression. Using these HERV levels alongside inflammatory markers, they developed a machine learning model that predicts COVID-19 hospitalization with 90% accuracy.

What Problem Does This Paper Address?

Predicting which COVID-19 patients will recover at home and which will require intensive care remains a significant clinical challenge. While blood-based biomarkers exist, they often appear after the disease has already progressed.

This paper investigates the nasal mucosa—the primary entry point for the virus—to find earlier genomic indicators. Specifically, it looks at Human Endogenous Retroviruses (HERVs). These are remnants of ancient viral infections integrated into the human genome millions of years ago. Usually silent, they can be "woken up" by modern viral infections, contributing to the "cytokine storm" that drives severe COVID-19.

Methods

The researchers employed a multi-pronged approach:

  • Molecular Profiling: They used qRT-PCR to measure the mRNA expression of three HERV families (K, W, and H) and 14 inflammatory mediators (including IL-6, TNF-α, and various Interferons) from patient swabs.
  • In Vitro Kinetics: To understand the timing of these markers, they stimulated a human hypopharyngeal cell line (FaDu) with the SARS-CoV-2 spike protein and measured expression at 3, 8, and 24 hours.
  • Machine Learning: They used Extreme Gradient Boosting (XGBoost) to classify patients as "hospitalized" vs. "not hospitalized." They utilized SHAP (Shapley Additive Explanations) values to determine which biological factors were the most influential in the model’s decision-making.

Datasets

  • Clinical Cohort: 63 nasopharyngeal/oropharyngeal swabs collected during the first pandemic wave (March–May 2020) at Policlinico Tor Vergata in Rome.
  • 20 SARS-CoV-2 negative controls.
  • 43 SARS-CoV-2 positive patients (29 non-hospitalized, 14 hospitalized).
  • Cell Line Data: Expression profiles from the FaDu cell line following spike protein exposure.

Key Insights and Ideas

  • HERVs as Early Responders: The in vitro experiments revealed that HERV-K and HERV-W peak just 3 hours after exposure to the spike protein—well before traditional inflammatory cytokines like IL-6. This makes them superior candidates for early-stage biomarkers.
  • The "Severe" Profile: Hospitalized patients exhibited a specific "unfavorable" profile: high levels of HERV-K, HERV-W, and pro-inflammatory cytokines (IL-6, TNF-α), but critically low levels of protective Type I Interferons (IFN-α/β) and TLR-4.
  • Oxygen Stratification: Nasal expression of HERV-W was significantly higher in patients who eventually required supplemental oxygen compared to those who did not, suggesting it reflects the degree of respiratory involvement.
  • Systemic Link: Nasal HERV levels weren't just local; they correlated with systemic clinical markers like fibrinogen (coagulation) and C-reactive protein (inflammation), linking the nasal entry point to whole-body pathology.

Key Results

  • High Predictive Accuracy: The machine learning model achieved an AUC of 0.85 and an overall accuracy of 90% in distinguishing between hospitalized and non-hospitalized patients.
  • Sensitivity: The model was particularly strong at identifying those at risk, with a sensitivity of 0.98.
  • Biomarker Importance: Factor analysis showed that the most predictive "components" were those combining HERV-K/W expression with the SARS-CoV-2 N gene and IL-6.

Limitations and Future Work

  • Small Sample Size: With only 63 patients, the study is a "proof-of-concept." Larger, multi-center cohorts are needed to validate these genomic biomarkers.
  • Variant Evolution: The study was conducted during the first wave. It is currently unknown if newer variants (like Omicron) trigger the same HERV response or if vaccination status alters this genomic signature.
  • Comorbidities: HERVs can be activated by other conditions like cancer or autoimmunity. While the study groups were age-matched, the influence of pre-existing conditions on nasal HERV expression requires deeper investigation.
  • Clinical Implementation: Moving from qRT-PCR research to a rapid diagnostic tool for emergency rooms requires significant analytical validation and simplification of the testing process.

Summary updated

Abstract

Overall, the present results suggest HERVs as contributing elements in COVID-19 and early genomic biomarkers to predict COVID-19 severity and disease outcome.
BibTeX
@inproceedings{petrone2023expression,
  author = {Vita Petrone and Marialaura Fanelli and Martina Giudice and Nicola Toschi and Allegra Conti and Christian Maracchioni and Marco Iannetta and Claudia Resta and Chiara Cipriani and Martino Tony Miele and Francesca Amati and Massimo Andreoni and Loredana Sarmati and Paola Rogliani and Giuseppe Novelli and Enrico Garaci and Guido Rasi and P Sinibaldi Vallebona and Antonella Minutolo and Claudia Matteucci and Emanuela Balestrieri and Sandro Grelli},
  title = {{Expression profile of HERVs and inflammatory mediators detected in nasal mucosa as a predictive biomarker of COVID-19 severity}},
  booktitle = {Frontiers in Microbiology},
  year = {2023},
  doi = {10.3389/fmicb.2023.1155624},
}
Figure