Gut microbiome during acute infection predicts risk of long COVID

A longitudinal study reveals that distinct gut microbiome profiles during acute SARS-CoV-2 infection can predict long COVID risk, offering insights into symptom clusters and potential biomarkers.

Study: Gut Microbiome Signatures During Acute Infection Predict Long COVID. Image Credit: Dragana Gordic/Shutterstock.com

*Important notice: bioRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.

In a recent pre-print study posted to bioRxiv*, a team of researchers investigated the predictive role of gut microbiome composition during acute Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infection in the development of Long Coronavirus Disease  (Long COVID) (LC) and its association with clinical variables and symptom clusters.

Background

LC affects 10–30% of non-hospitalized individuals infected with SARS-CoV-2, leading to significant morbidity, workforce loss, and an economic impact of $3.7 trillion in the United States (U.S.).

Symptoms span cardiovascular, gastrointestinal, cognitive, and neurological issues, resembling myalgic encephalomyelitis and other post-infectious syndromes. Proposed mechanisms include immune dysregulation, neuroinflammation, viral persistence, and coagulation abnormalities, with emerging evidence implicating the gut microbiome in LC pathogenesis.

Current studies focus on hospitalized patients, limiting generalizability to milder cases. Further research is needed to explore microbiome-driven predictors in outpatient populations, enabling targeted diagnostics and therapies for LC’s heterogeneous and complex presentation.

About the study

The study was approved by the Mayo Clinic Institutional Review Board and recruited adults aged 18 years or older who underwent SARS-CoV-2 testing at Mayo Clinic locations in Minnesota, Florida, and Arizona from October 2020 to September 2021. Participants were identified through electronic health record (EHR) reviews filtered by SARS-CoV-2 testing schedules.

Eligible individuals were contacted via email, and informed consent was obtained. Of the 1,061 participants initially recruited, 242 were excluded due to incomplete data, failed sequencing, or other issues. The final cohort included 799 participants (380 SARS-CoV-2-positive and 419 SARS-CoV-2-negative), providing 947 stool samples.

Stool samples were collected at two-time points: weeks 0–2 and weeks 3–5 after testing. Samples were shipped in frozen gel packs via overnight courier and stored at −80°C for downstream analyses. Microbial deoxyribonucleic acid (DNA) was extracted using Qiagen kits, and metagenomic sequencing was performed targeting 8 million reads per sample.

Taxonomic profiling was conducted using Kraken2, and functional profiling was performed using the Human Microbiome Project Unified Metabolic Analysis Network (HUMAnN3).

Stool calprotectin levels were measured using enzyme-linked immunosorbent assay (ELISA), and SARS-CoV-2 ribonucleic acid (RNA) was detected using reverse transcription-quantitative polymerase chain reaction (RT-qPCR).

Clinical data, including demographics, comorbidities, medications, and symptom persistence, were extracted from EHRs.

Machine learning models incorporating microbiome and clinical data were utilized to predict LC and to identify symptom clusters, providing valuable insights into the heterogeneity of the condition.

Study results

The study analyzed 947 stool samples collected from 799 participants, including 380 SARS-CoV-2-positive individuals and 419 negative controls. Of the SARS-CoV-2-positive group, 80 patients developed LC during a one-year follow-up period.

Participants were categorized into three groups for analysis: LC, non-LC (SARS-CoV-2-positive without LC), and SARS-CoV-2-negative. Baseline characteristics revealed significant differences between these groups. LC participants were predominantly female and had more baseline comorbidities compared to non-LC participants.

The SARS-CoV-2-negative group was older, with higher antibiotic use and vaccination rates. These variables were adjusted for in subsequent analyses.

During acute infection, gut microbiome diversity differed significantly between groups. Alpha diversity was lower in SARS-CoV-2-positive participants (LC and non-LC) than in SARS-CoV-2-negative participants.

Beta diversity analyses revealed distinct microbial compositions among the groups, with LC patients exhibiting unique microbiome profiles during acute infection.

Specific bacterial taxa, including Faecalimonas and Blautia, were enriched in LC patients, while other taxa were predominant in non-LC and negative participants. These findings indicate that gut microbiome composition during acute infection is a potential predictor for LC.

Temporal analysis of gut microbiome changes between the acute and post-acute phases revealed significant individual variability but no cohort-level differences, suggesting that temporal changes do not contribute to LC development.

However, machine learning models demonstrated that microbiome data during acute infection, when combined with clinical variables, predicted LC with high accuracy. Microbial predictors, including species from the Lachnospiraceae family, significantly influenced model performance.

Symptom analysis revealed that LC encompasses heterogeneous clinical presentations. Fatigue was the most prevalent symptom, followed by dyspnea and cough.

Cluster analysis identified four LC subphenotypes based on symptom co-occurrence: gastrointestinal and sensory, musculoskeletal and neuropsychiatric, cardiopulmonary, and fatigue-only.

Each cluster exhibited unique microbial associations, with the gastrointestinal and sensory clusters showing the most pronounced microbial alterations. Notably, taxa such as those from Lachnospiraceae and Erysipelotrichaceae families were significantly enriched in this cluster.

Conclusions

To summarize, this study demonstrated that SARS-CoV-2-positive individuals who later developed LC exhibited distinct gut microbiome profiles during acute infection. While prior research has linked the gut microbiome to COVID-19 outcomes, few studies have explored its predictive potential for LC, particularly in outpatient cohorts.

Using machine learning models, including artificial neural networks and logistic regression, this study found that microbiome data alone predicted LC more accurately than clinical variables, such as disease severity, sex, and vaccination status.

Key microbial contributors included species from the Lachnospiraceae family, such as Eubacterium and Agathobacter, and Prevotella spp. These findings highlight the gut microbiome’s potential as a diagnostic tool for identifying LC risk, enabling personalized interventions.

*Important notice: bioRxiv publishes preliminary scientific reports that are not peer-reviewed and, therefore, should not be regarded as conclusive, guide clinical practice/health-related behavior, or treated as established information.