Inflammation-type dysbiosis of the oral microbiome associates with the duration of COVID-19 symptoms and long COVID

Abstract

In the COVID-19 pandemic, caused by SARS-CoV-2, many individuals experience prolonged symptoms, termed long-lasting COVID-19 symptoms (long COVID). Long COVID is thought to be linked to immune dysregulation due to harmful inflammation, with the exact causes being unknown. Given the role of the microbiome in mediating inflammation, we aimed to examine the relationship between the oral microbiome and the duration of long COVID symptoms. Tongue swabs were collected from patients presenting with COVID-19 symptoms. Confirmed infections were followed until resolution of all symptoms. Bacterial composition was determined by metagenomic sequencing. We used random forest modeling to identify microbiota and clinical covariates that are associated with long COVID symptoms. Of the patients followed, 63% developed ongoing symptomatic COVID-19 and 37% went on to long COVID. Patients with prolonged symptoms had significantly higher abundances of microbiota that induced inflammation, such as members of the genera Prevotella and Veillonella, which, of note, are species that produce LPS. The oral microbiome of patients with long COVID was similar to that of patients with chronic fatigue syndrome. Altogether, our findings suggest an association with the oral microbiome and long COVID, revealing the possibility that dysfunction of the oral microbiome may have contributed to this draining disease.

Keywords: COVID-19; Inflammation; Neurological disorders.

Figure 1. Study enrollment flow chart.

Figure 2. Bacterial abundances predict ongoing symptomatic COVID-19.

RFC to identify predictors of ongoing symptomatic COVID-19 using 6 different combinations of data modalities. (A) F1 scores for the different RFC models trained on different sets of covariates. Box plot represents the median and IQR. (B) Ranking of forest predictors based on median permutated variable importance for the top-performing model. (C) Relative abundances for each bacteria found to be important in predicting ongoing symptomatic COVID-19 from the top-performing RFC model (on clinical data, microbiome species, and Shannon diversity). Violin plots showing the distribution of relative abundance for microbes in each patient with symptoms less than 4 weeks and 4 weeks or longer. 0 indicates “no” and 1 indicates “yes ongoing symptomatic COVID-19.” CC, clinical covariates; Abn, abundances; Div, diversity; RFC, forest classification modeling.

Figure 3. Bacterial abundances can predict long COVID.

RFC modeling to predict long COVID. (A) F1 scores for all subsets of trainable RFC models. (B) Ranking of top 29 predictors associated with long COVID based on median permutated variable importance from the top-performing model (on demographics, clinical data, and Shannon diversity). (C) Relative abundances for each bacteria identified by model (on demographics, clinical data, and Shannon diversity) as important for predicting long COVID are presented as violin plots. Long COVID is presented as orange plots. CC, clinical covariates; Abn., abundances; Div., diversity. RFC, forest classification modeling.

Figure 4. Bacterial metabolic pathways involving inflammation are significantly associated with ongoing symptomatic COVID-19.

Results from RFC modeling to predict ongoing symptomatic COVID-19 and long COVID from HUMAnN3 pathway abundances. (A) F1 scores for demographics, clinical covariates, and pathway abundances and only on pathway abundances. (B) Ranking of forest predictors based on median permutated variable importance from the top-performing model, pathways only, for each outcome. (C) Relative pathway abundances for each pathway found to be important in predicting ongoing symptomatic COVID-19 and long COVID, respectively, by RFC modeling using only pathway abundances. We report violin plots showing the distribution of the relative abundance of pathways in patients with symptoms less than 4 weeks (blue) and 4 weeks or longer (yellow). RFC, forest classification modeling.

Figure 5. Bacterial metabolic pathways involving inflammation are significantly associated with long COVID.

Results from RFC modeling to predict ongoing symptomatic COVID-19 and long COVID from HUMAnN3 pathway abundances. (A) F1 scores for (i) demographics, clinical covariates, and pathway abundances and only on pathway abundances. (B) Ranking of forest predictors based on median permutated variable importance from the top-performing model, pathways only, for each outcome. (C) Relative pathway abundances for each pathway found to be important in predicting long COVID, respectively, by RFC modeling using only pathway abundances. We report violin plots showing the distribution of the relative abundance of pathways in patients with symptoms less than 10 weeks (blue) and 10 weeks or longer (yellow). RFC, forest classification modeling.