Enterococcus and Eggerthella species are enriched in the gut microbiomes of COVID-19 cases in Uganda

The study received ethical approval from the Mulago Hospital Research & Ethics Committee and the Uganda National Council for Science and Technology in June 2020 (approval #s MHREC1868 & HS869ES). We used stool samples from the protocol “Establishment of a Quality Assured COVID-19 Specimen Repository to Support Research in Diagnosis, Prevention and Management of SARS CoV-2 in Uganda” [19]. The study recruited symptomatic, PCR-positive COVID-19 patients, and their household contacts in Uganda from June 2020 to December 2022. PCR-positive participants were enrolled from government-designated COVID-19 treatment centers around the country, with most study participants recruited in the capital city, Kampala. All study participants provided written informed consent before being included in the study.

Samples for SARS-CoV-2 PCR-positive patients were on average collected within four days of a positive PCR result. PCR-positive patients were further consented to recruit their household contacts, who were included as asymptomatic controls if they were free of COVID-19 symptoms (these individuals were not PCR tested due to regulations governing use of scarce testing supplies at the time). Blood (plasma, serum), saliva, urine, stool, and oropharyngeal/nasopharyngeal swabs were collected from each participant at baseline. Samples collected were processed and stored at the Integrated Biorepository of H3Africa Uganda (IBRH3AU). Aliquots of stool from 190 participants were used for this study.

Data was collected using an electronic questionnaire as described in Kamulegeya et al. [19]. This questionnaire collected data on participant sociodemographic factors, occupation, marital status, education background, travel history, and current and past health conditions. In addition to the questionnaire, an interview was also administered by a trained nurse. Available metadata is compiled in Supplemental Table 1.

Genomic DNA (gDNA) was isolated using the DNeasy PowerLyzer PowerSoil Kit from Qiagen (Cat. No./ID: 12855-50). Following the manufacturer’s protocol, one scoop of stool was taken from each stool aliquot for gDNA extraction. gDNA was quantified using a Nanodrop and diluted to 10 ng/μl. 16S rRNA amplification was conducted using the KAPA HiFi Hotstart kit (Cat. No. 50-196-5215) and the primers targeting the V4 region as described by Gohl et al. [20] (Supplemental Table 2). Cycling conditions were as follows: 95 °C for 5 min, followed by 25 cycles of 98 °C for 20 s, 55 °C for 15 s, and 72 °C for 1 min. Final extension for 10 min at 72 °C. A second PCR followed to add sample-specific indices and Illumina compatible flow cell adaptors (Supplemental Table 2). Cycle was as follows: 95 °C for 5 min; ten cycles of 98 °C for 20 s, 55 °C for 15 s, and 72 °C for 1 min; and a final extension at 72 °C for 10 min. A left-side selection (0.8×) using SPRI beads followed the PCR using the manufacturer’s protocol. The PCR product was quantified using a Qubit and pooled for sequencing on an Illumina MiSeq to generate 2 × 250 bp paired end reads.

Each of the 126 samples included in analysis contained a minimum number of 1000 reads per sample (Supplemental Table 1). Demultiplexed reads were pooled from multiple sequencing runs. Reads were demultiplexed using standard Illumina software or the iu-demultiplex package [21]. The Quantitative Insights into Microbial Ecology 2 (Qiime2) pipeline version 2023.5.1 was used for downstream analyses [22]. Potential chimeric reads were removed using consensus-based methods. Amplicon sequence variants (ASVs) were inferred using DADA2 [23] and a phylogenetic tree was built using MAFFT alignment [24, 25]. Taxonomic assignment was performed using the SILVA138 database [26]. For alpha and beta diversity analyses, samples were rarefied to 1000 reads per sample using the Qiime2 pipeline.

Alpha diversity was calculated using Faith’s Phylogenetic Diversity (PD) and the Shannon Index, with between group significance determined using a Wilcoxon rank sum test. Using the Qiime2 package, beta diversity was calculated using the Weighted UniFrac, Bray–Curtis, and Jaccard distance matrices. PERMANOVA tests were done to see if the distributions were different. The Weighted UniFrac value was set as the dependent variable, while SARS-CoV2 infection, participant location, antibiotic use, Enterococcus spp. presence, age, or sex were used as the independent variables. PERMDISP was used to calculate whether the variances between two groups were significantly different. The metadata file and Qiime2 files were imported into R version 4.2.1 and merged into a single Phyloseq object. Analysis of composition of microbiomes with bias correction (ANCOM-BC) [27] and microbiome multivariable associations with linear models (MaAsLin2) [28] were performed in R using the “ANCOMBC” [27, 29] and “Maaslin2” [28] packages. Bacterial taxa were determined to be differentially present based on an adjusted p-value of < 0.05. Plots were generated using the “ggplot2” [30] package and base R functions and edited in Adobe Illustrator.

The Qiime2 taxonomy feature table was used to determine which positive samples had Enterococcus species present. gDNA from these samples was used for shotgun whole genome sequencing using the published Hackflex protocol [31]. Samples were pooled and sequenced on an Illumina NovaSeq to obtain 2 × 150 bp reads. Sequences were uploaded onto Chan Zuckerberg’s ID (CZ ID) online pipeline and reads per million calculated using the NT database [32]. Sequences were analyzed using the metagenomics Illumina pipeline 8.3.0 and the antimicrobial resistance pipeline 1.3.2.

A meta-analysis of six metagenomic inflammatory bowel disease (IBD) studies from the United States was performed to evaluate the presence of Enterococcus species in additional patient populations and healthy controls [33‐38] (see Supplemental Table 4). All data was retrieved from the Sequencing Read Archive as raw fastq files and processed collectively through the CZ ID metagenomic pipeline which maps short read sequencing files to NCBI's bacterial nucleotide database. Sequences were uploaded onto the CZ ID online pipeline and analyzed as described above. Prior to analysis, samples were filtered to remove any IBD subjects in remission (non-active disease states) and any time points where subjects were undergoing a clinical intervention. Patients and controls in these studies were separated into antibiotic status based on whether antibiotics had been used within 3 months.

Participants were recruited from three COVID-19 treatment centers: Mulago National Referral Hospital, Entebbe Grade B Referral Hospital, and Masaka Regional Referral Hospital from June 2020 to December 2022 as described in Kamulegeya et al. [19] Individuals who tested PCR positive for SARS-CoV-2 and their asymptomatic household members were recruited to the study. The resulting COVID-19 biobank managed by the Integrated Biorepository of H3Africa Uganda (IBRH3AU) contains blood (plasma, serum), saliva, urine, stool, and nasal swabs. Participants also completed an electronic questionnaire for demographic data. Asymptomatic control individuals were not PCR tested for SARS-CoV-2 due to the global shortage of tests at the time, so their infection status is unknown.

We extracted DNA and generated 16S rRNA gene amplicon sequencing libraries from 190 stool samples. 126 libraries passed quality checks and were used for subsequent analysis (Supplemental Table 1). From these filtered samples, 101 samples had demographic metadata that allowed for more specific analyses, such as geographic location, sex/gender, age, and antibiotic exposure at the time of sample collection (Fig. 1, Table 1).

Bild vergrößern

We evaluated the composition of the gut microbiome of individuals positive for SARS-CoV-2 infection and asymptomatic controls using 16S rRNA gene amplicon sequencing of DNA isolated from stool samples. This allowed identification of most amplicon sequence variants (ASVs) at the genus level. Based on the abundances of these ASVs, we calculated alpha diversity for each sample using a metric that includes phylogenetic information (Faith’s Phylogenetic Diversity (PD)) and a metric that is phylogeny independent (Shannon Index). We identified significantly lower alpha diversity in the gut microbiomes of COVID-19 cases compared to those of asymptomatic controls using both metrics, suggesting that COVID-19 cases have less complex microbiomes than controls (Fig. 2A, B, Faith’s PD p*** < 0.001, Shannon p**** < 0.0001, Wilcoxon).

Bild vergrößern

We next investigated the similarity of gut microbiomes across individuals using beta diversity. We used the Weighted UniFrac distance metric which takes into account both ASV abundance and taxonomy. The gut microbiomes of individuals infected with SARS-CoV-2 formed an overlapping, but distinct distribution by principal coordinate analysis when compared to asymptomatic controls (Fig. 3A, PERMANOVA p < 0.01). Microbiomes of asymptomatic individuals were more similar to each other, and those of infected individuals were significantly more dispersed (p < 0.01, PERMDISP). We also compared microbiome diversity using the Bray–Curtis and Jaccard distance metrics, which consider abundance and presence/absence respectively (Supplemental Fig. 1A, B). We see similar results using these different distances metrics as we do when using Weighted UniFrac (Supplemental Fig. 1A, B, p < 0.01, PERMANOVA; p < 0.01 PERMDISP). No differences were observed between the microbiomes of male and female subjects (Supplemental Fig. 1C, p < 0.01). Age of subjects, however, was linked to increased dispersion (Supplemental Fig. 1D). This may be in part due to the fact that the median age of asymptomatic controls was much lower than that of the COVID-19 cases.

Bild vergrößern

We overlaid clinical and geographic data on the PCoA plot to see if these groupings drove the differences observed. We first wanted to test if antibiotic exposure explained the increased dispersion in COVID-19 cases. Comparing the distribution and dispersion of microbiomes from SARS-CoV-2 positive individuals with and without antibiotic exposure did not yield any significant differences (Fig. 3B, p = 0.577, PERMANOVA; p = 0.915, PERMDISP). We next tested the role of geography in microbiome composition. Both controls and COVID-19 cases were sampled from geographic regions around Uganda. Samples from individuals in the same district did not cluster with each other (Supplement Fig. 1E). Districts were classified into major cities (urban) or rural (all other areas) based on country data [39]. Using these criteria, gut microbiome beta diversity also did not differ significantly based on district urbanization (Fig. 3C, p = 0.127, PERMANOVA; p = 0.515, PERMDISP). Given that these variables are not significantly related to dispersion between microbiome samples, we conclude that COVID-19 cases are correlated with increased dispersion independent of the recorded metadata.

We next explored whether specific taxa are enriched in the microbiomes of COVID-19 cases or controls. Analysis of microbiome composition with bias correction (ANCOM-BC) [27] identified genera that were significantly enriched in asymptomatic controls. Species significantly enriched in controls with a log2(fold change) greater than 1.5 include Lactobacillus, Clostridium sensu stricto 1, Haemophilus, Romboutsia, Agathobacter, and Akkermansia (Fig. 4A, Supplemental Fig. 2). Additional differential abundance testing using microbiome multivariable associations with linear models (MaAsLin2) [28] identified Turicibacter and Lachnospira as species that are also significantly enriched in controls (Fig. 4B, adjusted p-value < 0.05). Akkermansia and Romboutsia, in particular, have been shown to be associated with beneficial outcomes in the context of inflammatory conditions such as diabetes and ulcerative colitis [40‐42].

Bild vergrößern

The gut microbiomes of COVID-19 cases were strikingly enriched for Eggerthella and Enterococcus (1.76-fold and 2.75-fold respectively, adjusted p-value < 0.05) (Fig. 4). Enterococcus species were completely absent from asymptomatic controls, whereas Enterococcus was found in 56.6% of COVID-19 cases. The range of Enterococcus relative abundance in the COVID-19 cases ranged from 0 to 85% (Fig. 5A). Eggerthella species were present in only 2% of asymptomatic individuals, whereas Eggerthella was found in 43.4% of COVID-19 cases. Eggerthella relative abundance in COVID-19 cases ranged from 0 to 4% (Fig. 5B). MaAsLin2 analysis additionally identified Lachnoclostridium as enriched in COVID-19 cases (Fig. 5C).

Bild vergrößern

We next tested if the presence of each species was linked to the heterogeneity in beta diversity observed above. We plotted the beta diversity PCoA and labeled the presence/absence of Enterococcus (Fig. 5D). The microbiomes of COVID-19 cases that had Enterococcus did not cluster together, but formed a distinct distribution to the microbiomes of COVID-19 cases with no Enterococcus (p < 0.01, PERMANOVA). Microbiomes with Enterococcus were quite dispersed. While this suggests that Enterococcus presence contributes to the broader dispersion seen in the microbiomes of COVID-19 cases, it was not sufficient to explain this increased dispersion. Enterococcus positive microbiomes were not significantly more dispersed than COVID-19 positive microbiomes without Enterococcus (p = 0.639, PERMDISP). Eggerthella positive COVID-19 cases also formed a statistically distinct distribution from positive COVID-19 cases without Eggerthella (Fig. 5E), while changes attributed to Lachnoclostridium did not reach significance by PERMANOVA (Fig. 5F).

Antibiotic exposure has been previously shown to cause blooms in Enterococcus abundance in hospitalized patients with diarrhea in Vietnam and cancer in the United States [43, 44]. Available metadata on antibiotic use allowed for us to test this hypothesis in the context of COVID-19 and Uganda. Individuals positive for SARS-CoV-2 were compared based on whether or not they received antibiotics at the time of sample collection. The relative abundance was also not significantly different based on antibiotic exposure within COVID-19 cases (Fig. 6A, p = 0.242 Wilcoxon test) suggesting that an increased risk of Enterococcus bloom in COVID-19 cases is not due to increased antibiotic exposure alone (16.8% of COVID-19 cases without antibiotics contained Enterococcus vs. 13.9% of cases exposed to antibiotics, Fig. 6B). Reported antibiotic exposure also did not result in significant changes in alpha diversity among COVID-19 positive individuals (Supplemental Fig. 3 A, B). Antibiotic exposure based on sex and geography also did not result in significant changes in alpha and beta diversity among COVID-19 positive individuals (Supplemental Fig. 3 C–H). Interestingly, the vast majority of Enterococcus positive microbiomes were sampled from individuals at Mulago hospital and antibiotic exposure within that particular site is not correlated with the abundance of Enterococcus (Supplemental Fig. 4).

Bild vergrößern

To determine the Enterococcus species contributing to these blooms, we performed shotgun metagenomic sequencing on five Enterococcus-containing samples. These samples were identified as Enterococcus-containing via the 16S rRNA amplicon sequencing. The metagenomes of the Enterococcus-containing samples were analyzed using the Chan Zuckerberg Infectious Disease (CZ ID) online platform [32]. These samples identified that the predominant species contributing to these blooms was E. faecium. E. faecium was identified in five of five sequenced samples (Fig. 6C). E. faecalis was also identified in all samples sequenced at a lower percentage (Fig. 6D). The Eggerthella species was identified as Eggerthella lenta, which was present in all metagenome sequenced samples (Supplemental Fig. 5). The antimicrobial resistance pipeline identified aminoglycoside, tetracycline, macrolide, streptogramin, lincosamide, and fluoroquinolone resistance reads associated with E. faecium (Supplement Table 3). Macrolide, fluoroquinolone, diaminopyrimidine, lincosamide, and pleuromutilin resistance reads were associated with E. faecalis (Supplement Table 3).

We next sought to understand the distribution of Enterococcus in individuals outside Uganda to compare to what we observed there. We hypothesized that if Enterococcus is linked to susceptibility to symptomatic COVID-19, we may observe more Enterococcus in populations with higher case fatality rates than Uganda, including the United States. We re-analyzed gut metagenome data to identify levels of Enterococcus species in six readily available studies from the United States which included healthy controls and patients with inflammatory bowel disease [33‐38] (Supplemental Table 4). We found that E. faecium and E. faecalis were found at a low level in some healthy controls (less than 2500 reads per million, Supplemental Fig. 6A, B). Although this data is not directly comparable to our amplicon sequencing data, it provides evidence that baseline Enterococcus levels in the population may vary by geography. Additionally, antibiotic exposure in this cohort also did not influence Enterococcus abundance, supporting our finding that antibiotics did not play a role in Enterococcal blooms in our population (Supplemental Fig. 6A). In fact, in the US datasets there was a trend toward antibiotic exposure suppressing Enterococcus in healthy individuals. Enterococcus was more prevalent in individuals with inflammatory bowel disease, although again antibiotic history did not influence the abundance significantly (Supplemental Fig. 6, not significant, Wilcoxon test). Thus, Enterococcus presence, like COVID-19 disease severity, may be heavily influenced by underlying medical conditions and geography.