Metagenomic Signatures of Gut Infections Caused by Different Escherichia coli Pathotypes

Abstract

Escherichia coli is a leading contributor to infectious diarrhea and child mortality worldwide, but it remains unknown how alterations in the gut microbiome vary for distinct E. coli pathotype infections and whether these signatures can be used for diagnostic purposes. Further, the majority of enteric diarrheal infections are not diagnosed with respect to their etiological agent(s) due to technical challenges. To address these issues, we devised a novel approach that combined traditional, isolate-based and molecular-biology techniques with metagenomics analysis of stool samples and epidemiological data. Application of this pipeline to children enrolled in a case-control study of diarrhea in Ecuador showed that, in about half of the cases where an E. coli pathotype was detected by culture and PCR, E. coli was likely not the causative agent based on the metagenome-derived low relative abundance, the level of clonality, and/or the virulence gene content. Our results also showed that diffuse adherent E. coli (DAEC), a pathotype that is generally underrepresented in previous studies of diarrhea and thus, thought not to be highly virulent, caused several small-scale diarrheal outbreaks across a rural to urban gradient in Ecuador. DAEC infections were uniquely accompanied by coelution of large amounts of human DNA and conferred significant shifts in the gut microbiome composition relative to controls or infections caused by other E. coli pathotypes. Our study shows that diarrheal infections can be efficiently diagnosed for their etiological agent and categorized based on their effects on the gut microbiome using metagenomic tools, which opens new possibilities for diagnostics and treatment.IMPORTANCEE. coli infectious diarrhea is an important contributor to child mortality worldwide. However, diagnosing and thus treating E. coli infections remain challenging due to technical and other reasons associated with the limitations of the traditional culture-based techniques and the requirement to apply Koch's postulates. In this study, we integrated traditional microbiology techniques with metagenomics and epidemiological data in order to identify cases of diarrhea where E. coli was most likely the causative disease agent and evaluate specific signatures in the disease-state gut microbiome that distinguish between diffuse adherent, enterotoxigenic, and enteropathogenic E. coli pathotypes. Therefore, our methodology and results should be highly relevant for diagnosing and treating diarrheal infections and have important applications in public health.

Keywords: 16S rRNA; Ecuador; Escherichia coli; clinical metagenomics; gut microbiome; infectious diarrhea; metagenomics; pathotypes.

FIG 1

Abundance of human reads and estimated coverage of the metagenomic data sets obtained in this study. (A) Assignment of recovered metagenomic raw reads to three groups: human (purple), discarded due to low quality (orange), and fraction passing quality control and not being of human origin (green). (B) Fitted Nonpareil curves and estimated average coverage for each metagenome after human and low-quality reads where removed from each data set. The horizontal dashed lines indicate 100% (upper red line) and 95% (bottom red line) coverage. Empty circles indicate the size (x axis) and estimated average coverage (y axis) of the data sets, and the lines after that point are projections of the fitted model. The inset plot shows the distribution of estimated coverage values in randomly drawn subsets of 1,000 reads per library for each pathotype and control group. Note that samples where DAEC was isolated showed less diverse communities (higher coverage) than other groups, including control samples.

FIG 2

Characteristics of samples where DAEC was most likely the causative agent. (A) Estimated metagenomic abundance of the reference commensal E. coli (strain HS, in light red) and the DAEC isolate (in red) recovered from the sample, along with the ELISA-based detection of rotavirus and bioinformatic detection of Adenovirus_F for each sample analyzed (rows). Samples where high-quality E. coli MAGs were recovered from the corresponding metagenome are denoted by a star. (B) Presence (detection) of four hallmark virulence factors in the metagenome, including the DAEC marker gene (afaB) and three enterotoxins, i.e., the hemolysin subunit B (hlyB), the heat-labile enterotoxin (eltA), and the secreted autotransporter toxin (sat). (C) Estimated E. coli intrapopulation diversity measured by ANIr of reads against the reference commensal strain HS (light orange) and the isolate obtained from the sample (dark brown). To avoid any potential bias by low in situ abundance, only samples where the average sequence depth of the reference genome was ≥1× were evaluated for ANIr. (D) Number of isolates that originated from cases of diarrhea (in red) versus control samples (in green) and were assigned in the same core-genome-based clonal complex as the isolate (epidemiology).

FIG 3

Core genome-based phylogenetic and ANI relatedness among DAEC (A) and ETEC (B) isolates and MAGs recovered from the same sample of infectious diarrhea. Annotations in blue denote pairs of genomes (isolates and MAGs) that clustered closely in the phylogenetic reconstruction, while annotations in red denote more divergent pairs of genomes. The environmental E. coli strain TW158338 was used an outgroup.

FIG 4

Correlation between recovered fraction of human metagenomic reads and DAEC pathogen abundance. The bar plot shows the observed percentage of the total metagenomic reads assigned to human (purple) and the estimated metagenomic abundance of the E. coli genome for the samples with strong evidence of DAEC infection. The inset plot shows the Pearson correlation analysis of the two variables, revealing a positive linear correlation.

FIG 5

Differentially abundant (diagnostic) taxa between DAEC and ETEC infections. Differentially abundant species were reported if they had a corrected P value of ≤0.05 and an effect size (the magnitude of the difference between groups) of ≥0.8. (A and B) Proportions of metagenomic sequences assigned to Fusobacterium mortiferum and Campylobacter concisus, respectively. (C and D) Proportions of sequences assigned to Bifidobacterium longum and Alloprevotella tannerae, respectively. (E) PCA plot based on the taxonomic composition of each metagenome (annotated at the species level using clade-specific marker genes with MetaPhlAn2) after removal of human and E. coli reads from the libraries.