Metagenomic signatures of extraintestinal bacterial infection in the febrile term infant gut microbiome

Abstract

Background

Extraintestinal bacterial infections (EBIs), e.g., urinary tract infection, bacteremia, and meningitis, occur in approximately 10% of febrile infants younger than 60 days. Although many EBI-causing species commonly reside in the infant gut, proof that the digestive system is a pre-infection habitat remains unestablished.

Results

We studied a cohort of febrile term infants < 60 days old who presented to one of thirteen US emergency departments in the Pediatric Emergency Care Applied Research Network from 2016 to 2019. Forty EBI cases and 74 febrile controls matched for age, sex, and race without documented EBIs were selected for analysis. Shotgun sequencing was performed of the gut microbiome and of strains cultured from the gut and extraintestinal site(s) of EBI cases, including blood, urine, and/or cerebrospinal fluid. Using a combination of EBI isolate genomics and fecal metagenomics, we detected an intestinal strain presumptively isogenic to the EBI pathogen (> 99.999% average nucleotide identity) in 63% of infants with EBIs. Although there was no difference in gut microbiome diversity between cases and controls, we observed significantly increased Escherichia coli relative abundance in the gut microbiome of infants with EBIs caused by E. coli. Infants with E. coli infections who were colonized by the putatively isogenic pathogen strain had significantly higher E. coli phylogroup B2 abundance in their gut, and their microbiome was more likely to contain virulence factor loci associated with adherence, exotoxin production, and nutritional/metabolic function.

Conclusions

The intestine plausibly serves as a reservoir for EBI pathogens in a subset of febrile term infants, prompting consideration of new opportunities for surveillance and EBI prevention among colonized, pre-symptomatic infants.

Video Abstract

Background

Infants younger than 60 days are at considerable risk for extraintestinal bacterial infections (EBIs), which frequently occur in the urinary tract, bloodstream, and meninges [1]. Although early-onset infections (< 7 days of life) have decreased following the introduction of intrapartum antibiotic prophylaxis for Streptococcus agalactiae (Group B Streptococcus) [2], the incidence of late-onset infections (≥ 7 days) remains unchanged. EBIs continue to affect an estimated 10% of febrile infants [3,4,5,6,7]. To date, little is known about how to prevent late-onset infections in full-term infants, motivating a better understanding of the risk factors present in this population.

EBIs can have substantial consequences in young infants, yet these infections remain challenging to diagnose. Fever and/or non-specific symptoms are often the only indicators of infection, and diagnosis requires a positive bacterial culture from a normally sterile site, which may take 24–48 h to identify, though rapid diagnostic tests have reduced this interval for bloodstream infections [8,9,10]. Because of the risk of clinical deterioration, infants are often admitted to the hospital and receive broad-spectrum antibiotics pending culture results [9]. Although a necessary practice to rapidly treat possible EBI, antibiotics risk perturbing an infant’s developing microbiome and selecting for antimicrobial resistance [11,12,13]. Furthermore, antibiotic administration may precede obtaining sterile site cultures, complicating their interpretation and often leading to extended treatments for “partially treated” presumed EBI [9, 14]. Clinical and laboratory characteristics (e.g., fever, white blood cell count, urinalysis, absolute neutrophil count, and serum procalcitonin) can inform EBI risk [9, 15], but specific biomarkers of EBI are lacking. Thus, improved detection of EBIs is needed to commence targeted, timely antibiotic therapy and reduce the burden of EBIs in vulnerable infants.

Bacterial species that commonly cause EBIs in infancy, including Enterobacterales (e.g., Escherichia coli and Klebsiella pneumoniae) and Enterococcus spp. [3, 5,6,7], often reside in the developing gut microbiome [16,17,18]. In adults and children, the gut microbiome can source urinary tract infections (UTIs) via the fecal-perineal-urethral route [19, 20] (ascending infection). Bacteremia is believed to arise during infancy via bacterial translocation from the gut or skin, especially in preterm infants with immature mucosal barriers [21], or from the urinary system [22], while meningitis may arise from bacteremia [23]. Thus, the gut microbiome may serve as a direct or indirect habitat for pathogens that cause EBIs in infants, and gut colonization may be an important precursor of community-acquired EBIs in otherwise healthy infants. However, the extent of gut colonization by the EBI-causing strain, along with microbial community correlates of EBI risk, has not been well characterized in febrile term infants < 60 days, who are in the most vulnerable age group [24]. Identifying both gut bacterial community- and strain-level attributes of EBIs using high genomic resolution, including alterations in the developing microbiome, may improve our understanding of EBI risk during early life.

Here, we sought to investigate the prevalence of gut pathogen colonization among infants with EBIs and define the correlates of invasion using isolate genomics and fecal metagenomics. Our data demonstrate that gut carriage of the EBI-causing strain is common in febrile term infants < 60 days old and can be associated with differences in virulence factor profiles, supporting a role for the intestine as a reservoir of EBI-causing organisms in term infants.

Methods

Participant selection

This study included infants with EBI and matched controls without EBI to investigate (1) whether the EBI-causing strain was present in the gut of cases based on genomic analysis of cultured isolates from cases and (2) whether EBI-causing species were enriched in the gut of cases compared to controls. Samples were obtained from a completed, prospective observational study of febrile infants presenting to the emergency department (ED) of 22 centers across the US in the Pediatric Emergency Care Applied Research Network (PECARN; Fig. S1). Details of the network have been published [15, 25, 26]. Thirteen centers provided samples for this study. Fever was defined as a rectal temperature ≥ 38 °C in the ED, in a prior healthcare setting, or at home as reported by the family in the preceding 24 h. Infants were excluded if they appeared critically ill, had received antibiotics in the preceding 48 h, were born preterm, or had major comorbidities as previously described [27] (e.g., serious congenital abnormalities, inborn errors of metabolism), indwelling devices, or soft-tissue infections. Patients were not excluded for otitis media. Clinical care was at the discretion of the treating clinician. Bacteremia and meningitis were defined as infants whose culture of blood and/or cerebrospinal fluid (CSF), respectively, yielded a single species of a known pathogen, while UTI was defined as growth of a pathogen above a certain threshold as previously described [25, 28]. Febrile infants without document EBIs were selected for use as controls; these infants were culture negative for known pathogens. Each EBI case was matched to up to 2 febrile controls (median: 2) by age (± 7 days of life, using the closest match available), sex (same sex when available ± 7 days of life), and race (data file S1). Infants whose culture grew a suspected contaminant (i.e., coagulase-negative Staphylococci or Streptococcus spp. not including S. agalactiae or S. pneumoniae) were excluded as cases but eligible as controls. All participants provided an enteric specimen (stool and/or rectal swab) in the ED, which was subsequently stored at − 80 °C without further processing until analysis in St. Louis, MO, USA.

Selective culturing

Extraintestinal bacterial isolates from EBI cases and enteric specimens from all cases and controls were sent to Washington University School of Medicine (St. Louis, MO, USA). Enteric specimens from EBI cases were selectively cultured for the extraintestinal pathogen species (Fig. S2). Less than 5% of the enteric specimen was removed without thawing by scraping frozen transport media or stool followed by resuscitation in brain heart infusion (BHI) broth for 1 h and incubation in selective broth until turbid and then plated on selective and differential agar. Samples were grown under the following conditions according to the identity of the organism previously isolated from extraintestinal culture (and as reported by the sending institution): Enterococcus faecalis, Enterococcosel broth with colistin (10 mg/L), amphotericin B (2 mg/L), and nalidixic acid (15 mg/L) followed by Enterococcosel agar (Becton Dickinson & Company); Klebsiella pneumoniae, MacConkey broth with vancomycin (20 mg/L) followed by Klebsiella ChromoSelect agar (Millipore 90925); Escherichia coli or Enterobacter cloacae, MacConkey broth with vancomycin (20 mg/L) followed by MacConkey agar with vancomycin; Proteus mirabilis, MacConkey broth with vancomycin followed by MacConkey agar with a 10 μg colistin disc; and S. agalactiae, tryptic soy broth (TSB) with added NaCl (6.5% w/v), aztreonam (8 mg/L) and amphotericin B (2 mg/L) followed by tryptic soy agar (TSA) and blood agar plates (BAP). Species identity was confirmed with matrix-assisted laser desorption/ionization-time-of-flight mass spectrometry (MALDI-TOF MS) and one colony matching the target species per intestinal sample was profiled by multi-locus sequence typing (MLST) using PCR amplification and sequencing of housekeeping genes, which were matched to known sequence types (STs) using the PubMLST.org database (data file S1) [29]. Potential matches were defined as extraintestinal and intestinal isolates of the same species and ST, except for Proteus mirabilis from one infant, for which neither isolate had a known ST at the time of screening; these isolates were later in silico typed (see below section). Isolates matching the extraintestinal species were retained and stored in glycerol stocks from pure culture at − 80 °C until sequencing. Antibiotic susceptibility testing was performed on Enterobacterales and Enterococcus spp. as previously described [30].

Isolate sequencing

Single isolates were recovered with the following media for subsequent whole-genome sequencing (WGS): Enterococcosel agar plates followed by BHI broth for E. faecalis, MacConkey agar followed by TSB for Enterobacteriaceae spp., and TSA followed by TSB for S. agalactiae. Genomic DNA was extracted from 5 mL overnight cultures using the QIAamp BiOstic Bacteremia DNA kit (QIAGEN). Short-read sequencing libraries were prepared using a Nextera DNA Library Prep kit (Illumina) with modifications [31] and sequenced at 2 × 150 bp on either the Illumina NextSeq or NovaSeq platforms to a target depth of 50 × coverage based on the expected genome size for each species. Reads were demultiplexed by index pair and quality-trimmed with Trimmomatic [32] (v0.39; SLIDINGWINDOW: 4:20, LEADING: 10, TRAILING: 10, MINLEN:60) followed by decontamination of human genomic reads using DeconSeq (v4.3, -dbs hsref38) and read repair with BBMap (v38.63). Oversequenced samples (> 150X coverage) were subsampled to 150 × coverage using SeqTK (v1.3). SPAdes (v3.15.3) was used to assemble genomes with defined parameters (-k 21,33,55,77,87,97,107,117,127 –careful) and short (< 1000 bp) contigs were subsequently removed. Genome assemblies were assessed via Quast (v5.2.0) and CheckM (v1.2.1) and species were identified using MASH (v2.3). Assemblies were confirmed to have < 500 contigs after filtering, > 90% completeness, and < 5% contamination.

Isolate genome analysis

Sequence types were identified in silico with mlst [33] (v2.22.1) (https://github.com/tseemann/mlst), specifying –scheme ecoli_achtman_4 for E. coli genomes. Proteus mirabilis isolates were typed by uploading their genome assemblies to PubMLST.org [29]. Whole-genome single-nucleotide polymorphism (SNP) quantification was performed in an all-versus-all manner for the isolates within each species using snippy [34] (v4.6.0). The more complete genome (based on CheckM output) for each pairwise alignment was selected as the reference genome, with the following tiebreaker criteria: (1) higher N50, (2) lower number of contigs, and (3) higher N90 (for two pairs, all four of these metrics were the same, so a reference genome was randomly chosen). All genomes were annotated using bakta [35] (v1.5.1) and antimicrobial resistance genes (ARGs) were profiled with AMRFinder (v3.10.42).

Metagenomic sequencing

One enteric specimen from each infant was selected for sequencing, prioritizing stool over swab samples when available; however, a pilot sequencing run indicated that community composition was similar for paired stool and rectal swab samples from a subset of infants (p = 0.001, PROTEST; Fig. S3). Frozen samples were chipped with a target quantity of 100 mg of stool (30–50 mg when quantity was limited) and 250 mg of transport media for rectal swabs. Metagenomic DNA was extracted using the DNeasy PowerSoil Pro kit (QIAGEN) with a modified homogenization step of 2 min at 2500 rpm on a bead beater followed by 5 min on ice and an additional 2 min at 2500 rpm. Libraries were prepared as for WGS and sequenced to a target depth of 5 M reads. Short reads were processed as for WGS.

Microbiome profiling

Preliminary analysis was carried out on a subset of samples using MetaPhlAn2 to determine the appropriate sequencing depth. Samples sequenced to > 10 M reads were subsampled to depths of 10 M, 5 M, 3 M, and 2 M reads using seqtk, and 2 M reads was used as a cutoff based on an insignificant difference of species α-diversity (richness and Shannon diversity) between 2 M reads and the original read set. MetaPhlAn4 [36] (v4.0.2) was used to profile the community taxonomic composition (mpa_vJan21_CHOCOPhlAnSGB_202103 database). Community diversity metrics were profiled with phyloseq (v1.41.1). MaAsLin2 [37] (v1.8.0) was used to investigate taxonomic associations with infection status using the formula Taxa ~ Group + (1 | SampleType) with default parameters. Groups included all cases or stratified by the causative species (of those with at least three cases) compared to all controls. To determine if intestinal S. agalactiae was enriched in S. agalactiae UTI, –min_prevalence was lowered to 1% [30]. Respiration mode was initially profiled at the species level using bacdive (https://bacdive.dsmz.de/); unassigned species or those with conflicting bacdive classifications (e.g., both “anaerobe” and “aerobe”) were subjected to further profiling at the species or a higher taxonomic level when applicable (e.g., all Lachnospiraceae are classified as anaerobes) using another source [38] as has previously been done [39]. Species without experimental evidence of respiration mode and those ≤ 1% abundance were filtered out. Obligate aerobes and facultative organisms were grouped as “aerobes,” and obligate anaerobes and aerotolerant organisms were classified as “anaerobes.” shortBRED [40] (v.0.9.4) was used to profile virulence factors (VFs) in metagenomes. A marker set was constructed from the VFDB 2022 [41] core protein sequences and complete UniRef90 [42] database (both downloaded in May 2024) using shortbred_identify.py (–clustid 0.95). VF-encoding loci were manually assigned and filtered to include those encoded by Enterobacteriaceae based on their species attribution. VF abundance was profiled using shortbred_quantify.py with default settings. MaAsLin2 [37] was used to identify VF associations with EBI status using the formula Pathways ~ Group + (1 | SampleType) with normalization set to “NONE.”

Strain-level analysis

BWA (v0.7.17) was used to index isolate genomes and align metagenomic reads in a pairwise all-versus-all manner. Alignments were provided to inStrain [43] (v1.6.3) as a .sam file along with the genome.fasta files, and inStrain profile was used with default parameters to determine strain presence and microdiversity metrics. If multiple extraintestinal isolates were available for an EBI case and were isogenic (> 99.999% ANI), the alignment to the higher quality assembly, as assessed above for SNP analyses, was used. For ease of visualization, population ANI (popANI), single nucleotide substitution (SNS) rate, and breadth were converted to log scale with -log10(1-metric); values of 1 were transformed to infinite values, so they were changed to a value of 7. Strain presence was defined as breadth ≥ 0.5 and popANI ≥ 99.999% [43]. Normalized coverage was calculated as (coverage/metagenomic read depth) * 10⁶. StrainGST from the StrainGE toolkit [44] was used to metagenomically profile E. coli strains. All 2,835 complete E. coli genomes on NCBI RefSeq on August 8th, 2023, were downloaded using ncbi-genome-download [45] (v0.3.1) and chromosomes were retained. Reference genomes were k-merized with the default k = 23 and subsequently clustered into 946 groups using a minimum Jaccard similarity of 0.90, with k-mer profiles from one representative genome from each group selected to form the database. Metagenomic read sets were k-merized and closely matching strains from the database were identified. Phylogroups were assigned to each reference and isolate genome with ClermonTyping [46] and STs were assigned to references using the mlst software as above. The predicted relative abundance (rapct) was summed for each phylogroup per sample; non-zero values were converted to 1 to determine prevalence. Matching E. coli EBI cases were defined as those with the same strain in their gut and cultured from extraintestinal site(s) based on isolate genomics and/or inStrain. Reference strains were annotated as described for isolate genomes, and a core gene alignment was produced via panaroo [47] (v1.2.10) using –core_threshold 0.98. An approximately-maximum-likelihood phylogenetic tree was generated with FastTree [48] (v2.1.10) and visualized using ggtreeExtra [49] (v1.4.2). FastANI [50] was used to assess relatedness between EBI-causing E. coli strains colonizing the gut and reference genomes called by strainGE that had discordant in silico MLST calls. The reference was presumed to represent the EBI-causing strain when it shared ≥ 99.5% pairwise whole genome ANI (average of the two measurements in which either genome was used as the query), a reported threshold for STs [51].

Statistical analysis

We performed statistical analyses in R Studio (R 4.1.0, RStudio 2024.04.1 + 748). We tested numeric distributions for normality using the Shapiro–Wilk test. Non-parametric distributions were compared using a two-tailed Wilcoxon rank-sum test. A PERMANOVA test (adonis2, vegan v2.6.4) was used to compare Bray–Curtis dissimilarity measures between cases and controls. Fisher’s exact tests were used to compare the distribution of categorical data between groups. The Benjamini-Hochberg (BH) correction was used to adjust p values for multiple hypotheses when applicable. Spearman correlations were used to determine the linear correlation between two variables (R) for one group. Linear regression using the lm() function in R was used to compare the relationship between continuous variables when multiple groups were present. p < 0.05 (BH-corrected when applicable) was considered statistically significant. For MaAsLin2 analyses, the default q < 0.25 was considered statistically significant except for the virulence factor analysis, in which q < 0.05 was used to reduce the number of observations for ease of visualization.

Results

The gut contains putatively identical bacteria to those isolated from extraintestinal sites in febrile infants < 60 days of life

Forty infants < 60 days of life with EBIs (cases) from whom EBI isolates were available were selected, with a median age of 31 days of life (Fig. S1, Table S1). UTIs without concomitant bacteremia or meningitis were the most common infection and were identified in 70% of infants with culture-proven EBI (28/40; Table 1). The remaining 12 cases had concomitant bacteremia and UTI (n = 7), bacteremia and meningitis (n = 1), bacteremia alone (n = 3), or meningitis alone (n = 1). Most EBIs were caused by E. coli (26/40), followed by E. faecalis (6/40), S. agalactiae (3/40), K. pneumoniae (2/40), E. cloacae (1/40), P. mirabilis (1/40), or E. coli and S. agalactiae (1/40), similar to previous reports of bacterial pathogens isolated from normally sterile sites in this population [3, 5,6,7] (Table 1).

Table 1 Distribution of EBI type by causative organism. Number of infants with EBIs based on extraintestinal site affected and causative species. Gut colonization is defined as detection via isolate genomics (culturing) and/or metagenomics. ^aone case was detected via metagenomics only

To investigate if the EBI-causing strain was present in the gut at the time of infection, we first screened intestinal samples by selectively culturing for the EBI-causing species (Fig. S2). Across all infants, intestinal samples were collected a median of 1.07 h before extraintestinal samples (Table S1, data file S1). We found at least one swab/fecal isolate to be concordant with the EBI-causing strain in 60% (24/40) of cases using MLST (Fig. 1A, Table 1). Further, 83% (5/6) of infants with UTIs caused by E. faecalis carried the presumptively pathogenic strain in their gut, consistent with an intestinal source of these infections. Of the infants with infections caused by E. coli, which is commonly found in the early-life gut microbiome [52], 56% (15/27) had a strain presumptively matching the pathogen cultured from their intestinal sample. The remaining infants for whom we observed gut pathogen colonization based on MLST had EBIs caused by P. mirabilis (1/1), K. pneumoniae (1/2), or S. agalactiae (2/4). These findings are consistent with the hypothesis that the gut is a reservoir for pathogenic bacteria in young infants [30] based on shared phylogeny of strains.

Fig. 1

High genomic relatedness between cognate extraintestinal and intestinal isolates. A Bar plot depicting the number of extraintestinal isolates that had a strain of the same species and ST cultured from the gut (MLST match) or not (no MLST match) by species and isolate source. B, C Histograms of whole-genome SNP counts computed through pairwise comparisons within (B) Escherichia coli or (C) other pathogen species. Colors indicate whether the SNPs represent comparisons of samples from the same or different infants and whether the STs matched. The dotted line at 12 SNPs indicates the upper limit observed for within-patient ST matches. CSF cerebrospinal fluid, ST sequence type, SNP single-nucleotide polymorphism

To strengthen confidence in the association between intestinal and extraintestinal isolates as inferred by species and MLST, we performed whole-genome sequencing of all isolates and assessed pairwise genomic distances between isolates of the same species. To set a benchmark for genomic relatedness, we included intestinal isolates not matching their cognate extraintestinal strain(s) by MLST and compared strains of the same species between infants. We found that a maximum of 12 whole genome SNPs separated within-infant ST matches (median 1 SNP; Fig. 1B, C, data file S2). This threshold corresponds to an average nucleotide identity (ANI) > 99.999%, indicating high genomic relatedness, i.e., isogenicity [43]. In comparison, the lower limit of SNPs observed between isolates from different infants was 114 for two E. coli ST404 isolates. Based on a stringent 12 SNP threshold, we confirmed that isolates of the same ST cultured from the gut and site of EBI were nearly identical by isolate genomics. However, given that only one colony was screened by MLST per intestinal sample, these findings may underestimate the true rate of gut pathogen colonization in young infants with EBI.

We next characterized the antimicrobial resistance profiles of strains isolated from sterile sites in our EBI cases. Specifically, we were interested in determining susceptibility against agents commonly used for empiric management in febrile term infants (i.e., ampicillin plus either gentamicin or a third-generation cephalosporin, such as ceftriaxone or ceftazidime [9]). 61% (19/31) of extraintestinal Gram-negative strains were resistant to ampicillin, while 13% (4/31) were resistant to gentamicin and 6% (2/31) to at least one third-generation cephalosporin (data file S2). Inferring susceptibility to ampicillin for S. agalactiae [53], across all EBI-causing strains, 90% (37/41) were susceptible to ampicillin and/or gentamicin, while 98% (40/41) were susceptible to ampicillin and/or a third-generation cephalosporin, mirroring results from similar cohorts [54, 55]. Through genome-wide annotation of antimicrobial resistance genes (ARGs), we also identified extended-spectrum β-lactamase (ESBL)-encoding E. coli strains colonizing the gut and causing UTIs in two infants: an ST69 strain encoding bla_CTX-M-14, which had intermediate resistance to ceftriaxone, and an ST131 strain encoding bla_CTX-M-15, consistent with resistance to ceftazidime, ceftriaxone, and cefepime (data file S2). While both strains were susceptible to gentamicin, these concordances indicate gut colonization and EBI caused by ESBL-producing E. coli that displayed resistance to third-generation cephalosporins commonly used in this population.

Gut metagenomic detection of high abundance EBI-causing strains

We next determined if the EBI-causing strains were prominent among the bacterial gut community in these colonized infants. To address this, we performed metagenomic sequencing of intestinal samples from EBI cases and age- and sex-matched febrile infants without EBI (n = 74; Fig. S1, Table S1). As we have done previously [30], we performed metagenomic read mapping using inStrain [43] to resolve strains within the gut microbial community (data file S3). We identified > 99.999% popANI as being specific for within-infant alignments, which indicated that the same EBI-causing strain was detected in the gut metagenome of the cognate infant (Fig. 2A). This exceptionally stringent threshold identified the pathogenic strain in the gut of 75% (18/24) of infants with EBIs for whom the isogenic strain had already been cultured from the gut (Table 1). Additionally, we identified one infant with concomitant E. coli UTI and bacteremia as having the EBI-causing strain present in the gut metagenome despite no E. coli being previously isolated from the intestine. We repeated selective culturing of this sample but were unable to recover E. coli, suggesting that although the organism eluded recovery, its genomic DNA was sufficient to infer presence within the intestinal microbial community.

Fig. 2

Metagenomic read mapping delineates gut pathogen colonization at high abundance in EBI cases. A Genome breadth of coverage (filtered by ≥ 0.5 for visualization) and popANI of pairwise metagenomic read mapping to EBI pathogens. Each dot represents an individual alignment, and colors correspond to whether the metagenome mapped to a genome from the same infant and whether that pathogen had been cultured from the gut (“culture match”). Dot size is scaled to read depth-normalized coverage. Dashed line indicates the minimum popANI (99.999%) that was determined to be specific to within-infant alignments. B Relative abundance of the EBI-causing species compared to normalized coverage for within-infant mappings of metagenomic reads to pathogen genomes for strains determined to be present in the gut via culturing and/or inStrain (as indicated by color). Dashed line indicates the minimum relative abundance (1%) of the pathogen species that enabled metagenomic detection of the EBI-causing strain. Dot size is scaled to popANI. Pearson correlation coefficient (R) and p value are shown

We observed that for infants with the EBI-causing strain detected in their gut metagenome, the fraction of the pathogen genome covered by metagenomic reads (breadth) exceeded that of infants whose EBI-causing strain was cultured from their gut but not detected metagenomically (median = 1.000 versus 0.696, p = 1.547e-4, Wilcoxon rank-sum test). Thus, we reasoned that lower pathogen abundance in these communities may explain why these strains eluded metagenomic detection. Indeed, on average, the EBI-causing species was present at a lower relative abundance in infants whose pathogen strain was detected in the gut solely by culturing compared to those detected metagenomically (median 11.1% versus 0.209%, p = 1.547e-4, Wilcoxon rank-sum test; Fig. 2B). In these seven infants, the EBI-causing species was present at < 1% relative abundance; in two of these cases, the EBI-causing species was present below the limit of metagenomic detection. While higher metagenomic sequencing depth may increase its sensitivity in these cases, four of these seven samples were already sequenced at > 10 million reads, suggesting that routine metagenomic detection is less practical for low abundance (< 1%) strains.

In sum, although isolating the target organism via culture was more sensitive than gut metagenomic sequencing for identifying the pathogen strain in the gut, in 72% (18/25) of cases, metagenomic DNA was sufficient to detect the presence of the EBI-causing organism. By combining both traditional microbial isolation and metagenomic sequencing approaches, we identified the EBI-causing strain in the gut of 63% (25/40) of cases in our cohort (Table 1). Metagenomic read mapping also suggested that, when present, the EBI-causing organism is often at high (≥ 1%) relative abundance, implicating microbiome community composition as a potential signature of EBI in febrile term infants, which we have previously observed with preterm infants [30].

Enrichment of pathogen species observed in the gut of E. coli EBI cases

To determine if a microbiome risk profile of EBI exists at the time of presentation, we compared the gut microbial community structure of EBI cases to those of controls. We observed that infants with EBI had on average higher serum procalcitonin (p = 0.005) and white blood cell counts (p = 0.032) than controls, consistent with prior reports [56, 57] and confirming systemic differences (Table S1). However, α-diversity of the gut microbiome, measured by richness and Shannon diversity at the species level, did not differ significantly between cases and controls (Fig. 3A, B). Similarly, we did not observe differences in β-diversity as measured by Bray–Curtis dissimilarity (PERMANOVA, p = 0.360; Fig. 3C).

Fig. 3

Microbiome compositional differences are age- and infection-dependent for infants with E. coli EBI. α-diversity as measured by A species richness and B Shannon diversity from the gut metagenomes of all EBI cases compared to control infants. P values determined through a Wilcoxon rank-sum test. C Principal coordinate analysis (PCoA) plot of Bray–Curtis dissimilarity between samples computed at the species level. PCoA axes have the percent variation explained shown parenthetically. Marginal x- and y-axis boxplots correspond to PCoA axis 1 and 2 values, respectively. D Association between E. coli relative abundance and E. coli EBI status compared to control infants as determined through generalized linear mixed effect models. The BH-adjusted p value (q) and coefficient of effect (coef) are shown. E E. coli relative abundance plotted by day of life for infants with E. coli EBI compared to controls with linear regression lines plotted for each group. The shaded region represents the 95% confidence interval

There were no overall taxonomic associations between EBI cases and controls using generalized linear mixed effect models, prompting us to stratify cases based on the species causing infection (for groups with ≥ 3 infants). Infants with EBIs caused by E. coli had greater relative abundance of E. coli in their intestine compared to control infants (q = 0.056, coefficient = 2.926; Fig. 3D, Table S2, data file S3). Likewise, infants with S. agalactiae UTI had significantly higher abundance of the EBI-causing organism detected in their gut metagenome (q = 4.449e-4, coef = 1.608; data file S3). However, this association was driven by the presence of S. agalactiae in only one of four S. agalactiae UTI cases (Table S2). Interestingly, we did not observe a significant enrichment of E. faecalis in the gut microbiome of infants with E. faecalis UTIs, despite the pathogenic strain having previously been cultured from the gut of 5/6 cases (Table 1). We also found that E. coli and P. mirabilis were more prevalent in the gut metagenome of infants with EBI caused by those organisms than controls (p = 2e-4 and p = 0.04, respectively; Table S2). Additionally, a relative abundance threshold of ≥ 1% E. coli was 74% sensitive and 57% specific for E. coli EBIs, highlighting its potential utility during clinical risk assessment. In sum, pathogen presence and enrichment in the gut was evident among certain EBI-causing organisms, suggesting that it may be related to infection risk in young infants.

Given that facultative bacteria, including E. coli, decline in relative abundance in the gut in the months after birth [16, 17], we postulated that there may be an age-related component to its enrichment in the gut of E. coli EBI cases. We compared the relationship of E. coli relative abundance with day of life for E. coli EBI cases and controls using linear regression. Indeed, there was a significant difference in this association between groups (p = 0.029), with E. coli EBI cases exhibiting a greater age-related decrease in E. coli relative abundance than controls (β = − 0.673 vs. β = − 0.155, Fig. 3E). We predicted that control infants may have greater α-diversity than infants with E. coli EBIs, which may protect against a bloom in the EBI-causing organism, especially in the youngest infants, when the difference in E. coli relative abundance is most apparent. There was a greater age-related increase in Shannon diversity (p = 0.020), but not species richness (p = 0.123) for infants with E. coli EBIs relative to control infants (Fig. S4A, B), suggesting that microbiome complexity may contribute to susceptibility to E. coli infections in this age group. Because we did not find differences in specific taxa other than E. coli associated with E. coli EBI, we assessed if controls had increased obligate anaerobic taxa, which could contribute to the exclusion of facultative bacteria such as E. coli. As expected, across both E. coli EBI cases and controls, aerobes (including strictly aerobic and facultatively anaerobic organisms) were depleted over time (R = − 0.280, p = 0.005), while the abundance of obligate anaerobes increased (R = 0.260, p = 0.009; Fig. S4C, data file S4). However, there were no significant differences in these rates between E. coli EBI cases and controls (p = 0.981 for aerobes, p = 0.533 for anaerobes). Hence, while a bloom in E. coli abundance in the gut was associated with E. coli EBI in an age-related manner, this could not be explained by a depletion in anaerobic taxa, leading us to consider whether there were strain-specific features associated with EBI susceptibility.

Phylogroup B2 is overrepresented in infants carrying EBI-causing E. coli

Although E. coli is a common cause of EBI, most gut-resident E. coli do not cause EBI. Based on this reasoning, we hypothesized that strains found in the gut of E. coli EBI cases may differ genetically from commensal E. coli found in controls. To test this hypothesis, we used the Strain Genome Search Tool (StrainGST) [44] to identify close representatives to a database of publicly available E. coli reference genomes and estimate their relative abundance based on metagenomic reads. Fifty-nine distinct E. coli genomes from phylogroups A, B1, B2, C, D, and F represented E. coli strains present across the 49 controls and 26 E. coli EBI cases whose samples had detectable E. coli by StrainGST (Fig. 4A, data file S4). Phylogroup B2 was most prevalent at 59% (44/75), but there were no statistically significant differences in phylogroup prevalence between groups (Table S3). When comparing relative abundance, however, phylogroup B2 was significantly more abundant in the stools of E. coli EBI cases who had a strain matching the EBI-causing isolate detected within their gut (“matches”) compared to those who did not (“non-matches”) or control infants (BH-corrected p = 0.042, Fig. 4B). We did not observe significant differences in abundance for other phylogroups (Fig. S5).

Fig. 4

Strain-level signatures of E. coli EBI matches in the gut. A Cladogram of all E. coli reference genomes for which a related strain was identified in at least one E. coli EBI case or E. coli-colonized control. Color strip denotes STs identified in at least three infants (the remainder are grouped as “other”). The bar chart indicates the frequency at which each reference genome was identified for E. coli EBI cases with a matching strain in their gut (match), those without a matching strain identified (non-match), or E. coli-colonized control infants. B Relative abundance of E. coli phylogroup B2 between groups as in A. C Virulence factor (VF) genes encoded by Enterobacteriaceae associated with E. coli EBI matches compared to E. coli-colonized controls as determined through generalized linear mixed effect models. Points are sized by their BH-adjusted p value (q) and colored by their VF functional category

We next asked if the EBI-causing strain was over-represented among E. coli in infants colonized by multiple strains [52]. This disproportion could provide evidence that abundance plays a key role in seeding infection; alternatively, EBI-causing strains may be more adept colonizers in the gut, or more capable of systemic invasion or retrograde movement through the urethra. Multi-strain colonization (defined as ≥ 2 STs predicted to be present) was less common than single-strain gut colonization, occurring in 33% (4/16) of E. coli EBI matches, 50% (5/10) of non-matches, and 18% (9/49) of controls colonized by E. coli; these rates did not significantly differ between groups (p > 0.050, Fisher’s exact test). Four E. coli UTI matches were co-colonized by multiple strains, including the EBI-causing strain, and in three of these infants, a reference genome presumed to be representing the EBI-causing strain had the greatest relative abundance. Due to the low incidence of co-colonization events, however, we were underpowered to draw conclusions about their relationship to EBI risk. Interestingly, in 15/16 E. coli EBI match cases, the presumptively pathogenic strain was estimated to be at high abundance (≥ 1% of all bacteria present), suggesting that increased abundance may facilitate infection. In sum, our metagenomic strain profiling results demonstrate that both the E. coli abundance and lineage are associated with concomitant gut colonization and invasive infection, identifying strain-level microbiome signatures of E. coli EBI in young infants.

Virulence factor signature in infants with gut colonization of EBI-causing E. coli

We predicted that the stools of E. coli EBI matches would contain a higher density of Enterobacteriaceae virulence factors (VFs) than E. coli-colonized controls, which may indicate strategies by which these pathogenic E. coli strains survive in multiple habitats. To test this hypothesis, we profiled the gut metagenomic VF composition of E. coli EBI matches and E. coli-colonized controls using a core set of experimentally verified VF genes encoded by Enterobacteriaceae [41]. The richness and abundance of these VF genes was not different between E. coli EBI matches and E. coli-colonized controls (p = 0.066 and p = 0.053, Wilcoxon rank-sum test; Fig. S6A, B). When comparing the abundance of specific genes, 11 VFs of the 225 queried were positively associated with E. coli EBI matches (q < 0.050, Fig. 4C). These included genes implicated in adherence (papB, papI, papX, draA, draD, draP), exotoxins (clbB, clbD, clbL, senB), and nutritional/metabolic factors (iucB), each of which have been associated with virulence in extraintestinal pathogenic E. coli [58,59,60,61]. To ensure that these associations were not merely due to differences in E. coli abundance, we compared gene prevalence between groups and confirmed that all 11 VF genes were more frequently found in the metagenome of E. coli EBI matches compared to E. coli-colonized controls (BH-corrected p ≤ 0.012 for all enriched VFs, Fisher’s exact test; Table S4). These findings point to a potential role for certain VFs in enabling E. coli strains to both inhabit the gut and infect normally sterile sites in young infants, which may heighten EBI risk in infants carrying those VFs in their metagenome.

Discussion

We and others have shown that gut pathogen colonization precedes bacteremia in hospitalized preterm (< 37 weeks gestational age) infants [30, 62, 63], and that the relative abundance of the bacteremia-causing strain in the gut microbiome of these infants increases immediately before infection onset [30]. However, hospitalized preterm infants are more commonly exposed to antimicrobials and have less developed gut physiology than term infants, which increases their risk of bacterial translocation into sterile sites [21, 64]. We extend these data to show that gut pathogen colonization is common among febrile term infants with EBIs who are otherwise healthy, supporting the concept that the intestine is a pre-dissemination habitat for these pathogens, which has previously been reported at lower molecular resolution [65]. Our observation that the EBI-causing organism was often at sufficient abundance (≥ 1%) to be detected metagenomically suggests that these strains may be resident, rather than transient, colonizers of the infant gut.

The human digestive system contains a microbiome that is highly dynamic in the months after birth. Aerobic and facultatively anaerobic organisms first seed this organ, followed by a gradual increase in diversity and strictly anaerobic organisms, such as bifidobacteria, as the community matures [16, 17]. However, perturbations to the infant gut microbiome may be associated with community-acquired EBIs, as others recently reported lower Bifidobacterium and higher Bacteroides abundance in otherwise healthy infants with late-onset sepsis compared to afebrile controls [66]. We observed that infants with E. coli EBIs had higher prevalence and relative abundance of E. coli in their gut compared to febrile control infants, consistent with previous reports of increased bacterial 16S rRNA reads mapping to either Escherichia or E. coli in older pediatric populations with UTI caused by E. coli [67, 68]. Our identification of E. coli relative abundance ≥ 1% as a risk factor for E. coli EBI should be validated in additional cohorts, as E. coli is the most common organism causing EBI in this population [9], and this finding may serve as a useful biomarker as microbiome profiling enters the clinical sphere [69]. Unlike a recent study that demonstrated that the gut microbiome of older infants (3–11 months of life) with E. coli UTI had reduced α-diversity compared to healthy controls [67], our cases did not have lower diversity compared to age-matched febrile infants without EBI. This difference may reflect the relatively low complexity of the gut microbiome in the first 2 months of life or differences in the selection of comparison infants. Specifically, the febrile state or the presence of another pathogen that caused the fever might temporarily diminish diversity. Our results support the concept that a relatively high abundance of pathogenic E. coli and/or VF carriage may facilitate dissemination to extraintestinal sites. Nonetheless, we do not find that microbiome diversity was associated with protection from EBI in these infants, at least when compared to controls who were studied during a febrile illness not attributable to EBI.

Metagenomic strain profiling demonstrated strain-specific attributes of E. coli inhabiting both the gut and extraintestinal site(s). Our findings support a role for increased relative abundance of E. coli, specifically the virulence-associated phylogroup B2 [70], with the onset of E. coli EBI in young infants. Phylogroup B2 is the most persistent E. coli lineage in the normal infant microbiome [71]. While these strains commonly have a wide VF repertoire [71], we found that certain VF genes were more frequently present in infants with E. coli EBIs who were colonized with the EBI-causing strain than E. coli-colonized controls. P (pap) and Dr (dra) fimbriae promote E. coli adhesion in extraintestinal sites [60], which may facilitate their persistence in the urinary tract. Similarly, the presence of machinery encoding the siderophore aerobactin (iuc) suggests that these strains may have adapted to survive in varied environmental conditions, such as those encountered in nutrient-limited extraintestinal sites. Interestingly, a set of genes involved in the synthesis of colibactin (clb) were more commonly found in infants colonized by the EBI-causing E. coli strain. Colibactin is a genotoxin that has been shown to contribute to both E. coli gut colonization and systemic infection in neonatal mice [59]. These functional associations potentially explain the difference in capacity to cause EBIs and suggest the likely sequence of dissemination (gut to bladder to bloodstream vs. gut to bloodstream). Future studies will need to determine if these VFs are involved in the seeding of E. coli EBI from the intestine and whether these may serve as therapeutic targets.

We note several important limitations of our findings. This study was cross-sectional in design and could not ascribe a causative role for the microbiome in seeding infection. Although a less likely explanation, our data cannot rule out the possibility that gut pathogen presence in stool represents seeding from an extraintestinal source, including the infected blood or bladder. Despite this possibility, longitudinal studies using strain tracking have identified the gut as a reservoir for EBI in other populations, including bloodstream infections in preterm infants [30, 62, 63] and hematopoietic cell transplantation recipients [72, 73] and UTIs in adults [74,75,76]. Future longitudinal cohort studies will be essential to confirm whether the gut is also a reservoir for organisms causing EBI in term infants. As mentioned above, our control group consisted of febrile infants without EBIs who may have instead had viral infections, which have been associated with a lower risk of bacterial co-infections [27]. Additionally, we cannot rule out the possibility that false negative blood cultures occurred in our control group due to infection with organisms that evaded culturing, which is a limitation of current diagnostics. Despite this possibility, our results present strong evidence for gut pathogen colonization in a subset of infants who also had EBI from those organisms, which we predict may extend to organisms not represented in our study. While we excluded infants with very recent (preceding 48 h) antimicrobial exposure, we did not account for exposures that are known to impact microbiome development in early life, including earlier antimicrobials, other medications, diet, and birth mode [77]. However, given the association between E. coli abundance and EBI, we propose that microbiome-targeting interventions, including breastfeeding [78], may provide a protective role by supporting the growth of non-pathogenic organisms. Our work also raises the possibility of screening for carriage of specific organisms (species and sub-species genotypes) with high risk of invasion in infants in the first 60 days of life, separate from the microbiome. The model for such an approach is prenatal screening for S. agalactiae carriage. However, demonstrating the efficacy of such targeted interventions against single pathogens would require considerable resources and very large cohorts.

Conclusions

Although many EBI-causing bacterial species colonize the healthy infant gut microbiome, their association with EBIs remains understudied. In this cohort of full-term infants younger than 60 days old presenting with fevers, we demonstrated that 63% of febrile infants with EBIs had the EBI-causing strain present in their gut at the time of presentation. We also identify age- and organism-dependent risk factors of EBI present in the microbiome, including several VFs associated with gut carriage of EBI-causing E. coli. In summary, we offer evidence that the intestine is a reservoir for EBI-causing pathogens in young full-term infants. Our work can inform future clinical studies and animal experiments to test the role of microbiome- and possibly pathogen-directed interventions in lowering the risk of EBI in young infants.