Angkor Hospital for Children is a non-governmental pediatric referral hospital with a dedicated neonatal intensive care unit (NICU) and special care baby unit (SCBU). The hospital has no maternity unit, thus all neonatal admissions are outborn. Over a 1 year period all infants aged ≤28 days were eligible for study enrolment on admission to the NICU or SCBU, excluding those who had any previous healthcare service exposure following delivery (e.g., overnight admission to a hospital/health center or transfer from another AHC ward).

Mothers provided written informed consent to participate in study. Those unable to read and write gave a thumbprint instead, with an impartial witness present during the consent process. Mothers consented on behalf of infants. A rectal swab was taken from the infant within 24 h of admission to the neonatal unit (at a median of 11 days of age [range 0–30]), and a maternal stool sample was collected as soon as possible after infant admission. Samples were cultured onsite on selective chromogenic ESBL detection agar (CHROMagar ESBL medium; CHROMagar, France) and all morphotypes followed up for analysis. Species identification was confirmed by standard biochemical tests and antimicrobial susceptibilities were determined by disk diffusion, following CLSI methodology and using 2017 breakpoints (CLSI, 2017). Double disk diffusion tests were done to confirm ESBL production (cefotaxime and ceftazidime with/without clavulanate; BBL, Becton Dickinson, United States).

96 isolates of E. coli and K. pneumoniae were cultured from frozen stocks (−80°C) and grown in LB broth at 37°C for 3 h. 900 μL of bacterial culture was spun down at 21,200 rcf for 5 min before the supernatant was removed. The bacterial cells were lysed with 600 μL of nucleic lysis solution (Promega, United States) at 80°C for 10 min. The cell lysate was then transferred to a new tube containing 250 μL of glass beads with 5 μL of RNase A solution (Qiagen, Germany), and the cells were further lysed by vortexing for 10 min. The cell lysate was incubated at 37°C for 30 min to remove the RNA. The mixture was purified using magnetic beads (Agencourt AMPure XP beads, Beckman Coulter, United States) with a 1:4 volume ratio of lysate mixture and magnetic beads. The extracted DNA was redissolved in distilled sterilized water and then stored in the freezer (−20°C) prior to library preparation for sequencing.

Sequencing libraries were prepared from extracted DNA using Nextera XT library preparation kits (FC-131-1096, Illumina, United States) and Nextera XT Index Kits 96 samples (FC-131-1002) and sequenced on an Illumina MiSeq with 300 bp paired-end reads. 36 isolates were selected for long-read sequencing. The Rapid Barcoding Kit (SQK-RBK004, Nanopore, United Kingdom) was used to prepare DNA libraries according to the Nanopore protocol with 12 barcoded samples per flow cell. The prepared libraries were pooled and purified using an additional step with magnetic beads (Agencourt AMPure XP beads, Beckman Coulter, United States) using a 1:1 volume ratio of pooled libraries and the magnetic beads. The purified libraries were then sequenced on a MinION R9.4 flow cell according to the Nanopore protocol. The sequences were deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under project accession number PRJEB37551 https://www.ebi.ac.uk/ena/browser/view/PRJEB37551.

Raw sequence data were processed using the Bactopia pipeline 1.4 (Petit and Read, 2020). Reads were mapped against E. coli K-12 substrain MG1655 reference sequence for all Escherichia isolates¹ and K. pneumoniae HS11286² for all Klebsiella isolates, and the genome assemblies were also used as references for confirmation of transmission. Mashtree 1.1.2 (Katz, 2021) was used to generate phylogenetic trees from all samples and visualized using ggtree 3.1.2 (Yu et al., 2017). Ska 1.0 (Harris, 2018) was used to perform split k-mer analysis using the default cutoffs (95% kmer identity and < 20 SNPs to call transmission clusters). Snippy 4.6.0 (Seemann, 2018) was used to produce variant calls, and snippy-core was used to mask repetitive regions and produce a core SNP alignment. SNP distances between samples were called with snp-dists 0.6.3 (Seemann, 2021a). MLST was determined using mlst 2.19.0 (Seemann, 2021b) and the PubMLST database (Jolley and Maiden, 2010). Genome assemblies were performed using Unicycler 0.4.8 (Wick et al., 2017). Escherichia species and phylogroups were determined using ClermonTyping (Beghain et al., 2018). Klebsiella phylogroups and virulence genes were obtained through Kleborate (Lam et al., 2021). Antimicrobial resistance genes and mutations were determined using NCBI AMRFinder+3.8.4 (Feldgarden et al., 2019), and upset plots were visualized using ComplexUpset (Krassowski, 2020). Platon 1.5.0 (Schwengers et al., 2020) was used to identify contigs originating from plasmids and the mobilization proteins, replication proteins, and incompatibility groups on each contig. The Platon identification was used to determine whether the AMR genes were part of the chromosome or present on plasmids. Flanker 0.1.5 (Matlock et al., 2021) was used to extract 20 kb genomic regions surrounding AMR genes, treating plasmids as circular if the Unicycler assembler reported them as circular. Isolates were removed if the AMR gene was on a contig less than 5 kb in length. Clinker (Gilchrist and Chooi, 2021) was used to visualize the regions, with links drawn between genes with 95% or greater identity.

We sequenced 96 isolates in total on Illumina MiSeq—68 isolates from 20 of the 21 pairs where both mother and infant were colonized, including eight of nine pairs considered as potential transmission pairs (one pair could not be retrieved for sequencing), and 28 other randomly selected isolates from the study. We sequenced up to four isolates per individual based on the number of distinct morphotypes identified during bacterial growth (Supplementary Table 2).

We performed long-read sequencing on the 18 isolates identified as possible transmission pairs based on the phenotypic data, as well as 18 other isolates which were observed to have closely related core genomes or similar patterns of antimicrobial resistance genes based on Illumina data. We used hybrid assembly of the long and short-read data to obtain complete genomes and resolve the plasmids in these strains in order to confirm transmission events and identify whether these transmission events involved complete isolates, or plasmid horizontal transmission between different isolates. We obtained between 1 and 9 contigs for each of the 36 isolates, including one contig of 4.5 Mb or larger assumed to be the chromosome, and 0–8 plasmid contigs per isolate.

The species of each isolate was determined from the whole genome sequencing using ClermonTyping (for Escherichia species) and Kleborate (for Klebsiella species). This showed that the K. pneumoniae isolates included three isolates of K. quasipneumoniae subsp. similipneumoniae and one isolate of K. quasipneumoniae subsp. quasipneumoniae, and the E. coli isolates included three isolates of E.fergusonii and two isolates of E. clade-1 (Figure 2).

No phylogroups or MLST were predominant in the isolates. Among E. coli isolates, the most common MLST was ST38 and the closely related ST3268 and ST3052 and ST318, as well as three isolates from the ST131 pandemic clade (Supplementary Figure 1). Of the three ST131 isolates, two were clade C and one was clade A. In the K. pneumoniae isolates there were no dominant sequence types, with ST17 the only sequence type present more than once in unpaired isolates, and four isolates had novel sequence types (Supplementary Figure 2). No virulence determinants were found in the K. pneumoniae isolates and all had a virulence score of 0 as determined by Kleborate.

95 of 96 isolates had an ESBL gene present in the genome, as expected as they were selected for the presence of ESBLs. One isolate (EbB064) had no ESBL genes present in the sequenced genome, despite having phenotypic resistance to third generation cephalosporins. Four K. pneumoniae and one K. quasipneumoniae subsp. similipneumoniae isolates had bla_SHV-2A genes predicted to cause ESBL resistance, with two cases of bla_SHV-2A carried on a plasmid, one case where the gene is chromosomal, and one pair of isolates carrying the gene on both a plasmid and in the chromosome. Multidrug resistance was extremely common in these isolates, with 64% (61/96) of isolates genotypically resistant to six antimicrobial classes, including two isolates resistant to eight classes (Figure 3). No carbapenemase resistance was found in this dataset. Colistin resistance was observed through presence of mcr-1.1 and mcr-3.5 genes in three E. coli isolates. A point mutation of pmrB (R256G) was reported in three K. pneumoniae isolates but this has been previously demonstrated not to cause colistin resistance in isolation (Cheng et al., 2015) and we did not consider these isolates resistant. AMRfinder analysis reported resistance to phenicol/quinolone via oqxAB and resistance to fosfomycin via fosA in K. pneumoniae, but Kleborate confirmed these were intrinsic and no acquired resistance was reported, and so these isolates were not recorded as resistant through presence of these genes (Lam et al., 2021).

A wide range of beta-lactamase genes were carried, including nine different bla_CTX-M genes, and genes from the bla_CMY, bla_OXA, bla_DHA, bla_SHV, and bla_VEB families (Figure 4). The most common were bla_CTX-M-15, found in 31 isolates, and bla_CTX-M-55, found in 24 isolates. 25 isolates had an ESBL gene in the chromosome, while seven isolates had ESBL genes present on plasmids as well as integrated into the chromosome.

For each potential pair of isolates from a mother-infant pair, we identified potential transmission events as either transmission of a strain with complete genome (chromosome only, or chromosome and plasmid), transmission of plasmids, or no transmission.

For eight of the nine pairs which had the same antibiogram in both strains, the mother and infant strains shared the same MLST and the same set of acquired antimicrobial resistance genes, and a further pair of isolates differed only by one antimicrobial resistance gene. We used split k-mer analysis to identify transmission clusters in the complete dataset, and confirmed this using the complete genome assembled from one of the paired strains as a reference for mapping and variant calling. We removed variant calls in repetitive regions but did not correct for recombination as we assume no recombination will occur in the timeframe of maternal transmission.

Eight of these nine pairs were identified as transmission clusters using ska with default filters (95% kmer similarity and a SNP distance threshold of 20), and had 0–4 SNP differences between them (Figure 5). To confirm these results, we used one strain from each pair as a reference genome, and performed mapping and variant calling. Mapping to one pair as a reference gave similar results in all cases with 0–4 SNP differences from the mapping-based approach. We identified one pair where complete strains of both E. coli and K. pneumoniae were transmitted.

We also looked at two other pairs of isolates from the whole collection which were identified as a possible transmission cluster by ska (Figure 5). In one case, the two isolates (EbB096 and EbB097) were from the same individual, and likely represent the same isolate which was sequenced twice due to appearing as different morphotypes during bacterial growth. The other pair of isolates (EbB002 and EbB018) were identified as a transmission cluster by ska, but with 14 SNP differences between the isolates. Mapping to one of the paired isolates as a reference gave similar results, with 13 SNP differences between isolates. This pair of isolates came from an unpaired mother and infant, and were collected 54 days apart.

For each of the transmission pairs we performed whole genome alignment using minimap to determine if the transmission included plasmids. Seven of the eight transmission clusters included plasmids, with only one pair where only the chromosome was present in both assemblies. While the plasmids appear to be transmitted in all strains, in three of the eight transmitted pairs there were rearrangements between plasmids. In one pair (isolates EbB065 and EbB066), one isolate had two circular contigs of 280 and 95 kb, while the other isolate has only one non-circular contig of 372 kb, which likely represents a misassembly of two plasmids into one contig (Supplementary Figure 3). In another pair (EbB043 and EbB045), an 88 kb non-circular contig from one isolate is present in the chromosome of the paired isolate, representing either a misassembly, or potentially insertion into the chromosome (Supplementary Figure 4). In a third pair (EbB104 and EbB105), the same plasmid material is assembled into a single plasmid in both strains, but with rearrangements between the plasmids, including a tandem duplication of the bla_CTX-M-65 gene seen in two copies in EbB104 and three copies in EbB105 (Supplementary Figure 5).

We also considered the possibility that maternal transmission of plasmids could occur in the absence of transmission of a complete strain, which would not be discovered using mapping-based approaches but may be identified by similar antimicrobial resistance profiles. To check for plasmid-borne resistance genes shared between strains, we compared the antimicrobial resistance profiles of all strains. There were no cases of identical or near-identical antimicrobial resistance profiles in paired strains which do not have similar core genomes. We did identify three clusters of unpaired samples with identical or near-identical antimicrobial resistance profiles which do not have similar core genomes: EbB004 and EbB039, EbB031 and EbB041, and EbB036/EbB037 and EbB091.

Strains EbB004 has a 127 kb plasmid carrying incFII and incFIB, while EbB039 has a 221 kb plasmid with incFII, incFIB, col156 and P0111 replicons. The two plasmids shared a 50 kb segment with an internal inversion, including the bla_CTX-M-27 gene, and eight other genes conferring resistance to aminoglycosides, trimethoprim, macrolides, sulfonamides, and tetracycline.

Strains EbB031 and Eb041 both belong to ST44 and have a SNP distance of 349 measured by ska. They also share a 170 kb plasmid with a similar resistance cassette carrying both bla_CTX-M-15 and bla_OXA-1, as well as resistance to aminoglycosides, trimethoprim, macrolides, sulfonamides, tetracycline, and chloramphenicol.

Strains EbB036 and EbB037 are a transmission pair which share an antimicrobial resistance profile with strain EbB091. However, none of these strains contain plasmids, and the shared resistance profiles are driven by chromosomal qnrS1 and bla_CTX-M-15 integrated in different locations.

Finally, we looked at the genomic context of ESBLs found in more than one strain to determine if shared mobile elements were driving the spread of ESBLs in this dataset. Observing the 20 kb regions around each ESBL, many of the ESBL genes were present in different contexts and were not present on shared plasmids or as shared mobile elements integrated into plasmids or chromosomes.