Characterization of antimicrobial-resistant Gram-negative bacteria that cause neonatal sepsis in seven low- and middle-income countries

Antimicrobial resistance in neonatal sepsis is rising, yet mechanisms of resistance that often spread between species via mobile genetic elements, ultimately limiting treatments in low- and middle-income countries (LMICs), are poorly characterized. The Burden of Antibiotic Resistance in Neonates from Developing Societies (BARNARDS) network was initiated to characterize the cause and burden of antimicrobial resistance in neonatal sepsis for seven LMICs in Africa and South Asia. A total of 36,285 neonates were enrolled in the BARNARDS study between November 2015 and December 2017, of whom 2,483 were diagnosed with culture-confirmed sepsis. Klebsiella pneumoniae (n = 258) was the main cause of neonatal sepsis, with Serratia marcescens (n = 151), Klebsiella michiganensis (n = 117), Escherichia coli (n = 75) and Enterobacter cloacae complex (n = 57) also detected. We present whole-genome sequencing, antimicrobial susceptibility and clinical data for 916 out of 1,038 neonatal sepsis isolates (97 isolates were not recovered from initial isolation at local sites). Enterobacterales (K. pneumoniae, E. coli and E. cloacae) harboured multiple cephalosporin and carbapenem resistance genes. All isolated pathogens were resistant to multiple antibiotic classes, including those used to treat neonatal sepsis. Intraspecies diversity of K. pneumoniae and E. coli indicated that multiple antibiotic-resistant lineages cause neonatal sepsis. Our results will underpin research towards better treatments for neonatal sepsis in LMICs.


Results
Enrolment in BARNARDS and isolation of pathogenic bacteria. The numbers of neonates recruited, clinically diagnosed with sepsis and with a confirmation of sepsis by positive blood culture are outlined in Fig. 1. Of 36,285 infants ≤60 d old (termed herein as neonates) enrolled in the BARNARDS study from November 2015 to December 2017, 2,483 had culture-confirmed sepsis. All 12 clinical sites used the same criteria for clinical diagnosis of sepsis ( Supplementary Fig. 1). We found that cases were mainly EOS for both sites in Pakistan, one site in Bangladesh (BC; see Methods for definitions of all two-letter site abbreviations) and the single site in Ethiopia. In Nigeria, India and South Africa, there were mainly LOS cases. In Rwanda, neonatal sepsis cases were equally split between EOS and LOS (Fig. 2).
Automated blood culture systems were used to detect microbial growth, with 2,620 microbial isolates recovered. These 2,620 isolates comprised 1,266 Gram-positive bacteria (GPB) isolates, 1,038 GNB isolates, 22 fungal isolates and 294 unassigned isolates (Fig. 1). The methods for collection and identification were standardized across all sites, with equipment and reagents purchased from uniform suppliers. The primary aim of the BARNARDS study was to characterize the extent of β-lactam resistance in GNB causing clinically diagnosed sepsis in infants <60 d old (see Supplementary  Figs. 2 and 3). However, at month 17 (during a BARNARDS network event), we anecdotally noted high rates of isolation of Staphylococcus species and therefore collected all of these isolates for analysis (to be reported elsewhere). Of 1,038 isolates, 916 GNB were analysed using WGS. For 122 isolates, identification beyond a Gram stain was not possible because the isolate was lost and/or purification for DNA extraction was unsuccessful (Fig. 1).
In total, 58 different species of GNB were identified across all sites by WGS (Supplementary Table 1  species (n = 36) and Ralstonia mannitolilytica (n = 21) (Fig. 2). Among the GNB characterized herein, 401 were Klebsiella, with six species identified: K. pneumoniae, K. variicola, K. quasipneumoniae, K. aerogenes, K. oxytoca and K. michiganensis ( Supplementary  Fig. 4). Burkholderia cenocepacia, K. michiganensis, R. mannitolilytica and S. marcescens were mainly isolated from samples obtained from single sites in Pakistan, Nigeria and Bangladesh (Fig. 2). Overall, R. mannitolilytica, K. michiganensis, Burkholderia species and Pseudomonas species caused more cases of EOS than other species (Supplementary Table 2 and Supplementary Fig. 5). Of note, similar proportions of fatal sepsis cases were caused by GNB on each continent (92/353 cases in Africa (21%) and 118/475 cases in Asia (20%)). We found that Burkholderia species or K. michiganensis sepsis infections were the most likely to be fatal (Supplementary Table  2 and Supplementary Fig. 5). However, there was a large proportion of missing data from certain clinical sites (due to data collection and entry error; Fig. 1), which limited the strength of conclusions.

Interspecies and intraspecies diversity across clinical sites.
To understand the extent of inter-and intraspecies diversity, we aimed to perform WGS on all GNB. Multilocus sequence typing (MLST) was used primarily as an unambiguous tool to examine bacterial relatedness. As MLST can be performed easily via PCR of housekeeping genes, in addition to in silico MLST via WGS, assessing relatedness via this tool (where applicable) provides a dataset that can be extrapolated to existing data on neonatal sepsis where WGS may not be available.
MLST revealed substantial intraspecies diversity, and 40 previously unknown STs were assigned in 12 species (Table 1 and  Supplementary Table 2). Fourteen STs were assigned in the Klebsiella genera (Table 1), including all three K. aerogenes STs (all from Africa; n = 2 from Nigeria and n = 1 from South Africa; ST194-196). K. michiganensis was mainly ST180 (from PP in Pakistan; Table 1). Such low ST diversity and close phylogenetic relatedness, as shown by the core genome phylogeny ( Supplementary Fig. 6), warrant further investigation. Similarly, we noticed that a single, previously unknown, B. cenocepacia ST (ST1621), also from PP, was dominant (Table 1 and Supplementary Fig. 7). B. cenocepacia ST1621 from PP and the S. marcescens isolates from BC were indistinguishable during core genome analysis (Supplementary Figs. 7 and 8). All local-level clusters will be studied further.
Enterobacter cloacae complex isolates were identified belonging to E. cloacae, E. hormaechei, E. kobei, E. asburiae and E. ludwigii. In total, seven different Enterobacter species (n = 80) with 28 STs were identified. The majority of Enterobacter species were found in Pakistan (n = 39), Nigeria (n = 19) and Rwanda (n = 14) ( Fig. 1 and Supplementary Fig. 9). ST171 was common across sites in Africa and Asia; however, ST346 was only detected in Rwanda, ST523 was only detected in Pakistan and ST850 was only detected in Nigeria. Within Enterobacter species, 13 STs were assigned (Table 1) to E. cloacae, E. hormaechei and E. ludwigii isolates.
Acinetobacter species were recovered from ten out of 12 clinical sites in both Africa and South Asia, and 38 out of 49 (78%) were Acinetobacter baumannii. Of these, 17 out of 38 (45%) belonged to international clones (Pasteur MLST) ST1 and ST2 (Supplementary  Tables 1 and 2).  The study sites are indicated by coloured squares. The African sites were located in Ethiopia (ES (green)), Nigeria (NK (cyan), NN (light blue) and NW (dark blue)), Rwanda (RK (dark purple) and RU (light purple)) and South Africa (ZAT; olive). The Asian sites were located in Bangladesh (BC (dark pink) and BK (light pink)), India (IN (orange)) and Pakistan (PC (peach) and PP (burgundy)). The numbers next to each clinical site location represent the total number of GNB identified. Inset: the stacked bar graph shows the distribution of the top ten GNB species recovered from blood cultures at the local sites. The onset of neonatal sepsis (EOS, LOS or ND) for GNB per clinical site is represented as a pie chart. The outcome of neonatal sepsis is shown for each continent.
treatments for neonatal sepsis, as well as carbapenems (the incidence of carbapenem-resistant GNB is increasing at an alarming rate), on 885 GNB (31 isolates were not recovered following storage at −80 °C after genomic DNA extraction for WGS; Supplementary Table 2 lists the isolates recoverable for WGS and/or MIC testing). Current data in LMICs focus on profiling AMR within certain species only, and are therefore not exhaustive across all pathogens causing sepsis.
Overall, GNB isolates resistant to at least one of the cephalosporins tested were less likely to cause LOS than EOS (P = 0.017; odds ratio (OR) = 0.63; 95% confidence interval (CI) = 0.43-0.92; Supplementary Table 3). For the statistical analysis, the outcome measurement was MFBS (deceased as response, alive as reference). Concomitant resistance to the three cephalosporins tested versus isolate susceptibility to all produced an odds ratio of 0.626; 95% CI 0.426-0.918; P = 0.017. In this way, concomitant resistance to the three cephalosporins tested among GNB isolates was less likely among infants who stayed alive compared to those who were deceased.
The total number of ARGs possessed by each bacterial isolate is shown in Supplementary Table 4. For each ARG increase among E. cloacae, we observed a 13.2% decrease in the likelihood of neonates having LOS (P = 0.016; 95% CI = 0.77-0.97). No other associations between ARG and onset were found (Supplementary Table 3).
Worryingly, we found that 529 (60%) of the GNB isolates tested were resistant to the first-line empirical treatment for neonatal sepsis (both ampicillin and gentamicin).
Plasmids and carbapenemase genes. As many ARGs are carried on mobile genetic elements such as plasmids, we searched for plasmid replicon types in isolates from the different geographical areas and, where possible, analysed linkages between plasmid type and the carriage of specific carbapenemase genes. We detected 1,124 plasmids with 45 inc gene variants, which we categorized into 18 plasmid groups. From these, 1,093 were found within E. coli (n = 169), K. pneumoniae (n = 623), K. michiganensis (n = 142), K. quasipneumoniae (n = 28), Enterobacter species (n = 87) and S. marcescens (n = 44) (Fig. 4). There were 12 plasmid types found within the seven S. enterica isolates, seven among Citrobacter species and five among K. variicola isolates.
Within the six dominant plasmid carriers (Fig. 4), the most frequently detected inc type was IncFIB, with 255 out of 440 hits within K. pneumoniae genomes. bla NDM-1 was found in IncA/C2 plasmids from K. pneumoniae and Enterobacter species from PP, as well as in IncFIB plasmids from K. pneumoniae from India ( Fig. 4 and Supplementary Fig. 10). We found IncX3 plasmids carrying bla NDM-5 in K. pneumoniae from NK, Nigeria, and multiple bla NDM-7 in K. pneumoniae and Enterobacter species from Nigeria and Pakistan ( Fig. 4 and Supplementary Fig. 10). Col plasmid types were identified within 82 genomes. We found ColKP3 plasmids carrying bla OXA-181 or bla OXA-232 in isolates of three different species: E. coli, K. pneumoniae and Enterobacter species (Fig. 4 and Supplementary Fig. 10).
Our bioinformatics analysis relied on the interrogation of short-read sequencing data; therefore, it was not possible to analyse the genetic context to link carbapenemase genes and inc type for all genomes. Instead, a representative genome of each species/ ST with the largest contig carrying the carbapenemase gene was chosen to maximize the analysis of other genetic material present, including the inc gene (n = 9; Supplementary Fig. 10). This analysis demonstrated cases where the same carbapenemase gene variant was detected in the same plasmid type across different GNB, suggesting that successful dissemination and acquisition within multiple species may be occurring. We also found cases where the same carbapenemase ARG was detected in multiple different plasmids, furthermore evidencing the spread of AMR.
Characterization of K. pneumoniae. K. pneumoniae is an important cause of neonatal sepsis in LMICs; however, there are few data analysing this species beyond antimicrobial susceptibility testing. Here, we have shown that K. pneumoniae was the most frequently identified GNB; therefore, the genomic diversity of this collection was scrutinized to contextualize these isolates, both within this study collection and within previously known collections 15,16,22,23 .  Table 5 for literature search inclusion criteria) and neonatal sepsis context 23 (Fig. 5b) revealed high diversity of K. pneumoniae, with 156 STs from 17 countries spanning five continents. BARNARDS isolates clustered with previously reported neonatal isolates, including ST45, ST48 and ST348 (refs. 15,22,23 ), but we also revealed distinct and new genetic lineages. The major AMR-related K. pneumoniae clades in Asia (ST11) and Europe (ST147 and ST307) 24 were also identified in this study. We only found ST307 in Rwanda (n = 6) and Nigeria (n = 2). However, ST258, a North American clade frequently associated with bla KPC 25,26 , was absent, which accords with the absence of bla KPC in this study. We did, however, detect one bla KPC-2 gene in a K. quasipneumoniae from Bangladesh ( Supplementary Fig. 10). While other studies have suggested that bla KPC K. pneumoniae causes neonatal sepsis, especially during nosocomial outbreaks 25 , there is currently little evidence from countries in Africa or South Asia.
BARNARDS' K. pneumoniae were disseminated throughout the global phylogeny, with 57 STs (Table 1 and Fig. 5a). ST35 and ST37 were predominantly found in Ethiopia (n = 38/39 and n = 29/30, respectively). We found four ST35 K. pneumoniae from other neonatal sepsis publications; however, these sit on a distinct branch in the core genome phylogeny (Fig. 5b) and were more closely related to the single ST35 isolated from RU, Rwanda. ST15 isolates were largely isolated from Pakistan and all carried both bla NDM-1 and bla OXA-181 (n = 22/27; Fig. 5b). ST15 was almost exclusively found at the South Asian clinical sites, with a single ST15 found at NN, Nigeria. ST442 (n = 6) and ST464 (n = 8) were only found in NN and all isolates contained either bla NDM-1 (ST442) or bla NDM-7 (ST464).
Multiple different capsule types (n = 47 KL loci and n = 12 O loci) were identified in silico. ST15 isolates in Pakistan (n = 23) and India (n = 1) were all the O1v1:KL112 serotype, whereas single ST15 isolates in Bangladesh and Nigeria had different serotype combinations of O3b:KL38 and O1v1:KL48, respectively. Similarly, the ST35 isolates from Ethiopia were all O1v2:KL108, whereas K. pneumoniae from RU in Rwanda were O2v1:KL113. Of the eight ST348 isolates, of which seven were from South Africa and one was from  In total, 115 isolates had a virulence score of 1, 3 or 4, indicating the presence of yersiniabactin and/or aerobactin/salmochelin virulence genes (see the Kleborate repository 27 and Source Data Fig. 5). The odds of LOS were 89% lower for infants with sepsis due to K. pneumoniae who had a virulence score of 3 or 4 compared with those with a score of 0 (P = 0.04; 95% CI = 0.014-0.90). Additionally, the odds of the outcome deceased were 14 times higher for infants with sepsis due to K. pneumoniae who had a virulence score of 3 or 4 compared with those with a virulence score of 0 (P = 0.001; 95% CI = 2.76-68.77; Supplementary Table 3). These results suggest that these genes may be involved in quicker onset and MFBS. Alternatively, they may reflect transmission of distinct isolates from the mother's microbiota 6 and from the clinical environment.
Collectively, this analysis reveals a large degree of intraspecies diversity within K. pneumoniae pathogens causing neonatal sepsis in LMICs. While certain ST groups previously shown to be dominant in particular geographic regions were found during the BARNARDS study, we also detected several different ST groups carrying different carbapenemase ARGs and virulence determinants.
Characterization of E. coli. E. coli has previously been reported as a dominant GNB cause of neonatal sepsis across many different LMICs 18 . In light of this, we aimed to further characterize the BARNARDS E. coli isolates and to compare the results with existing WGS datasets (Supplementary Table 5 displays the literature search inclusion criteria). The 2014 enterotoxigenic E. coli collection 28 displayed large diversity and, when analysed with the BARNARDS dataset, 90 STs across four continents and 21 countries were detected (Fig. 6a). The E. coli analysed herein fell within four main clades of the extended phylogeny. The greatest numbers of E. coli in the BARNARDS study (n = 75) were in Nigeria (n = 23; 31%) and Rwanda (n = 17; 23%), although E. coli neonatal sepsis was identified in 11 of the 12 clinical sites (excluding BK, Bangladesh). A phylogenetic analysis of BARNARDS E. coli revealed four main groups, each containing multiple clades (Fig. 6b), with 37 STs detected. ST10 (n = 9), ST131 (n = 6), ST410 (n = 5) and ST69 (n = 4) were the most common. In the Mentzer et al. 28 collection, ST10 was found in five countries across Africa, Asia and South America. Of the nine ST10 E. coli characterized here, seven were from clinical sites within Africa (Ethiopia, n = 1; Nigeria, n = 5; Rwanda, n = 1) and two were from PP, Pakistan. Although all ST10 E. coli belonged to the same phylotype (that is, group A), each isolate had a different O:H serotype profile and the phylogenetic tree (Fig. 6b) shows variability in the branch length, indicating genomic diversity within this ST group. Generally, we found highly variable O:H serotype classification (n = 57 O:H combinations), irrespective of the ST or phylogroup (A-F) (Source Data Fig. 6). Of the phylotypes, A (n = 22), B1 (n = 15) and B2 (n = 18) were the most common.
β-lactam and aminoglycoside ARGs were most commonly found in the South Asian isolates, with the exception of bla CTX-M-15 , which was also detected in E. coli from Africa. In total, 21 E. coli harboured bla CTX-M-15 and belonged to a variety of STs, including ST131, ST405, ST410, ST10 and ST167. Carbapenemase genes were detected in only three out of 75 E. coli. The three isolates were from South Asia, and all concomitantly carried bla CTX-M-15 plus bla NDM-5 (ST167; n = 2) or bla OXA-181 (ST410; n = 1).
No significant associations were found between phenotypic or genotypic AMR-related traits of E. coli and the clinical data assessed herein (Supplementary Table 3).
E. coli isolates were extremely diverse, suggesting that there are probably several important and worrisome lineages.

Discussion
In the BARNARDS study, we established a methodological framework to capture and extensively characterize GNB species causing neonatal sepsis in LMICs. We isolated 916 isolates of GNB, characterized them to species level, used WGS to probe genome composition and MLST to assess intra-and interspecies diversity, and documented extremely high rates of AMR.
Most of the Gram-negative isolates from neonates with sepsis were resistant to at least one β-lactam and one aminoglycoside (597/885; 67%), as has been reported previously for cohorts in India 29 and 26 countries in Africa 18 . World Health Organization guidelines 30 stipulate ampicillin plus gentamicin as the first line of empirical treatment for neonatal sepsis and third-generation cephalosporins as the second line of treatment. Of note, many of the blood culture isolates from our study were resistant to both lines of treatment, meaning that treatment options are unlikely to be curative.
The identification of 58 different GNB species suggests that the aetiology of neonatal sepsis is complex. We report multiple different lineages causing infection within single species, many of which carry either resistant or putatively virulent mechanisms and several of which have previously been shown to cause neonatal sepsis (for example, ST35 and ST37 K. pneumoniae) 23,31 . The identification of high-risk clones, such as ST15 in K. pneumoniae 32 and the global clones ST1 and ST2 in A. baumannii, which are notorious for nosocomial infection 33 , indicates the spread and persistence of problematic lineages in LMICs. In addition, through our comprehensive analysis, we identified 40 previously unknown STs, suggesting that well-known and previously unidentified lineages/ST groups are both co-existing and evolving.
A limitation of our study was the inability to follow up all neonates to 60 d (necessitating the exclusion of neonates who were lost to follow-up), which impacted our outcome data (Fig. 1). It is likely that additional local factors (such as the management of sepsis) contributed to mortality; therefore, we cannot attribute MFBS singularly to the presence/absence of genomic traits. Our statistical analyses were exploratory and should be interpreted as hypothesis generating only. Fig. 6 | Core genome characterization of E. coli isolates. a, Three-hundred-and-sixty isolates incorporating a global collection 28 . Blue shading indicates E. coli isolates from the BARNARDS collection (n = 75). The branch labels are coloured according to country of origin. b, Detailed core genome characterization of 87 E. coli isolates (n = 75 BARNARDS). Yellow shading represents isolates from other studies 16,22 causing neonatal sepsis. The colours on the right represent infant outcome (orange) and onset of sepsis (green), followed by the ST. The numbers and code names (coloured according to study site) of isolates are also given. The branch symbols denote the carriage of carbapenemase ARGs (bla NDM-5 (circles) and bla OXA-181 (squares). NF, not found.

Nature Microbiology
In summary, Klebsiella, E. coli and Enterobacter were the main GNB species responsible for sepsis in neonates. We report that 54% of isolated bacteria were resistant to at least one antibiotic within four to six classes of antibiotics, and observed widespread carriage of both resistance genes and virulence factors in GNB causing neonatal sepsis in LMICs. This large, observational study will inform  Table 6). Between November 2015 and December 2017, women in labour or immediately postpartum were recruited prospectively following consent, and their neonates were followed up for the first 60 d of life (at 3, 7, 14, 28 and 60 d of life) or until study withdrawal/death. Neonates admitted to clinical sites showing signs of sepsis were also enrolled. The BARNARDS sample collection workflow is shown in Supplementary Fig. 11. Although it was not strictly a neonatal population, during this study, we employed the term neonate for all enrolments, including those between 30 and 60 d post-birth. Documentation detailing the parameters for clinically diagnosing sepsis is available in Supplementary Fig. 1. The standard operating procedures for the laboratory processing of blood cultures and subsequent identification of bacteria, which were followed by all clinical sites (following agreement between clinical partners before the start of enrolment), are shown in Supplementary Fig. 2. Laboratory reagents (Liofilchem)-importantly, both agar media and antibiotic discs (used at the clinical sites)-were standardized throughout the network. Bacterial identification performed at each site was confirmed by WGS at Cardiff University. Antimicrobial susceptibility testing was performed twice: initially, at the local sites to guide treatment using antibiotic discs; and then at Cardiff University using the agar dilution method to establish the MICs (details below). The collected clinical data included onset of sepsis (EOS or LOS) and patient outcome following biological sepsis. Neonates that were lost to follow up were categorized along with neonates confirmed alive as 'not reported deceased' . For the purpose of this study, EOS and LOS were defined as sepsis occurring ≤72 h and >72 h after birth, respectively. If neonates showed clinical signs of sepsis at multiple time points within the first 60 d of life, additional blood cultures were analysed. All viable bacterial species were stored on charcoal swabs (Deltalab) for transport under UN3373 regulations to Cardiff University.
At Cardiff University, GNB isolates were plated onto chromogenic urinary tract infection media supplemented with vancomycin at 10 mg l −1 (Liofilchem) and incubated aerobically overnight at 37 °C. Isolates were identified using a Microflex LT MALDI-TOF MS (Bruker Daltonik) with α-cyano-4-hydroxycinnamic acid matrix (Sigma-Aldrich). Bacterial isolates were stored in TS/72 beads (Technical Service Consultants) at −80 °C and the original swabs were stored at 4 °C. MICs were determined by agar dilution for a panel of 19 antibiotics and interpreted according to the EUCAST guidelines 20,21 . E. coli ATCC 25922 and P. aeruginosa ATCC 27853 strains were used as quality controls for GNB tests. Supplementary Table 7 depicts the panels of antibiotics and additional control strains used. Supplementary Table 8 defines the EUCAST interpretations used. The MIC 50 and MIC 90 values for each antibiotic was determined. The phenotypic metadata included the following AMR-related counts: carbapenem resistance (taken as resistance to ertapenem); methicillin resistance (taken as an oxacillin MIC > 2) and an AMR score (the number of antibiotics to which an isolate was resistant) (Supplementary Table 7).

WGS.
A single bacterial colony was transferred into 1.8 ml LB broth and incubated at 37 °C and 180 r.p.m. for 18 h. Genomic DNA was extracted using a QIAamp DNA Mini Kit (Qiagen), with an additional RNAse step, on a QIAcube (Qiagen), and quantified using a Qubit Fluorometer 3.0. Genomic libraries were prepared using a Nextera XT V2 kit (Illumina) with bead-based normalization. A total of 48 isolates were multiplexed per sequencing run to provide a depth of coverage of >15×. Paired-end WGS was performed on an Illumina MiSeq using the V3 chemistry to generate fragment lengths of up to 300 base pairs (600 cycles).
Bioinformatics analyses. Bioinformatics analyses were performed using a high-performance computing cluster at Cardiff University (Advanced Research Computing at Cardiff (ARCCA)) and CLIMB (version 1.0) 34 . Paired-end reads (FASTQ) were subjected to quality control checks before downstream analysis. Trim Galore (version 0.4.3) 35 was used to remove the Nextera adapter sequences and low-quality bases. Reports before and after read trimming were generated using FastQC (version 0.11.2) 36 and collated using MultiQC (version 1.7) 37 . The mean read length and number of sequences provided on the MultiQC reports were used to determine the sequencing coverage. Paired-end reads were overlapped using Flash (version 1.2.11) 38 and assembled into contigs using SPAdes (version 3.9.0) 39 . The trimmed FASTQ reads were mapped to the contigs using BWA (version 0.7.15) 40 and SAMtools (version 1.3.1) 41 . Pilon (version 1.22) 42 was used to assess any misassemblies/errors in base calling in the resulting mapped BAM file. Final genome assembly metrics were generated using QUAST (version 2.1) 43 . Bacterial species were identified using both BLAST nt (version 2.2.25; https://blast. ncbi.nlm.nih.gov/Blast.cgi; input = contigs) 44 and PathogenWatch (version 3.13.10; https://pathogen.watch; input = contigs). MLST, virulence and plasmid genomic profiles were characterized using SRST2 (version 0.2.0) 45  Statistical analyses. Statistical associations between clinical outcomes (onset of sepsis (EOS/LOS) and MFBS (alive/deceased)) and phenotypical and genomic traits were explored using univariable logistic regression models with the Wald test in SPSS version 26. The outcomes for the analyses regarding sepsis onset were EOS and LOS and those for MFBS were alive and deceased. Depending on whether predictor variables were continuous (AMR or ARG) or categorical (resistance versus non-resistance to ampicillin and gentamicin, concomitant resistance versus non-concomitant resistance to the three cephalosporins tested, resistance to at least one of the three cephalosporins tested versus to none, resistance versus non-resistance to ertapenem (a marker for carbapenems resistance) or Klebsiella species virulence scores), they were treated as covariates or factors, respectively. For species group analyses, only groups with n = ≥50 isolates were included, except for B. cenocepacia isolates, which did not carry ARGs. Statistical significance was taken at P ≤ 0.05, and estimated ORs are presented along with 95% CIs.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

nature research | reporting summary
April 2020

Corresponding author(s): Kirsty M Sands
Last updated by author(s): Dec 6, 2020 Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Research policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted

Software and code
Policy information about availability of computer code Data collection No software was used in the data collection. At the low middle income countries, research nurses completed questionnaires with the women approaching labor. These questionnaires were either transcribed onto paper, due to availability of resources/infrastructure, i.e. Internet access, and later uploaded into Bristol Online survey (BOS) or directly entered into BOS using a tablet device provided by the project.