Ten Years of Population-Level Genomic Escherichia coli and Klebsiella pneumoniae Serotype Surveillance Informs Vaccine Development for Invasive Infections

Abstract Background The incidence of bloodstream infections (BSIs) caused by Escherichia coli and Klebsiella pneumoniae is increasing, with substantial associated morbidity, mortality, and antimicrobial resistance. Unbiased serotyping studies to guide vaccine target selection are limited. Methods We conducted unselected, population-level genomic surveillance of bloodstream E. coli and Klebsiella pneumoniae isolates from 2008 to 2018 in Oxfordshire, United Kingdom. We supplemented this with an analysis of publicly available global sequencing data (n = 3678). Results We sequenced 3478 E. coli isolates (3278 passed quality control) and 556 K. pneumoniae isolates (535 [K-antigen] and 549 [O-antigen] passed quality control). The 4 most common E. coli O-antigens (O1/O2/O6/O25) were identified in 1499/3278 isolates; the incidence of these O-types increased over time (incidence rate ratio per year [IRRy] = 1.14, 95% confidence interval [CI]: 1.11–1.16). These O-types accounted for 616/1434 multidrug-resistant (MDR) and 173/256 extended-spectrum beta-lactamase (ESBL)-resistant isolates in Oxfordshire but only 19/90 carbapenem-resistant isolates across all studies. For Klebsiella pneumoniae, the most common O-antigens (O2v2/O1v1/O3b/O1v2) accounted for 410/549 isolates; the incidence of BSIs caused by these also increased annually (IRRy = 1.09; 95% CI: 1.05–1.12). These O-types accounted for 122/148 MDR and 106/123 ESBL isolates in Oxfordshire and 557/734 carbapenem-resistant isolates across all studies. Conversely we observed substantial capsular antigen diversity. Analysis of 3678 isolates from global studies demonstrated the generalizability of these findings. For E. coli, based on serotyping, the ExPEC4V and ExPEC10V vaccines under investigation would cover 46% and 72% of Oxfordshire isolates respectively, and 47% and 71% of MDR isolates. Conclusions O-antigen targeted vaccines may be useful in reducing the morbidity, mortality, and antimicrobial resistance associated with E. coli and K. pneumoniae BSIs.

The incidence of bloodstream infections (BSIs) caused by Enterobacteriaceae (eg, Escherichia coli, Klebsiella pneumoniae) continues to increase globally [1][2][3]. Enterobacteriaceae are a significant antimicrobial resistance (AMR) threat as they can acquire new genes horizontally and therefore rapidly adapt to changing selection pressures, including antimicrobial use. With the limited development of new antimicrobials to treat multidrug-resistant (MDR) infections, the threat of untreatable disease is a significant risk [4]. To date, a nationally mandated set of infection prevention and control standards arising from the Health and Social Care Act of 2008 [5] has had limited success in reducing the incidence of BSIs caused by Enterobacteriaceae, as exemplified by national surveillance data [4,6].
Prophylactic vaccines represent an alternative approach to combating antimicrobial resistance (AMR) by reducing antibiotic usage and preventing infections caused by AMRassociated strains. For E. coli, the O-antigen, a component of lipopolysaccharide (LPS), has been considered the most promising vaccine target [7]. A recently developed bioconjugate vaccine (ExPEC4V; Janssen Pharmaceuticals) targeting 4 E. coli O-antigens (O1A, O2, O6A, and O25B) has subsequently shown promise in a phase II study [8]. A phase 1/2 clinical trial for a 10-valent vaccine (ExPEC10V) is currently recruiting [9]. For K. pneumoniae, vaccine development is lagging behind [10], and there are no vaccines currently in clinical trials.
For E. coli, recent serotyping studies have described limited O-antigen diversity in foodborne [11], enterotoxigenic E. coli/ enteropathogenic E. coli (ETEC/EPEC) [11], and extendedspectrum beta-lactamase (ESBL)-producing BSI isolates [12]. To our knowledge, however, there are no recent systematic and longitudinal studies characterizing the seroepidemiology of unselected BSIs. Recent studies of multicountry isolate collections, which were nonsystematically collected and/or enriched for antibiotic resistance, have suggested that the O-antigen could also be a promising vaccine target for K. pneumoniae. Follador and colleagues analyzed 4 sequencing data sets (1 pre-1950, 2 others enriched for AMR/clonal outbreaks) and found that 3 O-antigens (O1/2/3) accounted for 80% of isolates, whereas the diversity of K-antigens was an order of magnitude higher [13]. Similarly, the O1/2/3/5 antigens accounted for 90.1% of isolates in a recent global sequencing study [14], although there were only relatively small numbers of isolates over limited timeframes from each country, and it was unclear how these had been selected.
Large-scale, systematic typing studies could help to inform the rational selection of vaccine targets across population groups which best mitigate the risk of BSI (including those associated with AMR) and provide a baseline for monitoring the emergence of serotype variation in response to vaccine rollout. We performed a large, systematic study over 10 years in Oxfordshire and combined this with an analysis of publicly available global data sets to provide a robust basis for future vaccine development.

Sample Collection/Sequencing
All blood culture isolates identified by the Oxford University Hospitals clinical microbiology laboratory between September 2008 and December 2018 as being K. pneumoniae/E. coli (by matrix-assisted laser desorption/ionization time of flight [MALDI-TOF]) analysis (from 2013) and by growth on chromogenic/MacConkey agar and analytical profile index tests (prior to 2013) were included in the study. One E. coli isolate was excluded due to a cataloging error. Infection control procedures were in place for inpatients known to be colonized with ESBL/carbapenemase-producing Enterobacteriaceae (CPE) during the study, including isolation in side rooms where feasible, enhanced contact precautions, and screening of high-risk patients on admission (for CPE). To avoid double counting instances where patients had several positive blood cultures taken during a single episode of infection, we considered only the first available isolate per 90-day period per patient.

Statistics
Stacked negative binomial regression was used to compare incidence rate ratios over time (per year longer, incidence rate ratio per year [IRRy]) of BSI belonging to different antigen groups [24]. For each serotype or group of serotypes examined, we modeled the incidence of isolates with and without this trait over calendar time. In order to include as many isolates as possible in these models we defined calendar years as beginning in November (ie, isolates before November 2008/after November 2018 are excluded from this part of the analysis). A Wald test for heterogeneity was performed to test the null hypothesis that the incidence trend did not differ between isolates with and without the trait. Healthcare-associated (HA) infection was defined as either nosocomial infection (onset >48 hours after admission) or onset within 30 days of last discharge and community-associated (CA) as that occurring >30 days since last admission [25,26]. Statistical analysis was performed using R software, version 4.0.2, except for stacked negative binomial regression models, which were implemented in STATA version 15. Sequencing data were linked to electronic healthcare records via the Infections in Oxfordshire Research Database (IORD). IORD has generic Research Ethics Committee, Health Research Authority, and Confidentiality Advisory Group approvals (14/ SC/1069, ECC5-017(A)/2009).

Bioinformatics
De novo assembly was performed using Shovill (version 1.0.4) [27]. In silico serotyping was performed using the Ectyper tool [28][29][30] for E. coli and with Kleborate for Klebsiella spp. [31]. For E. coli serotypes we included only those isolates that passed the ECtyper quality control (QC) filter and had a high confidence locus call; similarly for Klebsiella spp. we included only isolates where the Kleborate locus confidence call was at least "good" (found in a single piece or with ≥95% coverage, with ≤3 missing genes and ≤1 extra genes) [31]. Genes (yersiniabactin, colibactin, and aerobactin) associated with hypervirulence (defined as the ability to cause severe invasive disease in otherwise healthy individuals in the community [31]) were detected using the Kleborate tool. Virulence score was part of the Kleborate output, namely: 0 = no acquired virulence factors, 1 = yersiniabactin only, 2 = yersiniabactin and colibactin (or colibactin only), 3 = aerobactin only, 4 = aerobactin and yersiniabactin, 5 = yersiniabactin, colibactin, and aerobactin.

Antibiotic Susceptibility Testing
Antibiotic susceptibility was measured using the BD Phoenix (Becton Dickinson, Franklin Lakes, New Jersey, USA) platform (EUCAST breakpoints [32]) for isolates after 2013 or disc diffusion (BSAC breakpoints [33]) for earlier isolates; all susceptibilities were measured using standard operating procedures in a UK Accreditation Service (UKAS) accredited laboratory.
Antibiotic susceptibility data were available through the Infections in Oxfordshire Research Database. For comparisons with existing studies, in silico antibiotic resistance gene detection (as called by Kleborate) was used instead of in vitro phenotyping because the latter results were not universally available.
Previous studies have suggested that clinically hypervirulent K. pneumoniae isolates (ie, causing CA invasive infection) are particularly associated with the KL1/2 capsular types [34]. There were only 2 isolates with the KL1 capsular antigen in our study, only 1 of which carried the virulence-associated genes encoding for yersiniabactin, colibactin, aerobactin, salmochelin, and rmpA. Although KL2 did have the most isolates carrying hypervirulence genes of any capsular type, there were also 13/35 KL2 isolates with no hypervirulence loci ( Figure 3). Also, 36/37 KL1/2 isolates had O1v1 or O1v2 antigens (the other had O1/ O2v2) meaning that a quadrivalent O-antigen-targeted vaccine would theoretically provide good coverage against this important group of isolates. Although there were more CA than HA isolates for the KL2 capsular type (19/15), this was not the only K-type where this was true, and the clinically hypervirulent phenotype was observed in 74 capsular types in total.
As the isolates from our study originated from a single region of the United Kingdom with a relatively low AMR burden, we compared our serotypes to those from recently published studies in the United Kingdom [17,19,21] and globally [16,18,20,22,23]. Overall, although there was some regional variation, findings supported the potential for O-antigen-targeted quadrivalent vaccines to contribute to substantial reductions in overall BSIs, as well as those caused by ESBL and MDR isolates, in both E. coli and K. pneumoniae (Table 1). For K. pneumoniae they would likely provide a good level of protection against carbapenem-resistant and (genetically) hypervirulent isolates.

DISCUSSION
A quadrivalent vaccine comprising the 4 most prevalent O-antigens for E. coli and Klebsiella spp. could theoretically have a significant impact on the incidence of BSIs and ESBL/ MDR infections, as well as offering an alternative approach to help tackle the global epidemic of AMR in Enterobacteriaceae. Ten-valent vaccines are at a slightly earlier stage of development for E. coli, but our data suggest that they have the potential to make a significant impact on the incidence of invasive disease and associated antimicrobial resistance. Such a vaccine for K. pneumoniae would have covered nearly all isolates (541/549 [99%]) associated with invasive disease over the past decade in Oxfordshire.
Vaccines should also provide some protection against non-BSI infections that may evolve into BSIs, in particular urinary tract infections. As well as reducing morbidity and providing economic benefits, this could also be expected to lower antibiotic selection pressures against E. coli and Klebsiella spp. The relatively much higher incidence of these primary infections versus BSIs may increase the economic viability of any prospective vaccine that, particularly for Klebsiella spp., may be a significant barrier to development given the difficulty in identifying groups of patients in which vaccination would be cost-effective [35].
We would expect the ExPEC4V/ExPEC10V vaccines to offer less protection against carbapenem-resistant E. coli infections. Although at first glance this looks disappointing, the prevention of ESBL infections would still reduce the carbapenem selection pressure. Additionally, it was notable that the majority of carbapenem resistance genes carrying E. coli isolates were not from the sequence types that have been found to cause most (predominantly CA) BSIs in other studies [16,17,19]. This might indicate that these isolates do not originate from the community reservoir and that they may be associated with nosocomial transmission as was found to be the case in K. pneumoniae [23]. Widespread dissemination of carbapenem resistance in sequence types associated with global epidemics of CA-MDR disease (eg, ST131) would be a public health challenge, and a vaccine may have a role in preventing this.
The main limitation of our study is that it is from a single region. To mitigate this we analyzed data from other published studies; however, in contrast to our long-term regional surveillance, most of these were either deliberately enriched for MDR isolates or at high risk of bias due to unclear sampling methodologies. Although direct comparison with our data is therefore difficult, the overall findings appear to be reflected both nationally and globally. We were unable to characterize the distribution of capsular (K) antigen types for E. coli because, to our knowledge, there is no curated database of sequences currently available. Additionally, we were unable to assign an in silico O-antigen call to ~5% of isolates in both E. coli and K. pneumoniae. This may be due to assembly errors or the presence of novel serotypes.
Overall, O-antigens targeted vaccines for E. coli and Klebsiella spp. could theoretically have a significant impact on the incidence of BSIs and ESBL/MDR infections, as well as offering an alternative approach to help tackle the global epidemic of AMR in Enterobacteriaceae. Our findings support future development and subsequent efficacy studies for such vaccines, provided that they are demonstrated to be safe and immunogenic. The substantial capsular antigen diversity observed for Klebsiella spp. BSIs would likely preclude this as a useful vaccine target. Surveillance following the implementation of any vaccine would be essential in identifying any evidence of strain replacement.