Molecular mechanisms of re-emerging chloramphenicol susceptibility in extended-spectrum beta-lactamase producing Enterobacterales

Infections with Enterobacterales (E) are increasingly difficult to treat due to antimicrobial resistance. After ceftriaxone replaced chloramphenicol (CHL) as empiric therapy for suspected sepsis in Malawi in 2004, ESBL-E rapidly emerged. Concurrently, resistance to CHL in Escherichia coli and Klebsiella spp. decreased, raising the possibility of CHL re-introduction. However, many phenotypically susceptible isolates still carry CHL acetyltransferase (CAT) genes. We used a combination of genomics, phenotypic susceptibility assays, experimental evolution and functional assays for CAT activity to understand the molecular mechanisms and stability of this re-emerging CHL susceptibility. Of 840 Malawian isolates, 31% had discordant CHL susceptibility genotype-phenotype, and we selected 42 isolates for in-depth analysis. Stable degradation of cat genes by insertion sequences led to re-emergence of CHL susceptibility. Our study suggests CHL could be reintroduced as reserve agent for critically ill patients with ESBL-E infections in Malawi and similar settings and highlights the ongoing challenges in inferring antimicrobial resistance from sequence data.


Introduction
Antimicrobial resistance (AMR) is a major threat to global health.Drug resistant bacterial infections were estimated to be associated with 4.95 million deaths in 2019, with sub-Saharan Africa the worst affected region [1].Among the most problematic drug resistant bacteria are extended-spectrum beta-lactamase (ESBL) producing Enterobacterales (E) which are resistant to 3 rd -generation cephalosporins (3GC) and classified as priority pathogens by the WHO [2].ESBL-E infections are associated with higher morbidity, mortality and economic burden in the treatment of bloodstream infections in Malawi [3,4], and in low resource settings such as Malawi access to effective treatment alternatives (e.g.carbapenems, or betalactam/beta-lactamase inhibitor combinations) is often lacking.The prevalence of ESBL-E rapidly increased in Malawi after ceftriaxone replaced chloramphenicol (CHL) as a first-line empiric therapy for suspected sepsis from 2004 [5,6].
CHL is an inhibitor of protein synthesis and broad-spectrum antibiotic discovered in 1947 [7], which is now rarely used, primarily because of the risk of severe adverse effects [8].These include dose-unrelated aplastic anaemia, which is irreversible and fatal [9] but rare (incidence between 1:19,000 -1:270,000 [10]), dose-related, reversible bone marrow suppression [11] and Grey Baby Syndrome [12].As effective and safer broad-spectrum beta-lactam antibiotics were introduced, they replaced CHL in most settings.The dominant mechanism of CHL resistance is mediated by CHL acetyltransferases (CAT) inactivating CHL by acetylation [13].
Other mechanisms include efflux pumps, inactivation by phosphotransferases, target site mutations and decreased membrane permeability [13].
While resistance to 3GCs, fluoroquinolones and aminoglycosides increased in Malawi, resistance to CHL decreased as its use declined [5].Escherichia coli and Klebsiella pneumoniae populations showed decreasing proportions of CHL resistant isolates, from around 80% in 1998-2004 to around 50% and below from 2012, sparking an interest in whether CHL can be re-introduced as a treatment option for ESBL-E infections in critically ill patients where there is no alternative therapy [5,14].Some bacteria isolated in Malawi, however, have a discordant CHL genotype-phenotype, i.e. they still harbour CHL resistance genes despite being phenotypically susceptible [15][16][17][18].Therefore, simple CHL resistance gene loss caused, for example, by lineage or plasmid replacement in the bacterial population is unlikely.With the prospect of using CHL as a potential reserve agent for 3GC resistant infections it is important to understand (i) the molecular mechanism(s) of this re-emerging CHL susceptibility, i.e., the molecular basis of CHL susceptibility genotype-phenotype mismatches, (ii) the stability of the phenotypic susceptibility and (iii) how widespread this phenomenon is.
Here, we investigated a collection of ESBL E. coli and K. pneumoniae with CHL susceptibility genotype-phenotype mismatches.Using functional assays in combination with genomic data, we determined the CHL sensitivities, CAT enzyme activity, and functional expression of cat genes, to understand the molecular mechanism of re-emerging CHL susceptibility.Further, we tested the stability of CHL susceptibility in several isolates employing experimental evolution with increasing concentrations of CHL and used co-occurrence analysis of AMR genes and investigated the spread of cat alleles in the context of sequence types to identify potential drivers to explain the high proportion of CHL susceptibility in ESBL-E populations in Malawi.

No CAT activity in susceptible isolates
Next, we tested all 42 isolates for CAT enzyme activity to determine whether cat genes were functionally expressed.We adapted the rapid CAT assay (rCAT) [19] as an indirect measure of enzyme activity.The rCAT assay measures free sulfhydryl groups of the CAT substrate and acetyl donor acetyl-coenzyme A in cell lysates and a colour change to yellow indicating CAT enzyme activity can be spectrophotometrically measured.
All phenotypically CHL-S isolates, irrespective of the presence of a cat gene, were negative for CAT activity (Fig. 2).All but five resistant isolates (CAC10A, CAC122, CAC10H, CAB119, CAC13Y) with a cat gene showed CAT activity.Three CHL-R/rCAT negative isolates carried catB4 genes and this was the only cat gene present.Isolate CAB199 (CHL MIC = 128 µg/mL) co-carried a floR gene which could contribute to the CHL-R phenotype.Isolates CAC13Y (CHL MIC = 64 µg/mL) and CAC10H (CHL MIC = 16 µg/mL) had no other detectable CHL resistance determinants whereas CAC10A had a catA1 but a weak CHL-R phenotype (MIC = 16 µg/mL).
We used the disc-diffusion (dCAT) assay [20] as a secondary and independent assay, which is based on the ability of CAT producing bacteria to cross-protect susceptible bacteria by inactivating CHL.Tested isolates (CAB119, CAC13Y) failed to cross-protect a sensitive strain, thus confirming that catB4 in those isolates is non-functional (supplementary Fig. 1).One isolate (CAC122) that carried cat_pC221, a cat gene originating from Staphylococcus aureus [21], did not show enzyme activity and its resistance phenotype is likely explained by cmlA1.
This cat gene was shown to be inducible and regulated through translational attenuation [22]; we thus tested this isolate with sub-MIC concentration of CHL in culture media but no CAT activity was observed.We concluded that resistant isolates with catA1, catA2 and catB3 have functional CATs whereas isolates with catB4 do not.Phenotypically susceptible isolates with or without cat genes did not show CAT activity demonstrating that cat genes in those CHL-S isolates are not expressed.

Functional expression of catB4 does not confer CHL resistance
To assess the effect of the genetic context of cat genes in our isolates, we cloned the coding sequence (CDS) of the dominant cat variants present in the collection (catA1, catA2, catB3 and catB4) into the expression vector pEB1-sfGFP [23] under the constitutive proC promoter, by replacing sfGFP, and tested their susceptibility to CHL in a clean genetic background of E. coli MG1655.MG1655 pEB1-catB4 had an MIC of 8 µg/mL identical to the MG1655 pEB1-sGFP control whereas all other cat variants had an MIC of >512 µg/mL confirming that catB4 is a non-functional variant (Table 1).

catB4 is a non-functional truncated catB3
Functional CAT assays (rCAT and dCAT) demonstrated that catB4 in the genetic context of tested isolates was non-functional and ectopic expression of catB4 confirmed that the CDS does not produce a functional product that confers CHL resistance.We previously observed that in addition to a full length assembled catB4 (549 bp) many isolates had several partial assemblies in their genomes of approximately 107 bp (Supplementary Fig. 2a, [24]).We investigated the 107 bp (position 443-549 of catB4) by BLASTn against NCBI's nucleotide collection and obtained many hits among different Enterobacterales with 100% query cover and sequence identity matching the "IS6-like element IS26 family transposase".We next compared catB4 with catB3 by pairwise sequence alignment.Both CDS share 100% sequence identity for the first 442 bases and only differ after position 443, corresponding to the last 107 bp of catB4 matching with the IS26 sequences (Fig. 3a).All but seven catB4 assemblies in our isolate collection share 100% sequence identity (Supplementary Fig. 2b) with catB3 but only until position 442 bp (full length catB3 is 633 bp).Therefore, we concluded that catB4 is a truncated variant of catB3 and suggest it should be called catB3∆ 443-633 .IS26mediated deletions of adjacent DNA have previously been reported [25,26].
We traced catB4 back in the literature and found it was first described, to the best of our knowledge, in two plasmids (pEK499 and pEK516) [27].Both plasmids had been moved into different E. coli lab strains and did not confer CHL resistance, consistent with our data showing that catB4 is not a functional CAT.Because this had been annotated as catB4 variant in the ARG-ANNOT (and SRST2) databases [28] we tested two additional databases for AMR genes.Both, the CARD database [29] and ResFinder [30] called the truncated catB3 correctly.
We selected five E. coli isolates for long-read sequencing to investigate the genomic context of catB3∆ 443-633 (Supplementary Table 3).Aligning four closed contigs containing catB3 and its truncated variant showed a similar and common genetic feature of aac(6')-Ib-cr -blaOXA-1 -catB3 flanked by two IS26 elements.There was an additional IS26 downstream of the wild type catB3 gene (Fig. 3b).In the four isolates this feature is located on a different replicon, on the chromosome and three different IncF plasmids suggestive of moving as transposable element independent of a single plasmid-type.As this genetic feature has previously been shown to be part of an integron cassette [31,32], we ran IntegronFinder and found attC sites upstream of blaOXA-1 and downstream of catB3, however, the truncation of catB3 removed the attC site, likely precluding its ability to move between or within integron(s).Further, the integrase gene is missing in all but one isolate (CAE137) where it is interrupted by IS26 (Fig. 3b).

IS5 insertion into catA1 promoter causes reversal of CHL resistance
Of 9/42 isolates with a catA1, only four had a CHL resistance phenotype consistent with its genotype (Fig. 2).We therefore PCR-probed these nine isolates with primers specific for catA1 to confirm its presence.Three isolates containing catA1 that were CHL-R had the expected amplicon size of 150 bp, however three CHL-S isolates and CAC10A had a larger amplicon than expected of ~1400 bp, (supplementary Fig. 3).We assembled the genomes of isolates with catA1 and aligned the contigs containing catA1.Two contigs (from isolates CAI10X & CAI10E) showed a presumed insertion of 1199 bp, starting 6 bp upstream of the CDS, matching the larger than expected PCR amplicon and BLASTn revealed a 100% match to an IS5 element.In addition, we purified and Sanger-sequenced the ~1400 bp PCR product and confirmed the insert as an IS5 element in three of the isolates.Long-read sequencing of isolate CAC10A confirmed the insertion of IS5 as a single element into the promoter region of catA1 (Fig. 3c).Overall, of the isolates with catA1, 3/9 were CHL-R with detectible CAT activity (Fig. 2 ) and had an expected catA1 amplicon size (150bp); isolate CAC10A was CHL-R with an MIC close to the breakpoint (16 mg/L), no detectable CAT activity and IS5 inserted into the catA1 promoter region.The remainder (5/9) were CHL-S, without detectable activity on the rCAT assay; in 3/5 this was likely mediated by IS5 insertion interfering with transcription.
We adjusted primers from a previously developed high-resolution melting (HRM) assay [33] to capture IS5-catA1 and the truncated catB3∆ 443-633 , and demonstrated that they can be clearly distinguished from the wild type genes (Supplementary Fig. 4), highlighting the potential of this molecular diagnostic test for CHL resistance to discriminate between functional and non-functional cat genes.

Degradation of cat genes by insertion sequences is stable
IS5 insertion into the promoter of catA1 and truncation of catB3 are the main mechanisms for the observed re-emerging CHL susceptibility in our collection of isolates.To test if these mutation and insertion events could be reversed and potentially result in a rapid re-emergence of cat expression and CHL resistance, we experimentally evolved three CHL susceptible isolates: CAI10Z (catA1, catB3∆ 443-633 ), CAM10K (catA1) and CAI10X (catA1) in LB broth with increasing concentrations of CHL for 7 days (n=3 per isolate).An equal number of replicates were evolved in LB without selection.All evolved populations under CHL selection grew until 64 µg/mL CHL (8-16 -fold increase in MIC).Increasing incubation for another 24 h enabled recovery of some populations in 128 µg/mL (i.e., clearly visible growth) but none of the CHL populations grew in 256 µg/mL CHL (Supplementary Fig. 5a).All control populations evolved in LB grew until the experiment was stopped at day 7. Since we were interested if CHL pressure can result in a re-activation of cat genes we tested all evolved populations with the rCAT assay (LB evolved from day 7, CHL evolved last surviving population, day 5, day 6 or day 7).All populations examined tested negative on the rCAT assay (Supplementary Fig. 5b) and PCR-probing of evolved strains showed that the IS5 insertion was still present upstream of catA1 (supplementary Fig. 5c), confirming that other mutations rather than (re-)expression of catA1 are responsible for the resistance phenotype.

The genomic locus of catB3∆ 443-633 is conserved and widespread
The truncated catB3∆ 443-633 is common among our isolates and seems to be highly conserved in the genomic context of aac(6')-Ib-cr -blaOXA-1 -catB3.To investigate potential drivers, we ran a co-occurrence analysis across 772 genomes in our collection from Malawi (68 of the 840 total isolates failed assembly quality check) (Fig. S6, supplementary Table 4).We applied the probabilistic model from [34] to determine whether each pair of genes had an observed cooccurrence which was significantly different from the expected co-occurrence.The cooccurrence relationships across all AMR genes with at least one other gene is depicted in Fig. S6.Both, catB3 and catB3∆ 443-633 had a positive co-occurrence relationship with blaOXA-1 and aac(6')-Ib-cr and the beta-lactamase genes blaCTX-M-15 and blaTEM-1B.The catB3 and catB3∆ 443-   633 genes had a negative co-occurrence as they were never found to co-occur on the same genome (Fig. 4).In our isolates, catB3∆ 443-633 is much more common than catB3, and thus we investigated if this pseudogene is restricted to Malawi or common elsewhere.We queried "catB3" in NCBI's Microbial Browser for Identification of Genetic and Genomic Elements (Micro-BIGG-E) [35] and investigated 46,667 isolates (data download: 4/8/2023) with an annotated catB3 gene.
Strikingly, 30,819 (66%) showed a coverage to the reference of 70%, which corresponds to the IS26 truncated variant described here (Supplementary Fig. 7a).We tried to trace back the emergence of this truncation.From 2006 onwards (only including years with > 100 isolates) the truncated variant was dominant over the wild type, and we could not conclude when it emerged (Supplementary Fig. 7b).Klebsiella spp., E. coli and Enterobacter spp.dominantly show the truncated gene whereas Salmonella, Acinetobacter and Pseudomonas, typically harbour wild type catB3 (Supplementary Fig. 7c).Differentiating by host where the isolates have been collected is heavily biased towards humans where the dominant truncated variant is much more common.This is also seen in companion animals (cats and dogs) whereas in most food animals, the wild type proportion is close to 100%, perhaps indicative of ongoing selection for chloramphenicol resistance through use of phenicols in veterinary medicine and agriculture (supplementary Fig. 7d).Most geographic locations show higher proportions of the truncated variant with a few exceptions in China and Australia (Supplementary Fig. 7e).
Next, we investigated the spread of catB3∆ 443-633 in the context of ST to potentially link clonal expansion and catB3∆ 443-633 carried by these clones in E. coli.The catB3∆ 443-633 gene was restricted to six STs (410, 44, 648, 405, 617, 131) in our Malawi isolates (Fig. 5), and five of those STs are among the top 8 most common STs in our collection (Supplementary Fig. 8).
We expanded our analysis and looked at a collection of 10k genomes consisting of the top 100 common STs in E. coli (previously curated and downloaded from Enterobase (2020) [36] where each ST was randomly sampled to select 100 genomes.All six STs with catB3∆ 443-633    in Malawi also dominantly harbour this cat gene in this extended genome collection (Fig. 5).
In phylogroup B2 (predominantly human isolates associated with urinary tract and bloodstream infections) catB3∆ 443-633 is common in ST131 and to a lesser extend ST1193 (Fig. 5), both of which represent recently emerged globally dominant antibiotic resistant clones [37,38].The ubiquitousness of catB3∆ 443-633 may thus be linked to its association with distinct and successful lineages.

Discussion
In Malawi, as in much of sub-Saharan Africa, the rapid spread of ESBL-E coupled with the scarcity of Watch and Reserve antibiotics (i.e.carbapenems) has rendered many severe bacterial infections untreatable, therefore the re-emergence of chloramphenicol susceptibility is potentially important.34.2% (1666/4874) of non-salmonella Enterobacterales were phenotypically CHL susceptible in bloodstream infections in Malawi from 1998-2016 [5].Here, we analysed a collection of phenotypically and genotypically characterised E. coli and KpSC from Malawi to understand the molecular basis of CHL-S genotype-phenotype mismatches.
Since many phenotypically susceptible isolates in our collection carried cat genes, we first investigated if they were still functional and expressed.None of the CHL-S isolates we tested had functional CATs and we found that cat gene interruption was caused by insertion sequences.This is in contrast to a previous study reporting an E. coli strain being susceptible to CHL despite a functional CAT [39] which was later attributed to low level of acetyl-coenzyme A linked to mutations in efflux pumps [40].Combined functional analysis of catB4 with a CAT enzyme assays, expression in a clean genetic background as well as genomic investigation confirmed that catB4 is non-functional and is in fact a catB3 that has been truncated by an IS26 element.In other cases, an IS5 element has integrated into the promoter region of catA1 and likely interfered with transcription.In the latter case, calling AMR genes from all databases with sequence data will yield a wild type gene and currently classify the isolate as resistant since the mutation is outside of the CDS.The truncated catB3 will be called correctly if the annotation of catB4 is removed in the ARG-ANNOT/SRST2 databases, which we strongly support.
To effectively re-introduce CHL as a reserve treatment option for Enterobacterales infections confirmed as 3GC-resistant, we must consider the potential for reversion to CHL-resistance.
Isolates carrying IS5-catA1 could potentially rapidly revert to high-level resistance since the catA1 CDS is still present in the genome.Our data using experimental evolution suggest that this is not the case and selection with CHL did not lead to expression of catA1.One of the evolved isolates co-carried a truncated catB3 and we expected no reversion to a functional catB3 upon CHL selection because the missing 3' prime end of the gene is no longer present in the genome.Indeed, no functional cat emerged.
A decrease in CHL resistant isolates has also been reported for Salmonella Typhimurium, which was caused via loss of CHL resistance genes by lineage replacement [41].Our study adds to a developing evidence base for CHL in treatment of MDR Gram-negative pathogens; several studies have reported high or increasing rates of phenotypic CHL susceptibility among MDR Gram-negative bacteria suggesting CHL as a viable alternative treatment option in those settings [42][43][44].Further, CHL is an affordable and useful antimicrobial in terms of bioavailability, tissue penetration and broad spectrum of action [45,46].However, the prevalence of CHL susceptibility and the rare but severe side-effects of CHL, preclude the use of CHL for empirical management of sepsis in our setting: we instead envisage CHL to be used in critically ill patients with confirmed ESBL-E and CHL-S infection as a reserve agent.
This has the added advantage of keeping selection pressure for CHL resistance low but does require rapid determination of CHL susceptibility phenotype.
In the Malawian isolates the gold-standard AST correctly determined phenotypic CHL susceptibility, however, rapid molecular diagnostics have the potential to be faster and low cost.We applied the HRM assay to distinguish between the cat mutants found in our study.
Co-occurrence analysis showed that catB3 and catB3∆ 443-633 nearly always co-occurred with aac(6')-Ib-cr and blaOXA-1.This conserved feature of aac(6')-Ib-cr-blaOXA-1-catB flanked by IS26, which was confirmed by long-read sequencing, has previously been associated with an integron [32] and was found to contain attC sites when catB3 was present.However, the truncation by IS26 removed one of the attC sites.This, in combination with loss of the integrase gene, likely led to functional loss of mobilisation of genes within the integron and the high conservation of this locus is confirmatory.The association of catB3∆ 443-633 with blaCTX-M-15, the most common ESBL gene found in E. coli isolates worldwide [47] and among our Malawian isolates [17,18], may point towards the co-location of these genes on a plasmid, though a larger selection of isolates will need to be (long-read) sequenced to determine the spread of such a plasmid.Selection for 3GC resistance and co-occurring blaCTX-M-15 with catB3∆ 443-633    could have led to a genetic hitchhiking of the latter which could explain the high levels of CHL susceptible isolates among ESBLs.However, our isolate collection is heavily biased towards ESBL and may thus limit the interpretation of this co-occurrence.
Expanding our analysis of catB3∆ 443-633 revealed that this single truncation is globally more common than the wild type gene.It is intriguing that a truncated non-functional gene is so widespread, and we hypothesise this is due to the specific genetic context of a loss of integron activity, and association with other AMR genes as well as with IS26.It has previously been shown that IS26 and pseudo-compound transposons (i.e.IS26 bounded transposons) move 50x more frequently if there is another IS26 to target [48] and move by co-integration [49].
This mode of mobility likely enhances persistence.Additionally, association of catB3∆ 443-633    with globally dominant lineages of E. coli (e.g.ST131) as seen in our ST analysis, could be a likely driver that has contributed to the widespread occurrence of this truncated variant.
Our study highlights that antibiotic susceptibility can re-emerge following a reduction or cessation in use.Plasticity and the high levels of horizontal gene transfer among Enterobacterales can result in degradation of AMR genes by mobile genetic elements which can persist in the absence of selection.Further, the context of the AMR gene determines the phenotypic resistance and database curation is crucial to infer resistance from the genotypes.
We strongly recommend integrating functional context to databases and including additional features such as genetic context, including promoters and transcription start site.Many AMR genes only cause clinically relevant resistance when they are moved to locations in the genome where expression is upregulated, such as downstream of a strong promoter [50].This is problematic with functional metagenomic discovery of AMR genes since the mere presence of an AMR gene is not indicative of phenotypic resistance.Lastly, our data support the reintroduction of CHL as a last-line treatment option for patients critically ill with ESBL-E infections in Malawi and similar settings.

Antimicrobial susceptibility testing
AST was performed using broth microdilution according the EUCAST guidelines (v.isolates with an MIC ≤ 8 µg/mL were classified as susceptible and isolates > 8 µg/mL as resistant.In EUCAST v.13.0,CHL breakpoints are no longer listed, stating a screening cut-off (CHL MIC > 16 µg/mL) can be used to distinguish wild type from acquired resistance.

Functional CAT assay (rCAT)
We adjusted the rapid CAT assay from [19] to enable read-out in 96-well plates using a spectrophotometer.Cultures of isolates to be tested were set up in replicates in 1-2 mL LB and incubated at 37 ºC with 220 rpm. 100 µL of overnight culture was transferred into a microcentrifuge tube containing 500 µL PBS and centrifuge at 8000 rpm for 3' to wash and pellet cells.The supernatant was completely removed using a pipette, the pellet resuspended in 250 µL lysis buffer (1 M NaCl, 0.01 M EDTA, 0.05 % SDS) and cells lysed while incubating for 1 h at 37 ºC with occasionally vortexing the tubes.Next, 50 µL of the lysed cells were transferred into each of 4 wells of a 96-well plate.To each well 50 µL reaction buffer was added (for 1 mL, 400 µL 0.2 M Tris; 400 µL 5 mM acetyl-coA; 200 µL 10 mM DTNB were mixed).DTNB was added last to the wells, since DTNB was reported to inhibit certain CAT enzymes [19].DTNB was always prepared fresh since stored DTNB (for even 1 day at 4 ºC in the dark) increased the background.Acetyl-coA was always prepared on ice and aliquots immediately frozen at -20 ºC and used within 2 weeks.50 µL 5 mM CHL was added to the first two wells and 50 µL ddH2O to the wells three and four of each isolate.The plate was incubated for > 10 min at 37 ºC -longer incubation led to a higher signal for positive samples while the background did not change -up to 1 hour tested.
The plate was read with a plate reader (GloMax® Discover Microplate Reader (Promega) or CLARIOstar Plus Microplate Reader (BMG Labtech)) at absorbance of 405 nm or 412 nm and absorption from lysed cells without CHL was subtracted from the value with CHL, hence, positive values indicate CAT activity, no activity give a value around zero.

Chemical competent cells and transformation
MG1655 were made chemically competent as follows: A single colony of MG1655 was picked from an LB plate and incubated in 3 mL LB broth on at 37 ºC with 220 rpm for 16 hours.The culture was diluted 1:100 (1 mL in 100 mL) in a 500 mL flask and incubated at 37 ºC with 220 rpm until OD600 reached 0.3 -0.4.The flask was immediately cooled in an ice-water slurry before the cells were transferred to pre-chilled 50 mL Falcon tubes.Cells were centrifuged at 4000 rpm at 4 ºC, supernatant discarded and the cell pellet resuspended in 10 ml ice-cold CaCl2 solution (60 mM CaCl2, 15% glycerol, 10 mM PIPES pH 7).The centrifugation step was repeated, and resuspension was kept on ice for 30 minutes.After another centrifugation step the cells were resuspended in 1 mL ice-cold CaCl2 solution and aliquots of 50 µL or 100 µL were frozen at -70 ºC.
For transformation, the chemically competent cells were thawed on ice. 2 µL of plasmid DNA (pEB1-variants; 0.5 ng/mL) was added and incubated on ice for 30 minutes.Cells were heatshocked for 30 seconds at 42 ºC in a water bath and immediately put back on ice for 2 minutes.900 µL SOC media was added to cells and incubated for 1 hour at 37 ºC 220 rpm before being spread on LB kanamycin (50 µg/mL) plates.

Cloning and functional expression of cat genes
Cat gene variants were amplified using primers spanning the coding sequence with 20 bp overhangs homologous to the pEB1-plasmid cloning site (supplementary Table S2).

Experimental evolution of CHL resistance
Isolates were retrieved from frozen stock on LB agar plates.A single colony was used to inoculate a 1 mL starting culture and incubated at 37 ºC with 220 rpm (the ancestor).Then, for each tested isolate the starting culture was diluted 1:100 into 3 tubes with 3 mL LB broth (control populations) and 3 tubes with LB and 0.5x the MIC of CHL (selected populations) of the isolate.All population were incubated for 24 h at 37 ºC with 220 rpm and then again diluted 1:100 in fresh LB broth with or without CHL.In the selected population the CHL concentration was doubled with each passage.This was repeated for 7 days or until the population became extinct.

DNA extraction
DNA was extracted with MasterPure™ Complete DNA & RNA Purification Kit (MC89010, LGC Biosearch Technologies).Isolates were grown overnight on blood agar (PP0120-9090, E&O labs) at 37°C and colonies were picked and washed in 500 µl of sterile PBS (10010023, Fisher Scientific) by centrifugation at 10,000 x g for 3 minutes.Pellets were resuspended in 300 µl Tissue and Cell Lysis Solution supplemented with 1 µl of Proteinase K solution.Tubes were incubated at 65°C shaking at 1000 rpm for 5 minutes.Samples were placed on ice for 3 minutes.1 µl of RNase A solution was added and the samples incubated at 37°C for 30 minutes.
Samples were placed on ice for 5 minutes then 150 µl of MPC Protein Precipitation Reagent was added and samples vortexed for 10 s.The resulting protein debris was pelleted by centrifugation at 10,000 x g for 10 minutes at 4°C.The supernatant was aspirated and added to 500 µl of isopropanol (15631700, Fisher Scientific) and the tube inverted 30-40 times to precipitate the nucleic acids which were then pelleted by centrifugation at 10,000 x g for 10 minutes at 4°C and supernatant discarded.The pellet was washed twice in 70% ethanol and resuspended in 50 µl of TE Buffer.The quantity of DNA was then measured using Qubit dsDNA Quantitation, Broad Range kit (Q32850, Thermo Scientific) on a Qubit 4 Fluorometer (Thermo Scientific).

Long-read sequencing
Five E. coli from the subset of 42 isolates with a genotype-phenotype mismatch, were restreaked from glycerol stocks onto antibiotic supplemented LB Agar and incubated at 37°C for 16 h.A single colony was picked and transferred to LB broth supplemented with antibiotics.
Antibiotic supplementation included chloramphenicol and/or ampicillin depending on the resistance profile of each strain.Genomic DNA was extracted using the Firemonkey High Molecular DNA Extraction Kit (Revolugen, UK) according to the manufacturers protocol.
The strains were long-read sequenced on a MinION device (Oxford Nanopore Technologies (ONT), UK).Library prep was carried out according to the manufacturer's protocol (ONT, UK) using the SQK-NBD114.24Ligation Sequencing and Native Barcoding Kit.The DNA library was quantified at several stages in the library prep using a Qubit Fluorometer (ThermoFisher Scientific, Massachusetts, USA), the TapeStation 4100 (Agilent, California, USA) was used to determine the molar concentration and 10-20 fmol were loaded onto the flow cell.Sequencing was carried out using a FLO-MIN114 (R10.4.1) flow cell (ONT, UK) on a MinION Mk1B sequencer, running for 72 h at a translocation speed of 400bp/s.Data acquisition used the MinKNOW software (v22.08.9).

Bioinformatic analysis
programming language v 4.3.1 was used to convert the abricate output tables into an appropriate format for further data analysis and visualisation, including a binary presence/absence AMR gene table, with the tidyverse (v1.3.0)package [63].Any annotation of catB3 with less than 75% coverage was labelled catB4.The cooccur package v1.3 [64] was used to create a co-occurrence matrix containing the probabilistic values which represent whether a co-occurrence relationship is observed significantly more or less than could have happened by chance.This used the probabilistic model of co-occurrence [34].If a pair of genes were observed to co-occur significantly less than expected by chance their relationship was termed negative, if they were observed to co-occur significantly more it was termed positive and any pair of genes which were observed to co-occur with no significant difference from their expected value were termed random.A significant difference was defined as having a p-value £ 0.05.Select genes conferring resistance to the phenicol, beta-lactam and aminoglycoside drug classes were then selected from the original presence/absence table for further analysis.Heatmaps to display co-occurrence were visualised using ggplot v 3.4.3[63] and pheatmap v 1.0.12.Clustering in the heatmap used hierarchical clustering with Euclidian distance and the complete method.

ST analysis
The genome assemblies of 100 randomly selected isolates from each of the 100 most common STs in Enterobase were downloaded on 18/12/2020.Extra genomes from ST391, ST44, ST940 and ST9847 with release date before 18/12/2020 were added later to match with the STs from the Malawi dataset.The STs of the Malawi dataset were determined using MLST (v2.23.0).The presence of cat genes was detected using AMRFinderPlus version

Fig. 1 |
Fig. 1 | Characteristics of isolates.a Chloramphenicol susceptibility of E. coli and KpSC isolates.b Number of CHL susceptible isolates carrying cat genes, floR or cmlA genes, both (cat + floR or cmlA) or no CHL resistance genes.c Number of CHL resistance genes present in the 840 Malawian isolates.

Fig. 2 |
Fig. 2 | CAT enzyme activity.CAT enzyme activity was measured using the rapid CAT assay.The difference ((signal + CHL) -(signal -CHL) for a single isolate (n = 2-6)) in absorbance at 405 nm is given for each of the 42 isolates.Colour indicates if the isolate is phenotypically resistant (R, blue) or susceptible (S, yellow) to chloramphenicol based on microbroth dilution and the shape indicates E. coli (circle) or K. pneumoniae (triangle).The Y axis is ordered according to presence of cat genes.

Fig. 3 |
Fig. 3 | Cat gene degradation by insertion sequences.a Schematic of IS26 truncation of catB3.b Alignment of four contigs from four different isolates containing either catB3 and catB3∆ 443-633 .c Alignment of two genomes containing catA1, with and without an IS5 insertion into the catA1 promoter.

Fig. 4 |
Fig. 4 | Co-occurrence networks of AMR genes.Heatmap displaying co-occurrence relationships between AMR genes as either positive (blue), random (grey) or negative (orange).These are probabilistic values based on the difference in expected and observed frequencies of cooccurrence between each pair of genes, these values were obtained by applying the probabilistic model from [34].Co-occurrence across select genes including cat genes, aac(6')-Ib-cr, blaOXA-1, blaCTX-M-15 and blaTEM-1.

Fig. 5 |
Fig. 5 | Cat genes occurrence and proportions per Sequence type in E. coli isolates from Malawi included in our study (top panel) and in a 10k genome collection of 100 randomly selected genomes of the top 100 most common STs from E. coli.
4.0) in cation adjusted MullerHinton broth (MH2, 90922, Merck) in duplicates.If results were inconsistent or different to the CHL susceptibility phenotype previously determined with disc diffusion, AST was repeated.Broth microdilution data was used to determine the phenotypic classification of the subset of 42 isolates.Isolate are classified according to EUCAST (v.12.0);