The distribution of fitness effects of plasmid pOXA-48 in clinical enterobacteria

Antimicrobial resistance (AMR) in bacteria is a major public health problem. The main route for AMR acquisition in clinically important bacteria is the horizontal transfer of plasmids carrying resistance genes. AMR plasmids allow bacteria to survive antibiotics, but they also entail physiological alterations in the host cell. Multiple studies over the last few years have indicated that these alterations can translate into a fitness cost when antibiotics are absent. However, due to technical limitations, most of these studies are based on analysing new associations between plasmids and bacteria generated in vitro, and we know very little about the effects of plasmids in their native bacterial hosts. In this study, we used a CRISPR-Cas9-tool to selectively cure plasmids from clinical enterobacteria to overcome this limitation. Using this approach, we were able to study the fitness effects of the carbapenem resistance plasmid pOXA-48 in 35 pOXA-48-carrying isolates recovered from hospitalized patients. Our results revealed that pOXA-48 produces variable effects across the collection of wild-type enterobacterial strains naturally carrying the plasmid, ranging from fitness costs to fitness benefits. Importantly, the plasmid was only associated with a significant fitness reduction in four out of 35 clones, and produced no significant changes in fitness in the great majority of isolates. Our results suggest that plasmids produce neutral fitness effects in most native bacterial hosts, helping to explain the great prevalence of plasmids in natural microbial communities.


Analysis of mutations affecting coding regions in cured clones.
The genomes of the cured strains presented in total eight-point mutations in coding genes (Table S3). Four SNPs were predicted to have neutral effects on protein function: a synonymous variant in the iron donor protein CyaY in C288c2, a synonymous variant affecting the putrescine ABC transporter ATP-binding subunit PotG in C325c1, and two missense variants affecting a phage capsid protein in CF12c1 (SNAP2 score -88, accuracy 93%) and the type VI secretion system baseplate subunit TssF in H53c1 (SNAP2 score -64, accuracy 82%). It is worth noting that these mutations could indirectly impact protein function by altering gene expression 1,2 . The remaining four SNPs were predicted to alter protein function. The frameshift mutation affecting gene pdeR in CF12c1 compromises protein secondary structure ( Figure S6), producing the elongation of the C-terminus (+22 amino acids), which comprises the EAL domain of PdeR.
Mutations in this protein, a member of the curli fimbrae biosynthesis cascade, are related to increased biofilm formation 3 . Another mutation restored the reading frame of the frameshifted znuB gene in H53c1, a subunit of the zinc ABC transporter complex. The third variant introduced a premature stop codon that results in the loss of more than a third of an integrase domain-containing protein in strain K147c1. The loss of function of this protein would probably only affect integron biology, impeding its excision. Thus, it should not produce large-scale physiological changes in the cured strain with respect to the wild-type strain. The fourth SNP also introduced a premature stop codon in the ompC gene of C288c2. This resulted in a truncated version of the protein, cutting down from 367 to 171 amino acids. The nucleotide sequences of the mutated genes were submitted to a BLASTn search against the NCBI database. In C288c2, the synonymous SNP in the cyaY gene was also found in other enterobacteria (e.g. CP077379.1). In the case of H53c1, SNPs present in both znuB (as seen earlier, Figure S7) and tssF were also present in other enterobacteria (e.g. CP043597.1 and CP081896.1, respectively), suggesting these are naturally-occurring variants. Lastly, we detected two probable insertion sequence (IS) rearrangements. One of them was a possible event of an IS jump from the IncF plasmid of J57c1 to its chromosome, inserting into the yhjQ gene and probably producing a knock-out. This gene is involved in the biosynthesis of cellulose, and its disruption can reduce bacterial aggregation and biofilm formation 4 . The other one was observed in K153c2: an IS1 element from its chromosome was integrated within a coding region of a putative AAA family ATPase, which are involved in a wide variety of functions, potentially producing a knock-out.

Analysis of mutations affecting intergenic regions in cured clones.
We analysed if the observed intergenic SNPs could be located within putative promoters.
In the case of CF12c1, the SNP located 50 bp upstream glpA did not apparently fall within predicted promoters in the sense strand that could control the expression of glpA and downstream genes ( Figure S8A). Still, this mutation could affect other regulatory regions or the transcription of non-coding RNAs. In J57c1, the SNP located 94 bp upstream the catecholate siderophore receptor fiu gene is positioned within a predicted promoter in the sense strand (final score 2.72) and near another promoter with higher score (2.79) ( Figure S8B). Therefore, the expression of the fiu gene and downstream genes could be altered. In the IncF plasmid of J57c1 there are multiple intergenic mutations between positions 90533-90554 bp that could be due to the possible excision and insertion of the IS element into the chromosome of J57c1. These SNPs fall near predicted promoters for the sense strand, and thus, could affect transcription of the ISNCY-like element ISKpn21 family transposase ( Figure S8C). Nonetheless, these mutations had low read coverage and alignment quality (multiple contiguous SNPs), and thus could represent false positive calls. In K153c2, another IS1 element was observed integrated into an intergenic region between the fyuA gene and a hypothetical protein, 77 bp upstream the latter. As with mutations in coding regions, the sequences of the mutated intergenic regions were also submitted to a BLASTn search against the NCBI database. However, we could not find the observed mutations in other organisms.
Analysis of possible off-target events. We aligned the sequences of the single-guide RNAs (sgOXA48 and sgPemK) to the regions surrounding the identified SNPs (see Methods). To recognize and produce off-target cuts, the alignment of the sgRNA must have less than eight mismatches, short or no gaps, and more importantly, the Cas9 protospacer adjacent motif (PAM) sequence NGG or NAG at the 3' of the sgRNA 5 . Most alignments had gaps and/or more than seven mismatches and lacked the PAM sequence. Only two alignments had less than eight mismatches and presented the PAM sequence NGG, constituting possible off-target events ( Figure S9). First, sgOXA48 aligned against the region encompassing the znuB mutation in H53 -that restores the reading frame of znuB in H53c1-with only five mismatches ( Figure S9A). However, the mutation affecting znuB in H53c1 is a naturally-occurring variant. In fact, we could not find any enterobacteria in the NCBI database that carried the frameshift variant of H53.
Second, the sgPemK guide aligned against the region surrounding the ompC mutation in C288 ( Figure S9B). However, the combination of gaps and/or mismatches located within the PAM-proximal region of the sgRNAs would most probably impede Cas9