Genomic inversion drives small colony variant formation and increased virulence in P. aeruginosa

Phenotypic change is a hallmark of bacterial adaptation during chronic infection. In the case of chronic Pseudomonas aeruginosa lung infection in patients with cystic fibrosis, well-characterised phenotypic variants include mucoid and small colony variants (SCVs). It has previously been shown that SCVs can be reproducibly isolated from the murine lung following the establishment of chronic infection with mucoid P. aeruginosa strain NH57388A. Here we show, using a combination of singlemolecule real-time (PacBio) and Illumina sequencing that the genetic switch for conversion to the SCV phenotype is a large genomic inversion through recombination between homologous regions of two rRNA operons. This phenotypic conversion is associated with large-scale transcriptional changes distributed throughout the genome. This global rewiring of the cellular transcriptomic output results in changes to normally differentially regulated genes that modulate resistance to oxidative stress, central metabolism and virulence. These changes are of clinical relevance since the appearance of SCVs during chronic infection is associated with declining lung function.


Introduction
Phenotypic variation is a hallmark of adaptation to the host during chronic bacterial infection.
P. aeruginosa is the major proven cause of mortality in patients with CF and chronic infection leads to a progressive decline in pulmonary function and inevitably respiratory failure 22 . Despite intensive anti-pseudomonal chemotherapy greatly improving the prognosis for CF patients 23 , the current median age at death for CF patients is around thirty years in developed countries 24 . The frequent failure of antibiotic therapy and host defences to eradicate P. aeruginosa from the CF lung is thought to be largely due to the increased antibiotic tolerance when growing in the biofilm state and the appearance of mucoid phenotyic variants that are a hallmark of adaptation in the chronically infected lung. A further complicating factor is the appearance of highly adherent SCVs that are adept at biofilm formation 18,19,25 . P. aeruginosa SCVs may display high intracellular c-di-GMP levels 19,20,[26][27][28] , enhanced biofilm formation, high fimbrial expression, repression of flagellar genes, resistance to phagocytosis and enhanced antibiotic resistance. Most importantly, the appearance of SCVs in the CF lung correlates with poor patient clinical outcome 11,17,29-31 . There are a range of genetic changes that have been shown to be responsible for the phenotypic switch to the SCV phenotype in P. aeruginosa, including mutations in the Wsp system and yfiBNR operon that form part of the c-di-GMP regulatory system in P. aeruginosa [32][33][34][35][36] . However, identification of the major clinically relevant pathways of conversion to the SCV phenotype are complicated by the unstable phenotype displayed by many SCVs with reversion to a normal colony phenotype frequently observed preventing successful comparative genetic studies on clinical SCVs and their closely related parent strains. In the case of S. aureus, which also forms clinically relevant SCVs, recent work has shown that a reversible large scale chromosomal inversion is the genetic basis of the switch between a normal colony and SCV isolated from the same patient 37 . In addition, S. aureus SCVs, which are commonly isolated from the CF lung, are highly resistant to oxidative stress, suggesting that conversion to the SCV phenotype may be an adaptation to environment present in chronically inflamed host tissue 38 .
In this work, we have determined the genetic basis of phenotypic conversion from the mucoid to the SCV phenotype for SCVs isolated from the chronic lung infection model described by Bayes et al 2016 39 . For two SCVs isolated from this work, we have shown through a combination of singlemolecule real-time (SMRT, Pacific Biosciences) and Illumina sequencing that a large and stable chromosomal inversion accompanies conversion to the SCV phenotype. Genome inversion is accompanied by transcriptional changes to a large number of genes that most notably includes downregulation of a number of genes encoding metabolic enzymes, DNA repair proteins and heat shock proteins and the upregulation of genes encoding proteins involved in the response to oxidative stress. The absence of other obvious genetic change suggests that this chromosomal inversion is the genetic basis of conversion to the SCV phenotype.

Results
P. aeruginosa SCVs are commonly isolated from patients with cystic fibrosis and have been isolated in vitro as well as from experimental infection models following aminoglycoside treatment 18,31,40 . The work of Bayes et al describes the isolation and partial characterisation of SCVs isolated from a chronic murine P. aeruginosa lung infection model 39 . In this model, animals were inoculated with P. aeruginosa strain NH57388A (NHMuc) a mucoid clinical isolate, embedded in agar beads. NHMuc has a known mutation in the gene encoding the anti-sigma factor MucA, that results in alginate overproduction 41,42 . Recovered bacteria from lung homogenate samples display two distinct colony morphologies: typical large mucoid colonies identical in morphology to the inoculating strain and SCVs. Mucoid colonies were evident after 24 hours of growth on agar plates at 37°C with SCVs visible only after 48 hours of growth on agar plates 39 .
To understand the genetic basis of this phenotypic change, we initially performed Illumina HiSeq whole-genome sequencing and genomic comparison between NH and two separate SCVs (SCVJan and SCVFeb) isolated from independent in vivo experiments. However, despite their gross phenotypic differences, this analysis failed to identify any genetic differences between the SCVs and the parent strain.
Next, we utilised the ultra-long reads produced by single-molecule real-time PacBio sequencing to attempt to identify any large scale genome rearrangements that could drive conversion to the SCV phenotype. Using this technique we identified a large scale genomic inversion accompanying conversion from the parent mucoid to SCV phenotype in both SCVJan and SCVFeb.  Figure 2b). There were no further differences in the number of protein coding genes (5619), rRNAs (12) or tRNAs (57) between SCVs and the parent strain. No SNPs could be identified in protein coding genes. A similar genome inversion was not identified (using a PCR based strategy) in a mucoid strain (NHMucJan) that was phenotypically identical to the parent strain (NHMuc) and isolated from the same chronic infection model as SCVJan ( Figure S1). Interestingly, comparison of the SCVJan and SCVFeb genomes with that of an SCV (SCV20265) isolated from a CF patient 43 , which has recently been sequenced by PacBio sequencing, revealed an almost identical chromosomal inversion. However, in the case of SCV20265 the inversion was not accompanied by truncation of the 16S rRNA gene in the third rRNA operon (Figure 1).

Transcriptional and phenotypic changes on conversion to the SCV phenotype
To determine the transcriptional changes associated with conversion to the SCV phenotype, we performed RNA sequencing (RNA-Seq) analysis of the parent strain, NHMuc, and two SCV strains grown in LB broth. Initial analysis showed that SCVJan and SCVFeb have highly similar gene expression profiles that are distinct from that of NHMuc. RNA-Seq data for all strains was collected in triplicate and data for SCVJan and SCVFeb were combined to compare with NHMuc. Relative to NHMuc, 190 genes showed >2-fold upregulation and 364 genes showed >2-fold downregulation in SCVJan/SCVFeb (Table S1). Interestingly, the transcriptional changes associated with genomic inversion and that drive conversion to the SCV phenotype are not restricted to genes close to or within the inversion breakpoints, with major upregulated and downregulated genes distributed relatively evenly throughout the genome (Figure 3).

Major functional classes of genes downregulated in SCVJan/SCVFeb include those involved
in energy metabolism, amino acid and protein biosynthesis, DNA replication and recombination and cell wall/LPS/capsule biosynthesis, which together are consistent with the slow growth rate observed for SCVs. Notably, genes encoding heat shock proteins and other molecular chaperones (IbpA, GrpE, HtpG, ClpB, DnaK, GroES, DnaJ and ClpX) are highly represented among the most strongly downregulated genes in the SCVs (Table 1). Conversely, genes that function in the response to oxidative stress and those that encode secreted virulence factors are largely upregulated in SCVJan/SCVFeb. Indeed, five of the ten most highly upregulated genes in SCVJan/SCVFeb are those associated with the response to oxidative stress (Table 2). Highly upregulated oxidative stress genes include, katA 44 which encodes the major catalase of P. aeruginosa, ahpB, ahpC and ahpF [45][46][47] that encode subunits of alkyl hydroperoxide reductase and trxB2 that encodes thioredoxin reductase 2 [48][49][50] . Consistent with the observed transcriptional changes, catalase activity was strongly increased in SCVJan relative to NHMuc ( Figure 4A).
Genes encoding a number of secreted virulence factors such as the proteases LasA, LasB and AprA, the frucose-binding lectin LecB 51 and the chitin binding protein CbpD and chitinase ChiC 52 were also highly upregulated. Similarly genes encoding hydrogen cyanide synthase and a number of enzymes that function in phenazine biosynthesis are also upregulated (Table S1) 28,53-57 .
Phenazines have previously been shown to enhance killing of Caenorhabditis elegans by P. aeruginosa 58

Discussion
In this work we show that P. aeruginosa SCVs isolated from a chronic murine lung infection model show a general upregulation of virulence associated genes, relative to the mucoid parent strain, and increased virulence which may begin to explain the link between the appearance of SCVs in chronic lung infection and the associated decline in lung function 60 . In addition, the upregulation of genes that mediate the response to oxidative stress immediately suggests why the isolated SCVs are rapidly selected for in a chronic infection model in which the host immune system is strongly activated.
A key strength of our study was the availability of the parent strain used to establish infection for sequencing. This allowed a meaningful comparative genetic analysis to be performed enabling the determination for the genetic basis of conversion to the SCV phenotype. Surprisingly, SNPs and INDELs were not identified in the SCV genome by Illumina sequencing and single-molecule realtime sequencing was subsequently used to show that the two sequenced SCVs carried a large genomic inversion within 16S rRNA genes. Interestingly, the transcriptional changes associated with genomic inversion are not restricted to genes close to or within the inversion breakpoints, but distributed relatively evenly throughout the genome. Instead the major changes in gene expression are largely restricted to specific functional classes of genes including those that mediate the response to oxidative stress, virulence, DNA repair and recombination, the chaperone network and metabolism. This global rewiring of the cellular transcriptomic output results in concerted transcriptional changes to these normally differentially regulated genes. However, the mechanisms that underlie these transcriptional changes are not clear. A recent study in E. coli suggests that position effects on gene expression may be due to local differences in chromosomal structuring and organization, with DNA gyrase playing an important role at certain high activity locations. Further studies will be needed to clarify such positional effects on gene expression in the SCV studied here 61 .
The mechanistic details of how the observed genomic inversion lead to the coordinated expression changes observed here is currently not known, but the observation that a clinical SCV strain (SCV20265) obtained from the lung of a CF patient strain recently sequenced using the same strategy of combining SMRT and Illumina sequencing possesses a similar 16S rRNA based inversion, indicates that this inversion may be a highly clinically relevant route to the SCV phenotype ( Figure 1) 43 . Other large scale genome rearrangements including large chromosomal inversions have previously been described in P. aeruginosa but these were not associated with conversion to the SCV phenotype 62,63 . A reversible genomic inversion has also recently been shown to mediate the reversible conversion between normal colony and SCV phenotypes in S. aureus 37 . However, in the case of the SCVs isolated in our work the SCV phenotype is stable and revertants to the parent phenotype were not observed. A possible explanation for this observation is that a number of genes encoding proteins involved in DNA repair and recombination, including RecA, are downregulated in the SCV relative to the parent strain (Table 1).
In conclusion, we have shown that a P. aeruginosa SCV originated in the lungs of an animal with chronic colonization appears to result from a large chromosomal inversion and associated largescale transcriptional changes. SCVs have a clear selective advantage in the context of the CF lung, and a better understanding of the drivers that produce the genomic rearrangement observed in this study may provide alternative therapeutic approaches to prevent the appearance of such damaging phenotypic variants.

Materials and Methods
Genome assembly and annotation. Bacterial suspensions were grown to early stationary phase to an OD600 of 1.8 in LB broth at 37 in a shaking incubator. 2 ml of each suspension was pelleted at 12000g for 10 minutes. RNA was extracted from samples using a bead beating/chloroform extraction method as previously Five-microliter aliquots of the serial dilutions were injected using a Hamilton syringe into G. mellonella larvae, via the hindmost left proleg as previously described 70 . Ten larvae were injected per dilution for each Pseudomonas strain tested. Larvae were incubated in 10-cm plates at 37 °C and the number of dead larvae scored 1 to 4 days after infection. For each strain, data from 3 independent experiments were combined. Larvae were considered dead when they displayed no movement in response to touch. A negative control was used in each experiment to monitor the killing due to physical injury or infection by pathogenic contaminants. Time to death was monitored every 24h post infection. In any instance where more than one control larvae died in any given experiment, the data from infected larvae were not used.     primer specific to birA, which is adjacent to the rRNA genes within the inverted region. Only the primer pair expected to give a PCR product for each strain is shown. The inverted sequence in the SCV strain is highlighted in red. b) Consistent with the hypothesis that genome inversion drives conversion we observed a PCR product only with birA-F/yedZ-R primer pair for NHMucJan indicating that the genome structure in this region is identical to the parent strain NHMuc.