Comparative genomics of non-pseudomonal bacterial species colonising paediatric cystic fibrosis patients

Sample collection and quantification

Bronchoalveolar lavage (BAL) samples were collected as part of the Australasian Cystic Fibrosis BAL Study that was approved by the Royal Children’s Hospital Ethics Committee RCH HREC PROTOCOL NO.: 1998/001. Informed consent was obtained from the children’s parents for sample collection, storage and subsequent testing. All BAL samples were collected under general anaesthesia using standard procedures. Five ten-fold serial dilutions from 500 µL of BAL fluid (BALF) were used for quantitative colony counts. 100 µL of undiluted BALF and each dilution were plated onto six different selective media: horse blood (bioMérieux, Marcy-l’Étoile, France), mannitol salt (BD Biosciences, San Jose, California, USA), MacConkey (bioMérieux, Marcy-l’Étoile, France), chocolate bacitracin (CBA, Oxoid, UK), cetrimide (Oxoid, UK) and Burkholderia cepacia (Mast Group Ltd, Bootle, Merseyside, UK) agar. For sputum samples, 10 µL of the undiluted sample was plated directly onto selective media plates. Plates were then incubated at 37 °C (5% CO₂ for CBA) and read at 24 and 48 h. The colony count was determined from plates bearing between 30 and 100 individual colonies. The final count is expressed as colony forming units per ml of BALF for each bacterial colony type of interest. Colony type identification tests were carried out as required: API 20 NE (bioMérieux, France, non-fastidious, non-enteric Gram-negative rods), VITEK GN-ID (bioMérieux, France, Gram-negative bacilli), coagulase production and latex agglutination (S. aureus), optochin susceptibility (Streptococcus pneumoniae) and X/V growth factor requirement (Haemophilus).

Purification and antibiotic susceptibility testing

Between 1 and 4 colonies from selective media plates of each sampled species were diluted to 0.5 McFarland standard and then replated, cultured and stored at −80 °C. This same dilution was used as the inoculum for antibiotic susceptibility testing: Clinical and Laboratory Standards Institute (CLSI) disk diffusion testing for Pseudomonas, Streptococcus and Haemophilus and VITEK (bioMérieux, France) instrument analysis and comparison with CLSI minimum inhibitory concentration breakpoints for other species.

DNA extraction and sequencing

DNA was extracted from pure cultures of each selected isolate using the PowerSoil DNA isolation kit (MO BIO Laboratories Inc, Carlsbad, California, USA). Indexed whole-genome sequencing libraries were prepared with the Nextera DNA sample preparation kit (Illumina, San Diego, California, USA). Libraries were pooled such that each isolate was approximately equal in concentration. This pool of libraries was submitted to the Queensland Centre for Medical Genomics (University of Queensland, Australia), and half a lane of 2 × 100-bp paired-end data was generated on an Illumina HiSeq2000 sequencer.

Draft genome assembly

Reads were trimmed using Trimmomatic v0.30 (Bolger, Lohse & Usadel, 2014) and assembled using SOAP de novo v1.05 (Luo et al., 2012). Contigs were ordered against reference genomes using CONTIGuator v2.7.3 (Galardini et al., 2011) and gaps were closed where possible using IMAGE (Tsai, Otto & Berriman, 2010) within the PAGIT wrapper v1.64 (Swain et al., 2012). Genomes were submitted to RAST for annotation (Aziz et al., 2008). Draft assemblies were compared with reference genomes using progressiveMauve (Darling, Mau & Perna, 2010). Contamination and completeness of each draft assembly were estimated using CheckM v0.9.7 (Parks et al., 2015). ANI was calculated using ANI calculator (http://enve-omics.ce.gatech.edu/ani/index, last accessed 2 June 2015).

For isolates of the same species, each strain was also mapped to a reference genome selected based on the lowest percentage of unmapped reads. Mapping to reference genomes was performed using BWA 0.6.2-r126 (Li & Durbin, 2009) using default settings. Duplicate reads were marked using Picard 1.41 (http://www.picard.sourceforge.net) and base quality score recalibration and realignment around indels was performed using GATK 2.4-9-g532efad (McKenna et al., 2010). SNPs and indels were identified using GATK 2.4-9-g532efad following best practice guidelines (DePristo et al., 2011). Larger variation was detected using Breakdancer 1.1.2 (Chen et al., 2009) and CREST (Wang et al., 2011). Mappings were visualised and SNP associated amino acid changes were determined using CLC Genomics Workbench (Qiagen, Velno, Netherlands). Unmapped reads were collected using SAMtools (Li et al., 2009) and regions of zero coverage were identified using BEDTools (Quinlan & Hall, 2010).

Comparative genome analysis

Antibiotic resistance and virulence related genes were identified using BLAST with the Comprehensive Antibiotic Research Database (McArthur et al., 2013) and the Virulence Factor Database (Chen et al., 2012) in addition to examining RAST output and manual comparison with various annotated reference genomes. For comparison with the CARD, a cut off of ≥40% identity over ≥60% of the query sequence was required. For comparison with the VFDB, a cut off of over 60% coverage of the target and over 80% identity was required. BLAST was used to identify candidate instances of horizontal gene transfer using a cut off of ≥95% over 500 bp. Genomic islands and prophage regions were detected using Island Viewer (Dhillon et al., 2013) and PHAST (Zhou et al., 2011). Maximum likelihood “genome” trees were inferred using MEGA (Tamura et al., 2013) with 500 bootstrap replicates based on concatenated alignments of 83 single copy marker genes from isolates and reference strains using CheckM (Parks et al., 2015).

Staphylococcus aureus

S. aureus is a commensal organism commonly found inhabiting the nasal passage (Wertheim et al., 2005). When disease occurs, it is usually caused by the patient’s own colonising strain (Von Eiff et al., 2001). In CF patients, S. aureus can contribute to inflammation (Sagel et al., 2009) and to the development of pneumonia (Koulenti et al., 2009). We sequenced six S. aureus isolates; four were obtained from patient A and a single isolate was obtained from both patient B and C (Fig. 1). Draft assemblies were of similar size with ∼170 kb (6%) separating the smallest, B6, from the largest, A1 (Table 1). Multilocus sequence type analysis classified A1 as sequence type ST1, A3, 4 and 5 as ST15, C9 as ST6 and B6 as ST953 (Enright et al., 2000). Contigs were ordered against MSSA476 (NCBI acc. no. BX571857), a community acquired strain isolated in 1998 from an immunocompetent child with a severe invasive infection (tibial osteomyelitis) and bacteraemia (Holden et al., 2004). The patient A strains A3–A5 appeared substantially different from the earliest isolated strain A1 while being very similar to each other suggesting the persistence of a single strain for a year (Figs. 1 and 3A). S. aureus infections are typically persistent, with over 80% of patients colonised by the same lineage for over two years (Branger, Gardye & Lambert-Zechovsky, 1996; Vu-Thien et al., 2010). SNP analysis within the patient A strains further confirmed the close relationship between A3, 4 and 5; A4 and A5 share one SNP that causes a synonymous change and A5 contains two SNPS, both of which are intergenic. Between A1 and A3, 4 and 5, large differences are associated with prophage regions and genomic islands (Fig. 3A).

Achromobacter xylosoxidans

A. xylosoxidans is an opportunistic, multi-drug resistant organism found in a variety of aqueous environments including water distribution systems (Amoureux et al., 2013a). While its role in CF is still under investigation, A. xylosoxidans has been shown to induce inflammation in these patients (Hansen et al., 2010). Chronic infection is also associated with declining lung function in some cases (Rønne Hansen et al., 2006). We sequenced seven A. xylosoxidans isolates, all from patient A. The initial strain originated from a BAL sample while all subsequent isolates were obtained from sputum samples. Assembly revealed a similar genome size for all strains except A8 which was ∼100 kb smaller than the other strains (Table 1). Contigs were ordered against NH44784-1996 (NCBI acc. no. HE798385), an isolate obtained from a CF patient in 1996 (Jakobsen et al., 2013). All A. xylosoxidans strains had a high degree of synteny (Fig. 3B) suggesting that they represent a single population that persisted from 2006 to 2011 (Fig. 1). This was confirmed via SNP analysis that revealed only a small number of unique mutations in all isolates except A7 (Table S5). A8–A13 also share five SNPs and A9–A13 share 17 SNPs and 3 indels. A11 contains considerably more mutations than the other strains which points to the possibility of an increased mutation rate in this isolate. Non-synonymous SNPs were found in three DNA repair associated proteins: RadC (A>G-Asp146Gly), MutL (C>T-Pro457Ser) and MutS (A>C-Thr810Pro). The transition:transversion ratio (∼10:1) supports a defect in MutS contributing to this mutation signature (Table S5) (Young & Ornston, 2001).

Stenotrophomonas maltophilia

S. maltophilia is an opportunistic, multi-drug resistant organism that is found in soil and aqueous environments also including water distribution systems (reviewed in Brooke, 2012). The incidence of S. maltophilia in CF patients varies (Goss et al., 2004; Dalbøge et al., 2011) but has been shown to be increasing in some locations (Emerson et al., 2010). Analyses of the impact of S. maltophilia infection are contradictory with some studies reporting little or no effect on lung function (Goss et al., 2004; Dalbøge et al., 2011) while others report poorer clinical status (Talmaciu et al., 2000) and an increased risk of pulmonary exacerbation with chronic infection (Waters et al., 2011). We sequenced five S. maltophilia isolates, three from patient B, one from patient A and one from patient C. Contigs were ordered against the reference strain K279a (NCBI acc. no. AM743169), isolated in 1998 from a bloodstream infection (Crossman et al., 2008). S. maltophilia strains were relatively divergent from each other compared to the S. aureus and A. xylosoxidans strains, even those from patient B (Fig. 3C) which is typical of S. maltophilia infection, particularly in young patients (Valdezate et al., 2001; Pompilio et al., 2011). Notably, average nucleotide identity (ANI) analysis revealed all strains lacked sufficient identity to the reference strain to be classified as the same species (<95%; Fig. 3C; Goris et al., 2007) despite forming a robustly monophyletic clade (Fig. S2). Comparison between available reference genomes also produced low identity e.g., 91.5% ANI between K279a and JV3 (NCBI acc. no. NC_015947). Based on this type of analysis, it has been suggested that strains currently designated as S. maltophilia actually represent a complex (Svensson-Stadler, Mihaylova & Moore, 2012) placing a caveat on interstrain comparisons.

S. maltophilia is intrinsically multi-drug resistant with resistance reported against multiple classes of antibiotics (reviewed in Brooke, 2012). A substantial degree of resistance is conferred by the presence of a number of efflux pump systems. The reference strain K279a has nine efflux pumps of which SmeZ confers resistance to aminoglycosides and SmeJ/K contributes minor aminoglycoside, fluoroquinolone and tetracycline resistance (Crossman et al., 2008). All nine are present within each of the analysed S. maltophilia isolates (Table S6). Other pumps are typically associated with resistance following mutations leading to over-expression; consequently their presence alone does not indicate resistance (reviewed in Brooke, 2012). We also examined the presence of other putative and known resistance genes annotated in the K279a genome (Crossman et al., 2008) (Table S6). All except two were found in at least one of the analysed isolates but with varying identity. This variation is known to contribute to differing degrees of antibiotic resistance (Avison et al., 2001) and further complicates resistance predictions based on genome data.

All S. maltophilia isolates were susceptible to sulfamethoxazole plus trimethoprim, the preferred treatment option for this species but against which resistance is becoming an increasing problem (Brooke, 2012) (Table 2). Patient B received sulfamethoxazole plus trimethoprim throughout the study period. Given the dissimilarity of the patient B strains it is possible the patient is being sequentially colonised by this species rather than there being a persistent population, suggesting in this regard antibiotic treatment was effective. However, it is also possible that there exists a diverse infecting population in this patient with a different lineage sampled at each time point. Patient A also received sulfamethoxazole plus trimethoprim after the isolation of A2 and no S. maltophilia was noted after this time. Like patient A, a single S. maltophilia isolate was obtained from patient C, however no sulfamethoxazole plus trimethoprim was administered in this case. Ticarcillin plus clavulanic acid was given following the isolation of C11 and may have cleared the infection in this patient.

Other CF isolates

Beyond the three main species described above, several other species were obtained that appeared sporadically in the patient profiles. Two Enterobacter cloacae strains were isolated, both from patient B. The genomes of the two E. cloacae isolates are very similar indicating persistence; we found only a single, intergenic SNP separating the two, however both contain unique sections associated with mobile elements (Fig. S3A). One of these regions in B3 is similar to antibiotic resistance elements from multiple species including E. coli (Tn2610, NCBI acc. AB207867; Tn21, NCBI acc. AF071413) and A. baumannii, (AbaR21, NCBI acc. KM921776) but with a different arrangement and contains an antiseptic resistance gene QacE delta 1, a sulfonamide insensitive dihydropteroate synthase Sul1 and a trimethoprim insensitive dihydrofolate reductase DfrA5. B3 was resistant to sulfamethoxazole plus trimethoprim supporting the activity of this element in the isolate (Table 2). Sulfamethoxazole plus trimethoprim had not been given to the patient for eight months prior to the isolation of B2 but was administered four times over the four month period separating B2 and B3, potentially providing the selective pressure for insertion of this element (Table S4). In addition, the similarity of the two E. cloacae isolates supports a lateral acquisition by the later isolate during infection. There was no similarity between the element and the other sequenced isolates from patient B and no BLAST hits were returned from Staphylococcus or Streptococcus species, which were also present in BAL samples during this period, indicating an unsampled source for this element. High sequence identity of the laterally transferred gene cassette with other members of the family Enterobacteriaceae (>95% identity), albeit with variable gene order (Fig. S4), suggest that this lineage likely harbours the donor species.

We also obtained two Haemophilus isolates, both from patient C. Neither strain contained the capsule biosynthesis locus and thus represent nontypeable H. influenzae (NTHi) which make up the majority of respiratory mucosal associated infections (Erwin et al., 2008). NTHi are a diverse population (Erwin et al., 2008), reflecting their ability to be both commensal and pathogenic, and this diversity is illustrated in the two analysed strains which were obviously dissimilar (Fig. S3B). Despite this dissimilarity, both strains displayed susceptibility to all three antibiotics tested: ampicillin, cefotaxime and sulfamethoxazole plus trimethoprim. Differences were observed, however, between the virulence factor genes present in the two strains. C10 contains the high-molecular weight adhesins HMW1 and HMW2 which are the major adhesins in ∼80% of NTHi strains (St. Geme et al., 1998) and are associated with acute otitis media (Krasan et al., 1999; Ecevit et al., 2004). Strains lacking HMW1 and HMW2 typically contain an alternative adhesin, hia (St. Geme et al., 1998), however C1 does not contain this gene either. C1 does however contain a haemagglutinating pili gene cluster consisting of five genes, absent from C10. These clusters are predicted to be important for nasopharyngeal colonisation (Krasan et al., 1999; Ecevit et al., 2004). C1 also contains a unique Type I CRISPR-Cas system (Makarova et al., 2011) resembling that of H. haemolyticus strain M21621 (NCBI acc. NZ_AFQQ01000000). Four spacer sequences were contained within the region. No CRISPR system was identified in C10. The divergent nature of these two isolates obtained from the same patient is typical of NTHi where sequential isolates are generally found to be different strains (Kaur et al., 2011; Cardines et al., 2012). Moreover, ANI and concatenated marker gene analysis of the genomes suggests that isolate C1 is not an H. influenzae strain (<95% ANI, Fig. S3B; Goris et al., 2007) having a closer ANI to H. haemolyticus (ANI 93.8 vs. 91.9%) and clustering with representatives of this species (Fig. S5).

Absence of lateral acquisition of antibiotic resistance genes and virulence factors between investigated species

The CF lung represents a diverse microbial community with sputum and BAL samples typically yielding multiple species per sample (Harris et al., 2007; Bittar et al., 2008). This cohabitation provides scope for lateral gene transfer both within and between species and the selective pressure applied by continual antibiotic therapy, coupled with the potential for antimicrobials to induce phage movement, means the CF microbial community needs to be considered as a whole (reviewed in Rolain et al., 2011). We were unable to identify any potential examples of lateral transfer occurring between isolates within the analysed dataset that could not be explained by shared evolutionary history. This may be due to the temporal isolation of most of the isolates, with few being obtained at the same time point. It is also likely a product of our relatively small sample size and the study of a single isolate at each time point. The E. cloacae example discussed above supports the possibility of identifying transfer events from outside the dataset, therefore we also examined predicted mobile elements in each of the strains for evidence of general acquisition. While we identified some elements with unique arrangements, we found no further evidence of recently acquired elements containing antibiotic resistance genes.