The involvement of the low-oxygen-activated locus of Burkholderia cenocepacia in adaptation during cystic fibrosis infection

Chronic infection with opportunistic pathogens including Burkholderia cepacia complex (Bcc) is a hallmark of cystic fibrosis (CF). We investigated the adaptive mechanisms facilitating chronic lung infection in sequential Bcc isolates from two siblings with CF (P1 and P2), one of whom also experienced intermittent blood-stream infections (P2). We previously showed increased lung cell attachment with colonisation time in both P1 and P2. WGS analysis confirmed that the isolates are closely related. Twelve genes showed three or more mutations, suggesting these were genes under selection. Single nucleotide polymorphisms (SNVs) in 45 regulatory genes were also observed. Proteomic analysis showed that the abundance of 149 proteins increased over 61-months in sputum isolates, and both time- and source-related alterations in protein abundance between the second patient’s isolates. A consistent time-dependent increase in abundance of 19 proteins encoded by a low-oxygen-activated (lxa) locus was observed in both sets of isolates. Attachment was dramatically reduced in a B. cenocepacia K56-2Δlxa-locus deletion mutant, further indicating that it encodes protein(s) involved in host-cell attachment. Time-related changes in virulence in Galleria mellonella or motility were not observed. We conclude that the lxa-locus, associated with anoxic persistence in vitro, plays a role in host-cell attachment and adaptation to chronic colonization in the hypoxic niche of the CF lung.


Results
Overview of whole genome sequencing of sequential B. cenocepacia isolates. Initial MLST of the sequential isolates used in this study (Supplementary information, Table S1) determined that these B. cenocepacia isolates all shared the same unique sequence type (ST867) 23 . Individual isolates were assembled de novo. Average coverage ranged from 23X(P2 late blood isolate, P2B3) to 171X(P2 sputum isolate, P2S) and the N50 scores ranged from 104238 (P2B3) to 246069 (P1 middle isolate, P1M). Total assembly lengths were approximately 7.7-7.8 million base pairs for individual species (Supplemental Table S1). We also mapped raw sequence reads against the B. cenocepacia J2315 reference genome, as these sequential ST867 isolates are in the same recA group as J2315 based on multi-locus sequence typing (MLST) sequence type. In addition, the J2315 genome is complete, well annotated 26 , actively curated (www.burkholderia.com) 27 and a widely studied Bcc strain. A mean of 3.08 million reads were produced per isolate of which a mean of 88.31% (87.3 to 89.1%) were mapped to the J2315 reference with a mean coverage of 86.45% of the reference.
Comparing the concatenated MLST gene fragments from our isolates with other strains clustering under the same B. cenocepacia recA subgroup to construct a MLST based neighbour-joining phylogenetic tree, confirmed that the ST867 isolates are closely related to recA subgroup of strains (Fig. 1A). Comparative analysis of the assembled sequences using BLAST Ring Image Generator (BRIG) indicated that comparable to previously reported B. cenocepacia CF isolates, several genomic islands appeared to be absent among the ST867 isolates, namely BcenGI2, BcenGI3, BcenGI6, BcenGI9, BcenGI10, BcenGI13. Furthermore our sequences had poor coverage in other genomic islands, e.g. BcenG15, BcenGI7, BcenG18, BcenGI12 and BcenGI14 (Fig. 1B) as has been previously shown for other patient isolates 26,28 . Obvious differences, including insertions and deletions of significant portions of genomic material, were not apparent over time of chronic lung infection from patient 1 isolates or between the blood and sputum associated isolations from patient 2 based on the BRIG comparison to B. cenocepacia J2315 (Fig. 1B). With respect to the B. cenocepacia J2315 reference genome 57,050 and 62,532 single nucleotide polymorphisms (SNVs) across our six isolates were located using Genome analysis toolkit (GATK) and mpileup variant callers respectively. Of these 48,263 and 55,093 were found in all six isolates. From the remaining non-intersecting SNVs we identified 1,132 variants that were called by both GATK and mpileup. We took a conservative approach and only considered the SNVs located by both variant callers as bona fide.
Examining the 1132 variants across the six isolates, 976 (88%) were within genes. In total, of the 823 SNVs in coding regions, 528 (64%) were non-synonymous (NS) SNVs, while 36% were synonymous, the remaining variants were stop loss or stop gain mutations, indels and non-coding transcript variants. Based on the confident identification SNVs, we constructed a phylogenetic tree (Fig. 1C) to determine the relationships among the isolates. The early and middle isolates from P1 (P1E and P1M) are likely to share a common ancestor, as expected. In addition, the late isolate from P1 (P1L) and the third blood isolate from P2 (P2B3) have evolved into a separate clade, based on SNVs and putative internal node observed (Fig. 1C). From this tree it appears that P2B1 and P2S1 are more closely related to each other than each is to either P1E and P1M or P1L and P2B3. A full list of all SNVs and indels observed in these isolates relative to the B. cenocepacia J2315 is provided in the Supplementary Information Table S2. The average evolutionary rate was 3.50 × 10 −6 SNVs/bp/year in the brothers isolates, which is higher than that observed in other studies, for example 5.3 × 10 −7 SNV/bp/year in Canadian CF B. cenocepacia isolates 28 . The high evolutionary rate may be the result of a nonsynonymous SNV in the mismatch repair gene mutS shared by P1L and all P2 isolates which conferred a hypermutation phenotype. The mutation rate was 1.1 × 10 −6 SNV/bp/year in the first patient prior to acquisition of the mutS mutation. The average mutation rate following the acquisition of the mutS mutation increased dramatically by three-fold in the later isolates, including all isolates from P2 which was first identified 8 months after the P1M isolate which lacked the mutS mutation.
Genes with multiple mutations or those which are fixed in the population are considered to be under selection 29 . There were 75 genes which had multiple mutations indicating that these genes were under strong selection pressure (Supplementary Information Table S2). Among these were 50 genes in which missense or frameshift mutations would result in an alteration of the gene product. There were 12 genes with three or more mutations (Table 1) which has a zero probability of occurring by chance 29 . In particular, the flagellar hook length control protein gene fliK showed 5 independent NS mutations, three in P2B3 and two in P2B1 which were unique to the individual isolates. There were 15 NS mutations in the BCAL1165 gene (which encodes Type VI secretion system base plate protein), identified in the P2S sample. This gene seemed highly mutable with 107 NS and S mutations in total across the six isolates. A total of eight independent mutations were observed in the proline transporter protein BCAL1252, seven of which were in P1L isolate. The ornibactin synthesis genes also found to have several mutations. The orbI gene showed 18  orbJ gene showed 9 NS mutations, one of which was unique to P2S and the remainder shared by P2S and P2B3 and therefore fixed over time in this patient's isolates. There were nine independent NS SNVs in the capsular polysaccharide transport gene (BCAM0209) in both late isolates, indicating selection pressure on this gene. Five of the mutations would result in substitutions of serines at three positions (Ser 52, Ser 47 and Ser 22). SNVs were also identified in 45 genes encoding for regulatory proteins, of which eight were fixed in later isolates, including lysR regulatory family protein genes (BCAL0707; BCAM0056; BCAM0404), two-component regulatory systems genes (BCAM0714) and TetR family regulatory protein gene (BCAS0083) ( Table S2). The Fis family transcriptional regulator BCAM0871 showed two mutations, a NS SNV (Arg to Cys) and a frameshift at Gly 321, suggesting it is also a gene under selection. Interestingly, all regulatory gene SNVs were absent from P1M (using P1E B. cenocepacia J2315 is included as an outgroup. The phylogeny was inferred using the Maximum Parsimony method as described in the methods. Bootstrap scores are shown at internal nodes. Input character data corresponded to all 1,132 SNVs called by both GATK and mpileup variant callers. as the reference), consistent with the appearance of the mutS mutation (Table S2). In order to validate the WGS sequencing, two genes were selected, amplified and the SNPs confirmed by Sanger sequencing. The missense variant at position 205 (G > A) in P2B1, was confirmed in BCAM0292 ( Supplementary Fig. S1A); the synonymous variant at position 222 (G > A) was confirmed in P2B1 and P2S and the missense variant at position 789 (C > T) in P2S ( Supplementary Fig. 2B) was confirmed in BCAL1700 ( Supplementary Fig. S1B).
B. cenocepacia ST867 proteome alterations over time of colonization. We previously demonstrated an increase in host cell attachment in both series of isolates from the two patients, which was independent of the location of infection 23 . Our primary aim was to identify global changes in bacterial protein abundance over time of colonization that may contribute to this increased host cell attachment. A comprehensive proteomic study was performed on the six isolates. Two types of data were obtained: (A) differentially abundant proteins with >1.5-fold alteration between isolates and (B) uniquely detected proteins, i.e. proteins whose presence was significantly increased from, or decreased to, non-detectable levels. The total numbers of altered proteins in each comparison are listed in Supplementary Information Table S3.
Proteins that altered over time of colonization in P1. Given that proteins are the functional molecules in cells, investigating alterations in protein abundance is likely to be a more informative indicator of phenotype alterations. Despite the high level of SNVs observed, there were only 149 proteins that showed a statistically significant (P < 0.05) increased abundance by >1.5-fold over the 61-month period of chronic lung infection examined for patient 1 (Supplementary Information Table S4). Differentially abundant proteins of particular interest are highlighted in Table 2. Consistent with the previously observed increase in attachment over time of colonization, a substantial number of cell surface-associated proteins were significantly increased from P1E to P1L including: trimeric autotransporter adhesin (TAA) (BCAM0219); fimbrial usher pilus protein (BCAL1828); WbxY, involved in O-antigen biosynthesis, and BCAL3149, an outer membrane lipoprotein-sorting protein. Two alkylhydroperoxide reductases, (AhpC and AhpD), involved in responses to oxidative stress, showed increased abundance in P1L relative to P1E. We have previously identified AhpC as being involved in host cell attachment 30 .
Of particular interest were a group of 19 proteins encoded within 50-gene cluster with significantly increased abundance over time of chronic infection. This cluster is designated as a low-oxygen-activated (lxa) locus (BCAM0275-BCAM0323) which was associated with persistence in B. cenocepacia J2315 when cultured in a limited oxygen environment 24 . These 19 proteins were substantially increased (up to 40-fold) over time of chronic lung infection in all comparisons in P1 (Table 2). Six universal stress proteins (USP) are encoded on this locus and all six USPs, together with a heat shock protein, showed increased abundance in the later isolates. A BON (bacterial OsmY and nodulation)-domain phospholipid-binding protein (BCAM0280) showed a 30-fold abundance increase over the 61-month period examined. Reduced abundance of 151 proteins over time of infection from the early to late sputum isolate from P1 was observed (Supplementary Information Table 4), including type VI secretion system (T6SS) proteins, TssF and TssG (Table 2), associated with virulence. Seventy-nine proteins were detected in one P1 isolate which were absent or undetectable in the other comparators (Supplementary  Information Table 5), including penicillin binding protein, DacB and virulence factor, ZmpA 31,32 which were both absent in the first two isolates. In contrast, another T6SS protein was only detected in P1E.
Proteins that altered over time of colonization in P2. Consistent with P1 isolates, there was also substantial time-associated increased abundance of 20 lxa-encoded proteins (Table 3) in the sequential P2 isolates, including the same 19 identified in P1 isolates with an additional uncharacterised protein (BCAM0307). Significantly the BON-domain phospholipid binding protein (BCAM0280) showed dramatically increased abundance (122-fold) in P2B3 relative to P2B1. These isolates had also previously shown increased attachment to lung epithelial cells over time of colonisation, despite the later isolate being a blood isolate 23 . Consistent with this, many time-dependent alterations in cell surface-related proteins were observed, including proteins involved in LPS synthesis: L-arabinose formyltransferase, UDP-glucuronic acid decarboxylase and a glycosyltransferase which showed reduced abundance over time. In contrast, WbxY, involved in O-antigen synthesis was increased 3-fold from P2B1 to P2S. Fimbrial usher protein and BamA were also elevated in the P2B3 blood isolate relative to P2B1. The quorum sensing (QS) regulatory protein CepR, was elevated in abundance in both P2S and P2B3 relative to P2B1, while the abundance of three proteins involved in chemotaxis and motility (CheB1, CheY and GspG) were also increased over time from P2B1 to P2B3.
Although only one sputum isolate was available for comparison, when P2 blood and sputum isolates were compared for source-dependent changes, obvious differences between cell-surface associated protein abundance were also observed. Mpl, involved in peptidoglycan synthesis and degradation, was strongly increased (34-to 38-fold) in P2S relative to both blood isolates (Table 3). TAA (BCAM0219) was elevated in both blood isolates relative to P2S. Other cell surface-related alterations include two putative OmpA family proteins (BCAL2645 and BCAL2958), which were higher in P2S relative to both blood isolates. Other proteins that were differentially abundant between blood and sputum isolates include a T6SS protein, BcsL, elevated in P2S relative to both of the blood isolates. Increased abundance of iron acquisition-associated proteins was also seen in P2S relative to blood isolates, including FUR, OrbF, OrbE and OrbG. Conversely, nematocidal protein, AidA, showed increased abundance in both blood isolates relative to P2S. In total there were 125 proteins that changed from being undetectable to detectable among the three P2 isolates (Supplementary Information Table S5). Both time-dependent and site of isolation-related alterations were apparent. QS-associated protein, BCAM0188, involved in virulence factor regulation, was only detected in the P2B3 isolate while the virulence factor, ZmpB 33 , was only detected in the P2B1 and P2S isolates (Supplementary  Information Table S5). Transmembrane protein, TolA and T6SS protein (BCAM0043) were only detected in the P2S and P2B3 isolates relative to P2B1 and consequently were time-dependent alterations. In contrast, iron acquisition proteins (OrbA and OrbB), OmpW and penicillin-binding protein, DacB, were not detected in either blood isolate but were present in P2S, indicating adaptations to blood.
Confirmation of up-regulation of selected genes. In order to substantiate the alterations in protein abundance, we examined the expression of genes encoding two consistently elevated lxa-encoded proteins, BCAM0280 and BCAM0276 by qPCR. Consistent with the proteomic analysis, a time-dependent increase in gene expression was observed for BCAM0280 and BCAM0276 genes from all six P1 and P2 isolates in all comparisons (Supplementary Information Table S6). This suggests that the observed protein adaptations were the result of altered gene expression over time of colonization, rather than due to altered protein processing. In addition, late in our study we obtained an additional isolate from P1 which was isolated 90 months after P1E (and which  we designated P190) and a second sputum isolate from P2 isolated 64 months after P1E. qPCR analysis of these additional late isolate also confirmed the two genes BCAM0280 and BCAM0276 were up-regulated over time in these two late additional isolates (Supplementary Table S6).

Virulence of B. cenocepacia isolates in Galleria mellonella.
In order to examine whether the alterations in proteomic profile resulted in other consistent phenotypic changes across the two series of isolates, virulence was examined. All six ST867 isolates examined were considerably less virulent than both positive control strains in the G. mellonella infection model and LD 50 values (CFU that resulted in 50% death of the larvae) could not be determined for every isolate at 48 h or 72 h. Consequently LD 30 values (CFU that resulted in 30% death of the larvae) at 48 h were used to enable comparisons of all isolates. The LD 30 values were comparable across P1 isolates (P = 0.084). Both blood isolates (P2B1 and P2B3) were 7-to 14-fold more virulent than P2S (Fig. 2) (P < 0.05), consistent with an adaptation between blood and lung in this patient.

Cytokine responses of CF epithelial cells to sequential B. cenocepacia isolates.
We previously demonstrated that Bcc strains promote potent IL-8 and IL-6 secretion from CF epithelial cells 34 . The alterations in proteins associated with LPS and O-antigen biosynthesis prompted us to compare proinflammatory cytokine secretions of cystic fibrosis bronchial epithelial cells (CFBE41o − ) in response to these sequential isolates. Despite their low virulence in G. mellonella, the sequential sputum isolates induced comparable levels of IL-8 compared with B. cenocepacia K56-2 strain (Fig. 3A) (P = 0.111). A trend for increased IL-8 chemokine induction by P1 isolates relative to time of colonisation (P = 0.058) was observed, consistent with the increase abundance of proteins involved in LPS and O-antigen biosynthesis. The blood and sputum isolates from P2 induced two-to four-fold less IL-8 than those induced by P1 isolates (Fig. 3A) (P = 0.001). Overall secretion of a panel of nine cytokines was low relative to B. cenocepacia K56-2 positive control. Comparable IL-6 secretion was observed in response to the sequential P1 isolates; although two-to four-fold lower than that of K56-2 (Fig. 3B). Both P2 blood isolates induced 3-to 7-fold less IL-6 relative to P2S (P < 0.01). Despite this, P2B3 induced more IL-6 than P2B1 (P < 0.05) (Fig. 3B). IL-4 induction by all isolates in CF cells was low (≤2 pg/ml) but within the dynamic range (0.02 to 154 pg/ml) and both P2 blood isolates induced lower IL-4 than the P2S isolate (P < 0.05) (Fig. 3C), consistent with IL-6 induction. Low levels of the anti-inflammatory cytokine, IL-10 were detected in all P2 isolates (0.24 to 0.55 pg/ml; limit of detection 0.03 pg/ml); while IL-10 was undetectable in response to P1 isolates (Fig. 3D). Secretion of the remaining six cytokines (IFN-γ, TNF-α, IL-12p70, Il-13, IL-1b and Il-2) were not significantly elevated over media controls in response to any of the isolates.

Susceptibility of P2 isolates to the bactericidal properties of serum.
In the bloodstream bacteria come in contact with the bactericidal components of serum. Serum bactericidal assays showed that P2S was more susceptible to serum killing relative to P2B1 (6.9-fold, P < 0.01) or P2B2 (2.9-fold, P < 0.05), indicating a bacterial response between these environments to overcome serum killing (Fig. 4). Heat inactivated serum, allowed increased survival of P2S (P < 0.005) and P2B3 (P < 0.05), demonstrating that heat sensitive components of serum contribute to serum resistance.

Exopolysaccharide (EPS) production and motility in sequential isolates.
Previous studies have shown differences in motility and EPS production in sequential isolates of P. aeruginosa, B. cenocepacia and B. multivorans 22 Information Fig. S2). All six ST867 isolates were non-motile for swimming, swarming and twitching compared to P. aeruginosa (data not shown).

Δlxa-deletion mutant shows greatly impaired host cell attachment.
Overall, only two consistent phenotype changes correlated with time of infection in the isolates from both patients: increased abundance of lxa-encoded proteins and our previous data showing increased host cell attachment. This prompted us to examine whether these two distinct phenotypes might be related and consequently, we examined the potential role of the lxa-locus in host cell interactions. An K56-2Δlxa-mutant strain 24 showed a 12-fold reduction in attachment to CFBE41o − cells relative to its parent K56-2 strain (Fig. 5A), which was confirmed by confocal microscopy (Fig. 5B,C). This suggests that at least one or more of the proteins encoded within the 50-gene cluster is either directly involved, or regulates proteins involved, in host cell attachment.

Discussion
Genetic flexibility of Bcc isolates in CF infection has been previously reported 20,28 . Among the SNVs identified, 12 were also identified in a separate group of three sequential isolates 28 , indicating common mechanisms of adaptation during CF infection. These included genes encoding siderophore synthesis or receptors orbA, orbK, BCAL1345; transcriptional regulator genes: BCAM1722, BCAM2452, BCAS0007; antibiotic resistance genes, BCAM1362; BCAM216 and BCAM21685; gyrA and gspD. Although the WGS analysis did not show many SNVs in the LXA locus, the consistent increased abundance of LXA-encoded proteins across both sets of patient isolates may be associated with one or more of the many mutations in regulatory protein genes. Many of the proteins which were undetectable in one or more of the isolates correlated to frameshift mutations in the specific isolate affected Table S2), which probably resulted in disrupted protein expression, including BCAL2037, BCAL2444, BCAL2605, BCAM0871, BCAM0881, BCAM1395 and BCAS0081. Despite a mutS mutation in the last P1 isolate and all P2 isolates and the abundance of SNVs in regulatory genes in these isolates, the phenotypes investigated appeared remarkably stable in this group of sequential isolates in contrast to other studies in Bcc 20 and P. aeruginosa 37 . We have examined five phenotypes across these sequential isolates (virulence in G. mellonella, cytokine stimulation, motility, mucoidy and, previously, host cell attachment) and host cell attachment was the only phenotype to consistently alter with time in both sets of isolates. The increased epithelial attachment of B. cenocepacia ST867 in the isolates from both siblings suggests that this may be a general strategy facilitating survival in the host, avoiding clearance and contributing to the challenges of Bcc eradication once chronic infection is established. Increased host cell attachment has also been shown in chronic B. multivorans isolates 29 and may also explain the previously observed increased bacterial invasion among other sequential B. cenocepacia isolates 38 . We therefore performed a proteomic analysis on six selected sequential isolates in order to identify a common mechanism which may mediate the niche adaptation associated with increased host cell attachment.
Among the altered proteins that could potentially mediate this increased host cell interaction is the BON-protein (BCAM0280), which was substantially elevated with time of colonization in both individuals' isolates. This phospholipid-binding protein is associated with osmotic shock protection and phospholipid-or host-interactions 39 , but has not previously been linked with host cell attachment. Studies are on-going to evaluate this protein and its role in pathogenesis. The 2-fold increase in TAA BCAM0219 in the sequential P1 isolates may contribute to the increased host cell attachment of later isolates. Two TAAs in the same cluster as BCAM0219, (BCAM0223 and BCAM0224) were both previously implicated in epithelial cell adhesion 40,41 . Increased expression of this TAA (and BCAM223) was previously shown in B. cenocepacia sequential isolates 42 . Other cell-surface alterations that may contribute to the increased adherence of the sequential P1 isolates include a protein associated with outer membrane biogenesis (BCAL3149). However, neither BCAM0219 nor BCAL3149 were consistently upregulated over time of colonization in the P2 isolates, suggesting that while they may contribute to increased lung cell attachment, these are unlikely to be part of a common mechanism of enhanced host cell interaction over time of colonization. Similarly, OmpA, OmpW and putative fimbrial usher protein were increased in abundance in the later P2 isolates and could contribute to enhanced adhesion, but they are unlikely to be part of a general mechanism of increased attachment over time due to the lack of alteration in P1 isolates over time. Ideally inclusion of a sputum isolate prior to the first blood isolate in P2 would have been helpful, however none was available. However, the parallel observations in the three sequential P2 isolates support the time-related alterations observed in P1.
A consistent and unexpected finding was the time-related increase in both sets of isolates of 40% of the proteins encoded by the lxa-locus which is associated with B. cenocepacia survival under oxygen limitation 24 . This gene cluster encodes six USPs in addition to proteins predicted to be involved in metabolism, electron transfer and regulation and probably provides B. cenocepacia with a fitness advantage under hypoxic conditions. The blood isolates were most likely adapted to hypoxia whilst in the lungs and this adaptation was maintained in the bloodstream. The fact that significantly elevated abundance of 19 proteins encoded by this lxa-locus was evident in sequential B. cenocepacia isolates from both individuals over time of infection indicates a concerted response to the host. This is supported by the fact that all six USPs encoded within this locus were increased over time in both sets of isolates. Interestingly, USPs are also induced under hypoxic conditions in P. aeruginosa 43 under  . Serum resistance of the P2 blood and sputum isolates on three independent occasions. Bars represent % survival of the bacteria after treatment with 30% normal human serum (NHS) or heat inactivated normal human serum (hNHS). Data represents % bacterial survival (CFU) relative to initial bioburden. Error bars represent the standard deviation of the mean. P value signifies a statistically significant difference between the isolates as determined by ANOVA (*P < 0.05, **P < 0.01).
Scientific RepORTS | (2018) 8:13386 | DOI:10.1038/s41598-018-31556-6 the regulation of the oxygen-sensing regulatory protein, Anr. Sass et al. 24 identified a potential oxygen-sensing regulator in B. cenocepacia (BCAM0049) which has 43% sequence similarity to P. aeruginosa Anr 24 . BCAM0049 is a cAMP-receptor regulatory protein (CRP) which showed time-dependent increased abundance in all comparisons from both patients (Tables 2 and 3). Recently, its homologue in B. dolosa, FixK, was shown to be under the control of an oxygen-sensing two component system, FixLJ which is under positive selective pressure during CF chronic infection 44 . Although neither mutations nor alterations in abundance of FixLJ homologues (BCAL2210/ BCAL2211) have been identified in these B. cenocepacia clinical isolates, BCAM0049/fixK may play a central regulatory role to the oxygen response in both species. Furthermore, the M. tuberculosis dormancy regulon, DosR, encodes 10 USPs among its 48 genes which are also upregulated during hypoxic conditions and implicated in long-term survival in anoxia 45 . The DosR regulon comprises genes with functional or phylogenetic homology to lxa-locus genes associated with increased protein abundance in the later isolates, such as α-crystallin (BCAM0278) and phosphofructokinase (BCAM0311), both of which were upregulated over time in all comparisons. Whether lxa-locus upregulation is common across other sequential Bcc isolates remains to be explored. It is likely that the increased abundance of lxa-encoded proteins during chronic infection in both individuals may be a competitive strategy essential for B. cenocepacia persistence in the hostile host environment. The consistent elevated abundance of lxa-locus encoded proteins plus other USPs co-regulated with the lxa-locus 24 , BCAM1495 and BCAM1500, and the potential oxygen-sensor, BCAM0049 probably contributes to niche adaptation in the CF lung. Furthermore, the lxa-associated proteins are the most likely candidates contributing to the increased attachment observed in both sets of isolates over time of colonization.
The bloodstream of P2 was acutely infected with Bcc, and consequently, the alterations observed between blood and sputum isolates are likely to be acute responses to the blood environment. Several adaptations were consistent in both blood isolates relative to P2S, despite being present in the bloodstream intermittently, suggesting that these were common responses to the bloodstream. The increases in larval virulence and serum resistance of P2 blood isolates relative to the sputum isolate may relate to the enhanced abundance of TAA, BCAM0219 40,41 . Likewise, expression of a complete LPS molecule has been associated with B. cenocepacia survival in the presence of serum 46 . Reduced abundance of four proteins involved in LPS synthesis in P2 blood isolates over 18 months, may contribute to the reduced serum resistance of P2B3 relative to P2B1 isolate. In addition, the pro-inflammatory properties of lipid A would suggest that reduced abundance might benefit B. cenocepacia blood isolates in vivo 47,48 . Inflammation is a significant problem in CF, leading to tissue damage. The lower pro-inflammatory cytokine responses elicited by the blood isolates relative to the sputum isolate from P2, suggests adaptations to avoid host detection, facilitating bloodstream infection. The increase in Mpl abundance, in P2S relative to blood isolates, may cause substantial differences in cell surface between these isolates, which might partially contribute to the reduced IL-6 responses of blood isolates. Overall, there seems to be a concerted effort to avoid clearance from the blood, with several proteins involved.
The reduced abundance of iron acquisition proteins, including five involved in synthesis of the siderophore, ornibactin, in the blood isolates relative to P2S are likely due to the reduced availability of iron in the lung relative to blood. These findings are consistent with a recent report that OrbA, among other iron acquisition proteins, was upregulated when B. cenocepacia was cultured under iron-restricted conditions. The reduced abundance of siderophore-associated proteins in blood isolates may represent a switch towards utilisation of alternative iron sources in blood, including haemin and ferritin, which Bcc use very effectively 49 . Given that iron acquisition proteins have been identified as virulence factors in Bcc 49,50 lower levels of these proteins by B. cenocepacia blood isolates may also benefit the organism in evading host detection in blood. Increased iron metabolism by P. aeruginosa in the lung epithelium has been implicated in increased production of free hydroxyl-radicals, contributing to pulmonary damage and host response 51 .
There was no significant change in virulence in the G. mellonella model observed in the isolates over time of infection, in contrast with P. aeruginosa 15,21 . Interestingly, the virulence factor, ZmpA, was not detectable in P1E or P1M but was detected in P1L, indicative of increased expression over time, while T6SS proteins which are also associated with virulence showed reduced abundance in later isolates. These and many other factors can impact on virulence, hence the lack of any observed alteration in virulence in G. mellonella may be a consequence of alterations being too subtle to be identified in this model.
It is clear that B. cenocepacia ST867 adapts over time of infection by upregulating lxa-locus-encoded proteins and increasing epithelial cell attachment in vitro. These were the only consistent time-dependent alterations across both patients' isolates and may provide a mechanism by which B. cenocepacia chronically persists, contributing to the challenge of Bcc eradication from CF patients. Although K56-2 is a different strain (which is regularly examined as it is more amenable genetic manipulation) and consequently direct extrapolation is limited, the dramatic impairment in attachment of the Δlxa mutant strain strongly suggests that this gene cluster is involved (directly or indirectly) in host cell attachment. In addition, this was an in vitro study and increased attachment to host epithelial cells would need to be validated in vivo. There is limited data on Bcc attachment in vivo or ex vivo. An in-depth study of lungs of 21 patients colonised with either P. aeruginosa, Bcc or both, clearly demonstrated that while P. aeruginosa was located in intraluminal mucus and not adherent to airway epithelial surfaces, B. cenocepacia behaved differently and appeared to grow as single cells or small clusters in macrophages, within mucus, or occasionally within epithelial cells 52 .
Sass et al. 24 also predicted that the lxa locus was also involved in metabolism, transport and electron transfer. Consistent with this, over 249 individual proteins that were classified as having a metabolism function showed altered abundance in the sequential isolates, with 52% of these showing reduced abundance. Among these proteins were 72 proteins classified as being energy metabolism and/or energy conversion proteins; 51 proteins involved in amino acid metabolism; 21 proteins classified as being involved in carbohydrate metabolism; 30 proteins involved in lipid metabolism and 10 proteins involved in fatty acid metabolism (Supplementary Information  Table S4) and highlight a substantial alteration in Bcc metabolism over time of chronic colonisation, some of which may be linked to the upregulation of the lxa locus.
More studies are required to determine if activation of the lxa-locus is common among larger cohorts of Bcc patient isolates. Furthermore the sputum from P2 showed markedly reduced virulence in G. mellonella, increased inflammatory properties, and enhanced serum susceptibilities relative to the blood isolates from the same patient, highlighting that Bcc also shows adaptive responses to the blood environment, even in the short term. The mechanisms behind the phenotypic differences in bacterial isolates at a given time point can be complex, but understanding the key adaptations that differentiate Bcc blood and sputum isolate types should contribute to the prevention of potentially life threatening bloodstream infections in these patients. Overall, these adaptations are likely to be beneficial during infection, contributing to bacterial survival during chronic infection in CF.

Methods
B. cenocepacia sequential isolates. Sequential isolates from two adult siblings with CF with the same unique sequence type (ST867) 23 were investigated. In order to carry out an in-depth analysis, the earliest available isolate from P1 (P1E) was compared with two randomly selected sputum isolates from that patient (P1M (middle) and P1L (late)) with an overall time span of 61 months. The earliest available isolate from P2 was a Scientific RepORTS | (2018) 8:13386 | DOI:10.1038/s41598-018-31556-6 blood isolate (P2B1) which was compared with the last available blood peripheral isolate (P2B3, 16 months later) together with a sputum isolate (P2S) from P2 (Supplementary Information Table S1). The blood infections were intermittent, so these isolates were considered to have adapted to the CF lung prior to colonising the peripheral blood. All isolates were routinely cultured overnight in Luria Bertani (LB) broth (Sigma-Aldrich) at 37 °C.
Ethics statement. The isolates analysed in this study were collected during routine clinical practice and as such this study was exempt from ethical review.
DNA isolation and sequencing. Bacterial genomic DNA was isolated using a DNeasy blood and tissue kit (Qiagen) as per manufacturer's specifications for Gram negative bacteria. A DNA sequencing library of bacterial genomic DNA was created using an Illumina Nextera XT preparation kit adding flow cell attachment and bar code through transposon insertion. These libraries were sequenced using Illumina V2 2 × 250 bp chemistry on the Illumina MiSeq Platform.
Genome assembly and SNV analysis. Low quality sequence reads and Illumina adaptors were removed from raw fastq files using Trimmomatic 53 . De novo genome assembly of all six isolates was undertaken using SPAdes genome assembler v.3.5.0. Reads for all six isolates were individually aligned against B. cenocepacia J2315 using BWA-MEM version 0.7.10 54 . Sorting of BAM files and duplicate removal were performed with PicardTools 2.8.2. SNVs were called independently using the GATK toolkit 55 following the best practices workflow 56 and also SAMtools mpileup 57 . SNPs that were called by both SNP callers were considered bona fide. Blast ring image generator (BRIG) was used to generate visual genome comparisons in terms of similarity and coverage to the B. cenocepacia J2315 reference genome 58 .
Phylogenetic analyses. Phylogenetic analyses were conducted in MEGA7 59 . The phylogenetic relationships of fourteen B. cenocepacia strains/isolates, including our six isolates of interest, were generated using the Neighbor-Joining method 60 with 100 bootstrap replicates undertaken to display branch supports 61 . The input alignment was a concatenated alignment of seven housekeeping genes (BCAL0036, BCAL0289, BCAL0421, BCAL0953, BCAL1003, BCAL1861 & BCAM0991). Each of these genes were individually aligned and subsequently concatenated to give a final superalignment of 13,827 nucleotides. We removed phylogenetically uninformative sites and all positions containing gaps resulting in a final input alignment of 1508 nucleotides. All positions containing gaps and missing data were eliminated. There were a total of 1508 positions in the final dataset. Branches corresponding to partitions reproduced in less than 50% bootstrap replicates were collapsed. The evolutionary distances were computed using the Maximum Composite Likelihood method and are in the units of the number of base substitutions per site. Additional evolutionary analyses of the isolates of interest were inferred using the Maximum Parsimony (MP) method. The input character data corresponded to SNP data located in this analysis. The MP tree was obtained using the Subtree-Pruning-Regrafting (SPR) algorithm with search level 1 in which the initial trees were obtained by the random addition of sequences (10 replicates). Branch supports were inferred using 100 bootstrap replicates.
Proteomic analysis of sequential B. cenocepacia isolates. Bacterial proteins were isolated as previously described 62 from isolates which had been cultured overnight in LB broth at 37 °C, pelleted at 2778 × g for 15 min and pellets resuspended in 0.25% (v/v) Triton X-100, 40 mM Tris pH 7.8, containing protease inhibitor cocktail (Roche) and sonicated on ice. Lysates were centrifuged 10,000 x g for 15 min, methanol precipitated at −80 °C overnight, centrifuged at 10,000 x g at 4°C for 30 min and proteins resuspended in 2 ml 8 M urea, 50 mM Tris-HCl pH8. Reduction, alkylation, dialysis and tryptic digestion were performed according to published methods 63 . Peptides (20 µl) were dried and resuspended in 0.5% trifluoroacetic acid (20 µl), sonicated and purified on C18 Zip-tips (Millipore) prior to mass spectrometry on a Q-Exactive hybrid quadrupole orbitrap LC-MS/MS (Thermo Scientific) 64 . Three biological replicates and one technical replicate of each digested sample were analysed.
MS files were analysed against the B. cenocepacia J2315 protein database (Uniprot). Comparative proteome abundance and data analysis were performed using MaxQuant software (Version 1.3.0.5) 65 , with Andromeda for database searching and Perseus (Version 1.4.1.3) for organisation and statistical analysis 66 with the following settings: fixed modification-carbamidomethylation of cysteines; variable modifications-oxidation of methionines and acetylation of N-termini; peptide/protein false discovery rates (FDR) set at 1% based on comparison to a reverse database. The Label-Free Quantification (LFQ) algorithm was used to generate normalised spectral intensities and infer relative protein abundance. Proteins were only retained in the final analysis if detected in at least three replicates from at least one sample. Proteins with significant abundance changes (t-test, p < 0.05; fold change ≥1.5) were included in the quantitative results. Qualitative analyses were also performed to detect proteins found in at least 3 replicates of a sample, but undetectable in the comparator sample. Differentially expressed proteins were searched in the Burkholderia database (http://burkholderia.com/) to identify potential functions 27 . Expasy software was used to compute pIs (http://web.expasy.org/compute_pi/) and subcellular locations were determined with Psort v3.0.2 (http://www.psort.org/psortb/index.html).
Virulence in Galleria mellonella. The virulence of the sequential isolates in Galleria mellonella larvae were determined as described previously 67 . Bacterial cultures (OD 600nm 0.6) were serially diluted in phosphate buffered saline (PBS) from 10°-10 −7 and 20 µl aliquots injected into the haemocoel of 10 healthy larvae per group. Negative control larvae were injected with PBS (20 µl). The percentage survival was determined up to 72 h. LD 30  Cytokine secretion by CFBE41o-cells. Cytokine secretion by CFBE41o − cells was determined as previously described 68 , with minor modifications. IL-8 levels were determined using an OptEIA ™ human IL-8 ELISA kit (Becton-Dickenson). A wider range of cytokines were analysed using a chemiluminescent ELISA system, (V-PLEX TM, MesoScale Discovery) to detect IFN-γ, TNF-α, IL-10, IL-12p70, IL-13, IL-1β, IL-2, IL-4 and IL-6 according to the manufacturer's instructions. Cytokines were determined in duplicate on three independent occasions.
Assessment of sequential isolates susceptibility to the bactericidal potential of serum. Serum bactericidal assays were performed as previously outlined 41 . Blood isolates (100 µl) were added to 30% (v/v) normal human serum (NHS) (Sigma-Aldrich) or to heat inactivated NHS (hNHS). Initial bioburden was determined with 1:10 dilutions of cultures in PBS. All samples were incubated at 37 °C for 1 h (30 rpm), chilled and plated onto LB agar in duplicate. The CFU at 48 h were expressed as % survival relative to initial bioburden. Each experiment was performed three independent times.
Real time PCR. B. cenocepacia cultures (OD 600 0.4-0.8) were incubated with RNAprotect ® (Qiagen), centrifuged and lysed with lysozyme (1 mg/ml) in 10 mM Tris-HCl, 1 mM EDTA (pH7.5) for 10 min. RNA was isolated by RNeasy procedure (Qiagen) and DNase treated with turbo DNA-free kit ™ (Ambion) and converted to cDNA using Superscript ® VILO ™ cDNA synthesis kit (Invitrogen). Primers (Bio-sciences) were designed using NCBI primer design tool (Supplementary Information Table S7). QPCR was performed using Power SYBR ® green master mix in 25 µl containing 250 ng cDNA, 300 nM forward and reverse primers. No-template controls were included on each plate and amplification performed on Applied Biosystems 7300 real-time PCR system: initial step 95 °C, 10 min; 40 cycles of 95 °C, 5 sec and 60 °C for 1 min 69 . Relative quantification (RQ) expression differences were determined using the 2 −ΔΔCt method.

Bacterial attachment to human CF epithelial cells. Bacterial attachment of wild type B. cenocepacia
strain, K56-2 or Δlxa mutant to CFBE41o − cells (5 × 10 4 cells) on coated chamber slides (multiplicity of infection (MOI) 50:1) was determined after 30 min incubation by lysing the cells with 0.5%(v/v) Triton X-100 for 20 minutes, serial dilution and plating on LB agar as outlined 70 on three independent occasions. Confirmation of attachment was performed by confocal microscopy by fixing the CFBE41o − cells in 3% paraformaldehyde and detection with rabbit anti-Bcc antibody in 1% BSA overnight at 4 °C.
Statistical analysis. All statistical analyses were performed using Minitab Statistical software package (v15) unless otherwise stated. Analysis of host cell attachment, cytokine secretion and serum resistance data were determined by one-way ANOVA. Virulence data were analysed by comparing LD 30 values between the isolates at 48 h by one-way ANOVA. Proteomic data were analysed by Perseus (Version 1.4.1.3) 66 with t-tests performed for pairwise comparisons of samples (p < 0.05; fold change >1.5).