Loss of O-Linked Protein Glycosylation in Burkholderia cenocepacia Impairs Biofilm Formation and Siderophore Activity and Alters Transcriptional Regulators

Protein glycosylation is increasingly recognized as a common posttranslational protein modification in bacterial species. Despite this commonality, our understanding of the role of most glycosylation systems in bacterial physiology and pathogenesis is incomplete. In this work, we investigated the effect of the disruption of O-linked glycosylation in the opportunistic pathogen Burkholderia cenocepacia using a combination of proteomic, molecular, and phenotypic assays. We find that in contrast to recent findings on the N-linked glycosylation systems of Campylobacter jejuni, O-linked glycosylation does not appear to play a role in proteome stabilization of most glycoproteins. Our results reveal that loss of glycosylation in B. cenocepacia strains leads to global proteome and transcriptional changes, including the repression of the quorum-sensing regulator cepR (BCAM1868) gene. These alterations lead to dramatic phenotypic changes in glycosylation-null strains, which are paralleled by both global proteomic and transcriptional alterations, which do not appear to directly result from the loss of glycosylation per se. This research unravels the pleiotropic effects of O-linked glycosylation in B. cenocepacia, demonstrating that its loss does not simply affect the stability of the glycoproteome, but also interferes with transcription and the broader proteome.

decreased stability of the majority of known glycoproteins, which in turn affects virulence (63,64). These data support a model whereby bacterial N-linked glycosylation contributes to protein stability, but it is unclear whether other glycosylation systems, such as O-linked glycosylation, have evolved to stabilize glycosylated proteins.
Previously, we reported B. cenocepacia possesses an O-linked glycosylation system responsible for the modification of at least 23 proteins with a trisaccharide glycan using the enzyme PglL (BCAL0960) (56). Building on this work, we recently identified the biosynthetic locus, the O-glycosylation cluster (OGC [BCAL3114 to BCAL3118]), responsible for the generation of the O-linked glycan, established the O-linked glycan structure as ␤-Gal-(1,3)-␣-GalNAc-(1,3)-␤-GalNAc, and demonstrated that glycosylation was required for optimal bacterial fitness and resistance to clearance in the Galleria mellonella infection models (65). Although these studies have demonstrated a link between glycosylation and bacterial fitness, the mechanism remains unclear. Using quantitative proteomic approaches, we sought to understand the proteome changes resulting from the loss of O-linked glycosylation in B. cenocepacia. We demonstrated that loss of glycosylation in B. cenocepacia resulted in global proteome alterations beyond the known glycoproteome, which are associated with widespread alterations in transcriptional regulation. We discovered that the HSL QS system CepR/I is repressed in glycosylation-defective mutants, and this coincides with defective biofilm formation and reduced siderophore activity. In contrast to the loss of glycosylation in C. jejuni, we also demonstrate that only a few glycoproteins are reduced in abundance in the absence of glycosylation, but they are not responsible for the glycosylation-null phenotypes. Together, our data indicate that the roles of glycosylation in B. cenocepacia extend beyond protein stabilization, and loss of O-linked glycosylation in B. cenocepacia causes dramatic physiological changes due to alterations in transcriptional regulatory systems and the proteome at large.

RESULTS
Loss of glycosylation in B. cenocepacia leads to global proteome alterations. We previously demonstrated that loss of glycosylation causes defects in motility (56), reduction of virulence in plant and insect infection models (56,65), and defects in carbon utilization (65). To better understand the role of glycosylation in B. cenocepacia, we assessed the effect of loss of glycosylation on the proteome. To achieve this, we generated markerless deletion mutations in the O-oligosaccharyltransferase pglL gene (⌬pglL [BCAL0960]) (56), the recently identified O-linked glycan cluster (⌬OGC [BCAL3114 to BCAL3118]) responsible for the generation of the glycan used for O-linked glycosylation (65), and a double-glycosylation-null strain (⌬pglL ⌬OGC). We also constructed a chromosomal pglL complemented strain (⌬pglL amrAB::S7-pglL-His 10 ) (see Fig. S1A in the supplemental material). The rationale for creating multiple glycosylationdefective strains was to eliminate potential confounding effects arising from blocking glycosylation at a specific step and the corresponding accumulation of unprocessed lipid-linked glycans. Western blot analysis using the glycoprotein acceptor protein DsbA Nm -His 6 (56,66) supported the loss of glycosylation in the ⌬pglL, ⌬OGC, and ⌬pglL ⌬OGC strains, as well as restoration of glycosylation in the ⌬pglL amrAB::S7-pglL-His 10 strain (Fig. 1A). In contrast to our previously reported plasmid-based PglL complementation approaches (56) chromosomal complementation lead to the restoration of glycosylation to near wild-type (WT) levels ( Fig. S1B) as well as restoration of motility ( Fig. S1C) compared to only partial restoration previously reported (56).
Using label-free quantification (LFQ)-based quantitative proteomics, 5 biological replicates of each strain were investigated, leading to the identification of 3,399 proteins with 2,759 proteins quantified in at least 3 biological replicates in a single biological group (see Fig. S2A and B and Data Set S1, tab 1, in the supplemental material). As expected, no glycopeptides were observed in the ⌬pglL, ⌬pglL ⌬OGC, and ⌬OGC strains, while multiple glycopeptides were observed in the wild-type and ⌬pglL amrAB::S7-pglL-His 10 strains (Fig. S1D). Hierarchical clustering of Pearson correlations of proteome samples demonstrated robust correlation between all samples (average  Pearson correlation of 0.98 [Data Set S1, tab 2]); yet three discrete proteome clusters were readily identified separating the wild-type K56-2 and ⌬pglL amrAB::S7-pglL-His 10 strains and the glycosylation-null strains (Fig. 1B). Examination of the most profound alterations, proteins with a -log 10 P value of Ͼ3 and a fold change greater than Ϯ2 log 2 units, revealed alterations in protein levels observed in the ⌬pglL mutant that were mirrored in the ⌬OGC and ⌬pglL ⌬OGC strains, which were restored by complementation (Fig. 1C). Consistent with the observed motility defects (Fig. S1C), the levels of proteins associated with flagellum-mediated motility and chemotaxis, including BCAL0114 (FliC), BCAL0129 (CheA), BCAL0524 (FliG), and BCAL0525 (FliF), were significantly reduced in glycosylation-null strains. Importantly, multiple known virulenceassociated proteins were also decreased in the glycosylation-null strains, including the heme receptor protein HuvA (BCAM2626 [67]) and nematocidal protein AidA (BCAS0293 [68]). Numeration of the overlap of all altered protein between glycosylation-null strains by Fisher exact enrichment analysis demonstrated a substantial enrichment between these three groups (Fisher's exact test, 6.7502 ϫ 10 177 and 4.3784 ϫ 10 245 for the ⌬pglL compared with ⌬OGC strain, and for the ⌬pglL compared with ⌬pglL ⌬OGC strain, respectively) (Data Set S1, tab 3, and Fig. S2C). These results revealed that the loss of glycosylation due to disruption of pglL or OGC leads to similar changes, which are largely complemented to parental levels by reintroduction of pglL in the chromosome. Loss of glycosylation results in reduction in CepR/I transcription and the levels of DNA-associated CepR. Enrichment analysis of the altered proteins in glycosylationnull strains demonstrate the over representation of a range of categorical groups based on GO (Gene Ontology) terms, protein localization, and virulence-associated factor assignments. These groups highlight that protein localization assignments and virulence-associated factors were similarly affected in ⌬pglL and ⌬OGC strains, recapitulating observations made at the individual protein level ( Fig. 2; Data Set S1, tab 3). Interestingly, enrichment analysis highlighted the link between the loss of O-linked glycosylation and changes that were broader than only motility and virulence. For example, differences also observed in proteins associated with DNA-sequence specific   Impact of Glycosylation on B. cenocepacia binding and transcriptional regulation ( Fig. 2; Data Set S1, tab 3). This observation suggested that loss of glycosylation results in alterations in the transcriptional landscape of B. cenocepacia. As virulence is coordinated by global regulators such as CciR, CepR, ShvR, and AtsR in B. cenocepacia (35,43,69,70), we assessed if known regulators could account for the observed proteome changes in glycosylation-null strains. As our data demonstrated minimal alteration of the regulator ShvR (BCAS0225; Data Set S1, tab 1) across the analyzed strains, and disruption of both atsR (BCAM0379) and cciR (BCAM0240) has previously been associated with increased motility (43,69), we reasoned that the regulator CepR (BCAM1868) may be responsible for the glycosylationdependent differences in our mutant strains. Although CepR is observed within our proteomic analysis, its low intensity prevented accurate quantitation across all strains (Data Set S1, tab 1). However, the stringently CepR-regulated AidA protein (BCAS0293 [45,71]) exhibited decreases of Ϫ2.9 and Ϫ3.1 log 2 within ⌬pglL and ⌬OGC strains compared to the WT (Fig. 1C), indicating reduced CepR levels. This observation prompted us to investigate regulation of other known CepR-regulated genes and proteins. Using available microarray data of CepR-regulated genes (43), we investigated the correlation of the proteome changes observed in the absence of glycosylation, with alterations observed in response to the disruption of CepR. We observed a statistically significant enrichment of CepR-regulated proteins altered in the absence of glycosylation (multiple hypothesis corrected P values of 1.79 ϫ 10 6 and 6.69 ϫ 10 6 for the ⌬pglL and ⌬OGC strains, respectively [Data Set S1, tab 3]), supporting a link between CepR and the alteration observed in glycosylation-null strains and suggesting that the loss of glycosylation may influence the B. cenocepacia CepR regulon.
To determine transcriptional changes in cepR/I genes, we introduced the cepR and cepI luciferase promoter reporter (pPromcepR [69] and pCP300 [72]) into the wild-type K56-2, mutant ⌬pglL, and complemented ⌬pglL amrAB::S7-pglL-His 10 strains. As expected from the proteomic results, the ⌬pglL strain showed decreased induction of both cepI and cepR over a 24-h period ( Fig. 3A; see Fig. S3 in the supplemental material) compared with the wild-type and ⌬pglL amrAB::S7-pglL-His 10 strains. Detailed examination at 12 h (log phase), 16 h (the transition from log to stationary phase), and 20 h (stationary phase) revealed higher levels of transcription in the wild type of both cepI and cepR at 16 and 20 h compared with transcription levels in the ⌬pglL mutant, despite comparable growth kinetics (see Fig. S4A and B in the supplemental material). As the C 8 -HSL levels affect the response of CepI and CepR in B. cenocepacia (39,44,73), we assayed cepR/I transcription in the absence and presence of additional C 8 -HSL (10 M [Fig. 3B]). In response to exogenous C 8 -HSL, cepI transcription increased in all strains ( Fig. 3B), consistent with the positive-feedback response expected to heighten C 8 -HSL levels (39,44). In contrast, while the addition of C 8 -HSL led to no change in cepR transcription in the ⌬pglL mutant, it resulted in reduced transcription of cepR to the level observed in the wild-type K56-2 strain. Complementation of pglL, using amrAB:: S7-pglL-His 10 , restored CepI transcription to wild-type levels but only partially restored CepR transcription (Fig. 3B). As expected from the reduction in cepR/I transcription resulting from the loss of glycosylation, cepR and cepI transcription was also compromised in ⌬OGC strains ( Fig. S4C to F). Together, these results indicate that both cepR and cepI transcription are altered in the loss of glycosylation, with the resulting cepR levels resembling the levels observed during C 8 -HSL-induced repression in wild type.
As the CepR protein autoregulates cepR's own transcription (48), we reasoned that the decreased transcription in the ΔpglL mutant would correspond to decreased levels of DNA-bound CepR. To directly assay DNA binding by CepR, we monitored the DNA-bound proteome using formaldehyde-based cross-linking coupled to DNA enrichment (74). Initial analysis of the DNA-bound proteome found glycosylation-null strains (⌬pglL and ⌬OGC) and glycosylation-proficient strains (wild type and ⌬pglL armAB::S7-pglL-His 10 ) possessed distinct proteome profiles with multiple uncharacterized transcriptional regulators (e.g., BCAL0946, BCAL1916, BCAS0168, BCAL2309, and BCAL0472) which were altered by the loss of glycosylation ( Fig. 3C; Data Set S1, tab 4). Although this analysis enabled the identification of CepR, its low abundance prevented its p-value: *** = <0.001  quantitation across biological replicates. To improve the monitoring of CepR, targeted proteomic analysis was undertaken using PRM assays, which confirmed the reduction in DNA-associated CepR in the ⌬pglL mutant compared with the wild-type and ⌬pglL armAB::S7-pglL-His 10 strains ( Fig. 3D [P ϭ 0.017 for wild type versus ΔpglL strain]; Data Set S1, tab 5). In agreement with the total proteome and lux reporter measurements, the DNA-bound proteome supports multiple transcription-associated proteins, including the global regulator CepR, that are altered in the absence of glycosylation.

WT pms402
The ⌬pglL mutant demonstrates a reduced ability to form biofilms and produce siderophores. The observed reductions in CepR/I transcription suggested that CepR/I-linked phenotypes may also be altered in glycosylation-null strains. To test this hypothesis, we assessed two phenotypes associated with CepR/I regulation: (i) the production of biofilm under static 24-h growth and (ii) siderophore activity (39,(43)(44)(45)48). Consistent with an impact of glycosylation on known CepR/I-regulated phenotypes, we observed a marked reduction in biofilm formation in the ΔpglL mutant, which was partially restored by complementation (Fig. 4A). Interestingly, we also observed that the method of complementation-i.e., expression of PglL-His 10 driven from the native pglL promoter (ΔpglL amrAB::native-pglL-His 10 ) or from the constitutive S7 promoter (ΔpglL amrAB::S7-pglL-His 10 )-affected the restoration of biofilm formation  ( Fig. 4A). Examination of independently created ΔpglL and ΔpglL amrAB::native-pglL-His 10 strains confirmed a link between biofilm formation through phenotype restoration by complementation (see Fig. S5A in the supplemental material). Chrome azurol S (CAS) assays, used to assess the global levels of siderophore activity, demonstrated a reproducible effect in the ΔpglL mutant, which was completely restored by complementation when PglL was expressed from either its native or the S7 promoter ( Fig. 4B and C). The ΔOGC and ΔOGC ΔpglL strains also demonstrate biofilm and siderophore alterations compared to the wild type, although these alterations were not completely identical to those observed in the ΔpglL mutant ( Fig. S5B to D). Together, we conclude that phenotypes associated with CepR/I regulation, including biofilm and siderophore activity, are affected by the loss of glycosylation. Except for BCAL1086 and BCAL2974, proteins that are normally glycosylated remain stable in the absence of glycosylation. As the loss of glycosylation in other bacterial glycosylation systems leads to protein instability (63,64,75), we examined whether protein instability in B. cenocepacia may be responsible for the phenotypic changes in glycosylation-null strains. Our proteomic analysis identified 21 out of 23 known glycoproteins (56), yet only 2 were altered in abundance in glycosylationnegative strains: BCAL1086 (Ϫ5.7 log 2 ) and BCAL2974 (Ϫ2.5 log 2 ) ( Fig. 5A; Data Set S1, tab 1). To confirm the observed decreases in abundance, endogenous BCAL1086 and BCAL2974 were His 10 tagged at the C terminus. While His tagging did not allow the detection of BCAL2974 by Western analysis (data not shown), the introduction of the His 10 epitope into BCAL1086 allowed quantification of endogenous BCAL1086 in the K56-2 wild-type, ΔpglL mutant, and ΔpglL amrAB::S7-pglL-His 10 complemented strain backgrounds and confirmed the loss of BCAL1086 in the ΔpglL mutant (Fig. 5B). We sought to directly assess whether BCAL1086 was subjected to increased degradation in the ΔpglL mutant, as a measure of instability. For this, we monitored the endogenous peptide pool (76), quantifying peptides derived from 783 proteins (Data Set S1, tabs 6 and 7) in the B. cenocepacia K56-2 wild-type, ΔpglL mutant, and ΔpglL amrAB::S7-pglL-His 10 complemented strain. Consistent with the degradation of BCAL1086, we observed an increase in the abundance of BCAL1086-derived peptides in the ⌬pglL mutant, while peptides from other known glycoproteins showed only modest changes ( Fig. 5C; Data Set S1, tab 7). Within this peptidomic analysis, we observed that multiple unique BCAL1086 peptides were present in the ⌬pglL mutant clustered around the central region of BCAL1086 (Fig. 5D), confirming that BCAL1086 was expressed in the ⌬pglL mutant, but subjected to proteolysis. Together, our data support that BCAL1086 becomes degraded in the absence of glycosylation, but the majority of known B. cenocepacia glycoproteins are unaffected.
Role of BCAL1086 and BCAL2974 in ⌬pglL phenotypes. As changes in the glycoproteins BCAL2974 and BCAL1086 coincided with an alteration in biofilm and siderophore activity, we investigated if the loss of BCAL2974 and BCAL1086 could be responsible for defects observed in the ΔpglL mutant. To answer this question, ⌬BCAL2974 and ⌬BCAL1086 strains were created and assessed for their effect on biofilm production and siderophore activity, as well as virulence in G. mellonella, a phenotype previously associated with ⌬pglL mutation (56). Both BCAL1086 and BCAL2974 have no known functions and lack homology to known domains but are present in multiple Burkholderia species. Assessment of 24-h static biofilm growth showed the ⌬BCAL1086 mutation had no effect on biofilm formation, while ⌬BCAL2974 resulted in a small but reproducible decrease in biofilm development. However, this effect is minimal compared to the defect observed in ⌬pglL and ⌬cepI mutants (Fig. 6A). The ability of ⌬BCAL1086 and ⌬BCAL2974 mutants to produce siderophores was unaffected ( Fig. 6B  and C). Similarly, while G. mellonella infections showed that ⌬pglL causes reduced mortality at 48 h postinfection compared to in the K56-2 WT (P ϭ 0.0015), ⌬BCAL2974, ⌬BCAL1086, and ⌬pglL amrAB::native-pglL-His 10 strains demonstrated wild-type levels of lethality in G. mellonella at 48 h (Fig. 6D). These results suggest that even though BCAL2974 and BCAL1086 are influenced by the loss of glycosylation, neither protein is solely responsible for the known defect observed in the ΔpglL mutant.
We also investigated whether the loss of either BCAL2974 or BCAL1086 drives proteome changes. Using label-free-based quantitative proteomics, we compared the proteomes of the K56-2 WT, ⌬BCAL2974, ⌬BCAL1086, ⌬pglL, ⌬cepR, ⌬cepI, and ⌬pglL amrAB::S7-pglL-His 10 to assess the similarity between the proteomes as well as the specific proteins affected by the loss of these proteins. Proteomic analysis led to the identification of 3,730 proteins, with 2,752 proteins quantified in at least 3 biological replicates in a single biological group (Data Set S1, tab 8). Clustering of the proteomic analysis revealed that ⌬BCAL2974 and ⌬BCAL1086 strains closely grouped with the WT strains, while the ⌬pglL, ⌬cepR, ⌬cepI, and ⌬pglL amrAB::S7-pglL-His 10 strains formed discrete clusters. This macroanalysis indicated that mutations in BCAL2974 or BCAL1086 had a minimal effect on the proteome (Fig. 7A; Data Set S1, tabs 9 and 10). Supporting this conclusion, analysis of the specific proteins that varied between the different strains demonstrated few proteome alterations in the ⌬BCAL2974 and ⌬BCAL1086 mutants compared with the ⌬pglL, ⌬cepR, and ⌬cepI mutants (Fig. 7B), with the ΔcepR, ΔcepI, and ΔpglL strains also demonstrating the expected similarity in their proteome changes (Fisher exact test, ⌬cepR versus ⌬pglL strain, P ϭ 3.25 ϫ 10 5 , and ⌬cepI versus ⌬pglL strain, P ϭ 6.95 ϫ 10 4 [Data Set S1, tab 11]). Taken together, the proteome analysis results support the contention that BCAL2974 and BCAL1086 have minimal effects on the proteome and are not responsible for the broad proteomic alterations observed in the ΔpglL mutant.

DISCUSSION
Although glycosylation is a common protein modification in bacterial species (49)(50)(51)77) our understanding of how this modification influences bacterial physiology and pathogenesis is unclear. Recent insights into how glycosylation impacts bacterial proteomes have been obtained through study of the archetypical N-linked glycosylation system of C. jejuni (78,79), yet it is unclear whether these observations are generalizable to other glycosylation systems such as O-linked glycosylation systems. Studies on the role of N-linked glycosylation within C. jejuni have revealed that defects associated with the loss of glycosylation stem from the loss of glycoproteins (78,79), suggesting that N-linked glycosylation extends protein longevity in C. jejuni. In contrast, we find here that loss of O-linked glycosylation in B. cenocepacia has a more limited effect on the proteins targeted for glycosylation with only a subset of the known glycoproteins being affected by the disruption of glycosylation (Fig. 5). Therefore, the defect associated with loss of O-linked glycosylation in B. cenocepacia cannot be merely explained by protein instability. Indeed, we demonstrate that loss of glycosylation leads to changes in the expression of nonglycosylated proteins whose expression is regulated by the CepR/I regulon (Fig. 3) (39,42,48). Therefore, our findings uncover a previously unknown link between loss of glycosylation and alterations in pathways controlled by global transcriptional regulators. The observation that biofilm formation is reduced in the ΔpglL mutant mirrors previous reports in Acinetobacter baumannii (55) and C. jejuni (63), but the link of this phenotype to alterations in regulations has not previously documented. Previous studies in B. cenocepacia have identified that not all CepR/I-regulated proteins are required for biofilm formation. However, BapA (BCAM2143) plays a major role in the formation of biofilms on abiotic surfaces, whereas the lectin complex BclACB (BCAM0184 to BCAM0186) contributes to biofilm structural development (45). Although BapA (BCAM2143) was not detected in any of our proteomic analyses, BclA and BclB (BCAM0186 and BCAM0184, respectively) were decreased in the ΔpglL mutant (both with a Ϫ1.0-log 2 decrease compared with the WT; Ϫlog 10 P Ͼ 3.05 [Data Set S1, tab 1]). Surprisingly, BclA and BclB increased in abundance in ΔpglL ΔOGC and ΔOGC strains (both 1.0 log 2 increases compared with WT; Ϫlog 10 P Ͼ 1.4 [Data Set S1, tab 1]), and these mutants formed extensive biofilms (Fig. S5B). This result agrees with recent work showing that with disruption of BCAL3116, the third gene in the OGC, resulted in enhanced biofilm formation (80). It also should be noted that within this study, we observed that the method of complementation of pglL also influenced the restoration of biofilm formation (Fig. 4A). As differences between the promoter used to drive pglL expression can influence some glycosylation-null phenotypes, this supports the hypothesis that pglL itself may be regulated under specific conditions. Concerning siderophore activity, our proteomic data reveal that siderophore-associated proteins were reduced in both ΔpglL and ΔOGC strains (Fig. 2), with glycosylation-null strains producing reduced zones of clearing in the CAS assays ( Fig. 4B and C; Fig. S5C and D). However, the magnitude of the reduction in the CAS assays differed in the mutant, since ΔOGC and ΔpglL ΔOGC strains presented significantly smaller zones of clearing than the ΔpglL strain ( Fig. S5C and D). These results highlight that although the proteome changes observed in the ΔpglL and ΔOGC glycosylation mutants are highly similar, they are not identical and show phenotypic differences. Therefore, a key question arising from our findings is how the loss of glycosylation alters gene regulation and whether the observed defects are simply the result of altered transcriptional control. The lack of any glycosylated signaling/receptor-associated proteins in B. cenocepacia (56) makes the identification of the link between a specific glycoprotein and transcriptional control unclear. It is possible the observed alterations in biofilm formation and siderophore activity are not solely driven by altered CepR regulation, but also reflect additional transcriptional alterations in the glycosylation-null strains. This conclusion agrees with our observations of many differences in the abundance of transcriptional regulators in the DNA-associated proteome of glycosylation-null strains ( Fig. 3C; Data Set S1, tab 4). Further, biofilm formation within B. cenocepacia is modulated by multiple transcriptional regulators (33), making CepR just one of a range of regulators that could be driving this phenotype. An additional driver of these pleiotropic effects may also be deleterious outcomes resulting from the manipulation of the O-linked glycosylation system. It has been suggested in C. jejuni that the disruption of glycosylation leads to undecaprenyl diphosphate decorated with N-linked glycan being sequestered from the general undecaprenyl diphosphate pool and that this depot effect may be a general phenomenon observed in all glycosylation mutants (64). Sequestration of undecaprenyl diphosphate was thought to drive an increase in the abundance of proteins in the nonmevalonate and undecaprenyl diphosphate biosynthesis pathways observed in glycosylation-null C. jejuni (64). However, in B. cenocepacia glycosylation mutants, we observe only minor alterations in the nonmevalonate (BCAL0802, BCAL1884, BCAL2015, BCAL2016, BCAL2085, BCAL2710, BCAM0911 and BCAM2738 [see Fig. S6A in the supplemental material]) and undecaprenyl diphosphate biosynthesis (BCAL2087 and BCAM2067 [ Fig. S6B]) pathways, which argues against this phenomenon being common to all glycosylation mutants. Furthermore, the similarity of the proteome changes in the ΔpglL, ΔOGC, and ΔpglL ΔOGC strains (Fig. S2C) supports the conclusion that proteome changes are independent of the sequestration of the undecaprenyl diphosphate pool as ΔOGC and ΔpglL ΔOGC strains are unable build the O-linked glycan on undecaprenyl diphosphate. Although our proteomic analysis shows similar protein levels across glycosylation-null and -competent strains, it is important to note that we have previously shown the loss of glycosylation reduces tolerance to oxidative and osmotic stresses (65). This suggests that additional off-target effects relating to lipidlinked glycan or membrane stress may occur that are driven by changes independent of protein abundance, such as changes in protein-protein interactions, protein localization, or protein folding.
Another explanation for the pleiotropic effects associated with loss of Oglycosylation could be the instability of the glycoproteins in the absence of the glycan. We identified two glycoproteins BCAL2974 and BCAL1086, both of unknown functions, which are reduced in abundance due to the loss of glycosylation. However, genetic experiments demonstrate that neither protein is responsible for the phenotypic and proteomic changes associated with loss of glycosylation ( Fig. 6 and 7). Furthermore, in the case of BCAL1086, endogenous tagging and degradomic analysis confirm the loss of this protein in the ⌬pglL background. Although these results support the breakdown of BCAL1086 as a consequence of the loss of glycosylation, an alternative explanation is that the changes in degradation arise from alterations in protease levels or activities in the ⌬pglL mutant. Previously, we reported that ⌬pglL results in enhanced casein proteolytic activity (65). However, our global proteome analysis shows only modest changes in protease levels. We also observed identical protease profiles from activity probe against multiple classes of protease in the wild-type, ⌬pglL, and ⌬pglL amrAB::S7-pglL-His 10 (Fig. S6C), suggesting all of these strains have similar protease activities. More importantly, aside from glycoproteins BCAL2974 and BCAL1086, the other proteins targeted for glycosylation remain consistently stable in the glycosylation-defective mutants. Although 23 glycoproteins are known in B. cenocepacia, additional glycoproteins may also exist that were missed in the initial characterization of B. cenocepacia glycoproteome. Regardless, although loss of glycosylation may affect the stability of some glycoproteins, the pleiotropic effect found in the glycosylation mutants cannot be explained by alterations in protein degradation.
In summary, this work provides a global analysis of the effect of O-linked glycosylation on B. cenocepacia traits. The application of quantitative proteomics enabled the assessment of nearly half the predicted proteome of B. cenocepacia K56-2 and revealed a previously unknown link between O-linked glycosylation and transcriptional alterations. The alteration in known transcriptional regulators, such as CepR, as well as its associated phenotypes, supports a model in which the defects observed for glycosylation-null strains arise from transcriptional changes and not from the direct result of glycosylation loss per se. This work challenges the idea that loss of glycosylation solely affects the stability and activity of the glycoproteome and instead shows that glycosylation can influence the bacterial transcriptional profile and broader proteome.

MATERIALS AND METHODS
Bacterial strains and growth conditions. The strains and plasmids used in this study are listed in Tables 1 and 2, respectively. Strains of Escherichia coli and B. cenocepacia were grown at 37°C in Luria-Bertani (LB) medium. When required, antibiotics were added to the following final concentrations: 50 g/ml trimethoprim for E. coli and 100 g/ml for B. cenocepacia, 20 g/ml tetracycline for E. coli and 150 g/ml for B. cenocepacia, and 40 g/ml kanamycin for E. coli. Ampicillin was used at 100 g/ml and polymyxin B at 25 g/ml for triparental mating to select against donor and helper E. coli strains. Antibiotics were purchased from Thermo Fisher Scientific, while all other chemicals unless otherwise stated were provided by Sigma-Aldrich.
Recombinant DNA methods. The oligonucleotides used in this study are listed in Table 3. DNA ligations, restriction endonuclease digestions, and agarose gel electrophoresis were performed using standard molecular biology techniques (81), with Gibson assembly undertaken according to published protocols (82). All restriction enzymes, T4 DNA ligase, and Gibson master mix were used as recommended by the manufacturer (New England Biolabs). E. coli PIR2 and DH5␣ cells were transformed using heat shock-based transformation. PCR amplifications were carried out using either Phusion DNA (Thermo Fisher Scientific) or Pfu Ultra II (Agilent) polymerases were used according to the manufacturer's recommendations with the addition of 2.5% dimethyl sulfoxide (DMSO) for the amplification of B. cenocepacia DNA due to its high GC content. DNA isolation, PCR recoveries, and restriction digest purifications were performed using the genomic DNA cleanup kit (Zmyo Research, CA) or Wizard SV gel and PCR cleanup system (Promega). Colony and screening PCRs were performed using GoTaq Taq polymerase (Qiagen) supplemented with 10% DMSO when screening B. cenocepacia. All constructs in Table 2 were confirmed by Sanger sequencing undertaken at the Australian Genome Research Facility (Melbourne, Australia).
Construction of unmarked deletion mutants, endogenous tagged BCAL1086, and complementation with pglL-His 10 . Deletions and endogenous tagging of BCAL1086 were undertaken using the approach of Flannagan et al. for the construction of unmarked, nonpolar deletions in B. cenocepacia K56-2 (83). Chromosomal complements of pglL were generated by introducing pglL-His 10 under the control of the B. cenocepacia S7 promoter (P S7 ) or the native pglL promoter (Ppgl; 660 bp upstream of PglL) inserted into amrAB using the pMH447 (23) derivative plasmids ( Table 2) according to the protocol of Aubert et al. (84).
Protein manipulation and immunoblotting. Bacterial whole-cell lysates were prepared from overnight LB cultures of B. cenocepacia strains. One milliliter of bacteria at an optical density at 600 nm (OD 600 ) of 1.0 were pelleted, then resuspended in a mixture of 4% sodium dodecyl sulfate (SDS), 100 mM Tris (pH 8.0), and 20 mM dithiothreitol (DTT) and boiled at 95°C with shaking at 2,000 rpm for 10 min.
Proteomic analysis. Whole-proteome sample preparation was undertaken as previously described (65), while peptidomic and DNA binding proteome analysis were undertaken according to the approaches of Parker et al. (76) and Qin et al. (85), respectively. For nonpeptidomic samples, isolated protein preparations were digested as previously described (86) and cleaned up using homemade stage tips according to the protocol of Ishihama and Rappsilber (87,88). Peptidomic samples were cleaned up using commercial tC 18 columns (Waters). Purified peptides were resuspended in buffer A* (2% acetonitrile [ACN], 0.1% trifluoroacetic acid) and separated using a two-column chromatography setup comprising a PepMap100 C 18 20-mm by 75-m trap and a PepMap C 18 500-mm by 75-m analytical column (Thermo Scientific). Data were acquired on either an Orbitrap Elite mass spectrometer (Thermo Scientific), an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Scientific), or a Q-exactive plus mass spectrometer (Thermo Scientific) and processed using MaxQuant (v1.5.5.1 or 1.5.3.30 [89]). Database searching was carried out against the reference B. cenocepacia strain J2315 (https://www.uniprot.org/ proteomes/UP000001035) and the K56-2 Valvano (90) (http://www.uniprot.org/taxonomy/985076) proteomes. Proteomic data sets have been deposited into the ProteomeXchange Consortium via the PRIDE (91) partner repository. A complete description of each PRIDE data set is provided in Table 4. A complete description of all proteomic-associated methods is provided in Text S1 in the supplemental material.
Motility assays. Motility assays were conducted using semisolid motility agar consisting of LB infusion medium supplemented with 0.3% agar as previously described (56). Plates were inoculated using 2 l of standardized (OD 600 of 0.5) overnight cultures of each strain. Motility zones were measured after 48 h of incubation at 37°C. Experiments were carried out in triplicate with 3 biological replicates of each strain.
Transcriptional analysis by luminescence assays. To assess transcriptional changes in CepR and CepI, luxCDABE reporter assays were performed using the B. cenocepacia K56᎑2 wild᎑type (WT), ⌬pglL, ΔcepI derivative of K56-2 created using pGPi-SceI-cepI This study ⌬OGC, and ⌬pglL amrAB::S7-pglL-His 10 strains containing pCP300 (CepI promoter luxCDABE reporter [72]), pPromcepR (CepR promoter luxCDABE reporter [69]) or pMS402 (promoterless luxCDABE reporter [92]) as a negative control. Overnight cultures were diluted to an OD 600 of 1.0, and 2 l was inoculated into 200 l LB supplemented with 100 g/ml trimethoprim in black, clear᎑bottom 96᎑well microplates (minimum of eight technical replicates per independent biological replicate). The OD 600 and relative luminescence were measured using a CLARIOstar plate reader at 10-min intervals for 24 h. Experiments assessing the effect of C 8 -HSL additions on CepR and CepI transcription were performed according to Le Guillouzer et al. (93). Briefly, cultures were supplemented with C 8 -HSL (Sigma-Aldrich) resuspended in acetonitrile (10 M final concentration) and added to cultures with acetonitrile added alone used as a negative control. Plates were incubated at 37°C with shaking at 200 rpm between measurements, with each assay undertaken 3 independent times on separate days. The resulting outputs were visualized using R (https://www.r-project.org/). Biofilm assay. Biofilm assays were performed according to previous reports (26,94,95) using protocols based on the approach of O'Toole (96). B. cenocepacia strains were grown overnight at 37°C and adjusted to an OD 600 of 1.0. Ten microliters of these suspensions was inoculated into 990 l of LB supplemented with 0.5% (wt/vol) Casamino Acids, and 100 l was added into 96-well microtiter plates (Corning Life Sciences [a minimum of eight technical replicates per independent biological replicate]). Microtiter plates were incubated at 37°C for 24 h in a closed humidified plastic container. The plates were then washed with phosphate-buffered saline (PBS) to remove planktonic cells then stained for 15 min with 125 l of 1% (wt/vol) crystal violet. Excess crystal violet was removed with two washes of PBS and 200 l of 33% (vol/vol) acetic acid was added for 15 min to release the stain. The resuspended stain was transferred to a new plate and measured on a CLARIOstar plate reader measuring the absorbance of the resulting solution at 595 nm. Three independent assays were undertaken on separate days.
Galleria mellonella infection assays. Infection of G. mellonella larvae was undertaken using the approach of Seed and Dennis (97) with minor modifications. B. cenocepacia strains were grown overnight at 37°C and adjusted to an OD 600 of 1.0, equivalent to 2 ϫ 10 9 CFU/ml. Strains were diluted with PBS to 4 ϫ 10 5 CFU/ml, with serial dilution plates undertaken to confirm inoculum levels. For each strain, 2,000 CFU in 5 l was injected in the right proleg of the G. mellonella larvae. Three independent challenges were performed with each strain injected into 8 to 10 G. mellonella larvae. For each independent challenge, 8 control larvae were injected with 5 l PBS. Postinfection, G. mellonella larvae were placed in  12-well tissue culture plates and incubated in the dark at 30°C. The number of dead larvae was scored at 24, 48, and 72 h after infection, with death of the larvae determined by loss of responsiveness to touch. The results visualized using R (https://www.r-project.org/), and statistical analysis of survival curves was undertaken with the survminer package (version 0.4.5). CAS siderophore assays. Alterations in activities of siderophores were assessed using the chrome azurol S (CAS) assay as previously described (98,99). Ten microliters of adjusted bacterial culture at an OD 600 of 1.0 was spotted on CAS agar plates and incubated at 37°C for 24 h. The diameter of the zone of discoloration from the removal of iron from the CAS dye complex was measured. Experiments were carried out with at least 3 biological replicates in technical triplicate.
Protease activity-based probes. K56-2 WT, ⌬pglL, and ⌬pglL amrAB::S7-pglL-His 10 strains were grown overnight on confluent LB plates. Plates were flooded with 5 ml of prechilled sterile PBS, and colonies were removed with a cell scraper. Cells were washed 3 times in chilled PBS and resuspend in 40 mM Tris-150 mM NaCl (pH 7.8) and then lysed by sonication. Samples were clarified by centrifugation at 10,000 ϫ g for 10 min at 4°C, and samples were diluted to a total concentration of 4 mg/ml. Reactivity to three classes of activity-based probes was assessed using PK-DPP (Cy5-tagged probe for trypsin-like proteases [100]), PK105b (Cy5-tagged probe for elastase-like proteases [101]), PK101 (biotin-tagged probe for elastase-like proteases [102]), and FP-biotin (biotin-tagged probe for serine hydrolases [103]), which were added at 1.3 M from a 100ϫ DMSO stock. Untreated control samples were prepared in parallel and left untreated to allow the assessment of autofluorescence and endogenous biotinylation in lysates. Samples were incubated at 37˚C for 15 min to allow labeling and then quenched by the addition of Laemmli sample buffer. Samples were then boiled, and proteins were resolved on a 15% SDS-PAGE gel. For Cy5-tagged probes, labeling was detected by directly scanning the gel for Cy5 fluorescence using a Typhoon 5 flatbed laser scanner (GE Healthcare). For FP-biotin, proteins were transferred to nitrocellulose and the membrane was incubated with streptavidin-Alexa Fluor 647 at 4°C overnight. Following three washes with PBS containing 0.05% Tween 20, the membrane was scanned on the Typhoon 5 in the Cy5 channel. Experiments were carried out in biological triplicate. All probes were synthesized in-house by the Edgington-Mitchell Laboratory according to published methods, with the exception of FP-biotin, which was purchased from Santa Cruz Biotechnology. Data availability. Proteomic data sets have been deposited into the ProteomeXchange Consortium via the PRIDE (91) partner repository with the data set identifiers PXD014429, PXD014516, PXD014581, PXD014614, and PXD014700.
TEXT S1, DOCX file, 0.1 MB.   We thank the Melbourne Mass Spectrometry and Proteomics Facility of The Bio21 Molecular Science and Biotechnology Institute at The University of Melbourne for the support of mass spectrometry analysis. We also thank Silvia Cardona for kindly providing the plasmids pMS402, pCP300, and pPromCepR, Mario Feldman for pKM4, and the Canadian Burkholderia cepacia research and referral repository for providing K56-2. We also thank David Thomas for critical evaluation of the manuscript.