Streptomyces coelicolor strains lacking polyprenol phosphate mannose synthase and protein O-mannosyl transferase are hyper-susceptible to multiple antibiotics

Polyprenol phosphate mannose (PPM) is a lipid-linked sugar donor used by extra-cytoplasmic glycosyl tranferases in bacteria. PPM is synthesiszed by polyprenol phosphate mannose synthase, Ppm1, and in most Actinobacteria is used as the sugar donor for protein O-mannosyl transferase, Pmt, in protein glycosylation. Ppm1 and Pmt have homologues in yeasts and humans, where they are required for protein O-mannosylation. Actinobacteria also use PPM for lipoglycan biosynthesis. Here we show that ppm1 mutants of Streptomyces coelicolor have increased susceptibility to a number of antibiotics that target cell wall biosynthesis. The pmt mutants also have mildly increased antibiotic susceptibilities, in particular to β-lactams and vancomycin. Despite normal induction of the vancomycin gene cluster, vanSRJKHAX, the pmt and ppm1 mutants remained highly vancomycin sensitive indicating that the mechanism of resistance is blocked post-transcriptionally. Differential RNA expression analysis indicated that catabolic pathways were downregulated and anabolic ones upregulated in the ppm1 mutant compared to the parent or complemented strains. Of note was the increase in expression of fatty acid biosynthetic genes in the ppm1- mutant. A change in lipid composition was confirmed using Raman spectroscopy, which showed that the ppm1- mutant had a greater relative proportion of unsaturated fatty acids compared to the parent or the complemented mutant. Taken together, these data suggest that an inability to synthesize PPM (ppm1) and loss of the glycoproteome (pmt- mutant) can detrimentally affect membrane or cell envelope functions leading to loss of intrinsic and, in the case of vancomycin, acquired antibiotic resistance.


INTRODUCTION
In the Actinobacteria, mannose is an important component of extracellular glycoconjugates including lipoglycans and glycoproteins [1,2]. Lipomannan (LM) and lipoarabinomannan (LAM) are essential constituents of the cell envelope in mycobacteria [3], and mutants of corynebacteria that lack these molecules grow poorly [4]. Membrane and secreted proteins are also modified by mannose residues in the Actinobacteria and species of Mycobacterium, Corynebacterium and Streptomyces have been shown to contain glycoproteins [5][6][7][8][9]. Extra-cytoplasmic glycosyl transferases require a lipid-linked sugar donor, which for the transfer of mannose is polyprenol phosphate mannose (PPM). The enzyme polyprenol phosphate mannose synthase, Ppm1, synthesizes PPM by catalysing the transfer of mannose from GDP-mannose to polyprenol phosphate on the cytoplasmic face of the plasma membrane and PPM is then flipped in the membrane, transporting the mannose moiety to the periplasm [10] (Fig. 1). In mycobacteria, Ppm1 is an essential enzyme as it is required for lipomannan biosynthesis [11], and Corynebacterium ppm1 mutants have a reduced growth rate [12].
We are interested in the role of protein O-glycosylation in bacteria and ultimately we aim to understand how the glycans affect protein function. The protein O-glycosylation pathway in the Actinobacteria is highly reminiscent of the O-mannosylation pathway in yeasts and humans [2,9,13]. In fungi, mutations that inactivate PMTs lead to loss of fitness and/or avirulent phenotypes [14][15][16][17][18][19]. In humans, loss of PoMT leads to a developmental phenotype [20]. Overall the evidence suggests that protein O-mannosylation is a fundamentally important protein modification in prokaryotes and eukaryotes.
S. coelicolor is a convenient model to study the protein O-glycosylation pathway in bacteria, as the pmt and ppm1 genes are not essential [21,22] and Streptomyces spp. are not known to generate the heavily glycosylated lipomannans and lipoarabinomannans that require Ppm1 for synthesis. In previous work we showed that S. coelicolor Ppm1 (SCO1423) catalyses the transfer of mannose from GDP-mannose to polyprenol phosphate to generate PPM, and Pmt (SCO3154) was shown to transfer mannose from PPM to target peptides [9] (Fig. 1). Ppm1 and Pmt activities are confined to the membrane fractions. The periplasmic phosphate-binding protein, PstS, part of a high-affinity phosphate uptake system, was shown to be O-glycosylated with a trihexose [9]. The enzymes Pmt and Ppm1 were first identified as being required for phage fC31 infection, suggesting that the phage receptor is a glycoprotein (s) [21,22].
Here we report on the global phenotypes of mutations that lead to an inability to synthesize PPM and the glycoproteome in S. coelicolor. Mutations in pmt and ppm1 confer pleiotropic phenotypes including increased susceptibilities to multiple antibiotics, most of which act at different stages of cell wall biogenesis. Global transcriptional analysis in the ppm1mutant compared to the parent (ppm1 + ) strain showed that this mutation conferred a general switch from catabolism to anabolism and a major change in fatty acid metabolism. Changes in lipid composition in the ppm1mutant were confirmed using Raman spectroscopy. We propose that these global effects on the cell envelope affect periplasmic and membrane protein function, leading to antibiotic susceptibility.

Disc diffusion assays
Difco nutrient (DN) agar plates were evenly spread with approximately 1Â10 7 Streptomyces spores. Sterile filter discs (5 mm width) were placed on the surface of the inoculated agar plates and 10 µl of an antibiotic stock solution was allowed to absorb to the disc. Plates were incubated at 30 C for 2 days and the zones of inhibition were measured. At least three biological replicates and at least two technical replicates were used for each strain.
Quantitative mRNA analysis and culture conditions Samples were prepared for quantitative reverse transcriptase-polymerase chain reaction (qRT-PCR) analysis by inoculation of 500 ml DN broth in a 2 l baffled flask to a starting OD 450 of 0.05 with heat-shocked pre-germinated spores. Pre-germination was performed according to Kieser et al. [24]; approximately 10 9 c.f.u. S. coelicolor spores were germinated for 10 h in 100 ml germination medium in 1 l Two enzyme activities, Ppm1 and Pmt (encoded by sco1423 and sco3154, respectively), required for glycosylation of the periplasmic phosphate-binding protein, PstS, are membrane bound [9]. In vitro Ppm1 transfers mannose from GDP-mannose to a nonaprenol to generate polyprenol phosphate mannose, which then acts as the sugar carrier for Pmt to glycosylate peptides containing a glycosylation site [9]. In Mycobacterium tuberculosis protein, O-glycosylation was associated with the secretion apparatus [13].
baffled flasks at 30 C. Cultures were then grown at 30 C for 6 h at 250 r.p.m. before the addition of vancomycin to a final concentration of 0.1 µg ml À1 . Next, 43 ml of culture was sampled and added to 7 ml stop solution (5 % phenol in ethanol, stored at -20 C) at time-points 0, 30, 60 and 90 min. RNA was extracted from samples as described below and cDNA was created using SuperScript III reverse transcriptase (Life technologies) using the manufacturer's instructions with the following altered temperature cycle: 25 C, 10 min; 42 C, 120 min; 50 C, 30 min; 55 C, 30 min; 85 C, 5 min. Primers used for qRT-PCR were designed using Primer3 software [25]. Reactions were carried out on a Ste-pOne PCR machine (Applied Biosystems) with each 20 µl qRT-PCR reaction containing 10 µl Fast SYBR Green Mast Mix (Applied Biosystems) 25 ng cDNA and optimized primer and DMSO concentrations. Optimized conditions were as follows: primers RH111 and RH112 300 nM no DMSO; RH107 and RH108 500 nM with 4.8 % DMSO; RH109 and RH110 200 nM with 4.8 % DMSO (Table S1, available in the online version of this article). The principal sigma factor of S. coelicolor (hrdB) was used as a representative constitutively expressed, housekeeping gene.

Fermentor cultivation conditions
Cultivations for RNAseq analysis were performed in 7 l fermentors (Applikon) containing 2.5 l of DN broth and 2 ml antifoam C emulsion (Sigma-Aldrich), with a starting pH adjusted to 7 with NaOH. Temperature was maintained at 30 C and pH maintained at 7 through the automatic addition of 2 M HCl. Dissolved oxygen was maintained at a minimum of 30 % by automatic adjustment of the agitation speed, with a minimum and maximum rate of 300 and 1000 ,respectively, and a constant aeration pressure of 2.5-3.5 bar. For the inoculum approximately 10 9 c.f.u. S. coelicolor spores were germinated for 10 h in 100 ml germination medium in 1 l baffled flasks at 30 C following heat shock as described in Kieser et al. [24], before being used to inoculate equilibrated fermentor culture medium to an OD 450 of 0.05. Growth was measured using dry weight while dissolved oxygen, pH, agitation speed and temperature were logged online throughout.
Isolation of RNA and transcriptome analysis RNA for RNAseq analysis was isolated from fermentorgrown cultures as described above. The equivalent of 40 OD 600 units were sampled and immediately added to 1/8 volume stop solution (5 % phenol in ethanol, stored at -20 C). Cultures were spun down at 12 000 r.p.m., 4 C for 5 min before pellets were flash-frozen in liquid nitrogen and stored at À80 C for future processing. RNA isolation was performed using a modified version of the Kirby mix protocol as follows [24]: 14 g of glass beads were added to defrosted pellets along with 5 ml of 30 C modified Kirby mix before being vigorously vortexed for 2 min. Acid phenol:chloroform 1 : 1 was then used to isolate nucleic acids before being precipitated with 4 M acetate and 100 % isopropanol. Nucleic acids were subject to an ethanol wash before final resuspension in water. Nucleic acids were repeatedly treated with DNase I using conditions described by the vendor (Roche) until DNA could no longer be amplified using polymerase chain reaction (PCR). RNA was then enriched for mRNA using MICROBExpress Bacteria beads, as described by the manufacturer (Ambion).
The use of stop solution to prevent artefacts due to variation in harvesting of samples was validated as follows: two samples were collected from the fermentor (containing J1929, time point 24 h) and stop solution was added to both. RNA was prepared immediately for one sample whereas the other was incubated on ice for 1 h prior to preparation of RNA. The RNA sequences of both samples was analysed for significant changes that might have resulted from RNA degradation. Subsequent linear regression analysis gave an R 2 of 0.959, indicating that the use of the stop solution was a valid procedure for sample preparation (Fig. S2).
Generation of cDNA and production of Illumina HiSeq data were performed by Vertis Biotechnologie AG (Germany). In brief, mRNAs were fragmented with ultrasound before Antarctic phosphatase treatment and re-phosphorylation with polynucleotide kinase. RNA fragments were then poly(A)-tailed using poly(A) polymerase, and an RNA adapter was ligated to mRNA 5¢ phosphate. First-strand cDNA synthesis was performed using an oligo(dT)-adapter primer and M-MLV reverse transcriptase before PCR amplification. The cDNA samples were pooled and size fractionated in the range 250-500 bp and run on an Illumina HiSeq 2000 system with a 100 bp read length.
Sequences were trimmed using Sickle and aligned to the S. coelicolor genome (NCBI, accession number NC_003888) using BWA MEM (http://bio-bwa.sourceforge.net). Normalized counts and differential gene expression analysis were performed with DESeq2 in RIBioconductor [26]. Gene ontology analysis was performed using GOEAST [27], and read counts were visualized using the Integrated Genomics Viewer [28,29]. RNA sequencing data is available at the GEO database accession number GSE107982.
Raman spectroscopy sample preparation S. coelicolor strains J1929, DT3017 (ppm1mutant) and DT3017:pDT16 (complemented mutant) were prepared for Raman spectroscopy by subjecting spores (approximately 10 9 c.f.u. in 10 ml TES 0.5 M, pH8) to heat shock (50 C, 10 mins) and then adding to 10 ml double strength germination medium in a 250 ml baffled conical flask [24]. After incubation (37 C, 180 r.p.m. for 8-9 h) and centrifugation to pellet cells, the pellet was resuspended in 40 ml DN broth to an OD 450 of 0.05 and incubation continued for a further 12-13 h. Ten millilitres was removed for dry weight measurement and the remaining culture was centrifuged, the cell pellet washed once in 30 ml high-purity deionized water and then plated via pipette-deposit directly onto small CaF 2 disks (13 mm diameter and 1 mm thickness). The bacteria were then gently spread over the disk surface, with excess water removed by pipette aspiration and then air-dried in a fume cupboard in preparation for Raman spectroscopy.

Raman spectroscopy experiments and analyses
Raman spectroscopy point spectra were collected from randomly selected cell populations of S. coelicolor J1929 (parent), DT3017 (ppm1mutant) and DT3017:pDT16 (complemented mutant) using an HORIBA XploRA micro-Raman instrument with 532 nm laser, Â100 objective (0.9 NA) and 2400 lines/mm spectral grating. Measurements were made in confocal mode using a laser spot size of 1 µm, with 3.5 mW laser power and 1 cm À1 spectral resolution.
Spectra were collected using the HORIBA LabSpec 6 software in the 600-1800 cm À1 cell-fingerprint and 2750-3110 cm À1 high-wavenumber ranges. Each spectrum was obtained using 90s acquisition time averaged over two spectral repetitions, with 40-55 spectra randomly collected across each cell line over three experimental repeats. The total number of spectra was averaged to obtain a single representative spectrum for each strain.
The averaged spectra were baseline-corrected from the first to the last spectral point in the fingerprint and high-wavenumber ranges using the 'Raman tool set' package [30]. Peak regions in the acquired spectral ranges were then linear baseline-corrected and Gaussian peak-fitted using IGOR Pro 6.35 to obtain the peak properties (peak positions, peak intensities, etc.). Biomolecular peak assignments were made by correlating the fitted peak positions to literature references [31,32]. The total unsaturated fatty acid (TUFA)/total fatty acid (TFA) peak intensity ratio for each cell line was determined with uncertainties calculated from the propagated standard error of the mean in the measured TUFA and TFA peak intensity values. The percentage standard error of the mean for the TUFA/TFA ratio was converged as a function of the number of spectra in each spectral average, ensuring that sufficient spectra were collected for accurate representation of each sample population.

RESULTS
Mutations in ppm1 and pmt in S. coelicolor confer hypersensitivity to cell wall-acting antibiotics S. coelicolor A3(2) J1929 strains with mutations in ppm1 (DT3017, DT1020, DT1029) and pmt (DT2008, DT1025) were isolated previously [21,22] (see also Methods). The ppm1 and pmt mutants display a small colony phenotype when growing on Difco nutrient agar or soya mannitol agar that could be complemented by plasmids pDT16 or pDT10, encoding the wild-type ppm1 or pmt, respectively (Fig. S1). Microscopic techniques revealed no obvious morphological differences in the mutant strains so, to probe the underlying physiological changes, the ppm1 and pmt mutants were subjected to antibiotic stress. Increased sensitivities to a number of antibiotics, measured using disc diffusion assays, were detected in multiple independently isolated ppm1and pmtmutant strains ( Fig. 2; Dataset S1).
Whilst both ppm1 and pmt mutants had increased susceptibility to ampicillin, imipenem, meropenem, rifampicin and vancomycin compared to the parent strain, the ppm1 mutants also had greater sensitivity to daptomycin, tunicamycin, bacitracin, nisin and ramoplanin. Complementation of the ppm1and pmtmutations with their corresponding wild-type alleles alleviated the antibiotic-susceptible phenotypes (Fig. 2). The ppm1 and pmt mutant strains showed no change in susceptibility to D-cycloserine, polymyxin, monensin, tetracycline, kanamycin, thiostrepton, chloramphenicol, ciprofloxacin and moenomycin, arguing against a general increase in permeability underlying the antibiotic susceptible phenotype. Overall, with the exception of rifampicin, which targets RNA polymerase, the antibiotics with increased efficacy on the ppm1 -and pmtmutants act at different stages of cell wall biosynthesis (Fig. 2).
The vancomycin resistance cluster is induced by vancomycin in the ppm1 and pmt mutants S. coelicolor encodes a well-characterized vancomycin resistance mechanism conferring a MIC of approximately 80 µg ml À1 [33,34]. This mechanism of resistance is shared with the pathogenic bacterium Enterococcus faecium and involves reprogramming of cell wall biosynthesis so that the vancomycin target of the peptidoglycan pentapeptide, the terminal D-Ala-D-Ala, is replaced by D-Ala-D-Lac [35][36][37]. The S. coelicolor resistance cluster consists of seven genes split into four operons: vanRS, a two-component signal transduction system, vanHAX, encoding enzymes required for the production of D-Lac, ligation of D-Ala to D-Lac and breakdown of D-Ala-D-Ala, respectively; vanK, encoding a Fem protein, a non-ribosomal peptidyltransferase that adds the cross-bridge amino acids to the stem pentapeptide ending in D-Ala-D-Lac; and vanJ, required for resistance to the structurally related antibiotic teicoplanin [33,34,38]. The addition of vancomycin, but not teicoplanin, induces expression of the van cluster genes. To test whether the van cluster was being induced in the ppm1 and pmt mutants, we measured the level of vanH mRNA ( Fig. 3; Dataset S2). As the ppm1 and pmt mutant strains are highly sensitive to vancomycin, we used a very low concentration of the drug to induce (0.1 µg ml À1 vancomycin). Even at this low concentration the van cluster in the parent strain and in DT2008 is induced between 10-and 100 -old. The ppm1 mutant strain also showed induction of vanH, but the extent varied widely from experiment to experiment leading to a large standard error. This observation and the low levels of induction in the complemented pmtstrain are most likely due to the use of a vancomycin concentration for induction that is at the lower limit. We conclude that failure to transcribe the van cluster in the ppm1 or the pmt mutants is not the reason why these strains are highly sensitive to vancomycin, and that delivery of resistance in these strains is blocked post-transcriptionally.
Loss of Ppm1 activity leads to a global switch from catabolism to anabolism, including a change in fatty acid metabolism In order to further our understanding of the underlying processes that led to a loss of intrinsic antibiotic resistance, global transcriptional profiling using RNAseq was performed in the ppm1 mutant, DT3017, the original isogenic parent, J1929, and the complemented strain, DT3017:pDT16. A ppm1mutant was chosen for RNAseq analysis because of the more extreme phenotype of the ppm1mutants. Samples were collected during growth under controlled fermentation conditions where DT3017 showed only a modest growth defect (Fig. S2). Deseq analysis revealed that expression of 658 genes had changed with a P<0.05 level of significance between DT3017 and J1929 (461 with a >2-fold difference); 574 of these changes showed at least some restoration towards J1929 levels in DT3017:pDT16 and were used for further gene ontology analysis (Dataset S3). The majority (69 %) of the 574 genes used for gene ontology analysis were down-regulated in the ppm1 mutant compared to the parent, J1929.
Gene ontology analysis showed that membrane and periplasmic functions had been severely affected in the ppm1 mutant. The most prevalent down-regulated genes are associated with the GO terms transport and catabolic processes and in particular organic acid, carbohydrate and amino acid transport, and carboxylic acid, fatty acid and lipid catabolism are significantly down-regulated (Dataset S4; Table 1). The GO terms that describe the up-regulated genes are broadly biosynthetic processes; these notably involve carboxylic acids (31 genes), which include fatty acid (9 genes) and amino acid biosynthesis (19 genes). Sixty-six genes reached the Benjamini-Hochberg level of significance and were at least partially restored through complementation (Table S2). Almost half of the genes in the Benjamini-Hochberg set for which a function is predicted have a putative role in lipid metabolism. Streptomyces coelicolor strains defective in protein glycosylation genes are hypersensitive to multiple antibiotics. Shown are diameters of growth inhibition zones from disc diffusion assays for the ppm1mutants, DT1020, DT1029 and DT3017 against the parent strain J1929 and the complemented strains DT2010:pDT16, DT1029:pDT16 and DT3017:pDT16 (a), and the pmtmutants, DT1025 and DT2008, the parent strain J1929 and the complemented strains DT1025:pDT10 and DT2008:pDT10 (b). Shown are the averages of 4 replicates with SEM. *Indicates a P<0.05 that the observed difference between the mutant strains versus J1929 has occurred by chance. The concentration of antibiotic used is shown in parentheses; the full set of antibiotic concentrations used is in Dataset S1. (c) Diagrammatic summary of antibiotic targets in cell wall biosynthesis.
Closer inspection of fatty acid metabolism revealed an upregulation of all the components of the S. coelicolor ACCase and FAS II complex (Fig. 4) and a general down-regulation of a variety of genes predicted to perform b-oxidation and fatty acid degradation. These results suggest up-regulation of fatty acid production in DT3017, and hence possible alterations to membrane composition. Alterations to membrane fluidity, topology and composition are all known to effect stability and sensitivity to stress agents such as antibiotics [39][40][41][42].
The enriched gene set was searched for genes that might be involved in cell wall biosynthesis, and/or antibiotic targets. Genes that were up-regulated and might be directly involved in peptidoglycan biosynthesis encode an FtsI-like PBP (sco3771), a putative muramoyl-pentapeptide carboxypeptidase (sco5467) and a putative deacetylase (sco2962). Genes that were down-regulated encode a putative PG deacetylase (sco6178), a putative AmpC bÀlactamase/ D-alanyl-D-alanine carboxypeptidase (sco5660), a putative D-alanyl-D-alanine dipeptidase (sco1396) and a putative murein DD-endopeptidase (sco0543). Enriched GO terms did not include cell envelope biogenesis, and the expression of the putative major PG biogenesis operon in S. coelicolor (sco2077-sco2092) was unchanged in DT3017. Amongst the down-regulated gene set, there were no obvious candidates that might contribute to enhanced antibiotic susceptibility. On the contrary, genes that are predicted to encode efflux systems were up-regulated (sco5950, sco5516, sco3206, sco1742 and sco0756) as well as one gene, sco4049, encoding a putative penicillin amidase/acylase. Two up-regulated genes have been linked with vancomycin resistance: sco6183 (cwgE), a GT1 putative heptosyl transferase that is part of the vancomycin-induced putative cell wall glycan operon (cwg) [43], and sco2472, a homologue of an uncharacterized protein, SanA, that affects membrane permeability to vancomycin [44].
The gene sco3736, encoding an ECF sigma factor and the two upstream genes sco3737 and sco3738, encoding a putative lipoprotein/sortase (homology to SrtE family -cd05829; E=3e À51 ) and membrane protein, respectively, are the most significantly up-regulated genes in the ppm1strain, DT3017 (13-, 37-and 32-fold, respectively) (Fig. 5). The ECF sigma factor sco3450 is also 1.9 fold up-regulated and there is a strong down-regulation (15-fold) of ECF sigma factor gene, sco4938. It is worth emphasizing that the expression of sco3356 encoding the known cell wall stress ECF sigma factor, SigE, [45] did not change (Dataset S3).
The Ppm1-defective strain contains a higher relative proportion of unsaturated fatty acids Raman spectroscopy was used to investigate changes in the fatty acid profile of the ppm1 mutant by means of quantitative lipid phenotyping. The Raman spectroscopy approach involves the inelastic scattering of light to produce high-discriminatory, molecular-scale cell characterization (i.e. fingerprinting). Raman fingerprinting provides detailed information on the cellular components including proteins, nucleic acids, lipids and carbohydrates [31,46], and has been used to characterize microbial species and subspecies, and phenotypic changes [32,[47][48][49].
Raman spectroscopy was previously applied to lipid profiling in several studies. Wu et al. correlated the relative intensities of the 1650 cm À1 C=C and 1440 cm À1 CH 2 Raman bands (I 1650 /I 1440 ratio) to the degree of lipid unsaturation in extracted lipids from algae cells [50]. Potcoava et al. used the degree of unsaturation (I 1655 /I 1440 and I 1655 /I 1294 ), lipid chain length (I 2930 /I 2959 ) and total unsaturated fatty acid (TUFA) to total fatty acid (TFA) (I 3015 /I 2851 ) ratios to characterize lipid droplets in treated and untreated breast and prostate cancer cells, noting that the TUFA/TFA ratio correlates to the number of C=C to CH 2 species in pure fatty acids [51]. The TUFA/TFA ratio has also been used by Nieva et al. to stratify the degree of malignancy in breast cancer cells, and is regarded as a key Raman biomarker for lipid phenotyping [52]. Whereas peak intensity ratios for the degree of lipid unsaturation and chain length also relate to proteins, the TUFA/TFA ratio comprises lipid molecule vibrations only [31] and is therefore a reliable biomarker to interrogate the fatty acids in intact cells. In this work, the TUFA/TFA ratio was used to determine changes in the lipid content of the parent S. coelicolor strain J1929, the ppm1 mutant derivative, DT3017 and the complemented mutant, DT3017:pDT16.
Averaged Raman spectra and peak assignments obtained for J1929, DT3017 DT3017:pDT16 (Figs 6 and S3, Table S3) agree with previously published spectra for Streptomyces cells [32] and other published works on bacteria [48]. The TUFA/TFA peak intensity ratios determined from these spectra showed that the parent strain, J1929, and the complemented mutant DT3017:pDT16 have similar lipid composition (Fig. 6c). For the DT3017 ppm1 strain, the TUFA/TFA value was outside of the J1929 and DT3017: pDT16 ranges, confirming that it has a larger relative unsaturated fatty acid component and hence different lipid composition. The converged percentage standard error of the mean for the TUFA/TFA values as a function of the number of randomly selected individual spectra in the spectral averages indicated that sufficient data were collected (N=40-55 spectra/strain) to establish reliable statistics for these measurements (Fig. S3).

DISCUSSION
Previous work has shown that mutations in the genes encoding Ppm1 and Pmt required for synthesis of polyprenol phosphate mannose (PPM; ppm1) and O-glycoproteins (pmt and ppm1) in Streptomyces coelicolor lead to phage resistance [9,21,22]. Here we show that pmtand ppm1 - mutations also lead to an increase in susceptibility to antibiotics, particularly those that target cell wall biogenesis. We propose that in ppm1strains, the extremely antibiotic-susceptible phenotype is due to global changes in lipid, membrane and periplasmic protein functions. The pmt mutants have a similar but less extreme phenotype to the ppm1 mutants, suggesting that loss of membrane and periplasmic glycoproteins contributes to antibiotic hyper-susceptibility.
The ppm1 mutants show greater sensitivity to vancomycin and b-lactams, and higher susceptibility to a greater range of antibiotics that target membrane or periplasmic enzymes than pmtmutant strains (Fig. 2). ppm1 mutants fail to generate the lipid-linked mannose donor, polyprenol phosphate mannose, PPM, whereas pmtstrains are able to generate PPM but fail to use PPM in protein O-glycosylation [9]. Streptomyces spp. have been shown to produce phosphotidylinositolmannosides (PIMs) whose synthesis could also require PPM, depending on the degree of mannosylation [53,54]. In mycobacteria myo-inositol is sequentially mannosylated generating PIM 1 to PIM 6 , with PPM acting as the sugar donor for the later steps in the pathway [55]. S. coelicolor contains close homologues of the M. tuberculosis genes required for PIM biosynthesis:PimA (Rv2610c, SCO1525; E=3e À122 ) and PimB (Rv2188c, SCO2132; E=9e À127 ) that synthesize myo-inositol with one (PIM 1 ) or two (PIM 2 ) mannose residues attached, respectively. Both PimA and PimB act on the inner face of the cytoplasmic membrane and use GDP-mannose as the sugar donor. PimE in M. tuberculosis uses PPM to transfer mannose to the growing mannose chain, i.e. PIM 4 to PIM 5 . SCO2335 is a distant (E=4e À15 ) homologue of M. tuberculosis PimE (Rv1159) and could be involved in the polymerization of the mannoside chain using PPM as the sugar donor. Another candidate protein from S. coelicolor that could utilize PPM is SCO4023, which is annotated as a dolichylphosphate-mannose protein mannosyltransferase (PMT_2, Pfam13231). The gene sco4023 is located adjacent to sco4022, which is a homologue of ppm1. In preliminary work we have knocked out both of these genes, and mutation in neither gene confers resistance to fC31 (Dr Anpu Varghese, unpublished work). According to the Carbohydrate Active Enzymes (CAZY) database (www.cazy.org), the S. coelicolor genome encodes 29 and 17 GT2 and GT4 family glycosyl transferases, respectively; members of both families can use polyprenol-containing glycoconjugates as substrates. Loss of PPM from the S. coelicolor membrane is therefore expected to affect synthesis of periplasmic O-glycoproteins, periplasmic mannosylation of PIMs and, possibly, a variety of as yet uncharacterized cell envelope macromolecules. While disruption of the O-glycosylation pathway by mutation of pmt leads to mild antibiotic hypersusceptibility, the extreme antibiotic hyper-susceptibility when PPM is depleted indicates that multiple pathways depend on PPM as a sugar donor.
RNAseq was used to provide insight into the global changes in the ppm1mutant strain compared to its parent. The ppm1strain has undergone a major shift from a largely catabolic metabolism to an anabolic one. Genes coding for membrane and periplasmic proteins, particularly transport proteins, were down-regulated in the ppm1 mutant, DT3017, whereas genes for biosynthetic cytoplasmic functions were up-regulated. This response might be expected after transfer of a bacterial culture to a minimal medium where biosynthetic functions would be up-regulated. However, the growth of all the strains for RNAseq was performed in rich medium (Difco nutrient broth). The logical interpretation of the switch from the parental catabolic metabolism to the ppm1 mutant anabolic metabolism is that some systems for sensing nutrients and their uptake are defective. At this stage we cannot, however, propose a direct mechanism through which loss of PPM might lead to the observed metabolic switch, other than multiple effects caused by a lack of mannose-containing glycoconjugates in the cell envelope and/or an accumulation of polyprenol phosphate.
Differential gene expression showed a switch in fatty acid metabolism in the ppm1strain, with genes coding for b-oxidation enzymes down-regulated and those for synthesis up-regulated (Table 1; Fig. 4) [25,56]. The up-regulation of the ACC complex might have increased the proportion of straight-chain fatty acids as ACC, but not the PCC complex, can use acetyl-CoA as a primer for FA biosynthesis [56]. To study lipids in the Streptomyces strains, we used Raman spectroscopy of whole cells. The ppm1mutant was shown to have a small but significant increase in the ratio of TUFA acids to total fatty acids (TFA) compared to the TUFA/TFA ratios in the parent strain or the complemented mutant. The only enzyme in Streptomyces known to introduce the C=C double bond in FA biosynthesis is FabA (SCO4636 and SCO4637), a hydroxyacyl-ACP dehydratase [57], and the corresponding genes are up-regulated in the ppm1 mutant. Up-regulation of fabA in E. coli is known to lead to a higher proportion of unsaturated FAs [25,58,59]. However, it is not clear whether up-regulation of the fabA genes in the S. coelicolor ppm1mutant led directly to the observed higher TUFA/TFA ratio or whether the up-regulation is simply in line with the overall up-regulation of the FA biosynthesis pathway. The net result, however, is a change in lipid composition in the ppm1strain DT3017 compared to its parent.
Changes in lipid composition can have profound effects on the activities of membrane proteins, perhaps providing a mechanism for the proposed defects in periplasmic nutrient sensing and transport protein function. Membrane composition is important in maintaining the topology of membrane proteins. Although the 'positive inside rule' is believed to be the major determinant of membrane protein topology, head group charge, fatty acid chain length and fatty acid unsaturation have also been shown to be important [60]. Pertinent to our work is that changes in membrane composition are also correlated with changes in sensitivity to antibiotics and/or detergents. For example, a B. subtilis strain lacking three extracytoplasmic functions, ECF, sigma factors (sM, sW and sX) is significantly more sensitive to cell wall-acting antibiotics than either the double mutants or the parents [61]. Expression of sW was shown to lead to a decrease in membrane fluidity with concomitant increase in resistance to detergents [42]. The S. coelicolor ppm1 mutants are highly susceptible to daptomycin, an antibiotic that targets the cytoplasmic membrane, possibly causing depolarization, and in Enterococcus faecalis inhibits cell division by concentrating at the septum [62]. Changes in membrane fluidity have been associated with daptomycin resistance in S. aureus; in vitroderived daptomycin-resistant strains have increased membrane fluidity, although highly fluid membranes also display increased resistance to cationic peptides including daptomycin, supporting the view that membrane fluidity is only one component of the cell envelope that mediates susceptibility to this antibiotic [39,62,63]. A decrease in the supply of undecaprenol phosphate (uppS mutant) in B. subtilis can lead to increased vancomycin resistance, and this was shown to be due to elevated sM expression. Mutation in a teichoic acid biosynthesis gene (TagF) can also increase sensitivity to vancomycin and lysozyme in Streptomyces [64].  (Table S3) with the spectra vertically shifted to aid visualization. Panel c; TUFA/TFA (I 3008 /I 2850 ) peak intensity ratios derived from the averaged spectra in panels a and b.
Gene regulation by ECF sigma factors is important in B. subtilis in sensing and reacting to cell wall stresses [65]. The RNAseq data on the S. coelicolor ppm1mutant showed significant changes in expression of regulators, but not where expected. For example, genes for sigma factors such as SigE (sco3356, linked to cell wall stress; [45]), SCO0600 (sigB; osmotic stress; [66]), SCO2954 (linked to increased expression of extracellular proteases; [67]), SCO4005 (upregulated by ppGpp; [68]) and SCO4938 were down-regulated in the ppm1strain DT3017. Three of four of the most significantly up-regulated genes in DT3017 were the consecutive genes sco3736-sco3738, probably co-transcribed and encoding a putative ECF sigma factor, a lipoprotein with putative sortase E function, and integral membrane protein, respectively (Fig. 5). Sortases are transpeptidases that attach proteins to the cell wall [69]. The major 'housekeeping' sortase normally recognizes a conserved pentapeptide sorting sequence in a variety of substrates that in S. coelicolor is predicted to be LAXTG [70]. Sortase E homologues are thought to have a more specific range of substrates, and two of these, SCO3849 and SCO3850, have been shown to be required for normal development in S. coelicolor [71]. SCO3737 belongs to a more divergent group of putative sortases and it has been proposed that genes for this group are co-transcribed with their substrates. In the case of sco3737, this would be sco3738 encoding a putative membrane protein [69,71]. A gene encoding a second putative ECF sigma factor, sco3450 (SigR2 transcribed by SigR, which mediates the major thiol-oxidative stress response) [72,73], is also upregulated. Both up-regulated sigma factors are expressed early in the life cycle of Streptomyces; sco3736 and sco3450 were identified in a study examining regulatory networks during germination [73].
The changes in the complement of sigma factors in DT3017 might be directly responsible for the global phenotypic response to growth without the ability to synthesize PPM. Changes in sigma factors might also explain why ppm1 and pmt mutants are more sensitive to rifampicin, an antibiotic inhibiting transcription initiation by binding within the DNA/RNA channel of the b-subunit of RNA polymerase [74]. Although rifampicin is considered a general inhibitor of RNA polymerase, there are examples of altered promoter responses to rifampicin for differing holo-enzymes. These include: the tdc operon is insensitive to rifampicin [75] and there is increased sensitivity to rifampicin in vivo and in vitro for s 70 -RNApol over s 32 -RNApol in E. coli [76]. In B. subtilis SigB dependent transcription is less sensitive to rifampicin [77], and transcription of a sigR-dependent promoter is more resistant to rifampicin than transcription of a hrdB-dependent promoter (rrn) [78].
The pmt mutants are, like the ppm1 mutants, considerably more sensitive to vancomycin and b-lactams (including ampicillin and imipenem) than the parent strain (Fig. 2). Our hypothesis, which is consistent with the mechanism of protein mannosyl transferases in yeasts and in another actinobacterium, M. smegmatis, is that secreted and membrane proteins are targets for O-glycosylation [2,9,13] and that the modification affects protein function. It has been established by others that protein glycosylation can have an impact on protein function [79,80], and it has recently been shown to have a role in regulating the function of a peptidoglycan hydrolase [81]. The relative activities of transpeptidase and carboxypeptidase cell wall biosynthesis enzymes are known to play a role in providing intrinsic vancomycin resistance in S. coelicolor, suggesting that subtle changes to protein activities could have a large effect on antibiotic sensitivities [82]. Our observations are reminiscent of the phenotypes of PMT mutants of fungi, where the most common phenotype is increased susceptibility to cell wall-acting inhibitors [2]. The consistency across kingdoms in the phenotypes of pmt mutants implies that O-glycosylation has similar roles in protein function in fungi and bacteria.
Both pmt and ppm1 mutants are highly susceptible to vancomycin, despite the presence in S. coelicolor of the vanRSJHAX gene cluster conferring resistance [33,34]. This cluster is induced by vancomycin, which binds the VanS sensor protein and, through phosphorylation of VanR, switches on the van promoters. We showed that the van genes are induced normally by vancomycin in the ppm1 and pmt mutants, and yet the strains are still very sensitive to the antibiotic. There are two possible explanations for the persistent sensitivity to vancomycin: either the substrates for the resistance enzymes to generate the D-ala-D-lac pentapeptides are in limited supply or resistance could be masked epistatically by membrane-associated steps such as flipping of lipid II, or the lipid II polymerization steps transglycosylation and transpeptidation. In the ppm1 mutant, changes in membrane physiology are predicted by RNAseq analysis and observed through Raman spectroscopy, arguing that perturbation of membrane-associated steps in peptidoglycan biosynthesis might be responsible for preventing the normal ability of van gene products to deliver resistance to vancomycin.
In summary we have provided insights into the complex phenotypes of mutants in S. coelicolor that lack the ability to synthesize PPM and O-glycoproteins. We have presented evidence for a remodelling of the plasma membrane in terms of the protein and fatty acid components, possibly mediated by changes in ECF sigma factors in the ppm1mutant. The result of these adaptations is a loss of intrinsic resistance to antibiotics that target cell wall biogenesis, particularly those that act outside the plasma membrane, and to rifampicin. As the pmtmutant strains have also lost some intrinsic resistance to a similar subset of antibiotics (particularly vancomycin, imipenem and rifampicin) as the ppm1strain, we conclude that loss of protein O-glycosylation contributes in part to the phenotype of the ppm1strains.

Funding information
This work was funded by project grants BB/J016691 and BBS/B/ 05990 from the Biotechnology and Biological Sciences Research Council, UK.