The growth rate of the bldA mutant in minimal media is as rapid as that of the wild-type (possibly it is even faster: Fig 1; D. W. Kim and K. J. Lee, personal communication). The mutant was therefore not expected to show significant changes in the abundance of mRNAs or proteins involved in primary metabolism. In general confirmation of this, the major biosynthetic pathways for amino acids, nucleotides and vitamins were apparently unaffected, with the exceptions of aromatic amino acid biosynthesis (SCO1496, chorismate synthase; and SCO2115, one of two aroH-like genes for the first step) and biotin biosynthesis (SCO1244, 1246); and among the central pathways of primary carbon metabolism (glycolysis, the pentose phosphate pathway, the citric acid cycle, the glyoxylate cycle, gluconeogenesis), the only major changes seen involved increased abundance in the bldA mutant of protein and mRNA from one of three genes annotated as glyceraldehyde-3-phosphate dehydrogenase (SCO7511, gap2) (of the two other paralogues, SCO1947 protein was detected in all samples and SCO7040 was not detected in any). SCO7511 is orthologous with a gap gene of Streptomyces aureofaciens that was reported to be induced at the time of aerial growth on glucose-free medium 17. In addition, a spot comprising a fragment of citrate synthase (SCO2784) was enhanced in the proteome of the mutant, though the major citrate synthase spot was unaffected.

However, some effects were seen on more peripheral aspects of primary and salvage metabolism, such as might be expected to be active after the main growth phase. The more basic of two SCO3073 (putative urocanate hydratase, hutU) spots was absent from the mutant; the product of a conserved hypothetical gene apparently translationally coupled to the downstream rocD (ornithine aminotransferase)-like gene was less abundant in the mutant; and a putative uracil phosphoribosyltransferase (SCO4041, nucleotide salvage) was likewise less abundant in the mutant. Changes in the abundance of both a basic and an acidic form of SCO4164, a homologue of thiosulphate sulphurtransferase (cysA, SCO3920), were seen in the mutant.

Interestingly, multiple protein forms were associated with ten of the 17 bldA-influenced protein spots potentially associated with primary/intermediary metabolic processes (central carbon metabolism, amino acid biosynthesis etc), suggesting that the majority of these effects may be indirect subtle changes in post-translational modifications (see Additional File 5).

Streptomycetes are very important producers of antibiotics and other valuable secondary metabolites. Bentley et al. 4 identified 21 genes or gene clusters in the genome sequence of S. coelicolor that are likely to determine secondary metabolite production. In our proteomic analysis, the abundance of gene products from seven of these was found to be affected by bldA, one of them, a Type III polyketide synthase (SCO7221), being more abundant in the mutant (Fig. 4 and see Additional File 10). Previously, only two of these gene sets (act and red) had been found to be influenced by bldA. In general, proteomic analysis here and in previous work 18 failed to detect some of the proteins specified by each cluster. In most cases this reflected the properties of the proteins themselves (size, charge, cellular location). For reasons discussed below, only two of the secondary metabolism gene sets (for Act and coelichelin) were among those identified as being affected by bldA in the transcriptome analysis (see Table 1).

Of the 147 genes identified as being bldA-influenced in Table 1 or Additional File 5, 87 are probably co-transcribed with other genes 31 in a total of 59 putative operons. Sixteen of these operons have more than one member listed in Table 1 or Additional File 5 (equivalent to 42 genes), reinforcing their classification as bldA-dependent. Apart from those involved in secondary metabolism (see above), the others include a cluster of function-unknown genes whose expression was particularly strongly dependent on bldA at the mRNA (SCO4246, SCO4252-3, SCO4256 and SCO4262; Fig. 5A) and protein levels (SCO4251-3; Fig.5B) 9. A search of the genome revealed a nearby TTA-containing gene, SCO4263, encoding a LuxR-family regulator. The effect of deleting SCO4263 on expression of the SCO4251-4253 operon was therefore analysed. S1 nuclease protection experiments indicated that transcription of SCO4253, the first gene in the putative SCO4251-4253 operon (Fig. 5C), starts 32 bp upstream of the translational start site, and is completely dependent on SCO4263 (Fig. 5D, E). As a control, transcription of another strongly bldA-dependent gene, SCO0762 8, was unaffected in the SCO4263 mutant. Proteomic analysis also demonstrated that the SCO4251-4253 proteins, readily detected in the parent strain, were absent from the SCO4263 mutant (data not shown), confirming the bldA-dependence of these three genes via the TTA-containing regulator. There was no obvious phenotypic change associated with the deletion of SCO4263.

The transcript of SCO6638, a TTA-containing gene of unknown function, was less abundant, and its gene product – the only one from a TTA-containing gene to be detected in the proteome analysis – completely absent, in the bldA mutant. The stop codon of SCO6638 overlaps the start site for the downstream SCO6637 gene, whose RNA and protein products were also less abundant in the bldA mutant, indicating that the two genes are cotranscribed and possibly translationally coupled. Possibly, a polarity effect on protein synthesis resulting from stalling of translation at the SCO6638 UUA codon upstream accounts for reduced SCO6637 expression, and the consequent uncoupling of transcription and translation may further result in exposure of the mRNA to RNase action, in turn reducing the half-life and overall abundance of mRNA corresponding to both genes. Such an explanation may also account for the similar situation described above for the deoxysugar cluster (SCO0381-0401).

Another group of linked genes, SCO0991-0995, most of which encode integral membrane proteins, was preferentially transcribed late in the wild-type, and transcription of SCO0991 (protein kinase), SCO0994 (function unknown) and SCO0995 (methyl transferase) was reduced in the bldA mutant. SCO0992, the TTA-containing gene encoding a putative cysteine synthase, and SCO0993 were not identified as being significantly affected by bldA mutation, presumably failing to meet the statistical criteria (though there are alternative explanations, including differences in half-lives of component transcripts or the presence of internal independent promoters, which are predicted to be common in Streptomyces: e.g. Laing et al. 32). The half-life of the UUA-containing putative polycistronic SCO0991-0995 mRNA may therefore also be affected by bldA-dependent uncoupling of transcription and translation. We note that the gene next to, and diverging from, the SCO0991-0995 cluster, SCO0990, and also encoding a membrane protein, was also bldA-dependent.

M145 genomic DNA was prepared by the 'Kirby mix' method described in Kieser et al. 35, except that sarkosyl replaced TPNS as the detergent. To retain RNA content and integrity, the mycelium from each 10 ml culture sample was treated with RNA Protect Bacteria Reagent (Qiagen), following the manufacturers' recommendations, and then stored at -20°C. The procedure used for RNA extraction was as described at 36 employing RNeasy purification columns from Qiagen. The RNA was eluted in a final volume of 250 μl, assessed for structural integrity using a Bioanalyser system (Agilent Technologies), and quantified spectrophotometrically using a NanoDrop ND-1000 instrument (Labtech). The A₂₆₀/A₂₈₀ absorption ratios of the extracted RNA were between 1.8 and 2.

The procedures for genomic DNA labelling and array hybridizations were as described at 36. Briefly, for the cDNA synthesis and labelling, 15 μg of total RNA was added to a reaction mixture containing random hexamer primers (Invitrogen), Superscript II Reverse Transcriptase (Invitrogen) and Cy3-dCTP (GE Healthcare), and the mixture was incubated at 42°C for 2–4 h. For the labelling of genomic DNA, ca. 3 μg DNA was labelled in a 50 μl reaction volume with Klenow enzyme and Cy5-dCTP. The incorporated Cy-dye-dCTP was quantified using a Nanodrop ND-1000 spectrophotometer. A volume equivalent to 45 pmol of Cy3-dCTP-labelled cDNA was pooled with 20 pmol of Cy5-dCTP-labelled genomic DNA, dried in a vacuum centrifuge and resuspended in 45 μl of Pronto! Universal Hybridisation Solution for long oligos and cDNA (Corning). For the array hybridisations of biological replicates 1 and 2, a salt-based hybridisation solution was used instead of a formamide-based solution 37.

The microarrays were coated glass slides (Corning GAPS II for replicates 1 and 2 and Corning UltraGAPS for replicate 3) spotted with PCR amplicons covering ca. 92% of S. coelicolor ORFs. The array design and layouts used are described at 36. Different array layout and print runs were used in this work: for the biological replicates 1 and 2 the arrays contained one probe spot for each gene (array batches Sc9 and Sc10) while the arrays for the third biological replicate contained duplicate probe spots (batch Scp28). Microarrays were scanned using an Affymetrix 428 dual laser scanner. The microarray spots were analysed and quantified using BlueFuse software (Version 3.1; BlueGnome Ltd, Cambridge), which uses statistical parametrical models for accurate quantification of the spot signal and background noise.

Mycelium was harvested from 25ml samples of cultures by centrifugation (30 sec, 4000 × g, room temperature) and immediately frozen in liquid nitrogen, with a transfer time from culture flask to frozen sample of 1.5 minutes. Mycelial pellets were stored at -80°C until use. Protein extracts, prepared as in Hesketh et al. 18, dissolved in denaturing isoelectric focusing buffer UTCHAPS (7 M urea, 2 M thiourea, 4% (w/v) CHAPS, 50 mM DTT, 4 mM Pefabloc SC protease inhibitor, and 40 mM tris pH9.0), were stored frozen in aliquots at -80°C.

Protein extracts were subjected to 2D gel electrophoresis over the pH4.5-5.5 or pH5.5-6.7 isoelectric point ranges as detailed in Hesketh et al. 18. For quantitative analysis of protein abundance profiles, gels were stained with Sypro Ruby (Bio Rad) according to the manufacturer's instructions, and scanned using the Perkin-Elmer ProXPRESS proteomic imaging system using excitation and emission wavelengths of 480 nm and 630 nm, respectively. To produce a quantitative analysis of protein abundance profiles, gel images were analysed using PHORETIX 2D version 5.1 (NonLinear Dynamics): spot detection was optimised automatically using the 'Spot Detection Wizard' and then manually edited; background subtraction was performed automatically using the 'Mode of Non-Spot' setting; and images were then normalised to the total spot volume for each gel for quantification. Spot filtering was not used, although all spots were manually edited. Histograms of normalised spot volumes displaying changes in spot abundance during growth and between M600 and the ΔbldA mutant were generated within the software. Protein spots of interest were excised from Sypro Ruby-stained gels using the Investigator ProPic robot from Genomic Solutions, and identified by tryptic digestion and MALDI-ToF mass spectrometry as previously described 18. Identification of proteins from peptide mass fingerprint data was performed using the MatrixScience 'Mascot' search engine 38, and was based on their 'Probability Based MOWSE Score' algorithm. A MOWSE score of 60 or higher is significant at the 5% level or better, and proteins in this work typically gave scores > 80 (frequently considerably higher). In addition, no identification was accepted unless at least 5 peptides representing at least 20% of the protein sequence were detected in the MALDI-ToF peptide mass fingerprint. Spots selected for identification included not only those showing significant differences between strains (using the non-statistical approach outlined below), but also some exhibiting only growth-phase dependent changes in abundance, and a smaller number of landmark spots that were neither growth-phase nor strain dependent.

For normalization methods the statistical computing environment R (Version 2.1.1) 39 and the package Limma 40 were used. Within-array global median normalisation of log₂ cDNA/gDNA ratios was applied to the data from each array in the analysis. Control and flagged spots were ignored. The log₂ ratios were then scaled to have the same median-absolute-deviation (MAD) across arrays 4041 of the same replicate series (I, II and III). Only un-flagged data (6,457 genes) were used in differential expression analyses and in the averaging of replicate data for the transcriptome-proteome correlation analysis.

The proteomics data were analysed using both statistical and mathematical approaches. For the mathematical methods, normalized spot volumes as generated within the Phoretix 2D software (see above) were used without further processing. For all statistical analyses, these normalized spot volume measurements were transformed to the log₂ scale, with an arbitrary value of 1 being assigned prior to transformation to protein spots too low in abundance to be detected.

The choice of analytical method when compiling a list of differentially expressed genes between conditions is dependent on numerous factors 43. For the transcriptomic data two independent methods were used: Welch t-test and Rank Product analysis 44. Whilst t-test based approaches are commonly used for analysing microarray data it is apparent that when there is large variance between biological replicates (as is the case with Streptomyces cultures) and/or crucial single time-point changes (in this analysis), the t-test has shortcomings [4344, this study]. Consequently, the more 'biology friendly' technique of Rank Products 44 was also used. Comparison of the gene lists produced when applying the respective thresholds (Welch t-test p-value < 0.05, Rank Product pfp value < 0.1) revealed little overlap between them. Hence, to avoid biasing any interpretation of the results the lists generated by the two techniques were combined.

The preprocessed data were imported into Genespring 7.2 (Agilent technologies) for further analysis and genes deemed to be 'non-changing' were filtered out from the un-flagged transcriptome data by removing those with log₂ expression values between -0.667 and 0.667 in all 16 time points (eight in M600, eight in ΔbldA). The remaining 5,983 genes were then analysed for significant (p < 0.05) differential expression between M600 and ΔbldA using the Welch t-test and Benjamini and Hochberg multiple testing correction 45.

The preprocessed data (6,457 genes) were imported into R (Version 2.1.1) 39 for analysis using the RankProd package 46 with default parameters. Though the microarray data produced in this study are two-colour, the use of a genomic reference allowed the use of the RankProd package analagously to one-colour data, the values compared being log₂ expression ratios of sample_cDNA/gDNA. Thus, to compare the bldA mutant to M600 at a particular time point six expression ratios were used, three from the mutant and three from M600. Rank products (and associated p-values, probability of false prediction (pfp) values and average fold-change) for both up- and down regulation were then calculated by using ranks for each pairwise comparison of bldA/M600 (i.e. bldA_repA/M600_repA, bldA_repA/M600_repB, bldA_repA/M600_repC etc). This was repeated for each time-point, comparing the bldA mutant replicates with their respective M600 counterparts. Differentially expressed up- and down-regulated genes were selected if their corresponding pfp value was ≤ 0.1.

To test statistically for proteins represented differentially between M600 and ΔbldA the non-parametric Mann-Whitney U-test was applied, comparing the data based on their ranks rather than absolute values. All 336 protein spots were tested for significance of differences between strains, with p-values being corrected by the Benjamini and Hochberg procedure. Differences were deemed significant if the corrected p-value was less than 0.05. It was evident that using these stringent criteria has the potential to lose interesting data, particularly spots absent in one strain and developmentally regulated in the other, since any protein spot completely absent from one would have to be present in at least three of the five time-points of the second strain to be called significantly different (based on simulations, see below). An additional complementary mathematical analysis was therefore undertaken in which protein spots were considered to be differentially represented between strains if their normalized volumes were changed two-fold or more in both biological duplicate growth curves in at least two (of the five) corresponding time points. The results from both approaches were ultimately combined to produce a master list of differentially represented proteins, with presence and absence of corresponding P-values indicating which protein spots were found in the statistical and mathematical analyses, respectively.

Normalized spot volume measurements for strain M600 samples over time for all 336 identified proteins were grouped into one of six abundance profile classes by applying the following set of simple mathematical rules. If in each replicate the largest value is less than or equal to the smallest value multiplied by 2, place into Group A (approximately equal abundance). If this is true only for one replicate, place in Group F (approximately equal abundance at each timepoint in one growth series, but not the other). The remainder exhibit two-fold or more changes in abundance in both replicates, and are placed into Group B if they reproducibly exhibit increasing abundance with time, Group C for those decreasing in abundance with time, and Group D for those with an initial increase followed by a decrease. Any remaining are placed in Group E (changing, but no reproducible trend). Ten of the 336 spots were not seen in M600 but were detected in the bldA mutant, and are given a separate class.

Spearman Rank correlation was used to compare the transcriptome and proteome mainly because it avoids scaling issues. As a non-parametric method, using the ranking of values to compute correlation Spearman rank also allows the use of arbitrary values representing very low abundance proteins. Thus, the Spearman rank correlation r between transcript and protein was calculated by

r = 1 − 6 ∑ s N d g , s 2 N ( N 2 − 1 ) MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGYbGCcqGH9aqpcqaIXaqmcqGHsislcqaI2aGndaaeWbqaamaalaaabaGaemizaq2aa0baaSqaaiabdEgaNjabcYcaSiabdohaZbqaaiabikdaYaaaaOqaaiabd6eaojabcIcaOiabd6eaonaaCaaaleqabaGaeGOmaidaaOGaeyOeI0IaeGymaeJaeiykaKcaaaWcbaGaem4CamhabaGaemOta4eaniabggHiLdaaaa@43FE@

where d is the difference between the ranks of protein and transcript data in sample s and N is the number of samples.

To test whether the observed Spearman rank correlation could be obtained by chance a t-test was calculated by

t = r 1 − r 2 N − 2 MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWG0baDcqGH9aqpdaWcaaqaaiabdkhaYbqaamaakaaabaWaaSaaaeaacqaIXaqmcqGHsislcqWGYbGCdaahaaWcbeqaaiabikdaYaaaaOqaaiabd6eaojabgkHiTiabikdaYaaaaSqabaaaaaaa@3842@

The resultant t-statistic was then converted to a p-value based on the t-distribution. All p-values were then corrected by application of the Benjamini and Hochberg multiple testing correction procedure 394547 and were classed as significant if p < 0.05. Correlation between transcriptome and proteome data was computed for each of the 5 time-points in each strain (a vector of 310 in each test) and for each gene across five time-points in M600, five time-points in ΔbldA and ten time-points across both M600 and ΔbldA. Analysis was only performed on 310 protein spots, as the other 26 spots did not have corresponding genes that passed through the stringent filtering applied to the transcriptome data. Where multiple protein products were identified from a single gene the correlation was calculated using the transcriptomic data for that gene versus proteomic data for each individual protein spot (i.e. multiple spots were not averaged or summed), because of the possibility that not all spots for any one gene product were detected.

Initial searches for conserved sequence motifs were conducted on the genes whose transcription was shown directly by Welch t-test analysis of microarray data to be influenced by bldA. A sequence set was constructed consisting of 500 nt segments (300 nt upstream and 200 nt downstream of the translational start codons) of each gene, in the correct transcriptional orientation (based on the annotated S. coelicolor genome GenBank (AL645882)). The sequence set was submitted to the MEME (Multiple Em for Motif Elicitation) server 48, applying default settings except for: minimum width of the motif to search for was set to 5 nts; maximum width to 30 nts; and maximum number of sites to find was set to 10. From the MEME output file an IUPAC consensus sequence was constructed for each motif. To estimate the distribution of each motif (i.e. how many genes in the genome have this motif in their upstream regions) the MEME-derived consensus sequence was used to search the S. coelicolor genome sequence for hits within the upstream regions of all genes (as defined above) using the genome-scale dna-pattern tool of RSAT (Regulatory Sequence Analysis Tools; 49) with default settings except that one nucleotide substitution was allowed.

A mutant SCO4263 allele, in which the entire coding region was replaced by an apramycin resistance cassette, was constructed by PCR-targeting essentially as described by Gust et al. 50. Briefly, oligonucleotide primers (5'-) with 5' ends homologous to the 5' and 3' ends adjacent to the SCO4263 coding sequence, and 3' (priming) ends designed to amplify the apramycin resistance disruption cassette, were used in a PCR with pIJ773 as template. The PCR product was introduced into E. coli BW25113/pIJ790 containing cosmid StD86A 51, preinduced for λred functions by the addition of arabinose, to obtain a SCO4263-disrupted version of the StD86A cosmid. The disrupted cosmid was introduced into E. coli ET12567/pUZ8002 by transformation, then transferred into S. coelicolor M600 by conjugation. Exconjugants were selected on MS agar 35 containing apramycin, and the products of double crossovers identified by screening for kanamycin sensitivity. Deletions were confirmed by PCR and Southern blotting.

Specific primers and probes for 14 of the 74 selected differentially expressed genes (SCO6808, SCO7657, SCO4295, SCO5013, SCO3088, SCO3285, SCO3286, SCO5166, SCO6958, SCO6638, SCO4648, SCO4702, SCO4705, SCO4717) were designed using either Primer Express v.2 software (Applied Biosystems) or Primer 3 software 52. Five μg samples of each of the 16 RNA preparations from replicate 3 were subjected to RNase-free DNase I treatment (Promega) in a 20 μl reaction volume. One μg of RNA was used as template for cDNA synthesis with 225 ng of random hexamers (Invitrogen) and 50 units of Superscript II reverse transcriptase (Invitrogen) in 20 μl reaction volumes, and incubated for 10 min at 25°C, 50 min at 42°C and then 15 min at 70°C. The samples were diluted 1:4 with distilled water and 5 μl was used in the quantitative PCR reaction with 10 pmol of forward and reverse primer and 2.5 pmol of FAM/TAMRA dual-labelled specific probe (Operon technologies) in 25 μl master mix (QPCR ROX, ABgene). Parallel reactions were performed in the same 96-well plate using different dilutions of genomic DNA in order to generate a standard curve for each selected gene. The reactions were analysed in an ABI PRISM 7000 Sequence Detection System (Applied Biosystems).

RNA was isolated from exponential and transition-phase cultures as described by Strauch et al. 12. For each S1 nuclease reaction 20 μg RNA was hybridised in NaTCA buffer 53 to about 0.2 pmol (approximately 10⁵ Cerenkov counts min^-1) of each of the following radiolabelled probes. For SCO4253, the oligonucleotide 5'-GACCGACGAAGGCCGCCACCGA-3', which anneals within the SCO4253 coding region, was uniquely end-labelled at its 5'- end with [γ-³²P]-ATP using T4 polynucleotide kinase. This was used in the PCR together with the unlabelled probe 5'-GATCTGACCGATCCTCCTGACACGCCGTCACCGT-3' (which anneals upstream of the SCO4253 gene) and cosmid StD49 as template, to generate a 432 bp probe. PCR was performed in the presence of 7% dimethylsulphoxide using the following conditions: 94°C for 4 min followed by 26 cycles of 45 s at 94°C, 45 s at 58°C and 60 s at 72°C, then held at 72°C for 5 min to finish. The probe for SCO0762 was made in a similar way, but using 5'- AGGCGATCCTAGTCGATCAAGAAACGCCCAGTTC-3' as the unlabelled primer, 5'-TCTCTCCGTGCCCCACGGTCAGCA-3' as the labelled primer, and cosmid StF81 51 as template. This produced a 365 bp uniquely end-labelled probe product. Hybridisations were carried out at 45°C for 14 h after denaturation at 65°C for 15 min. S1 nuclease digestions and analyses of RNA-protected fragments were performed as described previously 54. High-resolution mapping of the transcription start site of SCO4235 was achieved by generating a sequencing ladder with the same labelled primer as was used for the probe, using a Thermo Sequenase cycle sequencing kit (USB) according to the manufacturer's instructions.

Extraction and HPLC analysis of ppGpp, ATP and GTP from S. coelicolor cultures were carried out as described by Strauch et al. 12.