Two different S. coelicolor IJISS DNA microarrays were designed, featuring, respectively, 22,000 (Sco-Chip²-v1) and 44,000 (Sco-Chip²-v2) 60-mer oligonucleotide probes. In each case the same experimental optimization approach was used (Figure 1) where a large set (approximately 1 million) of 60-mer probes were printed in parallel with corresponding probes that had a 3-nucleotide mismatch. Cyanine-3 (Cy3) and Cyanine-5 (Cy5)-labeled S. coelicolor genomic DNA was hybridized against the test arrays and the probe performance was scored using the following equations. Firstly the Cy3 and Cy5 background-subtracted signals, designated 'g' and 'r', respectively, obtained by feature extraction of the arrays using the Agilent feature extraction software (Version 9.1.3.1) were entered into the following formula:

A = (gMM/gPM + rMM/rPM)/2

where the signal from the perfectly matched probe is designated 'PM' while that from the corresponding mismatched probe is designated 'MM'. For values of A greater than 1, A was set to 1 before entering it into the second equation:

R = [1 - arctan (A × π/2)] × [1 - exp(- (gPM + rPM)/2000)]

The resulting R-value was used to rank all tested probes. The higher the value, the better the probe performance. This method of ranking probes was developed within Oxford Gene Technology Ltd and has been applied to various prokaryotic organisms for empirical microarray probe design.

Sets of probes within a defined region, either gene or intergenic, were ranked. All probes were considered relative to each other without applying thresholds and the desired density of probe coverage was achieved by selecting top-ranked probes where possible. Performing the above experimental optimization approach is, in our opinion, a necessary step, given that approximately 40% of the in silico designed probes failed quality control. Sufficient probe coverage was obtained using fewer than 5% of probes ranked below the median value of the ranking distribution. The remaining 95% of the optimized probe set were picked from probes performing above average with a strong bias for very well performing probes. For both array formats the probes were deposited at random positions on the slide surfaces to minimize the risk of any position-specific artifacts.

All possible 60-mer probes for all targets (both coding and non-coding sequences) in the S. coelicolor genome (based on the S. coelicolor A3(2) [EMBL:AL645882.2]) were designed. For this version of the array all non-coding sequences upstream of protein-encoding genes were selected (sequences where transcription factors are most likely to bind) and multiple 60-mer probes targeting those regions were selected from the 'all possible probes' set. Following this initial selection, a total of 84,268 probes were experimentally tested and the best performing 21,064 probes that represented all upstream intergenic regions (an average of 3 approximately 110 bp spaced probes to each upstream site) in the genome were synthesized on the array. As this array design was developed specifically for ChIP-on-chip experiments, all probe sequences corresponded to one strand only (that in [EMBL:AL645882.2]) since the particular DNA strand was unimportant. (Note that intergenic regions flanked by transcription terminators for convergently transcribed genes were not selected for this array.)

From the 'all possible probes' set (see above), 964,820 60-mer probes were selected and printed to target all coding and non-coding sequences with minimal distance between the probes and maximal coverage of the genome. Following experimental validation of probe signal and specificity, 43,798 of the best performing probes were selected to give broad coverage. Probes within protein coding sequences corresponded to the mRNA strand for (cDNA-based) detection of gene expression. For intergenic regions the probe sequences corresponded to one strand only (that in [EMBL:AL645882.2]). The average spacing of probes in the genome was approximately 135 bp.

The experimentally optimized Sco-Chip²-v1 and Sco-Chip²-v2 arrays were used consecutively to identify in vivo targets of HspR. The latter array was designed to also enable quantification of gene expression. In order to validate the sensitivity and specificity of these arrays, we chose the well-studied transcriptional repressor HspR, which was previously known to bind to only three promoter regions in the genome of S. coelicolor: upstream of the dnaK operon; the protease-encoding gene lon; and the clpB gene, which is transcribed in an operon with SCO3660. These results were based on transcriptome analysis of an hspR disruption mutant and complementary in silico genome wide searches for HspR binding sites 6.

For the ChIP-on-chip experiments, samples of S. coelicolor cultures at early stationary phase were treated with formaldehyde and subjected to immunoprecipitation (IP) as described in Materials and methods. For these experiments, S. coelicolor was cultivated under non-heat-shock conditions in a rich liquid medium containing 10.3% sucrose to support dispersed growth of the mycelium and provide sufficient biomass for the ChIP protocol; this was to maximize formaldehyde penetration and to determine the genomic distribution of HspR under non-stressed conditions (the 'resting' state). Following the IP reaction, the DNA was recovered, labeled with Cy3-dCTP and then co-hybridized onto the Sco-Chip²-v1 and Sco-Chip²-v2 arrays together with the Cy5-dCTP-labeled total chromatin as reference (Sco-Chip²-v1) or with Cy5-dCTP labeled mock 'no-antibody' IP chromatin (Sco-Chip²-v2) (see Materials and methods). The results presented in Figures 2 and 3 (and Additional data file 3) represent the average of two biological replicates. They confirm that HspR does bind, in vivo, to the dnaK, clpB and lon promoter regions and, importantly, have served to identify additional putative HspR targets. The statistical approaches used to score probe enrichment ratios (gene targets) as significant differed between the two array formats because in Sco-Chip²-v1 the probes were focused only on promoter regions while in Sco-Chip²-v2 the probes were relatively evenly spaced across the genome (see Materials and methods). The targets scored as significant using Sco-Chip²-v1 were dnaK, clpB, lon and SCO5639 and those on Sco-Chip²-v2 were dnaK, clpB, lon and probe sequences between SCO3019-SCO3020 and SCO5549-SCO5550, corresponding, respectively, to the promoter region of the rrnD ribosomal RNA operon and a five-tRNA cluster encoding tRNA^Gln and tRNA^Glu species; if the cut-off threshold was slightly relaxed for the Sco-Chip²-v2 data, then SCO5639 was also identified.

The respective nucleotide sequences of the new putative stable RNA targets of HspR had been excluded from Sco-Chip²-v1 because they are non-protein-coding and are positioned between convergently transcribed genes. The discovery that HspR may regulate specific tRNA and rRNA genes is unprecedented and suggests a more global role for HspR in the stress response of Streptomyces. The HspR-specific probe enrichments in the previously known and the new putative HspR targets are shown in Figure 3 (and Additional data file 3).

The versatility of the Sco-Chip²-v2 design allowed us to detect gene expression using the same probe set. Thus, in Figure 3 the expression data from two pairs of comparisons are also superimposed on the ChIP enrichment data: the ratio of expression from hspR disruption mutants relative to the wild-type strain; and the ratio of expression from cultures heat-shocked at 42°C relative to non-heat-shocked control cultures. It is noted that the observed reduction of relative transcript levels of operonic genes more distal from the operon promoter (as in the dnaK operon; Figure 3a) is consistent with general observations of polarity of expression of Streptomyces operons 36.

The gene expression studies were conducted using RNA samples from strains cultivated on supplemented minimal medium agar plates, rather than rich liquid medium. This was for several reasons. First, the magnitude of the heat-shock response is relatively lower and less reproducible in heat-shocked mycelium cultivated in the rich YEME+10.3% sucrose liquid medium, compared with the heat-shock response of the surface-grown minimal medium cultures. Second, the hspR disruption mutants used in this study are unstable because hspR is an essential gene 627. The disruption of hspR is via a single integration event of a non-replicating plasmid and there is a strong selective pressure for its excision. Thus, in liquid culture the mycelium in which the disruption plasmid has excised outgrows the mutant mycelium, which is at a growth disadvantage, leading to a dominant wild-type revertant phenotype. In surface-grown cultures this reversion is markedly attenuated, where a low frequency of reversion maintains viability. Third, the RNA samples used here for comparison with the ChIP data have been extensively validated by other methods 627. The above experiment allowed, for the first time, a comprehensive identification of the heat-shock stimulon of S. coelicolor at the transcriptome level, where rank products analysis revealed 119 up-regulated genes (based on a probability of false prediction (pfp) threshold of <0.15 (see Materials and methods)) as a result of heat-shock (Additional data file 4). The use of such thresholds has been reported elsewhere 737.

Two genes on the heat-shock list with relatively high pfp values (SCO3202, pfp = 0.12; SCO4157, pfp = 0.13) were selected for independent validation by quantitative real time PCR (qPCR) to confirm true heat-shock induction (Additional data file 9); furthermore, SCO3660, a known member of the HspR regulon 6, had a pfp value of 0.12. This justified the use of the pfp threshold adopted here. The significantly up-regulated heat shock genes include all members of the dnaK operon, clpB, lon and the chaperonin-encoding groES-groEL1 operon and groEL2. Two more protease-encoding genes are also present in the heat-shock list: SCO4157, encoding the homologue of E. coli HtrA, a serine protease involved in degradation of periplasmic misfolded proteins, and SCO6515. Notably, eight oxidoreductase-encoding genes are present, some of them being strongly up-regulated by heat-shock and transcribed in an operon (SCO1131-SCO1134). The operon encoding different subunits of the nitrate reductase (SCO0216-SCO0219) and the nitrite/nitrate transporter-encoding gene SCO0213 are also heat-inducible together with the principal 'gas vesicle protein'-encoding operon (SCO6499-SCO6508) 38, the cytochrome oxidase-encoding genes SCO3945-SCO3946 and two genes of sigE operon (sigE (SCO3356) and the lipoprotein-encoding gene cseA (SCO3357)) 39. It is of interest that more than 10% of the heat-shock-induced genes (14) encode transcriptional regulators and included SCO0174 (the most induced), five sigma factors (HrdD (SCO3202), SigB (SCO0600), SigE (SCO3356), SigL (SCO7278) and SigM (SCO7314)) and an anti-sigma factor antagonist (SCO7325). A separate, complementary analysis of the heat-shock response in wild-type S. coelicolor cultivated under identical conditions in YEME to those used for the ChIP-on-chip studies demonstrated that most of the above 119 heat-shock induced genes (102/119) were also heat-induced in the YEME medium (Additional data file 11); however, the level of induction of the well-known molecular chaperone-encoding genes was attenuated relative to the surface grown SMMS cultures. It is interesting to note that 16 of the 17 genes not heat-induced in the YEME cultures are clustered in a discrete region at the left end of the chromosome between SCO162 and SCO219; this may reflect the differences in the widely different nutritional compositions of the two growth media and these genes may require additional transcription factors for their induction.

The list of 55 genes significantly up-regulated in an hspR disruption mutant relative to the wild-type is presented in Additional data file 5 (cut-off pfp < 0.15). It includes all previously known members of the HspR regulon and other notable genes that, on the basis of the ChIP-on-chip analysis, are not considered to be directly controlled by HspR; their induction could be a consequence of the up-regulation of molecular chaperone or protease-encoding gene expression in the HspR mutant. Genes for five putative transcriptional regulators are represented in the list.

The sensitivity of the IJISS arrays was deduced to be high since all previously known HspR targets were identified on both array designs. SCO5639 was not identified as belonging to the HspR regulon in a previous study 6. SCO5639 encodes a hypothetical protein of 176 amino acids in length and, unusually for streptomycete genes, has a low G+C content (approximately 53%) and is flanked by genes also of low G+C content: SCO5638 (55% G+C), which encodes an integral membrane protein, and SCO5640 (54% G+C), which encodes a hypothetical protein. Moreover, the adjacent gene, SCO5641, encodes a putative transposase, suggesting that SCO5639 could have been laterally acquired recently. Pfam 40 searches of the deduced amino acid sequence of SCO5639 returned the 'domain of unknown function' DUF1863, corresponding to a domain that adopts the 'flavodoxin fold' with "a probable role in signal transduction as a phosphorylation-independent conformational switch protein". Other proteins that contain this domain (37 known in total, including another actinomycete, Corynebacterium efficiens) are also uncharacterized. Similarly, blastp 4142 results identified further hypotheticals (at E-value < 1^e-10) and found a similarity, albeit low (35% identity), to the phosphorylation site of the calcineurin temperature suppressor (Cts1) of Cryptococcus neoformans (a yeast), which is responsible for restoring growth of calcineurin mutant strains at 37°C among other functions such as cell separation and hyphal elongation 43. This link with temperature would be consistent with SCO5639 being a target of HspR.

For the ChIP-on-chip studies the prototrophic S. coelicolor strain MT1110, a SCP1^- SCP2^- derivative of the wild-type strain, John Innes Stock Number 1147 57, was cultivated in YEME liquid medium plus 10% sucrose at 30°C in a rotary shaking incubator. For the gene expression studies the previously reported two independent hspR disruption mutants, MT1151 and MT1153, were used together with the two independent, otherwise isogenic, hspR⁺ integrants, MT1152 and MT1154 58. The heat-shock conditions were as reported previously 26.

In order to obtain Streptomyces chromatin of high quality, it was found that rapid, low temperature, physical disruption of the mycelium constituted a more reproducible method than the conventional lysozyme treatment methods. S. coelicolor MT1110 was cultivated at 30°C in 50 ml YEME liquid medium in 250 ml flasks with springs (supplemented with 10% sucrose, glycine and MgCl₂ as specified in 58 up to early stationary phase (OD₄₅₀ approximately 2.0). Cultures were divided into 20 ml aliquots and formaldehyde treated (final concentration, 1%) for 10 minutes at 30°C in order to in vivo crosslink proteins to DNA; glycine (final concentration of 0.5 M) was added to quench the formaldehyde and the culture was incubated for a further 5 minutes at 30°C. Mycelium was harvested by centrifugation, frozen in liquid nitrogen and then transferred to a 7 ml PTFE shaking flask with cap (which was also immersed in liquid nitrogen to cool it down). Mycelium was disrupted in a Mikrodismembrator U mechanical device (Sartorius Stedim Biotech, Epsom, Surrey, UK) for 2 × 1 minute at 2,000 rpm with one 10 mm diameter chromium steel grinding ball and contents of one tube of lysing matrix B (Q-BIOgene, Cambridge, UK). Chromatin processing and IP were based on previous methods 5960 with additional modifications. The pulverized mycelium was transferred to a tube containing 1 ml lysis buffer (10 mM Tris-HCl, pH 8, 20% sucrose, 50 mM NaCl, 10 mM EDTA, Protease Inhibitor Cocktail (Roche, Burgess Hill, West Sussex, UK); one tablet per 10 ml); 3 ml IP buffer (50 mM Tris-HCl, pH 8, 150 mM NaCl, 0.5% Triton X-100, plus Protease Inhibitor Cocktail) was added and the chromatin was sheared by sonication (Sonics VibraCell VCX130, CH-1217 Meyrin/Satigny, Switzerland) on ice. One 2 ml aliquot was sonicated 2 × 20 s power ON at 50% power, 40 s power OFF and the other 2 ml chromatin sample was sonicated 4 × 20 s power ON at 50% power, 40 s power OFF, to obtain the optimal DNA size range of 0.5-1.0 kb. Cell lysates were cleared by centrifugation at 12,000 rpm for 25 minutes. To assess chromatin quality, aliquots of the chromatin (70 μl) were treated with proteinase K (100 μg, Roche) for 2 h and the DNA-protein complexes were de-crosslinked at 65°C for 6 h; 5 μl aliquots were subjected to electrophoresis and the chromatin fraction(s) with optimal size range were subjected to IP either with specific antibody or with no antibody (mock IP) as control. The IgG fraction containing anti-HspR polyclonal antibodies 61 and the fraction from pre-immune serum from the same rabbit used for immunization were purified through Nab Protein A spin columns (Pierce, ThermoFisher Scientific, Cramlington, Northumberland, UK). Chromatin IP was carried out with 100 μl specific antibody added to 800 μl of chromatin overnight at 4°C on a rotating wheel at 12 rpm; 80 μl of either sepharose protein A (Sigma, Gillingham, Dorset, UK) or Ultralink immobilized Protein A/G beads (Pierce; previously washed twice in phosphate-buffered saline (PBS), once in PBS containing 5 μg/ml bovine serum albumin and resuspended in one half bead volume of PBS containing a protease inhibitor cocktail) were added to the immunoprecipitated chromatin and incubated for a further 3-4 h at 4°C on a rotating wheel at 12 rpm. The DNA-protein complexes bound to the beads were pelleted at 3,300 rpm for 1 minute and washed four times by resuspension in 1 ml ice cold IP buffer (wash 1), IP buffer plus salt (wash 2, as wash 1 but with 500 mM NaCl), wash buffer (wash 3, 10 mM Tris pH 8, 250 mM LiCl, 1 mM EDTA, 0.5% nonidet P-40 and 0.5% Na deoxycholate) and TE pH 7.6 (wash 4), with incubation at 4°C in a rotating wheel for 15 minutes and centrifugation at 3,300 rpm for 1 minute. After the first wash the protein A/G bound DNA protein complexes were transferred to a non-stick microfuge tube.

Immunoprecipitated complexes were eluted overnight at 55°C in Tris-EDTA, pH 7.6 (TE), 1% SDS, 100 μg Proteinase K (Roche) in 240 μl volume. A 170 μl aliquot of input chromatin (not subjected to IP) or mock-IP chromatin was incubated in parallel under the same conditions with 240 μl of elution buffer. Crosslinks were dissociated at 65°C for 30 minutes followed by centrifugation at 3,300 rpm for 1 minute. The protein A/G beads were washed in 50 μl TE and the supernatants were pooled. The immunoprecipitated and input chromatin/mock-IP samples were extracted twice with phenol/chloroform/isoamyl alcohol (25:24:1) pH 8, then once with chloroform and the DNA was ethanol precipitated in the presence of 20 μg glycogen as carrier, resuspended in 20 μl ultrapure water and quantified with a NanoDrop spectrophotometer.

Immunoprecipitated and input control DNA were labeled with Cy3-dCTP and Cy5-dCTP, respectively, using the BioPrime kit (Invitrogen, Paisley, UK). DNA (0.1-1 μg) was denatured at 94°C for 3 minutes in 40 μl including 20 μl 2.5× random primer mix and kept on ice. Nucleotide mix (5 μl; 2 mM dATP, 2 mM dGTP, 2 mM dTTP, 0.5 mM dCTP), 3.75 μl Cy3/Cy5-dCTP (Perkin Elmer, Beaconsfield, Bucks, UK) and 1.5 μl of Klenow fragment were added and the reaction was incubated at 37°C overnight. The labeled DNA was purified using the MinElute PCR purification kit (Qiagen, Crawley, West Sussex, UK) and the incorporated Cy3/Cy5-dCTP was quantified with the NanoDrop ND-1000 spectrophotometer. For gene expression analysis, cDNA synthesis and labeling were conducted as described previously 62.

For hybridization on Sco-Chip²-v1 arrays, 40 pmol of Cy3-labeled immunoprecipitated DNA was co-hybridized with the same amount of Cy5-labeled total input chromatin DNA in 500 μl Agilent hybridization buffer (1 M NaCl, 50 mM MES, pH 7, 20% formamide, 1% Triton X-100 buffer), in an Agilent Technologies hybridization chamber, rotated at 55°C for 60 h in an Agilent Technologies hybridization oven. For hybridization on Sco-Chip²-v2 arrays, 10-40 pmol of Cy3-labeled immunoprecipitated DNA were co-hybridized with the same amount of Cy5-labeled control mock IP DNA in 120 μl Agilent hybridization buffer as above. To control for Cy-dye bias, the hybridization was repeated with the same IP DNA samples labeled in the opposite dye orientation. Two biological replicates were hybridized on both array formats.

The arrays were washed once in 50 ml of 6 × SSPE, 0.005% N-lauryl sarcosine and once in 0.06 × SSPE, 0.18% polyethylene glycol 200, both for 5 minutes at room temperature. The arrays were briefly immersed in Agilent Technologies stabilization and drying solution prior to processing in an Agilent Technologies scanner. The probe signals were quantified using Agilent's Feature Extraction software (version 9.1.3.1).

Two different types of dual hybridizations were conducted on the arrays. With the Sco-Chip²-v1 arrays, HspR-IP chromatin was co-hybridized with Cy5-labeled total input chromatin as reference and the mock 'no-antibody' IP chromatin was also co-hybridized with total input chromatin on a separate array; the enrichment ratios for each probe were calculated as the signal from the former divided by that from the latter array. With Sco-Chip²-v2, the HspR-IP chromatin was co-hybridized directly with the mock 'no antibody' IP chromatin - the sample processed in the same way as the HspR-IP, but without the specific antibody. To compensate for any dye bias in the latter experiments, replicate hybridizations were conducted with both Cy3/Cy5 dye orientations on different arrays. It should be noted that the experimental design in terms of hybridizations are different for Sco-Chip²-v1 (IP or mock IP versus total chromatin) and Sco-Chip²-v2 (IP versus mock IP), being comparable, respectively, to 'traditional' microarray (expression) experimental designs of indirect and direct hybridizations. However, both designs are valid since ultimately the same output ratio of interest (IP/mock IP signal) is calculated. The direct hybridization approach, introduced for Sco-Chip²-v2, is preferred because there is likely to be a reduction in variance (as fewer arrays are used to derive the same ratios).

The RNA preparations used during this study were the same as those reported in 27. The RNA was isolated from 36 h S. coelicolor MT1110 cultures grown on SMMS agar, by the method reported previously 57; RNA from YEME plus 10% sucrose was isolated from 40 h batch cultures by the RNeasy method described in 27. RNA quality and integrity was re-assessed using the Agilent Bioanalyzer 2100 system. The Cy3/Cy5-dCTP labeled cDNA was synthesized from 10 μg RNA samples following the methods described in 27.

All (ChIP-on-chip and expression) hybridized arrays were scanned using an Agilent Technologies microarray scanner and resultant intensities calculated using Agilent Technologies Image Analysis and Feature Extraction software (version 9.1.3.1) with local background correction. Log₂ signal/reference ratios were calculated for all arrays, the reference channel either representing total input chromatin (Sco-Chip²-v1 array), mock 'no-antibody' IP (Sco-Chip²-v2 array) or cDNA (Sco-Chip²-v2 array).

ChIP-on-chip array data were not normalized as the typical microarray normalization assumptions do not hold 63. All expression arrays were normalized by the Loess method using the LIMMA package 6465 in R (version 2.5.0 6667). For the heat-shock experiment, across array normalization was applied such that the median absolute deviations (MADs) were similar (scale function in LIMMA); no further normalization was applied to the data from the hspR mutant/wild-type comparison.

Within the output of the Feature Extraction software (Agilent Technologies) there are four binary (1 for bad, 0 for good) variables (gIsFeatNonUnifOL/rIsFeatNonUnifOL, gIsBGNonUnifOL/rIsBGNonUnifOL, gIsFeatPopnOL/rIsFeatPopnOL and gIsBGPopnOL/rIsBGPopnOL) that describe outliers in each channel on an array. Spots on the Sco-Chip²-v1 array were flagged as poor quality if at least one of the four Feature Extraction variables for the total input chromatin reference channel had the value 1 (bad). Spots on each of the Sco-Chip²-v2 arrays were flagged if at least one of the two channels was classed as an outlier by Feature Extraction.

All data were filtered such that only those probes were retained for analysis that had good quality data (not flagged) in each replicate array (to control dye bias) within each independent experiment: 20,586 probes for Sco-Chip²-v1 array; 43,056 probes for Sco-Chip² (ChIP-on-chip); and 43,263 probes for the Sco-Chip²-v2 gene expression analysis.

The filtered data sets for the gene expression experiments were analyzed using rank products analysis 37 via the RankProd package in R (version 2.5.0) 68. This method has been shown to be robust in the identification of true differentially expressed genes in data sets where there are few replicates and/or large variance 6970. Differentially expressed genes were identified as having a pfp value 37 less than or equal to 0.15, equal to a false discovery rate of approximately 15%, a threshold value lower than that applied in the literature with this technique 37. The microarray-derived expression data have been deposited in ArrayExpress (accession numbers [E-MAXD-44], [E-MAXD-46] and [E-MAXD-49]).

Gel-shift assays were conducted using the Light Shift Chemiluminescence EMSA kit (Pierce), following the manufacturer's instructions. The 3'-biotinylated oligonucleotides comprising the HspR binding site, IR3 26 in the promoter region of the dnaK operon, and similar motifs in the promoter regions of SCO5639 and SCO4410 were annealed to their complementary strands and 200 fmol were used in the binding reaction together with either purified DnaK-refolded HspR (as described in 61) or a JM109 E. coli cell extract over-expressing hspR 26. In competition experiments a 200-fold excess of the same, but non-biotinylated, DNA was included in the electrophoretic mobility shift assay reaction. The dnaK biotinylated probe used in gel shift assays was 5'-TGCACACTTGAGCCTGTTCCACTCAAGTCAGCTGGAG; the SCO5639 biotinylated probe was 5'-TCGGATTGGAATTACTAAGATTCAGGATGCAGCACGCATCGT and the SCO4410 biotinylated probe was 5'-CGTTTCGGGTGAATCCCGAAAATTCCAGACGTTCCGACGAGG.