Silencing cryptic specialized metabolism in Streptomyces by the nucleoid-associated protein Lsr2

Lsr2 is a nucleoid-associated protein conserved throughout the actinobacteria, including the antibiotic-producing Streptomyces. Streptomyces species encode paralogous Lsr2 proteins (Lsr2 and Lsr2-like, or LsrL), and we show here that of the two, Lsr2 has greater functional significance. We found that Lsr2 binds AT-rich sequences throughout the chromosome, and broadly represses gene expression. Strikingly, specialized metabolic clusters were over-represented amongst its targets, and the cryptic nature of many of these clusters appears to stem from Lsr2-mediated repression. Manipulating Lsr2 activity in model species and uncharacterized isolates resulted in the production of new metabolites not seen in wild type strains. Our results suggest that the transcriptional silencing of biosynthetic clusters by Lsr2 may protect Streptomyces from the inappropriate expression of specialized metabolites, and provide global control over Streptomyces’ arsenal of signaling and antagonistic compounds.


INTRODUCTION 13
Chromosomes are remarkably dynamic molecules. In eukaryotes, chromosome structure is 14 governed largely by histones, while in bacteria, organization is provided by the nucleoid-associated 15 proteins. Collectively, these proteins function both in architectural capacities and in regulatory roles. 16 Chromosome evolution in bacteria can be being driven by mutation, genome rearrangement, and 17 horizontal gene transfer, and work over the last decade has revealed that many bacteria have co-opted 18 nucleoid-associated proteins to additionally serve as 'genome sentinels', suppressing the inappropriate 19 expression of newly acquired DNA (Dorman, 2007;Dorman, 2014). This is thought to maximize 20 competitive fitness by repressing the expression of foreign DNA until it is either incorporated into the 21 existing regulatory networks of the host, or decays to a point that it is lost from the chromosome 22 (Navarre et al., 2007). 23 Different bacteria employ distinct nucleoid-associated proteins as xenogeneic silencers, 24 including H-NS in the proteobacteria, MvaT/MvaU in the pseudomonads (Castang et al., 2008), Rok in 25 Bacillus species (Smits and Grossman, 2010), and Lsr2 in the actinobacteria (Gordon et al., 2008). None 26 of these proteins share sequence or structural homology, but all act by binding to AT-rich regions within 27 the chromosome (Navarre, 2006;Castang et al., 2008;Gordon et al., 2010;Smits and Grossman, 2010). 28 H-NS has been the best-studied of these proteins. In Escherichia coli and Salmonella, H-NS represses the 29 expression of pathogenicity islands, endogenous phage genes, as well as other genes needed to respond 30 to environmental changes (Lucchini et al., 2006;Navarre, 2006). H-NS binds DNA as a dimer, and can 31 either polymerize along the DNA to form a rigid filament (Liu et al., 2010), or bridge DNA to facilitate 32 chromosome compaction (Dame et al., 2000;Dame et al., 2006); both activities can limit the activity of 33 RNA polymerase. Lsr2 is thought to function similarly to H-NS. To date, its study has been confined to 34 the mycobacteria, where Lsr2 specifically binds and represses the expression of horizontally transferred 35 genomic islands and AT-rich regions, including major virulence factor-encoding genes (Gordon et al., 36 2010). 37 In contrast to many of the pathogens in which chromosome organization and genome silencing 38 has been explored, the streptomycetes are largely benign, sporulating soil bacteria (Flärdh and Buttner,39 2009) that are instead renowned for their ability to produce a wide array of specialized metabolites 40 (Hopwood, 2007; Barka et al., 2016). Notably, the metabolic output of this actinobacterial genus 41 includes the majority of naturally-derived antibiotics used to treat bacterial infections. The 42 streptomycetes encode two Lsr2 paralogues, unlike their mycobacterial relatives who possess a single 43 lsr2 gene. Streptomyces are additionally unusual in that they have linear chromosomes, where the 44 majority of the genes required for viability are clustered in the chromosome core, and more species-45 specific and laterally-acquired genes are located in the flanking chromosome arms (Bentley et al., 2002). 46 It is within these arm regions that most of the specialized metabolic clusters are found. Recent work has 47 revealed that specialized metabolic clusters are over-represented as horizontally-transferred elements 48 in the streptomycetes (McDonald and Currie, 2017), and that in the closely-related Salinospora, lateral 49 gene transfer is a major driver of specialized metabolism (Ziemert et al., 2014). 50 Specialized metabolic gene clusters are subject to complex, hierarchical regulatory control (van 51 Wezel and McDowall, 2011;Liu et al., 2013). Most Streptomyces clusters contain dedicated pathway-52 specific regulators, which in turn are controlled by a suite of more globally-acting transcription factors. 53 Interestingly, however, most clusters are poorly expressed under normal laboratory conditions, and in 54 many cases their associated metabolites remain uncharacterized. This is also the case for the 55 filamentous fungi, many of whom have a broad, untapped specialized metabolic repertoire, courtesy of 56 transcriptional silencing by histones (Pfannenstiel and Keller, 2019). Significant efforts are being made to 57 stimulate the production of these 'cryptic' metabolites in both bacteria and fungi, as they are widely 58 regarded as productive sources of new natural products (Craney et al., 2013;Ochi and Hosaka, 2013;59 Scharf and Brakhage, 2013; Yoon and Nodwell, 2014;Daniel-Ivad et al., 2017;Onaka, 2017). 60 We sought to investigate the role of the nucleoid-associated proteins Lsr2 and LsrL in gene 61 regulation in Streptomyces. We found that deleting lsr2 from the chromosome of Streptomyces 62 venezuelae had minor effects on S. venezuelae growth and development and major effects on 63 profiles, we examined where Lsr2 bound, relative to any potential cluster-specific regulators. For 160 approximately half (8 of 14), Lsr2 binding was associated with a regulatory gene (Table 1). For the 161 others, Lsr2 bound elsewhere in the cluster, suggesting an independent mechanism of regulation. 162 163 Trends in Lsr2 binding and regulatory control 164 To better understand how Lsr2 exerted its repressive effects, we undertook a more 165 comprehensive investigation into its binding and regulatory impacts. We first validated the specificity of 166 Lsr2 binding using electrophoretic mobility shift assays (EMSAs). We tested five ChIP-associated 167 sequences, and found that each of these effectively out-competed non-specific DNA probes for binding 168 by Lsr2. This indicated that Lsr2 preferentially bound the DNA regions identified in our ChIP assays 169 ( Figure S5). 170 Our ChIP-seq results suggested that Lsr2 bound to 223 sites across the chromosome. 171 Interestingly, these sites were not concentrated in the arm regions where more of the species-specific 172 (and presumably laterally-acquired) sequences were located, but instead were enriched in the 'core' 173 region of the chromosome (as defined by Bentley et al., 2002) (Figure 2B; Table S3). When considering 174 all Lsr2 binding sites, ~25% of the associated genes (where Lsr2 bound immediately upstream and/or 175 overlapping with their coding sequences) had altered transcriptional profiles, and of these, more than 176 30% were in specialized metabolic clusters (19 of 63) ( Table 1; Table S3) 177 We assessed whether there was any correlation between binding site position and regulatory 178 impact. We found that binding sites within the arm regions were more likely to have transcriptional 179 ramifications compared with those in the core [35% (left arm), 40% (right arm), 25% (core)] ( Figure 2B). 180 Binding sites associated with transcriptional changes were also, on average, larger than those that had 181 no direct effect on transcription, at least for the left and core regions (Table S3; Figure 2C), although  182 there were a number of large sites within the core region that had no direct effect on transcription. 183 We next sought to understand whether there was any specificity to Lsr2 binding in the 184 chromosome. We analyzed the in vitro-confirmed Lsr2 binding sites (from Figure S5) using the MEME 185 server (Bailey et al., 2009); however, no consensus motif could be identified. In examining the cluster-186 associated binding sites more broadly, we found these sequences had an average GC-content of 62.9%. 187 When all Lsr2 binding sites were considered, an average GC-content of 65% was observed (Table S3), 188 well below the chromosome average of 72.4%. 189 Previous in vitro analyses of binding preferences for Lsr2 from M. tuberculosis, had defined an 190 eight nucleotide AT-rich sequence as being optimal (Gordon et al., 2011). We analyzed both the S. 191 venezuelae genome, and our identified Lsr2 binding sites, for either AT-rich 20 nt segments (>50% A/T), 192 or AT-rich 'core' sequences (defined as 5 of 6 consecutive nucleotides being A/T). To first determine the 193 relative AT density in the S. venezuelae chromosome, we assessed the number of 20 nt AT-rich stretches 194 in 30 randomly selected sequences -15 that were 500 bp and 15 that were 1000 bp in length (Dataset 195 S1). We found that 7/15 of the shorter sequences lacked any AT-rich stretch (with the number of 196 stretches ranging from 0-20, with an average of 5), compared with 2/15 of the longer sequences (with 197 numbers ranging from 0-36, and an average of 10). In contrast, the vast majority (222/223) of Lsr2 198 target sequences possessed at least one AT-rich 20 nt stretch: shorter target sequences (<500 bp) 199 contained anywhere from 0 to 27 non-overlapping stretches (average of 7), while longer sequences 200 (>750 bp) contained between 8 and 291 (average of 64) (Dataset S1). 201 We next assessed the presence of AT-rich core sequences within both the random genome 202 sequences, and the Lsr2-bound sites (Dataset S1). For the random segments, 11/15 of the 500 bp 203 sequences lacked an AT-rich core (with numbers ranging from 0-10, with an average <1). This closely 204 mirrored the absence of an AT-rich core in 10/15 of the 1000 bp sequences (range of 0-8, with an 205 average of 1.5). This is in stark contrast to the Lsr2 target binding sequences: only 9 of 223 target 206 sequences lacked an AT-rich core, with shorter sequences (<500 bp) averaging three core sites (ranging 207 from 0-11), and larger sequences (>750 bp) averaging 25 (ranging from 1-124). This collectively 208 suggested that while the presence of an AT-rich core sequence and multiple AT-rich segments may not 209 be sufficient to promote Lsr2 binding, they appear to be near universal characteristics of Lsr2-bound 210 sequences. 211 To experimentally assess the importance of these AT-rich sequences for Lsr2 binding, we 212 focussed our attention on the Lsr2 binding site located in the intergenic region between the divergently 213 expressed sven_5106 and sven_5107 genes within the predicted butyrolactone biosynthetic cluster 214 (Table 1; Figure S6). Using EMSAs, we compared Lsr2 binding to the wild type sequence (58% GC), with 215 binding to mutant sequences having increasing GC content (63%, 64% and 70%). Lsr2 bound the AT-rich 216 sequences with much higher affinity than the more GC-rich sequences, with very little binding observed 217 for the 70% GC-containing probe ( Figure S7). Notably, there was little difference in binding seen for 218 sequences in which an AT-rich core was disrupted (64% GC), versus when the overall AT-content was 219 changed (63% GC) (Figures S6 and S7). To determine whether the Lsr2 preference for more AT-rich DNA 220 was also observed in vivo, we introduced these altered sequences in place of the wild type, within a 221 larger DNA fragment (spanning 9 kb and encompassing sven_5105-07) on an integrating plasmid vector, 222 and introduced these variants into the Lsr2-3×FLAG-expressing strain. We conducted ChIP experiments 223 for each strain, and quantified the amount of DNA specific for this region using quantitative PCR. We 224 observed far higher levels of the wild type sequence compared with any of the mutant sequences (>80% 225 less), presumably reflecting greater Lsr2 affinity for the wild type, AT-rich sequence ( Figure S7). We also 226 assessed the expression of the flanking genes in each case, and observed little expression from the wild 227 type sequence, while increased expression was associated with the mutant sequences. This indicated 228 that decreased binding by Lsr2 led to increased transcription of the flanking genes ( Figure S7).  Table 1). As expected, we found 238 that Lsr2 bound and repressed the expression of genes in many poorly conserved/recently acquired 239 clusters. However, not all S. venezuelae-specific clusters were controlled by Lsr2, and several of the best 240 conserved clusters (e.g. siderophore/desferrioxamine-encoding biosynthetic cluster and bacteriocin-241 encoding clusters) were under direct Lsr2 control (Figure 4; Table 1). This suggested that Lsr2 may 242 function both as a silencer of newly acquired clusters, and as a central regulator within the hierarchy 243 governing specialized metabolic cluster expression. 244 245 Deleting Lsr2 reprograms specialized metabolism and yields novel compounds 246 Given the abundance of specialized metabolic genes affected by Lsr2, we examined the 247 antibiotic production capabilities of the lsr2 mutant. Crude methanol extracts from wild type and lsr2 248 mutant cultures were initially tested against the Gram-positive indicator bacterium Micrococcus luteus. 249 We observed a significant increase in growth inhibition for extracts from the lsr2 mutant relative to the 250 wild type strain ( Figure 5A). Using activity guided fractionation and purification coupled with LC/MS 251 analyses, we identified chloramphenicol as being the major inhibitory molecule ( Figure 5B). 252 Chloramphenicol is a well-known antibiotic, but it is not expressed at appreciable levels by S. venezuelae 253 under normal laboratory conditions ( Figure 5B) (Fernández-Martínez et al., 2014). 254 We next compared the soluble metabolites produced by wild type and lsr2 mutant strains, and 255 found each had a unique metabolic profile. We further tested the metabolic effects of deleting lsrL, and 256 lsr2 in conjunction with lsrL, as the increased lsrL expression observed in the lsr2 mutant suggested that 257 a double mutant may have more profound metabolic consequences than the lsr2 mutation alone. 258 Comparing the metabolic profiles of these four strains revealed that the greatest effect stemmed from 259 the loss of lsr2, although the loss of lsrL (on its own, and in conjunction with lsr2) led to minor changes in 260 metabolic output ( Figure S8). In comparing the production of individual metabolites in a wild type and 261 lsr2 mutant strain, we first focussed our attention on compounds produced after 3 days of growth in 262 liquid MYM medium. We observed unique peaks in the lsr2 mutant for venemycin, a chlorinated 263 venemycin derivative, as well as thiazostatin and watasemycin ( Figure 5C). These compounds have all 264 been described recently; however, this is the first time they have been shown to be produced in S.  (Table S4). Of these new and 275 enhanced compounds, only one was a known molecule (ferrioxamine), produced by a well-conserved 276 cluster under Lsr2 control (Table 1). 277 Included amongst the unique compounds was a novel peak of m/z 281 in the lsr2 mutant ( Figure  278 5D). Based on fragmentation analysis, this compound was predicted to be N-acetyl-7-chloro-L-279 tryptophan. To determine the gene cluster responsible for producing this compound, we searched for 280 halogenase-encoding genes. We identified sven_6229 as a reasonable candidate, as it was dramatically 281 (>200×) upregulated in an lsr2 mutant (Table S1). It was also part of a large, otherwise transcriptionally 282 silent specialized metabolic gene cluster (the 'NRPS-ladderane' cluster in Figure 3 and Table 1) . We 283 mutated sven_6229, and found the m/z 281 peak disappeared ( Figure 5D 2017), we compared the volatile molecules produced by wild type and lsr2 mutant strains. After 296 eliminating peaks associated with the growth medium, 742 discrete peaks were detected for both 297 strains. Of these, 65 were reproducibly differentially expressed, with 38 being more abundant in the 298 wild type, and 27 more abundant in the lsr2 mutant ( Figure 5E; Table S5), suggesting that volatile 299 metabolites may not be subject to the same regulatory controls as other specialized metabolites. 300 Generally, those compounds present at higher levels in the wild type had terpene-like properties. 301 Notably, a terpene-encoding cluster (sven_7101-7117) was amongst a handful of metabolic clusters 302 whose expression decreased in the absence of Lsr2 (Table 1; Table S1). In contrast, the volatile 303 metabolites that were more abundant in the lsr2 mutant appeared to be enriched for by-products of 304 specialized metabolic precursors (e.g. derivatives of pyruvate and acetyl-CoA). 305 306

Modulating Lsr2 activity stimulates new metabolite production in diverse Streptomyces species 307
The dramatic increase in metabolic production by the lsr2 mutant in S. venezuelae prompted us 308 to test whether it was possible to exploit this activity and stimulate new metabolite production in other 309 streptomycetes. In M. tuberculosis, a dominant negative allele of lsr2 has been reported, in which a 310 conserved Arg residue in the C-terminal DNA-binding domain is changed to an Ala residue (Gordon et 311 al., 2008). We constructed an equivalent Streptomyces variant (R82A mutant). Using EMSAs, we 312 confirmed that this protein was defective in its ability to bind DNA, and that it interfered with DNA 313 binding by the wild type protein ( Figure S10). We also cloned this dominant negative allele behind a 314 highly active, constitutive (ermE*) promoter on an integrating plasmid vector (Figure 6A), and 315 introduced this 'Lsr2 knockdown' construct into wild type S. venezuelae to test whether it was able to 316 phenocopy the lsr2 mutant. Using a bioassay, we detected increased antibiotic production for this 317 strain, relative to one carrying an empty plasmid vector ( Figure S10). We also introduced the construct 318 into the well-studied S. coelicolor strain, and observed copious production of the blue pigmented 319 metabolite actinorhodin when grown on a medium where this compound is not typically produced 320 ( Figure 6B). Finally, we tested the construct in a small library of wild Streptomyces isolates. We 321 screened for new metabolite production using a bioassay to assess antibiotic production. We first 322 introduced the Lsr2 knockdown construct into strain WAC4718. This led to a significant increase in 323 growth inhibition of M. luteus, and new growth inhibition of B. subtilis, relative to the plasmid-alone 324 control strain (Figure 6C). We next introduced the knockdown and control constructs into four 325 additional wild isolates (Figure 6D), and tested their antibiotic production capabilities against the 326 indicator strain M. luteus. We observed new and/or increased antibiotic production for two strains 327 (WAC7072 and WAC7520), no change in growth inhibition for one strain (WAC5514), and reduced 328 activity in the final strain (WAC6377). Notably, these strains did not grow appreciably differently 329 compared with their empty plasmid-containing parent strain (e.g. Figure S11). These results suggested 330 our construct had the ability to downregulate Lsr2 activity in a wide range of streptomycetes, and could 331 serve as a broadly applicable means of stimulating antibiotic production in these bacteria. repressing the expression of laterally acquired sequences in Streptomyces, as well as in suppressing the 338 expression of antisense RNA, as has also been observed for H-NS. Uniquely in Streptomyces, however, it 339 appears that Lsr2 function has been co-opted for the control of specialized metabolism, and that the 340 cryptic/silent nature of many of these metabolic clusters is due to direct Lsr2 repression. 341 342

Mechanism of Lsr2-mediated repression 343
Previous work on Lsr2 from the mycobacteria has shown Lsr2 preferentially binds AT-rich DNA 344 (Gordon et al., 2010;Gordon et al., 2011), and our findings suggest that this is also the case in the 345 streptomycetes. Unlike more conventional transcription factors, we found that Lsr2 binding sites in S. 346 venezuelae tended to be quite broad, centring on AT-rich sequences, extending hundreds (or thousands) 347 of base-pairs, and frequently encompassing promoter regions (Table S3) in those regions. We also identified multiple specialized metabolic clusters having more than one Lsr2 355 binding site (see Figure 3). This was particularly notable within the right arm of the chromosome (Table  356 1). These sites were often smaller (Table S3), and it is possible that gene repression is achieved through 357 bridging between these sites. 358 Many of the Lsr2 binding sites identified here, however, were not associated with altered 359 transcription of their flanking genes. It is conceivable that these sites serve more of an architectural 360 role, with Lsr2 binding promoting chromosome organization and compaction. Binding at these sites may 361 also exert indirect effects on transcription, as a result of altered DNA structure and accessibility. 362 In this study, we focussed our attention on the DNA-binding activity of Lsr2, but it is worth 363 noting that post-transcriptional regulatory roles have been identified for related proteins. In particular,  (Table S3), suggesting that, like H-NS, it negatively regulates 409 its own expression. How lsr2 expression is activated, and whether it is also subject to post-410 transcriptional regulation remains to be seen. 411 At a protein level, H-NS activity can be modulated by interaction with a multitude of proteins, 412 including association with paralogous proteins like StpA (Müller et al., 2010). Intriguingly, all 413 streptomycetes encode a paralogous Lsr2-like protein, termed LsrL. Our data suggest that there exists 414 regulatory interplay between these proteins, with Lsr2 repressing lsrL expression. It will be interesting to 415 see whether LsrL associates with Lsr2 to form hetero-oligomers, and whether such an association alters 416 Lsr2 activity. Deleting lsrL did not have profound phenotypic consequences, at least under the 417 conditions we tested, so understanding its biological role in Streptomyces will require additional 418 investigation. Unlike the streptomycetes, the mycobacteria do not encode additional Lsr2-like proteins. 419 However, recent work in M. tuberculosis has suggested that Lsr2 can associate with the unrelated 420 nucleoid-associated protein HU (Datta et al., 2019); whether an equivalent interaction occurs in 421 Streptomyces has yet to be determined. Lsr2 also appears to be subject to post-translational 422 modification, having been identified in several phospho-proteome screens conducted in Streptomyces  Table S6, while oligonucleotide information is 465 provided in Table S7 An in-frame deletion of lsr2 (sven_3225) was created using the ReDirect PCR targeting method 475 (Gust et al., 2003). The lsr2 coding region was replaced with the aac(3)IV-oriT resistance cassette, which 476 was subsequently excised using the yeast FLP recombinase to leave an 81 bp scar. The aac(3)IV-oriT 477 cassette was amplified from pIJ773 using the primer pair Sven3225disruptF and Sven3225disruptR2 to 478 generate an extended resistance cassette (oligonucleotide sequences listed in Table S7). Cosmid 1-C1 479 (http://strepdb.streptomyces.org.uk/) was introduced into E. coli BW25113 containing pIJ790, and the 480 lsr2 coding region was replaced with the extended resistance cassette. Cosmid 1-C1∆lsr2::aac(3)IV-oriT 481 was confirmed both via PCR using the flanking primers sven3225F2 and sven3225R2, and through a 482 diagnostic restriction digest. The modified cosmid was then introduced into S. venezuelae by 483 conjugation. Two representative apramycin-resistant, kanamycin-sensitive null mutants were selected 484 for morphological analysis. Cosmid 1-C1∆lsr2::aac(3)IV-oriT was introduced into E. coli BT340 in which 485 the FLP recombinase was induced to excise the aac(3)IV-oriT cassette from the cosmid. The cosmid 486 backbone was then targeted to replace bla with the hyg-oriT cassette from pIJ10701 (Gust et al., 2004). 487 The resulting cosmid was checked using PCR (Table S7) and restriction digest, prior to being mobilised 488 into S. venezuelae ∆lsr2::aac(3)IV-oriT. Hygromycin-resistant exconjugants were selected, and then 489 screened for a double cross-over event resulting in aparamycin-sensitive, hygromycin-sensitive scarred 490 mutants that were confirmed by PCR (Table S7).  (Table S7) and were subjected to 505 morphological and metabolic analyses. 506 To mutate sven_6229, CRISPR-Cas-mediated mutagenesis was used (Cobb et al., 2015), with 507 minor alterations to the published protocol. Briefly, a 32 nucleotide deletion, along with an in-frame 508 stop codon, was introduced into sven_6229. The guide RNA sequence was cloned into the BsbI site of 509 pCRISPomyces2, following the annealing of the overlapping oligonucleotides Sven6229 GuideF and 510 Sven6229 GuideR (Table S7). The editing template was generated by first amplifying fragments 511 upstream (Sven6229 UpF/R) and downstream (Sven6229 DownF/R) of the guide RNA sequences. These 512 sequences were then joined by overlap extension PCR, before being digested and cloned into the XbaI 513 site of the guide RNA-carrying pCRISPomyces vector. Sequence integrity of both the guide RNA and 514 editing template was confirmed by sequencing. The resulting plasmid was conjugated into the lsr2 515 mutant (Table S6), and exconjugants were selected for using apramycin and nalidixic acid. Colonies were 516 then streaked on MS agar plates without antibiotic supplementation, and were screened for the desired 517 deletion using the Sven6229 GuideF and Sven6229 DownR primers. Candidate deletion mutants were 518 subjected to a final PCR check, using Sven6229 UpR and Sven6229 ConR, and the resulting product was 519 sequenced to confirm the mutation. 520 To investigate the effects of AT-content on Lsr2 binding and gene expression, we focussed on a 521 validated Lsr2 binding site between sven_5106 and sven_5107, where the expression of these genes was 522 increased upon loss of Lsr2, suggesting Lsr2 repression. To clone a ~9 kb DNA fragment spanning 523 sven_5105-5107, the TOPO® TA cloning kit was used as per the manufacture's instructions. Briefly, the 524 fragment was amplified using the Phusion proofreading polymerase (New England Biolabs) with 525 oligonucleotides Sven5105_5107F and Sven5105_5107R (Table S7), a 72°C annealing temperature, and 526 cosmid Sv-3-D04 (Table S6) as template. The amplified product was purified by gel extraction, and was 527 then incubated with Taq polymerase and dATP at 72°C for 15 min. Four microlitres of the resulting A-528 tailed product were mixed with salt solution and pCR®2.1-TOPO® vector provided in the cloning kit, 529 before being introduced into Subcloning Efficiency™ DH5α™ competent cells (ThermoFisher 530 Scientific). The sven_5105-5107 containing plasmid was verified using restriction enzyme digestion and 531 sequencing. To create mutant variants, synthetic gene fragments were generated and amplified using 532 oligonucleotides Sven5106_5107F and Sven5106_5107R ( Table S7). The amplified products were cloned 533 between unique NheI and AvrII sites within the sven5105-07 sequences. The designed mutations were 534 confirmed by restriction digestion and sequencing. All validated sven_5105-5107 variants (wild type 535 and mutants) were excised from the TOPO vector using XbaI and SpeI, and cloned into the SpeI site of 536 pRT801. Constructs were then conjugated into wild type S. venezuelae and Δlsr2 mutant strains (for 537 expression analysis), and the Δlsr2 mutant strains complemented with either lsr2 or lsr2-3×FLAG (for 538 ChIP analyses). 539 540

Streptomyces cell extract preparation, SDS-PAGE, and immunoblotting 541
Cell extracts were prepared from a 1 mL aliquot of S. venezuelae cells grown in liquid MYM 542 medium. The protein extracts were separated using 15% SDS-PAGE and were stained with Coomassie 543 brilliant blue R-250. Equivalent amounts of total protein were loaded onto a second 15% SDS-PAG, and 544 following transfer to PVDF membranes, were subjected to immunoblotting with anti-FLAG antibody 545 (1:1,500; Sigma) and anti-rabbit IgG horseradish peroxidase (HRP)-conjugated secondary antibodies 546 (1:3,000; Cell Signaling). 547 548 Lsr2 overexpression, purification and electrophoretic mobility shift assays (EMSAs) 549 lsr2 amplified using primers NdeISven3225F and BamHISven3225R (Table S7) was digested and 550 the product cloned into similarly digested pET15b (Table S7). After sequencing, this construct was 551 introduced into E. coli Rosetta cells ( Table S7). Overexpression of 6×His-lsr2 was achieved by growing 552 cultures at 37˚C to an OD600 of 0.5, and then adding 0.5 mM IPTG (isopropyl b-D-1-553 thiogalactopyranoside). Cells were grown for a further 3 h at 30˚C before harvesting and resuspending in 554 binding buffer (50 mM NaH2PO4, 300 mM NaCl and 10 mM imidazole, pH 8.0) containing 1 mg/mL 555 lysozyme and one complete mini EDTA-free protease inhibitor pellet (Roche). Cell suspensions were 556 incubated on ice for 20 min before sonication. 6×His-Lsr2 was purified by binding to 1 mL Ni-NTA 557 agarose (Invitrogen), after which the resin was collected and the bound protein was washed with 558 binding buffer supplemented with increasing concentrations of imidazole. Purified proteins were 559 ultimately eluted using 500 mM imidazole. Purified protein was exchanged into storage buffer (20 mM 560 Tris-HCl, pH 8, 150 mM NaCl, 25% glycerol and 1.4 mM -mercaptoethanol) using an Amicon Ultra-15 561 Centrifugal Filter with a 3 kDa cut-off. Bradford assays were conducted to measure protein 562 concentrations. 563 EMSAs were performed using 124 -280 bp probes amplified by PCR and 5'-end-labelled with [γ-564 32 P]dATP (primers prefixed "emsa" are listed in Table S7). Increasing concentrations of Lsr2 (0-5 µM) 565 were combined with either 1 or 10 nM probe, 1 mg/mL bovine serum albumin (BSA) and binding buffer 566 (10 mM Tris, pH 7.8, 5 mM MgCl2, 60 mM KCl and 10% glycerol). Each reaction was incubated for 10 min 567 at room temperature, followed by 30 min on ice prior to adding a glycerol-based loading dye and 568 running on a 12% native polyacrylamide gel. To test binding specificity, competition assays were 569 established in which increasing concentrations (0-160 nM) of unlabeled probe were added together with 570 4 nM labelled probe and 1 µM Lsr2, to the EMSA reactions described above. Gels were exposed to a 571 phosphor plate for 1 h, before being visualized using a phosphorimager (Typhoon FLA 9500). 572

RNA isolation and RT-(q)PCR 574
Wild type S. venezuelae and the Δlsr2 mutant strain were grown in 300 mL MYM cultures in 575 duplicate. After 8 h (vegetative growth), 12 h (early mycelial fragmentation) and 18 hours (late mycelial 576 fragmentation/ sporulation), density at OD450 was measured, and a 60-90 mL sample was harvested. 577 Subsequent experiments involved growing wild type and Δlsr2 mutant strains carrying sven_5105-578 sven_5107 variants on an integrating plasmid. These strains were grown in duplicate, in 50 mL MYM 579 liquid medium for 18 h. In all cases, RNA was isolated as described in Moody et al. (Moody et al., 2013), 580 using a modified guanidium thiocyanate protocol (Chomczynski and Sacchi, 1987). Primers HrdBF and 581 HrdBR, or SVEN4987F/SVEN4987R (Table S7) were used for PCR checks, alongside a quantified 582 chromosomal DNA control, to confirm any DNA contamination was <0.005%. 583 Reverse transcription (RT) reactions were performed as described previously (Haiser et al., 2009;584 Moody et al., 2013). In brief, gene-specific reverse primers (Table S7), or random oligonucleotides were 585 annealed to 1 g of total RNA prior to cDNA synthesis using SuperScript® III reverse transcriptase 586 (Invitrogen) (wild type and mutant) or Lunascript TM RT (New England Biolabs) (sven5105-5107 variants), 587 respectively. 588 To validating RNA-sequencing results, two microlitres of the resulting cDNA were used as 589 template DNA for PCR, with a 58C annealing temperature. The number of cycles was optimized to 590 ensure that amplification was occurring within the linear range of the reaction (33 cycles for sven_0514, 591 sven_6216, sven_6264 and hrdB and 30 cycles for sven_0493 and sven_5135). Negative control 592 reactions were run to confirm the absence of genomic DNA contamination in the RNA samples, and 593 involved adding an equivalent volume of a reverse transcription reaction in which nuclease free water 594 had been added in place of reverse transcriptase. All reverse transcription reactions and PCR 595 amplifications were carried out in duplicate, using RNA isolated from two independent cultures. 596 For the sven_5105-5107 variant-containing strains, 2.5 μL of cDNA (1:4) were used as template 597 for qPCR. Primers 5106F/5106R were used to amplify target gene from cDNA with a 55C annealing 598 temperature. 'No RT' samples were run to confirm no DNA contamination. All samples were assessed in 599 biological duplicate and technical triplicate. qPCR data were normalized to rpoB and were analyzed using 600 a relative quantification method (2 -DDC T) (Livak and Schmittgen, 2001). Raw sequencing reads were trimmed to remove low-quality 3' ends using PrinSeq (Schmieder 611 and Edwards, 2011). Trimmed reads were checked for quality using FastQC 612 (www.bioinformatics.babraham.ac.uk/projects/fastqc/) and aligned to the S. venezuelae ATCC 10712 613 genome sequence using Bowtie2 (Langmead et al., 2009). 614 The resultant SAM files were converted to BAM format, sorted by genomic position and indexed 615 to create bai files (Li et al., 2009). The BAM files were analyzed both visually using Integrated Genomics 616 Viewer (Version 2.3.60) (Robinson et al., 2011), and using Rockhopper2 (Tjaden, 2015). We assigned a 617 cut-off for significance using a p-value adjusted for multiple testing that was less than 0.01 (q-value), and 618 filtered for genes displaying a fold change greater than four. 619 620 Chromatin immunoprecipitation 621 S. venezuelae ∆lsr2 was complemented using an integrating plasmid (pIJ10706/pIJ82) carrying 622 either wild type lsr2 or lsr2-3×FLAG (Table S6). Each culture was then grown in 300 mL MYM cultures in 623 duplicate. After 18 h, the density at OD450 was measured and the developmental progression of each 624 strain was monitored by light microscopy. A 1 mL sample was then taken for immunoblot analysis, and 625 an 80 mL sample was transferred to a sterile flask. Formaldehyde was added to a final concentration of 626 1% (vol/vol) to cross-link protein to DNA, after which cultures were incubated for a further 30 min. The reads in the fastq files were first checked for quality using FastQC 638 (www.bioinformatics.babraham.ac.uk/projects/fastqc/), then aligned to the S. venezuelae ATCC 10712 639 genome (GenBank accession number NC_018750) using Bowtie2 (Langmead et al., 2009). The resultant 640 SAM files were converted to BAM format, sorted by genomic position and indexed to create BAI files (Li 641 et al., 2009). The BAM files were visualized using Integrated Genomics Viewer (Version 2.3.60) 642 (Robinson et al., 2011), and were subjected to quantitative analysis. 643 MACS2 was run from the command line to normalize all lsr2 and the lsr2-3×FLAG samples 644 against total DNA with the mappable genome size set at 7.92×10 6 (90% of the S. venezuelae genome) to 645 generate BED files (Zhang et al., 2008). The BED files were in turn used to generate a CSV sample sheet 646 that was read into the R package for statistical computing (Team, 2013)  Strains were grown in 10 mL of MYM medium overnight, before being subcultured in duplicate 657 into 50 mL of MYM medium. After incubating for 18 h, formaldehyde was added to a final concentration 658 of 1% (v/v) to cross-link protein to DNA. The cultures were incubated for an additional 30 min, at which 659 time glycine was added to a final concentration of 125 mM. Immunoprecipitation was performed as 660 described above, only using the FLAG M2 antibody (Sigma). 661 To quantify the abundance of target genes of interest in the ChIP DNA, 20 μL qPCR reactions 662 were prepared using Luna® Universal qPCR Master Mix (New England Biolabs) and 2.5 μL of ChIP DNA 663 (1:10) as template. Primers 0926F/0926R and 5105F/5105R (Table S7) were used to amplify target gene 664 from ChIP DNA with a 55C annealing temperature. qPCR data was then analyzed using DART-PCR 665 (Peirson et al., 2003). Antibiotic bioassays were performed by testing methanol extracts of S. venezuelae grown for 1 701 to 3 days against Micrococcus luteus. Twenty microliters of each extract was applied to a Whatman 702 antibiotic assay disc and applied to a lawn of LB inoculated with a 25-fold dilution of the M. luteus 703 indicator strain grown to mid-exponential phase. The plates were incubated overnight at 30°C before 704 measuring the size of the zone of clearing. Bioassays for wild Streptomyces were performed by growing 705 isolates (knockdown-and plasmid-control containing) on ISP4 supplemented medium for 6 days at 30 o C. 706 Overnight cultures of indicator strains (M. luteus or B. subtilis) were mixed 1% soft nutrient agar, which 707 was allowed to solidify before being overlaid atop the wild Streptomyces strains, after which the 708 cultures were incubated overnight at 37 o C. 709 710

Synthesis of N-acetyl-chloro-tryptophan standards 711
Synthesis of N-acetyl-5-chloro-L-tryptophan was achieved using Wang resin (50 mg, 1 mmol/g 712 loading), which was swollen in anhydrous dimethylformamide (DMF). Fmoc-5-chloro-L-Trp-OH (57.5 mg, 713 0.125 mmol, 2.5 eq.) was dissolved in 5 mL 9:1 DMF:dichloromethane (DCM) and cooled to 0°C. 714 Diisopropylcarbodiimide (DIC) (6.3 mg, 0.050 mmol, 1 eq.) was added in minimal DCM. The reaction was 715 stirred for 30 min at 0°C, in a flask fitted with a drying tube. The anhydride mixture was added to the 716 swollen resin, after which DMAP (0.6 mg, 5 µmol, 0.1 eq.) was added, and the flask was then periodically 717 agitated at room temperature for 2 h. The resin was washed in 3 × 10 mL DMF, followed by 3 × 10 mL 718 DCM. The Fmoc group was removed by the addition of 10 mL 20% (V/V) piperidine in DMF, after which 719 the suspension was agitated for 20 min. The resin was washed as above, before acetylation was carried 720 out with the addition of acetic anhydride (5.0 µL, 0.05 mmol, 1 eq.) and diisopropylethylamine (DIPEA) 721 (1 µL, 5 µmol, 0.1 eq.) in 5 mL DMF. The resulting suspension was agitated for 30 min at room 722 temperature. The resin was washed as above, prior to cleavage being carried out with 10 mL 95% TFA, 723 2.5% triethylsilane, 2.5% DCM for 30 min. The eluent was then collected and evaporated to dryness. 724 Analysis by LC-MS was completed without any further purification. 725 To synthesize N-acetyl-7-chloro-L-tryptophan, 7-chloro-L-Trp-OH (24 mg, 0.1 mmol) was 726 dissolved in 20 µL 50 mM ammonium bicarbonate. Fifty microliters of an acetylation mixture (20 µL 727 acetic anhydride, 60 µL methanol) were added to the amino acid, after which the mixture was agitated 728 for 1 h at room temperature. The solvent was evaporated to dryness and the resulting product was 729 analyzed without additional purification. 730 731

Analysis of volatile metabolites 732
Volatile metabolites in the headspace of culture supernatants were concentrated, analyzed, and 733 relatively quantified using headspace solid-phase microextraction coupled to two-dimensional gas 734 chromatography time-of-flight mass spectrometry (HS-SPME-GC×GC-TOFMS), as described previously 735 (Jones et al., 2017). Four millilitres of culture supernatants were transferred to 20 mL air-tight 736 headspace vials and sealed with a PTFE/silicone cap (Sigma-Aldrich). A 2 cm triphasic solid-phase 737 microextraction (SPME) fiber consisting of polydimethylsiloxane, divinylbenzene, and carboxen (Supelco) 738 was suspended in the headspace of the supernatant for 30 min at 37°C with 250 rpm shaking. 739 The SPME fiber was injected into the inlet of a Pegasus 4D ( ). An inter-chromatogram mass spectral match score ≥ 600 (out of 1000) and 752 maximum first and second dimension retention time deviations of 6 s and 0.15 s, respectively, were 753 required for peak alignment. Only peaks detected at a signal-to-noise ratio of ≥ 50:1 in one or more 754 chromatogram were considered for subsequent analyses. Mass spectra were compared with the 755 National Institute of Standards and Technology (NIST) 2011 mass spectral library, and a forward match 756 score ≥ 700 (out of 1000) was required for putative compound identification. When possible, putative 757 identifications were affirmed by comparing experimentally-determined linear retention indices (using C6 758 to C15 straight-chain alkanes, Sigma-Aldrich) with previously-reported values for both polar and non-759 polar column configurations. 760 Relative compound abundances (measured in total ion chromatogram (TIC)) were log10-761 transformed, mean-centered, and unit scaled prior to statistical analysis. The non-parametric  Whitney U-test (Mann and Whitney, 1947) with Benjamini-Hochberg correction (Benjamini and 763