Introduction

With his reputation for causing springs suddenly to gush from hillsides to the consternation of uninvited guests, Janus was a favourite deity in Ancient Rome, which provided no fewer than seven hills on which this act might be rehearsed. In addition to ‘God of Beginnings and Endings’ (creatively taken to include most civic celebrations) the dual-faced Janus held various other portfolios, including ‘Guardian of Gateways, Doorways and Bridges’. Janus was also Keeper of the Gateway to his own temple, which was kept open in times of war (presumably to encourage performance of his party trick) and closed only during peacetime, i.e., rarely. There he stood in the doorway with one face illuminated and the other dark. Hence, Janus imagery has been widely used to symbolise opposites or to place intellectual concepts in apposition (as, powerfully, in [10]). The present article discusses interplay between positive and negative control mechanisms that operate during antibiotic production.

Antibiotic-biosynthetic gene clusters

Genes encoding the biosynthesis of antibiotics and other secondary metabolites are typically clustered within the respective actinomycete genomes or, in a few known cases, on giant linear plasmids. In genetical terms, low molecular weight antibiotics are quite complex; commonly 10–50 genes might be required to encode synthesis of a molecule of Mr <1,000. In contrast, a single gene can readily encode a 100 KDa protein. Each antibiotic-biosynthetic gene cluster occupies up to 1% of the actinomycete genome and organisms that produce multiple antibiotics (probably the norm; see [8]) harbour multiple discrete gene clusters dedicated to that end. In addition to genes that encode enzymes required for antibiotic production (‘biosynthetic genes’ hereinafter), such clusters also typically include one or more antibiotic-resistance determinants [9] and usually, but not always, one or more regulatory genes. The latter commonly regulate the gene clusters in which they are found. As a deceptively simple example, the streptomycin-biosynthetic (str-sts) cluster in Streptomyces griseus includes a single regulator (strR) plus a single resistance determinant. (‘Regulator’ is used here interchangeably to mean either a gene or its product). In contrast, the tyl gene cluster of S. fradiae, producer of tylosin, is decidedly extravagant with three resistance genes plus at least five regulators (Fig. 1). Nevertheless, cohabitation of regulatory and biosynthetic genes is not an invariant feature of these clusters. There is no regulator associated with the much-studied ery genes of Saccharopolyspora erythraea and no other gene in that organism has yet been shown to control erythromycin production. The tyl cluster also includes ‘ancillary’ genes that are non-essential for tylosin production [7]. Orthologues of these are commonly associated with primary metabolism in other organisms; paralogues are presumably present elsewhere in the S. fradiae genome.

Fig. 1
figure 1

The tylosin-biosynthetic gene cluster of S. fradiae. Not drawn to scale. The cluster occupies a contiguous portion of the genome (approximately 85 kb). Regulatory genes are outlined in boxes. Tylosin-biosynthetic genes are represented by black arrows. Resistance determinants (designated ‘tlr’), ancillary genes and others that are unassigned are represented as grey arrows. The full complement of biosynthetic genes could, in principle, be expressed from three pairs of divergent promoters (stars) via operon control

Full size image

Positive control mechanisms

Antibiotic-biosynthetic gene clusters are subject to cascade regulation that ultimately (i.e., at the ‘lowest’ hierarchical level) involves positive control (for a review, see [6]). Wherever negative control is encountered, there is typically a lower level of positive control, involving direct activation of otherwise cryptic promoters that mediate transcription of antibiotic-biosynthetic genes or operons. At that basal level, control is pathway-specific with no associated effect(s) on other aspects of metabolism.

Pathway-specific transcriptional activators

Following the advent of DNA sequencing, a widespread family of pathway-specific regulatory proteins, later designated ‘SARPs’ (Streptomyces Antibiotic Regulatory Proteins; [33]), was soon revealed by comparison of their deduced aminoacid sequences. Founder members of the SARP club included ActII-ORF4 and RedD, transcriptional activators of the act and red gene clusters that, respectively, encode production of actinorhodin and undecylprodigiosin in S. coelicolor, and DnrI that controls daunorubicin-biosynthetic genes in S. peucetius. However, although many antibiotic-biosynthetic clusters among Streptomyces spp. harbour SARP-encoding genes (commonly, one per cluster), this is not a golden rule. The activator StrR is not a SARP and regulation of streptomycin biosynthesis in S. griseus is SARP-independent.

Positive control by SARPs and other pathway-specific regulatory proteins is consistent with the observation that deletion or inactivation of the respective regulatory gene usually abolishes antibiotic production. Interestingly, however, such proteins are not normally present in saturating amounts. This became evident when ‘self-cloning’ of DNA fragments (later shown to increase the copy number of specific regulator genes) caused elevated levels of antibiotic production, ostensibly reflecting elevated levels of regulatory proteins (for a review, see [5]). This strategy remains a panacea for enhancement of antibiotic production and even applies to organisms already subjected to empirical strain improvement. Thus [24], an advanced production strain of S. fradiae generated significantly more tylosin following self-cloning of the pathway-specific regulator, tylR, or the SARP-encoding gene (tylS) that activates expression of tylR.

Following transcriptional analysis of (a few) gene clusters to locate the respective promoters, allied with genetical analysis to identify candidate regulators, and complemented by DNA-binding studies with purified regulatory proteins, it now appears plausible that entire antibiotic-biosynthetic gene clusters can be activated by single regulatory proteins (one per cluster). For example, the actII-ORF4 gene is proposed to control transcription from all of the act biosynthetic genes [11] and the ActII-ORF4 protein binds to two pairs of divergently oriented promoters located at key intergenic sites within the act cluster [1]. Similarly, the dnrI gene of S. peucetius controls transcription from the (daunorubicin-biosynthetic) dnr cluster [17] which, from gene organisation alone, must involve at least ten promoters. Within that cluster, the DnrI protein binds to at least two sites, again involving pairs of divergent promoters [31]. Finally for now, transcription from the whole of the streptomycin-biosynthetic gene cluster of S. griseus is controlled by strR with extreme parsimony in promoter usage [32]. Binding of the StrR protein has been demonstrated at three out of eight postulated control sites, with sequence-based predictions indicating at least one further target [22]. With so many genes to account for in any given cluster it is not surprising that most analyses are incomplete but, nevertheless, the aforementioned generalization seems plausible. A notable feature of this model is extensive usage of operon control to minimise the number of promoters, as exemplified by control of streptomycin production, viz., turn on the strR promoter and the resultant StrR protein turns on 25 str-sts biosynthetic genes via activation of just 8 promoters. All blocks of codirectional genes are controlled as operons. Nothing could be simpler!

Higher levels of control

Before all of that, of course, industrial microbiologists were already familiar with pleiotropic control of antibiotic production, having routinely encountered non-producing mutants that had also lost the ability to sporulate. Such phenomena pointed to proteins that function within multiple pathways, most likely as control elements linking the regulatory cascades that govern secondary metabolism and morphological differentiation. In the hierarchy of regulation, such linkage operates at higher levels than pathway-specific control. At even higher levels, extracellular stimuli elicit intracellular responses so that to understand fully the regulation of antibiotic production would be to understand in detail how an organism responds to changes in its environment. Progress there is hard made and classic reviews of microbial physiology, written a quarter of a century ago, are still essential reading [18]. In comparison, the manipulation of regulatory genes, conveniently encountered in readily identifiable clusters (as typified by studies in the present author’s laboratory), is much more straightforward.

Positive control of tylosin production

When the tyl gene cluster of S. fradiae was first sequenced, three genes immediately presented themselves as candidate positive regulators. Two of them (tylS and tylT) were deduced to encode SARPs whereas the other, tylR, was clearly related to acyB2 from S. thermotolerans, producer of carbomycin. The latter gene had already been described as a positive regulator [2], the first to be associated with macrolide antibiotic production. The novelty of finding two SARPs apparently associated with a single biosynthetic pathway was diminished somewhat when tylosin production survived insertional inactivation of tylT, leaving the role(s) of its product yet to be determined [4]. However, knockouts (KOs) of tylS or, earlier, tylR [3] totally abolished production of tylosin or any of its chemical components. (Tylosin is a ‘macrolide’ antibiotic, consisting of a polyketide lactone substituted with three deoxyhexose sugars.) Loss of the tylR transcript in tylS-KO strains (but not vice versa) revealed that TylS controls expression of tylR, and restoration of tylosin production in tylS-KO strains, due to forced expression of tylR from a heterologous promoter, revealed that TylR is a global activator of the tyl cluster. At least in such engineered strains, TylR can apparently activate all of the tylosin-biosynthetic genes with no essential role for TylS [4]; N. Bate et al., manuscript in preparation, author’s laboratory). Hypothetically, TylR might do that from pairs of divergent promoters at just three intergenic target sites (see Fig. 1). This was the first demonstration that some SARPs (here, TylS) can act above the basal level in regulatory cascades that govern antibiotic production.

Activation of tylR is not the only direct role of TylS. Immediately adjacent to [tylS-tylT] in the regulatory ‘nerve centre’ of the tyl cluster lie two genes (formerly designated orf12* and orf11*; see Fig. 1) whose transcription, it now turns out, does not survive disruption of tylS [24]. When first encountered [3], orf12* was unlike any entry in the database although a few orthologues (with no assigned or even speculative functions) have since been reported. It now transpires that orf12* (tylU) is another tyl regulatory gene. Although there is no obvious phenotype associated with overexpression of tylU in S. fradiae wild type, disruption of tylU reduces tylosin production by about 80%. Normal production can be restored in such strains by introducing an intact copy of tylU, or by overexpressing an extra copy of tylR, but not at all by tylS (N. Bate et al., manuscript in preparation, author’s laboratory). Bearing in mind that overexpression of tylR similarly rescues the tylosin-production phenotype of tylS-KO strains (which also lack TylU) these data suggest that TylS and TylU combine (mechanism unspecified) in the activation of tylR. Western analysis with ‘anti HIS-tag’ antibody, aimed at detecting endogenously produced TylR in engineered strains of S. fradiae, has confirmed that model. The level of HIS-tagged TylR produced in tylU-KO strains is much lower than in isogenic tylU + strains (D. Bignell, unpublished data, author’s laboratory). In summary, TylS drives expression of tylU and the TylU protein acts in combination with TylS to facilitate TylR production, for which TylS (but not TylU) is indispensable.

One size does not fit all

By now it is clear that different SARPs act differently in different organisms. During daunorubicin production by S. peucetius, or synthesis of actinorhodin and undecylprodigiosin by S. coelicolor, SARPs control directly the promoters of antibiotic-biosynthetic genes and seem to operate exclusively at that lowest regulatory level. In contrast, TylS might not act at that level at all. In a further variation on this theme, a SARP (CcaR) encoded in the ‘cephamycin C cluster’ of S. clavuligerus controls (at least some) cephamycin C-biosynthetic genes by activating multiple promoters while also controlling, directly or indirectly, another regulatory gene (claR). The latter is located in, and is specific for, the ‘clavulanic acid cluster’ (the two sets of genes are immediately adjacent in the S. clavuligerus genome where they form a ‘supercluster’). As claR seems to control late, but not early, ‘clavulanic acid genes’, it is not clear in this context whether “one activator per cluster” remains intact. That aphorism, together with “one cluster per activator”, would again be violated in the (perhaps, likely?) event that CcaR were shown to activate directly any of the ‘clavulanic acid genes’ (for discussion and references; see [23, 28]). In these various ways, what you see depends on where you look. Janus might well concur with that!

Negative control mechanisms

Gamma-butyrolactone autoregulators

Almost 40 years ago, it was shown by Kokhlov and colleagues that a small, diffusible molecule of the gamma-butyrolactone (GB) family could influence secondary metabolism and morphological differentiation in S. griseus [for a review of GBs, see [34]). Non-sporulating mutants that also failed to make streptomycin were restored to wild type behaviour if supplied exogenously with a GB designated ‘A-factor’ (Fig. 2). Those effects of A-factor were elicited with nanomolar concentrations of the GB and were observed only in mutants, which later proved to be defective in A-factor biosynthesis. Supplementation of S.griseus wild type with A-factor did not enhance streptomycin production. Since then, involvement of GBs in regulation of antibiotic production by Streptomyces spp. has been studied quite widely, but with greatest impact in two laboratories.

Fig. 2
figure 2

Structures of gamma-butyrolactone autoregulators. There are three chemical classes that differ in substitution at the 6-position (-keto as in A-factor from S. griseus; -alpha hydroxyl, as in virginiae butanolides e.g., VB-A from S. virginiae; -beta hydroxyl, as in IM-2 from S. lavendulae). The length of the hydrophobic tail can also vary

Full size image

From the work of T. Beppu, S. Horinouchi and colleagues in Tokyo over the past 20 years, A-factor mediated control of streptomycin biosynthesis in S. griseus has been shown to involve a regulatory cascade triggered by classical induction (i.e., de-repression) of a single activator gene, adpA. De-repression of adpA involves binding of A-factor to a repressor protein, ArpA, with consequent dissociation of the latter from the adpA promoter [21]. That is the sole binding target for ArpA which, in turn, is the only known receptor for A-factor in S. griseus. The AdpA protein activates a regulon of multiple regulatory genes whose products separately trigger secondary metabolism and various aspects of morphological differentiation, including sporulation [35]. In the context of streptomycin production, AdpA activates the strR promoter, thereby triggering production of the StrR protein which activates the streptomycin-biosynthetic gene cluster. Again, nothing could be simpler! As it happens, other systems are typically much less simple.

Incisive studies by Y. Yamada, T. Nihira and colleagues in Osaka, mainly focused on virginiamycin production in S. virginiae and the involvement of GBs known as ‘virginiae butanolides’ (VBs; Fig. 2), have revealed significant differences from ArpA-mediated events in S.griseus. For example, a VB-binding protein, BarA, from S. virginiae represses multiple target promoters and thereby blocks virginiamycin production, without affecting sporulation [14, 20]. Such repression is relieved by binding of butanolides to BarA which, unlike ArpA, also regulates synthesis of the cognate GB. Again unlike ArpA, but similar to FarA (a GB-binding protein from S. lavendulae [16]), BarA targets the promoter of its ‘own’ gene, barA (i.e., BarA is autoregulatory).

Related studies in Leicester were strongly influenced by the secure demonstration in Osaka that BarA binds to, and blocks expression of, its downstream neighbour, barB, and that VBs could reverse such repression by binding to BarA and causing its dissociation from the barB promoter [14]. When aminoacid sequences of the tylP and tylQ products were first deduced [3], they were clearly similar to BarA and BarB, respectively, suggesting that TylP (a candidate GB-binding protein) might regulate expression of tylQ, although at that time no function could be suggested for the TylQ protein or, indeed, for BarB. The function of TylQ was resolved experimentally before that of TylP.

Negative control of tylosin production

TylQ is a repressor that is proposed to target the tylR promoter. This model was derived from two observations. When the tyl gene cluster of S. fradiae wild type was subjected to transcript analysis via reverse transcription-polymerase chain reaction, tylQ was found to be active at early stages of fermentation but was the only tyl gene to be silent following the onset of tylosin production [25]. In parallel, similar analysis was applied to a strain (‘tylQ-KO’) within which tylQ had been disrupted and to another engineered strain (‘tylQ-OE’) in which tylQ was overexpressed under control of a strong, constitutive promoter. The tylQ-OE strain failed to express many of the tyl genes and produced no tylosin, even after extended fermentation. In contrast, tylosin production began early in the tylQ-KO strain and, significantly, transcripts from tylQ and tylR were never detected concurrently.

Enhanced tylosin production by tylQ-KO strains has industrial consequences. When tyl DNA was obtained from an advanced S. fradiae strain of the Lilly industrial production lineage, no changes were found in the coding sequences of any of the regulatory genes (tylP,Q,R,S,T,U) except for a single T–A transversion within tylQ [26]. This would have changed a single aminoacid residue, histidine to glutamine, within the downstream helix (significantly, the DNA recognition moiety) of an N-terminal helix–turn–helix motif that is characteristic of many repressors. None of the tyl regulatory genes had suffered promoter mutations during decades of empirical strain improvement involving random mutagenesis nor had any changes occurred in other obvious promoter regions (for example, between divergent biosynthetic genes). That single base change in tylQ occurred quite early in the Lilly strain lineage (perhaps very early) and it clearly ‘killed’ the TylQ repressor. Thus, tylosin production by an engineered derivative of S. fradiae wild type was abolished by constitutive expression of a wild type copy of tylQ but was unaffected by overexpression of the mutant allele derived from the production strain [26]. And that’s not all. Disruption of tylQ, followed by overexpression of tylR, in S. fradiae resulted in sequential and significant enhancements of tylosin production (N. Bate, unpublished data, author’s laboratory).

Confirmation that TylP does indeed control tylQ was preceded by the observation of ‘PARE’ sequences (see below), characteristic of target sites for GB-binding proteins, located upstream of tylP, tylQ and tylS [27]. Again, this development was predicated on work from the Osaka group [15] who derived consensus recognition sequences from multiple authentic target sites for BarA (‘BARE’ sequences) and the autoregulatory binding site of FarA (‘FARE’ sequence). In engineered strains of S. lividans, TylP powerfully inhibited expression of a reporter gene fused to tylP or tylQ promoter DNA containing the respective ‘PARE’ sequences and also, but less strongly, inhibited expression from the tylS promoter [27]. The clear inference was that TylP controls not only tylQ (a repressor represses a repressor) but also tylS and, in autoregulatory fashion, tylP. That model has recently been confirmed directly. His-tagged TylP was shown to bind specifically to the three respective ‘PARE’ sequences in vitro and such binding was reversed by extracts, ostensibly containing GB(s), from S. fradiae fermentation broths (D. Bignell et al., manuscript in preparation, author’s laboratory).

The current model for regulation of tylosin production, summarized in Fig. 3, is clearly not yet complete. The model says nothing about positive control of tylS and, in its present form, it probably cannot adequately account for enhanced production of tylosin in tylP-KO strains [27] although it does predict an increase in mycelial TylS content under such conditions. Presumably, in the absence of TylP, there must be an another way of turning tylQ down or off.

Fig. 3
figure 3

Current model for regulation of tylosin production in S. fradiae. At early stages of fermentation, in the presumed absence of gamma-butyrolactone(s), tylP is subject to autorepression, TylQ represses tylR, and the tyl gene cluster is not expressed, although TylS activates expression of tylU at this time. Subsequently, following presumed derepression of tylP by uncharacterized gamma-butyrolactone(s), tylQ is silenced allowing TylS and TylU to activate expression of tylR. The TylR protein activates the cluster of tylosin-biosynthetic genes

Full size image

Questions for the future

Like BarA, TylP controls antibiotic production by regulating multiple genes, including its own determinant. Whether, like BarA and ScbR (a GB-binding protein from S. coelicolor [29]), TylP also influences cognate GB production remains to be established. In addition to TylP, a growing list of GB-receptors, including ScbR [30] and SpbR from S. pristinaespiralis [12], each regulate a ‘SARP gene’. How common will this turn out to be? There is also a SARP gene (vmsR), associated with virginiamycin production, that functions downstream of barB in the regulatory cascade [13]. BarB now appears to be a repressor [19]; does it control vmsR directly? ArpA in S. griseus has only a single target promoter [see 32]. How common is that among GB-binding proteins? And to what extent might other low Mr compounds, including intermediates or end products of the respective biosynthetic pathways, also influence antibiotic production in addition, or as alternatives, to GBs?

Conclusions

Observed interactions between regulatory elements during antibiotic biosynthesis are becoming ever more dissimilar and not suggestive of a common pattern, which brings us to the final keynote conclusion. In the theatre of antibiotic production, genomics might well set the regulatory stage by listing the actors. But their entrances and exits, and what goes on in between, will differ from play to play. To understand how any specific plot is thickened will still require detailed experimentation — that means good old-fashioned biochemistry and genetics. Best of all, throw in some good old-fashioned microbial physiology and one day we might really understand the regulation of antibiotic production by actinomycetes.