Where to Begin? Sigma factors and the selectivity of transcription initiation in Bacteria

Abstract

Transcription is the fundamental process that enables the expression of genetic information. DNA-directed RNA polymerase (RNAP) uses one strand of the DNA duplex as template to produce complementary RNA molecules that serve in translation (rRNA, tRNA), protein synthesis (mRNA), and regulation (sRNA). Although the RNAP core is catalytically competent for RNA synthesis, the selectivity of transcription initiation requires a sigma (σ) factor for promoter recognition and opening. Expression of alternative σ factors provides a powerful mechanism to control the expression of discrete sets of genes (a σ regulon) in response to specific nutritional, developmental, or stress-related signals. Here, I review the key insights that led to the original discovery of σ factor 50 years ago and the subsequent discovery of alternative σ factors as a ubiquitous mechanism of bacterial gene regulation. These studies form a prelude to the more recent, genomics-enabled insights into the vast diversity of σ factors in Bacteria.

ABBREVIATED SUMMARY

The sigma subunit of RNAP was first purified 50 years ago by Burgess et al. (1969) and shown to function as a dissociable subunit to allow recognition of specific transcription start sites (promoters). Subsequent studies have revealed that sigma factor replacement with alternative sigma factors (including ECF sigma factors) is potent transcriptional regulatory mechanism in Bacteria.

“Begin at the beginning,” the King said, very gravely, “and go on till you come to the end: then stop.”

Lewis Carroll, Alice in Wonderland

INTRODUCTION

The current revolution in biology is premised on our understanding of the role of DNA as the genetic material. The “central dogma” asserts that DNA is the repository of genetic information, and this information is used for expression of RNA and protein through the sequential processes of transcription and translation, respectively. DNA itself can be an enormous polymer, with bacterial chromosomes usually containing well in excess of 10⁶ bp, and chromosomes in eukaryotes typically much larger. The selective expression of this material implies there are well coordinated mechanisms that determine where and when specific genes and operons are transcribed into RNA. Similarly, RNA molecules must include signals that tell the ribosome where to begin and where to end during protein synthesis.

Here, I will review some of the key events that led to the discovery of bacterial σ factors and their role in promoter recognition, and the ability of σ substitution to serve as a powerful regulatory mechanism. My goal will be to put the discovery of ECF σ factors in a broad historical context. These early studies, particularly those leading to the discovery of σ (50 years ago), and leading to the identification of alternative σ factors in bacteria (40 years ago), provide insights into the challenges of dissecting complex enzyme mechanisms at a time when the tools that power contemporary biology (e.g. DNA cloning and restriction enzymes, polymerase chain reaction (PCR), and DNA sequencing) were yet to be invented.

The characterization of template-dependent nucleic acid polymerases

Perhaps the most logical place to begin this story is the landmark paper describing a proposed structure and likely mechanism of replication for DNA as published by Watson and Crick in 1953 (Watson & Crick, 1953). One key insight from the structure was famously summarized in the following sentence: “It has not escaped our notice that the specific pairing we have postulated immediately suggests a possible copying mechanism for the genetic material.” However, what was not apparent was how this would occur. The answer would emerge from the laboratory of Arthur Kornberg who purified the first DNA-directed DNA polymerase and demonstrated that this enzyme could copy DNA in the test tube. One might imagine that these studies simply looked for an enzyme that could “copy” DNA in the presence of all four substrates (deoxyribonucleoside triphosphates, dNTPs). The actual studies were far more intricate, and involved a complex fractionation to determine the components of an Escherichia coli cell extract needed for the incorporation of one added substrate (³²P-labeled dTMP) into a polymeric form precipitable by acid. Fractionation of the extract revealed the requirement for DNA, the four nucleotides, and associated nucleotide kinases (Kornberg, 1989, Lehman, 2003). As Kornberg recalls, the original rationale for including DNA in the in vitro reactions was based on the assumption it would serve as a primer for polymer synthesis, rather than as template, and that it would also help shield any newly synthesized material from DNases (Kornberg, 1989).

These early (1955) studies of DNA synthesis were the beginning of efforts that established several facts familiar to molecular biologists today: dNTPs are the substrates, synthesis is 5’ to 3’ and requires both a primer and a single-stranded DNA template, and synthesis is processive. That these early insights were gleaned from studies of an enzyme that turned out not to be the replicative DNA polymerase was of minor consequence, since these properties are universal. The enzyme activity that Kornberg initially characterized was, unexpectedly, found to not be the major replicative polymerase (De Lucia & Cairns, 1969), and functions instead in DNA repair and primer removal (DNA polymerase I). Replication involves a larger and more complex assemblage of proteins (the DNA polymerase III holoenzyme) (McHenry, 1991). Kornberg and legions of subsequent researchers would fill in the details of how DNA polymerase knows where to begin (at sites called origins), how primases, helicases, topoisomerases, and numerous other accessory factors enable efficient replication, and how replication terminates and chromosomes segregate.

The early history of RNA polymerase is in many ways intertwined with and motivated by these early biochemical studies of DNA polymerase. Indeed, the 1959 Nobel Prize in Physiology or Medicine 1959 was awarded jointly to Severo Ochoa and Arthur Kornberg “for their discovery of the mechanisms in the biological synthesis of ribonucleic acid and deoxyribonucleic acid.” Although Marianne Grunberg-Manago and Severo Ochoa had discovered an RNA polymerase activity, this enzyme (polynucleotide phosphorylase; PNPase) is involved in RNA degradation in cells, and not synthesis (Grunberg-Manago, 1989). Even at the time of the award, it was apparent that PNPase was likely not a template-directed enzyme. Nevertheless, PNPase provided a valuable tool for the in vitro synthesis of RNA copolymers of various composition that then proved essential to the rapid deciphering of the genetic code (by determining which amino acids where incorporated by ribosomes bound to RNAs of known base composition, but not sequence). Ochoa himself has stated that PNPase “may be considered to have been the Rosetta Stone of the genetic code” (Ochoa, 1980).

The first biochemical evidence for the template-directed RNA polymerase (RNAP) necessary for transcription emerged from purification of activities that synthesized RNA in a reaction requiring NTPs, and specifically stimulated by the addition of a DNA template (Fig. 1). By 1960, such activities had been detected independently in four different laboratories working with rat liver nuclei (Samuel Weiss), Escherichia coli (Jerry Hurwitz and Audrey Stevens), and pea extracts (James Bonner) (reviewed in Hurwitz, 2005, Kresge et al., 2006). The typical assay used in these studies was to monitor the incorporation of ³²P-radiolabelled NTPs into an acid-insoluble product that could be trapped on a filter and quantitated. A key challenge in these early studies was to demonstrate that the RNA was a product of template-directed transcription. Even demonstrating the DNA-dependence of the reaction was a challenge since this required the development of purification methods for RNAP that could efficiently separate it from DNA and ribonucleases also present in the extracts. Next, it was necessary to document that the nucleotide composition of the RNA product was broadly reflective of that of the substrate. As one approach, the synthetic copolymer poly-d(AT) could be used as template. Although initial efforts were met with mixed success due to nuclease contamination (Hurwitz, 2005), Michael Chamberlin and Paul Berg developed a robust purification method and demonstrated that the product was RNA with the expected composition (poly-r(AU)) (Chamberlin & Berg, 1962).

Fig. 1. DNA and RNA synthesis up to the discovery of σ.

Early studies of DNA and RNA synthesis were enabled by biochemical approaches, including the purification and characterization of the enzymes and their substrate requirements. Unlike DNA polymerase, RNA polymerase does not require a pre-existing primer for synthesis, and can recognize start sites (promoters). This latter activity requires the σ subunit, discovered in 1969 (inset is adapted from Burgess et al., 1969).

The demonstration that RNAP could catalytically copy a DNA template into RNA was ultimately supported by two major types of study. First, it could be demonstrated that the template composition of the RNA product was reflective of the DNA template using a nearest neighbor analysis of dinucleotide composition (modern sequencing methods were not yet invented). Moreover, Chamberlin and Berg showed that the amount of RNA product could exceed the amount of DNA template, indicative of a catalytic reaction in which each template could be used more than once (Chamberlin & Berg, 1962). Second, the RNA product could be shown to anneal specifically to the DNA template used for its synthesis, suggestive of specificity (Geiduschek et al., 1961). These were challenging experiments, and involved the in vitro synthesis of ³²P-radiolabelled RNA, the use of diverse DNA templates, and the separation of RNA, DNA, and RNA-DNA hybrids by cesium chloride (CsCl) ultracentrifugation. Having established that RNAP could be purified in an active form, and that RNA synthesis seemed to be template-directed, the stage was now set for studies to define the start and stop signals that must determine transcriptional selectivity in cells.

The discovery of sigma (σ) and the molecular basis for promoter recognition

With the availability of purified preparations of bacterial RNAP, and ample evidence that cells contained a short-lived RNA intermediate that could be used to program ribosomes for protein synthesis (mRNA), the challenge emerged of reconstituting the specificity of transcription initiation in vitro. The key observation, made independently by Richard Burgess and Andrew Travers (Harvard) and by John Dunn and Ekkehard Bautz (Rutgers), was that E. coli RNAP could be purified in two distinct forms that we would now recognize as the core and holoenzymes (Burgess et al., 1969). The core enzyme, obtained in a protocol culminating with phosphocellulose chromatography had subunit composition ββ’α₂ (Burgess, 1969), but was poorly active on purified phage T4 DNA while retaining activity on calf thymus DNA. Conversely, if the final purification was instead a glycerol gradient, the resulting enzyme retained high activity with the T4 template (Burgess et al., 1969). Moreover, further fractionation of this enzyme led to the separation of core and holoenzyme components, as visualized by the newly introduced technique of SDS-polyacrylamide gel electrophoresis (Fig. 1). When the σ factor was itself purified and added back to the core enzyme, high activity with the T4 template was restored, and σ could act catalytically to initiate RNA chains (Travers & Burgess,1969). This factor, then, seemed to answer the question: how does RNAP know where to begin? Presciently, even at this early stage Burgess noted that, “the interesting possibility arises that several similar factors could exist, each with a specificity for a different type of initiation site” (Burgess et al., 1969).

With these early insights into the specificity of initiation, attention was also drawn to the converse question: how does RNAP know where to stop synthesis? Jeff Roberts, working in the same Harvard Biological Laboratories as Dick Burgess, described the Rho transcription termination factor based on its activity on in vitro transcription of λ DNA (Roberts, 1969). These two Nature papers, reporting the discovery of σ and Rho, appeared (appropriately) at the beginning and end of 1969, respectively. Unlike σ, the Rho subunit is not required for all termination events and core RNAP itself recognizes intrinsic terminators at the end of many transcription units. However, Rho plays an important role in preventing transcription of horizontally transcribed DNA and long, untranslated stretches of RNA and is an essential protein in many bacteria (Mitra et al., 2017).

Biochemically, one of the key insights that emerged during this first decade of studying RNAP is that analysis of the biologically relevant transcription reaction required a well-defined template. In this era, prior to advent of restriction enzymes and molecular cloning technologies (let alone PCR), phage DNA provided such a template. Lytic phage can be purified in large quantities from lysates and the DNA extracted for biochemical use. Moreover, since many phage rely on host RNAP for early transcription, the corresponding early promoters are often highly active. In contrast, calf thymus and other more complex DNA preparations consist of randomly sheared fragments and are riddled with nicks, gaps, and single-stranded regions that bind RNAP and in some cases bypass the need for the σ subunit for initiation (Aiyar et al., 1994, Fredrick & Helmann, 1997). The use of purified phage DNA as template was also critical to the first successful efforts to identify alternative σ factors (see below).

In the decade following the discovery of σ factor the essential features of the bacterial transcription cycle would emerge, and the basic architecture of bacterial promoter sites was established (Chamberlin, 1976). Numerous studies, including many from the lab of Michael Chamberlin (UC Berkeley), established the distinction between closed and open complexes, and the temperature dependence of open complex formation led to the idea that this step was associated with localized melting of promoter region DNA to expose the template strand (Chamberlin, 1974). For these studies, the three strong promoters at one end of the phage T7 genome served as a model. The first insights into the nature of promoter recognition emerged when David Pribnow biochemically purified a DNA fragment bound to RNAP and then used this fragment to purify RNA sequences that annealed to this region and could be sequenced (sequencing RNA was easier than DNA at the time) (Pribnow, 1975). The conservation of the TATAAT sequence providing a first glimpse of what would later be known as the core consensus elements located near −35 and −10 relative to the transcription start site (Losick & Pero, 1981).

Alternative σ factors in bacteriophage gene regulation

Despite very active interest in bacterial RNA polymerase, evidence of alternative σ factors was not immediately forthcoming. Nevertheless, efforts towards defining mechanisms of altering transcriptional selectivity were successful. Of particular note, soon after the discovery of σ, Chamberlin sought to purify the transcription factor needed for late transcription in phage T7, expecting that this protein might be an alternative σ factor. The product of gene 1 was unexpectedly found to be a stand-alone, single-subunit RNAP rather than an alternative σ factor for the host RNAP (Chamberlin et al., 1970) (Fig. 2). The revelatory insight came when it was observed that the active fraction was insensitive to rifampicin (which inhibits all core RNAP activity); a notable example of how a “failed” control experiment foretold a major discovery. T7 RNAP proved to be a boon to biotechnology and was subsequently harnessed for numerous applications, including protein and RNA overproduction for biochemical analyses, RNA interference, and systems biology applications (Studier et al., 1990, Borkotoky & Murali, 2018, Wang et al., 2018). More recently, T7 RNA polymerase has been key to the synthesis of large amounts of RNA for clinical use (Jain et al., 2018).

Fig. 2. The discovery of alternative σ factors.

Efforts to determine if RNA polymerase can be re-programmed by alternative σ subunits focused initially on bacteriophage systems. These efforts led to the discovery of the single-subunit T7 RNAP, and specificity factors that function (more or less) as σ factors from phage T4 and B. subtilis phage SP01. By 1980, several alternative σ factors were identified in B. subtilis (inset adapted from Losick & Pero, 2018). A convergence of biochemical and genetic approaches soon revealed a rich diversity of bacterial σ factors and led to the first efforts to define subgroups within the growing σ⁷⁰ family (inset adapted from Gribskov & Burgess, 1986).

Evidence for the role of alternative σ factors regulating gene expression first emerged from studies of phage systems (Fig. 2). The phage T4 gp33 and gp55 proteins were found to be necessary for middle- and late-transcription, respectively, and both proteins can be purified bound to RNAP from infected cells. However, it was not initially clear if these proteins were, in fact, bona fide σ factors and the requirements for template activation proved to be complex (Geiduschek & Kassavetis, 2010). Biochemical evidence that phage encode regulators that can function in place of the host σ factor to redirect transcription was clearly documented by the studies of Tom Fox and Jan Pero working with Bacillus subtilis phage SP01 (Pero et al., 1975, Fox et al., 1976, Fox & Pero, 1974). In these landmark studies, Pero and her colleagues established that SP01 gp28 is a σ factor for middle genes, and gp33/34 appear to form a two-component σ (or σ-like) for late genes (Tijan & Pero, 1976). Thus, by the mid-1970s the concept of alternative σ factors had gained strong support (Fig. 2), but it remained unclear whether this same regulatory mechanism might be used by bacteria to regulate gene expression.

Key to each of these early studies was the availability of pure phage DNA templates that could be used for analysis of transcription selectivity. The challenge of identifying and characterizing promoters from the much more complex bacterial chromosome, as a needed assay for the purification and study of endogenous alternative σ factors, is emphasized by the title of a recent retrospective from Richard Losick and Jan Pero, “For Want of a Template” (Losick & Pero, 2018).

Biochemical identification of alternative σ factors involved in bacterial gene regulation

The first indications that bacteria may also employ alternative σ factors to regulate gene expression emerged from biochemical studies of RNAP purified from B. subtilis, an organism with a complex developmental program culminating in endospore formation. Within months of the original report of σ factor, Losick and Sonenshein had determined that RNAP from vegetative and sporulating B. subtilis cells differed dramatically in template selectivity; whereas both preparations were highly active with poly-d(AT), only the former was active with Bacillus phage ϕe DNA as template (Losick & Sonenshein, 1969). These observations led to the first suggestion that sporulation might be driven, at least in part, by a σ substitution mechanism. In a brief note added in proof, the authors also noted that E. coli σ factor (as purified by Burgess) could also restore transcription of ϕe DNA when added to the B. subtilis core enzyme. This remarkable result highlights the extent of RNAP conservation in these two bacteria, now thought to have diverged in evolution more than 3 billion years ago. In a series of studies over the next decade, the Losick laboratory documented the existence of sporulation-associated proteins that co-fractionated with RNAP, and a correlated loss of the primary σ factor during sporulation. However, demonstration of σ activity for these new, RNAP-associated factors could not be easily documented “for want of a template” (Losick & Pero, 2018).

The key breakthrough for defining the activity of alternative sporulation σ factors was the ability to identify and clone DNA fragments containing the cognate, developmentally regulated promoters. This led to initial identification of the σ³⁷(σ^B) and σ²⁹(σ^E), although the former turned out to be involved in general stress response regulation rather than dedicated to sporulation (Haldenwang et al., 1981, Haldenwang & Losick, 1979, Haldenwang & Losick, 1980) (Fig. 2). At this stage, the many newly identified candidate σ factors were identified by their molecular weight on SDS-PAGE gels (Fig. 2). Confirmation of σ factor activity was possible in many cases by simply excising the candidate σ factor band from after SDS-PAGE and renaturation of the polypeptide, which could then be added back to core RNAP to restore selectivity (Hager & Burgess, 1980). By 1981, the general theme that σ substitution was a driving force for sporulation had emerged, and it had become clear that σ switching directly altered the preferred sequences within the promoter at both the −35 and −10 elements (Lee & Pero, 1981, Moran et al., 1981). Thus emerged the first hints of the complex scheme of criss-cross regulation between the developing mother cell and forespore (Losick & Pero, 1981).

In parallel with the efforts in the Losick lab to understand the biochemical basis for transcriptional switching during sporulation, work in the Chamberlin lab was also directed at characterization of a newly defined σ factor, σ²⁸(σ^D). This activity was serendipitously discovered during efforts to ascertain the extent of conservation of RNAP selectivity amongst bacteria. Preparations of RNAP from seven divergent bacterial species were all found to recognize the same set of strong promoters on phage T7 DNA (Wiggs et al., 1979). However, the B. subtilis enzyme also gave rise to a transcript not seen with the other preparations (Jaehning et al., 1979), and subsequent fractionation revealed that this activity required a 28 kDa protein, σ²⁸(σ^D) (Wiggs et al., 1981).

Efforts to define the promoter selectivity of σ²⁸, and the biological role of this protein, were to occupy several members of the Chamberlin lab over the subsequent years. This includes five years I invested in finding and sequencing the corresponding gene, sigD, during my PhD studies. The strategy reflected the then ascendant approach of biochemistry; I purified RNA polymerase from kg quantities of bacterial cell paste to obtain mg quantities of RNAP that then allowed the isolation and renaturation of μg quantities of σ²⁸ polypeptide (Helmann et al., 1988b). With partial amino acid sequences in hand, it was possible to design degenerate DNA probes for recovery of the corresponding gene (Helmann et al., 1988a). This same approach was later used by Mark Buttner to clone the gene for Streptomyces coelicolor σ^E, which was a key enabling event in the discovery of the ECF σ factors (Lonetto et al., 1994).

In parallel with my ongoing efforts to clone the sigD gene, the corresponding consensus sequence for recognition by the σ²⁸ holoenzyme was becoming clear (Gilman et al., 1984, Gilman et al., 1981). The culmination of these studies was the realization that the characteristic promoter sequences recognized by σ²⁸ (which, I fear, are forever embossed on my mind) are highly conserved amongst bacteria, and that σ²⁸ factors are often involved in expression of genes related to chemotaxis and motility (Helmann & Chamberlin, 1987). Indeed, we subsequently found that expression of the B. subtilis sigD gene in a non-flagellated and non-motile E. coli fliD mutant, lacking the orthologous σ²⁸ protein, led to the growth of flagellar filaments and restoration of motility (Chen & Helmann, 1992). This result highlights the remarkable conservation of not only the core RNAP machinery, but also an alternative σ factor, which therefore likely evolved very early (likely >3 billion years ago) in the evolutionary diversification of Bacteria. To my knowledge, this type of genetic complementation of σ function between such divergent bacteria has not been seen for any other class of alternative σ factor. It makes sense that both σ structure and promoter selectivity might be highly conserved, since σ²⁸ controls a substantial regulon, and this presumably limits genetic drift at the level of promoter sequences, and therefore also σ structure.

Impact of DNA sequencing and genetics on identification of alternative σ factors

By the early-80s, the newly developed dideoxy-based (Sanger) technique for DNA sequencing (introduced in 1977; Sanger et al., 1977) was widely adopted, and the first complete sequences for genes encoding σ factors were starting to emerge. The first to be identified was E. coli rpoD, in studies from Carol Gross and Richard Burgess (Burton et al., 1981, Gross et al., 1979) (Fig. 2). By 1985, the primary σ factors from E. coli and B. subtilis were recognized as homologous proteins, despite their size differences (Gitt et al., 1985). The primary σ of E. coli migrates in SDS-PAGE at 90 kDa, but the protein mass is 70 kDa (and therefore named σ⁷⁰), whereas the B. subtilis primary σ (previously known as σ⁵⁵) was found to have an actual mass of 43 kDa (and renamed σ⁴³). This confusing situation with σ nomenclature was even worse with the B. subtilis alternative σ factors since many had very similar apparent molecular weights (σ²⁷, σ²⁸, σ²⁹). A solution was forthcoming, however, with the proposal from Losick et al. that σ factors be designed with letters rather than numbers, which had the added advantage that the corresponding gene names would adhere to the established conventions of bacterial gene nomenclature (Losick et al., 1986). Following this suggestion, the primary sigma factors were renamed as σ^A, with alternative σ factor σ³⁷ becoming σ^B (encoded by sigB), σ²⁸ becoming σ^D (encoded by sigD), and so forth. This system proved quite adaptable, and only faced a serious challenge with the expansion of the σ family to include ECF σ factors, which together with exponential increases in the sequence databases, led to the realization that some bacteria have more (and sometimes far more) than 26 σ factors. There were some exceptions to this naming system, for largely historical reasons. In E. coli, for example, the σ factors are encoded by rpo genes (e.g. rpoD for σ^A and rpoH, rpoS, etc. for alternative σ factors).

As Sanger sequencing technology was widely adopted by bacteriologists, the sequences emerged of many genes known to encode factors that activated transcription, but by unknown mechanisms. This led to the emergence of a whole new set of σ factors, which in many cases were first glimpsed as DNA sequences rather than as bands on an SDS-gel of a purified RNAP fraction. One early example was the realization that htpR, an E. coli gene known to encode a high-temperature protein regulator, encoded an alternative σ factor (σ³², and later σ^H or RpoH) (Landick et al., 1984, Grossman et al., 1984). Another E. coli regulator encoded by katF controls catalase expression and encodes the stationary phase regulator, σ^S (renamed rpoS) (Mulvey & Loewen, 1989), and the whiG developmental regulator from Streptomyces coelicolor was found to encode a close homolog of B. subtilis σ²⁸ (Chater et al., 1989). Conversely, a central nitrogen regulator, ntrA, was found to encode a protein that looked nothing like other σ factors, despite having demonstrable σ activity (Hirschman et al., 1985). This σ serves as the prototype for the large family of activator-dependent σ factors of the so-called σ⁵⁴ family (Merrick, 1993, Danson et al., 2019).

One significant outcome of the rapid growth of the σ family tree was the ability to define the key protein sequence features that would come to define σ factors. Comparisons of the DNA sequences of the first available genes encoding σ factors led to a seminal paper from Gribskov and Burgess (1986), which noted that primary σ factors have four conserved regions, numbered 1 through 4 (Gribskov & Burgess, 1986) (Fig. 2). However, there was considerable variability, and some sequences appeared to lack a well conserved region 1 and/or 3, and in the case of the more divergent proteins that function as alternative σ factors for phage, sometimes only a single region was recognizable. This early recognition of the high potential variability within the larger σ family foreshadowed the 1994 description of the extracytoplasmic function (ECF) sub-family as comprising those σ factors with only regions 2 and 4 (Lonetto et al., 1994).

With a combination of clever bacterial genetic approaches, and DNA sequencing to identify the resulting sequence changes, the late 1980s also witnessed the emergence of a detailed model for promoter recognition by bacterial σ factors. In acknowledgement of the unusual σ⁵⁴ family proteins (which have a different overall promoter structure), these studies are generally described as applying to σ⁷⁰ family proteins (the large family of proteins with similarity to σ⁷⁰, including all alternative σ factors except the σ⁵⁴ family). The resulting model posits that σ⁷⁰ conserved region 4 functions as a helix-turn-helix DNA-binding module that recognizes the −35 element (as duplex DNA), whereas σ⁷⁰ region 2 has elements important for both binding to core RNAP and recognition of the −10 element (Gardella et al., 1989, Siegele et al., 1989, Zuber et al., 1989). These interactions correlate with conserved regions in bacterial σ⁷⁰ family proteins, and their consideration led to a more refined nomenclature for the conserved σ regions dividing, for example, region 2 into regions 2.1 through 2.4 (Helmann & Chamberlin, 1988). While these divisions proved a useful shorthand for discussing the function of discrete modules (or sequence units), the true relationship between these conserved elements and protein structural domains was not immediately clear. In this early review, we also introduced the notion that σ factor, and in particular conserved aromatic amino acids in conserved region 2.3, participates directly in DNA strand separation (promoter melting) during the closed- to open-promoter transition (Helmann & Chamberlin, 1988). Specifically, we postulated that conserved aromatic amino acids could provide a protein scaffold for stacking with the bases of the displaced non-template in the transcription bubble, a prediction largely supported and refined by a number of subsequent genetic and structural studies (Feklistov et al., 2014).

The comparative sequence analysis of bacterial σ factors was further extended using a substantially larger dataset of sequences in 1992 (Lonetto et al., 1992). This study revealed, for example, that σ region 2.3, involved in DNA-melting, was much less well conserved in alternative σ factors. Much later, it was appreciated that for some of these alternative σ factors a lack of strong conservation in region 2.3 is correlated with weaker promoter melting activity, and consequently a stronger reliance on the initial promoter recognition binding event. This also correlates with an overall higher conservation of cognate promoter consensus sequences (Koo et al., 2009).

Expansion of the σ family to include the ECF σ factors

The studies above, encompassing the first ~25 years of σ history, form the backdrop for the discovery of the extracytoplasmic (ECF) σ factors. Many of the genes that encode ECF σ factors were first identified for their regulatory role, and even when their sequences were determined, many were not immediately recognized as σ factors despite, in some cases, apparent similarity to the larger σ⁷⁰ family. In some ways, this is surprising since these genes encode proteins more closely related to conventional σ factors than the phage σ factors included in the original (1986) Gribskov and Burgess alignments (Gribskov & Burgess, 1986). The key convergence that led to the integration of these sequences as part of the σ family tree, and as founding members of the ECF sub-family, came from the realization that Streptomyces coelicolor σ^E (as characterized by traditional protein purification and reconstitution methods; Buttner et al., 1988) was encoded by a gene clearly related to several other “transcription factors” (Myxococcus xanthus CarQ, Pseudomonas aeruginosa AlgU, Pseudomonas syringae HrpL, E. coli σ^E and FecI, Alcaligenes eutrophus CnrH, and Bacillus subtilis SigX) not yet shown to function as σ factors (Lonetto et al., 1994) (Fig. 3). It was quickly appreciated that some bacteria contain multiple ECF σ factors, a phenomenon first explored systematically in the B. subtilis model system (Helmann, 2002).

Fig 3. ECF σ factors and the age of genomics.

With increasing availability of DNA sequence information, the genetic diversity of σ factors became appreciated. The ECF subgroup was identified in 1994 (inset from Lonetto et al., 1994), and 1995 marked the beginning of the era of bacterial genomics. By 2009, phylogenomics-based approaches revealed that ECF σ factors were a highly diverse and often numerically dominant subset of alternative σ factors that play major roles, in many cases yet to be determined, in bacterial gene regulation (inset from Staron et al., 2009).

In studies in my group over the last two decades we have generated mutant strains lacking each of the seven ECF σ factors encoded by B. subtilis. These ECF σ factors are both individually and collectively dispensable (Luo et al., 2010), but the resulting mutant strains have a greatly increased sensitivity to specific stresses (Mascher et al., 2007). In our earliest studies, we took advantage of the fact that many ECF σ factor genes are autoregulatory to identify and characterize cognate promoters (Huang et al., 1999, Huang & Helmann, 1998). Detailed promoter mutagenesis studies yielded insights into the conserved bases that defined the −35 and −10 elements (Huang et al., 1998), which then allowed sequence-based identification of candidate promoters from the genome sequence (Huang et al., 1999, Huang & Helmann, 1998). Remarkably, some ECF σ promoter consensus sequences differed by as little as a single base in the −10 element, and substitutions at this position could transfer a promoter from one regulon to another (Qiu & Helmann, 2001). The high conservation of promoters recognized by ECF σ factors also facilitated computational approaches to predict regulons in several other systems, including the σ^E regulon in E. coli and its close relatives (Rhodius et al., 2006) and in Rhodobacter sphaeroides (Dufour et al., 2008).

By carefully delineating the regulatory scope of each B. subtilis ECF σ factor (the regulon), often using a combination of complementary approaches (Cao et al., 2002a, Eiamphungporn & Helmann, 2008), we were able to derive insights into the likely role of each σ and predict relevant inducing conditions (Cao et al., 2002b, Guariglia-Oropeza & Helmann, 2011). Our results, together with those from others, indicates that σ^X controls cell surface modifications that impart resistance to antimicrobial peptides (Cao & Helmann, 2004) and has been coopted by phage SPβ to regulate sublancin production (Luo & Helmann, 2009), σ^W protects cells against a variety of antibacterial agents including bacteriocins and cell membrane disruptive agents (Cao et al., 2001, Butcher & Helmann, 2006, Luo et al., 2010, Kingston et al., 2011), σ^M is critical in regulating cell envelope synthesis and in resistance to agents that interfere with peptidoglycan synthesis (Luo & Helmann, 2012, Meeske et al., 2015), and σ^V is specifically implicated in defense against lysozyme (Guariglia-Oropeza & Helmann, 2011, Ho et al., 2011). Although we now have a reasonably complete understanding of the regulatory scope for these four ECF σ factors (Helmann, 2016), the roles of the other three (which appear to have much more proscribed regulons; Asai et al., 2003) are less well understood.

As the large family of σ factors has rapidly expanded, so too has the sophistication of our understanding of all aspects of RNA polymerase structure and function. Structural studies, many using X-ray crystallography, have gradually emerged for σ factors and their domains, RNA polymerase holoenzyme, and ultimately holoenzyme:promoter complexes (Feklistov et al., 2014). These efforts have allowed insights into the processes of both promoter recognition and DNA strand separation for σ⁷⁰ holoenzyme (Zhang et al., 2012), and recently, for two holoenymes containing mycobacterial ECF σ factors (Li et al., 2019, Lin et al., 2019). Three decades ago, we predicted that conserved aromatic amino acids σ region 2.3 (a TYATWW motif in σ^A) would be involved in DNA strand separation (Helmann & Chamberlin, 1988), and subsequently demonstrated that mutations of the corresponding aromatic amino acids formed holoenzymes specifically defective in the DNA-melting step accompanying open complex formation (Juang & Helmann, 1994). Gratifyingly, these interactions are now understood to be central for DNA-melting and the accompanying sequence-selective recognition of the non-template strand during the transition from the initial closed to the open complex (Murakami & Darst, 2003, Bae et al., 2015). The situation with ECF σ factors is somewhat more complicated, and also involves an adjacent loop region (Campagne et al., 2014), but generally follows a process reminiscent of that for σ^A holoenzyme (Li et al., 2019, Lin et al., 2019). This is just one example of how structural studies have helped inform our understanding of RNA polymerase, and its interactions with its template, substrates, accessary subunits and regulators, and transcription factors (Feklistov et al., 2014).

The advent of high throughput DNA sequencing methods has also drastically altered our perspective on σ factors and their diversity. The original dideoxy-method of Sanger held sway for over 25 years as the dominant technology, although the introduction of fluorescent chain-terminators and automated, capillary-based sequencing machines has spared individual investigators from the tedium of pouring and running sequencing gels. The subsequent development of new, next-generation sequencing technologies has vastly increased our access to the universe of genetic information (Loman & Pallen, 2015, Shendure et al., 2017). This immense treasure-trove of information has enabled scientists to answer questions in ways that could scarcely have been dreamt of in prior decades, a time when σ factors were visualized by the small cadre of researchers in this area as bands on an SDS-PAGE gel that correlated with distinct transcriptional activities.

Bacterial genomics has led to the recognition of the ECF σ family as a very large and diverse group of σ factors. In a seminal analysis of the family, the group of Thorsten Mascher determined that these proteins form a “third pillar”, together with one- and two-component regulator systems, of bacterial gene regulation (Staron et al., 2009). This paper also led to the first iteration of the current ECF σ classification system based on phylogenomics (Fig. 3). Analysis of nearly 3000 ECF σ sequences identified 43 distinct groups (designated ECF01 through ECF43) that encompassed 2/3 of the available sequences, with the remaining forming many smaller clusters (Staron et al., 2009). This classification scheme has since grown to ~100 distinct groups of ECF σ factors, as defined based on their protein sequences, genomic context, and target promoter sequences (Pinto & Mascher, 2016).

Despite this large expansion of the σ factor family tree, it remains likely that there are branches yet to be discovered (Mascher, 2013). Moreover, the diversity of gene regulation mechanisms in phage never ceases to amaze, and there are likely divergent regulators in these systems that function as σ factors, through a σ-like mechanism, or by modifying host σ activity in interesting ways (Tabib-Salazar et al., 2019). We also described an unusual “split” σ in B. subtilis, in which two polypeptides (one with σ region 2 like functions, and the other corresponding more closely to region 4) cooperate in promoter recognition and activation (MacLellan et al., 2009). This is reminiscent of the earliest efforts to define alternative σ factors in which it was noted that late transcription in B. subtilis phage SP01 required the cooperation of two polypeptides (Tijan & Pero, 1976). With the current explosion of DNA sequence information, it remains possible that there are new branches of the σ⁷⁰ family tree, or even new types of σ like regulators entirely.

This brief survey of the unfolding story of nucleic acid polymerases over the last ~60 years, and of bacterial σ factors since their discovery 50 years ago, provides one example of how changing times and new technologies have dictated and constrained the dominant approaches during each phase of discovery. For the first half of this period, nucleic acid enzymology generally, and the early studies of RNA polymerase specifically, were driven by classical biochemical approaches developed by Arthur Kornberg and his contemporaries, including Paul Berg and his student Mike Chamberlin. Key early contributions from Jan Pero, Richard Losick, Richard Burgess, and their scientific progeny set the stage for the next major leaps forward, enabled and facilitated by the new technology of dideoxy-based (Sanger) sequencing. The merging of biochemical studies with genetic and molecular biology approaches was foundational for this next phase, including key contributions from Carol Gross, Mark Buttner, Tim Donohue, and many others. The newest phase can again be traced to a change in enabling technologies: the explosion of sequence information from high-throughput sequencing has led to insights into the vast diversity of ECF σ factors, and motivated efforts to harness this diversity for applications in synthetic biology (Pinto et al., 2019, Pinto et al., 2018, Rhodius et al., 2013).