*Review* **Light-Triggered Carotenogenesis in** *Myxococcus xanthus***: New Paradigms in Photosensory Signaling, Transduction and Gene Regulation**

**S. Padmanabhan 1,\*, Antonio J. Monera-Girona 2, Ricardo Pérez-Castaño 2, Eva Bastida-Martínez 2, Elena Pajares-Martínez 2, Diego Bernal-Bernal 1, María Luisa Galbis-Martínez 2, María Carmen Polanco 2, Antonio A. Iniesta 2, Marta Fontes <sup>2</sup> and Montserrat Elías-Arnanz 2,\***


**Citation:** Padmanabhan, S.; Monera-Girona, A.J.; Pérez-Castaño, R.; Bastida-Martínez, E.; Pajares-Martínez, E.; Bernal-Bernal, D.; Galbis-Martínez, M.L.; Polanco, M.C.; Iniesta, A.A.; Fontes, M.; et al. Light-Triggered Carotenogenesis in *Myxococcus xanthus*: New Paradigms in Photosensory Signaling, Transduction and Gene Regulation. *Microorganisms* **2021**, *9*, 1067. https://doi.org/10.3390/ microorganisms9051067

Academic Editor: David Whitworth

Received: 29 April 2021 Accepted: 12 May 2021 Published: 15 May 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

**Abstract:** Myxobacteria are Gram-negative δ-proteobacteria found predominantly in terrestrial habitats and often brightly colored due to the biosynthesis of carotenoids. Carotenoids are lipophilic isoprenoid pigments that protect cells from damage and death by quenching highly reactive and toxic oxidative species, like singlet oxygen, generated upon growth under light. The model myxobacterium *Myxococcus xanthus* turns from yellow in the dark to red upon exposure to light because of the photoinduction of carotenoid biosynthesis. How light is sensed and transduced to bring about regulated carotenogenesis in order to combat photooxidative stress has been extensively investigated in *M. xanthus* using genetic, biochemical and high-resolution structural methods. These studies have unearthed new paradigms in bacterial light sensing, signal transduction and gene regulation, and have led to the discovery of prototypical members of widely distributed protein families with novel functions. Major advances have been made over the last decade in elucidating the molecular mechanisms underlying the light-dependent signaling and regulation of the transcriptional response leading to carotenogenesis in *M. xanthus*. This review aims to provide an up-to-date overview of these findings and their significance.

**Keywords:** photoreceptor; photosensitizer; photoregulation; singlet oxygen; plasmalogens; CarF; vitamin B12; CarH; ECF-sigma; CarD-CdnL

#### **1. Introduction**

Light is an important and ubiquitous signal in terrestrial and aquatic ecosystems, and the ability to sense, respond and adapt to light is crucial for most living organisms, including bacteria. Photosynthetic bacteria capture and convert light, an essential energy source, to chemical energy for cellular utilization, but light is also important for several other cellular processes in both phototrophic and non-phototrophic bacteria [1–5]. Thus, light is linked to many bacterial responses such as phototaxis, development, virulence, circadian rhythms and UV-induced DNA damage repair [4–7]. However, light can be harmful and cause cell damage and death. This stems from excitation of photosensitizing biomolecules, such as porphyrins, chlorophyll or flavins, to generate highly reactive oxygen species (ROS) like singlet oxygen (1O2), superoxides, peroxides and hydroxyl radicals that can destroy cellular DNA, protein and lipid components [4,8–11]. Consequently, bacteria have evolved ingenious mechanisms and machineries to mount a protective response to counter photooxidative stress.

A commonly used defense mechanism against photooxidative damage is through the biosynthesis of carotenoids, which quench and dissipate as heat the excess energy of 1O2 and other ROS produced upon illumination [4,8,9,11–13]. Carotenoids constitute a major class of lipophilic isoprenoid derivatives that are characterized by an extended, typically all-*trans*, conjugated polyene chain (usually C40 and some C50, C45 and C30 terpenes) with acyclic, monocyclic or bicyclic ends. Their oxygenated (hydroxy, aldehyde, keto, carboxyl, methoxy, epoxy, oxy and glycosidic) derivatives are called xanthophylls. Most carotenoids are richly colored (light yellow to deep red), since they absorb blue-violet light (400–500 nm range) owing to their extended conjugated double bonds that also determine the molecular conformation and reactivity [13]. Carotenoids also fulfill biological roles other than in photoprotection, such as in photosynthetic light harvesting, signaling and as precursors of photosensory molecules and hormones [13,14].

Carotenoid biosynthesis de novo occurs in all photosynthetic organisms (plants, algae or bacteria) and in many non-photosynthetic fungi, archaea and bacteria, whereas animals, save some strikingly few exceptions, do not synthesize carotenoids but obtain them exogenously [13,14]. Given that carotenoids are in the frontline of the defense against photooxidative stress, light and oxygen-related species like 1O2 are among the principal environmental factors involved in signaling and triggering carotenoid biosynthesis. This has been amply demonstrated in several studies from plants [15,16] and fungi [17,18] to bacteria [4,8,9,12]. Light-induced carotenogenesis and its regulation in the Gram-negative soil bacterium *M. xanthus* is undoubtedly one of the best studied and characterized among bacteria. We last reviewed this topic over a decade ago when many questions remained open [4,12]. Since then, considerable progress has been achieved largely from work in our group on the mechanistic, structural and photochemical aspects of light-regulated carotenogenesis in *M. xanthus*. Our work has uncovered new and large protein families, such as an entirely new class of photoreceptors with their novel mode of action that we specifically reviewed elsewhere [19–22]. It has also revealed the participation of "eukaryotic-like" proteins, including one found in *M. xanthus* and related myxobacteria, but absent in the vast majority of other bacteria, that turned out to be a long-sought human enzyme conserved across metazoa [23]. Our present review aims to provide a timely update of these findings, from signal reception and transduction to the transcriptional regulation underlying the photooxidative stress response and carotenoid biosynthesis in *M. xanthus*, and to discuss their mechanistic and evolutionary significance.

#### **2. Biosynthesis of Carotenoids**

Carotenoid biosynthesis occurs via a well-established and largely conserved pathway involving a number of genes and their products [13,14]. The pathway is considered to begin with the condensation of the universal five-carbon (C5) isoprenoid precursors isopentenyl diphosphate (IPP) and dimethylallyl diphosphate (DMAPP), themselves products of either the mevalonate (MVA) pathway (see Figure 1) or the non-mevalonate 2C-methyl-Derythritol-4-phosphate (MEP) pathway [13,14,24]. Most bacteria and plastids are equipped with the MEP pathway, the MVA pathway is prevalent in animals, archaea, fungi and some bacteria including *M. xanthus* and the majority of myxobacteria, while plants and some select bacterial species use both pathways [13,14,24]. Condensation of IPP and DMAPP, the first committed and usually rate-controlling step in the core carotenoid biosynthesis pathway, produces geranylgeranyl diphosphate, two molecules of which then condense to generate the colorless C40 isoprenoid phytoene. A series of phytoene isomerization and desaturation steps generates the red carotenoid lycopene, from which carotenes and xanthophylls are produced in further desaturation, isomerization and hydroxylation reactions. Carotenoid biosynthesis and its regulation at levels from transcription, which is among the earliest and most crucial steps, to post-translation, degradation and feedback have been studied in many organisms [13,14]. Here, we discuss our current understanding of the *M. xanthus* carotenoid biosynthesis pathway, the structural and regulatory genes involved and how their transcription is induced and regulated.

**Figure 1.** Carotenoid biosynthesis pathway and genes in *M. xanthus*. (**a**) *M. xanthus* colony color in the dark and in the light. Wild-type strains are yellow in the dark and red when exposed to blue light. (**b**) Structural genes for carotenogenesis characterized in *M. xanthus*. The *carB* locus encodes nine structural genes for carotenoid synthesis and two transcription regulatory factors, CarA and CarH, expressed from the primary σA-dependent PB promoter. The isolated structural *crtIb* gene is expressed from a promoter that depends on the ECF-σ factor CarQ (see text). The four-digit number above or below the corresponding gene indicates the original genome locus tag (MXAN\_xxxx) of each of these genes. (**c**) Carotenoid synthesis pathway derived from the mevalonate (MVA) pathway in *M. xanthus*, with enzymes and products indicated.

#### **3. A Brief History of Early Findings in** *M. xanthus* **Light-Induced Carotenogenesis**

*M. xanthus* cells are yellow (Figure 1a) in the dark due to noncarotenoid, light-sensitive pigments that were identified and named DKxanthenes just fifteen years ago [25]. Four decades earlier, Burchard, Dworkin and coworkers reported that *M. xanthus* cells, when grown in the light, suffered photolysis or developed an orange/red color attributed to carotenoids and resisted photolysis, with the extent of illumination and growth phase determining the accumulation of carotenoids [26]. The action spectrum for photoinduction of carotenoids mirrored those for photolysis and for the absorption spectrum of protoporphyrin IX (PPIX), a hydrophobic cyclic tetrapyrrole and immediate precursor of heme in its biosynthesis, which accumulates in the *M. xanthus* cell membrane especially during stationary phase [26,27]. Photoinduction of carotenogenesis was maximal under blue light (405–410 nm), with lower maxima in the green light region (510–580 nm). Blue light excites the photosensitizer PPIX to 3PPIX, a very reactive high-energy triplet state that can directly cause cell damage or transfer its energy to other molecules [8,9]. Energy transfer from 3PPIX to molecular oxygen generates 1O2, an extremely reactive ROS that is relatively longlived and diffusible in membrane environments [28]. Light-generated 1O2 was therefore proposed as the signal for carotenoid biosynthesis in *M. xanthus* [29], and later validated experimentally [30].

Two decades after these early findings, isolation and genetic analysis of *M. xanthus* spontaneous mutants, or ones generated by chemical, UV or Tn*5*-*lac* insertions, helped identify key genetic loci involved in light-induced carotenogenesis and established it as a transcriptional response [31–34]. The distinctive color change from yellow in the dark to red in the light (Figure 1a) due to light-induced carotenogenesis (the wild-type Car+ phenotype) provided a valuable visual tool for facile genetic analysis. This helped identify mutants that synthesize carotenoids constitutively (CarC) and are always orange/red, and these mutations mapped to two loci, *carA* and *carR*, which were inferred to encode negative regulators. On the other hand, mutants that never turn red in the light (Car−) were also identified and these mutations mapped to two loci encoding carotenogenic enzymes, or to various loci encoding putative positive regulators [32–37]. In these and subsequent studies, the loci were further mapped, epistatic relationships between them established, and the stage set for cloning, sequencing, gene expression assays and chemical analysis of carotenoids. This uncovered most of the structural and regulatory genes, and more recent biochemical, biophysical, genome-level and high-resolution structural analyses have provided profound insights into the molecular mechanisms underlying this light response.

#### **4. Structural Genes Encoding** *M. xanthus* **Light-Induced Carotenoid Biosynthetic Enzymes**

Genetic analysis, cloning and sequencing of the loci involved revealed that structural genes encoding the carotenoid synthesis enzymes (gene names usually prefixed *crt*) were located at the unlinked *carB* and *carC* loci [31–34,38–40]. The *carB* locus groups nine structural genes and two regulatory genes organized as *crtE-crtIa-crtB-crtD-crtC-orf6-crtYccrtYd-orf9-carA-carH*, and the *carC* locus corresponds to a single gene, *crtIb* (Figure 1b). These annotations were based on analysis of sequence and of carotenoids accumulated in different mutants, as well as on heterologous expression in *E. coli* [33,38–42]. Genes *orf6* and *orf9* at the *carB* locus may also be structural ones whose functions remain to be established experimentally, while *carA* and *carH* encode transcription factors that regulate expression of *carB* genes (see below). The proposed carotenoid biosynthesis pathway for *M. xanthus* and the enzyme(s) involved in each step (Figure 1c) leads to synthesis of the final product, myxobacton ester, a monocyclic carotenoid with a keto group in the ring at one end of the molecule and a glycosyl group esterified to a straight-chain fatty acid at the other [43].

CrtE (geranylgeranyl pyrophosphate synthase), encoded by the first gene at the *carB* locus, catalyzes conversion of farnesyl diphosphate to geranylgeranyl diphosphate, two molecules of which condense to phytoene through the action of CrtB (phytoene synthase). The colorless phytoene is isomerized and transformed in four successive dehydrogenation steps to the red acyclic lycopene by the synergistic action of the CrtIa and CrtIb phytoene dehydrogenases [38,39,41]. Lycopene is cyclized at one end to monocyclic γ-carotene by the concerted action of CrtYc and CrtYd, members of the heterodimeric lycopene monocyclase family encoded by adjacent genes at *carB* [42]. The γ-carotene is subsequently hydroxylated by hydroxyneurosporene synthase (CrtC), desaturated by hydroxyneurosporene dehydrogenase (CrtD), and appended with a sugar moiety possibly by the action of the *orf6*

gene product, a predicted glycosyltransferase. Finally, a putative acyltransferase encoded by *orf9* may act in myxobacton esterification [12]. Regulation of the carotenoid biosynthesis pathway can occur at the levels of transcription, post-transcription, modulation of enzyme activity through crosstalk and cooperation between them (such as the synergistic action of CrtIa and CrtIb mentioned above) and feedback regulation by the final carotenoid product and/or precursors. Moreover, how pathways intertwined with that for carotenogenesis, such as the MVA or the PPIX/heme biosynthesis pathways, are regulated can be important. Clearly, regulation at the transcription level is among the earliest and most crucial, and light-dependent regulation of transcription of *crtIb* and of the genes at *carB* has been intensely studied in *M. xanthus*.

#### **5. Two Modes of Light Sensing and Signaling in** *M. xanthus* **Carotenogenesis**

Since light triggers expression of *carB* and *crtIb*, understanding how light is sensed and converted to a cellular signal to mount the transcriptional response in *M. xanthus* is critical. In most living organisms including bacteria, the crucial task of sensing and transducing the light signal depends on photoreceptors, which are specialized proteins equipped with covalently or noncovalently bound light-sensing cofactors called chromophores. Photoreceptors have been classified into ten families thus far based on the specific chromophore and the protein photosensory domain [5,21,44–47]. One or more of these proteins occur in various bacteria, some are more widely distributed than others, and some occur even in species with minimal genomes or lifestyles that might suggest an absence of photoreceptors. Yet, surprisingly, given its well-established light response and a genome that is among the largest and most complex across bacteria, *M. xanthus* appeared to lack known photoreceptors. Consequently, blue light sensing through PPIX and the corresponding, rather convoluted, signaling pathway (discussed in Section 7) were considered as the sole mechanism to trigger carotenogenesis in *M. xanthus*. This changed about ten years ago with the discovery of the first member of an entirely new family of photoreceptors, which established a second light sensing and signaling mechanism that is simpler and more direct [20,21].

#### **6. Direct Light Sensing, Signal Transduction and Gene Regulation by the B12-Based CarH Photoreceptor**

Discovery of the more direct light-sensing/signaling pathway and of a new photoreceptor family emerged from studies of the two adjacent and most downstream genes of the *carB* cluster, *carA* and *carH,* whose expression is significantly enhanced in the light [20,21,38,40]. Whereas mutations at *carA* yielded a Car<sup>C</sup> phenotype, linking it to a negative regulator [32,34], a *carH* deletion had no apparent effect [48], even though the corresponding gene products of comparable sizes (CarA: 288 residues; CarH: 299 residues) share ~48% similarity (~35% sequence identity) and a similar two-domain architecture [38,48]. In both proteins, the ~70-residue N-terminal region resembles the DNA-binding domain (DBD) of MerR family proteins [38], which are widespread transcription factors in bacteria that repress or activate gene expression in response to diverse environmental stimuli such as oxidative stress, heavy metals or antibiotics [49]. MerR proteins bind as dimers via their winged-helix DNA binding domains to specific (pseudo)palindromic sites located within or overlapping their target primary σA-dependent promoters, and binding of a ligand (metal/drug) to a C-terminal module or oxidation of a redox center in it, enables these proteins to modulate transcription [49–52]. Notably, the ~200-residue C-terminal domain in CarA and CarH resembles a domain that binds to methylcobalamin (MeCbl) [48], one of the two biological forms of vitamin B12, in the methionine biosynthesis enzyme MetH, a methionine synthase. The MetH B12-binding domain (B12-BD) houses a signature motif, E/DxHx2Gx41SxTx22-27GG, whose His supplies the lower axial ligand in the so-called base-off/His-on binding to B12 [53]. Prior to CarA and CarH, such B12-BDs were reported only in enzymes using B12 as a cofactor [54,55]. The combination of a B12-BD and a DBD in CarA and CarH was therefore unprecedented and hinted at a pair of unusual transcription factor paralogs. Identifying a role for B12 and its mode of action, however, turned out to be less than straightforward.

#### *6.1. CarH and Vitamin B12 Regulate Light-Induced Expression of Carotenoid Genes*

Mapping the transcription start site at the *carB* locus identified a light-inducible primary σA-dependent promoter, PB, with a consensus TTGACA –35 element and a less conserved TACCTC –10 element [38], which was recognized by σA-bound RNA polymerase (RNAP) in vitro [56]. CarA was found to dimerize via its B12-BD [20,57] and use its Nterminal DBD (which indeed structurally resembles MerR DBDs [58]) to bind cooperatively, as two dimers, to a large ~55-bp DNA segment at the PB promoter region (from positions –70 to –19 relative to the transcription start site) [56,57,59]. Since the operator overlaps with the –35 PB promoter element, CarA binding can block promoter access to RNAP-σ<sup>A</sup> and repress transcription [56]. Surprisingly, even though CarA could bind B12, consistent with the presence of a canonical B12-binding motif at its C-terminal domain, it neither required B12 for operator binding in vitro nor did mutating key residues in its B12-binding motif impair PB repression in the dark in vivo [60]. Key to unmasking the role of CarH was the finding that the Car<sup>C</sup> phenotype caused by deleting *carA* could be reverted to wild-type behavior upon addition of exogenous vitamin B12 to the growth medium (*M. xanthus* takes up and assimilates B12 but cannot synthesize it de novo) [60]. CarH was shown to orchestrate this B12-dependent repression of PB in vivo and its relief in the light, and this activity of CarH required an intact CarA operator [60]. Thus, CarA and CarH both target the same operator at PB to control light-induced expression of all but one of the carotenogenic genes in *M. xanthus*, but only CarH absolutely required B12 for activity. These findings not only established a functional link between B12 and CarH but also revealed a novel facet of this vitamin: its use in a cellular light response.

#### *6.2. Molecular Architecture and Mode of Action of the B12-Based CarH Photoreceptor*

Answers to what specific B12 form was required by CarH and its molecular mechanism of action, as well as why and how CarH differs from its paralog CarA, began to emerge with a seminal study ten years ago [20]. CarA and CarH remained the first and only known transcription factors with a B12-binding motif until homologs of unknown function were revealed in bacterial genomes covering a vast taxonomical range beyond myxobacteria [20,21]. This allowed comparative studies and better molecular understanding of these proteins. Whereas CarH has thus far resisted purification in the native form, two of its homologs from bacteria unrelated to *M. xanthus* have been purifiable in a native soluble form and could therefore be well-characterized in vitro. Both homologs turned out to be B12-dependent like CarH. Studies of the homolog in the Gram-negative *Thermus thermophilus*, CarHTt, yielded valuable biochemical [20], structural [19] and photochemical insights [61–63] that were further extended with CarHBm, the homolog in the Gram-positive *Bacillus megaterium* [64,65]. These findings, reviewed elsewhere [21,22], are briefly highlighted here.

The specific B12 form required in CarH-mediated regulation of light-induced carotenogenesis in *M. xanthus* was established as 5'-deoxyadenosylcobalamin (AdoCbl) or coenzyme B12 (Figure 2a), which binds to the CarH C-terminal domain and directs its oligomerization and function [20]. AdoCbl is a complex organometallic molecule with a central cobalt, generally Co3+/Co(III), coordinated to: (a) four equatorial pyrrolic nitrogens of the corrin ring; (b) a lower axial nitrogen from the 5,6-dimethylbenzimidazole (DMB) moiety linked to the corrin ring (so-called base-on or DMB-on conformation), or histidine from the B12-binding motif in a protein (base-off/His-on binding, mentioned earlier); (c) an upper axial 5 -deoxyadenosyl (Ado) group; this upper ligand is methyl (Me) in MeCbl or cyano (CN) in vitamin B12, a nonbiological form. The Co-C bond to an alkyl carbon in AdoCbl or MeCbl confers some unique and useful chemical properties. Its enzyme-catalyzed cleavage, which enables the use of AdoCbl in mutases, dehydratases, deaminases and ribonucleotide reductases and of MeCbl in methyltransferases, has been extensively studied and reviewed elsewhere [54,55]. Cleavage of the Co-C bond, by near-UV and visible light of wavelengths <530 nm, also underlies the use of AdoCbl as a chromophore for light sensing and response by CarH proteins (Figure 2b,c), which now represent a separate, large and widespread photoreceptor family among the ten currently known [19–22,61–63].

**Figure 2.** Light sensing and gene regulation by the B12-based CarH photoreceptor. (**a**) Chemical structure of AdoCbl, the light-sensing chromophore of the CarH photoreceptor, with the upper axial 5 -deoxyadenosyl group in cyan and the rest of the molecule in magenta as depicted in the structures below. (**b**) CarH-mediated regulation at PB. In the dark, AdoCbl-bound CarH binds to its operator at PB to block access to RNAP-σ<sup>A</sup> and repress transcription; and light (UV, blue or green) inactivates CarH to prevent its binding to operator, allowing PB access to RNAP-σ<sup>A</sup> and transcription initiation. (**c**) Molecular mechanism of CarH-mediated regulation at PB. AdoCbl (filled red diamonds) bind to apo form monomers (CarHTt) or molten globule tetramers (CarHBm) to produce active, properly folded, compact tetramers that bind in the dark to an operator overlapping with a σA-dependent promoter (shown for –35 region but can be –10 or both) and thereby block transcription. UV, blue or green light photolyzes CarH-bound AdoCbl and disrupts DNA-bound tetramers to monomers (CarHTt) or dimers (CarHBm) that retain photolyzed AdoCbl (open red diamonds), leading to loss of operator binding and transcription. Upon photolysis, the upper ligand of AdoCbl is released as 4 -5 -anhydroadenosine (open triangles). Structures for AdoCbl-bound CarHTt tetramer, free and DNA-bound, and for the light-exposed monomer are shown below. The protomer structure (left) with the DBD in cyan (recognition helix and wing, dark blue) and the AdoCbl-BD, with its four-helix bundle subdomain in golden, Rossmann fold subdomain in green, AdoCbl colored as in (**a**). PDB accession codes; 5C8D (tetramer in the dark), 5C8E (DNA-bound tetramer in the dark), 5C8F (light-exposed monomer). Below are close-ups of the Trp, Glu and His of the Wx9EH motif capping the Ado group of AdoCbl in CarHTt, with the lower axial His (green) in the dark state (left) and light-exposed (right) state. In the latter, a His adjacent to the Trp in the Wx9EH motif becomes the upper axial ligand in a bis-His linkage. The 4 -5 -anhydroadenosine product of AdoCbl-CarHTt photolysis is also shown (far right bottom).

Studies of CarHTt and CarHBm revealed that light-dependent regulation of transcription relies on modulation of their oligomeric state by AdoCbl and light (Figure 2c) [19–22,64–66]. AdoCbl-free apoCarHTt is a monomer and apoCarHBm is a loosely folded molten globule tetramer, and both bind poorly to operator DNA. Both proteins form AdoCbl-bound tetramers in the dark that bind tightly to a large operator, which overlaps with the target gene promoter, to thwart access to RNAP-σ<sup>A</sup> and block transcription; in the light, cleavage of the Co-C bond frees the upper axial Ado group and provokes tetramer disassembly to photolyzed CarHTt monomers or CarHBm dimers that detach from the operator to allow RNAP-σ<sup>A</sup> binding and transcription initiation [19,20,64]. Cleavage of the AdoCbl chromophore with release of the Ado group is irreversible, in contrast to the usually reversible light-induced molecular changes observed with other photoreceptor chromophores [44,45,47], suggesting that there may be pathways to recover and reuse the chromophore that remain to be identified. Available data suggest that, like CarHTt, CarH is a monomer in the light-exposed AdoCbl-bound and apo forms, and that the dark AdoCbl-bound form is oligomeric but its stoichiometry remains to be defined [20].

Crystal structures of the AdoCbl-CarHTt tetramer, free or DNA-bound, and of the light-exposed AdoCbl-CarHTt monomer provided detailed molecular snapshots of CarH architecture and its light-dependent mechanism of action [19]. It confirmed the twodomain CarH modular architecture, with a MerR/CarA-like winged-helix N-terminal DBD connected by a flexible, disordered linker to a C-terminal AdoCbl-binding domain (hereafter AdoCbl-BD), in which AdoCbl is sandwiched between a four-helix bundle and a Rossmann fold subdomain (Figure 2c). The AdoCbl-BD is structurally similar to the MetH MeCbl-binding domain but has a critical Wx9EH motif in the four-helix bundle that caps the upper axial Ado, which is absent in MetH (Figure 2c). In addition to the classic ExHx2Gx41SxV/Tx22-27GG B12-binding motif, the Wx9EH motif is absolutely conserved in all CarH homologs studied thus far and its critical role in AdoCbl-binding and function has been experimentally demonstrated [19,64]. Thus, absence of the motif in CarA can largely account for its B12-independent activity. Indeed, the presence of both signature motifs defines CarH homologs, and several hundreds of these now assigned from genome data are broadly distributed across diverse bacterial taxa [21,22]. The dark state AdoCbl-CarHTt tetramer is a dimer of two dimers, each of which is itself assembled by head-to-tail packing of two monomers via their AdoCbl-BD, with Trp of the Wx9EH motif playing a crucial role. Since tetramer formation is very favorable, dimers are detected only by disruption of the dimer-dimer interface, such as by mutation [19,20,66]. In this unusual AdoCbl-CarHTt tetramer assembly, the DBDs of neighbouring monomers point away from each other on the tetramer surface, which results in an unexpected DNA binding mode, wherein one DBD contacts a 11-bp direct repeat (DR) with a consensus nAnn**T**nn**ACA**n sequence (*n* = any base). Hence, it differs from the typical (pseudo)palindromic DNA sites of MerR proteins, yet it conserves most of the DNA contacts. Whereas three such tandem 11-bp DRs comprise the CarHTt operator, four of these constitute the CarHBm operator [19,64] and, likely, the ~55-bp *M. xanthus* CarH operator [20], suggesting a notable DNA-binding plasticity. Comparing the tetramer structure with that determined for the photolyzed CarHTt monomer yielded molecular insights into light-induced tetramer collapse and loss of DNA binding. The light-exposed form revealed bound photolyzed AdoCbl (without the upper axial Ado) and a large shift (>8 Å) of the four-helix bundle relative to the Rossmann fold (Figure 2c), which disrupts the head-to-tail dimer interface, and thereby the tetramer, leading to loss of DNA binding.

The photochemistry of CarHTt-bound AdoCbl examined by analyzing photolysis products [61], by ultrafast spectroscopy [62,63] and by theoretical calculations [67] suggested it may differ significantly from that established for free or enzyme-bound AdoCbl. Photolytic cleavage of free AdoCbl, often a model for that in AdoCbl-dependent enzymes, is homolytic and generates reactive cob(II)alamin and Ado• radical species that rapidly react to yield specific products depending on the presence or otherwise of molecular oxygen [21,61,68]. In AdoCbl-dependent enzymes, which also rely on homolytic Co-C bond

cleavage, the cob(II)alamin and Ado• radical species are generated in carefully controlled protein environments to ensure the difficult radical-based enzyme action and cofactor recovery, and to simultaneously limit enzyme damage and unwanted side reactions [69]. Since CarH controls a cell response (carotenogenesis) precisely to combat reactive ROS like 1O2, its use of an AdoCbl chromophore with an underlying irreversible photolytic Co-C cleavage that releases reactive radicals seemed paradoxical. Remarkably, CarH appears to resolve this problem by altering AdoCbl photochemistry for its safe use as a photoreceptor chromophore. It was found that photolysis of CarHTt-bound AdoCbl avoids release of Ado• radicals by generating 4 ,5 -anhydroadenosine, a harmless product undetected upon cleavage of free or enzyme-bound AdoCbl ([61]; Figure 2c). Based on ultrafast spectroscopy data, it has been proposed that CarH enables an unprecedented heterolytic cleavage of the AdoCbl Co–C bond to bypass radical formation and release [62] or stabilizes an excited state long enough to ensure the reactions that yield the 4 ,5 -anhydroadenosine product [63]. The molecular mechanism for how CarH alters AdoCbl photochemistry is still unclear. It has been speculated that molecular oxygen and residues around the Ado group, notably of the Wx9EH motif, may be important.

#### **7. Blue Light Sensing, Signaling and Gene Regulation in the B12-Independent Pathway**

#### *7.1. Light Is Perceived through Photoexcitation of PPIX, Which Leads to 1O2 Production*

Although the blue light-PPIX sensing and signaling mechanism to induce carotenogenesis in *M. xanthus* was the first to be identified, it is also the more complex one. Genetic evidence for the role of PPIX came from analysis of *M. xanthus* strains bearing specific deletions of genes in the heme biosynthetic pathway that resulted in elimination or overproduction of endogenous PPIX [30]. Thus, a strain with a deletion of *hemB*, which encodes an early enzyme in the heme biosynthetic pathway was Car−, and the Car<sup>+</sup> phenotype could be restored by supplying PPIX exogenously. On the other hand, a strain with a deletion of *hemH*, whose product incorporates ferrous iron into PPIX in the final step of the heme biosynthetic pathway, exhibited a markedly enhanced light-induced carotenogenesis. The light response thus requires PPIX and correlates with the photosensitizer levels. The need for blue light and PPIX to induce carotenogenesis could be bypassed using the phenothiazinium dye methylene blue and red light, which also generates 1O2, and was suppressed by 1O2 quenchers [30]. The blue light signal is thus transduced via PPIX to 1O2 and then relayed via a recently identified (and unprecedented) mechanism, whose molecular details continue to be unfurled.

#### *7.2. CarF and Plasmalogen Lipids in M. xanthus Blue Light-PPIX-1O2 Signaling*

Signaling by 1O2 produced by blue-light photoexcitation of PPIX absolutely requires CarF [30], which was found in an analysis of Tn*5*-*lac* mutants and mapped to a locus unlinked to those previously identified in *M. xanthus* [70]. CarF is a 281-residue membrane protein with a four transmembrane-helix topology (Figure 3a), and its expression is not light-dependent [70,71]. Sequence homology searches [23] revealed that bacterial CarF-like proteins are present only in myxobacteria and a few Leptospiraceae and Alphaproteobacteria but, intriguingly, they are widespread in animals (invertebrates and vertebrates including humans, where the homolog is named TMEM189 or Kua [72]) and in plants. Protein phylogenetic analysis clearly indicated that CarF homologs from animals and from Leptospira are more related to those in *M. xanthus* and other myxobacteria, and those from Alphaproteobacteria and plants group together and are less related to CarF (Figure 3b). Until very recently their functions were largely unknown, except for the fact that CarF was required in the *M. xanthus* light response, and that a plant CarF homolog was a chloroplast fatty acid desaturase (FAD4) that generates an unusual trans double bond in the *sn*-2 acyl carbon chain [73].

**Figure 3.** *M. xanthus* CarF and plasmalogen synthesis. (**a**) Cartoon representation of the *M. xanthus* CarF protein depicting its experimentally established membrane topology with four transmembrane helices (delimiting residues of each helix numbered in black). Numbered dots correspond to the 12 histidines in CarF. Nine of these (in red) are essential for CarF function but not the rest (in green); all nine essential histidines are conserved in animal CarF homologs, and all of these except His113 in plant homologs. The inner membrane plasmalogens are depicted in red. (**b**) Maximum-likelihood unrooted phylogenetic tree based on selected CarF homologs in metazoa, bacteria and plants (distributed in different colored sectors as indicated; branches in red, ≥75% confidence values from 200 bootstrap replicates; scale bar, number of substitutions per residue). (**c**) The *M. xanthus* plasmalogen biosynthesis pathway highlighting the early pathways and the final step, in which CarF mediates the desaturation that converts its alkyl ether lipid AEPE (1-*O*-(13-methyltetradecyl)-2-(13-methyltetradecanoyl)-glycero-3-phosphatidylethanolamine), to the plasmalogen VEPE (1-*O*-(13-methyl-1-*Z*-tetradecenyl)-2- (13-methyltetradecanoyl)-glycero-3-phosphatidylethanolamine). (**d**) Blue light-PPIX generated 1O2 cleaves the vinyl ether bond of plasmalogens (VEPE) to yield a lyso-PE (2-monoacylglycerophosphoethanolamine) and an (*n-1*) fatty aldehyde (and formic acid, not shown).

A notable feature of CarF is its many (12) histidines, all cytoplasmic, with nine being essential for function ([23]; Figure 3a). The distribution of these histidines, some as HxxxH and HxxHH motifs, resembles that in membrane-associated diiron fatty acid desaturases and hydroxylases of otherwise low overall sequence similarity to CarF [23,70–72]. Hence, these observations hinted that CarF might be a fatty acid desaturase, like FAD4, but probably of a different kind, given that FAD4 lacks one of the crucial histidines in CarF [23].

The exact function of CarF and its role in *M. xanthus* light-induced carotenogenesis has only now been established [23]. It was discovered that CarF and its homologs in animals from worm and fly to fish, mouse and human, but not those in plants, correspond to the long-sought plasmanylethanolamine desaturase (now named PEDS1). This enzyme converts plasmanylethanolamine or alkyl ether phosphatidylethanolamine (glycerophospholipids with the *sn*-1 hydrocarbon chain linked by an ether bond instead of the typical ester bond; hereafter, AEPE) to plasmenylethanolamine, the alkenyl or vinyl ether phosphatidylethanolamine (hereafter, VEPE; Figure 3c). VEPE and analogs with

choline instead of ethanolamine, collectively called plasmalogens, are found in animals and some anaerobic bacteria but not in plants, fungi or most aerobic bacteria except, notably, myxobacteria [23]. Human brain, heart and leukocytes are rich in plasmalogens, which occur in all subcellular membranes, and their deficiency or abnormal levels correlate with many disorders including cancer and Alzheimer's disease [74–76]. As a result of their vinyl ether bond, plasmalogens can affect membrane fluidity and function, and have a proposed antioxidant role given their sensitivity to cleavage by 1O2 and other ROS [77]. However, plasmalogens had never been implicated in signaling photooxidative stress, a role that has now been clearly demonstrated in the *M. xanthus* light-induced carotenogenic response. Thus, deletion of *carF* annuls plasmalogen biosynthesis [23] as well as lightinduced carotenogenesis [30,70], and the latter can be restored by supplying exogenous plasmalogens, even those from human cells that are distinct from the natural ones in *M. xanthus* (in that they have *sn*-1 and *sn*-2 moieties that differ from those in the *M. xanthus* VEPE). Furthermore, deleting genes (*elbD* and MXAN\_1676) implicated in synthesis of the precursor AEPE (Figure 3c) impaired light-induced carotenogenesis, but was rescued by exogenous plasmalogen or by AEPE, which CarF converted to VEPE [23]. In sum, CarF is crucial in the response to light because it is indispensable for the biosynthesis of plasmalogens.

The role of plasmalogens in a blue light-PPIX-1O2 signaled response is both very recent and unprecedented, and identifying the underlying molecular mechanism of action is still being pursued. Breakage by 1O2 of the vinyl ether bond in the plasmalogen yields lyso-PE (2-monoacylglycerophosphoethanolamine) and a fatty aldehyde (Figure 3d; [23,77–79]). This may perturb local membrane structure, environment and properties and affect the function(s) of downstream effector(s) in the pathway. The cleavage products might also function as signaling lipids or second messengers to modulate (or inactivate) effector activity through establishing noncovalent interactions, or covalent adducts between the reactive fatty aldehyde product and target nucleophiles (lysines, cysteines or histidines in proteins). Plasmalogens may themselves bind to specific membrane proteins or complexes to directly modulate their functions through interactions with 1O2. These mechanisms, frequently invoked to link plasmalogens and cellular signaling [77,80], may also operate in *M. xanthus*.

#### *7.3. Light-Induced Expression of the carQRS Operon and Gene crtIb*

Early genetic analysis established that the *carR* locus encodes a negative regulator acting downstream of CarF [32,70], and that *carQ* and *carS*, closely linked to *carR*, encode positive regulators [33–37]. Subsequent DNA sequencing and transcription start site mapping revealed three translationally coupled genes, *carQ*, *carR* and *carS*, forming the *carQRS* operon and expressed from the light-inducible PQRS promoter, which has –35 and –10 promoter elements divergent from typical *M. xanthus* RNAP-σ<sup>A</sup> promoters [36]. Mutations at *carQ* are epistatic over those at *carR* and block activation of *carQRS* as well as of *crtIb*, the structural gene for carotenogenesis unlinked to the *carB* cluster [35–37,39]. Furthermore, *crtIb* expression is driven by a light-inducible promoter PI, with –35 and –10 promoter elements similar to PQRS [39,81]. These findings therefore implicated CarQ in activating *carQRS* and *crtIb* expression from similar light-dependent promoters, and CarR in their downregulation.

While CarS turned out to be the trans acting antirepressor of CarA [57,82], CarQ was identified as the founding member of a new, large and diverse group of alternative σ factors known as the extracytoplasmic function or ECF-σ factors, which were first discovered over 25 years ago [83,84]. Usually, ECF-σ act in a gamut of cellular responses to a variety of extracytoplasmic stimuli (hence the name) and are negatively regulated by association with cognate anti-σ factors, which are often membrane-bound and coexpressed with their ECF-σ partner [85]. CarR was shown to be such a membrane-bound anti-σ, as it specifically and stoichiometrically sequestered CarQ and rendered it inactive in the dark [37] through direct, physical interactions ([71,86]; Figure 4). With six transmembrane helices [36,86,87], CarR belongs to a small group of anti-σ factors with similar membrane topology, largely restricted to proteobacteria, and classified as DUF1109 in the conserved protein domain family database [85]. Some of these other anti-σ act in stress responses to ROS or to heavy metals [88–90] and, interestingly, transcription of both *carQRS* and *crtIb* is activated in the dark by copper [91]. The molecular basis for this copper-mediated action, which bypasses both CarF and light, is still unknown and remains to be elucidated.

**Figure 4.** Model for the blue light-PPIX-1O2 signaling and transduction pathway, and regulation by the CarA-CarS repressor-antirepressor pair in *M. xanthus*. In the dark, the anti-σ factor CarR (with a six transmembrane-helix topology; IM: inner membrane) sequesters its cognate ECF-σ factor CarQ. Blue light excites PPIX to the high-energy 3PPIX state, from which energy transfer to molecular O2 generates the highly reactive 1O2. CarF produces plasmalogens (VEPE), which are required to transmit the 1O2 signal and cause the inactivation of CarR by a mechanism that remains to be elucidated. Plasmalogen cleavage by 1O2 might perturb the local membrane environment of CarR, or its cleavage products may interact with CarR, to alter its activity. This liberates CarQ, which associates with RNAP to activate promoters PQRS (which also requires the CarD-CarG global regulatory complex and IHF) and PI to drive expression of the regulatory *carQRS* operon and of the carotenogenic *crtIb* gene, respectively. CarS, expressed from PQRS in the light, counteracts repression of PB by CarA (see below) to drive expression of the *carB* operon, containing all but one of the carotenogenic genes, leading to the synthesis of carotenoids.

CarQ must be first liberated from its cognate anti-σ CarR, which sequesters it in the dark [37], to associate with RNAP and initiate transcription of its target genes ([86]; Figure 4). Light triggers this liberation of CarQ from CarR, but the exact molecular mechanism remains elusive. CarR was reportedly unstable when exposed to light, especially when cells enter the stationary phase of growth [86], which also correlates with PPIX accumulation. Increased PPIX levels, however, do not activate PQRS and *carQRS* expression in the absence of CarF [30], and since the actual role of CarF is in plasmalogen synthesis, this lipid must somehow mediate inactivation of CarR by light [23]. Various mechanisms, as noted before, can be hypothesized for how plasmalogens mediate CarR activation. Plasmalogen cleavage by 1O2 might perturb the local membrane environment of CarR, or a cleavage product may interact with CarR to alter its activity. Other unknown player(s) or mechanism(s) cannot also be ruled out. These questions will have to be resolved in future work.

#### *7.4. Regulation of CarQ Activity in Light-Induced Expression of carQRS and crtIb*

Negative regulation by CarR is the key determinant of CarQ activity since this controls its availability for association with RNAP. Nonetheless, additional factors required for CarQ activity have also been discovered. An early screen of Tn*5*-*lac* Car− mutants identified two constitutively expressed genes, *carD* and *ihfA*, acting directly in light-induced activation of *carQRS* and, through expression of CarQ and CarS, indirectly in those of *crtIb* and the *carB* operon, respectively [92,93]. The *ihfA* gene encodes the α subunit of the integration host factor (IHF) heterodimer, a nucleoid-associated, histone-like architectural factor that functions as a global regulator [94,95]. Gene *carD* encodes a 316-residue DNA-binding transcriptional factor and is translationally coupled to a downstream gene, *carG*, whose product forms with CarD a tight heteromeric complex that functions as one regulatory unit ([96–98]; Figure 4). CarG is therefore essential for CarD function and the two always coexist. Interestingly, the pair occurs exclusively in *M. xanthus* and related myxobacteria. Thus, at least three other proteins besides CarR, namely IHF, CarD and CarG, regulate CarQ activity at PQRS.

Both CarD and CarG are unusual transcription factors. CarG is a monomer with no DNA-binding capacity, which coordinates two zinc atoms via a His-Cys rich segment (HQx2Hx2Ex2HCx4CxMx16Cx2C; x is any amino acid) [96]. The motif is similar to one found in zinc-metalloproteases called metzincins [99] but an E essential for protease activity is replaced by Q in the motif in CarG, which has no protease activity [96]. In short, CarG can be considered to be one more among the few bacterial transcriptional factors that do not bind DNA [100,101], but which appears exclusively in myxobacteria. CarD is also a rather singular protein. One striking feature is its ~136-residue C-terminal segment comprising a highly acidic ~50-residue region flanked by a C-terminal segment containing four repeats of the RGRP "AT-hook" DNA-binding motif (Figure 5; [102]). Interestingly, these motifs are rare in bacteria but occur in eukaryotic proteins such as high-mobility group type A (HMGA), a relatively abundant, nonhistone architectural factor that remodels chromatin in various DNA transactions [97,102–104]. Similar to HMGA, the CarD C-terminal domain is intrinsically disordered and binds to the minor groove of appropriately spaced AT-rich DNA tracts; and two such tracts at PQRS (at –63 and –77 relative to the transcription start site) to which CarD binds are implicated in CarQ activity [96,98,103,105]. In line with this, the minimum PQRS segment required for CarQ activity is a ~145 bp upstream stretch starting from the transcription start site [106]. By comparison, CarQ activity at its other target promoter, PI, which CarD and IHF affect indirectly, requires a shorter stretch extending to position –54 upstream of the transcription start site [81]. Interestingly, CarD can function in *M. xanthus* even when its natural HMGA-like domain is replaced by human HMGA, histone H1 or the intrinsically disordered H1 C-terminal region, indicating that a basic, structurally disordered C-terminal domain is sufficient for CarD function [98]. Surprisingly, even without its HMGA-like domain, CarD functions in vivo, albeit with diminished activity [107]. By contrast, the remaining N-terminal region of CarD is indispensable for function [108].

**Figure 5.** *M. xanthus* CarD and CdnL. Schematics summarizing structural and functional domains of *M. xanthus* CarD (left) and CdnL (right). Numbers correspond to the residues delimiting the indicated domains, which interact with the partners listed below. Bottom: Structures determined for the RNAP interacting module of CarD (PDB accession code 2LT1) and full-length CdnL (PDB accession code 2LWJ).

Unlike its intrinsically disordered HMGA-like C-terminal domain, the CarD Nterminal domain, CarDNt (Figure 5), is a structurally defined module with sequence similarity to the RNAP-binding domain of bacterial transcription repair coupling factors or TRCFs [103], which repair lesions in the transcribed strand by interacting with RNAP [109]. Indeed, CarDNt has an N-terminal subdomain with a five-stranded β-sheet Tudor-like tertiary structure (Figure 5) similar to its counterpart in TRCF [107] and interacts specifically with the RNAP β subunit [105,110]. Moreover, the C-terminal part of CarDNt, not involved in the interaction with RNAP, binds to CarG [96,98,107]. CarDNt is thus a protein– protein interaction hub directing interactions with both RNAP and CarG, while the CarD HMGA-like C-terminal domain mediates DNA binding.

Remarkably, although CarD homologs are restricted to myxobacteria closely related to *M. xanthus*, CarDNt is a defining member of a large family of bacterial RNAP-interacting proteins (PF02559 or CarD\_CdnL\_TRCF protein family; http://pfam.sanger.ac.uk, accessed on 29 April 2021) that includes not only CarD and TRCF homologs but also a large group of standalone proteins similar to CarD without its HMGA-like domain. These proteins, denoted CdnL (for CarD N-terminal Like), are widely distributed in bacteria and occur in *M. xanthus* and other δ-proteobacteria, α-proteobacteria, Actinomycetes, Firmicutes, Deinococcus-Thermus and Spirochaetes, but not in β-, γ- or ε-proteobacteria, Chlamydiae or Cyanobacteria [108,110,111]. Whereas knocking out *carD* does not affect normal growth or viability, CdnL is indispensable for normal growth and survival of *M. xanthus* [110,112,113]. CdnL has also been reported to be an essential gene in *Borrelia burgdorferi* (spirochaetes), *Mycobacterium tuberculosis* (mycobacteria) and *Rhodobacter sphaeroides* (α-proteobacteria), and to impair normal growth in *Caulobacter crescentus* (αproteobacteria) [111,114–116]. This is because transcription of the essential RNAP-σAdependent rRNA genes in these bacteria requires CdnL in the critical step of open promoter complex (RPo) formation, and CdnL directly or indirectly impacts expression of important biosynthetic genes [111,113,116–118]. In contrast to CarD, which interacts with both RNAP and CarG via CarDNt [105,107], CdnL interacts only with RNAP [110] and does not bind DNA, since it lacks the HMGA-like domain [113]. Nevertheless, the CdnL N-terminal region conserves both the structure and contacts with the RNAP β subunit of its equivalent in CarDNt (Figure 5; [107,113]). When associated to RNAP in RPo, the compact C-terminal region of CdnL can interact with promoter DNA from positions –14 to –10 to stabilize the transcription bubble [101,117]. Some of the functionally important residues in this CdnL domain, which comprises five well-packed α-helices (Figure 5; [113]), are conserved and

important for CarD function as well, and mutating these affects CarD function at target promoters even though binding to CarG remains unaffected [107].

Importantly, the CarD-CarG complex affects processes other than light-induced carotenogenesis in *M. xanthus*. It was implicated in regulating the expression of some early genes in the starvation-induced development to multicellular fruiting bodies [92,96] and of various vegetatively expressed genes of mostly unknown functions [119], none CarQ-dependent. A later study showed that the CarD-CarG complex affects the activities of at least twelve ECF-σ/anti-σ pairs besides CarQ-CarR in *M. xanthus*, suggesting that the complex may control many of the ∼45 putative ECF-σ factors in this bacterium [87]. Except for the light-induced CarQ-CarR pair, the signals that activate each of the other ECF-σ/anti-σ pairs that depend on CarD-CarG are unknown. One pair was, however, recently shown to direct the expression of one of the three CRISPR-Cas systems (type III-B) in *M. xanthus*, which suggested that this bacterial defense system is triggered by a phage [120]. Thus, CarD-CarG is a global regulator, like CdnL, but targets different genes. The parallels with CdnL (despite differences) and the finding that the CarD-CarG complex targets various ECF-σ promoters suggests that CarD-CarG may have a role at these promoters analogous to that of CdnL at RNAP-σA-dependent promoters [107]. This remains to be further explored in future studies.

#### *7.5. Derepression of PB by Light-Induced Expression of the CarS Antirepressor*

The photoregulatory switch controlling expression of the *carB* cluster from the PB promoter relies on repression by both CarH and CarA, and their inactivation by light. Whereas CarH is a photoreceptor that directly senses light, CarA repression is relieved by physical interaction with the CarS antirepressor, whose expression is induced by light (Figure 6; [56–59,82,121]). A series of biochemical, structural and mutational studies demonstrated that the CarA N-terminal domain is an autonomous folding unit with the wingedhelix topology of MerR family DBDs (Figure 6), and that it contains the determinants for specific binding to operator DNA as well as to CarS [57–59]. Further structural-mutational analysis revealed that the highly acidic, 111-residue CarS adopts a five-stranded, antiparallel β-sheet fold resembling SH3 domains (protein–protein interaction modules prevalent in eukaryotes but rare in prokaryotes) and contains a solvent-exposed hydrophobic pocket lined by acidic residues that mimics operator DNA to bind tightly to the DNA recognition helix of CarA and sequester it (Figure 6; [121]). Interestingly, a gain-of-function *carS* mutant (*carS1*) lacking the 25 C-terminal residues results in constitutive, light-independent expression at PB [36], presumably because the variant CarS1 is more acidic than CarS and thus binds more tightly to CarA [57]. Given that CarH recognizes the same operator as CarA and both proteins have similar DBDs and recognition helices, CarH also physically interacts with CarS, albeit with lower affinity than CarA [20,60,121]. Thus, repression of PB by CarA is counteracted by CarS expressed only under light, while PB repression by CarH is relieved mostly by the direct effect of light on the AdoCbl chromophore. CarS homologs occur only in myxobacteria related to *M. xanthus*, and likely play an analogous antirepressor role.

**Figure 6.** Molecular mechanism of CarA-CarS repressor-antirepressor mode of regulation at PB. In the dark, CarA dimers bind cooperatively to its operator at PB, which blocks access to RNAP-σ<sup>A</sup> and represses transcription. CarS, expressed from the *carQRS* operon in the light, acts as a DNA mimic to sequester the CarA DBD and prevent its binding to operator, thereby enabling transcription initiation by RNAP-σ<sup>A</sup> at PB. Bottom: structures of the CarA DBD and CarS1 (PDB accession codes 2JML and 2KSS, respectively) and structural models for CarA-DNA and CarA-CarS1 complexes.

#### **8. Conclusions**

Delving into how *M. xanthus* "sees" and mounts a photooxidative stress response that triggers carotenogenesis uncovered two novel pathways in bacterial light sensing, signal transduction and gene regulation. One pathway relies on a form of vitamin B12 and its association with a single photoreceptor-cum-transcriptional factor, and the other is a B12 independent, more complex route that requires various singular factors. Many worthy firsts can be credited to elucidation of the two pathways, including the discovery of one of the first ECF-σ factors, CarQ [83,84]; the founding members of large protein families, notably the B12-based CarH photoreceptor family [19–22] and the CarD\_CdnL family of RNAPbinding transcription factors [102,103,110]; the long-sought human desaturase involved in plasmalogen biosynthesis through its *M. xanthus* CarF homolog [23]. Insights specific to *M. xanthus* and closely related bacteria, but also ones more broadly conserved across bacteria, have emerged. This photooxidative stress response is linked, directly or indirectly, to that of copper and to heme and fatty acid biosynthesis, and shares global regulators with processes as diverse as fruiting body development and activation of CRISPR-Cas systems. Future work will undoubtedly reveal new, possibly surprising, interconnections to other cellular activities.

Beyond bacterial physiology, signaling and gene regulation, the findings from *M. xanthus* light-induced carotenogenesis have had other important ramifications. How this response and its unique factors are conserved across bacteria and other organisms provides valuable evolutionary insights. Some of the factors involved, which are more typical of eukaryotes, yield phylogenetic signals that may be supportive of the hypothesis that an ancient myxobacterium may have contributed in eukaryogenesis [122]. This hypothesis, known as the Syntrophy hypothesis for the origin of eukaryotes, posits that the eukaryotic cell evolved from symbiosis or syntrophy between a complex early myxobacterial-like deltaproteobacterium (host), an endosymbiotic Asgard-like archaeon (future nucleus) and an alphaproteobacterium (future mitochondrion) [122]. The role of a myxobacterium proposed in this hypothesis was based on the many myxobacterial-like genes in eukaryotes. These phylogenetic signals include, among various others, isoprenoid biosynthesis enzymes, HMGA proteins (CarD) [122], CarF and plasmalogens [23].

Satisfyingly, CarH has now been exploited as one of the few green-light responsive optogenetic tools for light-controlled: (a) gene expression in *M. xanthus* and transgene expression in mammalian and plant cells; (b) receptor interactions and signaling in human cells and zebra fish embryos; (c) generation of protein hydrogels that enable facile encapsulation and release of cells and proteins, and cell adhesions [123–128]. Notably, this last application was very recently adapted to address challenges in regenerative neurobiology to engineer metal-coordinated protein hydrogels for sustained delivery of neuroprotective cytokines aimed at neuronal survival and axon regeneration in vivo [126].

The discovery that CarF and its human and animal homologs are identical lipid desaturases essential in plasmalogen synthesis has not only revealed a remarkable conservation of this enzyme across a vast evolutionary distance, but also has important implications in human health and disease [23]. Plasmalogens have been linked to various human disorders including cancer and Alhzeimer´s disease but the unknown identity of plasmanylethanolamine desaturase had been an impediment in directly assessing the role of these lipids in diverse pathologies. This is now possible with the identity of the enzyme in hand, and has already proved useful in studies of mitochondrial metabolism [129] and ferroptosis [130,131].

**Author Contributions:** Conceptualization, writing—original draft preparation, supervision and funding acquisition S.P. and M.E.-A.; writing—review and editing, all authors. All authors have contributed to the work, read and agreed to the published version of the manuscript.

**Funding:** This research was funded by grants PGC2018-094635-B-C21 (to M.E.-A.) and PGC2018- 094635-B-C22 (to S.P) from the Agencia Estatal de Investigación (AEI)-Spain and European Regional Development Fund (FEDER), and by grant 20992/PI/18 (to M.E.-A.) from Fundación Séneca (Murcia)- Spain. The Ministerio de Educación y Cultura-Spain funded Ph.D. fellowships to A.J.M.-G, E.P.-M. and E.B.-M., and AEI-Spain funded that to R.P.-C.

**Data Availability Statement:** Not applicable.

**Acknowledgments:** We thank Francisco J. Murillo, all past and present members of our group, and our various research collaborators for their valuable contributions.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


### *Article* **Predatory Bacteria Select for Sustained Prey Diversity**

**Ramith R. Nair 1,2\* and Gregory J. Velicer <sup>2</sup>**


**Abstract:** Predator impacts on prey diversity are often studied among higher organisms over short periods, but microbial predator-prey systems allow examination of prey-diversity dynamics over evolutionary timescales. We previously showed that *Escherichia coli* commonly evolved minority mucoid phenotypes in response to predation by the bacterial predator *Myxococcus xanthus* by one time point of a coevolution experiment now named MyxoEE-6. Here we examine mucoid frequencies across several MyxoEE-6 timepoints to discriminate between the hypotheses that mucoids were increasing to fixation, stabilizing around equilibrium frequencies, or heading to loss toward the end of MyxoEE-6. In four focal coevolved prey populations, mucoids rose rapidly early in the experiment and then fluctuated within detectable minority frequency ranges through the end of MyxoEE-6, generating frequency dynamics suggestive of negative frequency-dependent selection. However, a competition experiment between mucoid and non-mucoid clones found a predationspecific advantage of the mucoid clone that was insensitive to frequency over the examined range, leaving the mechanism that maintains minority mucoidy unresolved. The advantage of mucoidy under predation was found to be associated with reduced population size after growth (productivity) in the absence of predators, suggesting a tradeoff between productivity and resistance to predation that we hypothesize may reverse mucoid vs non-mucoid fitness ranks within each MyxoEE-6 cycle. We also found that mucoidy was associated with diverse colony phenotypes and diverse candidate mutations primarily localized in the exopolysaccharide operon *yjbEFGH*. Collectively, our results show that selection from predatory bacteria can generate apparently stable sympatric phenotypic polymorphisms within coevolving prey populations and also allopatric diversity across populations by selecting for diverse mutations and colony phenotypes associated with mucoidy. More broadly, our results suggest that myxobacterial predation increases long-term diversity within natural microbial communities.

**Keywords:** pretator-prey coevolution; antagonism; mucoidy; predatory bacteria; bacterial predation; prey diversity; negative frequency dependence; experimental evolution; MyxoEE-6

#### **1. Introduction**

Predation is one of the most common forms of inter-specific antagonism [1]. Under predation pressure, prey face the dual challenges of optimizing their own acquisition and use of resources for growth and reproduction while avoiding being killed or injured by predators. This dilemma has been shown to play an important role in the ecology and evolution of diverse prey species, including among plants [2], animals [3] and microorganisms [4].

Bacteria fall prey to a wide variety of predators, including unicellular eukaryotes [5], amoebae [6], nematodes [7], and even other bacteria [8,9]. Several studies have shown that over short time periods, microbial predators can elicit phenotypic responses providing resistance against predatory killing. These include filamentation [10], biofilm formation [11], sporulation [12] and production of various extracellular compounds [13] (reviewed in detail in [14]). Although these mechanisms can help thwart predation, many are part of a repertoire of responses that also protect against other stresses, making the degree to

**Citation:** Nair, R.R.; Velicer, G.J. Predatory Bacteria Select for Sustained Prey Diversity. *Microorganisms* **2021**, *9*, 2079. https://doi.org/10.3390/ microorganisms9102079

Academic Editor: David Whitworth

Received: 17 July 2021 Accepted: 23 September 2021 Published: 2 October 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

which they are selected specifically by predation unclear. Longer studies of predator-prey interactions make it possible to identify evolutionary-scale responses of prey specific to interaction with predators and to characterize their temporal evolutionary dynamics.

Long-term coevolution studies have examined the emergence and subsequent evolution of adaptive defensive traits of bacteria under attack by phage [15–17], but few have done so with predatory bacteria. Such coevolutionary studies with phage have shown that bacteria adapt to phage predation by modifying molecules involved in phage adsorption or by producing extracellular polysaccharides to restrict access to the cell surface. However, the reproductive rate of phage is generally much higher than that of its prey, such that dynamics and mechanisms of coevolution between bacteria and phage likely differ greatly from those between bacterial prey and slower-growing predatory bacteria.

Myxobacteria are soil- and sediment-dwelling bacteria that kill and consume diverse other microbes, including both Gram+ and Gram– bacteria and fungi, by mechanisms that remain poorly understood [18–20]. Because of this broad prey range, predation by myxobacteria is predicted to play significant roles in shaping the composition, structure and evolution of complex microbial communities [21].

In a predator-prey coevolution experiment with *Escherichia coli* as prey and the myxobacterium *Myxococcus xanthus* as predator recently named MyxoEE-6 [22], we previously showed that coevolving prey were under selection both for parallel losses of function in a prey outer-membrane protein (OmpT) and favoring the increase of genotypes that generate a mucoid-colony phenotype [23]. Mucoid colonies are characterized by increased opacity, generally lighter pigmentation and convex colony surfaces relative to non-mucoid colonies, although detailed mucoid-colony phenotypes may vary among conspecifics. Mucoidy is achieved by increased secretion of extracellular polysaccharides and can evolve in response to a number of biotic and abiotic challenges, including bacteriophages [15], antibiotics [24], macrophages [25] and menthol [26]. Mucoid colony-forming *E. coli* cells were found to arise frequently in MyxoEE-6 populations co-evolving with *M. xanthus*, but were largely absent from prey-only control populations [23]. Additionally, mucoidy was associated with reduced swarming and killing by the predator.

Our first study of MyxoeEE-6 and many earlier studies have shown that predation often strongly impacts prey diversity [23,27–30], but longer-term fates and dynamics of such predation-induced diversity are underexplored. Here we test whether mucoid lineages in MyxoEE-6 were on frequency trajectories predictive of long-term fixation of the phenotype or rather reveal the operation of evolutionary mechanisms preventing fixation, for example negatively frequency-dependent selection or clonal interference [31–33]. We then test for negative frequency dependence (NFD) of fitness in one clone pair and for cost of mucoidy that might promote extended maintenance of mucoid/non-mucoid polymorphisms. Finally, we characterize inter-population genetic and phenotypic diversity mediated by predation.

#### **2. Materials and Methods**

#### *2.1. Predator-Prey Coevolution*

MyxoEE-6 was conducted as described in Nair et al. (2019) [23]. Briefly, replicate populations of *M. xanthus* strain DK3470 and *E. coli* strain MG1655 (along with prey-only and predator-only controls) were paired and spread with glass beads on 8 mL prey-growth agar (1× M9 salts, 2 mM MgSO4, 0.1 mM CaCl2, 0.2% glucose, 1.5% agar) in 50 mL conical flasks. The flasks were incubated at 32 ◦C for 84 h, after which they were harvested by adding 5 mL TPM buffer and shaking on an orbital shaker set at 300 rpm for 15 min. 1% of the surviving community was transferred to fresh minimal media and the cycle was repeated 25 times, allowing ∼166 generations of growth.

#### *2.2. Mucoid Frequency Estimation and Clone Isolation*

*Mucoid frequency estimation.* Evolved *E. coli* colonies that were of lighter pigmentation, more opaque and more convex relative to the ancestral colony phenotype on LB agar were identified as mucoid. To determine the frequency of mucoids in each replicate MyxoEE-6

community/population at different time points, aliquots of frozen stocks from respective lineages and time-points were thawed in 100 μL TPM buffer, diluted and plated onto LB agar plates. Following overnight incubation at 32 ◦C, 90% humidity, mucoid and nonmucoid colonies were distinguished and counted. Cycle 18 and cycle 25 populations were sampled over successive sampling periods, with each replicate sampling for each cycle performed independently at different times. Samples from cycles 6, 10, 14, 16, 20 and 22 of *E. coli* populations ME4, ME8, ME11 and ME12 were collected and plated together in the same sets of at least 3 temporally separated biological replicates. When zero mucoid colonies were present on the plate selected for counting from a dilution series, we estimated a hypothetical maximum mucoid frequency as [1/(colony *N* + 1)] to mimic a scenario in which the actual colony count *N* for the plate was increased by one and the additional hypothetical colony was mucoid. The four populations with the highest mucoid frequencies at cycle 18 (ME4, ME8, ME11 and ME12) were selected for examination at more time points to allow the greatest opportunity to accurately resolve frequency dynamics. The colonies scored to assess mucoid frequencies were not subsequently cultured or stored frozen.

*Isolation of cycle 18 mucoid and non-mucoid clones.* One mucoid and one non-mucoid clone each were isolated from cycle-18 co-evolved populations in which mucoids were detected and subsequently stored frozen. The clone pair from ME4 was used for the direct competition experiment measuring mucoid vs non-mucoid relative fitness and all nine clone pairs were used in the survival-under-predation and productivity assays and for colony-phenotype imaging. Single colonies of each phenotype were picked and streaked onto fresh LB agar plates to isolate a sub-colony and thereby purge any genetic variation that may have been present in the colony from the original plating. Isolated colonies of each phenotype were picked, grown to high density in LB liquid and samples of the resulting cultures were stored as frozen stocks (20% glycerol) for further experiments. The same mucoid and non-mucoid clone from each co-evolved population were used in the productivity and predation assays described below and were used for imaging colony phenotypes. For colony-phenotype imaging, aliquots of frozen stocks from mucoid and non-mucoid clones from nine different lineages were thawed in 100 μL TPM buffer and 10 μL of the suspension was spotted on an LB agar plate. The resulting colony was imaged after overnight growth at 32 ◦C using Olympus SZX16 stereomicroscope at 0.8X magnification. The same mucoid and non-mucoid clones from cycle 18 population ME4 were also used in the relative fitness assay.

#### *2.3. Productivity and Predation Assays*

Experiments measuring *E. coli* mucoid vs. non-mucoid relative fitness in the presence and absence of *M. xanthus*, survival in the presence of *M. xanthus* and population size after 25 h of incubation (productivity) were performed under the same abiotic conditions as MyxoEE-6 [23]. *E. coli* cells of each type were grown overnight in LB liquid at 32 ◦C, 300 rpm, then diluted to OD600 1.0 with LB and subsequently diluted 1:100 with TPM liquid buffer immediately prior to initiating the relevant assay. Initial and final population sizes were determined by dilution plating onto LB agar and counting colonies after overnight incubation at 32 ◦C, 90% humidity. *M. xanthus* was grown in 8 mL CTT liquid (10 g/L casitone, 10 mM Tris pH 8.0, 8 mM MgSO4, 1 mM KPO4) in 50 mL flasks at 32 ◦C, 300 rpm until mid-exponential phase, when cultures were centrifuged (5000 rpm, 15 min) and resuspended in TPM buffer to a density of ∼109 cells/mL. 50 <sup>μ</sup>L of the diluted prey suspension were then either mixed with 50 <sup>μ</sup>L of *M. xanthus* cell suspension at ∼109cells/mL or 50 <sup>μ</sup><sup>L</sup> TPM buffer (productivity assays and competitions without predator) before being spread onto MyxoEE-6 prey-growth agar [23]. For the relative-fitness competition experiment, one mucoid and one non-mucoid clone each isolated at the end of MyxoEE-6 cycle 18 from the coevolution treatment ME4 were used. In those experiments, cultures of the paired *E. coli* competitors were mixed at the specified ratios after the adjustment to OD600 1.0 prior to proceeding as described above. Cultures were harvested as during MyxoEE-6. The selection rate constant was calculated as in [34] using the following formula:

$$\text{sij} = \frac{1}{t} \left[ \ln \left( \frac{N\_l(t)}{N\_l(0)} \right) - \ln \left( \frac{N\_{\dot{f}}(t)}{N\_{\dot{f}}(0)} \right) \right],\tag{1}$$

where *Ni(0)* and *Ni(t)* are initial and final population densities of the mucoid clone, while *Nj(0)* and *Nj(t)* are initial and final population densities of the competing non-mucoid clone respectively.

#### *2.4. Genomic Data*

Candidate mutations for causation of mucoidy were determined by comparing wholegenome sequences from two non-mucoid clones and one mucoid clone from four coevolved populations after cycle 25 (ME4, ME8, ME11, ME12). The genomic data for these 12 clones have already been published [23] and made available elsewhere (deposited in the SRA database under BioProject accession PRJNA551936 (BioSample accessions SAMN12169214—SAMN12169315)).

#### *2.5. Statistical Analysis*

Frequency dependence of the selection rate constant for competition experiments between mucoid and non-mucoid clones from population ME4 was tested using one-way ANOVA with starting frequency (0.1, 0.25, 0.5) as a factor. We tested for differences in average productivity as well as killing resistance between mucoid vs. non-mucoid clones using Welch's two-sided, two-sample *t*-tests. We tested for differences in productivity and killing resistance among mucoid clones from different coevolving populations with one-way ANOVA with population ID as a factor in both cases. Statistical analyses were performed in R [35] using R studio (version 1.3.1093) and plotted with the package ggplot2 version 3.3.4 [36].

#### **3. Results**

#### *3.1. Mucoid Phenotypes Rise to and Remain within Intermediate Frequency Ranges*

In the first study of MyxoEE-6, we reported parallel emergence of mucoid colonyforming variants in *E. coli* prey among ten out of twelve replicate populations that had co-evolved with *M. xanthus* for 18 cycles [23], whereas mucoids were largely absent from control *E. coli* populations that evolved without predators. This pattern strongly suggested that most or all mucoid genotypes detected among the co-evolved populations rose to high frequency due to selection rather than neutral drift. Indeed, mucoidy was found to be associated with reduced susceptibility to predation. However, mucoid colonies were present only as a minority after cycle 18, leaving open the question of whether, toward the end of MyxoEE-6, mucoids were (i) sweeping to fixation, (ii) decreasing toward loss due to clonal interference from non-mucoid adaptive mutants, or (iii) being maintained long-term by balancing selection while fluctuating around intermediate equilibrium frequencies.

In this study, we first compared mucoid frequencies in all MyxoEE-6 *E. coli* populations, including the 12 replicate populations that co-evolved with *M. xanthus* (ME1–ME12) and the six control replicate populations that evolved in the absence of *M. xanthus* (E1–E6), between cycles 18 and the end of the MyxoEE-6 experiment at cycle 25. Although mucoids appear to have generally decreased in frequency in many co-evolved prey populations from cycle 18 to 25, they were nonetheless present above our levels of detection in nine of the twelve co-evolved populations at cycle 25 (Figure S1). Between cycles 18 and 25, mucoids dropped below the limit of detection in two populations (ME7 and ME9) and newly rose to a detectable level in one population in which no mucoids had been found at cycle 18 (ME2, Figure S1). Thus, in total across the two time points, mucoid colony variants were observed in eleven out of the twelve coevolving prey populations and were at frequencies above the limit of detection at both time points in most. In contrast, among the six prey-only control populations, mucoidy was detected in only one replicate sample of one population at one time point (Figure S1).

The cycle 25 results did not strongly suggest trajectories toward either fixation or loss of mucoids. We thus addressed the above hypotheses more rigorously for the four prey populations with the highest mucoid frequencies at cycle 18 by examining their mucoid frequencies at six additional MyxoEE-6 timepoints (Figure 1). In all four populations, mucoids increased several orders of magnitude from zero to detectable frequencies already by the end of cycle 6, indicating strong selection favoring mucoidy. After this commonality of rapid early increase, mucoids in the four populations fluctuated within an intermediate range of detectable minority frequencies (Figure 1).

**Figure 1.** Mucoid prey rose to and persisted within intermediate frequency ranges in MyxoEE-6 in four focal populations. Frequencies of mucoid variants among the four focal coevolving prey populations estimated hypothetically at the start of MyxoEE-6 (cycle 0, open circles) and estimated directly by sampling at the end of eight MyxoEE-6 cycles (see Methods). A mucoid frequency of <sup>∼</sup>10−<sup>5</sup> at cycle 0 is shown for a hypothetical scenario in which there was one mucoid cell among the <sup>∼</sup>10<sup>5</sup> starting prey cells for each *E. coli* population. Grey dots show individual replicate estimates, black dots indicate cross-replicate means and error bars show 95% confidence intervals (*t*-distribution, three temporally separated biological replicates).

After cycle 6, however, each of the four focal populations showed a qualitatively unique pattern of subsequent mucoid-frequency dynamics. Mucoids in the ME4 community exhibited the most stable dynamics, remaining at frequencies near 0.01 from cycles 6–16, rising ∼10-fold by cycle 18 and remaining at frequencies near 0.1 for the duration of MyxoEE-6. Mucoids in ME8, ME11 and ME12 showed strikingly similar dynamics through cycle 16, all increasing to frequencies above 0.1 by cycle 10, subsequently decreasing by cycle 14 and increasing in parallel again by cycle 16 before diverging more in their detailed dynamics through the remaining cycles.

It is noteworthy that on the one hand, none of the four populations showed any indication of mucoids approaching fixation and, on the other hand, after cycle 25, mucoids in all populations remained within or very near the respective mucoid frequency range covered from cycles 6–22. Thus, despite late decreases in three populations (ME8, ME11 and ME12), the observed patterns are consistent with, and in our view suggestive of, extended maintenance of minority mucoids by NFD of mucoid fitness.

#### *3.2. A Single-Cycle Competition Experiment Reveals a Predation-Specific Advantage to Mucoidy but Not NFD*

NFD of fitness might be generated by direct interactions between mucoid and nonmucoid cells within a given MyxoEE-6 growth cycle. For example, costly production of exopolysaccharides by mucoids might confer some degree of social protection from predation and thus a fitness advantage to non-mucoids, with the degree of benefit conferred correlating with mucoid frequency. In this cheating scenario, NFD-mediated maintenance of the mucoid/non-mucoid polymorphism should be manifested by a reversal of fitness ranks (e.g., [37,38]) between mucoids and non-mucoids simply as a function of frequency within a single MyxoEE-6 growth cycle.

In our first study of MyxoEE-6, we reported that fewer cells of a mucoid clone isolated from the coevolving ME4 prey population at cycle 18 are killed by the ancestral predator compared to a contemporary non-mucoid isolate [23], but direct competition experiments between these prey isolates were not performed. Here we directly competed these same two isolates in the presence and absence of predators to both confirm a predicted direct relative fitness advantage of the mucoid isolate specific to predation pressure and to test whether this fitness advantage is frequency dependent within a single competition cycle due to social cheating.

The mucoid and non-mucoid prey clones were mixed at initial starting mucoid frequencies of 0.1, 0.25 and 0.5 and allowed to compete while growing under the same the experimental conditions as MyxoEE-6, except that the predator (a clone of *M. xanthus* also isolated from the ME4 community after cycle 18) was either absent or added at a population size of ∼<sup>5</sup> × <sup>10</sup><sup>7</sup> predator cells to impose predation pressure. In these assays, the mixed *E. coli* populations generally increased more than 100-fold during the competition period. As expected, the mucoid genotype had higher fitness than the non-mucoid clone in the presence of *M. xanthus* but not in its absence (Figure 2, one-way ANOVA, F1,16 = 235.1, *<sup>p</sup>* = 5.49 × <sup>10</sup>−11). However, NFD of fitness within a single competition cycle was not detected for this particular pair of competitors (one-way ANOVA, F2,6 = 0.644, *p* = 0.558). These results thus do not support the hypothesis that mucoids fail to reach fixation due to cheating-mediated NFD. However, we note that this outcome does not generally exclude the NFD hypothesis for other experimental conditions or for other mucoid and non-mucoid competitors.

#### *3.3. Mucoidy Is Associated with Low Prey Productivity in the Absence of Predators*

Adaptations of prey that decrease susceptibility to predation often come at a cost to other components of fitness. We previously found no cost of mucoidy to growth rate from 5–20 h after inoculation into the MyxoEE-6 selective regime for a pair of ME4 cycle 18 clones in the absence of predators [23]. However, final population-size productivity as well as rate of population increase can impact overall fitness and we hypothesized that mucoidy might come at cost to productivity. To explore this possibility, we examined one mucoid and one non-mucoid clone each from nine of the cycle 18 coevolving populations in which mucoids were detected (ME3, ME4, ME5, ME6, ME7, ME8, ME9, ME11, ME12).

We first confirmed that the mucoid clones are on average less susceptible to killing by *M. xanthus* than the corresponding non-mucoid clones from the same population (Figure 3, Welch's two-sample *t*-test: *t*15.8 = 2.558, *p* = 0.011), as was expected from previous results with the clone pair from population ME4 [23]. To examine the productivity-cost hypothesis, the clones were grown under MyxoEE-6 abiotic conditions for 25 h in the absence of predators and their final population sizes (productivity) determined. The average productivity of mucoid clones was found to be ∼27% lower than that of nonmucoid clones (Figure 4, Welch's two-sample *t*-test: *t*14.18 = −3.5636, *p* = 0.0015). Thus, mucoidy protects against predation, but at a cost to total productivity in the absence of predation. This productivity cost of mucoidy led us to speculate that it may mediate a form of NFD in which mucoid frequency impacts total *E. coli* productivity, which, along with and decreased susceptibility to predation by mucoids, impacts predator productivity at the

end of a growth cycle, which in turn impacts mucoid vs non-mucoid relative fitness in the subsequent growth cycle. We elaborate this hypothesis further in the Discussion.

**Figure 2.** A mucoid clone outcompetes a non-mucoid clone only under predation pressure but does not exhibit negative frequency dependence. Estimates of the fitness of a mucoid clone isolated from ME4 at cycle 18 in competition relative to a contemporary non-mucoid clone (as represented by the selection-rate constant) across three initial frequencies are shown. x-axis values depict the estimated starting frequencies for the mucoid clone at the beginning of the experiment. Positive and negative values indicate estimates that the mucoid clone has higher or lower fitness than the non-mucoid clone, respectively. Smaller and larger dots represent individual-replicate and mean values, respectively. Error bars show 95% confidence intervals from three biological replicates (*t*-distribution).

**Figure 3.** Mucoidy is associated with higher survival of encounters with *M. xanthus*. Frequencies of prey surviving encounters with *M. xanthus* are shown. Grey dots are mean survival frequency for clones from each of nine coevolving populations from which the clones were isolated at the end of cycle 18. Three independent estimates were generated for each population (Figure 5b). Black dots are means across the nine populations for each prey type and error bars show 95% confidence intervals (*t*-distribution).

**Figure 4.** Mucoids have lower population productivity than non-mucoids in the absence of predators. Log-transformed (productivity) is shown. Grey dots are mean values from nine coevolving populations from which the clones were isolated at the end of cycle 18. Three independent estimates were generated for each population (Figure 5c). Black dots are means across the nine populations for each prey type and error bars show 95% confidence intervals from nine replicate populations (*t*-distribution).

#### *3.4. Mucoidy Is Phenotypically Variable*

We observed visually that colonies of the nine mucoid clones isolated from different lineages at the end of cycle 18 were phenotypically diverse, varying with regard to colony size, opacity, color and degree of phenotypic differentiation from non-mucoid colonies from their respective populations (Figure 5). We therefore tested for quantitative variation among these clones with respect to susceptibility to predation and productivity. While the mucoid clones are collectively less susceptible to killing by *M. xanthus* than non-mucoids (Figure 3), the degree of resistance to killing varied significantly among the nine mucoid clones tested here (Figure 5b, one-way ANOVA, F8,18 = 3.988, *p* = 0.007). Additionally, while mucoids collectively have lower productivity than non-mucoids in the absence of predation (Figure 4), the isolated mucoid clones varied significantly in their productivity relative to contemporaneous non-mucoid clones isolated from the same prey population (Figure 5c), one-way ANOVA, F8,18 = 4.4158, *p* = 0.004). Thus, while mucoidy in general is clearly selected by bacterial predation, the broader categorical phenotype can be associated with a diversity of detailed colony phenotypes and predation-related parameter values.

#### *3.5. Candidate Mutations for Mucoidy*

The observed variation in mucoid phenotypes suggested that distinct prey lineages may have followed different genomic routes to mucoidy, with distinct causal mutations having different phenotypic effects. Alternatively, the variable mucoid phenotypes might have resulted from differences in epistatic interactions between shared mutations or mutation targets (genes or gene pathways) and other evolved mutations in the same genetic background. To generate hypotheses regarding genetic causation of mucoidy in MyxoEE-6, we compared whole-genome sequences of one mucoid and two non-mucoid clones each from the four focal coevolved prey populations examined here for mucoidy dynamics across MyxoEE-6 (ME4, ME8, ME11, ME12) [23]. Mucoid clones from all four populations each had a mutation in one gene in the *yjbEFGH* operon, with only one gene (*yjbH*) mutated in more than one population (Table 1). In contrast, no mutations in this operon were found

in any of the non-mucoid colonies, strongly suggesting that mutations in this operon confer mucoidy. This operon encodes proteins involved in the production of an uncharacterized extracellular polysaccharide [39]. Additionally, the mucoid clone from ME4 also has a mutation upstream of the gene *rcsA* (Table 1). Because *rcsA* is a transcriptional regulator of colanic acid capsular biosynthesis [40], this mutation may contribute to mucoidy in this clone.

**Figure 5.** Allopatric variation among mucoid prey. (**a**) Observed mucoid (bottom row) and nonmucoid (top row) colony phenotypes on LB agar for the respective clones isolated from nine coevolving populations at the end of cycle 18 and used in the experiments reported in Figures 3, 4 and 5b,c. Phenotypic differences between the mucoid isolates and between mucoid vs. non-mucoid clones from the same population are yet more pronounced upon direct visual and microscopic observation than in the two-dimensional images shown here. (**b**) Frequencies of prey surviving encounters with *M. xanthus* are shown. (**c**) The relative productivity (population size after 25 hr) of single mucoid vs non-mucoid clones isolated from nine populations at the end of cycle 18 is shown. A value of 1 (solid black line) indicates equal productivity. (**b**,**c**) Small dots are values from individual replicates and large dots are inter-replicate means. Errors bars show 95% confidence intervals (*t*-distribution, n=3).

**Table 1.** Mutation candidates for causation of mucoidy in cycle 25 mucoid isolates from the four focal co-evolved populations.


#### **4. Discussion**

By imposing predatory selection on prey, predators can increase diversity within prey communities and populations by diverse mechanisms. For example, keystone predators can promote diversity by targeting the most dominant prey species and thereby increasing resources available for other prey species [27–29,41,42]. Predation can ameliorate competitive exclusion among competing prey species by promoting prey that are less competitive at low predation pressure but less susceptible to predation by a dominant predator [10,43]. This can result in a negative feedback loop in which predation drives increase of less-susceptible prey, eventually resulting in reduced predator population size (and thus reduced predation pressure), which in turn reverses prey fitness ranks back in favor of prey that are more competitive at low predation pressure [28]. Predators can also maintain polymorphisms among prey through apostatic selection [30,44,45] and across environmental gradients by selecting for different prey alleles in different environmental contexts [46]. While such effects of predation on prey diversity have been investigated in a broad array of organisms, predatory bacteria are understudied in this regard.

Having previously shown that predatory bacteria select for mucoid variants of prey during coevolution [23], here we tested whether mucoid frequency dynamics indicate or suggest longer-term fixation, loss or intermediate persistence of mucoidy, especially in four focal populations of MyxoEE-6. Mucoid frequencies increased several orders of magnitude early in the experiment until reaching detectable frequencies already by cycle 6 (Figure 1). At this rate of increase, mucoids should have easily reached fixation during MyxoEE-6 in the absence of clonal interference or negative frequency dependence, but did not. Among the four focal populations, mucoids ceased increasing after reaching detectable frequencies and never reached majority status at sampled time points. After reaching their maximum frequencies, rather than decreasing continuously to undetectable frequencies, as might be expected under clonal interference by non-mucoid adaptive genotypes, mucoids fluctuated within detectable frequencies ranges to the end of MyxoEE-6, an outcome suggestive of negatively frequency-dependent selection.

In light of the above results, we designed an experiment to test for such NFD between a mucoid and a non-mucoid clone from population ME4 (cycle 18) resulting from immediate effects of prey-type frequency on prey-type fitness that would be observable over the course of a single cycle of MyxoEE-6 community growth. Such single-cycle NFD could result, for example, if non-mucoids can cheat on mucoids with respect to EPS production [47], not incurring the cost of excess EPS production borne by mucoids but receiving some benefit of such production by neighboring mucoid cells with respect to predation susceptibility. However, this experiment did not detect such NFD between these two focal prey clones, despite confirming a general fitness advantage to the mucoid clone specific to the presence of predation pressure (Figure 2). This outcome might suggest that mucoid vs. non-mucoid fitness is in fact not generally dependent on frequency in a predation-dependent manner, but it is premature to exclude the NFD hypothesis absent further experiments. It remains possible that the observed fitness relationships between the particular ME4 clones chosen for this experiment are not representative of those between most mucoid and non-mucoid genotypes and/or that details of the experimental conditions under which the competition experiment was performed (e.g., starting predator density) prevent the manifestation of NFD that did actually occur during MyxoEE-6. Alternatively, the productivity cost of mucoidy demonstrated here (Figures 4 and 5c) suggested to us the possibility of a distinct form of NFD that, if it occurs, would only play out over multiple growth cycles.

We speculate that the productivity cost and predation-resistance of mucoidy have the potential to collectively mediate NFD of mucoid fitness by reducing predator numbers when mucoids are sufficiently frequent in one growth cycle, which may in turn impact mucoid vs. non-mucoid fitness in the subsequent growth cycle. Under MyxoEE-6 conditions, *E. coli* grows much faster than *M. xanthus* and *M. xanthus* is completely dependent on *E. coli* for growth. For these reasons, we hypothesize that *E. coli* populations generally reached carrying capacity within each growth cycle before *M. xanthus* predator populations

increased sufficiently to impose strong predation pressure that would selectively favor mucoids. Due to the productivity cost of mucoidy, non-mucoids may have generally outcompeted mucoids upon completion of prey growth earlier in each cycle, while mucoids may have often outcompeted non-mucoids later in each cycle, after predator density had increased sufficiently to impose strong predation pressure and thus favor mucoidy. In this scenario, sufficiently high late-cycle predation pressure promotes the net increase of mucoids over the entire cycle, despite a disadvantage of mucoids upon completion of prey growth before predators increase to large population sizes. However, mucoid frequency in one MyxoEE-6 cycle might impact the degree of late-cycle predation pressure in the next cycle if predators exhibit lower productivity on mucoid cells than non-mucoids.

This latter scenario would require first that predators exhibit lower productivity on mucoids than on non-mucoids, a plausible scenario given that mucoid prey have lower productivity and are less susceptible to predatory death. It would also require that lower predator productivity caused by relatively high mucoid frequency in one cycle decreases total predation pressure over the next cycle sufficiently to give non-mucoids a net wholecycle advantage over mucoids in that subsequent cycle. This hypothetical scenario involves a negative feedback loop analogous to such loops proposed or demonstrated to maintain diversity in other predator-prey systems [28]. A test of this hypothesis with respect to MyxoEE-6 would require additional experiments to examine whether these requirements are often met across a range of mucoid vs non-mucoid genotype combinations.

Mucoid phenotypes in *E. coli* are associated with increased production of exopolysaccharides, especially colanic acid (CA-EPS) [48,49] and have been linked with mutations in *rcs*, *yjb* and *yrf* genes [40,48]. Given that all sequenced mucoid *E. coli* clones carried mutations in the *yjbEFGH* operon, while none of the non-mucoid clones had such a mutation, it appears that mucoidy in MyxoEE-6 was achieved predominantly by mutation of this operon, which is involved in the synthesis of an uncharacterised exopolysaccharide [39]. Interestingly, deletion of this operon in a Δ*rpoS* background has been shown to result in CA-EPS synthesis and mucoidy [48]. Apart from the mutations in *yjbEFGH*, the mucoid strain from lineage ME4 had an additional mutation upstream of *rcsA*, which is a known positive regulator for colanic acid synthesis [40,50]. Thus, mutations in *yjbEFGH* and *rcsA* are likely to mediate the apparent tradeoff between resistance to *M. xanthus* predation and productivity, although additional genetic experiments would be required to demonstrate this directly. Collectively, mucoidy seems to have been achieved largely through mutating the same operon, but in different genes and sites. These distinct mutations may have themselves generated the phenotypic variation among mucoid genotypes documented here, or they may have interacted with other mutations to generate such diversity.

Natural microbial communities contain myriad species of prey and predators coexisting in complex food chains and webs [51–53]. The shifting balance of selection imposed by predation vs resource utilisation is likely to contribute to this diversity [28,41,43]. Our results with simple synthetic bacterial communities initially derived from just one prey genotype and one predator genotype suggest that both sympatric and allopatric prey diversity that is evolutionarily induced by bacterial predators can be long-lived. This suggests more strongly than previous studies that predators of microbes and myxobacteria in particular promote greater intra-specific diversity across many prey types over long evolutionary time scales than would co-exist in the absence of predation. Future experiments with more complex communities are required to test this prediction.

At the inter-specific level, it is now well understood that variable levels and forms of intraspecific diversity impact community evolution (e.g., [54–56]). Thus, myxobacteria are likely to shape microbial community composition and structure both directly through differential predation of distinct species [21] and indirectly by shaping long-term patterns of intraspecific diversity that in turn influence inter-specific fitness relationships.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/microorganisms9102079/s1, Figure S1: Mucoid-frequency estimates for all evolved MyxoEE-6 *E. coli* populations after cycles 18 and 25.

**Author Contributions:** R.R.N. and G.J.V. designed the study and analysed the data. R.R.N. conducted the experiments. R.R.N. wrote the initial draft. Review and editing by G.J.V. and R.R.N. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Raw data presented in figures available at Figshare (doi:10.6084/ m9.figshare.16726465).

**Acknowledgments:** We thank Dominique Schneider, Alex Hall and Sébastien Wielgoss for helpful discussions and inputs, Marco La Fortezza for help with microscopy and Lisa Freund for experimental assistance.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

