Evolution of Superinfection Immunity in Cluster A Mycobacteriophages

Many aspects regarding superinfection, immunity, virulence, and the evolution of immune specificities are poorly understood due to the lack of large collections of isolated and sequenced phages with a spectrum of genetic diversity. Using a genetically diverse collection of Cluster A phages, we show that the classical and relatively straightforward patterns of homoimmunity, heteroimmunity, and virulence result from interactions between homotypic and heterotypic phages at the extreme edges of an evolutionary continuum of immune specificities. Genetic interactions between mesotypic phages result in more complex mesoimmunity phenotypes and virulence profiles. These results highlight that the evolution of immune specificities can be shaped by homotypic and mesotypic interactions and may be more dynamic than previously considered.

also defending against other superinfecting phages and escaping other prophage defenses.
Phage immunity systems, critical components of the temperate life cycle, are a target of these evolutionary forces. Immunity systems are genetic circuits that control the temporal expression of genes required for lytic growth (Fig. 1). Coliphage harbors the most highly characterized system, a single genetic locus comprised of two DNAbinding transcriptional regulatory genes, cI and cro, that compete for binding of two tripartite operator sites (7). During lysogeny, CI binds to the operators to block transcription initiation of cro, which is required for lytic growth. Immunity systems are diverse and vary in complexity, sometimes consisting of multiple genetic loci (7)(8)(9).
within promoters, such as the P left early lytic promoter. However, most sites are not promoter associated, and the repressor binds these "stoperators" to block transcription elongation (9,29). L5 and Bxb1 typify the relationships between phages of different subclusters in that they are heteroimmune, their stoperators/operators have distinct consensus sequences, and their repressors show strong binding preferences for their cognate binding sites (26).
Here, we take advantage of the large and diverse group of Cluster A phages to investigate their evolving immune specificities. In addition to homoimmunity and heteroimmunity, we find a surprisingly broad spectrum of mesoimmune specificities, including partial (or incomplete) and asymmetric superinfection immunities (Fig. 1). The variety of immunity phenotypes provides a rich landscape for mutational variation with push-pull dynamics, moving toward either shared immunity or escape from immunity. The evolutionary trajectories are thus likely to be nonlinear as they navigate a complex spectrum of phage relationships.

RESULTS
Characterization of the Cluster A immunity system. All Cluster A phages, including L5, exhibit similar genomic architectures (Fig. 2a). The left arm contains structural and assembly genes, the right arm contains genes associated with lytic growth such as DNA replication, and the genome center contains prophage inheritance genes such as integration or partitioning systems (16,30). The immunity repressor can be readily identified at syntenic positions, and the early lytic promoter, P left , is near the right genome terminus (Fig. 2a). Despite the conserved synteny, these phages are genetically diverse and have been further subdivided into 19 subclusters (Fig. 2b). Phages from distinct subclusters, such as Bxb1 (Subcluster A1), L5 (Subcluster A2), and Peaches (Subcluster A4), have highly divergent repressors and stoperator motifs and are heteroimmune (16). However, the genetic diversity within and between subclusters is not homogenous, and there is a clade of nearly 100 phages representing 10 subclusters that are more closely related to L5 than Bxb1 or Peaches (Fig. 2b). All phages in this "L5 clade" (except for those in Subcluster A15) infect Mycobacterium smegmatis mc 2 155, and they exhibit a spectrum of genetic diversity based on their gene content and nucleotide sequence (Fig. 2c). We therefore focused on the superinfection immunity relationships of phages in this clade, beginning with analysis of the sequence relationships and their divergence.
Immunity repressors in the L5 clade are similar in size (see Fig. S1a in the supplemental material) and exhibit a genetic spectrum that correlates with whole-genome gene content distances (Fig. 2d). As seen with L5, Bxb1, and Peaches, a set of stoperator sites can be identified in each genome (see Materials and Methods). Similar numbers of stoperators are present in each genome, and they predominantly exhibit one orientation relative to the direction of transcription ( Fig. S1b and c) (9,26). Sequence motifs representing each genome's cognate stoperators are similar, but not identical, to each other (Fig. 2e), and they exhibit a genetic spectrum that also correlates with wholegenome gene content distances as well as repressor distances (Fig. 2f).
Gene expression profiles. The genetic diversity of phages in the L5 clade raises questions as to whether they exhibit similar expression profiles and whether they carry additional genetic elements that may interfere with superinfection, such as secondary ) between phages, as described above for panel c. D Nuc ranges from 0 (100% sequence similarity) to 0.5 (no sequence similarity). D GC ranges from 0 (all phams are identical) to 1 (no phams are identical). D Rep ranges from 0 to 100 amino acid substitutions per 100 amino acids. D Stop motif equals zero for identical motifs, and larger values reflect higher degrees of dissimilarity. See Materials and Methods for further description of distance measurements.
In each of these phages, repressor expression initiates in the upstream intergenic region, extends across rep, and substantially decreases across the adjacent gene of unknown function and the Cas4-family gene (Fig. 3a). A conserved repressor binding site is located in this intergenic region, which in L5 is proposed to be involved in autoregulation of repressor synthesis ( Fig. 3a and b) (28). One or more stoperator sites are located in the downstream genes and may act to reduce the level of transcription ( Fig. 3a and b). We observed some expression at the right end of the genomes near P left (Fig. S1d), which likely results from low levels of spontaneous lytic induction rather than lysogenic expression per se (31). This region is very highly expressed in early lytic growth (Fig. 3c), and the presence of multiple repressor binding sites, including at least one operator site at the P left promoter, downregulates this region during lysogeny ( Fig. 3d and Fig. S1e) (9).
Rep Trixie discriminates between stoperators of different L5 clade phages. Although the stoperator motifs of these phages are similar, they also vary at several positions, which may correspond to different repressor binding specificities (Fig. 2e). To demonstrate this, we overexpressed and purified Rep Trixie (Fig. S2a) and determined its binding affinities for a syntenic site in several phages (Fig. S2b). Rep Trixie has the strongest affinity for Trixie and RedRock stoperators (equilibrium dissociation constant [K D ] of ϳ6 to 7 nM, which is comparable to that of Rep L5 and Rep Bxb1 [9,26]), a somewhat reduced affinity for a Gladiator stoperator, and a very low affinity for Alma, Rockstar, and Peaches stoperators (Fig. S2c to e). Furthermore, the Rep Trixie binding affinity progressively diminishes when the substrate's sequence is incrementally changed from a Trixie to a Peaches stoperator ( Fig. S2e and f).
Taken together, these data show that these L5 clade phages have similar genome organizations and expression profiles, that the repressor system is likely the only influence on superinfection phenotypes, and that variations in the repressor and stoperator sequences are likely to contribute to differences in superinfection properties.
L5 clade phages exhibit diverse infection phenotypes. To determine how these diverse regulatory systems relate to superinfection immunity, we selected 19 phages from 7 subclusters across the L5 clade representing various degrees of genetic diversity based on their gene content, immunity system regulatory elements, and prophage inheritance strategies (Table 1). Lysogens were generated with each phage as well as with Dreamboat (Subcluster A1) as a heterotypic control. Superinfection immunity assays were performed against these lysogens using a variety of phages, including the parent temperate phages from which the lysogens were created, several naturally occurring L5 clade isolates that are obligately lytic derivatives of temperate parents, and several heterotypic Cluster A phages, including Peaches (Subcluster A4), Bxb1 (Subcluster A1), and Petruchio (Subcluster A1) ( Table 1).
The superinfection phenotypes are complex. Obvious examples of both homoimmunity and heteroimmunity are observed ( Fig. 1 and Fig. 4a and b), but there are also numerous examples of intermediate behaviors (including partially reduced infection efficiencies and changes in plaque size and morphology). These are illustrated by infection of several different lysogens either by phage L5 (Fig. 4c) or by several other phages (Fig. S3). In some instances, plaques increase in size (Fig. S3a), and this can make it appear as though the efficiency of infection is higher on a particular lysogen than on the nonlysogenic strain (Fig. S3d). Additionally, some reciprocal infection tests do not produce symmetric phenotypes (Fig. 1). For example, an L5 lysogen is sensitive to Trixie superinfection, but a Trixie lysogen completely defends against L5 infection (Fig. 4d).
To evaluate this complex set of phenotypes, we devised a scoring system (Table 2) to reflect the superinfection phenotype of each phage on each lysogen, relative to its infection of mc 2 155 (Fig. 4c) (see Materials and Methods). The infection score (⌱) for each assay reflects the range of phenotypes from complete immunity (in which no plaques or spots are observed [⌱ ϭ 0]) to complete superinfection (in which the phenotypes on the lysogen and mc 2 155 are identical [⌱ ϭ 5]) as well as enhanced superinfection (⌱ ϭ 6) ( Table 2). Specific examples are illustrated in Fig. 4c and Fig. S3,  and a matrix of all scores is shown in Fig. 5a. There are three key features of this matrix. First, although there are examples of complete immunity (⌱ ϭ 0) or complete superinfection (⌱ ϭ 5), there are many examples of intermediate (⌱ ϭ 1 to 4) or enhanced (⌱ ϭ 6) phenotypes ( Fig. 1 and Fig. S4a). Second, the diverse phenotypes do not correlate with subcluster designations (Fig. 5a and Fig. S4b). Third, there are many examples of asymmetric phenotypes, in which the infection scores from reciprocal assays (Δ⌱) differ depending on which phage is superinfecting and which is defending (Fig. 1, Fig. 5a, and Fig. S4c).
To investigate the causative factors of the diverse phenotypes, we compared the infection scores (⌱) and the reciprocity of infection scores (Δ⌱) to several genomic metrics. The scores correlate with changes in whole-genome gene content and nucleotide sequence (Fig. S5a). The spectrum of phenotypes occurs among phage pairs regardless of prophage inheritance genes (i.e., integration or extrachromosomal replication) ( Fig. S5b and c). Interestingly, immunity scores correlate, albeit weakly, with changes in repressors or stoperator motifs (Fig. 5b) and more so than with other highly conserved genes (Fig. S5d). Also, pairwise correlations between phage defense or superinfection profiles decrease as stoperator motif distances increase (Fig. 5c). Taken together, these data suggest that the repressor-stoperator interactions play an important role in the immune phenotypes.
Diverse infection phenotypes are repressor mediated. To explore further the role of the repressors, we constructed a series of recombinant strains carrying the repressor genes and upstream regulatory regions of L5, StarStuff, Trixie, Gladiator, Et2Brutus, and DaVinci (Table 3), as described previously for L5 (24), and compared their infection profiles to those of their cognate lysogens (Fig. 6). In general, there is a close correlation between infection of the "cloned-repressor strain" (CRS) and its cognate lysogen (Fig. 6a to c), consistent with the interpretation that the repressor-operator/stoperator system is the primary determinant of these diverse phenotypes. This is in contrast to phages where additional defense systems also contribute to phage susceptibilities (1,(32)(33)(34).
Although the CRS and lysogen infection scores generally correlate, there are two types of notable departures. In general, a CRS confers weaker immunity than the analogous lysogen against the homotypic superinfecting phage. For example, Gladiator infection of its CRS forms clearings or tiny plaques at high titers, even though the lysogen shows complete immunity (Fig. 6a). This was also reported previously for L5 (24). In contrast, some CRSs confer stronger immunity than the analogous lysogen ( Fig. 6c), with the most discrepant phenotypes occurring at greater genetic distances of the stoperator motifs (Fig. 6d). The most notable difference between a Cluster A lysogen and a cognate CRS is the presence of as many as 30 stoperator sites. Although the roles of these sites during superinfection immunity are unclear, the relative affinities of the repressors for the stoperator and operator sites of both the lysogen and the infecting phage may be important. We note that for DaVinci, a CRS containing a larger DNA segment from the repressor locus restores homotypic immunity to that observed in the lysogen, which may be due to either low-level expression of the three genes downstream of the repressor, stabilized expression of the repressor itself, or the presence of an additional stoperator (Table 3 and Fig. S6).
Isolation and characterization of defense escape mutants. In general, the L5 clade phages described here show strong immunity to themselves. Virulent mutants that escape homotypic immunity are not observed and likely arise only at very low frequencies, similar to phages (35), P22 (13), and P1 (14). However, among infections d Type of mutant phage, including an unintentional isolate, a recombinant by bacteriophage recombineering of electroporated DNA (BRED), or a defense escape mutant (DEM), from either the indicated lysogen (Lys) or cloned-repressor strain (CRS). e Mutations relative to the indicated parent phage are reported as top-strand genomic coordinates in the parent phage genome sequence from the Actinobacteriophage_1321 database and include point mutations (pm), insertions (ins) (with coordinates indicating nucleotides flanking the insertion), deletions (Δ) (with coordinates indicating the first and last nucleotides of the deleted region), and recombinations (rec) (with coordinates indicating regions of crossover within each genome). The mutational impacts on select genes and proteins (such as the rep locus) for mutants or natural isolates are indicated in parentheses, including sense mutations, missense mutations, nonsense mutations (*), frameshifts (fs) (with the first amino acid impacted), and complete gene deletions (Δ). of mesotypic lysogens and CRSs, it is common to observe reduced efficiency of plating and individual mutant plaques at high titers (Table S1). These potentially represent mutational pathways to escape immunity and evolve new immune specificities. We thus isolated, sequenced, and characterized nine defense escape mutants (DEMs) that escape defense from six different lysogens or CRSs ( Table 1).
The DEMs have acquired different types of mutations (Table 1 and Fig. 7a and b). An L5 mutant, phiTM41, escapes a Trixie lysogen after acquiring a single missense mutation in the first coding sequence downstream of P left , gene 89 ( Fig. 4d and Fig. 7a). More substantial deletions are incurred by Pioneer and EagleEye mutants, phiTM35 and phiTM36, to escape each other's lysogen ( Fig. 7a and Fig. S7a). The most dramatic mutation is observed in phiTM42, isolated from a RedRock infection of a Trixie lysogen (Fig. 7b). This DEM is a recombinant hybrid of the two phages, in which Trixie has lost the rightmost ϳ10 kb of its genome (including P left and rep) and has acquired the analogous locus from RedRock via two recombination events. Over 4 kb of the RedRock fragment has been deleted, and rep has acquired a 1-bp insertion. Unlike lambdoid virulent mutants (7), all DEMs (except for phiTM41) contain a mutation at the rep locus, and none of them involve mutations within operators ( Fig. 7a and b and Fig. S7b). Thus, they may escape immunity by utilizing different pathways compared to lambdoid phages.
Defense escape mutants exhibit different degrees of virulence. We next examined how the escape mutations impact virulence against other related systems. We compared the infection strengths between each DEM and parent phage across a panel of strains. DEMs nearly always exhibit infection strengths equal to or greater than those of their parent phages on both lysogens and CRSs (Fig. 7c). Similar to naturally occurring temperate phages, DEMs occasionally exhibit increased infection on lysogens compared to CRSs (Fig. 7d and e). However, although DEMs have escaped one immunity system, they are not able to escape all immunity systems (Fig. 7f). Instead, different degrees of virulence are observed, and we highlight these distinctions below with several examples.
The expanded host range (or "virulence specificity") is variable, as observed with phiTM41 and phiTM42. The mutation in phiTM41 confers narrow mesotypic virulence, as it does not impact infection of any lysogens other than Trixie ( Fig. 4d and see Fig.  10c). phiTM41 is the only DEM with no mutation in rep, its plaque morphology is not substantially impacted, and it remains temperate (Fig. S8f). phiTM41 lysogens exhibit the same defense profile as an L5 lysogen ( Fig. 4d and see Fig. 10c), indicating that the missense mutation in gp89 (which has no known function) abolishes the asymmetric infection observed between Trixie and L5 without altering other infection or defense capabilities. In contrast, phiTM42 is obligately lytic and exhibits broad homotypic and mesotypic virulence; it escapes immunity from both parent phages (Trixie and RedRock) and a Trixie CRS (Fig. 8a) as well as every other lysogen tested (Fig. S8a). However, Spots of lysis at highest 1-2 titers but no plaques 2 Superinfecting phage produces plaques with an efficiency of plating of less than ϳ10 Ϫ3 -10 Ϫ4 or spots of lysis at highest 3 titers but no plaques 3 Superinfecting phage produces plaques with an efficiency of plating from 10 Ϫ1 -10 Ϫ3 or spots of lysis at highest 4-5 titers but no plaques 4 Superinfecting phage produces plaques with an efficiency of plating of 1, but spots/plaques exhibit increased turbidity or reduced size compared to infection of mc 2 155 5 Superinfecting phage produces plaques with an efficiency of plating of 1, and there is no phenotypic difference compared to infection of mc 2 155 6 Superinfecting phage produces plaques with an efficiency of plating of 1, but spots/plaques exhibit reduced turbidity or increased size compared to infection of mc 2 155 Evolution of Phage Superinfection Immunity ® phiTM42 plaques are noticeably smaller on Trixie and RedRock lysogens than on mc 2 155, suggesting that some degree of superinfection inhibition remains (Fig. 8a).
Mutations may confer nuanced virulence specificities, as observed with Et2Brutus escape mutants phiTM39 and phiTM40 isolated on L5 and Trixie lysogens, respectively. phiTM40 has an A38V substitution in the rep DNA-binding domain, whereas phiTM39 has a nonsense mutation at codon 102 of rep (although both have additional mutations which could influence their phenotypes [ Table 1]). Both phages infect the CRS strains cognate to the lysogens on which they were isolated (Fig. 8b), but phiTM39 has a more relaxed specificity than phiTM40, exhibiting stronger infection of several strains (Fig. S8b), including an L5 lysogen (Fig. 8b). It even produces tiny, faint plaques on the homotypic Et2Brutus lysogen (Fig. 8b). Thus, the type of repressor mutation alters the superinfecting phage's behavior, depending on the nature of the lysogen being infected.
An additional example of the complexities of the escape phenotypes is illustrated by DEMs phiTM46 and phiTM47 (Table 1), which are derivatives of the closely related phages DaVinci and Gladiator, respectively, and have similar stoperator motifs (Fig. 2e and Fig. S8c). Both phiTM46 and phiTM47 have a 1-bp insertion within rep that presumably generates a truncated, inactive product ( Fig. 7a and Fig. S7b). However, the infection properties of these DEMs depart from those of their parents in distinct ways (compare infection profiles in Fig. S8d), suggesting that features other than the repressor per se influence these behaviors. One plausible explanation is that variation among the individual stoperator sites plays an important role.
An especially striking phenotype is exhibited by the DEM phiTM38. phiTM38 is a derivative of phage D29, an obligately lytic phage due to a deletion of the 5= end of the repressor gene (Fig. S7b) (31). Unlike its parent, phiTM38 efficiently escapes immunity of the homotypic StarStuff lysogen and CRS (Fig. 8c) as well as the mesotypic Et2Brutus lysogen on which it was isolated (Table 1). In contrast, its infection profile on other lysogens is unchanged (Fig. S8e), as illustrated with the mesotypic Larenn lysogen  Evolution of Phage Superinfection Immunity ® (Fig. 8c). Surprisingly, the single point mutation in phiTM38 maps to the remaining 3= segment of the inactive repressor gene (Fig. S7b). The simplest explanation is that this influences the expression of the downstream genes during superinfection and that these also play a role in the complex immunity relationships. Both the rep DNA-binding domain and C terminus confer immune specificity. The immunity repressors of lambdoid phages contain a helix-turn-helix domain near the N terminus responsible for DNA binding and a C-terminal domain responsible for dimerization that impacts sequence specificity (36)(37)(38). Cluster A Rep similarly has an N-terminal helix-turn-helix DNA-binding domain (16,24) and a distinct C-terminal region (27). Nucleotide sequence alignment of the immunity repressors from the L5 clade suggests that these two regions are under markedly different evolutionary pressures (Fig. S7b), with greater diversity in the C-terminal region than in the N-terminal region (Fig. 9a). Surprisingly, differences in the divergent C-terminal regions correlate with the immunity phenotypes rather than differences in the N-terminal DNA-binding domains (Fig. 9b) (Fig. 9c), but their stoperator motifs are distinct ( Fig. 2e and Fig. 4c), and they exhibit asymmetric superinfection phenotypes on lysogens and CRSs (Fig. 9d). Thus, both the N-terminal DNA-binding domain and the C-terminal region influence Cluster A Rep immune specificity.
An engineered L5 mutant exhibits narrow homotypic virulence. As part of a study of repressor function, we constructed two derivatives of phage L5 in which either a FLAG (phiTM6) or a hemagglutinin (HA) (phiTM1) tag is added to the C terminus of the repressor (Fig. 10a). These derivatives exhibit only subtle differences in plaque morphology (Fig. S8f), and they retain the ability to lysogenize. Their infection and immunity profiles are similar to those of their L5 parent (Fig. 10b and c and Fig. S8g and  h). During purification of a phiTM1 lysogen, a homotypic virulent mutant derivative, phiTM4, was isolated. phiTM4 has acquired a point mutation within gene 70, a gene of unknown function located immediately downstream of rep (Table 1 and Fig. 10a). The point mutation changes the last amino acid in gp70, but it also occurs within a stoperator site, although this site deviates from the consensus and is not bound by Rep in vitro (9). phiTM4 superinfects lysogens of all L5 derivatives as well as a StarStuff lysogen and L5 and StarStuff cognate CRSs (Fig. 10b and c). However, it remains unable to superinfect lysogens generated from other phages, including its closest relative, Serenity ( Fig. 10b to d). The single point mutation confers the narrowest homotypic virulence observed: phiTM4 escapes homotypic immunity but remains subject to closely related immunity systems.
Et2Brutus relatives exhibit nonlinear immune specificity evolution. To understand the evolutionary history of immune specificities, we compared superinfection phenotypes within a phylogenetic context. A phylogeny was constructed from all rep nucleotide sequences in the L5 clade, and a subclade representing several subclusters and exhibiting a robust tree topology was evaluated (Fig. 11a). An Et2Brutus lysogen and CRS exhibit immunity to Et2Brutus and Mulciber (both Subcluster A11). They exhibit sensitivity to other phages across this subclade, including DaVinci (Subcluster A6), but they are immune to Gladiator (also Subcluster A6). These immunity patterns are not congruent with the phylogeny, suggesting that immune specificities have switched more than once (Fig. 11a). Escape from Et2Brutus immunity may have emerged in a distant ancestor near the root of the tree, followed by immunity reemerging among some Subcluster A6 phages. Alternatively, escape from immunity may have independently emerged multiple times across this subclade.
Function and evolution of other Cluster A phages. There are over 200 Cluster A phages outside the L5-related clade. The majority of these phages also contain syntenically positioned immunity repressors and predicted stoperators (data not shown). A Bxb1 (Subcluster A1) CRS, constructed analogously to the L5 clade CRSs, is immune to Bxb1 superinfection, and a Bxb1 mutant (phiTM45) escapes this CRS after acquiring a single nonsense mutation in rep (Table 1 and Table S1). Therefore, the function and evolution of the immunity system within L5 clade phages may extend to all Cluster A phages.

DISCUSSION
Classical models of superinfection, immunity, virulence, and the evolution of new immune specificities were primarily developed with limited collections of coliphages related to , P22, and P1 (7,14). However, the genetic diversity among Cluster A phages illustrates how an immunity system can gradually evolve into homologous, mesotypic circuits with regulatory elements that exhibit a spectrum of interactions (Fig. 1). Interactions between mesotypic systems generate superinfection immunity patterns that are not necessarily binary or symmetric. In this more complex genetic environment, virulence and specificity can be shaped by both homotypic and mesotypic phages (Fig. 1). For instance, the mutations acquired in phiTM38 and phiTM39 conferring escape from mesotypic immunity systems also enable homotypic virulence. Additionally, there are different degrees of virulence within a mesoimmunity group. Both phiTM38 and phiTM4 exhibit narrow homotypic virulence, with little to no impact on escape from mesotypic systems (Fig. 11b). phiTM39 exhibits weak homotypic virulence but enhanced mesotypic virulence. phiTM41 exhibits narrow mesotypic virulence, while phiTM42 exhibits broad homotypic and mesotypic virulence (Fig. 11b). Thus, evolutionary divergence between repressor-mediated immune specificities may not be a linear trajectory toward heteroimmunity. Instead, genetic interactions with homotypic and mesotypic phages may result in a meandering evolutionary path in which homoimmunity, mesoimmunity, and heteroimmunity emerge more than once (Fig. 11c).
Homotypic virulence can occur by disrupting interactions between the repressor and cognate binding sites, as observed for and P22 (7), or by disrupting interactions between factors at secondary immunity loci, as seen in P1 and P7 (8). In contrast to these systems, the majority of Cluster A DEMs have acquired mutations that inactivate the repressor, and none of them (with the possible exception of phiTM4) contain a mutation within an operator or stoperator. Rep is the only identified transcriptional regulator involved in initiating and maintaining lysogeny for Cluster A phages (24). Branch support values reflect data from an approximate likelihood ratio test. (Right) Immunity assays involving representative phages from across the phylogeny (dashed lines) against mc 2 155, an Et2Brutus lysogen, and CRSs (pMH94, empty vector; pTM33, Et2Brutus). Peaches serves as a heterotypic control and is not represented in the phylogeny. (b) Diagram summarizing three different types of virulence observed among DEMs, defined by their ability to superinfect lysogens carrying prophages across a genetic spectrum of immune specificities, using a color spectrum as in Fig. 5a. (c) Model of immune specificity evolution. Circles represent phage genomes on a theoretical landscape of genetic diversity, and distance from the reference phage (centered circle) represents increasing genetic diversity of immunity system regulatory elements. Circles are shaded based on reciprocal superinfection immunity phenotypes relative to the reference phage and range from homoimmunity (black) to heteroimmunity (white). Genetic relationships to selected phages across the genetic landscape (red arrows) highlight that homoimmunity, mesoimmunity, and heteroimmunity may emerge multiple times as immunity systems diverge.
Evolution of Phage Superinfection Immunity ® However, we do not fully understand how the Cluster A immunity system functions, and other factors may be involved, similar to systems in P1, P7, and N15. The diverse types of mutations in phiTM39, phiTM40, phiTM41, and phiTM42 conferring escape from a Trixie lysogen may be targeting different aspects of the immunity system. phiTM39 and phiTM40 may escape with a modified Rep, while phiTM41 escapes with a modified gp89. The dramatic recombination in phiTM42 enabling escape from all lysogens may combine discordant regulatory elements in Trixie and RedRock that no individual prophage is able to properly regulate at the same time. Meanwhile, the mutation in phiTM4 appears to disrupt a very specific interaction present within L5, derivatives of L5, and StarStuff, such that it does not disrupt interactions within other prophages.
The diverse range of infection phenotypes observed on lysogens or CRSs could be caused by several factors related to Rep expression and specificity. Immunity asymmetry may be caused by repressors with sufficiently similar, but distinct, binding affinities or specificities (39), combined with subtle differences in the sequences and positions of stoperators, such that only one of the repressors is able to prevent lytic gene expression in both genomes. Gradual fading of spot dilutions could reflect weak affinity or specificity of the prophage's Rep for stoperators in the superinfecting phage genome that is sufficient to interfere with, but not completely defend against, superinfection and lytic growth, resulting in smaller or more turbid plaques and spots. Enlarged spot dilutions or plaques (suggesting enhanced infection) could reflect negative interactions between the prophage's and the superinfecting phage's regulatory elements that reduce the efficiency of lysogenization without inhibiting lytic growth. Differences in CRS and lysogen immunity may result in differences in Rep expression levels or in the number of available stoperators. Other factors may be involved in this immunity system that we have not yet identified, and the evolution of these factors may also contribute to the diverse infection phenotypes.
Mesoimmunity groups are likely to be common in nature. There are several groups of actinobacteriophages infecting Gordonia, Rhodococcus, and Streptomyces hosts that harbor immunity systems similar to those of the Cluster A mycobacteriophages (40,41). Additionally, we note that there are examples of asymmetric and incomplete infection among phages related to (42) and P2 (15). The complex pathways of immunity evolution are likely common features of temperate phages although readily apparent only when comparing large groups of related temperate phages known to infect a common host bacterium.

MATERIALS AND METHODS
Phamerator database construction. The database Actinobacteriophage_1321 was created using Phamerator (43), consisting of 1,305 manually annotated genomes of actinobacteriophages isolated from the environment and 16 engineered or isolated mutants as described below. Genes are grouped into phamilies ("phams") based on amino acid sequence similarity using kClust implemented in the Phamerator pipeline (23). The database is available online (http://phamerator.webfactional.com/databases _Hatfull).
Identification and analysis of stoperator sequences. Stoperator sequences were automatically identified in all Cluster A genomes using MEME (44), using the following parameters: site distribution of any number of repetitions, maximum of 2 motifs, motif length of 12 to 16 bp, consisting of 10 to 50 sites, and derived from both strands. The motif that most closely resembled empirically determined L5 and Bxb1 stoperator sites was selected. All sites representing each motif were converted to the sense strand and manually aligned in Excel. Motif logos representing the aligned sequences were created with WebLogo (45). Stoperator sequences were compared in R using the Biostrings and TFBSTools packages (46). Position weight matrices (PWMs) of the core 13-bp sequence were created using the PFMatrix and toPWM functions, using the log2probratio method and default values for background and pseudocount settings. Pairwise PWM-normalized Euclidean distances were computed using the PWMSimilarity function, and larger distances represent more dissimilar PWMs (46,47). Similar to whole-genome distance metrics (described below), stoperator motif distances can be computed between all L5 clade phages. Stoperators were determined to be oriented in the direction of transcription (syn oriented) if they were located on the top strand to the left of the genome center or on the bottom strand to the right of the genome center. The center of the genome was defined as the coordinates of the integrase (for integrating phages) or parA (for extrachromosomal phages) gene. To generate genomic distributions of stoperators in L5 clade phages, coordinates of all stoperators in each phage were adjusted relative to the coordinates of the genomic feature of interest in that specific phage, and histograms were created using adjusted coordinates for all L5 clade phages.
Computation of whole-genome distances. Pairwise nucleotide similarity and gene content dissimilarity between all phage genomes were computed, as previously described (48). For pairs of phages, gene content dissimilarity ranges from 0 (all gene phams are identical) to 1 (no gene phams are identical), and nucleotide distance ranges from 0 (identical sequence) to 0.5 (unrelated sequence).
Genetic distance of specific Cluster A genes. Amino acid sequences for 336 full-length homologs of the Cluster A immunity repressor present in the database (represented by phams 3247, 38916, and 38877) were aligned using MAFFT (49). The alignment was manually trimmed at the N terminus in SeaView (50) and split into N-terminal and C-terminal regions as previously reported (27). Uncorrected distances between taxa in the full-length, N-terminal, and C-terminal alignments were computed using the EMBOSS distmat tool with no gap weight and reported as a normalized distance reflecting the number of substitutions per 100 amino acids (https://www.ebi.ac.uk/Tools/emboss/). The 20-amino-acid helix-turn-helix domain was identified in all taxa from the MAFFT alignment based on previous reports (16,26). Uncorrected distances for full-length proteins of 311 homologs of the Cas4-family gene (pham 29663), 306 homologs of Endonuclease VII (EndoVII) (pham 39443), 311 homologs of DNA polymerase (pham 39585), and 311 homologs of Portal (pham 38438) genes present in Cluster A phages were computed in the same way. Hamming distances between helix-turn-helix domains were computed using the stringdist R package. Unlike whole-genome distances and stoperator motif distances, gene-specific distances are limited to phages that carry a homolog of the gene of interest.
Repressor nucleotide alignment and phylogeny. Nucleotide sequences for 79 immunity repressors from L5 clade phages were aligned by codon using webPRANK (51), and a phylogenetic tree was constructed using maximum likelihood in SeaView (50) and annotated using Evolview (52).
Preparation of phage lysates and lysogens. A diverse set of phages was selected for immunity assays, representing multiple subclusters, utilizing different prophage inheritance strategies (integration or extrachromosomal partitioning), and carrying complete or mutant repressor genes ( Table 1). All phages used for immunity assays were plaque purified at least twice. Lysates were expanded for one round from a plaque pick by plating phage with mc 2 155, incubation at 37°C for 24 to 36 h, incubation with 5 ml phage buffer at room temperature for 4 to 5 h, and filtering with a 0.22-m filter. Lysates were confirmed to have the expected phage by PCR using primers that amplify near the right genome terminus, in which there is substantial sequence diversity among Cluster A phages, to generate phage-specific amplicons. Lysogens for many purified phages were created by spotting high-titer phage lysates on a lawn of mc 2 155, picking cells from the center of the spot after 3 to 7 days, and performing colony purification at least two times. Strains were confirmed as lysogens by PCR using the same primers as those used for lysate confirmation, by verifying that cells exhibit spontaneous phage release when spotted onto a lawn of mc 2 155, and by verifying that the strain is immune to infection from the parent phage. Lysogens for phages Echild (30), Journey13, and Piro94 could not be generated ( Table 1). Lysogenization of some phages was not tested (Table 1).
Repressor overexpression and EMSAs. Rep Trixie was amplified from the Trixie genome (coordinates 45599 to 46174) with primers oTM13 and oTM14 and cloned into the expression vector pET21a using the NdeI and HindIII sites to create the plasmid pTM1, which carries Rep Trixie C-terminally tagged with His and a short linker (KLAAALEHHHHHH). pTM1 was transformed into NEB5␣ cells. Sequence-verified plasmid constructs were transformed into BL21 Star(DE3) cells, and single colonies were grown in LB medium supplemented with carbenicillin. Repressor expression was induced with 1 mM isopropyl-␤-Dthiogalactopyranoside (IPTG) for 3 h, and cells were lysed by resuspension in lysis buffer (50 mM Tris [pH 8.0], 300 mM NaCl, 10% glycerol), treatment with 1 mg/ml lysozyme for 30 min on ice, and light sonication (54). C-terminally His-tagged Rep Trixie was purified using a nickel-nitrilotriacetic acid (NTA) matrix, dialyzed overnight with storage buffer (50 mM Tris [pH 8.0], 150 mM NaCl, 0.1 mM EDTA, 0.1 mM dithiothreitol [DTT], 50% glycerol), and quantified at ϳ1 mg/ml using a NanoDrop instrument (Thermo Fisher). DNA substrates for electrophoretic mobility shift assays (EMSAs) were designed to be 30 bp long, consisting of a 13-bp stoperator sequence flanked by 8 to 9 bp of sequence. Complementary 30-bp oligonucleotides were synthesized, radiolabeled at the 5= end with ␥-32 P, and annealed (54). Oligonucleotides for each substrate in Fig. S2 in the supplemental materials are as follows: oTM21 and oTM22 for Alma, oTM23 and oTM24 for Gladiator, oTM17 and oTM18 for Peaches, oTM31 and oTM32 for RedRock, oTM19 and oTM20 for Rockstar, oTM33 and oTM34 for Trixie, oTM29 and oTM30 for the L5 gene 31 negative control, oTM43 and oTM44 for C 9 , oTM41 and oTM42 for C 9 G 10 , oTM39 and oTM40 for C 9 C 11 , oTM37 and oTM38 for C 9 A 12 , oTM49 and oTM50 for C 9 G 10 C 11 , oTM47 and oTM48 for C 9 G 10 A 12 , oTM45 and oTM46 for C 9 C 11 A 12 , and oTM35 and oTM36 for C 9 G 10 C 11 A 12 (Table S2). The sequences of the 30-bp substrates to test syntenic stoperator sites, including the L5 gene 31 negative control, are derived directly from the genome sequence. For the 30-bp substrates in which the Trixie stoperator site is progressively converted to a Peaches stoperator site, the variable 13-bp sequence is flanked by invariable 8 to 9 bp derived from the Trixie substrate. EMSAs were performed with serially diluted Rep, which was electrophoresed on an 8% polyacrylamide gel and imaged, as previously described (54). The K D for each substrate was calculated with nonlinear regression in Prism software using the one-site-specific binding option and least-squares fit.
The immunity repressor from DaVinci was cloned into the extrachromosomal multicopy vector pJV44 (57). The locus was amplified by PCR using primers containing XbaI and HindIII sites. For constructs containing only rep, analogous to the integrated repressor constructs described above, a segment from DaVinci (coordinates 42748 to 43932) was amplified using primers oTM257 and oTM265. For the construct containing the extended repressor locus (from rep to gene 73), a segment from DaVinci (coordinates 41377 to 43932) was amplified using primers oTM257 and oTM258. pJV44 and the amplicons were digested with XbaI and HindIII, gel purified and cleaned up using the NucleoSpin gel extraction kit, ligated with T4 DNA ligase, and transformed into NEB5␣ cells. Since this cloning strategy removes the hsp60 promoter in pJV44, a religated vector backbone that lacks the hsp60 promoter was constructed as an empty vector control using an amplicon directly generated from self-amplifying primers (oTM266 and oTM267). The following plasmids were constructed: pTM44 (empty vector), pTM48 (DaVinci rep), and pTM51 (DaVinci rep-73) ( Table 3). Sequence-verified plasmid constructs were transformed into electrocompetent mc 2 155 cells, and positive transformants were selected using Middlebrook 7H10 medium supplemented with gentamicin and clonally purified.
L5 repressor tagging. Rep L5 was C-terminally tagged in vivo with either a 27-bp HA (phiTM1 [TACCCATACGACGTCCCAGACTACGCT]) or a 24-bp FLAG (phiTM6 [GACTACAAGGACGACGATGACAAG]) (58) sequence using recombineering with an L5 lysogen, similar to previous reports (59). The FLAG oligonucleotide (oTM51) and HA oligonucleotide (oTM52) were PCR amplified using primers oTM53 and oTM54 to create ϳ200-bp recombineering substrates that overlap the 3= end of gene 71 and that contain the tag sequence (Table S2). Amplicons were purified using the GeneJet PCR purification kit, and the DNA was cotransformed with pJV44 into electrocompetent mc 2 155(L5)pJV53, as previously described (59). Successful pJV44 transformants were selected on Middlebrook 7H10 medium supplemented with gentamicin, and successful L5 recombinants were subsequently identified by PCR. Positive recombinants were clonally purified and patched onto a lawn of mc 2 155, and spontaneously released phages were picked and plaque purified. From one of the HA-tagged recombinant phage picks, a spontaneous mutation was acquired (phiTM4).
Immunity assays. Fresh 10-fold serial dilutions of each phage lysate were generated using phage buffer (10 mM Tris [pH 7.5], 10 mM MgSO 4 , 68 mM NaCl, 1 mM CaCl 2 ), and 3 l of each dilution was spotted onto a top agar layer of the indicated strain. For immunity tests involving lysogens, strains were plated in Middlebrook 7H9 top agar on Middlebrook 7H10 medium. For immunity tests involving strains carrying pMH94-derived cloned-repressor constructs, strains were plated in Middlebrook 7H9 plus kanamycin top agar on Mycobacteria 7H11 plus kanamycin medium. For immunity tests involving strains carrying pJV44-derived cloned-repressor constructs, strains were plated in Middlebrook 7H9 plus gentamicin top agar on Middlebrook 7H10 plus gentamicin medium. Lysates were always spotted onto an accompanying nonlysogen or empty vector control strain (mc 2 155, mc 2 155pMH94, or mc 2 155pTM44) for reference. Plates were incubated at 37°C for 3 days and photographed with ImageLab using a 1.5-to 2.0-s exposure. Individual assays were quantitatively scored by comparing the qualitative infection phenotypes of the phage on the strain of interest to those for the control strain, including efficiency of plating, turbidity, the presence of plaques, and plaque size (Table S1). Results were processed in R using custom scripts. More than 3,000 immunity assays were performed and manually scored, representing 1,050 unique comparisons, 239 reciprocal comparisons, and 164 lysogen-CRS paired comparisons. The bacterial densities for each culture used are approximately 1 ϫ 10 9 to 4 ϫ 10 9 CFU per 1 ml of culture. We estimate that spots from the highest titers of phage represent phage infections that occur (on average) with a multiplicity of infection (MOI) of between 2 and 20. Therefore, the infection phenotypes for the majority of spot dilutions represent phage infections at an MOI of less than 1.
Isolation of defense escape mutants. Mutant phages able to escape prophage or cloned-repressor defense were isolated by picking plaques from immunity assays in which the challenging phage exhibits a substantial reduction in efficiency of plating, performing plaque purification at least twice on mc 2 155, and confirming the ability to infect the original strain (Table 1). DNA was extracted from both the DEM and parent phage lysates and sequenced as previously described (1). Mutations were identified by whole-genome alignment. In some cases, the parent phage genome contained a mutation(s) relative to the published sequence. Only mutations that are present in the DEM compared to the parent are reported.
R data analysis. Infection data were analyzed and visualized in RStudio using custom scripts with the reshape2 and stringdist packages (Table S1). More than 65% of unique comparisons were measured with two or more replicate assays, and replicate infection scores were averaged. Although the infection score can vary between replicates, more than 80% of comparisons with two or more replicates exhibit a range of infection scores lower than 2. All R 2 correlations between genetic elements and immunity phenotypes were determined with linear regression only using intra-L5 clade comparisons, unless otherwise indicated, using the lm function.
Data availability. Raw fastq data have been deposited in the Gene Expression Omnibus (GEO) under accession number GSE123612. The R code used for data analyses is available upon request.

SUPPLEMENTAL MATERIAL
Supplemental material for this article may be found at https://doi.org/10.1128/mBio .00971-19.  We thank Rebecca Garlena for genome sequencing and assembly of phage mutants and Valerie Villanueva and Pallavi Ghosh for assistance with EMSAs.
T.N.M. performed the experiments and analyses. T.N.M. and G.F.H. designed the project, interpreted the results, and wrote the paper.
We declare no competing interests.