A competence-regulated toxin-antitoxin system in Haemophilus influenzae

Natural competence allows bacteria to respond to environmental and nutritional cues by taking up free DNA from their surroundings, thus gaining both nutrients and genetic information. In the Gram-negative bacterium Haemophilus influenzae, the genes needed for DNA uptake are induced by the CRP and Sxy transcription factors in response to lack of preferred carbon sources and nucleotide precursors. Here we show that one of these genes, HI0659, encodes the antitoxin of a competence-regulated toxin-antitoxin operon (‘toxTA’), likely acquired by horizontal gene transfer from a Streptococcus species. Deletion of the putative toxin (HI0660) restores uptake to the antitoxin mutant. The full toxTA operon was present in only 17 of the 181 strains we examined; complete deletion was seen in 22 strains and deletions removing parts of the toxin gene in 142 others. In addition to the expected Sxy- and CRP-dependent-competence promoter, HI0659/660 transcript analysis using RNA-seq identified an internal antitoxin-repressed promoter whose transcription starts within toxT and will yield nonfunctional protein. We propose that the most likely effect of unopposed toxin expression is non-specific cleavage of mRNAs and arrest or death of competent cells in the culture. Although the high frequency of toxT and toxTA deletions suggests that this competence-regulated toxin-antitoxin system may be mildly deleterious, it could also facilitate downregulation of protein synthesis and recycling of nucleotides under starvation conditions. Although our analyses were focused on the effects of toxTA, the RNA-seq dataset will be a useful resource for further investigations into competence regulation.

Natural competence allows bacteria to respond to environmental and nutritional cues by taking up free DNA from their surroundings, thus gaining both nutrients and genetic information. In the Gram-negative bacterium Haemophilus influenzae, the genes needed for DNA uptake are induced by the CRP and Sxy transcription factors in response to lack of preferred carbon sources and nucleotide precursors. Here we show that one of these genes, HI0659, encodes the antitoxin of a competence-regulated toxin-antitoxin operon ('toxTA'), likely acquired by horizontal gene transfer from a Streptococcus species. Deletion of the putative toxin (HI0660) restores uptake to the antitoxin mutant. The full toxTA operon was present in only 17 of the 181 strains we examined; complete deletion was seen in 22 strains and deletions removing parts of the toxin gene in 142 others. In addition to the expected Sxy-and CRP-dependent-competence promoter, HI0659/660 transcript analysis using RNA-seq identified an internal antitoxin-repressed promoter whose transcription starts within toxT and will yield nonfunctional protein. We propose that the most likely effect of unopposed toxin expression is non-specific cleavage of mRNAs and arrest or death of competent cells in the culture. Although the high frequency of toxT and toxTA deletions suggests that this competence-regulated toxin-antitoxin system may be mildly deleterious, it could also facilitate downregulation of protein synthesis and recycling of nucleotides under starvation conditions. Although our analyses were focused on the effects of toxTA, the RNA-seq dataset will be a useful resource for further investigations into competence regulation. PLOS

Introduction
Bacterial toxin-antitoxin gene pairs were originally discovered on plasmids, where they function to promote plasmid persistence by killing any daughter cells that have not inherited the plasmid. Typically, one gene of the pair encodes a relatively stable toxic protein that blocks cell growth, and the other encodes a labile antitoxin (RNA or protein) that blocks the toxin's activity and limits its transcription [1,2]. Toxin-antitoxin gene pairs have also been discovered on many bacterial chromosomes, where they are thought to be relatively recent introductions that in some cases have been co-opted to regulate cellular functions or provide other benefits [3]. Here we describe one such system, which is induced in naturally competent cells and whose unopposed toxin completely prevents DNA uptake and transformation. Many bacteria can become naturally competent, able to take up DNA from their surroundings and-when sequence similarity allows-recombine it into their genomes [4,5,6]. In most species, DNA uptake is tightly controlled, with protein machinery specified by a set of co-regulated chromosomal genes induced in response to diverse cellular signals. Genes in the competence regulon encode not only components of the DNA uptake machinery that moves DNA across the outer membrane of the cell, but proteins that translocate DNA across the inner membrane, proteins that facilitate recombination, and proteins of unknown function. Haemophilus influenzae has an unusually small and well-defined competence regulon (26 genes in 13 operons) induced by signals of energy and nucleotide scarcity [7,8]. Induction of these genes begins in response to depletion of phosphotransferase sugars. The resulting rise in cyclic AMP (cAMP) activates the transcription factor CRP, and the CRP/cAMP complex then stimulates transcription of genes with canonical CRP-promoter elements (CRP-N genes). Most of these genes help the cell to use alternative carbon sources, but one encodes the competence-specific transcriptional activator Sxy. However, efficient translation of sxy mRNA occurs only when purine pools are also depleted [9,10]. If both signals are active, Sxy then acts with CRP at the promoters of competence genes, stimulating their transcription and leading to DNA uptake and natural transformation. These competence promoters are distinguished by the presence of 'CRP-S' sites (formerly called CRE sites), variants of standard CRP sites that depend on both CRP and Sxy for activation [11]. Development of competence thus requires the CRP/cAMP complex twice, first for sxy transcription (at its CRP-N promoter) and then for transcription of the competence genes (at their CRP-S promoters). As is common in competence systems, only some of the cells in the population become competent (typically 10-50%).
Of the fifteen Sxy-regulated H. influenzae genes needed for DNA uptake, all but one encode typical competence proteins-membrane-associated proteins homologous to known components of the Type IV pilus-based DNA uptake machinery present in nearly all known naturally competent species [5]. The one exception is HI0659, which instead encodes a predicted 98 amino acid cytoplasmic protein with no similarity to known DNA uptake proteins. It shares a competence-inducible CRP-S promoter with an upstream gene encoding another short cytoplasmic protein (HI0660, 119 aa) (Fig 1, top). Although a knockout of HI0659 eliminates detectable DNA uptake and transformation, a knockout of HI0660 has no effect [8].
Here we show that HI0660 and HI0659 comprise a horizontally transferred operon that encodes a toxin-antitoxin pair, and that expression of the toxin in the absence of the antitoxin completely prevents DNA uptake and transformation. Surprisingly, expression of this toxin has only modest effects on induction of competence genes and on cell growth and viability.

HI0659 and HI0660 show homology to type II toxin/antitoxin systems
Our original analyses of competence-induced genes did not identify any close homologs of HI0659 or HI0660 [7,8]. However subsequent database searches and examination of BLAST results revealed that these genes' products are homologous to genes annotated as belonging to Type II toxin/antitoxin families, which consists of a two-gene operon encoding the toxin and antitoxin components [2]. HI0660 is annotated on NCBI as a phage-related protein containing the COG4679 region found in members of the ParE toxin family. The Pfam and TAfinder databases assign the H. influenzae HI0660 protein to the ParE/RelE toxin superfamily, whose characterized members include both gyrase inhibitors and ribonucleases that arrest cell growth by cleaving mRNAs and other RNAs [2,12,13]. The HI0659 protein is annotated in NCBI as containing a helix-turn-helix-XRE DNA-binding domain, which is commonly found in promoterbinding antitoxins [14]. Structurally, the Phyre2 modelling software predicts that the best match of known structure for HI0660 is the HigB toxin of Streptococcus pneumoniae (100% confidence across 95% of the protein sequence), and the best match for HI0659 is the HigA antitoxin of Streptococcus pneumoniae (99.9% confidence across 92% of the protein sequence) [15].

HI0659 and HI0660 act as a toxin-antitoxin system that affects natural competence
If HI0660 and HI0659 do encode a toxin-antitoxin pair as suggested by in silico analyses, then ΔHI0659's DNA uptake defect would likely be caused by unopposed expression of a HI0660encoded toxin protein that prevents DNA uptake, and knocking out this toxin gene would restore competence to the HI0659 (antitoxin -) mutant. We tested this by constructing an HI0660/HI0659 double mutant (Fig 1) and examining its ability to be transformed with antibiotic-resistant chromosomal DNA (MAP7) compared to wild type or either single mutant. The double mutant had normal transformation (Fig 2, black bars), showing that mutation of HI0660 suppresses the competence defect of an HI0659 mutant, and also that neither HI0660 nor HI0659 is directly needed for the development of competence. This supported the postulated antitoxin function of HI0660, so we named the HI0660 and HI0659 genes toxT (toxin) and toxA (antitoxin) respectively. using NCBI's tBLASTn function to search for homologs of a protein by querying translated nucleotide sequence databases; this function was use because toxin and antitoxin proteins are often missed in protein annotations due to their small size. Complete or partial toxTA operons were found at the same genomic location in most H. influenzae genomes and in the closely related H. haemolyticus (see below), but there were no recognizable homologs in most other bacteria, including most other members of the Pasteurellaceae. Instead, most identifiable homologs (with about 60% amino acid identity) were in a very distant group, the Firmicutes, mainly within the streptococci. Ninety-six of the top 100 tBLASTn hits to ToxT outside the Pasteurellaceae were to diverse Streptococcus species. This suggests that the toxTA operon may have been transferred from a Firmicute into a recent ancestor of H. influenzae and H. haemolyticus. When we excluded Streptococcus from the BLAST search, sporadic matches were found in a wide variety of other taxa. In addition, toxTA operons with about 50% identity were found in one other small distant Pasteurellacean subclade (Actinobacillus sensu stricto), and on two 11kb plasmids (pRGRH1858 and pRGFK1025) from an uncultured member of a rat gut microbiome and an uncultivated Selenomonas sp. The genes flanking the A. pleuropneumoniae homologs are unrelated to those flanking HI0659 and HI0660, and an A. pleuropneumoniae toxA knockout mutant had normal growth and competence ( Figure A in S1 File). The distribution of toxTA homologs across taxa is summarized in Fig 3A. To examine the potential for of gene transfer events in the two Pasteurellaceae sub-clades, we created an unrooted maximum likelihood phylogeny from the concatenated alignment of toxT and toxA amino acid sequences from selected species where both genes were found ( Fig  3B). There was 99% bootstrap support for a node which branches into the Haemophilus, Actinobacillus, and Streptococcus clades (which also group with the plasmid samples). In isolation,  this branching pattern could be consistent with a single Pasteurellacean origin of the toxinantitoxin pair, but the absence of homologs from all other Pasteurellaceae makes a single Pasteurellacean origin unlikely, since it would require multiple deletions in other Pasteurellacean subclades or a near-simultaneous within-clade lateral transfer. Since the Actinobacillus sequences showed higher identity to Streptococcus sequences than to the Haemophilus sequences, the two Pasteurallacean groups are more likely to have acquired their toxTA operons by independent lateral transfers, perhaps from Firmicutes. However, because we were unable to root our tree, we cannot unequivocally determine the presence or direction of any gene transfer events. We were unable to use synteny to determine whether the gene pair was gained in the same insertion event because in Actinobacillus species, the upstream and downstream genes flanking ToxTA in Haemophilus influenzae are not adjacent to one another nor to the Actinobacillus ToxTA homologs. Examination of flanking genes within the Streptococci was broadly consistent with inheritance of the ToxTA homologs from a common ancestor, but was not possible in all cases due to genomic rearrangements which separated the flanking genes.

Deletions in H. influenzae toxT are common
The presence of an intact and syntenic toxTA operon in H. haemolyticus shows that the presence of toxTA is the ancestral H. influenzae state. Of the 181 H. influenzae genome sequences available in public databases (described in Table B of S1 File), 162 had recognizable toxA sequences. All of these encoded full length ToxA proteins, but all except 24 had one of two common deletions affecting toxT. The extents of these deletions are shown by the grey bars at the bottom of Fig 4. The most common deletion (n = 97) removed 178 bp of toxT coding sequence but left both promoters intact. The second (n = 45) removed 306 bp of sequence including both toxTA promoters and the toxT start codon. The 19 genomes that lacked recognizable toxA sequences all had the same 1015 bp deletion removing both toxT and toxA but leaving the flanking genes intact. In place of the missing sequences were 87 bp with no highscoring BLAST alignments in GenBank. This collection of strains is from diverse times, locations and body sites, and the deletion distribution suggests that deletions inactivating toxT arise rarely but are favoured by selection.  genes is 0.106, slightly higher than the average of all genes with one copy per strain (0.088). The d N /d S ratio of 0.037 is lower than that of the average gene (0.243), indicating mild purifying selection on toxA. However, this may underestimate the strength of selection, since most toxAs lack functional toxT, no strains with toxT only were seen, and the operon may not be expressed due to upstream deletions. Both sequence divergence and the high frequency of toxT deletions agree with expectations for a toxin/antitoxin system whose antitoxin protects against a toxin that is at least mildly deleterious.
Variation in toxTA does not correlate with strain-specific variations in DNA uptake and transformation Maughan and Redfield [17] measured the ability of 34 H. influenzae strains to both take up DNA and become transformed, so we examined this data for correlations with the presence of toxTA in the 19 of these strains whose toxTA genotypes we were able to determine from genome assemblies (Table C of S1 File). All but one of the 19 strains had a complete toxA coding sequence but only five had intact toxTA operons. Of the rest, four had the large deletion that removed both toxTA promoters, nine had the smaller deletion internal to toxT, and one had the 1015 bp complete deletion. No obvious correlation was seen between the toxTA genotypes and the DNA uptake, transformation or growth phenotypes, but there was insufficient data for a high-powered analysis using allelic variation.

Growth and competence phenotypes of H. influenzae toxTA mutants
Earlier investigation of DNA uptake and transformation by the H. influenzae toxA knockout strain found that both were below the limit of detection (>100-fold reduction and >10 6 -fold reduction respectively) after the standard competence-inducing treatment [8] (see also Fig 2), although there was no apparent growth defect in rich medium. Figure B in S1 File confirms the non-transformability of ΔtoxA cells both in log phase growth and at intermediate time points during competence induction.
A simple explanation for this defect would be that unopposed ToxT prevents competence, when not opposed by ToxA, by killing or otherwise inactivating the (competence-induced) cells in which it is expressed. To detect effects of unopposed expression of toxT toxin, we analyzed growth rates of wildtype and ΔtoxA strains ( Fig 5) before, during, and after transfer to the competence-inducing starvation medium MIV. The first 60 min of Fig 5 show that unopposed expression of toxT toxin slightly slows exponential growth in rich medium. A more detailed analysis of growth of rich-medium cultures wells is provided in Figure C in S1 File, using 20 replicate wells of a BioScreen culture plate for each strain (see Methods for details); the growth defect of the ΔtoxA strain is barely detectable under this condition. This lack of a severe growth defect is not surprising; because the toxTA promoter is regulated by a CRP-S site, its expression (and thus ToxT production) might be limited to competent cells even in the absence of ToxA [7]. The grey-shaded portion of Fig 5 shows cell growth after transfer to MIV (samples between 70 min and 165 min). Cells transferred to MIV usually undergo only one or two doublings, and deletion of toxA delays this but does not eliminate it (see also Figure D in S1 File). Cells returned to rich medium from MIV might be delayed in their recovery, if unopposed toxin expression during competence development kills cells or halts growth. However, both strains had similar recovery kinetics after a fraction of their culture was returned to rich medium, although ΔtoxA cells again grew slightly slower than wildtype (175-230 min in Fig  5). Figure E in S1 File shows the OD 600 readings corresponding to the CFU/ml data shown in Fig 5. These agree well, showing that differences in cell growth can be explained by the differences in cell division.
Since cyclic AMP is required for normal induction of the competence genes, and addition of cAMP induces partial competence during exponential growth [18], we also tested the effect of cAMP on the ΔtoxA knockout. Addition of cAMP did not rescue its transformation defect (Figure F in S1 File), so failure to transform is not caused by defective cAMP production in the antitoxin mutant.
Since some chromosomal toxin-antitoxin systems have acquired beneficial roles in modulating cell growth [19], we also examined whether the absence of toxin changed growth and competence under various conditions. The grey line in Figure C in S1 File shows that, under BioScreen growth assay conditions, ΔtoxT's growth was indistinguishable from that of wildtype cells (blue line), and Fig 2 shows that its MIV-induced competence is also unchanged. Figure G in S1 File shows that the kinetics of ΔtoxT competence development and loss during growth in rich medium were also indistinguishable from wildtype. We conclude that ToxT's normal expression in cells expressing antitoxin does not detectably regulate growth or the development or loss of competence. Unfortunately, inferring the relationship between competence and growth is complicated because many cells in 'competent' cultures are unable to transform. In some species the non-transformable cells are known to have not induced their competence genes, but this has not been investigated for H. influenzae.

Transcriptional control of competence
Since these phenotypic analyses gave little evidence of MIV-specific toxicity or insight into the cause of the competence defect, we used RNA-seq to investigate how toxTA is regulated and how mutations affect transcript levels of competence genes. In these experiments, samples for RNA preparations were taken from three replicate cultures at four time points, first when cells were growing in log phase in the rich medium sBHI (t = 0), and then at 10, 30 and 100 minutes after each culture had been transferred to MIV. We first examined how competence induction in wildtype and regulatory-mutant cells changed transcript levels of genes known to be regulated by CRP and CRP+Sxy (CRP-N and CRP-S genes respectively). Fig 6 gives an overview of the results. The top row shows competence-induced changes in transcript abundances in wildtpe cells at 10, 30 and 100 minutes, and the lower rows show that some of these changes do not occur in Δcrp and Δsxy cells. Each coloured dot represents a gene, colour-coded by function. Its horizontal position indicates its transcript level in rich medium (T = 0) and its vertical position indicates how this level changed in MIV (top row-wildtype cells, lower rows, Δcrp and Δsxy cells). Thus in Fig 6A the higher positions of the dark blue dots (genes regulated by CRP-N sites) and the red diamond (the competence regulator sxy) indicate that they were strongly induced after 10 min in MIV. Induction of sxy was followed at 30 and 100 minutes by strong induction of the known competence-regulon genes (higher positions of CRP-S genes; light green dots) (Fig 6B and 6C). Note that transcript levels are relative; because of population heterogeneity they may underestimate the degree of induction or downregulation in the cells that go on to become competent. Consistent with prior studies [7], induction of all these genes was blocked by deletion of the crp gene (Fig 6D-6F), and induction of the competence regulon (CRP-S) genes was blocked by deletion of sxy (Fig 6G-6I). Analysis of other competence-associated changes in transcript levels in normal cells and in Δcrp and Δsxy mutants is provided in the Supplementary Materials.

Transcriptional control of toxTA
RNA-seq analysis confirmed that toxTA is regulated as a typical competence operon. In wildtype cells, baseline RNA-seq levels of toxT and toxA transcripts were very low during log phase in rich medium, with approximately tenfold induction after 30 minutes incubation in MIV (green lines and points in Fig 7 (toxT) and Figure H in S1 File (toxA). As expected, this increase was eliminated by knockouts of CRP and Sxy (brown and blue lines and points in Fig  7 and Figure H in S1 File). Like other CRP-S genes, both toxT and toxA were also strongly induced in rich medium by sxy and murE mutations known to cause hypercompetence [20,21] and by a newly identified hypercompetence mutation in rpoD (Figure I in S1 File). Thus the toxTA operon is regulated as a typical member of the competence regulon.
RNA-seq analysis also showed that the toxTA operon is regulated as a typical type II toxinantitoxin operon. In such operons, the antitoxin protein usually binds to the toxin protein, which protects cells from the toxin in two ways. First, antitoxin binding inactivates the toxin. Second, it also activates the antitoxin component as a repressor of the toxTA promoter [2,22]. H. influenzae ToxA has a HTH-XRE DNA-binding domain, which is commonly found in promoter-binding antitoxins [1,14], and the RNA-seq analysis in Fig 7 strongly suggests that it represses toxTA transcription. The ΔtoxA mutant retains an intact toxTA promoter and toxT coding sequence (see Fig 1); it had 9-fold increased baseline transcript levels of toxT in log phase cells (red line and points in Fig 7). Transcript levels increased further during A competence-regulated toxin-antitoxin system competence development, with the same kinetics as in wildtype cells, suggesting independent contributions from baseline repression by antitoxin and competence induction by CRP and Sxy. (Values for toxA transcript levels are shown by the red points and line in Figure H in S1 File, but are underestimates because most of the gene has been deleted).
Because most antitoxins have only weak affinity for DNA in the absence of their cognate toxin, ToxA was predicted to repress toxTA only when bound to ToxT. Therefore, we were initially surprised that knocking out toxT or both toxT and toxA did not increase RNA-seq coverage of residual toxT sequences (Fig 7, grey and black lines) and that knocking out both genes did not increase coverage of residual toxA (grey line in Figure H in S1 File). These deletion mutants retain all the toxTA upstream sequences and the toxT start codon, and enough sequence of the deleted genes to identify them in the RNA-seq analysis. An explanation was suggested by a recent study of the Escherichia coli hicAB toxin/antitoxin system [23], and confirmed by more detailed analysis of toxTA transcripts. The HicA (toxin) and HicB (antitoxin) proteins have no detectable sequence homology to ToxT and ToxA, but their operon is similarly regulated by Sxy and has the same atypical organization (toxin before antitoxin) [24]. Turnbull and Gerdes [23] showed that the hicAB operon has two promoters. Promoter P1 has a CRP-S site regulated by CRP and Sxy, which is not repressed by the HicB antitoxin. A secondary promoter P2 is very close to the hicA start codon; it is repressed by HicB independently of HicA, and its shortened transcripts produce only functional HicB, not HicA. Promoter P1 of this hicAB system thus resembles the CRP-S regulation of the toxTA operon, and the presence of a second antitoxin-regulated internal promoter similar to P2 would explain the high toxTA operon transcript level seen in the toxA knockouts. This finding prompted us to do a more detailed analysis of toxTA transcription patterns in wildtype and mutant cells to determine whether the toxTA transcripts expressed in the absence of toxA were similarly truncated. Fig 8A shows RNA-seq coverage of the toxTA promoter region and the 5' half of toxT, in wildtype cells (green) and in the toxA deletion mutant (purple). As expected, the predicted CRP-S promoter upstream of toxTA was only slightly active at T = 10 but strongly induced at T = 30 and T = 100 (note log scale); its activity was not affected by deletion of toxA. Deletion of toxA instead caused strong constitutive transcription from a second promoter ('P2'), with reads beginning about 30 bp downstream of the toxT start codon. Transcripts with this 5' end are unlikely to produce active ToxT; the only other in-frame AUG in toxT is 30 bp from the end of the gene, and it and the first GUG (position 35) lack Shine-Dalgarno sequences. This supports the hypothesis that the H. influenzae toxTA operon is regulated similarly to the E. coli hicAB operon, with Sxyinduced transcription from a CRP-S promoter and antitoxin-repressed transcription from a downstream 'P2' promoter whose transcript produces antitoxin but not toxin.
In the E. coli hicAB system, P2 is repressed by HicB antitoxin alone, binding of HicB to the P2 operator is destabilized when HicA toxin is abundant, and transcription from P2 in plasmid constructs is elevated when the chromosomal hicAB operon is deleted [23]. To look for parallels in H. influenzae's toxTA, we more precisely measured transcription in wildtype and toxTA mutant cells by scoring the coverage at two positions in the toxTA operon; each is indicated by a red vertical line at the bottom of Fig 8A. Position 0 is the toxT start codon, 34 nt downstream from the CRP-S promoter (P CRP-S ) but upstream of the putative P2 promoter, and position 100 is 70 nt downstream from P2 (P2 and position 100 are deleted in ΔtoxT). To eliminate read-mapping artefacts arising from failure of reads that span an insertion or deletion to align to the reference sequence, each mutant's reads were instead mapped onto its own toxTA sequence. Comparison of Fig 8B and 8C shows that coverage at position 100 was always higher than coverage at position 0, consistent with the presence of a second promoter between positions 0 and 100. Fig 8B also shows that coverage at position 0 (transcription from P CRP-S ) was reduced by all of the toxTA deletions. This was unexpected, and suggests that this promoter may have unusual properties, since coverage of other CRP-S genes was not similarly affected. The toxA deletion caused the predicted increase in coverage at position 100 (Fig 8C), but the toxTA deletion unexpectedly reduced rather than increased coverage at this position~3-fold from the wildtype level, even though this construct retains the first 150 bp of the operon, including P2. This reduction was not accounted for by the reduction in transcripts from P CRP-S , suggesting that high-level transcription from the toxTA P2 promoter only occurs when ToxT is present and ToxA is absent. This could mean either that ToxT directly binds the P2 promoter to induce transcription, which seems unlikely given its lack of DNA-binding domain, or that the presence of ToxT disrupts binding of a secondary repressor of the operon, such as a noncognate antitoxin [25]. Alternatively, it is possible that that the reduced transcript levels in the ΔtoxTA mutant instead reflect reduced transcript stability in this mutant.

Unopposed ToxT does not block induction of the competence regulon
Transcript levels of the competence operons that these regulators induce were also normal or near-normal in the ΔtoxA mutant at 30 min after transfer to MIV, the time when competenceinduced gene transcription is normally highest (Fig 9). Modest decreases were seen for some operons, but these are not expected to cause the absolute competence defect, for two reasons. First, competence gene transcript levels at this time were very similar between the ΔtoxA mutant, which cannot take up any DNA or produce any transformants (dark blue bars), and the ΔtoxT and ΔtoxTA mutants (other blue bars), which have normal competence. Second, an unrelated regulatory defect mutation, a knockout of hfq, causes a more extreme reduction in transcript levels at 30 min (brown bars) but this causes a much less extreme competence defect (only 10-20 fold rather than >10 6 -fold; green line in Figure B in S1 File). Mutation of toxTA genes also did not substantially change transcript levels of the sxy, crp and cya genes encoding the competence regulon regulators Sxy, CRP, and adenylate cyclase (Figure J in S1 File). At t = 100 the ΔtoxA mutant showed a stronger reduction in RNA-seq coverage of competence genes ( Figure K in S1 File), however this cannot explain the competence defect and its significance is unclear. The severe competence defect caused by unopposed toxT expression was also not explained by transcript level changes in other genes (Table E in S1 File).

Related toxins may suggest mechanism of action for ToxT
Since examination of gene expression shed little light on how the ToxT toxin prevents competence, and it has no close homologs with characterized functions, we considered the modes of action of other Type II toxins. The most common type II toxins (e.g. RelE) act as translationblocking ribonucleases, but several alternative modes of action are also known, and some newly discovered toxins lack identified activities [14]. The Pfam and TAfinder databases assign the H. influenzae ToxT protein to the ParE/RelE toxin superfamily, whose characterized members include both gyrase inhibitors and ribonucleases that arrest cell growth by cleaving mRNAs and other RNAs [2,12,13].

Unopposed toxin does not inhibit gyrase
If ToxT inhibited gyrase we would expect the RNA-seq data to show that transfer to MIV caused increased transcript levels of gyrA (HI1264) and gyrB (HI0567) and reduced levels of topA (HI1365), since these genes have opposing activities and compensatory regulation by DNA supercoiling [26]. However, these genes' coverage levels were similar in wildtype and all toxTA mutants, during both exponential growth and competence development. The SOSresponse genes lexA, recA, recN, recX, ruvA and impA [27] were also not induced.

Unopposed toxin does not cleave competence-induced mRNAs at specific sites or sequences
The best-studied homologs of the toxT toxin act by cleaving mRNAs at positions near their 5' ends during their translation on the ribosome [28,29]. Because the resulting 'non-stop' mRNAs lack in-frame stop codons and cannot undergo the normal ribosome-release process, this causes a general block to translation [30] which is predicted to arrest cell growth until normal translation can be restored [31]. Thus we considered whether ToxT might prevent competence by one of two mechanisms. First, ToxT might specifically cleave the 5' ends of competence-gene transcripts, eliminating their function without significantly changing their overall RNA-seq coverage levels or otherwise interfering with essential cell functions. Visual inspection of RNA-seq coverage of all positions within the competence operons did not reveal any anomalies that might indicate that the mRNA in ΔtoxA cells had been inactivated either by cleavage at specific sites or by random cleavage near the 5' end [32]. As an example, Figure L in S1 File compares read coverage across the comNOPQ operon in wildtype and ΔtoxA cultures after 30 min in MIV.
Because mRNA sequences preferred by an RNA-cleaving toxin [33] are expected to be depleted from RNA-seq reads, we also used DE-kupl [34] to look for differences in kmer frequencies between RNA-seq samples from cells with and without active toxin. No significant differences were found.
To detect cleavage that was neither position-specific or sequence-specific, we examined the insert sizes of our RNA-seq sequencing libraries by comparing the spanning-length distributions of paired-end sequencing reads among strains. Because independent library preparations had different insert sizes, comparisons were limited to samples prepared at the same time. Fig  10 shows that the ΔtoxA samples from library batch 1 had shorter fragment sizes than the KW20 samples from the same batch, and that the difference increased as the time after competence induction increased. This supports the hypothesis that the extreme lack of competence in ΔtoxA cultures is due to non-specific ToxT cleavage of mRNAs. This mechanism is consistent with those of homologous HigB/RelE toxins, which inhibit translation by cleavage of mRNA at the ribosomal A-site [35,36].

Discussion
Our investigation into why a HI0659 knockout prevents competence has provided a simple answer: HI0659 encodes an antitoxin (ToxA) needed to block the expression and competence-preventing activity of the toxin encoded by HI0660 (ToxT). But this answer has generated a number of new questions that we have only been able to partially answer. Why is competence controlled by a toxin/antitoxin system? How does it completely abolish DNA uptake and transformation without causing significant cell death? Do its effects in wildtype cells confer any benefit to the cells, either generally or specific to competence? Several findings support the conclusion that HI0660 and HI0659 encode proteins that function as a toxin/antitoxin pair. First is the similarity of the encoded ToxT and ToxA proteins to biochemically characterized toxin and antitoxin proteins of the RelE/ParE families. Second, and the strongest evidence, is the restoration of normal DNA uptake and transformation to antitoxin-knockout cells when the putative toxin is also knocked out. Third is the regulatory similarity between this system and the hicAB system of E. coli.

How did the toxTA operon come to be in the H. influenzae genome and under competence regulation?
Phylogenetic analysis showed that H. influenzae acquired its toxTA operon by horizontal transfer, either into a deep ancestor of the Pasteurellaceae or more recently by independent transfers into ancestors of H. influenzae and A. pleuropneumoniae. The closest relatives of the H. influenzae toxTA genes are in the distantly related Firmicutes, with homologs especially common in Streptococcus species. Since the Streptococci and Pasteurellaceae share both natural competence and respiratory-tract niches in many mammals, there may have been frequent opportunities for horizontal transfer between them.
We do not know how the toxTA operon came to be under CRP-S regulation. The toxTA operon's strong regulatory parallels with the E. coli hicAB system suggest that toxin-antitoxin systems with similar regulation and function have adopted similar roles in separate instances, a phenomenon which is more likely in toxin antitoxin systems as they undergo frequent horizontal transfers and are often under strong selective pressure. The sxy gene and the CRP-S promoters it regulates are not known outside of the Gamma-Proteobacteria sub-clade that contains the Vibrionaceae, Enterobacteraceae, Pasteurellaceae and Orbaceae [11]. Thus, it would be interesting to examine the regulation and function of the toxTA homologs outside the Pasteurellaceae to determine when and where it adopted a regulatory role and the mechanism of the toxic activity. Examining these homologs could give insight into both the mechanism of action of the H. influenzae toxTA system, and its evolutionary history.

How does unopposed ToxT prevent DNA uptake and transformation?
The transformation defect caused by deletion of the antitoxin gene toxA is very severe, so it was surprising that RNA-seq analysis detected only few and minor changes in transcript levels of competence genes. Instead, the best explanation is that ToxT is an mRNA-cleaving ribonuclease, whose activity causes a general block to translation that prevents functioning of the induced competence genes. The most direct evidence is the decrease in insert size distributions seen in ΔtoxA mutants, but this conclusion is also supported by the combination of regulatory similarities between the toxTA and hicAB systems and by sequence and predicted structural similarities between the ToxT protein and HigB ribonuclease toxins.

Why then does the ΔtoxA mutant not suffer from growth arrest or toxicity?
Part of the explanation is that mRNAs encoding functional ToxT are only expressed after cells have been transferred to competence-inducing starvation medium, a condition that severely slows cell growth and division even in wildtype cells. Detecting the predicted competencespecific toxicity is further complicated by the uneven distribution of transformability in competence-induced cells. Co-transformation experiments using multiple unlinked markers consistently show that no more than half, and sometimes as little as 10%, of the cells in a MIVtreated culture produce recombinants [6]. We do not know whether only the transforming cells express the competence genes or all cells express them but some fail to correctly assemble the DNA uptake or recombination machinery. If only a modest fraction of the cells in a competent culture are expressing the toxin then any toxic effect on culture growth and survival will be more difficult to detect.

Does this operon confer any benefit (or harm) on H. influenzae?
Why have a competence-regulating toxin/antitoxin system at all, when it has no detectable effect on competence unless its antitoxin component is defective? Although regulatory parallels with the hicAB system suggest that CRP-S regulation is not incidental, we found no direct evidence of any toxin-dependent alteration to the normal development of competence. Production of Sxy is subject to post-transcriptional regulation by the availability of nucleotide precursors [9,10], and we have elsewhere proposed that DNA uptake is an adaptation to obtain nucleotides when nucleotide scarcity threatens to arrest DNA replication forks [6]. In this context, competence-induction of the toxTA operon may be a specialization to help cells survive, by slowing or arresting protein synthesis until the nucleotide supply is restored.
On the other hand, the high frequency of deletions that remove either complete toxTA or both promoters (35%) indicates that the operon is dispensable. And the even higher frequency of toxin-inactivating deletions in the presence of intact antitoxin genes and CRP-S promoter (51%), coupled with the absence of any deletion that inactivates antitoxin but preserves toxin, indicates that unopposed toxin is indeed harmful under some natural circumstances. This may indicate that the toxin-antitoxin system represents merely a selfish genetic element which has become integrated into the H. influenzae genome and coincidentally fallen under CRP-S regulation [37]. This element may therefore be simply "junk DNA" which serves no function. Under this hypothesis, the mild purifying selection we see on the antitoxin gene would be lost in the H. influenzae strains with nonfunctional copies of the toxin gene. It is also possible that the toxA gene is beneficial even in the absence of its cognate toxin, which is under purifying selection even though most strains analyzed do not have a functional toxT counterpart. For example, the presence of a genomic antitoxin gene may counter the "plasmid addiction" effect of a plasmid bearing a similar toxin-antitoxin gene [37], perhaps limiting the risk of taking up foreign DNA. Unfortunately, the function of most toxin-antitoxin systems is poorly understood, and in this paper we are not yet able to conclusively determine the role and function of the toxTA system.
We have examined the toxTA operon from many angles and answered our initial question of why toxA knockout prevents competence in H. influenzae, but have raised new questions whose eventual answers we hope will give us greater insight not just into the toxTA system, but competence regulation in general. A number of desirable follow-up experiments could improve understanding of the toxTA system, particularly complementation experiments to determine whether ToxA expression in ΔtoxA can restore competence, and whether ToxT expression in ΔtoxTA can block competence. In future work, it would also be valuable to express ToxT and ToxA in E. coli on separate plasmids to examine the system's toxicity and effect on competence.

Bacterial strains, plasmids, and growth conditions
Bacterial strains used in this work are listed in Table A of S1 File. Escherichia coli strain DH5α [F80lacZ #(lacIZYA-argF) endA1] was used for all cloning steps; it was cultured in Luria-Bertani (LB) medium at 37˚C and was made competent with rubidium chloride according to the method provided in the QIAexpressionist manual protocol 2 (Qiagen). When antibiotic selection was required, 100 μg/mL ampicillin and 50μg/mL spectinomycin were used.
Haemophilus influenzae cells were grown in sBHI medium (Brain Heart Infusion medium supplemented with 10mg/mL hemin and 2mg/mL NAD) at 37˚C in a shaking water bath (liquid cultures) or incubator (plates). H. influenzae strain Rd KW20 [38], the standard laboratory strain, was used as the wild type for all experiments. Mutant strains used in this study were marked deletion mutants in which the coding region of the gene was replaced by a spectinomycin resistance cassette, as well as unmarked deletion mutants derived from these strains; the generation of these mutant strains is described in [8]. Specifically, we used an unmarked deletion of HI0659 (HI0659-), marked and unmarked deletions of HI0660 (HI0660::spec, HI0660-), and a marked deletion of the whole operon (HI0659/HI0660::spec). Knockout mutants of crp and sxy have been described previously [20,39].
Actinobacillus pleuropneumoniae cells were grown in BHI-N medium (Brain Heart Infusion medium supplemented with 100μg/mL NAD) at 37˚C. A. pleuropneumoniae strain HS143 [40] was used as the wild type for all experiments. Marked deletion mutants in which the gene of interest was replaced by a spectinomycin resistance cassette strains were generated for this study as described below. The HS143 genome region containing the homologs of the Actinobacillus pleuropneumoniae serovar 5b strain L20 APL_1357 and APL_1358 genes, plus approximately 1 kb of flanking sequence on each side, was PCR-amplified, ligated into Promega pGEM-T Easy and transformed into E. coli. Plasmid regions containing APL_1357, APL_1358, or both genes were deleted from the pGEM-based plasmid by inverse PCR, and the amplified fragments were blunt-end ligated to the spectinomycin resistance cassette [41] from genomic DNA of a H. influenzae comN::spec strain [8]. Plasmids linearized with ScaI were transformed into competent A. pleuropneumoniae HS143 and transformants were selected for spectinomycin resistance using 100μg/mL spectinomycin after 80 minutes of growth in nonselective medium.

Generation of competent stocks
To induce competence, H. influenzae and A. pleuropneumoniae were cultured in sBHI or BHI-N respectively and transferred to the competence-inducing starvation medium MIV [42] when they reached an optical density at 600nm (OD 600 ) of approximately 0.25 [43]. After incubation with gentle shaking at 37˚C for a further 100 min (H. influenzae) or 150 min (A. pleuropneumoniae), cells were transformed or frozen in 16% glycerol at -80˚C for later use.

Transformation assays
Transformation of MIV-competent cells. Transformation assays were carried out as described by Poje and Redfield [43]. MIV-competent H. influenzae or A. pleuropneumoniae cells were incubated at 37˚C for 15 minutes with 1μg/ml DNA, then DNaseI (10μg/mL) was added and cultures were incubated for 5 minutes to ensure no DNA remained in the medium. H. influenzae cultures were transformed with MAP7 genomic DNA [44], which carries resistance genes for multiple antibiotics, while A. pleuropneumoniae cultures were transformed with genomic DNA from an A. pleuropneumoniae strain with spontaneous nalidixic acid resistance (generated in this lab). Cultures were diluted and plated on both plain and antibioticcontaining plates (2.5ug/mL novobiocin for H. influenzae cultures, 20ug/mL nalidixic acid for A. pleuropneumoniae cultures) and transformation frequencies were calculated as the ratio of transformed (antibiotic-resistant) cells to total cells. For A. pleuropneumoniae, transformed cells were given 80 minutes of expression time in BHI-N before plating.
Time courses in rich medium. H. influenzae cells from frozen stocks of overnight cultures were diluted in fresh sBHI and incubated with shaking at 37˚C. Periodically, the OD 600 was measured, and at predetermined optical densities aliquots of the culture were removed and transformed with MAP7 DNA and plated as described above.
Bioscreen growth analysis. The Bioscreen C apparatus (BioScreen Instruments Pvt. Ltd.) was used to measure growth. Cells frozen from overnight cultures were pre-grown at low density in sBHI, and 300μL aliquots of 100-fold dilutions were placed into 20 replicate wells of a 100-well Bioscreen plate. Wells at the edges of the plate were filled with medium alone as controls. Cells were grown in the Bioscreen at 37˚C for 18 hours with gentle shaking, and OD 600 readings were taken every 10 minutes. Readings were corrected by subtracting the OD 600 measured for medium-only controls, and replicates for each strain were averaged at each time point to generate growth curves. Doubling times were calculated for each strain from the subset of time points that represents exponential growth phase, as determined by linearity on a semi-log plot of time versus OD 600 .
Competence growth and survival time course. Cells were grown in sBHI to a density of 2x10 8 cfu/ml (OD600 = 0.075) and transferred to MIV. After 100 min (time for maximum competence development, an aliquot of each culture was diluted 1/10 into fresh sBHI for recovery and return to normal growth. A fraction of each culture was incubated in a shaking water bath, and aliquots of the initial and 'recovery' sBHI cultures were also grown and monitored in a Bioscreen incubator.
Cyclic AMP competence induction. H. influenzae cells in sBHI were incubated with shaking to an OD 600 of approximately 0.05. Cultures were split and 1mM cAMP was added to one half. At an OD 600 of approximately 0.3, aliquots were transformed with MAP7 DNA and plated as described above.
Phylogenetic analysis. A nucleotide BLAST search (discontinuous MEGABLAST) and a protein BLAST search against translated nucleotide databases (tBLASTn) were used to identify homologs of the HI0659 and HI0660 genes [45]. Protein sequences found by the tBLASTn search were retained for analysis if they showed greater than 60% coverage and greater than 40% identity to the H. influenzae query sequence. For species with matching sequences in multiple strains, the sequence from only one strain was kept.
For species in which homologs of HI0659 and HI0660 were found next to one another, amino acid sequences of concatenated matrices were aligned by multiple-sequence alignment using MAFFT, version 7.220 [46], run from modules within Mesquite version 3.02 [47]. The L-INS-I alignment method was used due to its superior accuracy for small numbers of sequences. After inspection of the alignments, poorly-aligning sequences were removed from the analysis, and alignment was repeated.
Phylogenetic trees were generated using the RAxML [48] maximum likelihood tree inference program, run via the Zephyr package of Mesquite. For each gene, 50 search replicates were conducted, using the PROTGAMMAAUTO option to allow RAxML to automatically select the best protein evolution model to fit the data. Since these trees were found to correspond exactly to a set of trees generated using the PROTGAMMAJTT model, this faster model was used to generate a majority-rules consensus tree from 1000 bootstrap replicates for each gene.
Analysis of natural deletions. 181 publicly available H. influenzae genomes were downloaded from NCBI and the Sanger centre. Genomes were re-annotated using Prokka v1.11 [49], and the pangenome was calculated using Roary v3.5.1 [50] with a minimum blastp threshold of 75. The toxA gene cluster in the pangenome was identified by finding the gene cluster that contained the toxA gene from Rd KW20, and the hicA cluster was identified by finding the gene cluster that contained the hicA gene from PittAA. 2300 bp genome sequences centered on toxA and/or hicA were extracted from all H. influenzae genomes containing recognizable toxA and/or hicB genes, and aligned by MAFFT. For strains that lacked recognizable toxA or hicB, sequences adjacent to the genes that normally flanked each operon were extracted. K a /K s and pairwise distance were calculated for each gene using SeqinR v 3.4-5 [51] with codon aware gene alignments were made using Prank (v.100802).

RNA-seq analysis
Sample preparation. Cell cultures of H. influenzae strain Rd, Δcrp and Δsxy derivatives, and ΔtoxTA mutants were grown in sBHI to an OD 600 of 0.2-0.25, then transferred to MIV. Aliquots of cells were removed just prior to transfer to MIV, and after 10, 30, and 100 minutes in MIV, and immediately mixed with Qiagen RNAprotect (#76526) to stabilize RNA. Cells were pelleted and frozen, and RNA was later extracted from thawed pellets using the Qiagen RNeasy Min-elute Cleanup Kit (#74204). Contaminating DNA was removed with Ambion Turbo DNase (#AM2238), and ribosomal RNA was depleted using the Illumina Ribo-Zero rRNA Removal kit (#MRZMB126). Sequencing libraries were prepared using TruSeq mRNA v2 library preparation kit, according to manufacturer's instructions (Illumina). Libraries were pooled and sequenced on a HiSeq 2500, generating paired-end 100 bp reads.
Data analysis pipeline. FASTQ files were analysed using the FASTQC tool (Andrews, 2015) to confirm read quality. Reads were aligned to the H. influenzae Rd KW20 reference genome sequence using the Burrows-Wheeler Alignment tool (BWA) algorithm bwa mem [52]. Differential coverage analysis was performed using the DESeq2 package, v.1.6.3 [53]. Specifically, the function DESeqDataSetFromMatrix() was used to generate a dataset to compare reads from each mutant strain reads from the wild-type control based on their strain, sample time point, and the interaction between the two parameters. The function DESeq() was called to determine which genes were differentially expressed based on these parameters, using p-values adjusted for a B-H false-discovery rate [54] of 0.1 as a cut-off to determine significance, after normalizing total read counts and variances.