Gut Symbiont Bacteroides fragilis Secretes a Eukaryotic-Like Ubiquitin Protein That Mediates Intraspecies Antagonism

ABSTRACT Human gut Bacteroides species produce different types of toxins that antagonize closely related members of the gut microbiota. Some are toxic effectors delivered by type VI secretion systems, and others are non-contact-dependent secreted antimicrobial proteins. Many strains of Bacteroides fragilis secrete antimicrobial molecules, but only one of these toxins has been described to date (Bacteroidales secreted antimicrobial protein 1 [BSAP-1]). In this study, we describe a novel secreted protein produced by B. fragilis strain 638R that mediated intraspecies antagonism. Using transposon mutagenesis and deletion mutation, we identified a gene encoding a eukaryotic-like ubiquitin protein (BfUbb) necessary for toxin activity against a subset of B. fragilis strains. The addition of ubb into a heterologous background strain conferred toxic activity on that strain. We found this gene to be one of the most highly expressed in the B. fragilis genome. The mature protein is 84% similar to human ubiquitin but has an N-terminal signal peptidase I (SpI) signal sequence and is secreted extracellularly. We found that the mature 76-amino-acid synthetic protein has very potent activity, confirming that BfUbb mediates the activity. Analyses of human gut metagenomic data sets revealed that ubb is present in 12% of the metagenomes that have evidence of B. fragilis. As 638R produces both BSAP-1 and BfUbb, we performed a comprehensive analysis of the toxin activity of BSAP-1 and BfUbb against a set of 40 B. fragilis strains, revealing that 75% of B. fragilis strains are targeted by one or the other of these two secreted proteins of strain 638R.

T he gut microbiota of healthy humans is comprised of many different microbes, with members of the order Bacteroidales being the most abundant Gram-negative bacteria. Numerous Bacteroidales species colonize the human gut simultaneously at high density, and colonization with more than one strain of the same Bacteroidales species is common (1,2). Some factors that may account for the ability of so many closely related species and strains to colonize the same ecosystem include their ability to utilize different nutrients (3) or to prioritize the utilization of different nutrients (4,5), their occupation of different spatial niches (6,7), and their ability to cooperate in the utilization of dietary polysaccharides (8,9). Bacteroidales species also physically contact each other in the human gut and have been shown to exchange more than 100 kb of DNA between strains with the transfer of a single conjugative element (10,11).
Despite characteristics of the Bacteroidales that permit or promote cocolonization, these bacteria have also evolved mechanisms to antagonize each other. Antagonism or interference competition is likely an important factor dictating the composition of diverse microbial communities. Bacteroidales have been shown to elicit two different types of antagonistic systems: contact-dependent type VI secretion systems (T6SSs) (12)(13)(14)(15) and secreted antimicrobial protein toxins (16,17). Most Bacteroides fragilis strains have genetic loci encoding T6SSs (12), and some of these systems have been shown to antagonize nearly all gut Bacteroidales species tested (13). As T6SSs are contact dependent, this antagonism may largely occur when nutritional niches overlap and/or when dietary nutrients are limiting and Bacteroidales species are forced to utilize host mucins, one of the preferred carbon sources of B. fragilis (5).
In contrast to the B. fragilis T6SSs, the two identified Bacteroidales secreted (noncontact-dependent) antimicrobial proteins (BSAPs) each targets a subset of strains of the same species (16,17). Both described BSAPs contain membrane attack/perforin (MACPF) domains found in immune molecules, such as complement components and perforin, that lyse bacteria or virally infected cells by pore formation. BSAP-1 and BSAP-2 are the only bacterially produced MACPF proteins shown to kill other bacteria. BSAP-1 is produced by a subset of B. fragilis strains and mediates its toxicity through pore formation following recognition of a specific outer membrane ␤-barrel protein on sensitive (non-BSAP-1-producing) B. fragilis strains. BSAP-2 is produced by a subset of Bacteroides uniformis strains and recognizes the lipopolysaccharide (LPS; core polysaccharide or short O antigen [O-ag]) of sensitive (non-BSAP-2-producing) B. uniformis strains. The genes encoding both BSAP-1 and BSAP-2 were acquired with adjacent genes encoding orthologs of their receptors, replacing the receptor and rendering the strain resistant to the newly acquired toxin (17).
In studying secreted antimicrobial molecules produced by B. fragilis, we found that several strains, such as 638R, inhibited the growth of many B. fragilis strains, whereas other strains had no secreted antimicrobial activity against the panel of strains analyzed (16). We also showed that a mutant in which the BSAP-1-encoding gene of B. fragilis 638R is deleted retains the ability to inhibit the growth of a subset of B. fragilis strains (16). The present study was designed to identify and characterize the additional secreted antimicrobial molecule of strain 638R. Here, we describe a novel eukaryoticlike ubiquitin molecule that mediates potent antimicrobial activity against B. fragilis strains.
strains were not susceptible to secreted molecules of 638R (Fig. 1A). Ten strains were sensitive to BSAP-1 only, as His-BSAP-1 produced a zone of inhibition and no inhibitory activity remained in the 638RΔ1646 mutant (Fig. 1B). Four strains were not antagonized by BSAP-1 but were inhibited by a different molecule(s) secreted by 638R (Fig. 1C), and 19 strains were antagonized by both BSAP-1 and an additional molecule(s) secreted by this strain (Fig. 1D).
Identification of a gene necessary for inhibitory activity in 638R⌬1646. To identify the molecule or molecules responsible for the second antimicrobial activity, we performed transposon mutagenesis of the 638RΔ1646 strain and screened for loss of inhibition of the sensitive B. fragilis strain CL03T00C08. Two transposon mutants of 638RΔ1646 were identified that were severely attenuated in their ability to inhibit the growth of this strain ( Fig. 2A). In both of these mutants, the transposon inserted within the transcribed region of an unusual bacterial gene previously recognized as encoding a very close ortholog of eukaryotic ubiquitin, termed B. fragilis Ubb (BfUbb) (18,19). One transposon (Tn1) inserted 6 bp into the predicted coding region of BF638R_3923 (ubb) (Fig. 2B) and abrogated its ability to antagonize strain CL03T00C08. The second mutant (Tn8) had the transposon inserted 76 bp downstream from ubb, resulting in severely attenuated activity. We analyzed our previously generated high-throughput RNA sequencing (RNA-Seq) data on in vitro-grown mid-log-phase B. fragilis NCTC 9343, which contains a ubb genetic region identical to that of strain 638R (18,19). These analyses predicted that the ubb transcript begins just following the Ϫ7 region of the Bacteroides sigma 70 binding site (20) and terminates 140 bp downstream from the ubb stop codon (Fig. 2B). Based on this prediction, Tn8 has the transposon inserted within the ubb transcript. We tested whether these transposon mutants retained the ability to antagonize three other strains sensitive to 638RΔ1646 and found that growth inhibition of these strains was also abrogated or severely attenuated ( Fig. 2A). The RNA-Seq data revealed that ubb is the 47th most highly expressed gene of the NCTC 9343 genome, with a fragments per kilobase of transcript per million mapped reads (FPKM) value of 7,453.
To confirm that the transposon insertions in the ubb transcript accounted for the loss of toxic activity, an internal deletion mutant with the deletion of ubb was created in the 638RΔ1646 background. This strain lost the ability to antagonize the four sensitive strains tested, and the toxic activity was restored when the gene was cloned into a Bacteroides expression vector and added to the deletion mutant in trans (Fig. 2C). Therefore, ubb is required for the antimicrobial activity against these strains.
B. fragilis ubiquitin is the inhibitory factor. BfUbb was previously shown to have a 27-amino-acid (aa) signal sequence and to be secreted extracellularly (18), even though the gene is annotated as a smaller open reading frame (ORF) without this sequence. Alignment of the mature 76-aa BfUbb with the 76-aa human ubiquitin shows that the similarity begins immediately after the signal sequence and the proteins are 84% similar along their lengths, with the exception of the last 4 amino acids (Fig. 3A). The fact that BfUbb is secreted suggests that it may mediate the toxic activity itself, either directly or indirectly. To eliminate the possibility that BfUbb may be modifying or acting on another gene or gene product of strain 638R that mediates the activity, we placed ubb in trans in B. fragilis strain CM11, which does not have BfUbb activity and is also not sensitive to it (Table S1). We found that the acquisition of ubb by strain CM11 conferred upon it the ability to antagonize BfUbb-sensitive strains (Fig. 3B). These data also suggest that BfUbb-producing strains do not require an immunity protein for protection, as CM11 was not noticeably affected by the addition of ubb.
To further establish that BfUbb is the inhibitory factor of strain 638R, we had the 76-aa BfUbb peptide (lacking the signal sequence) synthesized and tested it in the assay. We found that the synthesized protein has very potent activity against sensitive strains, where as little as 250 ng of the protein resulted in a detectable zone of inhibition (Fig. 3C). Strain B. fragilis CL07T12C05, which is not affected by wild-type 638R, is not inhibited by the BfUbb peptide (Fig. 3C). In addition, purified 76-aa bovine ubiquitin, which is identical to human ubiquitin, has no activity against these strains, even when a much greater dose is used (Fig. 3D). We also tested the remaining 36 B. fragilis strains used in this study for growth inhibition by the BfUbb peptide. Of the 40 strains analyzed, 13 are inhibited by BfUbb (Table S1). All of these strains are antagonized by the 638RΔ1646 mutant, as expected (Table S1). Unlike our findings with BSAP-1, where the B. fragilis strains analyzed either produce the toxin or are sensitive to it, there are many B. fragilis strains that do not synthesize BfUbb and are also not sensitive to it (Table S1).
Species-wide analysis of the genetic architecture of the ubb region. To investigate the diversity of the ubb genetic region among B. fragilis strains, we analyzed the ubb region or the corresponding region in strains lacking ubb. We analyzed 97 B. fragilis genomes contained in our curated genome database and retrieved and analyzed DNA from mutL to the gene encoding the first protein with a ␤-propeller motif (Fig. 4A). This analysis revealed that this area of the B. fragilis genome is heterogeneous, with three major genetic types identified (Fig. 4A). Each genome has a similar mutL; however, the DNA between strains begins to diverge significantly 39 bp downstream from this gene. Of the 97 genomes analyzed, 13 contain a ubb and surrounding DNA nearly identical to that of strain 638R (Fig. 4A, top). Seventy-three of the strains completely lack ubb and Each of the genomes encodes proteins with ␤-propeller motifs outside the divergent region ( Fig. 4A, genes colored yellow), but these sequences are not conserved, even within a genetic type. A 52-bp direct-repeat element (Fig. 4A, green boxes) was identified flanking the higBA genes, one located 61 bp upstream from a ␤-propellerencoding gene and the other 257 bp downstream from mutL. A single copy of this 52-bp element is present in the other two genetic types, in both cases 61 to 67 bp upstream from their respective ␤-propeller-encoding genes (Fig. 4A). The 52-bp element in ubb-containing genomes is an exact match with the element in the higBA genetic group. However, the element differs somewhat in the genomes with no genes in the region; for example, strain 3_1_12 has 8 mismatches. The other small genes unique to the ubb genetic region do not encode proteins that appear to be of eukaryotic origin: BF638R_3921 is a peptidase of the S41 superfamily, BF638R_3922 is the N terminus of a truncated ␤-propeller protein, and BF638R_3934 is a hypothetical protein with no predicted function. Comparisons of genomes within a genetic group revealed that the genomes without any genes between mutL and the downstream ␤-propeller-encoding gene (represented by strain 3_1_12 in Fig. 4A) are the most divergent within a group. To determine whether the ubb-containing genomes may be more phylogenetically related to each other than to the genomes of the other two genetic variants, we analyzed by BLAST all 97 genomes for their phylotype of five conserved genes that are commonly used for phylogenetic analyses: dnaJ, groEL, gyrB, recA, and rpoB. It was shown many years ago that the species B. fragilis is comprised of two genetic groups, with each group having distinct ␤-lactamase-encoding genes, one with cepA and the other containing cfiA (ccrA) (21)(22)(23). In these analyses, we also found that B. fragilis genomes segregate into two distinct branches and correlate perfectly  with the presence of cepA or cfiA. In analyzing our three genetic types in the ubb region, we found that the ubb-containing genomes are indistinguishable from genomes containing the higBA toxin-antitoxin pair using this phylotyping method. Both genetic types contain cepA, and the five conserved genes are Ͼ99.6% identical to each other. However, the 11 B. fragilis genomes that do not have any genes in this region are all of the other subspecies, those which contain cfiA. The five conserved genes are only 90 to 92% identical to the orthologous gene in B. fragilis genomes containing ubb or higBA.
Both BSAP-1 and BSAP-2 target surface molecules in sensitive cells encoded by genes that were replaced by DNA containing the incoming BSAP-encoding gene. Therefore, the BSAP receptor genes are in the same genetic region of sensitive strains as the BSAP gene in producing strains. We therefore considered that BfUbb may be affecting the HigBA toxin-antitoxin (TA) system, which is not present in Bfubbcontaining strains. Among possible mechanisms for such an activity is the binding of BfUbb to the HigA antitoxin so that it can no longer interact with the HigB toxin to prevent self-intoxication. Three lines of evidence suggest that the toxicity of BfUbb may not involve the HigBA TA system. We first attempted to delete either higBA or just the higB toxin gene from several sensitive B. fragilis strains. Despite repeated attempts, we were unable to construct these deletions, possibly due to the difficulties of creating deletion mutants in some B. fragilis strains or an intolerance to the deletion of these genes. In lieu of a deletion mutant, we cloned higBA into a Bacteroides expression vector with a constitutive promoter, thereby eliminating the normal regulation of these genes, and this construct was conjugated into strain 638RΔubb. The presence of constitutively expressed higBA did not render 638RΔubb sensitive to BfUbb (Fig. 4B). We also cloned higA into a vector for high gene expression in Bacteroides to determine whether excess quantities of the antitoxin could overcome the effects of BfUbb in a sensitive strain. We observed no difference in the ability of various quantities of BfUbb to antagonize B. fragilis 078320-1 in the agar spot assay when this sensitive strain contained either the empty vector or the highly expressed antitoxin gene (Fig. 4C, showing results for one concentration). Another finding indicating that the HigBA system may not be the target of BfUbb is that we identified four B. fragilis strains with higAB genetic regions nearly identical to those of other sensitive strains, and yet, these four strains are not sensitive to BfUbb (Table S1).
Analysis of human gut metagenomic data sets for ubb. Among the 97 sequenced B. fragilis genomes in our curated genome collection, 13 genomes contain ubb. These analyses estimated the frequency of ubb in B. fragilis genomes at approximately 13%.
To determine the frequency of ubb in the human gut microbiota, we analyzed the 3CGC human gut metagenomic set, a subset of the recently compiled integrated gene catalog (IGC) that contains 1,267 metagenomes, for the presence of ubb. We detected evidence of the species B. fragilis in 370 of these metagenomes, of which ubb was detected in 44 (Table S2). ubb was only detected in metagenomes identified as containing B. fragilis (Table S2). Based on these data, approximately 12% of metagenomes with evidence of B. fragilis contain ubb, a percentage relatively consistent with that of our sequenced genome collection. In addition, we found numerous metagenomes that contain both the BSAP-1 encoding gene and ubb and some metagenomes that contain one or the other. In total, 130 metagenomes contain at least one of the two secreted antimicrobial-protein-encoding genes, and 20 contain genes encoding both BSAP-1 and BfUbb.

DISCUSSION
Ubiquitin is found in eukaryotic organisms from fungi to humans. Ubiquitin is attached to eukaryotic proteins by a series of three enzymes that create isopeptide bonds. Ubiquitination of these substrate proteins regulates their cellular fate in numerous and distinct ways. Ubiquitination or polyubiquitination of eukaryotic proteins can affect protein degradation (reviewed in reference 24), localization (25), activity, and interactions with other molecules (26). Ubiquitination of proteins has not been demonstrated in bacteria, largely due to the fact that most bacteria do not produce a Intraspecies Antagonism by B. fragilis Ubiquitin ® eukaryotic-like ubiquitin. However, some pathogenic bacteria produce proteins that alter eukaryotic cell function by interfering with the ubiquitin signaling pathways in host cells (reviewed in reference 27). Therefore, coevolution of bacteria with eukaryotic hosts has resulted in ubiquitin-related processes involved in toxicity, where either the ubiquitin molecule itself is involved in antagonistic interactions or, as described above, bacterial effectors alter host ubiquitin processes.
B. fragilis is one of a few bacteria that encode a eukaryotic-like ubiquitin. BfUbb is an interesting ubiquitin ortholog with many similarities to human ubiquitin but with some important distinctions. One of the major differences is in the C termini of these molecules. The last 4 amino acids are distinct, and BfUbb lacks the critical terminal glycine residue involved in the isopeptide linkage to substrate proteins in the eukaryotic system. Therefore, BfUbb would not serve as a substrate for eukaryotic E3 ubiquitin ligase. Another distinction is that BfUbb is glycosylated; Patrick et al. (18) previously noted that BfUbb has two Bacteroidetes glycosylation sites (28). Our prior study showed that all analyzed secreted proteins of B. fragilis that have glycosylation motifs are in fact glycosylated (29), strongly suggesting that BfUbb is glycosylated. As synthetic BfUbb has potent activity, glycosylation of this protein is not necessary for its ability to antagonize target strains. We previously showed that glycosylation of some Bacteroides secreted proteins increases their stability (28); therefore, glycosylation of BfUbb may be important in the context of the bacterial cell.
One of the most interesting distinctions of BfUbb is the N-terminal signal sequence dictating its processing by signal peptidase I and subsequent secretion. The cleavage site of this signal peptide is located such that mature BfUbb aligns exactly with the first amino acid of its human counterpart (Fig. 2B). Although mammalian ubiquitin does not contain such a signal sequence, there are ubiquitin orthologs described in several nematode species with this feature. The best described is produced by the nematode Globodera rostochiensis, a parasite of plants that creates a syncytium for successful parasitism. The secreted ubiquitin of this organism is produced exclusively in the gland of the nematode and serves as an effector for host syncytium formation (30). In addition to the N-terminal signal sequence, the nematode ubiquitin has a 12-aa C-terminal extension that is cleaved from the ubiquitin molecule in the plant. The 12-aa peptide suppresses effector-triggered immunity. The remaining core ubiquitin molecule also plays a role in parasitism, possibly by perturbing ubiquitin levels, thereby affecting the host 26S proteasome (30). Therefore, the unique ubiquitin molecules produced by some nematodes are similar to BfUbb in that they are secreted and result in effector/ toxic activity in recipient cells.
The source from which ubb was acquired is not clear from existing genomic sequences. There are several amoeba-infecting giant viruses that encode ubiquitin molecules (31,32) that are the closest orthologs of mature BfUbb in the databases. One such ubiquitin-encoding giant virus was identified in human stool (33). Giant virusinfected cells are on occasion coinfected with small virophage that are eukaryotic viruses but have properties of prokaryotic phage and are predicted to transfer genes between giant viruses during coinfection (34). It is possible that such a virus or phage may have introduced an ortholog of ubb into the B. fragilis genome. We did not detect any obvious signs in the B. fragilis genomes to hint at how this region was acquired. However, we did identify a 52-bp element in the divergent regions of all three genetic variants (Fig. 4A). This 52-bp element is duplicated in higBA-containing genomes and flanks these genes. Therefore, this region may represent an integration site for chromosomal insertions.
How BfUbb antagonizes specific B. fragilis strains is not readily obvious from the protein sequence or genomic analyses. A potential target is the antitoxin protein of the HigBA toxin-antitoxin system that is present in the majority of non-ubb-containing B. fragilis strains. Our experimental and genomic data did not confirm a role for the HigBA TA system in the activity of BfUbb, but its involvement has also not been excluded. BSAP-1 and BSAP-2 each contain an MACPF domain and bind surface receptors, leading to pore formation. BfUbb may function by a completely different mechanism. For BSAP-1, B. fragilis strains typically produce the toxin or are sensitive to it. This occurs because BSAP-1 binds a surface protein necessary for gut colonization, and in BSAP-1-producing strains, the gene encoding the protein conferring sensitivity is replaced by an ortholog that serves its function in gut colonization but also renders the strain resistant to the BSAP-1. Therefore, most non-BSAP-1-producing strains have the BSAP-1 surface target by default. For BfUbb, no such pattern emerged, as we found many strains without ubb that were not sensitive to it. Our data also suggest that, similar to BSAP-1 and BSAP-2, BfUbb does not require an immunity protein to protect the producing cell. Therefore, it is likely that sensitive cells contain a specific molecule that BfUbb targets rather than certain B. fragilis strains producing an immunity protein for resistance. It is possible that BfUbb is transported into cells by a protein-specific nutrient uptake system, where it would act on an intracellular target rather than at the bacterial surface. Continued analysis of BfUbb will likely reveal a novel mechanism of action.
This is the first bacterially produced eukaryotic-like ubiquitin molecule shown to intoxicate bacterial cells, whether directly or indirectly. To date, the three described secreted antimicrobial proteins produced by Bacteroides all have eukaryotic-like features. We previously showed that Bacteroides species synthesize another mammalianlike enzyme, termed Fkp, that charges fucose with GDP for its addition to surface glycans of these bacteria. Fkp is necessary for the bacteria to colonize the mammalian gut (35). These secreted antimicrobial molecules are additional examples of these host-associated bacteria likely acquiring and adapting eukaryotic molecules to increase their fitness in the human gut.
The importance of antagonism to bacterial colonization of the mammalian gut is evident by the fact that a single B. fragilis strain produces not only a T6SS that targets nearly all Bacteroidales species (13) but also at least two secreted antimicrobial proteins to further antagonize strains of the same species. Identifying and characterizing molecules of gut bacteria that mediate competitive interactions will allow the rational design of engineered probiotic-type bacteria to successfully colonize the host to deliver health-promoting functions.

MATERIALS AND METHODS
Primers. All primers used in this study are listed in Table S3 in the supplemental material. Bacterial strains and growth conditions. The Bacteroides strains used in this study were previously described (1,36). All Bacteroides strains were grown in supplemented basal medium (37) or on supplemented brain heart infusion (BHIS) plates. Antibiotics (5 g/ml erythromycin or 3 g/ml tetracycline) were added where indicated below. Escherichia coli strains were grown in L broth or L plates with antibiotics added where appropriate (100 g/ml ampicillin, 100 g/ml trimethoprim, and 50 g/ml kanamycin).
Agar spot test for growth inhibition analysis. The ability of Bacteroides strains to inhibit the growth of other strains was assayed using the agar spot test (38). In brief, Bacteroides strains were resuspended from plates into phosphate-buffered saline (PBS) at a density of approximately 10 10 /ml, and 5-l volumes were spotted onto BHIS plates and grown anaerobically at 37°C overnight. The bacteria were removed with swabs, and the residual bacteria remaining on the plates were killed by exposing them to chloroform vapor for 15 min. Strains to be tested for growth inhibition were grown to an optical density at 600 nm (OD 600 ) of 0.6, and then 100-l amounts were mixed with 4 ml top agar and overlaid onto the chloroform-treated plates. The zones of inhibition were analyzed after anaerobic overnight incubation at 37°C.
Growth inhibition assays using synthetic BfUbb, mammalian ubiquitin, and His-tagged BSAP-1. The 76-aa BfUbb peptide corresponding to the mature molecule without the signal sequence was synthesized by LifeTein (Hillsborough, NJ). Agar overlay assays using synthetic BfUbb were performed the same as the regular agar spot assays except that the purified protein in PBS was added to the BHIS plates (2.6 g in a 5-l volume unless otherwise indicated), allowed to dry, and then overlaid with strains as described above. Purified bovine ubiquitin (76 aa, Ͼ98% pure) from erythrocytes was purchased from Sigma (U6253) and resuspended in PBS. Five micrograms of protein in 5 l was spotted onto the BHIS plates for the overlay assay. The N-terminally His-tagged BSAP-1 protein was purified and processed as previously described (16). For the agar overlay assay, 2.5 g of His-BSAP in a 5-l volume of PBS was added to the plates and allowed to dry before performing the overlays.
Transposon mutagenesis and deletion of BF638R_3923 (ubb). Random mutagenesis of B. fragilis 638RΔ1646 was performed using the transposon-containing plasmid pYT646b as described previously (39), using tetracycline selection. The insertion sites of transposon mutants were identified by cloning the junctional DNA as described previously (16,39).
Intraspecies Antagonism by B. fragilis Ubiquitin ® A deletion mutant with the deletion of ubb was constructed such that 255 bp of the 312-bp gene was removed. DNA segments upstream and downstream from the region to be deleted were PCR amplified, and the PCR products were digested with BamHI and EcoRI and cloned by three-way ligation into the BamHI site of pNJR6 (40). The resulting plasmid was conjugally transferred into wild-type B. fragilis 638R or 638RΔ1646 using helper plasmid R751, and cointegrates were selected by erythromycin resistance. Following growth under nonselective conditions, erythromycin-sensitive colonies were screened by PCR for the mutant genotype.
Cloning ubb into a conjugal expression vector. ubb was PCR amplified using primers with BamHI ends (Table S3). The PCR product was digested with BamHI, cloned into the BamHI site of Bacteroides expression vector pFD340 (41), and screened for correct orientation in relation to the plasmid-borne promoter. The resulting plasmid was conjugated into B. fragilis 638RΔ1646 and B. fragilis CM11 by conjugal mating using an E. coli strain containing helper plasmid RK231 and selected by acquisition of erythromycin resistance.
Cloning of the higBA toxin-antitoxin genes. The putative toxin-antitoxin genes HMPREF1067_ 00095 and -96 were PCR amplified from sensitive strain CL03T12C07, cloned into the BamHI site of pFD340, and screened for correct orientation in relation to the plasmid-borne promoter. The antitoxin gene (HMPREF1067_00096) was cloned into pMCL140 (42), a Bacteroides expression vector for high expression of cloned genes. Both plasmids were conjugated into Bacteroides strains as described above.
Determination of ubb transcript size and expression level. We reanalyzed the two biological replicates of wild-type B. fragilis NCTC 9343 from our previously generated RNA-Seq data (13) by first adapter and quality trimming the two sets of paired-end reads using BBDuk (see "Genome sequencing of additional B. fragilis strains" below). Read alignment, transcript prediction, and statistical calculations, including normalization and measures of relative expression (fragments per kilobase of transcript per million mapped reads [FPKM]), were achieved using HISAT2 (version 2.1.0) (43), samtools (version 1.6) (44), and StringTie (version 1.1.3) (45). We used the sequence information for B. fragilis NCTC 9343 available from NCBI (GenBank accession number NC_003228) as a scaffold, except that we corrected the start coordinate of the ubb gene (BF9343_3779) to reflect the true beginning of the open reading frame.
Detection of B. fragilis and B. fragilis genes in human gut metagenomes. Metagenomic analyses were performed using a subset (3CGC) of the recently compiled integrated gene catalog (IGC) (46). This subset comprises 1,267 human gut metagenomes. To detect B. fragilis in these metagenomes, we used DNA sequences of single-copy genes known previously to differentiate B. fragilis from other species, namely, dnaJ, groL, gyrB, recA, and rpoB. The DNA sequences of these five genes were collected from each of the three B. fragilis strains with unique genetic types in the ubb or corresponding region. The genes from B. fragilis 638R are BF638R_1741 (dnaJ), BF638R_3250 (groL), BF638R_0298 (gyrB), BF638R_1245 (recA), and BF638R_4052 (rpoB); the genes from B. fragilis 078320-1 are CQW37_03448 (dnaJ), CQW37_03630 (groL), CQW37_02683 (gyrB), CQW37_01563 (recA), and CQW37_01330 (rpoB); and the genes from B. fragilis 3_1_12 are BFAG_01078 (dnaJ), BFAG_03510 (groL), BFAG_02716 (gyrB), BFAG_00519 (recA), and BFAG_04295 (rpoB). These fifteen gene sequences were used as queries against a blastn database created using the makeblastdb program. The output from the blastn command (executed with switches -task megablast -evalue 1eϪ5 -dust no -best_hit_score_edge 0.05 -best_hit_ overhang 0.25) was parsed, and the best hit (by highest bit score) returned from each metagenome was retained if that metagenome had a hit reaching the threshold levels indicated. Hits that survived the filter cutoff values were defined as evidence of the presence of B. fragilis in the subject metagenome (Table S2). The search for ubb, higBA, and the BSAP-1-encoding gene in the metagenomes was performed in the same way, using the DNA sequences of CQW37_03188 and CQW37_03189 (the higB and higA genes, respectively, from B. fragilis 078320-1), BF638R_3923 (the ubb gene from B. fragilis 638R, including the region encoding the heretofore unannotated signal sequence), and BF638R_1646 (the BSAP-1encoding gene from B. fragilis 638R) as queries (Table S2).
Genome sequencing of additional B. fragilis strains. Chromosomal DNA from B. fragilis strains 12905, CL04T03C20, US326, CM13, 1284, and 078320-1 was fragmented using the Covaris S2 instrument and analyzed for fragment distribution with a high-sensitivity D1K TapeStation machine and for sufficient quantity by a SYBR quantitative PCR (qPCR) assay. The DNA was sequenced using an Illumina MiSeq sequencer, producing paired-end reads of 150 bp. Genomic sequencing was performed by the Biopolymers Facility, Harvard Medical School. The raw paired Illumina reads were processed to remove adapter sequences and quality trimmed using BBDuk, part of the BBTools (version 37.50) suite of programs distributed by the Department of Energy's Joint Genome Institute (https://jgi.doe.gov/data-and-tools/ bbtools/). NCBI's UniVec_Core database (build 10.0) was downloaded (ftp://ftp.ncbi.nlm.nih.gov/pub/ UniVec), entries originating in GenBank were removed, and the Illumina reads were further screened against this data set using blastn, removing any read that returned a significant hit. The reads passing these screens (including orphans) were used to assemble the genomes. Velvet Optimizer (version 2.2.5, http://www.vicbioinformatics.com/software.velvetoptimiser.shtml) was utilized to determine the optimal k value (among other settings), and the genomes were assembled de novo using Velvet 1.2.10 (47). The draft genomes were annotated using an in-house-customized version of Prokka version 1.12 (48).

ACKNOWLEDGMENTS
We declare that we have no conflicts of interest. We thank the BEI for providing some of the strains used in this study.
M.C.-L. performed experiments, analyzed data, and wrote the paper. M.J.C. performed all bioinformatics analyses and wrote the paper, K.G.R. performed experiments and analyzed data, R.R.G. performed experiments, J.M.C. performed experiments, and L.E.C. performed experiments, analyzed data, and wrote the paper.
This work was supported by Public Health Service grant R01AI093771 from the NIH/National Institute of Allergy and Infectious Diseases. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.