De Novo Synthesis of a Conjugative System from Human Gut Metagenomic Data for Targeted Delivery of Cas9 Antimicrobials Authors: Thomas A Hamilton, Benjamin R Joris, Arina Shrestha, Tyler S Browne, Sébastien Rodrigue, Bogumil J Karas, Gregory B Gloor, David R Edgell Published: 2023 ABSTRACT AbstractMetagenomic sequences represent an untapped source of genetic novelty, particularly for conjugative systems that could be used for plasmid-based delivery of Cas9-derived antimicrobial agents. However, unlocking the functional potential of conjugative systems purely from metagenomic sequences requires the identification of suitable candidate systems as starting scaffolds for de novo DNA synthesis. Here, we developed a bioinformatics approach that searches through the metagenomic “trash bin” for genes associated with conjugative systems present on contigs that are typically excluded from common metagenomic analysis pipelines. Using a human metagenomic gut data set representing 2805 taxonomically distinct units, we identified 1598 contigs containing conjugation genes with a differential distribution in human cohorts. We synthesized de novo an entire Citrobacter spp. conjugative system of 54 kb containing at least 47 genes and assembled it into a plasmid, pCitro. We found that pCitro conjugates from Escherichia coli to Citrobacter rodentium with a 30-fold higher frequency than to E. coli, and is compatible with Citrobacter resident plasmids. Mutations in the traV and traY conjugation components of pCitro inhibited conjugation. We showed that pCitro can be repurposed as an antimicrobial delivery agent by programming it with the TevCas9 nuclease and Citrobacter-specific sgRNAs to kill C. rodentium. Our study reveals a trove of uncharacterized conjugative systems in metagenomic data and describes an experimental framework to animate these large genetic systems as novel target-adapted delivery vectors for Cas9-based editing of bacterial genomes.Keywords: CRISPR/Cas9, human metagenome, bacterial conjugation, DNA assembly, antimicrobial Introduction Metagenomic sequencing has revealed the diversity of microbial species that compose the human microbiome and the widespread occurrence of antimicrobial resistance (AMR) genes and mobile genetic elements that can spread AMR.1−3 The development of multidrug-resistant bacteria has reduced the efficacy of traditional antibiotic strategies, necessitating alternative approaches to combat AMR and microbial dysbioses.4,5 Recently, Cas9 and related nucleases derived from the clustered regularly interspaced short palindromic repeat (CRISPR) system have been repurposed as antimicrobial agents to target specific bacteria for elimination based on the introduction of double-strand breaks in the chromosome that lead to replication fork collapse and cell death.6−8 A key unresolved issue in using Cas9 as an antimicrobial tool is delivery. Current strategies that use phage,9−11 phagemids,12,13 or conjugative plasmids6,7,14 as delivery vectors are based on characterized systems that replicate in Escherichia coli or other model bacteria and that are amenable to genetic manipulation. Delivery vectors are limited in their application by a number of factors, including host range and stability in natural bacteria and, in the case of phage vectors, development of resistance.15 Metagenomic data represent an untapped source of potential mobile genetic elements, and conjugative systems in particular, that could be repurposed as host-adapted antimicrobial delivery tools. Conjugative plasmids are well suited as delivery vectors for CRISPR nucleases or other genetic tools; they have a large coding capacity, they often encode CRISPR systems,16,17 they encode factors that promote biofilm formation to enhance cell-to-cell contact and DNA transfer,18 they can carry antirestriction systems to promote resistance to restriction-modification systems,19−22 and they generally do not require a cellular receptor that would provide a mechanism for bacterial resistance,23 although some conjugative systems do require a receptor.24,25 Conjugative systems initiate the unidirectional transfer of DNA from a donor to a recipient cell through the actions of the relaxase protein (MOB) that nicks DNA at a defined origin of the transfer sequence (oriT).26,27 Subsequent interaction of the relaxosome with the type IV coupling protein (T4CP), mating pair formation, and type 4 secretion (T4SS) proteins catalyze DNA transfer to the recipient.28−30 Conjugative systems can exist as complete systems that are self-transmissible or as partial systems that are dependent on conjugative proteins encoded by other elements. Upon DNA transfer, some conjugative systems integrate into the chromosome as integrative conjugative elements (ICEs).31−33 Currently, the classification of conjugative systems is based on the identity of the MOB protein.34 Metagenomic sequencing coupled with bioinformatics analysis, DNA synthesis, and synthetic biology optimization of gene expression can provide access to the biochemical repertoire of a community.35 Identification of conjugative systems (and other mobile genetic elements) from metagenomic sequencing studies where individual contigs (often generated by short-read sequencing) are assembled into metagenome-assembled genomes (MAGs) is problematic.36,37 Conjugative systems often have GC content that is different from the parent genome38 and can be excluded from metagenomic binning algorithms that use GC content as a parameter for clustering. Moreover, plasmids that carry conjugative systems are not necessarily maintained in the same unit copy number as the parent chromosome, causing differential sequence coverage and exclusion of the contigs that encode these systems from MAGs. Thus, our current understanding of the distribution and diversity of conjugative systems in metagenomes is likely an underestimate, and it is unknown if metagenomically identified systems represent functional conjugation systems. Here, we elaborate a synthetic biology approach to capitalize on the diversity of conjugative systems from human metagenomic data for the delivery of CRISPR nucleases. We developed a bioinformatics pipeline that searches for genes associated with conjugative systems in contigs that are excluded from assembly into MAGs. Mapping of the identified genes to short-read sequences from human gut metagenomic data revealed differences in the distribution and abundances between human cohorts.1 We selected a putative ∼54 kb conjugative system identified on a Citrobacter spp. plasmid and synthesized it de novo to create a functional conjugative plasmid, pCitro. Crucially, pCitro conjugates with a higher frequency to Citrobacter species than to other bacteria, is compatible with resident Citrobacter plasmids, and is stable over multiple generations. Programming pCitro with a TevCas9 dual nuclease and sgRNAs resulted in the efficient killing of Citrobacter. This work provides proof of principle that conjugative systems identified in metagenomic data can be synthesized de novo and functionalized for the targeted killing of bacteria. Results Identification of Conjugation-Associated Genes from the Human Gut Microbiome Reveals Differential Geographical Distribution To overcome problems associated with the identification of conjugative systems from MAGs, we searched the sequences of contigs that were excluded from binning into MAGs, rationalizing that these contigs would be enriched for genes associated with conjugative systems (Figures 1A and S1). We used a data set that represents a nonredundant picture of the human gut microbiome1 of 2805 taxonomically distinct units and searched 92143 contigs that did not pass quality filtering and were not binned into MAGs. Open-reading frames (ORFs) were predicted from the contigs and aligned to the UniRef90 database,39 with the results subsequently searched by keyword for conjugation proteins. We found a total of 1598 contigs from 787 taxonomically distinct units that matched UniRef90 annotations for relaxase/MOB and T4SS/T4CP proteins (Data S1 and S2). From these contigs, 3216 subregions where conjugative protein annotations were concentrated on the contig and representing the minimal conjugative regions were extracted, retaining 2413 that were >1 kb in size (Figure 1A). These minimal conjugative regions represent an aggregate picture of the distribution of conjugative functions because the human gut microbiome data used here itself are an aggregate of short-read metagenomic data from many different samples. Legend on the following page. Identification and distribution of conjugative functions in the human gut metagenomic data. (A) Workflow (on the left) for identifying genes and functions associated with conjugative systems from metagenomic data. On right side shows a classification of identified conjugative functions by bacterial phylum (outer ring) and by relaxase (MOB) homology (middle ring), with GC content shown on the inner ring. (B) Distribution of identified conjugative functions in different human cohorts shown by an Anvi’o plot. The outer ring denotes the bacterial phylum, and the vertical and horizontal color bars indicate the human cohort. The inner rings represent the abundance of a conjugative system in each cohort, as indicated by a black dot (plotted on a log2 scale), with darker points representing a higher relative abundance of a conjugative system. Using MobScan40 (Figures 1A and S2) we predicted the mobility (MOB) families of the relaxases that were present in the identified minimal conjugative regions. All 9 of the relaxase Mob families were present, but interestingly, MOBP1-type relaxases were found to be the majority with 865 represented systems. We also used CONJscan41 to classify the minimal conjugative regions. CONJscan identified 1334 systems as having MOB proteins associated with plasmids and 134 with MOB proteins identified as chromosomal but only classified 18 as belonging to the type B T4SS and 4 to the type C T4SS families (Table S1). We used PlasFlow42 to predict whether the conjugative systems were encoded on a plasmid (Figure S2) and found that 57 were predicted to be plasmid-based. Of the remaining systems, 1353 were predicted to be chromosomal, and the genetic location of 179 conjugative systems could not be determined. To provide a higher-resolution picture of the minimal conjugative regions in different human cohorts, we aligned short-read sequencing data from 785 samples from 8 different cohorts to the minimal conjugative regions (Figure 1B, Data S3 and S4). This approach allowed us to determine the abundances of each conjugative region relative to all other regions based on the read counts in the short-read sequences. Surprisingly, we found that some minimal conjugative regions were common to all cohorts whereas other conjugative regions were private to specific cohorts. This was most notable for the North American Preterm Infant cohort that was almost exclusively composed of conjugative regions from 16 different genera of Pseudomonadota. In contrast, these conjugative regions were essentially absent from the Western European Infant cohort. Constructing a Conjugative System De Novo The analyses in Figure 1 revealed a significant diversity of minimal conjugative regions in the human gut metagenome, but whether these regions were associated with functional conjugative systems was unknown. The identified minimal conjugative regions could represent partial systems, systems that function in trans or orphan genes that are not associated with full conjugative systems. To address these questions, we selected a 53.7 kb conjugative system (c_000000000801, hereafter called 20298) with a MOBP1 family relaxase that was extracted from a larger ∼150 kb Citrobacter contig for DNA synthesis (Figure 2A, Table S2, Datas S2 and S5). We chose this system because the ∼150 kb contig appeared to be of plasmid origin as it contained genes for a toxin/antitoxin system and a repA homologue. The 20298 conjugative system, or a system(s) similar to support read mapping using our stringent parameters, was found in virtually every human cohort in Figure 1B, but under-represented in indigenous populations and absent from infant populations not exposed to antibiotics (Figure S3). This distribution suggests that the system 20298 is functional. Moreover, Citrobacter rodentium is a model for enterohemorrhagic E. coli.6,44−46 Strategy for the synthesis and assembly of the 20298 contig. (A) Schematic of the 20298 contig (to scale) with genes that have similarity to known conjugative functions indicated by blue-filled arrows. The putative relaxase is indicated by an orange-filled left-facing arrow and the putative oriT sequence by a pink rectangle. The contig was split into tiles for DNA synthesis as indicated. Tile 9 was further split into 4 smaller synthetic fragments that were assembled to create a complete tile 9’. Tiles 1 and 11 had homology hooks (red and green rectangles) for assembly into the pAGE or p15a vectors by in vivo yeast assembly. (B) Identification and functional testing of the 20298 oriT sequence. The putative oriT sequence and flanking nikA and mobC genes were cloned to create p298-ori. Below the plasmid map are direct repeats identified in the 20298 oriT sequence.43 At right is the strategy to test conjugation of p298-ori from E. coli to C. rodentium. pCitro-15a is a sequence-verified clone of the 20298 conjugative contig and pCitro-15-M has inactivating mutations in the traV and traY genes. The plot shows the conjugation frequency of p298-ori from E. coli harboring either pCitro-15 or pCitro-15-M to C. rodentium. Data points are three biological replicates. The limit of detection for exconjugants is 10–8. To synthesize the 20298 conjugative system for assembly into destination plasmids by in vivo yeast assembly,47,48 we divided the 53.7 kb contig into 11 tiles of ∼5 kb that were commercially synthesized (as described in the Methods section; Data S6). The end tiles (1 and 11) contained “homology hooks” to the destination plasmids, and the internal tiles had a minimum of 100 bp homology overlap with the adjacent tile (Figure 2A). One tile (number 9) that included the coding regions for traV, traW, traX, and traY could not be synthesized commercially and was subsequently synthesized in-house on a Codex DNA BioXP 3200 printer as 4 smaller fragments that were assembled into a complete tile (tile 9′) (Data S6). At the same time, we constructed a vector (p298-ori) that contained the putative origin of transfer (oriT) for the 20298 conjugative system (Figure 2B, Data S7). We first assembled the 20298 contigs into the pAGE149 backbone to create pCitro-AGE and screened for correct assembly by diagnostic restriction digest and then by whole plasmid sequencing (Figure S4, Table S3, Data S8). Of the five clones sequenced, all contained frameshift mutations in traV (pilus assembly) and traY (Integration Host-Factor (IHF) association and oriT nicking) suggesting either a systematic issue with DNA synthesis of this region or that these genes were toxic in E. coli when this particular DNA tile was cloned independent of the adjacent tiles. Attempts to conjugate one clone (pCitro-AGE-5) from an E. coli donor to an E. coli or C. rodentium recipient yielded no transconjugants (Figure 3A). We considered two reasons that conjugation of pCitro-AGE-5 was unsuccessful. First, mutations in the traV and traY genes could abolish conjugation of pCitro-AGE to both E. coli and C. rodentium. It was also possible that clones contained mutations (other than obvious frameshifts) in essential conjugation genes but that these mutations would be difficult to identify based solely on the sequence of the metagenomically assembled 20298 contig. Second, while the oriV and F ′ origins of the pAGE backbone function in E. coli, they may not support replication in Citrobacter, possibly due to incompatibility with resident plasmids. We subsequently identified 4 resident plasmids by sequencing our C. rodentium strain that are similar to previously described Citrobacter plasmids.50 Conjugation of the pCitro variants. (A) Conjugation of pCitro-15a, pCitro-AGE, and pNuc-cis from an E. coli donor to an E. coli DBH179 KanR or C. rodentium DSB100 RifR recipient. (B) Conjugation of pCitro-15a from a C. rodentium DSB100 donor to an E. coli DBH179 KanR or C. rodentium DSB100 RifR recipient. For panels (A,B) points represent biological replicates. (C) Timecourse conjugation from 30 min to 24 h of pCitro-15a an E. coli EPI300 dapA- donor to an E. coli DHB179 KanR or a C. rodentium DBS100 RifR recipient. Individual dots represent biological replicates and the shaded regions indicate the 95% confidence interval. (D) Strategy for testing sequential conjugation of pCitro-15a (on the left) with the frequency of conjugation from a C. rodentium DBS100 RifS donor to a C. rodentium DBS100 RifR recipient shown on the right. C. rodentium DBS100 RifS transconjugants were selected after conjugation from an E. coli Epi300 dapA- donor. RifS transconjugants were prepared as donors for subsequent conjugation to a C. rodentium DBS100 RifR recipient. (E) Conjugation of pCitro-15a from E. coli EPI300 dapA- to different bacteria. Each point represents a biological replicate. The limit of detection for exconjugants is 10–8. To resolve these issues, we attempted to complement the functions of pCitro-AGE in trans with different conjugative plasmids that could replicate in E. coli with pCitro-AGE (Figure S5). We found one conjugative plasmid, MIP231, belonging to the IncY incompatibility group that promoted conjugation of pCitro-AGE to an E. coli recipient with a conjugation frequency of ∼10–4–10–5 (Figure S5). Second, we used electroporation rather than conjugation to move pCitro-AGE or pAGE1 into C. rodentium but could not recover transformants for either plasmid, whereas transformants were readily obtainable by using a p15a-based plasmid. Collectively, these experiments showed that the inability of pCitro-AGE to conjugate was due to mutations in the traV and traY genes because the pAGE1 origin did not support replication in C. rodentium. Using pCitro-AGE-5 as a template, we reamplified the 20298 conjugative system using PCR primers that simultaneously corrected the traV and traY genes. These new fragments were then assembled into the p15a backbone in yeast to generate pCitro-15a (Figure 2A, Data S9, and Table S4). Additionally, we constructed pCitro-15a-M that changed the plasmid backbone but maintained the inactivating mutations in the traV and traY genes that were present in pCitro-AGE. Preferential Conjugation of p20298-15a to Citrobacter Strains Using an E. coli EPI300 dap- auxotrophic donor strain we successfully conjugated pCitro-15a to E. coli and C. rodentium recipients (Figure 3A). Crucially, conjugation from the E. coli EPI300 donor occurred at an approximately 30-fold higher frequency to C. rodentium than to E. coli after a 1 h conjugation (p-value = 1.1 × 10–6, Welch’s two-sided t-test). We obtained a similar result when we conjugated pCitro-15a from a C. rodentium donor to either E. coli EPI300 or a rifampicin-resistant C. rodentium recipient; a 13-fold higher conjugation frequency to C. rodentium than to E. coli was observed (p-value = 0.008, Welch’s two-sided t-test) (Figure 3B). We also performed a time-course conjugation experiment with the E. coli EPI300 donor and an E. coli or C. rodentium recipient over 24 h (Figure 3C). A higher frequency of conjugation to C. rodentium than to E. coli was observed over the first 4 h, whereas the 12- and 24 h time points showed similar conjugation frequencies. We also found that pCitro-15a had a higher frequency of conjugation from E. coli to C. rodentium than pNuc-cis, an IncP RK2-based conjugative plasmid that we previously constructed for Cas9-killing of S. typhimurium(7) (Figure 3A). However, pNuc-cis had a much higher frequency of conjugation from an E. coli donor to an E. coli recipient than did pCitro-15a (Figure 3A). We used pCitro-15a to confirm that the p298-ori plasmid (with the putatively identified origin of the transfer sequence) could be conjugated in trans from E. coli to C. rodentium, whereas the pCitro-15-M plasmid with mutations in the traV and traY genes did not support conjugation of p298-ori (Figure 2B). We also showed that pCitro-15a could be sequentially conjugated from an E. coli donor to a rifampicin-sensitive C. rodentium recipient, and then from C. rodentium transconjugants that became donors to a rifampicin-resistant C. rodentium recipient (Figure 3D). The frequency of reconjugation to the rifampicin-resistant C. rodentium recipient occurred with a frequency similar to that observed for E. coli to C. rodentium conjugation experiments (Figure 3A). We analyzed 4 clones each of 6 original pCitro-15a C. rodentium transconjugants that had been passaged on solid and liquid media by restriction digests, finding no evidence of gross rearrangements in pCitro-15a (Figure S6). The same analyses revealed restriction fragments diagnostic of the resident C. rodentium plasmids. Further evidence of plasmid function and stability was obtained by mapping reads from RNA-seq of RNA extracted from C. rodentium or E. coli harboring pCitro-15a (Figure S7 and Data S10). In both cases, reads could be mapped to all annotated genes in the 20298 contig, but we noted a higher abundance of reads that mapped to the coding versus the noncoding strands for RNA isolated from C. rodentium. This differential mapping could reflect more stringent regulation by Citrobacter-specific regulatory elements. To examine the host range of pCitro-15a within the Citrobacter genus, we performed conjugations of pCitro-15a to seven additional Citrobacter spp. strains (Figure 3E). After a standard one hour conjugation from the E. coli Epi300 donor, moderate conjugation frequencies between 2.9 × 10–6 and 7 × 10–6 were found for four of the seven strains. The remaining three strains yielded minimal or no transconjugants. We also tested conjugation to other Gammaproteobacteria, finding moderate conjugation to Morganella morganii but not to other bacteria. Conjugation was performed with C. rodentium in parallel and occurred at a higher frequency (5.9 × 10–4) than with any of the other species tested. Collectively, these data show that we successfully synthesized and assembled a functional conjugation system that was identified from metagenomic data. The 20298 conjugative system functions outside of the context of the large contig that was identified. Crucially, the pCitro-15a plasmid has a significantly higher frequency of conjugation to Citrobacter species than to E. coli, implying that the 20298 conjugation system is optimized for the Citrobacter genus. Moreover, pCitro-15a appears to stably replicate in C. rodentium and is compatible with the resident plasmids. Functionalizing p20298 to Kill C. rodentium with TevCas9 To facilitate cloning of genetic cargo into pCitro-15a, we installed a destination cassette for gateway recombination cloning51 to generate pCitro-dest (Figure 4A, Data S11). We also created pENTR-TC9 (Data S11), an entry plasmid for pCitro-dest that contains the TevSpCas9 dual nuclease under the control of an arabinose-inducible promoter and an sgRNA expression cassette driven by a constitutive promoter derived from the tetracycline resistance gene (Figure 4A). Together, these plasmids allowed us to create pCitro-dest clones with different sgRNAs targeting the C. rodentium genome and where correct assembly could be quickly confirmed by simple diagnostic restriction digests (Figure 4A). Functionalizing pCitro to kill C. rodentium. (A) pCitro was modified to include a gateway destination cassette for cargo insertion to create pCitro-dest. pENTR-TC9 was constructed to facilitate the recombination of TevCas9 and sgRNA into the pCitro-dest plasmid. (B) Mechanism of chromosomal targeting by a conjugated pCitro-TC9 plasmid. pCitro-TC9 is conjugated from E. coli to C. rodentium. TevCas9 targeting the C. rodentium genome is either induced with arabinose or repressed with glucose. Transconjugants are plated and grown overnight and killing efficiency is calculated by comparing CFUs. (C) Killing efficiency of C. rodentium with pCitro-TC9. Eight sgRNAs targeting the C. rodentium genome (blue circles), a nontargeted sgRNA (red triangles), and an empty sgRNA cassette control (NG, black diamonds) were tested in triplicate. Killing efficiencies were calculated as described above and expressed as a percent. We cloned eight sgRNAs targeted to the C. rodentium genome that were predicted to have high activity52 and one sgRNA not targeted (NT) to the C. rodentium genome into pCitro-dest (correctly assembled clones were called pCitro-TC9) (Figure 4B, Data S13, Table S5). We conjugated pCitro-TC9 from E. coli to C. rodentium with glucose supplementation to repress TevSpCas9 expression. Individual transconjugants were grown in liquid cultures and then plated on solid media under inducing (+0.2% arabinose) or repressing (+0.2% glucose) conditions. We determined the killing efficiency by comparing colony counts between the two conditions (Figure 4B) and observed killing efficiencies of 82–99.9% for all sgRNAs targeted to the C. rodentium genome (Figure 4C). In contrast, killing efficiencies for the nontargeted sgRNA, as well for pCitro-TC9 with no sgRNA (NG), ranged from 0 to 20% (Figure 4C), agreeing with previous observations that the expression of TevSpCas9 or SpCas9 without an sgRNA can be partially toxic suggesting that expression of TevSpCas9 by itself was partially toxic as previously observed.52−56 Taken together, these data show that pCitro can successfully deliver a TevSpCas9 nuclease for targeted killing of C. rodentium. Discussion Strategies to access the functional potential of metagenomic sequences can be broadly classified into untargeted approaches that involve functional- or sequence-based screening of metagenomic DNA libraries for activities of interest,57 or targeted approaches that identify candidate genes based on sequence or structural similarity for DNA synthesis and characterization. Indeed, a number of studies have utilized the latter, synthetic biology approach to functionalize single genes, small biosynthetic pathways, or regulatory elements from reference databases or metagenomic surveys.58−61 Our strategy of searching contigs typically excluded from metagenomic assembly pipelines identified a trove of uncharacterized, multigene conjugative systems in the human gut metagenome as starting scaffolds for DNA synthesis and assembly. To our knowledge, our study is the first to reanimate a large multigene system de novo from metagenomic data. Recent studies have mined the reference sequence databases of bacterial isolates to highlight the diversity of plasmid-based conjugation systems.62,63 Thus, a primary motivation for our study was to broadly reveal the diversity of uncharacterized conjugative systems in metagenomic data that could be used as scaffolds for DNA synthesis rather than to provide an exhaustive description of conjugative systems. Two notable findings emerged from our analyses. The first was that conjugative systems are differentially distributed between human cohorts, which could reflect sampling bias or differences in diets, antibiotic exposure, or other environmental differences that could influence bacterial composition.64,65 The second was the differential distribution of conjugative systems in bacterial phyla, particularly the low relative abundance of systems in Pseudomonadota relative to those found in Bacillota and Bacteroidota. This difference is striking because many previously characterized plasmid-based conjugative systems were isolated because they could replicate in and conjugate to model bacteria (often Pseudomonadota) cultivable in a laboratory environment. However, by using the synthetic biology strategy outlined here, uncultivable plasmid-based conjugative systems could be animated in E. coli (or other hosts). Interestingly, only 3.5% of identified systems were classified as plasmid-based, reflecting the limitations of current annotation tools, while 85% were classified as chromosomal, which likely included integrated-conjugative elements. In the future, identification of plasmid-based conjugation systems from metagenomic communities could be enhanced by using long-read sequencing technologies, or a hybrid of long- and short-read sequencing,66 to capture the sequences of intact plasmids,67 as opposed to short-read sequencing that was primarily used to generate the human gut metagenome and that is not well-suited to capturing plasmids or other sequences with differential copy number.68 While significant at project initiation, the declining costs for DNA synthesis and assembly of large DNA fragments suggest that the animation of uncharacterized conjugative systems from metagenomic data is not insurmountable. We found, however, that the repeated occurrence of frameshift mutations in tile 9 of the 20298 contig, specifically in the traV and traY genes, was problematic. This observation suggested either a systematic issue with DNA synthesis of this tile or that these conjugation components were toxic in E. coli when cloned on a multicopy vector. In the IncI1 plasmid R64, both gene products are involved in conjugal transfer,69 with the TraY being a predicted integral membrane protein and TraV involved in the assembly of the conjugative pilus.70 Potential mechanisms for TraY and TraV toxicity are unclear. Regardless, the use of compatible conjugative plasmids to complement inactive pCitro assemblies suggests a general strategy to ascertain whether assembled conjugative systems are functional, particularly those that do not contain obvious errors that create frameshifts or other mutations. Moreover, by using plasmid libraries of single genes from known, functionally active systems, it would be possible to narrow inactivating mutations on the assembled plasmids. Alternatively, the use of a different “storage host” for the synthesized tiles, such as Saccharomyces cerevisiae or Bacillus subtilis,71,72 could avoid issues of toxicity. Our data suggest that target-adapted conjugative systems are abundant in the human gut metagenome. Interestingly, we observed a higher frequency of conjugation of pCitro-15a to C. rodentium than to other bacteria, agreeing with studies that showed specificity in the conjugation process.24,73 A number of groups have utilized conjugative systems with broad host ranges in an attempt to access a significant fraction of the targeted microbiome for editing.74−76 While promising, low conjugation rates to diverse bacteria in the targeted microbiome and the stability of conjugative plasmids with resident plasmids are confounding factors. Bacterial resistance to Cas9-killing has also been documented and could be minimized by multiplexing sgRNAs to different targets and tighter regulation of Cas9 expression.7,77 In many cases, conjugative systems required optimization by in vitro evolution,6 gene minimization,78 or other manipulations to enhance conjugation to desired recipients.79 In contrast, by tapping the natural diversity of conjugative systems optimized for specific bacterial genera, or specific species, issues such as conjugation efficiency, compatibility with resident plasmids, and stability have been evolutionary preoptimized for the microbial communities of interest. In summary, we constructed a large synthetic genetic system de novo from metagenomic data that contains at least 47 predicted genes and showed the conjugative function of the system. Codon optimization, removal of destabilizing repetitive sequences and toxic genes, and addition of biocontainment systems could be included in the synthetic biology strategy to fine-tune each conjugative system for its intended use. Similar approaches to those described here could be used to identify and functionalize other mobile genetic systems for microbiome editing. Methods Bacterial and Yeast Strains All yeast assemblies were performed using S. cerevisiae VL6-48 (MATa, his3Δ200, trpΔ1, ura3-52, ade2-101, lys2, psi + cir°). E. coli EPI300 (F′ λ–mcrA Δ(mrr-hsdRMS-mcrBC) ϕ80dlacZΔM15 Δ(lac)X74 recA1 endA1 araD139 Δ(ara, leu)7697 galU galK rpsL (StrR) nupG trfA dhfr) (Epicenter) as well as a diaminopimelic acid (DAP) auxotroph of the strain49 were used for cloning and as conjugative donors. C. rodentium DBS100, Salmonella enterica Typhimurium LT2, and kanamycin-resistant E. coli DBH179 strain were used as primary recipients. Seven additional Citrobacter spp. strains were used as recipients. A complete list of plasmids and strains used can be found in Table S6. Synthesis and Cloning of the 20298 Contig A list of oligonucleotides used in this study is listed in Table S7. The conjugative system identified from the conjugative contig “20298” was synthesized in 11 fragments from Twist Bioscience as clonal genes (Data S6). One fragment of the sequence was additionally synthesized on a Codex BioXP 3200. Fragments ordered from Twist Bioscience were released from their vectors using a PmeI digest, while the tile constructed using the BioXP 3200 was PCR amplified. Each tile contained 86-476 base pairs of homology with the adjacent fragment. The terminal DNA fragments contained homology with the backbone of plasmid pAGE1.0,49,80 which was originally derived from p0521s.81 pCitro-AGE was constructed with these fragments using a modified yeast assembly,47,48 which we have previously described in detail.7 Once it was identified that pCitro-AGE contained mutations in the traV and traY genes and that the pAGE replicon was nonfunctional in C. rodentium, the conjugative contig was recloned into the backbone derived from pNuc-trans.7 This contains a p15a origin, chloramphenicol-acetyltransferase, and CEN6-ARS4-HIS3. The conjugative contig was reamplified from pCitro-AGE with PCR, using primers that fixed the mutations in the traV and traY genes. The remainder of the plasmid was amplified in fragments with overlapping homology to the adjacent fragment. These fragments were then assembled in a yeast assembly as described above to generate pCitro-p15a. Sequences of all assembled plasmids are available in the Supporting Information. sgRNA Cloning A gateway cloning donor vector was constructed into the pDONR221 backbone, containing an arabinose-inducible TevCas9 endonuclease82 and an sgRNA cloning cassette flanked by attL recombination sites to form pENTR-TC9. A corresponding gateway destination cassette containing a ccdB toxin gene and ampicillin resistance gene flanked by attR sites was cloned into pCitro-15a, to generate pCitro-15a-dest. We used crisprHAL52 to identify sgRNAs predicted to have high killing efficiency against Citrobacter (Table S5). We cloned eight sgRNAs into pENTR-TC9 using the Golden Gate assembly at the BsaI cassette. A gateway LR reaction was performed and the resulting plasmid was transformed into E. coli EPI300 (Dap-) to insert pENTR-TC9 variants containing sgRNAs into pCitro-15a-dest, forming pCitro-TC9. Bacterial Conjugation Assays Donor and recipient strains were grown to saturation overnight in selective media. Saturated cultures were then diluted 1:50 into 5 mL nonselective LSLB (10 g/L tryptone, 5 g/L yeast extract, 5 g/L NaCl) supplemented with 60 μg/mL diaminopimelic acid (DAP) and grown to an A600 of 0.5–1.0 and adjusted to an A600 of 0.5. Cultures were centrifuged at 4000 RCF for 5 min and resuspended in 500 μL of 10% glycerol. Cells were aliquoted and frozen at −80 °C. For conjugation experiments, cells were thawed on ice, and 50 μL of donor strain was mixed with 50 μL recipient strain on LSLB plates supplemented with 60 μg/mL DAP. Conjugations proceeded on plates at 37 °C for 1 h. Cells were scraped into 400 μL of LSLB with a glass spreader. These cell suspensions were then serially diluted and plated on selection for donors, recipients, and transconjugants. Plates were incubated for 16–20 h at 37 °C and colonies were counted manually. Plasmid Stability Assays pCitro-15a was conjugated to C. rodentium, as described above. Transconjugants were isolated and passaged on selective plates with overnight growth at 37 °C for 16–20 h. Colonies were isolated from these plates and passaged once again on selective plates at 37 °C for 16–20 h. Individual colonies were picked from these plates and grown overnight in selective liquid LSLB at 37 °C. Plasmids were extracted and diagnostic digests were performed with NotI-HF and NcoI-HF (New England Biolabs). Digests were analyzed on agarose gels to check for plasmid rearrangements. C. rodentium Killing Assays pCitro-15-TC9 plasmids with sgRNAs cloned were conjugated from E. coli (Dap-) to C. rodentium on LSLB + Dap 60 μg/mL + 0.2% glucose plates at a 4:1 donor/recipient ratio for 1 h. Cells were scraped into 500 μL of LSLB and added to an additional 500 μL LSLB. 250 μL portion of the resulting cell suspension was diluted 1:2 into 2x selective and inducing LSLB (50 μg/mL chloramphenicol and 0.4% arabinose) and selective and 2x repressing LSLB (50 μg/mL chloramphenicol and 0.4% glucose). Cell suspensions are therefore at final concentrations of 25 μg/mL chloramphenicol and 0.2% arabinose for induced, and 25 μg/mL chloramphenicol and 0.2% glucose for repressed. Cells were then incubated at 37 °C with shaking for 2 h to induce TevCas9 expression. After 2 h of growth, the suspensions were centrifuged and resuspended in LSLB supplemented with 25 μg/mL chloramphenicol. Cell suspensions were then serially diluted and plated on LSLB supplemented with 25 μg/mL chloramphenicol and 0.2% glucose. Colonies were counted, and killing efficiency was calculated using the following formula. Identifying Conjugative Plasmids in Metagenomic Data Sets A detailed workflow is available at https://github.com/bjoris33/ce_synthesis_paper. In brief, a near-complete and nonredundant set of human gut microbiome genomes were downloaded from the European Bioinformatics Institute FTP site.1 A set of genomes that represent a near-complete array of bacterial species capable of colonizing the human intestinal tract was downloaded from the European Bioinformatics Institute FTP site.1 Open reading frames were predicted in the genome by Prodigal version 2.6.383 and the predicted protein sequences were then aligned to the UniRef90 database39 using the Diamond protein aligner version 0.9.14.84 Contigs were extracted from the genomes if they contained annotations for a relaxase/mobilization protein and a type IV secretion/type IV coupling protein by using a word-search strategy. Data S2 is a FASTA formatted file of the 1538 conjugative contigs and is available at https://figshare.com/authors/David_Edgell/15442121. Short-read data from 785 samples (Data S4)85−92 were downloaded from the Sequence Read Archive using the SRA toolkit version 2.9.2. Reads were deduplicated with “dedupe.sh”,93 and trimmed with Trimmomatic version 0.3694 with options “LEADING:10 TRAILING:10”. Subregions of the contigs where annotations for conjugative proteins were present with no more than 20 ORFs between successive UniRef90 annotations for conjugative proteins were extracted. The processed read data were mapped to the extracted conjugative systems using Bowtie2 version 2.3.595 with the settings “–no-unal–no-mixed–no-discordant”. Extraction of the subregions was to avoid an artificially high proportion of reads mapping in samples where the bacterial species is present, but the ICE has not integrated into its genome (Figure S1). The proportion of reads mapping to the conjugative systems was extracted from the Bowtie2 output, and the mapping data was visualized using Anvi’o.96 Identity as either a plasmid or chromosomal sequence was predicted using PlasFlow.42 Relaxase mobility grouping of the conjugative element was predicted using MOBscan.40 RNA-Seq Total RNA was prepared from three replicates of pCitro-15a electroporated into E. coli EPI300 (dap-) and C. rodentium. A single colony from each transformation was grown overnight to saturation under selection. Saturated cultures were diluted 1:50 into nonselective LSLB, and LSLB supplemented with 60 μg/mL DAP for C. rodentium and EPI300 respectively. Cultures were grown for 2 h at 37 °C with 225 rpm shaking, reaching OD600 between 0.4 and 0.6. 1.5 mL of culture were centrifuged was centrifuged at 16,000 RCF for 5 min at 4 °C. The supernatant was aspirated and the pellet was resuspended in 1 mL TRIzol reagent (Invitrogen). Samples were incubated at 65 °C for 10 min, and 0.2 mL of chloroform was added and mixed by inverting for 15 s. Samples were incubated at room temperature for 3 min before being centrifuged at 16,000 RCF for 5 min at 4 °C. The aqueous phase was transferred to a clean tube, and 1 volume of ethanol was added to each. An NEB Monarch cleanup kit was used by adding samples to an RNA cleanup column and centrifuging at 11,000 RCF for 1 min at room temperature. The column was washed twice by adding 500 μL of RNA cleanup wash buffer and centrifuging at 11,000 RCF. The column was spun for an additional 1 min at 11,000 RCF to dry before eluted into 50 μL nuclease-free water by centrifuging at 11,000 RCF for 1 min. Twelve libraries were prepared and sequenced using an Illumina NextSeq High Output 75 cycle sequencing kit on an Illumina NextSeq Reads were trimmed with Trimmomatic version 0.3697 with options LEADING:10 TRAILING:10. Processed reads for E. coli EPI300 samples were mapped to the E. coli strain K-12 NEB 5-alpha genome (CP017100.1) and pCitro-15a plasmid reference sequences using Hisat2 version 2.2.0.98 Processed reads for C. rodentium samples were mapped to the C. rodentium strain DBS100 genome (CP038008.1) and pCitro-15a plasmid reference sequences using the same workflow. Positional read coverage was determined using Samtools version 1.10.99 The RNaseq data set generated in this study has been deposited in the Sequence Read Archive with the accession number PRJNA971224. Acknowledgments This work was supported by a Project grant (PJT-159708) from the Canadian Institutes of Health Research to D.R.E., G.B.G., and B.J.K. T.A.H. was supported by a Postgraduate Doctoral Fellowship from the Natural Sciences and Engineering Research Council of Canada. We thank Emma Allen-Veroe (University of Guelph) for providing bacterial strains. Data Availability Statement The RNAseq data set generated in this study has been deposited in the Sequence Read Archive with the accession number PRJNA971224. Supporting Information Available The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acssynbio.3c00319. Strategy for mapping coverage of integrated conjugative elements; distribution of MOB family relaxes in the human gut metagenome; abundance and distribution of the 20298 conjugative system in human cohorts; diagnostic screening of pCitro assemblies; complementation of pCitro with compatible conjugative plasmids; stability of pCitro after passaging; and RNaseq analysis of pCitro in E. coli and C. rodentium (PDF) Strategy for mapping coverage of integrated conjugative elements; distribution of MOB family relaxes in the human gut metagenome; abundance and distribution of the 20298 conjugative system in human cohorts; diagnostic screening of pCitro assemblies; complementation of pCitro with compatible conjugative plasmids; stability of pCitro after passaging; and RNaseq analysis of pCitro in E. coli and C. rodentium (PDF) Predicted functions of the 20298 conjugative contig; oligonucleotides used in this study; CONJscan results; error report for pCitro-AGE assemblies; error report for pCitro-15a assemblies; predicted activity scores for Citrobacter sgRNAs; and strains and plasmids used in this study (ZIP) Predicted functions of the 20298 conjugative contig; oligonucleotides used in this study; CONJscan results; error report for pCitro-AGE assemblies; error report for pCitro-15a assemblies; predicted activity scores for Citrobacter sgRNAs; and strains and plasmids used in this study (ZIP) Identified contigs from human gut metagenomic data; FASTA file of identified conjugative systems; read coverage of conjugative contigs in short-read data; key for mapping short-read sample data to cohort information; sequence of the 20298 conjugative contig; sequences of the 20298 tiles for DNA synthesis; pOri-298 sequence; pCitro-AGE sequence; pCitro-15a sequence; RNaseq read coverage for E. coli and C. rodentium with pCitro-15a; pCitro-DEST sequence; pENTR-TC9 sequence; and pCitro-TC9 sequence (ZIP) Identified contigs from human gut metagenomic data; FASTA file of identified conjugative systems; read coverage of conjugative contigs in short-read data; key for mapping short-read sample data to cohort information; sequence of the 20298 conjugative contig; sequences of the 20298 tiles for DNA synthesis; pOri-298 sequence; pCitro-AGE sequence; pCitro-15a sequence; RNaseq read coverage for E. coli and C. rodentium with pCitro-15a; pCitro-DEST sequence; pENTR-TC9 sequence; and pCitro-TC9 sequence (ZIP) Author Contributions T.A.H., B.R.J., B.J.K., G.B.G., and D.R.E. conceived of the experiments, analyzed data, and wrote the paper. A.S. performed experiments and analyzed data. T.S.B. performed bioinformatics analyses of RNA-seq data. E.A.-V. and S.R. provided bacterial strains and plasmids, and wrote the paper. All authors reviewed the manuscript. The authors declare no competing financial interest. Supplementary Material sb3c00319_si_001.pdf sb3c00319_si_002.zip sb3c00319_si_003.zip References Associated Data Supplementary Materials sb3c00319_si_001.pdf sb3c00319_si_002.zip sb3c00319_si_003.zip Data Availability Statement The RNAseq data set generated in this study has been deposited in the Sequence Read Archive with the accession number PRJNA971224.