Multiplexed Competition in a Synthetic Squid Light Organ Microbiome Using Barcode-Tagged Gene Deletions

Beneficial microbes play essential roles in the health and development of their hosts. However, the complexity of animal microbiomes and general genetic intractability of their symbionts have made it difficult to study the coevolved mechanisms for establishing and maintaining specificity at the microbe-animal host interface.

. It is clear that molecular communication between animal hosts and their microbial partners leads to selection and retention of the cognate microbiome: while many microbes are obtained from the environment, the composition of mature microbiomes is often largely stable and resilient within members of a host species (9,10). While microbial communities have been characterized using metagenomic, transcriptomic, and metabolomic approaches (11), the complexity of animal-associated microbiomes and the inability to culture and genetically manipulate many symbionts make it difficult to study the precise molecular mechanisms that establish specific relationships.
The binary symbiosis between genetically tractable Vibrio fischeri and the Hawaiian bobtail squid Euprymna scolopes serves as a model system to study the molecular interactions underlying microbiome assembly (11)(12)(13)(14)(15)(16). The squid hatch aposymbiotically (without symbiont) and are colonized by V. fischeri in a multistep process that leads to the specific recruitment of the symbiont from a marine environment in which the bacteria are ,0.1% of the bacterioplankton (14,17). The symbionts are housed in the dedicated light organ (LO) within the squid's mantle cavity, where they generate light that the host uses for counterillumination to hide its shadow while hunting at night (18). The host provides the symbionts with a protected niche, nutrients, and oxygen (15). Once the symbiosis is irreversibly established in juvenile squid, a daily cycle proceeds where 90 to 95% of the bacteria are expelled from the LO at dawn. The remaining symbionts grow during the day until they fill the LO, and at night the dense population of symbionts provides light (19). Because the aposymbiotic hatchlings can be cultured in the lab and infected with genetically tractable V. fischeri, colonization experiments can be performed to study the molecular factors that play a role during this process (12,16,20). In addition, the translucent nature of the LO in squid hatchlings allows for visualization of the colonization process by microscopy (21)(22)(23).
Microbe-host signaling mechanisms and developmental transitions ensure specificity during colonization (14,17). Upon detection of bacterium-derived peptidoglycan, the ciliated appendages on the surface of the LO secrete mucus that traps bacteria circulating within the mantle cavity (13,24,25). In the mucus field, V. fischeri bacteria bind to cilia and form aggregates by expressing symbiosis polysaccharide (syp) genes, a locus of 18 structural and regulatory genes whose products contribute to biofilm formation (26,27). Approximately 3 to 4 h postinoculation, bacterial aggregates migrate through the host mucus toward the pores that lead into the LO ducts (25). While the initial migration is independent of flagellar motility (28), at the pores squid-produced chitin oligosaccharides serve as a chemoattractant to direct the symbiotic bacteria into the host crypts (21). Motility and chemotaxis are required for colonization, and strains with mutations in genes required for these processes-such as cheA, flrA, and rpoNare unable to successfully colonize the squid LO (21,28). Once within the LO V. fischeri generates light through expression of the lux operon in a quorum sensing-dependent process (29). Symbionts that fail to produce luminescence, such as strains with mutations in the autoinducer synthase genes ainS and luxI, or deletions of the lux operon, are unable to persist in the symbiosis (30,31).
To identify novel colonization factors in V. fischeri, our lab previously used a global transposon insertion sequencing approach (INSeq) to identify bacterial mutants that were depleted after 48 h in the squid host (32). This approach successfully identified previously known colonization factors, such as rscS, rpoN, ompU, various motility factors, and the syp biofilm locus, and in addition revealed 344 putative novel colonization factors. Twenty candidates were tested in competitive colonization assays of wild-type (WT) versus mutant strains, and the results showed that nine factors had colonization defects. Some of the validated factors encompass roles in protein quality control (DnaJ and DegS) and copper detoxification (CopA and CusC), inner membrane proteins predicted to play a role in secretion of autotransporters (TamB/YtfN), and other poorly characterized factors (YdhC, YafD, and YhcB). This global approach was crucial in identifying putative colonization factors. However, further study is required to address which genes are true colonization factors, when they act during colonization, and how their products modulate the interaction with the host.
Approximately 32% of putative colonization factors identified by INSeq did not fall into a curated Clusters of Orthologous Groups (COGs) category, suggesting that the ability to interrogate the function of these colonization factors will reveal novel biology. Traditional genetic engineering techniques in V. fischeri are either random (transposon mutagenesis) or labor-intensive (plasmid-based allelic exchange) (32)(33)(34). We therefore considered approaches by which we could isolate mutants and examine phenotypes in a multiplexed fashion. One possible approach was to retrieve transposon insertions of interest from an arrayed library (35)(36)(37). A second approach we considered was to adapt a newly developed method for transformation-mediated mutagenesis using linear DNA (38) with an in-frame barcoding strategy to facilitate precise mutations. The latter option was attractive in that we hoped that it would limit the effects of polar mutations and provide a set of defined deletions that can be characterized by amplicon PCR. Barcode sequencing (BarSeq), in which each strain is uniquely labeled and identified using high-throughput next-generation sequencing, has been used successfully to track population dynamics in multiple systems, including in yeast genomic libraries, during Vibrio cholerae infection, and to track and phenotype laboratory-evolved Escherichia coli (39)(40)(41)(42)(43). Here, we describe an approach to generate barcode-tagged gene deletions in V. fischeri and perform high-throughput colonization experiments using BarSeq. We also describe the barseq python computational package used to analyze the results.

RESULTS
Generation of barcoded gene deletions. To generate barcoded deletions of specific V. fischeri genes, we designed an approach that takes advantage of splicing by overlap extension PCR (SOE-PCR) and tfoX-based natural transformation ( Fig. 1) (38,(44)(45)(46). The first step uses PCR to amplify DNA fragments upstream and downstream of the gene targeted for deletion, fused to the left and right linker sequences, respectively (Fig. 1A). A separate PCR is performed with pHB1 as a template to generate the central fragment of DNA containing the linker sequences, a selectable marker (erm, conferring erythromycin resistance) surrounded by FLP recombination target (FRT) sites, and the semirandom barcode sequence. The barcode is provided by the reverse primer, which contains a region of semirandomized sequence. The three resulting DNA fragments-upstream, central, and downstream-are then fused into one fragment via their overlapping linker sequences by SOE-PCR (46) and transformed into V. fischeri upon tfoX induction (44). Finally, the erm cassette is removed via FLP recombinase (45).
The resulting 138-bp deletion scar (barcode scar, or "bar" scar, Fig. 1B) lies between the deleted gene's first codon and last seven codons (i.e., the final six amino acidencoding codons plus the stop codon). The scar is designed to be in-frame to prevent polar effects on gene expression when targeting genes within operons. The terminal codons were retained in case they contain a ribosomal binding site for downstream gene(s) (47). In addition to the barcode, the additional sequence in the scar includes left and right linker sequences that are shared among all of the mutants, which allows us to identify and quantify the abundance of each barcoded strain using amplicon sequencing, while minimizing amplification bias by using common primers that amplify the same size product.
To test this new approach, we investigated the copA gene. Among Gammaproteobacteria, CopA is the main exporter of cytoplasmic copper and is the most widely conserved copper detoxification factor (48,49). Although our laboratory previously demonstrated that copA is a squid colonization factor, its role in copper resistance has not been examined (32). We therefore targeted copA for deletion using our mutagenesis approach as a proof of concept and subsequently tested its role in copper resistance in V. fischeri. To ensure that the deletion process worked as intended, we used four sets of diagnostic PCR primers that would report on correct erm insertion, subsequent removal of the erm cassette by FLP recombinase, and absence of the targeted gene from the bacterial chromosome. PCR with various pairs of oligonucleotides that target the copA gene and its deletion constructs produced amplicons of the expected size in each strain ( Fig. 2A). These results show that the erm cassette was successfully inserted into copA generating DcopA::erm-bar and subsequently removed by FLP recombination to generate the in-frame deletion scar in DcopA::bar. Furthermore, sequencing of the deletion scar for several DcopA::erm-bar candidates showed that after a single round of mutagenesis, multiple uniquely barcoded deletion strains were generated (Fig. 2B). These results demonstrate that our deletion method is successful in generating uniquely barcoded mutant strains of V. fischeri. The presence of a barcode within a gene deletion does not alter mutant phenotypes. To test that the barcoded scar does not affect the mutant phenotypes, we measured the copper sensitivity of strains deleted for copA using various methods. In addition to the mutants generated using our deletion approach (DcopA::erm-bar and DcopA::bar), we constructed a deletion of copA using plasmid-based allelic exchange Vibrio fischeri Barcoded Deletions and BarSeq (DcopA) (33) and obtained a copA transposon mutant isolated from our previous study (copA::Tnerm) (32). We then tested the growth of these copA mutants in the presence of various amounts of copper. Our results show that, regardless of the mutagenesis method, the growth of copA mutants is similarly impeded in the presence of copper, with the severity of the growth defect increasing in proportion to the concentration of copper: at 0.2 mM Cu 21 , the copA mutants were able to grow slightly, whereas at 20 mM Cu 21 , these strains were unable to grow (Fig. 3A). This is in contrast to the WT strain that achieved the same growth yield regardless of the concentration of copper present. The DcopA::erm-bar and DcopA::bar mutants showed the same degree of copper sensitivity (Fig. 3A).
To corroborate that the observed growth defects were due specifically to excess copper, we measured the growth of the DcopA::bar mutant in the presence of copper, with and without the copper chelator bathocuproinedisulfonic acid (BCS). As expected, DcopA::bar was unable to grow in the presence of 20 mM Cu 21 , whereas the WT is unaffected (Fig. 3B). However, growth of DcopA::bar in the presence of copper was rescued by addition of 80 mM BCS (Fig. 3B), suggesting that free copper is indeed responsible for the observed lack of growth in the mutant. To verify that the absence of CopA was responsible for susceptibility to copper toxicity, we complemented copA at the chromosomal attTn7 site in the DcopA::bar strain and observed that growth was rescued in the presence of copper (Fig. 3B). On the basis of these results, we conclude that CopA is required for resistance to copper in V. fischeri, consistent with its function in other Gammaproteobacteria (48).
To further test our deletion method, we generated mutants in multiple genes required for V. fischeri motility-rpoN, flrA, and flaA-and tested the resulting strains' motility phenotypes on soft agar plates (28,50,51). We also included a WT strain tagged with the deletion scars at the attTn7 site (WT-1) and copA mutants as controls. While the motility of WT V. fischeri resulted in a migration disc with a diameter of 26 mm from the inoculation point on soft agar plates, deleting flaA resulted in a drastic reduction in migration (9 mm), while deleting either flrA or rpoN resulted in no motility (1.5 mm) (Fig. 4). We observed that both the erm-bar and bar versions of the gene deletions displayed equivalent phenotypes, showing that the strains behave as null alleles regardless of whether the scar contains the erm cassette (Fig. 4). The motility of both the WT-1 and copA strains is comparable to that of the WT, showing that motility defects are due to the deleted loci and not to the insertion of the deletion scars.
Removing the erythromycin resistance cassette minimizes polar effects of the barcoded deletions. While the presence or absence of the erm cassette does not prevent deletion strains from manifesting the corresponding phenotypes ( Fig. 3A and 4), we were concerned about polar effects on downstream gene expression upon insertion of the 1,049-bp heterologous erm cassette (52)(53)(54). To test the effect of the erm cassette on downstream gene expression, we used reverse transcriptase quantitative PCR (RT-qPCR) to measure expression of genes immediately upstream and downstream of a targeted gene deletion for three different predicted operons. In each case, we measured the ratio in expression levels of the downstream versus upstream genes in the mutant, normalized to the ratio in WT V. fischeri (defining this normalized value as the "polarity ratio"). For both rpoN and cheA, deletion scars of either erm-bar or bar resulted in negligible changes in the polarity ratio (Fig. 5). In contrast, the polarity ratio of DcusA::erm-bar was 26-fold higher than WT, whereas removal of the erm cassette to form the in-frame deletion scar restored the polarity ratio to basal levels (Fig. 5). We conclude that, in at least some cases, gene::bar deletion scars can alleviate collateral effects on flanking genes that are caused by inserting an antibiotic resistance cassette.
Development of a computational pipeline to analyze V. fischeri BarSeq data. With the ability to quickly generate precise barcoded deletions, we next sought to compete the Dgene::bar deletions en masse during host colonization. We therefore developed a BarSeq sample preparation protocol, and an accompanying computational package to analyze the data (Fig. 6). To accomplish this, we mixed barcoded strains to generate an input library (i.e., a synthetic microbiome). This library was then used to inoculate media and/or squid hatchlings, which were then sampled at the FIG 4 Assaying the phenotype of strains deleted for motility factors. Representative TBS agar trays showing the migration of strains from the inoculation point after incubation at 28°C for 4 h. WT is MJM1100 (ES114), while WT-1 represents the attTn7-marked MJM1100 strain with barcode 1 (either WT::erm-bar1 or WT::bar1). Bar graph shows the quantified data from five independent replicates with error bars showing the standard deviations of the means (n = 5). Statistical analysis was performed using a two-way ANOVA test. ****, P , 0.0001. desired time points. Samples were then processed to extract genomic DNA (gDNA), and PCR was performed with dual-index Illumina sequencing primers to obtain double-stranded DNA (dsDNA) fragments containing the barcoded deletion scars. The resulting library was then sequenced on an Illumina MiSeq and demultiplexed based on the unique dual indexes (55). The resulting sequence data were then analyzed using the barseq package, which identifies and counts the barcodes present in the samples, assigns strain identity, normalizes strain counts, and calculates relative frequency and the competitive index (CI) for each strain within the samples. The BarSeq protocol provides a streamlined and effective way to measure population dynamics throughout squid colonization.
BarSeq enables sensitive multiplex competition experiments. To test our BarSeq protocol in tracking individual strains within a population, we performed an in vitro competition and a competitive colonization experiment using an input library of seven different barcoded strains mixed in an equivalent ratio. In addition to several mutant strains, we included three WT::bar strains that had the bar scar inserted at the attTn7 site that could be similarly tracked using amplicon sequencing but without affecting the phenotypes of the strains (WT-1, WT-2, and WT-3 in Fig. 7). After 15 generations of growth in vitro, the proportion of most strains relative to the WT::bar strains remained stable except for flrA and rpoN deletion strains, which were fourfold higher and lower, respectively, compared to WT::bar (Fig. 7A). In contrast, following 48 h of squid colonization-which corresponds to approximately 15 bacterial generations (32)-we observed reduced levels of the flaA flagellin mutant and severely reduced levels of the flrA, rpoN, and cheA strains, all of which were near the limit of detection (Fig. 7B). This result is consistent with their previously known roles as necessary factors for squid colonization, although we did observe higher levels of flaA in the competitive colonization than are observed when a transposon insertion is competed against the wild-type strain (28,50,51,56). We note that there was relatively little variability among the WT:: bar strains in the analysis, whereas the 4-to 5-log-unit scale in which to identify colonization defects provided a substantially greater range over which to identify and refine colonization phenotypes in vivo (Fig. 7B). Taken together, these results show that our method for targeted barcoded deletions, multiplex squid colonization, and analysis by BarSeq allows for reproducible competition experiments in vitro and in vivo with high sensitivity.

DISCUSSION
In this study, we developed a method to quickly generate gene deletions where the resulting strains are tagged with unique DNA barcodes. We demonstrated the utility of these strains in performing BarSeq high-throughput competitive colonization experiments and introduced a software package to analyze the resulting sequencing data. BarSeq provides a sensitive method to track population dynamics of squid colonization by V. fischeri.
Generation of targeted, barcoded gene deletions that minimize effects on neighboring genes. Our approach builds upon previous SOE-PCR/tfoX mutagenesis techniques to incorporate a unique barcode in each deletion strain, which enables high-throughput experiments via barcode sequencing (BarSeq). Since BarSeq relies on amplicon sequencing, library preparation is straightforward and allows for a large number of samples to be processed in parallel. The method we have employed to design the barcode and flanking sequences was planned to minimize disruption on flanking genes. Expression of bacterial genes is frequently organized by their genetic   Table S2). Sequencing reads are analyzed by the barseq package to obtain counts for individual strains in a sample based on their unique barcodes. Those counts are then used to calculate the relative frequencies of individual strains at each time point and the competitive index (CI) as described in Materials and Methods. arrangement into operons where expression of operon members is driven by a common promoter (57). However, given that some regulatory regions overlap in neighboring genes, deletion of one gene can alter the expression level of another nearby cistron. These off-target effects on gene expression could obfuscate the analysis of experimental results. Similar to the approach used by Baba et al. (53), our deletion approach reduces off-target effects on gene expression by ensuring the formation of an in-frame open reading frame within the deletion scar and including several codons at the end of the deletion target where the ribosome binding site of downstream genes is frequently located (Fig. 1B and Fig. 5). We do note a potential limitation any time that foreign codons are introduced. The rarest codon within the scar (normalized to the set of codons for that amino acid) comprises 2% of the arginines encoded in the genome. It is formally possible that under growth conditions not tested here, for example, a different tRNA pool in the cell could theoretically lead to distinct effects. Our polarity assay results suggest that this is not an issue under typical laboratory growth conditions (Fig. 5). Nonetheless, there may be conditions in which the introduced DNA leads to substantive effects.
BarSeq enables detailed studies of the molecular mechanisms that result in establishment of the Vibrio-squid symbiosis. Using an INSeq screen, our lab previously identified 344 putative novel squid colonization factors in V. fischeri (32). Our deletion approach, combined with BarSeq, will enable the high-throughput characterization of these factors during squid colonization by allowing multiplexing of colonization factor mutants and tracking of individual strains. By enabling a more precise study of colonization factors, BarSeq has several potential applications.
BarSeq can be applied to the study of strain variation and evolution of colonization mechanisms in the Vibrio-squid symbiosis. V. fischeri strain variation is an important consideration when studying the mechanisms of colonization of the squid LO (58,59). Previous studies have shown that multiple strains can cocolonize the squid LO and that they do so at different rates (60)(61)(62). More recent studies have focused on deciphering the specific mechanisms that result in differing colonization behavior (63,64). The barcode-tagged mutagenesis method presented here can be applied to generate uniquely tagged WT or mutant strains of the various phylogenetically distinct V. fischeri strains and assayed in multiplexed format during squid colonization using BarSeq. We have already successfully used our SOE-PCR/tfoX mutagenesis approach to make targeted deletions in strain SR5, showing that this method is applicable to V. fischeri strains that are evolutionarily distant to the frequently used ES114 strain (63). BarSeq can also be used in directed evolution experiments to examine the functional evolution of colonization factors. Directed evolution experiments have recently been applied to study colonization factors in V. fischeri (65). The ease of tracking large numbers of individual V. fischeri strains using BarSeq could enable tracking of strain lineages in long-term evolution experiments, as has been conducted in other organisms (39,42,43).
Phenotypes of rpoN and flrA mutants during competitive growth in media. Both rpoN and flrA deletion strains showed a statistically significant fourfold decrease and increase, respectively, during competitive growth in media compared to WT (Fig. 7A). Due to the nature of the factors they encode, the observed growth defects are likely due to changes in energetic and nutritional requirements when rpoN and flrA are deleted. The rpoN gene encodes the alternative s 54 factor that is responsible for expression of various systems involved in squid colonization, including Syp biofilm formation, flagellar motility, and luminescence (51,66,67). Therefore, it is not surprising that deleting the gene encoding s 54 has pleiotropic effects on gene expression due to misregulation of the RpoN regulon and could reduce the ability of the mutant strain to effectively compete for growth in vitro, though further experiments are necessary to define the precise mechanism for the defect. FlrA is the s 54 -dependent transcription factor that activates expression of the flagellar biosynthesis cascade and is required for motility and squid colonization (50). The high energetic cost of expressing all genes related to flagellar biosynthesis (68), which in V. cholerae requires FlrA-dependent regulation of 52 genes (69) and in V. fischeri between 39 and 131 genes (28), is consistent with the observed increase in growth of the flrA deletion strain compared to WT during competitive growth in media (Fig. 7A). Nonetheless, even though the changes observed in competitive growth of the rpoN and flrA mutants in vitro are in opposing directions (less versus more growth, respectively), both are severely defective in squid colonization (Fig. 7B). Future experiments using BarSeq to probe bacterial growth in vitro and during colonization have the potential to elucidate heretofore hidden phenotypes.
Discrepancy in flaA colonization efficiency measured by BarSeq versus traditional competitive squid colonization experiments. In our BarSeq experiment, the known colonization factor flaA shows only a small (ca. twofold) colonization defect after 48 h post-squid inoculation (Fig. 7B). In contrast, Millikan and Ruby showed that a flaA deletion made by insertion of a Kan r cassette is severely defective during competitive colonization against WT V. fischeri (56). Using confocal microscopy, their work showed that LO colonization by flaA is delayed compared to WT by ;8 h. However, because our competitive colonization experiment using BarSeq was done at 48 h postinoculation, this delayed colonization is not enough to explain the observed discrepancy. Previous work has shown that the concentration of V fischeri in the inoculum can affect the number of different strains that can cocolonize the squid LO (70). This raises the possibility that the inoculum amount or the ratio of strains within the synthetic microbiome might affect the observed colonization defect. To address this, future experiments should examine how inoculum amount and the ratio of mutant strains to WT within a synthetic barcode-tagged population affects colonization efficiency for the different strains in the population.
In summary, we provide a new method for constructing barcoded deletions of V. fischeri genes, we demonstrate the utility of this method for generating in-frame deletions and discovering new functions of squid colonization factors, and we combine this approach with a computational tool to conduct multiplex animal colonization assays using barcode sequencing.

MATERIALS AND METHODS
Bacterial strains, growth conditions, plasmids, and primers. Bacterial strains used in this study are listed in Table 1, with Table S1 in the supplemental material containing an extended Table 1 showing the oligonucleotides used to generate the specified barcode-tagged gene deletions. Plasmids are listed in Table 2, and DNA oligonucleotides are listed in Table S2. DNA oligonucleotides were synthesized by Integrated DNA Technologies (Coralville, IA), and Sanger DNA sequencing was performed through the University of Wisconsin-Madison Biotechnology Center. Escherichia coli strains were grown in Luria-Bertani (LB) medium (per liter, 25 g Difco LB broth [BD], in distilled water) at 37°C with aeration. Unless otherwise indicated, V. fischeri strains were grown in Luria-Bertani salt (LBS) medium (per liter, 25 g Difco LB broth [BD], 10 g NaCl, and 50 ml 1 M Tris buffer [pH 7.0] in distilled water) at 25°C with aeration. When necessary, growth media was solidified by adding 15 g Bacto agar (BD) per liter. For growth of V. fischeri, antibiotics (Gold Biotechnology) were added at the following concentrations: 5 mg/ml erythromycin, 5 mg/ml or 2.5 mg/ml chloramphenicol as indicated, and 100 mg/ml kanamycin. For E. coli, the antibiotic concentrations used were 100 mg/ml carbenicillin, 25 mg/ml chloramphenicol, and 50 mg/ml kanamycin. The E. coli strain p 3813 containing pKV496 is a thymidine auxotroph and was grown in LB with 50 mg/ml kanamycin supplemented with 0.3 mM thymidine (38,71).
The unmarked deletion of copA in V. fischeri MJM1100 was made by allelic exchange as described previously (63). Briefly, 1.6-kb upstream (US) and 1.6-kb downstream (DS) sequences of copA were amplified by PCR using oligonucleotides HB44 and HB45 and oligonucleotides HB46 and HB47, respectively, and were cloned using Gibson Assembly (NEBuilder HiFi DNA assembly cloning kit) into the linearized vector pEVS79 (linearized using oligonucleotides HB52 and HB53) ( Table S2). The Gibson mix was transformed into E. coli NEB5a chemically competent cells and selected on chloramphenicol. The resulting pEVS79-DcopA candidates were screened using PCR with oligonucleotides HB54 and HB55 and DH5a lpir Ff 80lacZDM15 D(lacZYA-argF)U169 supE44 hsdR17 (r K m K 1 ) endA1 recA1 gyrA96 thi-1 relA1 uidA::pir 1

38, 71
a Thymidine auxotroph, growth conditions in Materials and Methods.
confirmed by sequencing, generating pHB3, which was conjugated into V. fischeri MJM1100 (ES114) via triparental mating with MJM534, which contains the helper plasmid pEVS104 (33). Single recombinants of pHB3 into the chromosome were screened and selected by growth on chloramphenicol (MJM3400), and double recombinants were screened and selected by loss of the antibiotic resistance cassette and copA (MJM3401). The resulting constructs were verified by PCR and sequencing (Table S2).
Construction of barcode-tagged gene deletions. The deletion protocol demonstrated in Fig. 1A is based on splicing by overlap extension PCR (SOE-PCR) and tfoX transformation (38,(44)(45)(46) to directly delete and tag targeted genes with a randomized sequence (barcode). Our protocol was in development prior to publication of the previous method (38), so while it is conceptually similar, the sequences of the linkers and primers are distinct. First, several oligonucleotides were designed specific to the targeted genes to amplify 1 kb of US (F1 and R1-LL) and DS (F2-RL and R2) DNA tagged with the left linker (LL) and right linker (RL) sequences, respectively, to screen the deletion scar via PCR (FO and RO), and to assay for the absence of the targeted gene (FW and RW) (Fig. 1A, Table S1, and Table S2). FW and RW were designed to amplify a fragment of 500 to 1,000 bp, depending on the size of the gene. The F1 and R2 oligonucleotides were designed to anneal 1-kb US and DS, respectively, of the targeted gene. The R1 oligonucleotide was designed to anneal starting at the start codon of the targeted open reading frame (ORF) going upstream, then the reverse complement of the LL sequence (LL reverse complement, 59-CTGGCGAAGCATATATAAGAAGCTCGTCTCGT-39) was attached to the 59 end of the R1 oligonucleotide, resulting in R1-LL. The F2 oligonucleotide was designed to anneal at the last seven codons (6 amino acids [aa] and stop codon) on the 39 end of the targeted ORF going downstream, then the RL sequence (RL, 59-GACTTGACCTGGATGTCTCTACCCACAAGATCG-39) was attached to the 59 end of the F2 oligonucleotide, resulting in F2-RL. The FO and RO oligonucleotides (forward outside and reverse outside, respectively) were designed to anneal 500 bp away from the annealing sites of F1 and R2, respectively, and were used to probe the targeted genomic region for insertion of the desired deletion scar.
The middle dsDNA fragment containing the erm cassette flanked by FRT sites and the randomized barcode was obtained by PCR with Phusion Hot Start Flex 2Â master mix (New England BioLabs [NEB]; catalog no. M0536L) and pHB1 as the template, which contains the LL-FRT-erm-FRT-spacer sequences and was built as described previously (63), and oligonucleotides HB42 and HB154. Oligonucleotide HB154 is a reverse primer and contains the RL sequence, 18 bp of randomized sequence composed of six trimers of "NNB" to prevent formation of stop codons (results in "VNN" codons in the forward direction), and the spacer sequence (Table S2). The resulting 1,049-bp product containing LL-FRT-erm-FRTspacer-random barcode-RL was purified by gel extraction using a QIAquick gel extraction kit (Qiagen; catalog no. 28706). The flanking 1-kb US and DS fragments for each targeted gene were then fused to this middle DNA fragment via the homology between the LL and RL sequences and using SOE-PCR with the F1 and R2 oligonucleotides, resulting in the 3-kb mutagenic dsDNA. The reaction mixture contained 10 ng of each of the middle, US, and DS DNA fragments, 200 nM concentration of the corresponding F1 and R2 oligonucleotides (Table S2) The 3-kb mutagenic DNA fragments were purified using a QIAquick PCR purification kit (Qiagen; catalog no. 28106) and transformed into V. fischeri ES114 via natural transformation with pLostfoX (MJM1538) (32,44) where the flanking sequences guide the barcoded erm cassette to substitute the targeted gene. Mutant candidates were selected on LBS with 5 mg/ml erythromycin (LBS-Erm5) and screened by PCR with oligonucleotide pairs F1/R2, FO/HB8, and FW/RW (as shown in Fig. 2A). The insertion of the erm-bar scar was confirmed by Sanger sequencing with primers HB8, HB9, HB42, and HB146, and the unique barcode sequence was recorded for each strain.  (38) and helper strain MJM534 (CC118 lpir/pEVS104) into recipient V. fischeri strains containing the erm-bar scar and selection on LBS containing 100 mg/ml kanamycin (LBS-Kan100). Plasmid pKV496 contains the FLP recombinase that removes the erm cassette and fuses the two surrounding FRT sites into one, resulting in the final bar scar as shown in Fig. 1B. The plasmid was eliminated by growing the candidates on LBS without selection twice and selecting colonies that were Erm s and Kan s . The gene::bar candidates were screened by PCR using oligonucleotide pairs F1/R2, FO/HB146 (RL), and FW/RW, and the deletion scar was verified by Sanger sequencing using oligonucleotides HB42 and HB146. The barcode sequences were verified to match the barcode within the parental strains containing the gene::erm-bar scar.
The barcoded WT V. fischeri strains (WT::bar) were constructed using the same procedure as outlined above for the gene deletions but targeting a site next to the attTn7 site in the intergenic region of yeiR and glmS. The 1-kb US and DS arms were amplified using PCR with ES114 gDNA and oligonucleotide pairs HB239/HB240 and HB241/HB242. After SOE-PCR to form the mutagenic DNA and tfoX transformation, the WT::erm-bar candidates were screened by PCR with oligonucleotide pairs HB243/HB244 and HB243/HB8. Sanger sequencing was used to confirm insertion of the erm-bar scar and record the unique barcode sequences. Triparental mating as described above was performed to remove the erm cassette using pKV496. The bar scar was confirmed by PCR with HB243/HB146 and Sanger sequencing.
Growth assays in the presence of copper. Colonies from freshly streaked LBS plates of the indicated V. fischeri strains were inoculated into 3 ml LBS with the appropriate antibiotics and grown for 8 h at 25°C with shaking. Three microliters of the LBS cultures was subcultured into 3 ml Tris minimal medium [per liter, 500 ml defined seawater (DSW; 2X), 50 ml of 1 M Tris base [pH 7.5], 1 ml of 5.8% K 2 HPO 4 , 1 ml of 10 mM FeSO 4 , and 20 ml of 10% N-acetylglucosamine (GlcNAc), in distilled water; DSW (2Â) = 100 mM MgSO 4 , 20 mM CaCl 2 , 600 mM NaCl, and 20 mM KCl] and incubated at 25°C overnight for #16 h. Overnight cultures were diluted to an optical density at 600 nm (OD 600 ) of 0.5 in 200 ml, and then 2 ml of solution with an OD 600 of 0.5 was transferred into 198 ml of fresh Tris minimal medium containing the appropriate amounts of copper and/or bathocuproinedisulfonic acid (BCS) in a 96-well plate.  8.63 g MgSO4, and 3 g agar, in distilled water) by stabbing the toothpick into the media at a single spot. The trays were incubated at 28°C for 4 h, and the outer diameter of swimming cells was measured.
Measuring polarity ratio via RT-qPCR. The indicated bacterial strains were grown in 3 ml LBS with the appropriate antibiotics and grown at 25°C overnight. On the day of the experiment, 15 ml of the overnight cultures was transferred into 3 ml of fresh LBS, and growth was continued at 25°C with aeration. Samples were harvested at an OD 600 of 0.2 to 0.4 (mid-log phase) by transferring 800 ml of culture into a 2-ml screw-cap tube containing 100 ml of a cold 95% ethanol (EtOH)-5% phenol solution that inactivates RNases (74). RNA extraction and RT-qPCR were performed as described previously (75). Briefly, cells were lysed in Tris-EDTA (TE) buffer (10 mM Tris-Cl [pH 8.0], 1 mM EDTA) containing lysozyme (Epicentre; catalog no. R1804M) and 1% sodium dodecyl sulfate (SDS). RNA was extracted using the hot phenol method (74) and digested with DNase I (NEB; catalog no. M0303S).
cDNA was synthesized from 0.5 mg of total RNA using the iScript Advanced cDNA synthesis kit (Bio-Rad; catalog no. 1725037) following the protocol 25°C for 5 min, 46°C for 20 min, and 95°C for 1 min. Quantitative PCR was performed using 1:10 dilutions of cDNA synthesis products with the iTaq Universal SYBR green supermix (Bio-Rad; catalog no. 1725121) on a CFX Connect real-time PCR detection system (Bio-Rad). The qPCR protocol was 95°C for 30 s and 40 cycles, with each cycle consisting of 95°C for 5 s and 58°C for 30 s with a final melt curve analysis to ensure specificity in the reaction. The mRNA levels of rpoD, lptB, hpf, cheZ, cheB, cusB, and cusF were measured using the oligonucleotide pairs listed in Table S2. Expression levels for each gene were normalized to that of rpoD, and the mutants were normalized to WT using the 2 2DDCT method (76). The polarity ratio of rpoN, cheA, and cusA was calculated as "expression of the downstream gene/expression of the upstream gene" using the respective flanking genes in each putative operon-lptB-rpoN-hpf, cheZ-cheA-cheB, and cusB-cusA-cusF. Operons were predicted using the BioCyc database for "Aliivibrio fischeri strain ES114, version 24.1," which is based on the sequenced genome in Mandel et al. (77,78).
barseq bioinformatic tool. To quantify barcodes within each sequenced sample, we developed barseq (https://github.com/mjmlab/barseq), a python package that identifies putative barcodes in the sequenced reads and matches them to a user-provided barcode library. The program iterates through each sample and uses regular expressions to search within the reads for flanking sequences on the left (GCTCATGCACTTGATTCC; spacer sequence) and the right (GACTTGACCTGGATGTCT; right linker sequence) of the barcode (Fig. 1B), while also allowing for 18 random nucleotides that represent a candidate barcode. The putative barcode sequence is then mapped against the reference barcode library and increases the count for the matched strain. barseq outputs a tab-delimited table with the barcode/strain counts for each of the samples analyzed.
Barcode sequencing and multiplexed competitive experiments. Cells of the indicated strains (Fig. 7) were grown in 3 ml LBS at 25°C overnight with aeration. The cultures were then diluted (1:80) into 3 ml fresh LBS and grown to mid-log phase (OD 600 of 0.2). Equivalent ODs of cells from each strain [volume to mix calculated as volume (in microliters) = (1.25/OD 600 ) Â 50] were mixed, resulting in a multiplexed population with each strain present at a 1-to-1 ratio. A sample from this input library was harvested by collecting cells from 700 ml by centrifugation and storing the cell pellet at -80°C. The input library was then used to inoculate hatchling Euprymna scolopes squid at 5 Â 10 3 to 9 Â 10 3 CFU/ml for 3 h in FSIO (filter-sterilized Instant Ocean) as previously described (20). Squid samples (n = 24, per replicate) were harvested at 48 h postinoculation and surface sterilized by storing at -80°C. Concurrently to squid colonization, the input library was competed for growth in vitro for 15 generations by diluting the library 1:181 into LBS, growing at 25°C with aeration back to the starting OD 600 , repeating this process once more, and harvesting samples as described above. Individual squid were homogenized in 700 ml of FSIO, 500 ml of each homogenate was mixed in a 50-ml conical tube, diluted 1:20 in 70% IO (Instant Ocean), and 50 ml plated onto LBS plates in triplicate. After a 17-h overnight incubation at 25°C, the bacterial colonies from each plate were scraped with a sterile cell scraper into 1 ml of 70% IO and collected by centrifugation. Cell pellets were stored at -20°C prior to DNA extraction.
Genomic DNA from the cell pellets was extracted and purified using the Qiagen DNeasy blood and tissue kit (Qiagen; catalog no. 69506) following the Gram-negative bacteria protocol and quantified using a NanoDrop spectrophotometer (Thermo Fisher Scientific). The barcoded scars were amplified together with dual-index Illumina sequencing primers (55). The reaction mixtures contained 50 ng of gDNA, 200 nM each oligonucleotide (Table S2), 1Â Phusion Hot Start Flex Master Mix (NEB; catalog no. M0536L), and H 2 O up to a total volume of 50 ml. PCR conditions were 98°C for 30 s, 20 cycles with each cycle consisting of 98°C for 10 s, 60°C for 10 s, and 72°C for 10 s, with a final extension step at 72°C for 5 min. PCR products were visualized using a 2% agarose gel to confirm the dual-indexed amplicon of 231 bp and purified using a QIAquick PCR purification kit (Qiagen; catalog no. 28106). Purified PCR products were quantified using a Qubit 3 fluorometer (Life Technologies), pooled in equimolar amounts, and diluted to 4 nM. The pool was sequenced on an Illumina MiSeq using the 2 Â 250 bp v2 kit with a 10% PhiX control following the manufacturer's protocol (Illumina, Inc., San Diego, CA) and using custom primers developed from reference 55. The sequencing data were processed using the barseq python package to obtain strain counts per sample, and mutants that were in the input library but still being validated were removed from the data set. The relative frequency (RF) for each strain in a sample was calculated, normalized to the RF in the input library and the average RF in the sample, and the competitive index (CI) was then calculated using the formula: CI = log 10 [(RF mutant /average RF WT ) Sample /(RF mutant / average RF WT ) Input ].

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.