Tn-Seq Analysis Identifies Genes Important for Yersinia pestis Adherence during Primary Pneumonic Plague

Colonization of the lung by Yersinia pestis is a critical first step in establishing infection during primary pneumonic plague, a disease characterized by high lethality. However, the mechanisms by which Y. pestis adheres in the lung after inhalation remain elusive. Here, we used Tn-seq to identify Y. pestis genes important for adherence early during primary pneumonic plague. Our mutant enrichment strategy resulted in the identification of genes important for regulation and assembly of genes and proteins rather than adhesin genes themselves. These results reveal that there may be multiple Y. pestis adhesins or redundancy among adhesins. Identifying the adhesins regulated by the genes identified in our enrichment screen may reveal novel therapeutic targets for preventing Y. pestis adherence and the subsequent development of pneumonic plague.

YPO3903, which encodes a hypothetical protein. We demonstrated a significant role for YPO3903 in Y. pestis adherence in the lung and in regulating transcript levels of psaA, the deletion of which had a minor effect on Y. pestis adherence in the lung. The enrichment of mutants that likely have altered gene expression and levels of multiple proteins suggests functional redundancy or the presence of multiple adhesins in Y. pestis.

RESULTS
Yersinia pestis genes important for adherence in the lung are unknown. The earliest events of primary pneumonic plague, particularly how Y. pestis interacts with and adheres to cells in the small airways, are unknown. To characterize the initial adherence of Y. pestis in the lung, we used our mouse model of primary pneumonic plague to determine the proportion of adherent bacteria early after inoculation (2). Mice were inoculated intranasally with 1 ϫ 10 4 CFU Y. pestis CO92 (wild type), and 2 h postinoculation (hpi), we performed a bronchoalveolar lavage of the lungs to remove any nonadherent Y. pestis and enumerated the proportion of Y. pestis that remained in the lung. We observed that approximately 75% of Y. pestis was adherent in the lung 2 h after inoculation (Fig. 1A, black bar).
The attachment invasion locus (Ail) has been implicated in adherence for Y. enterocolitica and Y. pestis to various cell lines, and a Y. pestis ail mutant exhibits reduced virulence following intranasal inoculation (16,20,30). Therefore, ail may play a role in adherence of Y. pestis in the lung early after inoculation. Additionally, it was recently shown that the Y. pestis type III secretion system (T3SS) needle tip protein, LcrV, binds the N-formylpeptide receptor (FPR1) on immune cells for delivery of T3SS effectors (31). It is therefore possible that T3SS binding to host FPR1 mediates Y. pestis adherence in the lung. To test the role of ail and the T3SS in Y. pestis adherence early during primary pneumonic plague, we inoculated mice intranasally with 1 ϫ 10 4 CFU Δail Y. pestis or pCD1 Ϫ Y. pestis, which lacks the plasmid encoding the T3SS needle apparatus and effectors (25). At 2 hpi, we performed a bronchoalveolar lavage and calculated the percentage of adherent Y. pestis for each strain. Approximately 75% of Y. pestis Δail and pCD1 Ϫ was adherent in the lung, similar to the levels of adherence for wild-type Y. pestis (Fig. 1A). Total bacterial burdens in the lung for the three Y. pestis strains were also similar (ϳ10 4 CFU) at 2 hpi (Fig. 1B). Thus, despite the putative role of ail in Y. pestis adherence in vitro and LcrV binding to FPR1, our data suggest that these surface structures do not individually play a major role in mediating Y. pestis adherence in the lung early during primary pneumonic plague.
To identify genes important for Y. pestis adherence in the lung, we took an unbiased genetic approach using Tn-seq to screen comprehensive libraries of Y. pestis mutants for genes required for adherence. We predicted that mutants with transposon insertions in genes important for adherence in the lung would be enriched in the bronchoalveolar lavage fluid (BALF) collected from mice following intranasal inoculation with a Y. pestis transposon library. To enhance the enrichment of nonadherent mutants, we also created a comprehensive library of transposon insertional mutants in a Y. pestis Δcaf1 strain in addition to the wild-type Y. pestis mutant library. The caf1 gene encodes the F1 antigen, which forms a large capsule-like structure around Y. pestis at 37°C (26,27). The presence of the F1 antigen on the surface of Y. pestis masks surface structures and adherence in vitro (23,28). Therefore, the Δcaf1 strain background may provide greater enrichment of nonadherent mutants without the masking effect of the F1 antigen.
A schematic depicting our enrichment strategy is shown in Fig. 1C. Briefly, we generated transposon input libraries of approximately 233,000 mutants for wild-type Y. pestis and approximately 208,000 mutants for Y. pestis Δcaf1. We then inoculated two groups of three mice each intranasally with either 1 ϫ 10 6 CFU Y. pestis wild-type transposon mutants or 5 ϫ 10 6 Y. pestis Δcaf1 transposon mutants. Using a higher inoculum than our standard dose of 1 ϫ 10 4 CFU ensured delivery of all transposon mutants to the lung. At the same time, a portion of each input library was inoculated into two replicate in vitro liquid cultures and grown with shaking at 37°C for in vitro growth controls. At 2 hpi, we collected BALF (containing Y. pestis mutants with adherence defects) from mice and a portion of each in vitro culture, and we plated these groups of mutants separately on selective media. Portions of these enriched mutants were then used to start liquid cultures for the next round of infection, keeping the two biological replicates for the in vivo and in vitro groups separate. This enrichment protocol was repeated three more times for a total number of four rounds of enrichment in vivo and in vitro. After each round of enrichment, genomic DNA was isolated from each group of Y. pestis mutants for sequencing of the transposon junctions to identify the enriched mutants.
Enrichment of less-adherent Y. pestis mutants after serial passaging of transposon libraries. During our Tn-seq screen, we calculated the proportion of adherent The Yersinia pestis genes important for adherence in the lung are unknown. Mice were inoculated intranasally with 1 ϫ 10 4 CFU of the Y. pestis wild-type, Δail, or pCD1 Ϫ strains, and bronchoalveolar lavage was performed at 2 hpi. CFU were enumerated in both the bronchoalveolar lavage fluid (BALF) and the lung. (A) Percent adherence was calculated for each strain by determining the proportion of Y. pestis in the lung compared to total CFU in both the BALF and the lung. Significance was determined by one-way ANOVA with Tukey's multiple correction. ns, not significant. (B) Total Y. pestis CFU enumerated in both the lung and the BALF. Data are representative of two independent experiments with 3 mice per group and are means and SD. (C) Schematic depicting the screen method for serial enrichment of nonadherent Y. pestis transposon mutants in both wild-type and Δcaf1 strains. The input library was inoculated into 2 groups of 3 mice each or 2 in vitro culture flasks. At 2 hpi, bronchoalveolar lavage was performed on the mice to collect nonadherent Y. pestis mutants and a portion of the in vitro cultures were removed. These mutants were then used to inoculate new groups of mice and in vitro cultures, repeating the in vivo enrichment or in vitro controls, and this process was repeated for a total of four times to collect mutants for high-throughput sequencing to determine the identity of the enriched mutants.
bacteria after each round of enrichment in vivo for both Y. pestis wild-type and Δcaf1 groups. The proportion of adherent wild-type Y. pestis insertional mutants in the lung decreased from 70% after two rounds of enrichment to ϳ60% by three rounds and ϳ50% by four rounds of enrichment ( Fig. 2A). There was also a reduction in adherence for Y. pestis mutants in the Δcaf1 background over four rounds of enrichment. There was approximately 85% adherence after two rounds of selection, and this decreased to ϳ70% adherence by three rounds and ϳ65% adherence by four rounds of enrichment (Fig. 2B). Together, these data indicate that there was significant enrichment of nonadherent mutants by serially passaging both the Y. pestis wild-type and Δcaf1 transposon libraries through the lungs of mice. Additionally, the Y. pestis Δcaf1 mutants were more adherent in the lung than the wild-type Y. pestis mutants at each round of selection, supporting previous observations that the F1 antigen masks Y. pestis adherence factors.
Correlated with the enrichment of nonadherent mutants after each in vivo passage of the transposon libraries, we also observed an increase in the proportion of an altered colony morphology for Y. pestis mutants collected in the BALF after each round of enrichment. The colonies were larger with a less-defined border compared to those of wild-type Y. pestis, which is indicative of an ail mutant (Fig. 2C) (21). For the nonadherent mutants in the wild-type background, approximately 6% of the colonies recovered in the BALF had the Δail-like colony morphology after two rounds of enrichment, which increased to 40% after three rounds and 70% after four rounds of enrichment (Fig. 2D, red bars). The enrichment for Δail-like mutants was stronger in the Δcaf1 background, with 10% of nonadherent mutants having the Δail-like colony morphology after two rounds, increasing to 70% by three rounds and 90% by four rounds of FIG 2 Enrichment of less-adherent Y. pestis mutants after serial passaging of transposon libraries. Mice were inoculated intranasally with 1 ϫ 10 6 CFU of the Y. pestis wild-type transposon library or 5 ϫ 10 6 CFU of the Y. pestis Δcaf1 transposon library. Bronchoalveolar lavage was performed at 2 hpi, and CFU were enumerated in both the bronchoalveolar lavage fluid (BALF) and the lung. Percent adherence was calculated after two, three, and four rounds of in vivo enrichment by determining the proportion of Y. pestis transposon mutants in the lung compared to the total CFU of transposon mutants in both the BALF and the lung for wild-type (A) and Δcaf1 (B) backgrounds. P values (A and B) were determined by one-way ANOVA with Tukey's multiple correction. (C) Colony morphology of a wild-type Y. pestis colony and a Δail Y. pestis colony, both in the pCD1 Ϫ strain background, grown at 26°C for 2 days on BHI agar. (D) The proportion of colonies in the BALF with a Δail-like morphology at two, three, and four rounds of enrichment in vivo and four rounds of enrichment in vitro for both wild-type (red bars) and Δcaf1 libraries (blue bars). Data are means and SD. enrichment (Fig. 2D, blue bars). However, we also observed enrichment of Δail-like mutants in the in vitro control groups. After four rounds of enrichment in vitro, ϳ25% of the colonies in the wild-type in vitro controls had a Δail-like colony morphology, whereas nearly 75% of colonies in the Δcaf1 background had a Δail-like morphology (Fig. 2D). The in vitro enrichment suggests that Δail mutants may have a competitive growth advantage during the in vitro culture steps between inoculations that contributed to their enrichment during the in vivo screen, particularly in the Δcaf1 background. Due to the large proportion of Δail-like colonies (and thus likely mutants with transposon insertions in ail or genes that regulate ail) saturating the mutant pools after four rounds of enrichment, we performed high-throughput sequencing and subsequent analysis of the pools of nonadherent mutants collected after two and three rounds of selection.
Significant enrichment of mutations in genes involved in the regulation and assembly of macromolecules. We performed high-throughput sequencing using the MiSeq platform, sequencing the transposon junction to identify which mutants were enriched by the screen. We determined that approximately 25% of all TA sites in the genome contained a transposon insertion in the Δcaf1 input library, and 77% of annotated genes in the genome had at least one TA site disrupted. For the wild-type input library, approximately 37% of all TA sites in the genome contained a transposon insertion, and 83% of annotated genes had at least one TA site disrupted. Considering that about 10 to 15% of Y. pestis annotated genes were calculated to be essential (32), our input libraries had relatively good coverage of the genome. The number of unique reads for each gene was counted, upper quartile normalized, and log 2 transformed. As there were two replicates for the in vivo and in vitro groups, we compared the similarity of the read count for each gene between the two replicates for each condition (Fig. S1). We observed a very strong correlation between the two replicates for each condition, with R 2 values of Ͼ0.99 for every condition except Δcaf1 2 rounds in vitro, which has an R 2 value of 0.94. This indicates a high level of reproducibility among the technical replicates in the screen. Thus, we averaged the replicate read count for each gene.
We then calculated the enrichment value for each gene. The enrichment value equals the fold change in the read count for a given gene after two and three rounds of enrichment relative to the read count for that gene in the input library. The goal of our screen was to identify significant enrichment of mutants with insertions in genes involved in Y. pestis adherence, so we focused on genes with enrichment values that were greater than two standard deviations (SD) from the mean after both two rounds and three rounds of enrichment in vivo. In the wild-type background, there were 51 genes that had enrichment values Ͼ2 SD from the mean (Fig. 3A, red dots). In the Δcaf1 background, there were 67 genes with enrichment values Ͼ2 SD from the mean (Fig. 3B, blue dots). For these significantly enriched mutants, we wanted to eliminate any that were enriched due to enhanced in vitro growth. To this end, we subtracted the in vitro enrichment values from the in vivo enrichment values for each significantly enriched gene ( Fig. 3C and D). In the wild-type background, there were 18 genes that had greater enrichment values in vivo over in vitro (Fig. 3C, purple dots). In the Δcaf1 background, there were 34 genes with greater enrichment values in vivo relative to in vitro (Fig. 3D, green dots). The identity and enrichment values for each of these genes are listed in Table 1.
Of the 18 genes identified in the wild-type background and the 34 genes identified in the Δcaf1 background, six genes were shared between the two groups: clpX, tig, iscR, rnc, YPO3903, and YPO3904 (Fig. 3C). This small overlap in the identity of the significantly enriched mutants is reflective of comparing the enrichment values for every gene in the wild-type background relative to the enrichment value for the same gene in Δcaf1 background, where there was poor correlation at both two and three rounds of enrichment (Fig. S2). Because mutations in the six genes were significantly and independently enriched in both strain backgrounds, we hypothesized that these genes may play the greatest role in Y. pestis adherence in the lung. The six genes are all encoded on the Y. pestis chromosome, and none of these genes are annotated as adhesins, fimbriae, or other similar structures (33). Instead, these genes are all involved in either the regulation or assembly of macromolecules. The genes tig, which encodes trigger factor, and clpX both encode proteins that exhibit chaperone functions for structures that assemble in the bacterial cytoplasm (34,35). The genes iscR and YPO3904 encode transcriptional regulators, rnc encodes an endoribonuclease, and YPO3903 encodes a hypothetical protein. By enriching for mutations in genes that are involved in regulating expression of multiple genes or assembly of multiple proteins, these data suggest that there may be several Y. pestis adhesins important for adherence during primary pneumonic plague.
YPO3903 contributes significantly to Y. pestis early adherence in the lung. To evaluate the effect on Y. pestis adherence for the six significantly enriched mutants in both the wild-type and Δcaf1 backgrounds, we first created individual clean deletion strains for each gene, except rnc, in the wild-type Y. pestis background. RNase III, the endoribonuclease encoded by rnc, regulates rRNA processing, and rnc deletions result in global transcriptome changes that would be difficult to distinguish from other mRNAs that rnc may more specifically regulate (36)(37)(38). Therefore, due to the global, nonspecific effects of the rnc deletion, we chose not to further explore the effects of rnc in Y. pestis adherence. We inoculated mice intranasally with 1 ϫ 10 4 CFU of wild-type Y. pestis and the ΔclpX, ΔiscR, and Δtig deletion strains. At 2 hpi, we performed a bronchoalveolar lavage to remove nonadherent bacteria and calculated the proportion of adherent Y. pestis. Y. pestis ΔclpX, Y. pestis ΔiscR, and Y. pestis Δtig all had similar or even slightly higher levels of adherence (65 to 70% adherence) in the lung relative to the ϳ65% adherence observed for wild-type Y. pestis (Fig. 4A). We next inoculated groups of mice with the single-deletion mutants Y. pestis ΔYPO3903 and Y. pestis ΔYPO3904 and compared the adherence at 2 hpi to that in mice inoculated with wild-type Y. pestis. Y. pestis ΔYPO3903 had a significant reduction in adherence, with an Tn-Seq Analysis of Y. pestis Pulmonary Adherence average of 55% adherence compared to 75% for wild-type Y. pestis (Fig. 4B). Y. pestis ΔYPO3904 had a small but statistically insignificant reduction in adherence relative to wild-type Y. pestis (Fig. 4B). Additionally, each Y. pestis strain had a similar total bacterial burden in the lung at 2 hpi, with ϳ10 4 CFU recovered ( Fig. 4C and D). These data suggest that YPO3903 plays a significant role in early Y. pestis adherence in the lung. a Values are in vivo enrichment values for the genes that were Ͼ2 SD from the mean and had greater in vivo enrichment after two and three rounds of enrichment, with the corresponding in vitro values in parentheses. The first group of genes (no highlighting) fit these criteria for both the wild-type and the Δcaf1 backgrounds, the second group (highlighted in light gray) fit these criteria only for the wild-type background, and the third group (highlighted in dark gray) fit these criteria only for the Δcaf1 background.
YPO3903 regulates psaA levels, which contributes to Y. pestis adherence in the lung. Of the five mutations we analyzed for their role in Y. pestis adherence in vivo, only deletion of YPO3903 had a significant effect on Y. pestis adherence in the lung. We explored the mechanism by which YPO3903 could modulate adherence in the lung early after inoculation. YPO3903 encodes a hypothetical protein that has 74% identity to the Escherichia coli protein YifE/MaoP (UniProt accession no. A0A384LBX8) (39). MaoP is conserved in Enterobacteriaceae and is involved in control of chromosome conformation and segregation (40). As chromosome organization could affect transcription, we hypothesized that the function of YPO3903 could be to regulate the expression of multiple genes involved in Y. pestis adherence in the lung. This hypothesis aligns with our prediction that there may be multiple factors involved in mediating adherence in the lung rather than a single Y. pestis adhesin.
To test if YPO3903 regulates genes involved in adherence, we measured transcript levels of a small subset of Y. pestis genes previously implicated in in vitro adherence: ail, caf1, pla, and psaA (21,23,28). RNA was isolated from wild-type Y. pestis or Y. pestis ΔYPO3903 cultures grown at 37°C. The RNA was converted to cDNA, and we performed quantitative reverse transcription-PCR (qRT-PCR) to measure ail, caf1, pla, and psaA transcript levels in each strain. While the ail, caf1, and pla transcript levels in ΔYPO3903 were similar to wild-type levels (no more than a 2-fold change), the levels of psaA were reduced over 10-fold in ΔYPO3903 Y. pestis relative to the wild-type strain (Fig. 5A). The reduction in psaA transcript levels correlated with a loss of PsaA protein levels in ΔYPO3903 Y. pestis cultures compared to wild-type Y. pestis when strains were grown at 37°C (Fig. S3).
If YPO3903 regulates levels only of psaA, we would predict that a psaA mutant would be as adherent as a YPO3903 mutant in vivo. If YPO3903 regulates levels of psaA in Tn-Seq Analysis of Y. pestis Pulmonary Adherence addition to other genes involved in adherence, we would expect that the loss of psaA would have a minor effect on Y. pestis adherence in vivo. To test the role of psaA in Y. pestis adherence early during primary pneumonic plague, we inoculated mice intranasally with 1 ϫ 10 4 CFU wild-type Y. pestis, Y. pestis ΔYPO3903, or Y. pestis ΔpsaA. At 2 hpi, bronchoalveolar lavage was performed, and the proportion of adherent Y. pestis bacteria for each strain was calculated. As seen previously, the YPO3903 mutant had a significant defect in adherence relative to wild-type Y. pestis, with approximately 75% adherence calculated for wild-type Y. pestis and 55% adherence for ΔYPO3903 Y. pestis (Fig. 5B). However, we observed only a minor loss of adherence in the psaA mutant, at ϳ70% adherence, which was not statistically significant (Fig. 5B). The minor effect of psaA on adherence in the lung is also observed in the Δcaf1 Y. pestis background (Fig. 5B). Each strain had a similar total Y. pestis burden recovered from the lung at 2 hpi of approximately 10 4 CFU (Fig. 5C). These data support the hypothesis that PsaA contributes to some but not all of the Y. pestis adherence in the lung mediated by YPO3903. Thus, YPO3903 controls Y. pestis adherence likely by regulating levels of multiple adhesins at the transcriptional level, including psaA.

DISCUSSION
Colonization of mucosal surfaces is a critical first step in the pathogenesis of Yersinia species. Following inhalation of Yersinia pestis, the bacteria colonize the lower airways, where unrestricted bacterial growth during an immunosuppressive phase of disease ultimately leads to severe inflammation and a fatal acute pneumonia (1,2). Although several proteins have been proposed to mediate adherence in vitro, phenotypes for individual mutants are slight and the roles of these proteins in Y. pestis adherence in the lung have not been tested. The goal of our study was to identify genes important for Y. pestis adherence in the lung during primary pneumonic plague. We used Tn-seq to screen pools of wild-type and Δcaf1 Y. pestis insertional mutants for defects in adherence in the lung.
For our screen, we used a lung lavage assay in which nonadherent bacteria are collected by bronchoalveolar lavage (41). Mutants with insertions in six genes were significantly enriched in both Y. pestis strain backgrounds: YPO3904 and iscR, encoding transcriptional regulators; clpX and tig, encoding proteins that function as molecular chaperones; rnc, encoding an endoribonuclease; and YPO3903, encoding a hypothetical protein. While none of these genes are annotated to encode adhesins, each one encodes a protein involved in the regulation or assembly of macromolecules. The inactivation of these genes could result in the dysregulation of multiple adhesins. Overall, our Tn-seq results suggest that there may not be one predominant adhesin that Y. pestis utilizes for adherence in the lung. Identification of how each gene regulates the expression, assembly, or display of various surface proteins may reveal the various adhesins Y. pestis employs during infection.
A similar serial enrichment strategy was successfully applied to screen Caulobacter crescentus transposon mutants for defects in adherence in vitro, highlighting the utility of this approach in characterizing bacterial adherence (42). However, in our assay, the enrichment of mutants with transposon insertions in genes involved in gene regulation or protein display rather than adhesins also reflects a limitation of the Tn-seq approach: only single insertional mutants are created by transposon mutagenesis. If there is redundancy among adhesins, or if the loss of a single adhesin can be complemented by the presence of others, then we would not enrich for those particular mutants. Additionally, a unique issue with performing Tn-seq during primary pneumonic plague was identified previously: Y. pestis transposon mutants with a wild-type phenotype will transcomplement any avirulent mutants for growth in the lungs (1). The transcomplementation is dependent on the type III secretion system creating a permissive environment in the lungs to allow bacterial outgrowth of avirulent mutants. Since we were evaluating the ability of individual bacterial mutants to bind to the lung epithelium during the first 2 h after inoculation, and because the pCD1 Ϫ strain was just as adherent as wild-type Y. pestis, it is unlikely that transcomplementation affected the results of the screen in this study.
We created clean genetic deletions in our wild-type Y. pestis strain for five of the shared enriched mutants to determine their contribution to Y. pestis adherence. Of the deletion mutants tested for adherence in vivo, only the YPO3903 mutant had a significant loss of adherence in the lung relative to wild-type Y. pestis (Fig. 4). Although Y. pestis ΔYPO3904 did not have a statistically significant defect in adherence compared to wild-type Y. pestis, it was slightly less adherent in the lung. YPO3904 encodes a transcriptional regulator with homology to HdfR, which is best characterized for its regulation of the flagellar master operon flhDC (43). However, HdfR has flagellumindependent activities, regulating expression of the glutamate synthase gene subunits gltBD in E. coli and colonization of the bladder during Proteus mirabilis infection (44,45). HdfR may play a minor role in the regulation of genes required for Y. pestis adherence. We did not observe any significant loss of adherence for the clpX, iscR, and tig mutants compared to wild-type Y. pestis. These genes may play even subtler roles in regulating adherence that could be detectable if the genes were deleted in the more adherent Δcaf1 strain background or when assayed during coinfection with other mutants, similar to the conditions of the Tn-seq screen.
YPO3903 encodes a hypothetical protein, and we determined that YPO3903 has 74% identity to the E. coli protein YifE/MaoP. MaoP is conserved in Enterobacteriaceae and is involved in control of chromosome conformation and segregation, which can affect gene expression (40,46). We hypothesized that YPO3903 may perform a similar function in Y. pestis and regulate expression or display of adhesins on the cell surface. Transcriptional analyses using quantitative PCR for a small subset of genes previously implicated in Y. pestis adherence in vitro revealed that transcript levels of psaA were about 10-fold lower in the ΔYPO3903 strain than in the wild-type strain (Fig. 5A). The pH 6 antigen, encoded by psaA, increases adherence of Y. pestis to respiratory epithelial cells by binding phosphatidylcholine, but in the absence of the F1 antigen, a psaA mutant is just as adherent as wild-type Y. pestis (23,47). Additionally, a Y. pestis psaA mutant has a minor virulence defect in the lung during primary pneumonic plague at 12 hpi that is not sustained as disease progresses (48). We observed a small but statistically insignificant loss of adherence in the lung for a psaA mutant in both the wild-type and Δcaf1 Y. pestis strains (Fig. 5B). These data again suggest that multiple surface structures may be involved in Y. pestis adhesion in the lung and that YPO3903 likely regulates the expression of multiple genes involved in adherence, including psaA.
During our in vivo screen, we observed strong enrichment of mutants with a Δail colony morphology. This was surprising given our observations that an ail mutant was as adherent as wild-type Y. pestis early after inoculation (Fig. 1A). However, we observed a concurrent increase in the prevalence of the Δail colony morphology after in vitro enrichment, particularly in the Δcaf1 background (Fig. 2D). Sequencing results reveal that ail mutants were significantly enriched in the Δcaf1, but not the wild-type, background (Table 1). Therefore, the enrichment of ail mutants in vivo could be due to enhanced growth or enrichment during the in vitro culture steps between passages in mice. Additionally, previous studies using in vitro assays demonstrated moderate defects in Y. pestis adherence to human epithelial cell lines at 37°C when ail was deleted individually, and these defects were enhanced when ail was deleted in combination with other surface protein-encoding genes, such as pla and psaA (16,21). Taken together, these data suggest that ail may play a minor (if any) role in early Y. pestis adherence in the lung, which was magnified due to enhanced in vitro growth and multiple rounds of enrichment in our screen. The ail gene is one of the most highly expressed genes in Y. pestis at 37°C, and Ail constitutes a large fraction of the outer membrane (19,49,50). It is possible that a complete deletion of ail causes disruption of the outer membrane of Y. pestis. Outer membrane perturbations can nonspecifically alter serum sensitivity in E. coli (51). Therefore, the phenotypes we observed for a Y. pestis ail mutant in vitro (loss of adherence and serum sensitivity) may result from outer membrane perturbations due to the loss of Ail. However, molecular analyses of Ail revealed that different residues in the outer loops of the Ail protein contribute to Y. pestis adherence and serum resistance in vitro (52)(53)(54). Experiments to determine the role of these adherence-specific residues in vivo would provide a more sophisticated analysis of the contribution of Ail to Y. pestis adherence in the lung without the pleiotropic effects that result from the complete deletion of ail.
Adherence is not completely lost in a Y. pestis YPO3903 mutant, suggesting that there are likely other proteins outside the control of YPO3903 that may play a role in adherence. Although we focused our analysis on the mutations in genes that were enriched in both the wild-type and Δcaf1 backgrounds, there are many mutants with insertions in genes that were significantly enriched and unique to either wild-type or Δcaf1 Y. pestis. We observed a relatively poor correlation between mutants enriched in the wild-type background and those in the caf1 background (Fig. S2). These data imply that the presence of the F1 antigen dramatically alters display or availability of various surface structures for Y. pestis binding to host cells. Because Y. pestis produces the F1 antigen during primary pneumonic plague, mutants that were enriched in the wildtype background may be the most relevant to natural infection. There was significant enrichment of mutations in micF, which is a small regulatory RNA. In Y. pestis, MicF posttranscriptionally regulates levels of OmpF, an outer membrane protein (55). We also observed enrichment of mutations in waaQ, which encodes a lipopolysaccharide core biosynthesis protein (56). Since lipopolysaccharide (LPS) structure can affect the function of outer membrane proteins such as Ail, alterations in LPS resulting from a waaQ mutation may also affect adherence in Y. pestis (30,57). Mutants with insertions in the gene YPO3880 were also enriched in the wild-type Y. pestis background. YPO3880 is predicted to be an exported protein as part of a chaperone-usher system. The deletion of the operon containing this chaperone-usher system results in slight attenuation of a Y. pestis KIM strain following intranasal, but not intravenous, inoculation (29,58). Characterizing the genes identified by our in vivo enrichment for nonadherent mutants, as well as further defining the role of YPO3903 in regulating expression of Y. pestis genes involved in adherence, may reveal novel bacterial targets for inhibiting initial Y. pestis colonization of the lung during primary pneumonic plague.

MATERIALS AND METHODS
Bacterial strains and plasmids. The fully virulent Yersinia pestis strain CO92 was obtained from the U.S. Army, Ft. Detrick, MD. The presence of pCD1 and the pgm locus was confirmed by PCR before use. Y. pestis was grown on brain heart infusion (BHI) agar (Difco Laboratories) at 26°C for 2 days. All bacterial strains used in this study are listed in Table S1.
Animals and animal infections. Naive 6-to 8-week-old female C57BL/6J mice were obtained from Jackson Laboratories and housed in animal biosafety level 3 facilities at the University of North Carolina (UNC) at Chapel Hill prior to inoculation. All experiments involving mice were reviewed and approved by the Institutional Animal Care and Use Committee at UNC Chapel Hill under protocol number 15-022 or number 17-258. Y. pestis CO92 growth from a BHI agar plate was used to start liquid cultures in 2 ml BHI broth, which were grown for 12 h at 26°C. Cultures were then diluted 1:200 in 10 ml BHI broth supplemented with 2.5 mM CaCl 2 and grown for 12 to 16 h at 37°C with constant shaking at 250 rpm. Mice were lightly anesthetized with 50 to 100 mg/kg ketamine and 5 to 10 mg/kg xylazine and then inoculated intranasally with a lethal dose of bacteria suspended in 20 l sterile phosphate-buffered saline (PBS). Two hours after inoculation, mice were euthanized with an overdose of sodium pentobarbital.
Determination of nonadherent Y. pestis. Groups of mice were euthanized at 2 h postinoculation, and lungs were inflated via tracheal cannulation with 1 ml PBS. The PBS was then retracted to collect bronchoalveolar lavage fluid (BALF). This process was repeated until a total of 3 ml BALF was collected from each mouse. Lungs were then removed, placed in 1 ml PBS, and homogenized with a tissue homogenizer (Dremel). Serial dilutions of both the BALF and lung homogenate were plated on BHI agar to enumerate CFU. To calculate the percent nonadherent Y. pestis, the number of CFU in the BALF was divided by the sum of CFU in the lung homogenate and the BALF.
Transposon mutagenesis. The transposon mutant libraries were generated using the Himar1-based transposon system encoded on pPP47, which has been used previously for transposon mutagenesis in Yersinia pestis (1). E. coli S17 pir carrying pPP47 was mated with Y. pestis wild-type and Δcaf1 strains twice. A 500-l portion of liquid culture for each strain was washed with 1 ml 10 mM MgSO 4 , and then each Y. pestis strain was combined with E. coli in 50 l 10 mM MgSO 4 , and this mixture was spotted on BHI agar. The wild-type Y. pestis transposon libraries are designated YP473Tn and the Δcaf1 Y. pestis transposon libraries are designated YP475Tn. For YP473Tn and YP475Tn libraries, only one mating was performed for each strain. For YP473Tn3 and YP475Tn2 libraries, 4 individual matings were performed for each strain. Plates were incubated for 5 h at 26°C. A cotton swab was used to collect the resulting bacterial growth from each plate, which was resuspended in 1 ml PBS. For YP473Tn and YP475Tn, this was diluted 1:100 in PBS, and 250 l was plated on 5 BHI agar plates containing kanamycin for Y. pestis selection with the transposon integrated into the genome and polymyxin B for E. coli counterselection. For YP473Tn3 and YP475Tn2, the resuspended mating was diluted 1:10 in PBS, and 200 l was plated on 8 BHI agar plates with kanamycin and polymyxin B. Serial dilutions were also plated for each mating resuspension to determine the concentration of mutants in each mating. After 2 days incubation at 26°C, colonies were collected from the plates by flooding the top of the plate with BHI broth and dislodging colonies with a cell spreader. The medium was removed with a serological pipet and pooled in a 50-ml conical tube. Approximately 48,000 individual colonies for YP473Tn, ϳ43,000 colonies for YP475Tn, ϳ185,920 colonies for YP473Tn3, and ϳ165,760 colonies for YP475Tn2 were collected. The colonies were resuspended in the BHI broth by repeated pipetting, and glycerol was added to a final concentration of 15%. Aliquots (500 l) of each mating were kept at Ϫ80°C until use.

Enrichment of transposon mutants in vivo and in vitro.
For the following protocol, the BHI agar or liquid broth used was supplemented with kanamycin (50 g/ml) to maintain selection of transposon insertional mutants (kanamycin resistance). One frozen library aliquot from each transposon mutagenesis library (YP473Tn and YP473Tn3 for the wild-type background and YP475Tn and YP475Tn3 for the Δcaf1 background) was thawed. YP473Tn was mixed with YP473Tn3 and YP475Tn was mixed with YP475Tn2. A 500-l portion of the combined transposon libraries was added to separate 10-ml BHI broth cultures supplemented with 1 mM CaCl 2 to generate the wild-type (WT) input (YP473TnϩYP473Tn3) and the Caf1 input (YP475TnϩYP475Tn2). The two libraries from the separate matings were combined to maximize diversity of insertional mutants in both input libraries. These cultures were incubated shaking at 37°C for 4 h. A 200-l portion of each culture was added to 10 ml of fresh BHI broth supplemented with 1 mM CaCl 2 and incubated with shaking for 15 h. The optical density at 620 nm (OD 620 ) for each culture was determined, and 2 groups (designated WT in vivo group 1 and group 2) of 3 mice each were inoculated intranasally with 1 ϫ 10 6 CFU WT input, and 2 groups (designated Caf1 in vivo group 1 and group 2) of 3 mice each were inoculated intranasally with 5 ϫ 10 6 CFU Caf1 input. At the same time, 200 l of the WT and Caf1 input cultures were added to 2 separate flasks of 10 ml BHI broth supplemented with 1 mM CaCl 2 and were incubated with shaking at 37°C. Each flask was designated WT or Caf1 in vitro group 1 or group 2. At 2 hpi, mice were euthanized, lungs were inflated via tracheal cannulation with 1 ml ice-cold PBS, and the liquid was retracted to collect BALF. A total of 3 ml BALF was collected from each mouse. Lungs were then removed, placed in 1 ml PBS, and homogenized with a tissue homogenizer (Dremel). Serial dilutions of both the BALF and lung homogenate were made and plated on BHI agar to enumerate CFU. The remaining BALF (containing nonadherent Y. pestis mutants) was plated on BHI agar. The OD 620 was determined for the in vitro cultures and 1 ϫ 10 6 CFU from each was diluted in 1 ml PBS and plated on BHI agar.
After 2 days of incubation at 26°C, the single colonies were collected from each plate. For each set of plates from a single mouse, BHI broth was poured on the surface of the agar plates. A plate spreader was used to dislodge the colonies from the surface of the agar into the medium, and this was collected with a serological pipet. The colony suspension was transferred to a 50-ml conical tube. This process was repeated for each set of plates from each mouse and each in vitro group. The colony suspensions for the 3 mice from group 1 were pooled, and the suspensions from group 2 were pooled for both the WT and Caf1 libraries. The colony suspensions from in vitro groups 1 and 2 were kept separate. A 500-l portion of each pooled suspension was added to 500 l of 50% glycerol and kept at Ϫ80°C. The strains generated from freezing aliquots of each library after enrichment are listed in Table S2. A 500-l portion of each pooled bacterial suspension was added to 10 ml of fresh BHI broth with CaCl 2 and incubated with shaking at 37°C for 4 h. Then, 200 l of each culture was added to 10 ml of fresh BHI broth with CaCl 2 and shaking at 37°C for 15 h. The process of inoculating groups of mice and the in vitro cultures and collecting mutants was repeated for a total of 4 rounds of enrichment in vivo and in vitro.
Preparation of bacterial genomes for sequencing. Frozen aliquots of the WT and Caf1 libraries before, during, and after enrichment were thawed, and 500 l of the stock was added to 10 ml BHI broth and grown with shaking at 37°C for 6 h. A 500-l portion of this culture was added to fresh 10 ml BHI broth and grown shaking at 37°C for 12 h. Genomic DNA was isolated from 500 l of culture using the Wizard Genomic DNA purification kit (Promega) following the manufacturer's protocols. The names of all samples of DNA isolated from the corresponding pool of transposon mutants are listed in Table S2. Each DNA pellet was rehydrated overnight in 100 l Tris-EDTA (TE) buffer at 4°C. An additional 100 l TE buffer was added to each genomic DNA sample; then each sample was sheared using an EpiShear Probe sonicator (Active Motif) with 1 s on/1 s off for 30 s at 30% amplitude for 6 cycles, with 1 min of incubation on ice between cycles. Following sonication, each sample was purified using PCR purification with a QiaQuick kit (Qiagen) and eluted in 50 l nuclease-free distilled H 2 O (dH 2 O). Samples were stored at Ϫ20°C until preparation for sequencing.
Each sheared DNA sample was diluted to 250 ng in a total volume of 50 l nuclease-free dH 2 O. The KAPA Hyper Prep kit (KAPA Biosystems) was used to perform end repair, A-tailing, and adaptor ligation for each sample. Seven microliters of end repair and A-tailing buffer and 3 l of end repair and A-tailing enzyme mix were added to the sheared DNA samples. The following thermocycler program was run: 30 min at 20°C, 30 min at 65°C, then hold at 4°C. Next, adaptor ligation was performed. The end repair and A-tailing product was added to 30 l ligation buffer, 10 l DNA ligase, 5 l PCR water, and 5 l of a 15 M mix of adaptors A01 and A02. These samples were incubated at 20°C for 15 min. A postligation bead cleanup was performed using Agencourt AMPure XP beads, using a 0.8ϫ bead selection according to the manufacturer's protocols. The sample was eluted in 50 l buffer EB (10 mM Tris-Cl, pH 8.5). A second bead cleanup was immediately performed with a 1ϫ bead selection and elution of the DNA in 20 l buffer EB. The samples were kept at Ϫ20°C until the PCRs.
PCR to amplify the transposon junctions was performed using the KAPA Robust 2G kit (KAPA Biosystems). The first PCR amplified sheared DNA sequences that had successfully ligated the adaptors. The 20-l sample created with the KAPA Hyper Prep kit was added to 10 l 5ϫ buffer A, 5ϫ enhancer, 10 mM deoxynucleoside triphosphates (dNTPs), 5 M primer R1, 5 M primer KAPA Prim 1, and 0.2 l polymerase. The reaction was run in a thermocycler with the following program: 1 cycle of 98°C for 45 s; 10 cycles of 98°C for 15 s, 55°C for 30 s, and 72°C for 30 s; 1 cycle of 72°C for 1 min; then a hold at 4°C. A bead cleanup was performed for each sample with Agencourt beads, using a 1ϫ bead selection according to the manufacturer's protocols and eluting in 20 l buffer EB. The second PCR further amplified the read fragments and added the sample-specific barcode sequence to allow pooling of the libraries for sequencing. This PCR was performed using KAPA HiFi Taq from the KAPA Hyper Prep kit (KAPA Biosystems). The 20-l sample from the previous bead cleanup was added to 25 l Ready Mix (2ϫ), 5 M index primer, and 5 M primer KAPA Prim 1. The index primer used for each library is listed in Table S2. The reaction was run in a thermocycler with the following protocol: 1 cycle of 98°C for 45 s; 3 cycles of 98°C for 15 s, 63°C for 30 s, and 72°C for 15 s; 10 cycles of 98°C for 15 s, 65°C for 30 s, and 72°C for 30 s; 1 cycle of 72°C for 1 min; then a hold at 4°C. A bead cleanup was performed for each sample with Agencourt beads, using a 1ϫ bead selection according to the manufacturer's protocols and eluting in 20 l buffer EB.
Each sample was analyzed on a Bioanalyzer (Agilent) to determine the molarity of the amplicons in the size range of 350 to 600 bp, which was the desired range of fragment lengths for sequencing. Each sample was diluted to 15 nM, and 5-l portions of each sample was pooled, yielding a pooled sample volume of 90 l. A double size selection was performed on the pooled sample using a 0.5 to 0.7ϫ bead ratio with Agencourt beads following the manufacturer's protocol. The double size selection further enriched the pooled sample for amplicons ranging from 200 to 800 bp. The sample was eluted in buffer EB, the concentration was determined by Qubit analysis, and the molarity was determined with a Bioanalyzer (Agilent). The sample was diluted to a final concentration of 30 nM in a 25-l volume of buffer EB, which was sequenced using the Illumina MiSeq with 150ϫ paired-end reads.
Analysis of transposon mutants. (i) Read count and normalization. The paired end reads generated 2 FASTQ files for each amplicon sequenced, named Read 1 (R1) and Read 2 (R2). Each read in the R2 FASTQ file was modified such that the four random nucleotides (NNNN in primer R1, which creates a random barcode to aid in identification of PCR duplicates), were removed and appended to the R2 identifier line. BWA (Burrows-Wheeler alignment tool) alignment (v0.7.17-r1194-dirty) to the Yersinia pestis CO92 full genome (33) was then performed on each paired-end set of FASTQ files and sorted using samtools (v1.8), after which duplicates were filtered using biobambam's bammarkduplicate program (v2.0.33). This generated SAM (sequence alignment/map) files for each read.
Reads from the duplicate-filtered alignment were then inspected to find the insertion locations of the transposon. Most alignments allowed quick determination of the transposon insertion site, where a TA/AT dinucleotide was present followed by the transposon sequence. Other alignments resulted in a portion of the end sequence of the transposon aligning to the genome, thus obscuring the true start site. These states were suggested by the presence of soft clips adjacent to a match (denoted by an nSnM, nMnS, or nSnMnS pattern in the CIGAR string, where n denotes the number of nucleotides to which the designation of S, denoting a soft clip, or M, denoting a match, applies) of reads in the SAM file.
To confirm each suggested location, NCBI's BLASTn (ncbi-blast v2.7.1ϩ) program was used to determine the true start site of transposon insertion, where BLASTn was run with the transposon sequence against each of the suggested reads. BLAST, being a greedy aligner, would give results matching the entire transposon sequence if possible. For reads for which there was only an nM pattern in the CIGAR string, BLASTn would return a position such that the transposon insertion position complemented the start of the read as reported by BWA. For instances where a read's CIGAR string contained a combination of soft clips and matches, any unaligned portion from BLASTn could be truly attributable to the Y. pestis genome. This allowed the resolution between BWA's reported read and the exact location of the transposon insertion site in the genome. The number of unique insertions for each gene was quantified, and upper-quartile normalization was performed to give a read count for each gene.
(ii) Calculating top hits. The normalized read counts for each gene were averaged between the replicate samples, and this average was divided by the normalized read count for the input library and log 2 transformed to give an output/input ratio for the following groups: WT in vivo 2 rounds, WT in vivo 3 rounds, WT in vitro 2 rounds, WT in vitro 3 rounds, Caf1 in vivo 2 rounds, Caf1 in vivo 3 rounds, Caf1 in vitro 2 rounds, and Caf1 in vitro 3 rounds. The mean and standard deviation of the output/input ratios for every gene within each individual group were calculated. Genes that were more than 2 standard deviations away from the mean in both in vivo 2 rounds or in vivo 3 were pursued for further analysis.
Lambda red recombination. Deletion of clpX, iscR, tig, YPO3903, and YPO3904 was performed in Y. pestis wild-type and pCD1 Ϫ backgrounds using a modified form of lambda red recombination. Briefly, 500 bp upstream and 500 bp downstream sequences for the desired region to delete were amplified by PCR and combined in splicing by overhang extension (SOE) PCR with a Kan r cassette flanked by FLP recombination target (FRT) sites for allelic replacement of the wild-type open reading frame (ORF) (59). The products were transformed into Y. pestis strains harboring pWL204, a plasmid containing the recombinase genes (10). Following successful recombination verified by PCR, the Kan r cassette was resolved by the introduction of pSkippy, a Tet s derivative of pFLP3 harboring an Amp r cassette and sacB and carrying the FLP recombinase gene under the control of the lac promoter (60). Correct resolution of the Kan r cassette, as well as presence of the virulence plasmids (for mutants in the Y. pestis wild-type background) and the pgm locus, was confirmed by PCR. The oligonucleotides used for lambda red recombination and the strains generated are listed in Table S1.
Transcript quantification by qRT-PCR. Two-milliliter portions of BHI broth cultures were inoculated with either wild-type or ΔYPO3903 Y. pestis in the pCD1 Ϫ background and incubated with rolling overnight at 26°C. A 250-l portion of each culture was added to 10 ml of fresh BHI broth supplemented with 1 mM CaCl 2 and incubated with shaking for 6 h at 37°C. Total RNA was then purified from 1 ml of culture using the TRIzol reagent manufacturer's protocol, treated with Turbo DNase (Ambion), and reverse transcribed with the iScript cDNA synthesis kit (Bio-Rad) according to the manufacturer's instructions. cDNAs were used as the templates for amplification and detection of the Y. pestis genes ail, caf1, pla, and psaA with SYBR green dye (Bio-Rad) in an iCycler thermocycler (Bio-Rad). For each gene, the calculated threshold cycle (C T ) was normalized to that of gyrase B (gyrB) from the same sample prior to calculation of the fold change using the ΔΔC T method (61). Oligonucleotides used for qRT-PCR are listed in Table S1.
Western blot analysis. Two-milliliter portions of BHI broth cultures were inoculated with either wild-type or ΔYPO3903 Y. pestis in the pCD1 Ϫ background and incubated with rolling overnight at 26°C. A 250-l portion of each culture was added to 10 ml of fresh BHI broth supplemented with 1 mM CaCl 2 and incubated with shaking for 6 h at 26°C or 37°C. Whole-cell lysates were prepared from cells at an OD 620 of 2 that had been pelleted, washed once with ice-cold PBS, and resuspended in Laemmli buffer containing 5% ␤-mercaptoethanol. Samples were boiled for 10 min, and a portion corresponding to an OD 620 of 0.2 was separated via SDS-PAGE and transferred to polyvinylidene difluoride (PVDF) membranes for Western blot analysis. Loading was qualitatively assessed by Ponceau S staining of the PVDF membrane. Anti-PsaA serum was used to probe for PsaA. Prior to use, the anti-PsaA serum was absorbed against E. coli lysates and used at a titer of 1:2,500 (62). Anti-IgG horseradish peroxidase (HRP)-conjugated secondary antibodies were used at a titer of 1:20,000.
Statistical analysis. Data were graphed and analyzed for statistical significance in Prism 7.0b. Data were analyzed by one-way analysis of variance (ANOVA) with Tukey's multiple-comparison test or by Student's t test and are presented as means with standard deviations (SD). Details for statistical analyses, such as numbers of replicates, group numbers, the statistical test used, and the definition of statistical significance, are given in the figure legends.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.

ACKNOWLEDGMENTS
We thank Marcia Sanders and members of the Vironomics Core for Qubit and Bioanalyzer analysis of libraries. We thank Tara Skelly and members of the High-Throughput Sequencing Facility for assistance with library preparation and sequencing on the MiSeq.
This work was supported in part by the NIH under award numbers AI26667 to K.R.E. and AI119558 to W.E.G.