Capsule Production and Glucose Metabolism Dictate Fitness during Serratia marcescens Bacteremia

ABSTRACT Serratia marcescens is an opportunistic pathogen that causes a range of human infections, including bacteremia, keratitis, wound infections, and urinary tract infections. Compared to other members of the Enterobacteriaceae family, the genetic factors that facilitate Serratia proliferation within the mammalian host are less well defined. An in vivo screen of transposon insertion mutants identified 212 S. marcescens fitness genes that contribute to bacterial survival in a murine model of bloodstream infection. Among those identified, 11 genes were located within an 18-gene cluster encoding predicted extracellular polysaccharide biosynthesis proteins. A mutation in the wzx gene contained within this locus conferred a loss of fitness in competition infections with the wild-type strain and a reduction in extracellular uronic acids correlating with capsule loss. A second gene, pgm, encoding a phosphoglucomutase exhibited similar capsule-deficient phenotypes, linking central glucose metabolism with capsule production and fitness of Serratia during mammalian infection. Further evidence of the importance of central metabolism was obtained with a pfkA glycolytic mutant that demonstrated reduced replication in human serum and during murine infection. An MgtB magnesium transporter homolog was also among the fitness factors identified, and an S. marcescens mgtB mutant exhibited decreased growth in defined medium containing low concentrations of magnesium and was outcompeted ~10-fold by wild-type bacteria in mice. Together, these newly identified genes provide a more complete understanding of the specific requirements for S. marcescens survival in the mammalian host and provide a framework for further investigation of the means by which S. marcescens causes opportunistic infections.

IMPORTANCE Serratia marcescens is a remarkably prolific organism that replicates in diverse environments, including as an opportunistic pathogen in human bacteremia. The genetic requirements for S. marcescens survival in the mammalian bloodstream were defined in this work by transposon insertion sequencing. In total, 212 genes that contribute to bacterial fitness were identified. When sorted via biological function, two of the major fitness categories identified herein were genes encoding capsule polysaccharide biogenesis functions and genes involved in glucose utilization. Further investigation determined that certain glucose metabolism fitness genes are also important for the generation of extracellular polysaccharides. Together, these results identify critical biological processes that allow S. marcescens to colonize the mammalian bloodstream.
S erratia marcescens is a member of the Enterobacteriaceae family that is found ubiquitously in the environment. This species is known to be associated with plants, animals, insects, freshwater, and soil and is also recognized as an opportunistic pathogen of humans (1). As an opportunistic pathogen, S. marcescens causes a wide range of both health care-associated and community-acquired infections, including bacteremia, keratitis, wound infections, and urinary tract infections (1). The diverse environments in which this organism thrives suggest a remarkable physiologic flexibility. Recent analysis of the S. marcescens pangenome has demonstrated that a large number of poorly conserved accessory gene elements exist between isolates of this species, suggesting a genetic basis by which S. marcescens can occupy diverse niches (2). Furthermore, a comparison of the gene contents of an insect isolate and a human isolate of S. marcescens demonstrated a large degree of genetic diversity (3). This physiologic flexibility and genetic diversity underscore the need to assess the requirements for bacterial fitness of specific isolates under conditions that closely model their niche of origin. Of particular interest to this work is the ability of S. marcescens to survive during mammalian bloodstream infection. Bacteremia caused by S. marcescens is a serious disease associated with a mortality rate of 23 to 52% (4)(5)(6).
Previous work to define genes required for S. marcescens survival during infection has focused primarily on individual gene targets or genetic screens using invertebrate hosts. One virulence factor that has emerged from these studies is the ShlA hemolysin, which has been demonstrated to have cytolytic activity against multiple eukaryotic cell types and contributes to pathogenesis in a murine model of pneumonia (7)(8)(9)(10). Other S. marcescens proteins that are proposed virulence factors include secreted proteases (11)(12)(13), the PhlA extracellular phospholipase (14), and iron acquisition proteins (15). Two studies have also assessed in vivo fitness with Caenorhabditis elegans as a host to identify S. marcescens transposon mutants that exhibit decreased nematode killing. By using a derivative of insect isolate Db10 (16), Kurz et al. identified 23 transposon mutants that were attenuated in the nematode model (10). In similar work, Iguchi et al. also identified 12 transposon mutants of Db10 that were attenuated in C. elegans, corresponding to eight genetic loci not previously identified in the Kurz screen (3). This same study identified a putative polysaccharide capsule biosynthesis locus and determined that the wza gene within this locus contributed to C. elegans infection.
Polysaccharide capsules are produced by multiple Enterobacteriaceae species and are important mediators of immune resistance by interfering with opsonophagocytosis and complement-mediated killing (17,18). S. marcescens is known to produce a capsule consisting of acidic polysaccharides, the specific composition of which varies between isolates (19,20). Despite the potential importance of S. marcescens capsule during mammalian bloodstream infection, the genetic basis of capsule biosynthesis remains understudied in this organism.
To further define S. marcescens genes required for in vivo bacterial fitness, specifically within a mammalian host, we conducted a transposon insertion sequencing (INSeq) screen with a bank of~32,000 mutants in a murine model of bacteremia. We report herein the identification of 212 genes, encompassing a wide range of predicted functions that contribute to S. marcescens in vivo fitness. Two of the major biological processes identified in this work are capsule biosynthesis and glucose metabolism functions.

RESULTS
Establishment of a murine model of S. marcescens bloodstream infection. Eleven S. marcescens bloodstream infection isolates were obtained from the University of Michigan Health System for this study, and the genome sequence of each isolate was determined. From this collection, strain UMH9 was selected for further characterization on the basis of the ability to reproducibly colonize mice at a high level and an antibiotic susceptibility profile that was compatible with available genetic manipulation tools (see Fig. S1 in the supplemental material). Strain UMH9 is a nonpigmented isolate that exhibits intermediate resistance (32 g/ml) to the clinically relevant piperacillintazobactam drug combination.
The overall objective of this work was to define the bacterial genes necessary for survival in the mammalian bloodstream; therefore, a murine model of bacteremia was established for strain UMH9. Infection of C57BL/6 mice with 1 ϫ 10 7 CFU via tail vein injection resulted in robust colonization of the spleen, liver, and kidneys at 24 h (Fig. 1A). In contrast, inoculation with 1 ϫ 10 6 CFU resulted in a low overall bacterial burden and mice infected with 1 ϫ 10 8 CFU were moribund prior to 24 h (data not shown). The 1 ϫ 10 7 CFU dose was used for all subsequent experiments. The numbers of colonizing bacteria in the spleens, livers, and kidneys of infected mice were determined 4 h postinfection to assess the proportion of the inoculum that was capable of establishing infection. The greatest density of bacteria was found in the spleen at this early time point, indicating a decreased likelihood of colonization bottlenecks at that site (Fig. 1B). Although UMH9 replicates to high levels in the murine kidney for Ն48 h, the low initial bacterial density in this organ was considered problematic with regard to the potential loss of individual mutants from a large inoculum.
To further assess the potential for colonization bottlenecks, a spontaneous nalidixic acid-resistant mutant was isolated and used in coinfections with parent strain UMH9. Mice were infected with different ratios of the two strains, with strain UMH9 Nal Mice were sacrificed at 24 h postinfection, and the number of viable bacteria in each organ was determined. The limit of detection is indicated by the dotted line. (B) Mice were inoculated as described for panel A with 1 ϫ 10 7 CFU of strain UMH9, and the bacterial burden was monitored over time. Lines indicate the median bacterial burden in each organ. (C) Evaluation of colonization bottlenecks in the spleen with a spontaneously resistant strain (UMH9 Nal ) during coinfection with the UMH9 parent. Mixed inocula, at the ratios indicated, were used to infect mice with a total of 1 ϫ 10 7 CFU. After 24 h, mice were sacrificed and the CI was determined by measurement of viable counts. A CI of 1.0 (dotted line) indicates a lack of fitness advantage for either strain. CI values that differed significantly from a hypothesized median value of 1.0, as determined by a Wilcoxon signed-rank test (P Ͻ 0.01), are indicated with asterisks.
In Vivo Fitness of S. marcescens ® representing any given clone in a mixed population of transposon insertion mutants. If colonization was limited by bottleneck effects, UMH9 Nal would be expected to exhibit stochastic loss from the population and a competitive disadvantage (competitive index [CI] of Ͻ1.0). At input ratios as low as 1 ϫ 10 Ϫ4 (UMH9 Nal to UMH9), no spontaneous loss of the resistant strain was observed from the spleen-colonizing population (Fig. 1C). Since an in vitro outgrowth step was necessary to recover the low proportion of UMH9 Nal from spleen homogenates, the observation that UMH9 Nal seemingly outcompeted the parent strain in vivo may be due to a marginal growth advantage of strain UMH9 Nal (see Fig. S2A). In support of this conclusion, a 1:1 ratio of the two strains that was plated directly from spleen homogenates exhibited a neutral CI (~1.0) (see Fig. S2B). Together, these data demonstrate that the UMH9 bacteremia infection model can accommodate a high population diversity without the stochastic loss of individual clones due to bottlenecks during colonization. Transposon insertion sequencing screen for in vivo fitness genes. A marinerbased transposon carried by suicide plasmid pSAM-Cm (21) was used to generate a random library of transposon insertion mutants in strain UMH9 (see Fig. S3). Five pools containing equivalent numbers (~6,400) of mutants were used to infect four mice per pool. Spleens were collected from infected mice after 24 h. The insertion sites from input and output pools were PCR amplified and sequenced via Illumina to determine the relative abundance of each transposon mutant (22). The ESSENTIALS pipeline (23) was used to map sequence reads and identify essential and conditionally essential (fitness) genes. A newly developed R package called TnseqDiff, which uses insertion level information to identify fitness genes, was also applied to the data (L. Zhao, M. T. Anderson, W. Wu, Y. Li, H. L. T. Mobley, and M. A. Bachman, submitted for publication). In total, the library consisted of~32,000 unique insertions (~25,000 in open reading frames [ORFs]), and the two methods combined identified 212 UMH9 transposondisrupted genes that exhibited a Ն2 (range, 2.9 to 154.3)-fold decrease (adjusted P Ͻ 0.05) in bacterial fitness (see Data Set S1). The majority of the fitness genes were identified by both methods, although the TnseqDiff package identified a larger number of fitness genes that did not meet the significance criteria of ESSENTIALS (Fig. 2). Sequence reads from the input population were also used to predict that 712 genes in the UMH9 genome are essential (adjusted P Ͻ 0.05) (23). Complete essential scores (see Data Set S2) and conditionally essential scores (see Data Set S3) for each gene, as well as the genomic locations of all transposon insertions (see Data Set S4) can be found in the supplemental material.
Seven fitness genes identified by both the TnseqDiff package and ESSENTIALS were chosen for further study and for validation of the INSeq screen. Deletion-insertion mutations were constructed for each of the genes described in Table 1, and the resulting strains were tested for in vivo fitness defects in competition with the UMH9 parent strain by using the murine bacteremia model. Three of the mutants tested exhibited significant fitness defects in the spleen, and five were outcompeted in the murine kidney (CI of Ͻ1) (Fig. 3). The results of these experiments validated the findings of the genetic screen and confirmed that six of the seven genes tested contribute to S. marcescens fitness in the mammalian host. Importantly, none of the seven mutants exhibited a general growth defect when cultured in vitro (see Fig. S4). Of the strains tested, only the rcsB mutant failed to exhibit a significant fitness defect in either the  In Vivo Fitness of S. marcescens ® spleen or the kidneys. The genes listed in Table 1 are predicted to encode a wide range of biological functions; therefore, subsequent in vitro assays were designed to explore the role of each gene product during S. marcescens infection.

S. marcescens capsule is a major fitness determinant.
In competition infections with the wild-type strain, the wzx mutant exhibited 9-and 450-fold losses of fitness in the spleen and kidney, respectively (Fig. 3). Wzx flippase proteins are best studied with regard to their role in inner membrane translocation of lipid-linked O-antigen repeat units (24,25); however, Wzx homologs also function in sugar translocation with wzy-dependent capsule synthesis systems (18). The UMH9 wzx gene is located within an 18-gene locus with similarity to predicted capsule biosynthesis loci from other S. marcescens isolates (3). The hypothesized function of this genetic region in capsule production, rather than O-antigen synthesis, is reinforced by the presence of wza, wzb, and wzc outer membrane capsule transport gene homologs (18). A total of 191 unique transposon insertions were identified in the 22-kb capsule locus, and in addition to wzx, 10 other genes were determined by INSeq to contribute to bacterial fitness, underscoring the critical importance of this genetic region for S. marcescens infection (Fig. 4). Further analysis of the capsule locus among the 11 different clinical isolates sequenced in this study demonstrated a high degree of heterogeneity, consistent with the known variation of S. marcescens capsule composition and structure (see Fig. S5) (20). Fitness genes outside this region with predicted roles in capsule biosynthesis were also identified; notably, rfaH transposon mutants exhibited an~9-fold fitness defect. The RfaH antitermination factor is known to control the transcription of multiple extracytoplasmic component genetic loci in other bacterial species, including the group 1 capsule genes of Escherichia coli (26,27).
To assess capsule production, S. marcescens strains were stained with Congo red (CR) dye and visualized by light microscopy. Wild-type strain UMH9 produced a robust negative-stain area surrounding bacterial cells when cultured at physiologic temperatures, indicative of capsule production (Fig. 5A). This result is in contrast to that of the wzx mutant, which appeared to yield a smaller negative-stain area. Quantitative analysis demonstrated that wild-type bacteria produced a median negative-stain area of 2.10 m 2 compared to 1.28 m 2 for the wzx mutant harboring the empty vector (Mann-Whitney test P Ͻ 0.0001), indicating a defect in capsule production (Fig. 5B). Attempts to complement the wzx mutation by providing wzx on a multicopy plasmid (wzx ϩ ) failed to restore capsule production in this assay, a result that may be due to the polarity of the wzx mutation on transcriptionally linked capsule-associated genes (Fig. 4). The area of crystal violet-stained S. marcescens strains were also measured as a control, the results of which demonstrate a more similar overall distribution between strains compared to CR (Fig. 5C). As expected, the largest difference between CR and crystal violet stain areas was observed with the fully encapsulated wild-type strain.
Uronic acids are carboxylated sugars that can be found in the capsular polysaccharides of certain Gram-negative bacterial species, including E. coli strains with group 1 capsules and Klebsiella pneumoniae (18,28). Previous work to characterize the capsule serotypes of S. marcescens identified glucuronic acid and galacturonic acid as common components of acid polysaccharide fractions from several different reference strains (19,20). As a further assessment of capsule production in UMH9, extracellular uronic acids from the relevant S. marcescens strains were quantitated. A significant reduction in uronic acid levels was observed in the wzx/vector strain compared to those of the wild type, reinforcing the lack of capsule production in this mutant (Fig. 5D). As In Vivo Fitness of S. marcescens ® observed with the CR stain assay, genetic complementation of the wzx mutation was unable to restore uronic acid production. The importance of glucose metabolism during infection. Niche-specific acquisition and utilization of nutrients are critical aspects of bacterial physiology during infection. Four S. marcescens gene products predicted to function in the enzymatic manipulation of glucose were identified as fitness factors by INSeq. PfkA catalyzes the first committed step in glycolysis by converting fructose-6P into fructose-1,6P 2 , whereas GpmA catalyzes the interconversion of glycerate-3P and glycerate-2P further downstream in the pathway (Fig. 6A). A third gene, pgm, encodes phosphoglucomutase, which promotes the bidirectional conversion of glucose-6P to glucose-1P and is not involved in the conversion of extracellular glucose into pyruvate. Lastly, the galU gene product catalyzes the conversion of glucose-1P into the nucleotide sugar UDP-glucose (29). The UMH9 galU gene is located within the capsule biosynthesis locus and exhibited a 10-fold fitness defect by INSeq (Fig. 4). While the loss of GalU function is expected to disrupt capsule biosynthesis in these bacteria, S. marcescens galU mutants are likely pleiotropic because of the central role of UDP-glucose in bacterial metabolism.
Since loss of gpmA or pfkA is likely to result in the same phenotype with regard to the utilization of glucose as a sole carbon source, initial efforts were focused on characterization of just the pfkA null mutant. Assessment of growth in M9 medium containing glucose demonstrated that pfkA was required for glucose utilization, as expected (Fig. 6B). In contrast, the pgm mutant exhibited no growth defect under the same conditions. Since pfkA was identified as a fitness factor in a bloodstream infection model, it was hypothesized that a functional glycolysis pathway would be necessary for growth of S. marcescens in serum. M9 medium containing 20% heat-inactivated (HI) human serum promoted the robust growth of both the wild-type UMH9 strain and the pgm mutant; however, S. marcescens lacking pfkA showed a severe growth defect under these conditions (Fig. 6C). S. marcescens cultured in the presence of serum formed macroscopic aggregates during exponential growth (data not shown), resulting in a decrease in optical density (OD) during this phase. To confirm that the loss of OD was not due to decreased viability, the CFU counts of UMH9 and the pfkA mutant in serum culture were determined by colony counting. The results of this experiment confirmed the requirement for pfkA in the presence of serum and showed no loss of viability of the wild-type strain (Fig. 6D). Interestingly, after 24 h of culture, the pfkA mutant achieved the same number of viable bacteria as UMH9, suggesting that while glucose may serve as an S. marcescens carbon source in serum, it is likely that alternative carbon sources may also be utilized. Together with the INSeq analysis findings, these results indicate that glycolysis is a critical pathway during S. marcescens bloodstream infection. The minimal growth defect observed with the pgm mutant in both M9 containing glucose and M9 with human serum (Fig. 6B and C) suggested that glucose-1P was not a major glycolytic substrate under these conditions. To address the contribution of pgm to S. marcescens fitness, the relative survival of bacteria incubated in the presence of 10% normal human serum and HI human serum was tested. Wild-type bacteria were largely resistant to the bactericidal activity of human serum under these conditions, exhibiting a minimal loss of viability over a 90-min incubation (Fig. 7). In contrast, the number of viable pgm mutant bacteria harboring the empty vector decreased by Ͼ2 orders of magnitude after exposure to serum, with serum resistance restored by genetic complementation of the pgm mutation (pgm ϩ ). Treatment with HI serum resulted in minimal changes in bacterial viability over the course of the experiment, indicating that the loss of pgm mutant viability was likely due to complement-mediated bactericidal activity. On the basis of the observed serum sensitivity of the pgm mutant and the predicted function of Pgm in the production of nucleotide sugar precursors (Fig. 6A), it was hypothesized that the enhanced susceptibility of the pgm mutant to complement components was due to altered production of cell surface polysaccharides. Therefore, mutations that disrupt capsule or lipopolysaccharide (LPS) biogenesis functions would be predicted to have a similar phenotype. The UMH9_0939 ORF is predicted to encode a ␤-1,3-glucosyltransferase and is located within the putative O-antigen biosynthetic gene cluster (3). Mutation of UMH9_0939 resulted in a robust fitness defect (Fig. 3) and conferred an even greater loss of serum resistance than the pgm mutation (Fig. 7). The wzx mutant strain also exhibited a complete loss of serum resistance, and surprisingly, partial restoration of viability was observed upon genetic complementation. The reason for the discrepancy in the complementation phenotypes of the wzx mutant is unclear; however, we speculate that the level of capsular polysaccharide required for detection by the CR stain and uronic acid measurement may differ from that required for serum resistance. Together, these results demonstrate that surface polysaccharides are required for S. marcescens serum resistance. To test more directly if pgm contributes to surface polysaccharide biogenesis, pgm mutant bacteria were assayed for loss of capsule. Indeed, the pgm/vector strain produced In Vivo Fitness of S. marcescens ® significantly less capsule than the wild type by CR staining (median ϭ 1.37 m 2 , P Ͻ 0.0001) (Fig. 5A and B) and had lower levels of extracellular uronic acids (Fig. 5D). Genetic complementation of the pgm mutation resulted in partial restoration of both the capsule and uronic acid phenotypes.
The UMH9_0544 gene is located~1.9 kb upstream of the pgm ORF and is predicted to encode an uncharacterized protein containing a ribbon-helix-helix transcriptional repressor motif (InterPro accession no. IPR010985). Although mutation of UMH9_0544 resulted in an~3-fold loss of fitness in the murine spleen (Fig. 3), no change in sensitivity to human serum was detected in comparison to wild-type bacteria (data not shown). As a result, the function of the UMH9_0544 gene product remains uncharacterized.
A magnesium transport protein contributes to fitness. Magnesium is a divalent cation that has multiple biological functions, including as an enzymatic cofactor and a neutralizer of negatively charged macromolecules. There are four known types of bacterial magnesium transporters (30). On the basis of sequence identity with characterized proteins from other species, the S. marcescens UMH9 genome encodes at least one homolog of each type, namely, CorA, MgtE, MgtA, and MgtB (data not shown). However, among the predicted UMH9 magnesium transporter genes, only the mgtB homolog was identified as a fitness factor during murine bloodstream infection (Table 1). Analysis of an mgtB deletion mutant during competition infections with the wild-type parent demonstrated a 7-fold fitness defect in murine kidneys, confirming the role of mgtB as an S. marcescens fitness gene (Fig. 3B). The function of UMH9 MgtB as a magnesium transporter was investigated by culturing bacteria in Chelex-treated M9 medium supplemented with different concentrations of magnesium. In the presence of low (4 M) and intermediate (16 M) concentrations of magnesium, the mgtB mutant grew to a lower density than the UMH9 parent strain (Fig. 8). Interestingly, a difference between the two strains was observed only in the postexponential phase. When bacteria were cultured in medium containing as little as 32 M magnesium, there was ® a negligible growth difference between the mgtB mutant and wild-type strains. The reduction in bacterial density observed in the mgtB mutant strain in 4 M magnesium was partially restored by providing mgtB on a multicopy plasmid (mgtB ϩ ) (Fig. 8B). Although the growth difference between the two strains in the presence of low concentrations of magnesium is consistent with a role in magnesium transport, the phenotype appears to be specific to the postexponential phase and was not observed in the presence of Ն32 M magnesium. These observations suggest that additional transport mechanisms may compensate for the loss of MgtB at higher concentrations of magnesium and that MgtB is not required during rapid S. marcescens growth. It is also possible that MgtB functions most efficiently only under specific environmental conditions, since there was a clear difference in the requirement for mgtB in the spleen and in the kidneys during competition infections (Fig. 3).

DISCUSSION
The global emergence of Enterobacteriaceae species harboring metallo-␤-lactamase and serine ␤-lactamase genes that provide resistance to "last-resort" antibiotics of the carbapenem class underscores the urgency with which additional treatment options are needed to combat infections caused by these versatile pathogens (31,32). The identification and characterization of genes that are required for bacterial survival in vivo provide a strategy through which novel targets for antimicrobial therapy may be identified. In this study, we identified S. marcescens genes that contributed significantly to bacterial fitness in the mouse using an INSeq-based genetic screen. Two previous in vivo genetic screens with an S. marcescens insect isolate were conducted in a nematode model, together resulting in the identification of 30 transposon insertion mutants that were attenuated for C. elegans killing (3,10). Two common fitness pathways were identified by the nematode screen and our murine screen. An mgtB promoter mutation was shown to attenuate C. elegans killing but did not differ from the wild-type strain during infection of Drosophila melanogaster or a pulmonary epithelial cell line (10). The wza gene, encoding a capsule polysaccharide export protein, was also identified in a C. elegans screen (3) and this study, indicating that capsule production is important for infection of both invertebrate and vertebrate host species.
One gene notably absent from our INSeq results was shlA. The ShlA extracellular hemolysin (7) facilitates cytotoxic activity against cultured cell lines (8), and multiple studies have demonstrated a role for ShlA in pathogenesis (9,10,33). The absence of shlA identification in our study was not due to a lack of appropriate mutants, since multiple shlA transposon insertion mutants were detected in our library (data not shown). Rather, it is likely that shlA mutant bacteria in a larger population of hemolysinproducing organisms may benefit from the secreted hemolysin of neighboring bacteria, and thus, no lack of fitness would be detected.
The Rcs phosphorelay is a well-studied regulatory network that controls the expression of numerous genes and is conserved among the members of the family Enterobacteriaceae (34,35). Colonic acid and group 1 capsules are mutually exclusive extracellular polysaccharides that are produced by different strains of E. coli, and the biosynthesis of each involves the transcriptional activation of different target genes by the RcsB response regulator (26,36). The S. marcescens rcsB mutant did not exhibit enhanced sensitivity to serum bactericidal activity (data not shown) and did not have a significant fitness defect in competition infections (Fig. 3), inconsistent with a role in capsule production. However, 50% of the mice infected with a 1:1 mixture of wild-type and rcsB mutant bacteria succumbed to infection prior to 24 h. This is in contrast to mice infected with the wild-type strain alone (Fig. 1) or in combination with other fitness gene mutants (Fig. 3), which routinely survived to 24 h and beyond. These results suggest that the rcsB mutation increases the mouse mortality rate without altering the fitness of the bacterium. In S. marcescens, RcsB is known to repress shlAB expression both through the FlhDC regulatory pathway and by directly binding to the shlAB promoter (37). Therefore, the increased mortality rate of mice infected with rcsB mutants is potentially due to increased expression of the shlAB hemolysin-encoding In Vivo Fitness of S. marcescens ® locus. Consistent with this hypothesis, overexpression of shlAB through deletion of an independent two-component regulatory system, RssAB, also results in an increased murine mortality rate (33). The discrepancy between the identification of rcsB as an INSeq fitness gene and the lack of a fitness defect in competition infections with the wild-type strain remains to be determined.
In the case of the mgtB and pfkA genes, the fitness defect observed by INSeq in bacteria harvested from the spleen was not replicated in engineered mutants in competition with the wild type (pgm conferred a fitness defect that was not statistically significant) (Fig. 3A). One possible explanation for this result is that the INSeq experiment may be more sensitive to loss of fitness from any single mutant in the population, given that the ratio of each mutant (~1:6,000) is much lower than in the competition infections (1:1). It is also important to note that the pgm, mgtB, and pfkA mutants did exhibit significant fitness defects when bacteria from the kidneys were assayed (Fig. 3B). The reason for the organ-specific differences in the fitness defects of the mgtB and pfkA mutants remains untested. However, it was determined that wild-type S. marcescens colonizing the spleen gradually decrease in abundance over 48 h, in contrast to the prolific expansion of bacteria in the kidneys in the same time period (Fig. 1B), suggesting that the selective pressures in these two environments likely differ.
The S. marcescens UMH9 capsule biosynthesis locus shares genetic similarity with group 1 capsule and colonic acid biosynthesis loci of E. coli and certain capsule loci of K. pneumoniae (18,28). A notable difference in comparison to the group 1 and K. pneumoniae capsules is the lack of a wzi gene within the Serratia capsule locus. Wzi is an outer membrane protein involved in tethering of the capsule polysaccharide to the bacterial surface, and the presence of wzi is one factor that differentiates cellassociated capsular polysaccharide loci from secreted exopolysaccharide loci, such as colonic acid (18,38,39). Interestingly, a predicted homolog (61% identity) of the E. coli B44 (group 1, K30) Wzi protein is located at a distal chromosomal locus in UMH9 and other S. marcescens isolates. Previous characterization of extracellular polysaccharides from several S. marcescens strains demonstrated that the O antigen of LPS and capsule polysaccharides are composed of neutral and acidic sugars, respectively (19,20). The sugar acids glucuronic acid and galacturonic acid were commonly observed constituents of the different S. marcescens K serotypes tested but were lacking among different O-antigen serotypes (20). Therefore, the decrease in extracellular uronic acids observed in the pgm mutant strain (Fig. 8D) is likely specific to decreased or modified capsule production. This hypothesis is also consistent with the observed decrease in extracellular uronic acids in the wzx flippase mutant. However, it should be noted that because Pgm acts upstream in the pathway converting glucose-6P to the polysaccharide precursors UDP-glucose and UDP-glucuronate (Fig. 5A), it is possible that pgm mutation results in LPS defects in addition to those associated with the capsule.
In summary, this work represents a comprehensive mutational analysis of an S. marcescens clinical isolate to identify fitness genes required for mammalian infection. We have demonstrated the importance of capsule, glucose metabolism, and a magnesium transport protein by using this system and have identified many additional S. marcescens fitness genes that await functional analysis.

MATERIALS AND METHODS
Bacterial strains and culture conditions. All of the S. marcescens isolates used in this study (see Text S1) originated from patient bloodstream infections and were collected by the University of Michigan Hospital System Clinical Microbiology Laboratory. Bacterial isolates were identified as S. marcescens as part of routine clinical laboratory procedures. The 16S rRNA locus from each isolate was also PCR amplified and sequenced (data not shown) by previously described methods (40). Strain UMH9 Nal was obtained by isolating a spontaneously resistant mutant of wild-type strain UMH9 following culture on LB agar containing 50 g/ml nalidixic acid. LB medium (41) was used for the routine culture of S. marcescens and E. coli strains. Antibiotics were used at the following concentrations: kanamycin, 50 g/ml; gentamicin, 10 g/ml; ampicillin, 100 g/ml; streptomycin, 100 g/ml.
Construction of S. marcescens mutant strains. Recombineering-based mutagenesis of S. marcescens was performed in accordance with the protocol of Thomason et al. (42). S. marcescens UMH9 harboring plasmid pSIM19 (43) was used as the parent strain, and the oligonucleotide primers used are listed in Text S1. For additional details and a description of the methods used for genetic complementation, see Text S1.
Murine model of S. marcescens bacteremia. S. marcescens strains were cultured overnight in LB medium and then subcultured to an OD at 600 nm (OD 600 ) of 0.1 in LB and incubated at 37°C for 2 to 3 h. Bacteria were then collected by centrifugation and resuspended in an appropriate volume of phosphate-buffered saline (PBS). Female 6-to 8-week-old C57BL/6 mice were infected via tail vein injection with 0.1 ml of the bacterial suspension containing~1 ϫ 10 7 CFU unless indicated otherwise. Mice were euthanized at 24 h or at the postinfection time indicated, and the spleen, liver, and kidneys were harvested in accordance with institution-approved protocols. The bacterial burden of infected mice was determined by plating serial dilutions of organ homogenates onto LB agar. For competition infections and bottleneck assessment experiments, wild-type bacteria were mixed with antibioticresistant mutant strains at the ratios indicated prior to infection. The density of each strain was determined for both the inoculum (input) and organ homogenates (output) by serial dilution and plating on LB containing or lacking antibiotics. The CI was calculated by determining the following ratio: (CFU mutant /CFU wild type ) output /(CFU mutant /CFU wild type ) input . Colonization bottleneck experiments are described in Text S1. All animal experiments were conducted by using protocols approved by the Institutional Animal Care and Use Committee.
In vivo screen for S. marcescens fitness mutants. Frozen stocks from five mutant pools, containing 10,000 pSAM_Cm transconjugants each (see Text S1), were used to inoculate independent 20-ml LB cultures at a dilution of 1:100. Cultures were incubated for 2.5 h at 37°C before bacteria were collected by centrifugation and resuspended to a density of 1 ϫ 10 8 CFU/ml in PBS. Suspensions were used to infect four replicate C57BL/6 mice per mutant pool, for a total of 20 mice. Aliquots (1 ml) of each inoculum suspension were collected by centrifugation and stored at Ϫ80°C for subsequent isolation of genomic DNA to serve as the input samples. Two mice succumbed to infection prior to 24 h and were excluded from further analysis (pools 3 and 4, n ϭ 3). The remaining mice were sacrificed, and spleens were homogenized in 1 ml of PBS. A 20-l sample of each homogenate was removed for CFU determination, and the remaining volume was plated onto LB. The total bacterial growth recovered from each individual mouse was pooled to serve as the output samples. Bacteria from these pools were collected by centrifugation, and aliquots were frozen at Ϫ80°C for subsequent isolation of genomic DNA. The quantitative sequencing of transposon insertion sites and the identification of fitness genes and essential genes are described in Text S1.
Microscopic examination of capsule production. Capsule production by S. marcescens strains was assessed by negative staining with CR dye. Bacteria from LB plates incubated at 37°C were suspended in an aqueous solution of 1% CR and allowed to dry on the surface of a slide. As a control, bacteria were stained with a 1% solution of crystal violet for 2 min. Images were captured on an Olympus BX60 microscope with a 100ϫ objective. To determine the size of capsule negative staining and of crystal violet-stained bacterial cells, images were converted to binary and quantitated with Fiji (44). Three fields per slide were captured, and the areas from~300 cells per condition were plotted as a frequency distribution. Only cells that were well isolated were measured.
Quantitation of extracellular uronic acids. Extracellular uronic acids were measured on the basis of previously published methods (45)(46)(47) and extracted from overnight cultures of S. marcescens strains cultured in LB medium. For additional details, see Text S1.
Growth of S. marcescens in magnesium-depleted medium. The growth of the mgtB mutant in the presence of defined concentrations of magnesium was measured in M9 medium (48). The base medium contained 0.4% glucose, 0.1% Casamino acids, 0.1 mM CaCl 2 , 36 M FeSO 4 , and 5 g/ml thiamine. Magnesium-replete M9 was additionally supplemented with 1.0 mM MgSO 4 . Magnesium-depleted M9 medium was prepared by treatment of the base medium with Chelex 100 resin (Bio-Rad) prior to the addition of divalent cations. Following treatment, CaCl 2 and FeSO 4 were added along with MgSO 4 at the concentrations indicated. S. marcescens strains were incubated in magnesium-replete M9 overnight, washed once, and then subcultured into magnesium-depleted M9 for 3 h. Strains were subcultured a second time in 100-well microplates containing magnesium-depleted M9 supplemented with different concentrations of magnesium. Growth was measured by OD 600 determination every 15 min with a BioScreen C Analyzer at 37°C with continuous shaking.
Growth and survival of S. marcescens in human serum. The sensitivity of S. marcescens strains to serum bactericidal activity was tested by incubating bacteria in the presence of 10% normal human serum (Innovative Research). Bacteria were harvested from exponential-phase cultures by centrifugation and washed in PBS. Approximately 2 ϫ 10 6 CFU were added to serum diluted in PBS and were then incubated for 90 min at 37°C. Following incubation, bacterial suspensions were serially diluted and plated on LB agar for CFU determination. Serum HI by treatment at 56°C for 30 min was used as a negative control.
Growth of S. marcescens in human serum was tested with the M9 base medium described above supplemented with 20% HI serum. Overnight LB cultures of S. marcescens were harvested by centrifugation and washed once prior to the inoculation of 0.3-ml culture volumes into M9 basal medium supplemented with 20% HI human serum at an OD 600 of 0.01. The OD 600 of cultures was determined every 15 min with a BioScreen C Analyzer. Additional growth curves in M9 medium supplemented with 0.4% glucose and 0.1% Casamino acids were generated in the same manner. Bacterial aggregation in cultures containing 20% serum limited the accuracy of OD 600 measurements, and thus, the density of viable bacteria over time was determined from independent 25-ml cultures grown under the same conditions.
In Vivo Fitness of S. marcescens ®