Genome-Wide Identification of Fitness Factors in Mastitis-Associated Escherichia coli

ABSTRACT

Virulence factors of mammary pathogenic Escherichia coli (MPEC) have not been identified, and it is not known how bacterial gene content influences the severity of mastitis. Here, we report a genome-wide identification of genes that contribute to fitness of MPEC under conditions relevant to the natural history of the disease. A highly virulent clinical isolate (M12) was identified that killed Galleria mellonella at low infectious doses and that replicated to high numbers in mouse mammary glands and spread to spleens. Genome sequencing was combined with transposon insertion site sequencing to identify MPEC genes that contribute to growth in unpasteurized whole milk, as well as during G. mellonella and mouse mastitis infections. These analyses show that strain M12 possesses a unique genomic island encoding a group III polysaccharide capsule that greatly enhances virulence in G. mellonella. Several genes appear critical for MPEC survival in both G. mellonella and in mice, including those for nutrient-scavenging systems and resistance to cellular stress. Insertions in the ferric dicitrate receptor gene fecA caused significant fitness defects under all conditions (in milk, G. mellonella, and mice). This gene was highly expressed during growth in milk. Targeted deletion of fecA from strain M12 caused attenuation in G. mellonella larvae and reduced growth in unpasteurized cow's milk and lactating mouse mammary glands. Our results confirm that iron scavenging by the ferric dicitrate receptor, which is strongly associated with MPEC strains, is required for MPEC growth and may influence disease severity in mastitis infections.

IMPORTANCE Mastitis caused by E. coli inflicts substantial burdens on the health and productivity of dairy animals. Strains causing mastitis may express genes that distinguish them from other E. coli strains and promote infection of mammary glands, but these have not been identified. Using a highly virulent strain, we employed genome-wide mutagenesis and sequencing to discover genes that contribute to mastitis. This extensive data set represents a screen for mastitis-associated E. coli fitness factors and provides the following contributions to the field: (i) global comparison of genes required for different aspects of mastitis infection, (ii) discovery of a unique capsule that contributes to virulence, and (iii) conclusive evidence for the crucial role of iron-scavenging systems in mastitis, particularly the ferric dicitrate transport system. Similar approaches applied to other mastitis-associated strains will uncover conserved targets for prevention or treatment and provide a better understanding of their relationship to other E. coli pathogens.

INTRODUCTION

Escherichia coli is a diverse species, encompassing a wide variety of strains that originate through frequent genetic exchange. While many E. coli strains are commensals that form part of the normal intestinal flora, there are several pathotypes that cause intestinal or extraintestinal illness. Extraintestinal pathogenic E. coli (ExPEC) strains include those that cause urinary tract infections, neonatal meningitis, pneumonia, and sepsis in humans, those that cause airway infections and septicemia in birds, and those that infect mammary glands and cause mastitis in lactating mammals. Of these strains, we understand the least about mastitis-associated E. coli, despite its critical impact on the dairy industry, where it causes annual losses totaling nearly $2 billion and endangers animal health (1, 2).

A mammary pathogenic E. coli (MPEC) pathotype has been proposed (3), but the defining characteristics of this group have not been identified. While virulence genes have been functionally associated with the intestinal E. coli pathotypes (e.g., Shiga toxin, translocated intimin receptor, etc.), we lack a similar understanding of the basis for virulence in the unique environment of the lactating mammary gland despite many years of investigation (4–10). This has led to the conclusion that strains causing mastitis do not have specific virulence factors and that it is merely their introduction into the mammary gland and the subsequent inflammatory response that result in mastitis symptoms (10, 11). In this model, disease severity is primarily dependent on host factors. However, it is unlikely that E. coli strains causing mastitis are random since not all E. coli strains are pathogenic in experimental models of the disease (12). Additionally, mastitis-associated strains are less genotypically diverse than environmental strains, indicating selective pressure for specific virulence or fitness factors. Nevertheless, several recent genome comparison studies have failed to consistently identify genes present more often in strains associated with mastitis than in nonpathogenic strains (13–17), and functional genomic studies aimed at identifying genes required by MPEC to colonize the mammary gland have not been reported.

In order to cause mastitis, bacteria must gain access through the teat canal and overcome the immune defenses constitutively present in the mammary gland. The mammary gland is a transiently mucosal tissue, producing mucosal secretions only during lactation. Mammary gland epithelial cells secrete numerous soluble antimicrobial defenses to impede bacterial growth, and these are amplified by an influx of inflammatory cells during infection. Milk contains components of innate immunity such as antimicrobial peptides, lysozyme, lactoferrin, and complement. Milk also provides a unique nutritional environment for bacteria, being high in fats and sugars and with limited availability of free amino acids and iron (18–20). Shortly after infection, bacteria are usually eliminated by the inflammatory response that follows, resulting in transient mastitis. However, some MPEC strains cause severe, acute disease resulting in permanent mammary gland dysfunction, sepsis, or death. Other strains may cause persistent subclinical infection. MPEC strains that cause transient or persistent mastitis may have distinct in vitro phenotypes such as motility, suggesting that subtypes defined by carriage or differential expression of specific genes may exist (21, 22). It is also possible that some MPEC strains could cause disease in other tissues, but as yet the genes associated with virulence in multiple hosts or colonization sites are unknown.

We hypothesize that, in addition to differences in host immune factors, variability in the colonizing strains plays a significant role in determining the severity or nature of the mastitis that develops in individual cases. Targeted interference of bacterial virulence factors as well as a better understanding of the most effective host defenses could lead to improved treatment options. However, we currently do not know enough about MPEC virulence factors to develop these approaches in an informed way. There is clearly a need for an unbiased, broad-scale approach to identify genes required by MPEC for colonization and dissemination. In this study, we demonstrated the utility of the lepidopteran host Galleria mellonella for predicting MPEC virulence. G. mellonella has been used to measure the virulence of several bacterial and fungal pathogens (23–25). We then used massively parallel transposon insertion sequencing (Tn-seq) (26–28) to identify genes that allow MPEC to survive in whole unpasteurized milk, in G. mellonella, and in murine mammary glands and spleens. We confirmed the relevance of the Tn-seq data through targeted deletion mutations. Our results reveal a large number of genes important for MPEC survival under these distinct conditions. We confirmed the essential role of the MPEC ferric dicitrate transport system in obtaining iron and counteracting host nutritional immunity.

RESULTS

Virulence of MPEC isolates in Galleria mellonella.

G. mellonella has been used previously to measure infectivity of ExPEC clinical isolates (23, 24). To determine whether these insects would be applicable to studying MPEC pathogenesis, we screened 20 MPEC strains that were isolated from clinical cases of bovine mastitis for their ability to kill G. mellonella. The MPEC strains, which were primarily of phylogroup A, exhibited a broad range of virulence levels in these assays (Table 1). The most virulent strain (M12) killed all the larvae at the lowest inoculum (10³ CFU). In subsequent experiments with lower infectious doses, we determined that the 50% lethal dose (LD₅₀) for this strain is between 500 and 1,000 CFU. Conversely, none of the larvae were killed by several of the strains (such as isolate M6) even at the highest dose (10⁵ CFU). Additional infections with less virulent isolates showed that both M6 and M11 were able to kill some of the larvae at later time points but were still attenuated in virulence compared to that of strain M12 (Fig. 1A).

TABLE 1.

Mortality of G. mellonella larvae infected with MPEC strains^a

^aPercentage of dead larvae determined 24 h after infection (n = 10). Virulence for each strain was classified as low (green), moderate (orange), or red (high). Highly virulent strains killed >50% of larvae at the lowest infectious dose. Moderately virulent strains killed >50% of larvae at the highest dose.

FIG 1.

Virulence of MPEC strains in G. mellonella predicts virulence potential in a mouse mastitis model. (A) G. mellonella larvae were infected with 10³ CFU of MPEC strains M6, M11, M12, and the control strain E. coli DH5α. Mortality of infected larvae was determined over a 72-h period. M12 virulence was significantly greater than that of the MPEC strains M6 (P < 0.0001) and M11 (P = 0.0003) (log rank [Mantel-Cox] test). (B) Infection of mouse mammary glands (MG) with strains M6 and M12. Bacteria (250 CFU) were inoculated directly into the teat canal of lactating mice. After 24 h, strain M12 grew to significantly higher numbers in the mammary glands (*, P = 0.029, by Mann-Whitney test) than strain M6. Strain M12 also colonized the spleens of the infected mice whereas strain M6 was not detected.

We next sought to determine if the results of the G. mellonella infection assay could be predictive of the virulence of MPEC isolates in mammary gland infections. Because of the sharp contrast in their virulence levels in G. mellonella, strains M6 and M12 were selected and injected into the mammary glands of lactating mice. The M12 strain grew to significantly higher numbers in mammary glands than strain M6. In addition, strain M12 spread and replicated in the spleens whereas M6 did not (Fig. 1B). These results suggest that the G. mellonella infection assay may be capable of predicting the virulence potential of MPEC strains in a mouse model of mastitis, but testing a larger number of strains is required to fully determine this relationship. Our results further suggest that G. mellonella could also be used to identify virulence factors relevant to mastitis or disease in other host environments.

Genome sequencing and annotation of MPEC strain M12.

The ability of M12 to infect G. mellonella and grow to high numbers in mouse mammary glands and disseminate to the spleen suggested that this strain would be particularly useful in discovering genes that are needed for multiple aspects of MPEC infection. We therefore sequenced and partially assembled the genome of strain M12 (see Table S1 in the supplemental material). Annotation of the assembled sequence revealed that, in common with most other MPEC isolates, M12 possesses genes for type VI secretion systems, long polar fimbriae, and iron scavenging (13, 14, 17, 29). Unlike most MPEC strains, M12 belongs to phylogroup D and to the multilocus sequence type 69 (ST69) complex. Some ST69 strains have been associated with fecal contamination of surface waters and with human ExPEC infections (30). They have also been isolated from cattle and livestock attendants and have shown enhanced ability to colonize, persist, and adapt to different hosts (23, 31–33).

Genome-wide identification of genes required for MPEC growth in milk, G. mellonella, and mice.

We sought to make a comprehensive comparison of genes required for M12 growth in unpasteurized cow's milk, G. mellonella larvae, and mouse mammary glands and spleens. A Tn-seq approach was used to measure the contribution of each gene to the relative fitness in these environments. A library consisting of ∼400,000 unique M12 transposon insertion mutants derived from 40 independently derived pools was created and passaged through the four experimental conditions (milk, G. mellonella larvae, and mouse mammary glands and spleens) (see Materials and Methods), followed by recovery in Luria-Bertani (LB) broth cultures. The mutant populations were then collected and sequenced in parallel with the control population.

Between 17.8 and 34.8 million reads were generated for each sample, with 88.8 to 93.7% of the reads mapping to single locations in the M12 genome. In total, there were 89,747 unique insertion sites which were detected across the different conditions, which corresponded to an average density ranging from 70 to 158 nucleotides between transposon insertion sites in the libraries. The transposon insertion sites were then tabulated for each condition (Data Set S1). Reproducibility between the replicate samples for each condition was very high, with R² values between 0.90 and 0.98 (Fig. S1).

To identify genes that may contribute to the fitness of M12 under each of the conditions, input (LB medium) and output libraries were compared using the DESeq package (34, 35). Genes that had a log₂ fitness score of −2 or lower (i.e., at least 4-fold fewer insertions in the output pool compared to the input pool) and a P value of 0.05 or lower were identified as candidate genes (Data Set S1). We compared the candidate fitness genes for each of the test conditions according to their functional categories (Fig. 2). The 25 genes wherein insertions caused the greatest decreases in fitness under each of the four conditions are listed in Table S2.

FIG 2.

Functional classification of genes required for M12 fitness in milk, G. mellonella, and mouse mammary glands and spleens. Genes with a log₂ fitness score of −2 or lower (i.e., at least a 4-fold reduction in insertions in the output pool relative to the input pool) and a P value of 0.05 or lower were identified as candidate genes and grouped according to their annotated function.

It appears that there is general consistency in the proportions of genes that fall within each functional category (with some exceptions discussed below), but the individual genes within the categories varied. However, fewer genes are required for growth in milk than under the in vivo conditions (Fig. 2). The nucleotide and amino acid biosynthesis categories contained a large proportion of the putative fitness genes (Fig. 2) and were prominent among those with the greatest effects on fitness (Table S2). Many of those required for growth in milk or in mice are predicted to be involved in purine metabolism and the biosynthesis of phenylalanine, histidine, and serine. Genes involved in nucleotide metabolism have previously been shown to be required for E. coli growth in blood and spleens (36, 37). Our results indicate they are also required for growth in milk, a further indication that de novo synthesis pathways are crucial virulence factors in mammals.

The virulence, disease, and defense category represented a larger proportion of the total genes under in vivo conditions than during growth in milk (Fig. 2). These genes primarily encode membrane transport functions, including metal resistance and efflux genes, and are listed in Table S3. Genes required for resistance to cellular stresses were also prominent, especially under the in vivo conditions. M12, like other E. coli strains, encodes three superoxide dismutase (SOD) enzymes that utilize different metals as cofactors. The sodA (encoding Mn-SOD), sodB (Fe-SOD), and sodC (Cu/Zn-SOD) genes were each predicted to enhance M12 fitness in G. mellonella (Fig. S2). In addition to SOD enzymes, the Dps protein helps bacteria survive oxidative stress by binding to DNA and iron to prevent damage from iron-catalyzed Fenton reactions (38). Insertions in the dps gene reduced M12 fitness under each of the in vivo conditions but not in milk (Fig. S2). The HtrA-family protein DegP removes misfolded periplasmic proteins during oxidative or heat stress (39) and has been shown to be required for E. coli survival during mouse bladder infections (40). DegP is activated by the sigma factor RpoE, which our Tn-seq data indicate is essential for viability in M12. RpoE is repressed by the anti-sigma factor RseA (41). The degP and rseA genes also appeared to be required by M12 in the in vivo environments but not in milk.

In addition to oxidative stress resistance, the transcriptional regulatory protein NsrR and the nitric oxide reductase flavorubredoxin are required for E. coli responses to nitric oxide (42). Insertions in the nsrR gene reduced M12 fitness in mice and G. mellonella, and insertions in the norV gene reduced fitness in mice but not in G. mellonella (Fig. S2). The requirement for several oxidative and nitrosative stress resistance genes particularly in vivo but not in whole milk may be attributed to the initiation of inflammatory responses. Influx of neutrophils or hemocytes, activation of prophenoloxidase or degranulation cascades, and upregulation of the oxidative or nitric oxide burst occur only in the presence of living phagocytic cells, and thus bacterial defense mechanisms are most critical under these conditions. Many of these genes are known to be virulence factors for ExPEC in other environments, and these results suggest they may also be important for MPEC mastitis.

Genes encoding a type III polysaccharide capsule are required for virulence in G. mellonella.

We were interested in identifying genes encoding bacterial surface structures that have direct interactions with host cells or surveillance proteins. One cluster of genes which appeared to contribute significantly to fitness only in G. mellonella was predicted to encode the synthesis and transport of a polysaccharide capsule (Fig. 3A). Examination of the assembled M12 contigs showed that these genes are found on a putative genomic island in M12 not previously associated with MPEC strains. Within the 33-kb island are kpsS, kpsC, and kpsDEMT genes ordered according to group III capsule synthesis and export systems (43). The island also contains nucleotide sugar biosynthesis, glycosyltransferase, and modification genes in this region in an arrangement that does not appear in any other publicly available bacterial genome sequence, suggesting lateral gene transfer and/or extensive recombination, possibly resulting in a novel capsule type.

FIG 3.

A unique genomic island in MPEC strain M12 encodes a type III capsule that contributes to virulence. (A) Genetic organization of the 33-kb island containing capsule biosynthesis genes, including kpsS and kpsC which were deleted in the ΔkpsCS mutant strain. (B) Tn-seq determination of fitness costs (log₂ fitness score) associated with disruptions to M12 kpsC and kpsS genes in milk, G. mellonella, or during mouse infections (***, P < 0.001). (C) Gel electrophoresis and Alcian blue staining of capsular material shows that the mutant strain fails to produce capsule that is detectable in the wild-type strain and complement (pMO1). (D) The ΔkpsCS mutant strain is avirulent in G. mellonella, with no lethality evident in larvae injected with up to 30,000 bacteria in contrast to complete killing by the wild-type and complemented strains (**, P = 0.0043; ****, P < 0.0001, by log rank test). (E) Competition between the M12 wild-type strain and ΔkpsCS mutant (500 CFU) was measured during mammary gland infection in lactating mice. After 24 h, mice were sacrificed, and the competition index (means ± standard errors of the means) was determined (*, P = 0.0002, by one-sample t test). ns, not significant.

The Tn-seq data indicated that the capsule genes were particularly important for G. mellonella infection but were detrimental to fitness during colonization of mouse mammary glands (Fig. 3B). To confirm the validity of these Tn-seq results, the contribution of this putative capsule to survival in G. mellonella and mouse mastitis was assessed. A mutant strain with a deletion of the kpsCS genes predicted to be necessary for initiation of capsule synthesis was generated (44). The production of capsule in the wild-type M12 bacteria was examined using SDS-PAGE and alcian blue staining (Fig. 3C). In contrast to the abundant capsule produced by the wild-type strain, as expected, the mutant strain failed to produce detectable capsule, and replacement of the kpsCS genes on a multicopy plasmid partially restored capsule production. In G. mellonella infections, the ΔkpsCS mutant strain was much less virulent than the wild-type strain (Fig. 3D), consistent with the Tn-seq result. Complementation restored virulence to nearly wild-type levels, correlating with capsule production. Fitness of the ΔkpsCS mutant during mouse mammary gland infections was then tested in competition assays with the wild-type strain. Unlike the Tn-seq results which predicted that the mutant would have enhanced fitness, the ΔkpsCS mutant had a slight but statistically significant decrease in survival after a 24-h infection in the mammary glands of mice (Fig. 3E).

Nutrient-scavenging genes.

The genome sequence of strain M12 and the Tn-seq results together suggest a critical role for nutrient scavenging in mastitis pathogenesis. M12 encodes multiple iron acquisition systems, including receptors for the enterobactin (fepA) and yersiniabactin (fyuA) siderophores. It also encodes the sitABCD and mntH uptake systems for iron and manganese, as well as a high-affinity zinc uptake transport system (znuABC). The Tn-seq results indicate that genes encoding the enterobactin receptor and uptake system are necessary for bacterial growth in G. mellonella and in spleens (Fig. 4A). While other siderophores have been previously shown to contribute to ExPEC virulence, a role for enterobactin has not been identified (45). Conversely, the yersiniabactin receptor genes do not appear important for survival in G. mellonella or in mammary glands but do appear to be important for survival in spleens (Fig. 4A). The transport systems for zinc (znuABC) and iron/manganese (sitABCD) also appear to be required for fitness in G. mellonella and in mice (Fig. 4A).

FIG 4.

Nutrient-scavenging genes are required for fitness in G. mellonella, mice, and growth in milk and are highly expressed during growth in milk. (A) Fitness costs (log₂ fitness score) associated with disruptions to M12 nutrient-scavenging genes in milk, G. mellonella, or during mouse infections as measured by Tn-seq (*, P < 0.05; **, P < 0.01; ***, P < 0.001). (B) The change in transcription levels of iron receptors fecA, fepA, fhuE, and sitA was determined 4 h after transfer to whole unpasteurized milk. The change in gene expression levels was determined relative to that of control cultures in LB medium (*, P = 0.0273; **, P = 0.0092; ***, P = 0.0019, by one-sample t test).

Transposon insertions in the fecA gene also resulted in significantly less growth in milk, G. mellonella, and mouse mammary glands (Fig. 4A). The fecABCDE operon encodes an uptake transport system for ferric dicitrate (46). This chelated form of iron effectively reduces the iron availability in environments such as milk and insect hemolymph. Thus, citrate can be considered an iron sequestration factor that may prevent bacterial growth in these environments (47–49). The ferric dicitrate receptor FecA, along with its accessory proteins FecB and FecC, enables bacteria to transfer ferric dicitrate across the outer membrane and to incorporate essential iron (50, 51). Insertions in the fecA, fecB, and fecC structural genes, as well as the sigma factor fecI which activates transcription of the operon in response to ferric dicitrate, decreased fitness in milk and under the in vivo conditions.

The requirement for iron scavenging systems for growth in milk indicates the iron-limited nature of this environment. To further investigate which of the iron acquisition systems might be important for growth, we measured expression of four iron-scavenging genes following transfer of actively growing M12 culture into whole unpasteurized cow's milk (Fig. 4B). These genes are all known to be regulated by the availability of iron at least partially through the ferric uptake regulator protein Fur (52–55). Three of four genes (fecA, fepA, and sitA) were strongly upregulated 1 h after transfer to cow's milk. These results confirm that iron availability is severely limited in milk, likely as a consequence of its sequestration by lactoferrin and citrate, and that the ferric dicitrate system is especially highly expressed.

Ferric dicitrate transporter (fecA) enhances M12 growth and virulence.

Previous genome sequencing and PCR-based screens of MPEC strains have found that these strains are more likely than other pathotypes to carry genes for type VI secretion systems, long polar fimbriae, and the ferric dicitrate uptake system (fecABCDE). According to our Tn-seq results, the type VI secretion system and long polar fimbriae are not required for fitness under any of the experimental conditions. The Tn-seq data indicated a strong fitness defect when the fecA operon was disrupted (Fig. 4A), and the fecA gene was highly upregulated during growth in milk (Fig. 4B), which suggested an important role for this system.

To verify the role of the ferric dicitrate iron-scavenging system in overcoming iron limitation, we created a deletion mutant of the outer membrane receptor fecA gene in strain M12. Compared to the wild-type strain, the ΔfecA mutant grew significantly less well in bovine or mouse milk in single-culture experiments (Fig. 5A). The growth defect of the mutant strain was much more pronounced during growth in cow's milk than in the mouse milk, which may be due to the higher abundance of citrate in cow's milk (16, 56, 57). We then complemented the mutation by replacing the fecABCDE genes on a multicopy plasmid and tested the fitness of the mutant and complemented strains in 1:1 competition assays with the wild-type strain in cow's milk (Fig. 5B). The mutant had an approximately 3-log reduction in fitness, and the competitive fitness was restored to wild-type levels in the complemented strain.

FIG 5.

The fecA gene is required for optimal M12 growth in whole milk and for virulence in G. mellonella and in murine mammary glands. (A) Growth of the M12 wild-type or ΔfecA mutant strain was measured in whole unpasteurized milk (cow or mouse) after 4 h. The mutant strain had significantly decreased growth in both cow and mouse milk (*, P = 0.033; **, P = 0.0046, by Student's t test). (B) Competitive fitness of the M12 wild-type strain, ΔfecA mutant, or complemented (pMO2) strain was determined by competition assays in cow's milk. The mutant strain was significantly less fit (*, P < 0.0001, by one-sample t test) than the wild-type strain whereas the complemented strain was not. (C) Virulence of M12 wild-type or ΔfecA mutant bacteria measured in G. mellonella infections. Groups of larvae were infected with the indicated number of bacteria, and survival was measured 24 h after infection. The ΔfecA mutant strain was significantly less virulent in the larvae (**, P < 0.01; ***, P < 0.001, by log rank test). (D) Virulence of M12 wild-type or ΔfecA mutant bacteria during mouse infections. Lactating mice were infected with ∼300 CFU of either the wild-type or ΔfecA mutant strain, and bacterial numbers in mammary glands or spleens were determined after 24 h. Growth of the mutant strain was significantly different from that of the wild type in mammary glands (**, P = 0.0022, by Mann-Whitney test) but not in spleens. (E) Competitive fitness levels of the ΔfecA mutant strain were compared during infection of lactating and nonlactating mice. Mice were infected with mixtures of mutant and wild-type M12 strains, and bacterial growth in the mammary glands was measured after 24 h. The fitness of the ΔfecA mutant strain was significantly lower during infection of lactating mice (*, P < 0.0001, by one sample t test) but not in nonlactating mice. ns, not significant.

We then tested the virulence of the ΔfecA mutant in G. mellonella and in mouse mastitis infections. Virulence of the mutant was severely attenuated in the larvae compared to that of the wild-type strain over a range of inoculum concentrations (500 to 50,000 CFU) (Fig. 5C). The virulence of the mutant was restored by complementation with the fecABCDE plasmid. In mouse mammary gland infections, numbers of the mutant strain bacteria were 10- to 100-fold lower than those of the wild-type strain 24 h after infection (Fig. 5D). However, the mutant and wild-type strains achieved similar concentrations in the spleens of the infected mice. This suggested that the ΔfecA mutation prevents the acquisition of iron from citrate complexes in milk but does not impact iron acquisition from other sources. To more fully test this hypothesis, we compared the fitness of the ΔfecA mutant strain in competition with the wild-type strain in lactating and in nonlactating mouse mammary glands (Fig. 5E). As with the single-infection experiments, the mutant was less fit in lactating mice in the competition experiments. In contrast, the growth of the mutant strain was not significantly different from that of the wild type during infection of nonlactating mice. Therefore, in environments where citrate is present (in vitro in milk, lactating mouse mammary glands, and insect hemolymph), fecA enables significantly more bacterial growth but is dispensable in environments where citrate is not present (nonlactating mammary glands and spleens).

DISCUSSION

In this study, we utilized Tn-seq to identify genes required by MPEC in environments that are relevant to its natural history, including growth in milk, exposure to an effective innate immune system, and colonization of mammary tissues. We showed that a G. mellonella larva model was able to discriminate between strains with high and low virulence potentials in mammary glands, providing a relevant, convenient method to investigate the molecular pathogenesis of MPEC strains. The variability in virulence levels toward G. mellonella among this relatively small set of MPEC strains was unexpectedly high, as mastitis-associated E. coli bacteria are commonly perceived as generic inducers of inflammation. Most previous studies investigating the nature of mastitis-associated strains concluded that they are commensals belonging mainly to phylogroups A and B1 (5, 9). While mastitis-associated strains show increased growth in milk relative to that of other strains (4, 13), they are thought to possess few known virulence factors that would be predicted to influence the severity of disease in mammary glands or to allow them to cause disease in other contexts (10, 11). Our results challenge this interpretation. Instead, they support the idea that some mastitis-associated strains may share features in common with other phenotypic groups of E. coli and that more detailed studies are required to elucidate virulence genes that may be strain specific (58). This is reminiscent of other recent findings where strains classified as nonpathogenic have been isolated from the bloodstream of immunocompromised patients (59), and intestinal pathogenic O157:H7 strains have been recovered from urinary tract infections (60). It is possible that the variation among MPEC strains in their ability to infect and kill G. mellonella correlates with virulence in mammary gland infections only. Alternatively, this variation may reflect a previously unappreciated ability of some mastitis-associated strains to cause disease in other contexts, which is deserving of further investigation.

Similar to milk, G. mellonella hemolymph contains iron-sequestering proteins and citrate, which severely limit free iron available to invading pathogens (47, 61). It also contains lysozyme and antimicrobial peptides that inhibit or kill bacteria (62). Their hemocytes phagocytize and kill bacteria in a similar fashion as neutrophils (63). They also activate a proteolytic phenoloxidase cascade functionally similar to complement at the surface of bacteria, leading to engulfment and deposition of toxic reactive oxygen molecules (64). It is not surprising, then, that a large number of genes important for survival in mouse mammary glands, including genes related to stress resistance and detoxification as well as nutrient acquisition functions, were also required in G. mellonella. For instance, the inner membrane zinc uptake genes znuABC, which are also required for Acinetobacter baumannii virulence in G. mellonella (65), appear important for M12 survival in mice and G. mellonella. High-affinity transport by bacteria can combat the effects of the zinc- and manganese-chelating protein calprotectin, which is abundant in neutrophils and is released as a component of extracellular traps (66–68). Neutrophil influx, degranulation, and extracellular trap formation are early critical events in the mammary gland immune response to mastitis (69, 70), and mammary epithelial cells may also release calprotectin (71). Zinc and manganese are cofactors for the superoxide dismutases sodA and sodC, which our data suggest are required for MPEC virulence. Additional experiments need to be carried out to determine if G. mellonella has a calprotectin-like protein and to clarify the role of znuABC in dealing with oxidative stress in the context of mammary gland infections.

The mastitis-associated strain M12 was particularly virulent in the G. mellonella model and grew prolifically in mouse mammary glands. We found that a significant portion of its virulence in G. mellonella can be attributed to production of a type III capsule encoded on a genomic island. A mutant unable to produce capsule was avirulent in the larvae and slightly attenuated in mouse mammary gland infections. Some group II or III capsules have been shown to contribute to urinary tract or bloodstream ExPEC survival (72–75). However, capsule production has not been previously shown to affect virulence of E. coli in mastitis. Microscopically visible production of capsule material by mastitis isolates has been observed (76), but phage or PCR-based screens for capsule genes has generally shown only sporadic carriage by mastitis-associated strains (6, 9, 15). Capsule production could provide a slight advantage to strain M12 in the mammary gland by masking bacterium-associated molecular patterns and delaying initial detection, enhancing complement resistance, or providing protection against neutrophils and macrophages. The much larger reduction in fitness of the ΔkpsCS mutant in the G. mellonella infections suggests that there may be environments outside the mammary gland where this group III capsule confers a more significant advantage. The reason for the dramatically reduced virulence of the mutant during G. mellonella infections requires further investigation, but it suggests that this capsule may have a more prominent role in enhancing survival of the bacteria in their intestinal reservoir or during infections in other extraintestinal niches (58).

Our study confirms the role of the ferric dicitrate transport system encoded by the fecA operon, at least for one mastitis-associated E. coli strain. The fecA gene has been one of very few that consistently appears in screens of MPEC strains, which has led to speculation that it confers a key advantage for growth in mammary glands. Based on its antigenic consistency among mastitis-causing strains, Lin et al. proposed vaccinating cattle with recombinant FecA protein in order to elicit anti-FecA antibodies that could limit bacterial growth (56). However, they found that vaccinated cattle produced antibody that was insufficient to affect bacterial growth in mammary glands of the vaccinated animals (77). It is unclear whether the antibodies in their study were able to completely block FecA function or if the bacteria were able to acquire iron via other methods (78). Our Tn-seq data suggest that the ferric dicitrate and enterobactin siderophore iron acquisition systems are nonredundant, at least for the M12 strain in mouse mammary gland infections.

Our results clearly demonstrate that a mutation eliminating FecA function has a strong detrimental effect on bacterial replication in environments that contain citrate, including cow's milk and mouse mammary glands. Citrate is an iron chelator, and bacteria with the ability to utilize this iron source have a growth advantage. In addition to ferric dicitrate, ferric-lactoferrin is the other major iron source potentially available to bacteria in milk. Siderophores such as enterobactin or yersiniabactin could remove iron from lactoferrin, enabling bacterial acquisition through ferric-siderophore receptors. However, production of siderophores is an energetically expensive process and is susceptible to interference by other bacteria. Therefore, strains may achieve faster growth by fecA-mediated iron acquisition than by siderophore-mediated acquisition. Levels of citrate and lactoferrin in the milk of dairy cattle vary depending on the time of lactation (49). Citrate levels are highest relative to those of lactoferrin early in lactation, a time at which cattle are most susceptible to severe mastitis. Later in lactation the levels of citrate fall, and typically mastitis is less severe. It is possible that utilization of ferric citrate as an iron source contributes to the propensity for severe mastitis early in lactation, but the increased energy costs associated with using lactoferrin reduce bacterial growth rates later in the lactation cycle, tipping the balance toward resolution of infection and milder disease.

While the majority of mastitis-associated strains may be fitter in the mammary gland environment because they express FecA, it is clearly not a universal requirement. Substantial variation exists among ExPEC isolates even within the same niche, and the emergence of tropism for mammary glands has likely evolved among many separate lineages. Epidemiological and experimental approaches aimed at identifying genes that explain the properties of MPEC are unlikely to identify any single ubiquitous locus (15). Rather, like other ExPEC types (79, 80), there may exist a spectrum of overlapping virulence gene requirements. The necessity of any single gene for survival in mammary glands may depend on the genetic background of the individual strain. Such variability should be considered in the development of therapeutic strategies that will be broadly effective in the treatment of E. coli mastitis.

MATERIALS AND METHODS

Bacterial strains and growth conditions.

E. coli strains (Table 1) were generously supplied by Paolo Moroni (Cornell University Animal Health Diagnostic Center, Ithaca, NY). They were isolated from individual quarter milk samples from cattle with clinical mastitis. Putative E. coli bacteria were identified based on colony morphology on blood and MacConkey agar plates, followed by potassium hydroxide, cytochrome oxidase, citrate, ornithine decarboxylase, lysine, motility, and indole biochemical tests. E. coli strains were routinely grown in Luria-Bertani (LB) medium at 37°C. To select for mutant strains or carriage of plasmids, growth medium was supplemented with kanamycin (50 μg/ml), chloramphenicol (10 μg/ml), or ampicillin (100 μg/ml) as appropriate. For milk cultures, whole, unpasteurized cow's milk was obtained from a local supplier and used immediately or stored at −80°C until use. Mouse milk was obtained as described previously (81) and used immediately after collection.

Phylogroup analysis of the isolated E. coli strains was performed as described by Clermont et al. (82). Multiplex PCR products were separated on agarose gels, and carriage of the genes chuA and yjaA and the DNA fragment TspE₄ was determined. Primer sequences used in this study are given in Table 2.

TABLE 2.

Primers and plasmids used in this study

Name	Sequence (5′–3′) or descriptiona	Purpose
Primers
TspE4C2.2	CGCGCCAACAAAGTATTACG	Phylogroup analysis
TspE4C2.1	GAGTAATGTCGGGGCATTCA
YjaA.2	TGGAGAATGCGTTCCTCAAC
YjaA.1	TGAAGTGTCAGGAGACGCTG
ChuA.2	TGCCGCCAGTACCAAAGACA
ChuA.1	GACGAACCAACGGTCAGGAT
fecA-Cm F	TTCTCGTTCGACTCATAGCTGAACACAACAAAAATGATGATGGGGAAGGTGTTGATCGGCACGTAAGAGG	Knockout of fecA
fecA-Cm R	CAACATAATCACATTCCAGCTAAAAGCCCGGCAAGCCGGGCGTTAACACAAAAATTACGCCCCGCCCTG
fecA F	ATGACGCCGTTACGCGTTTT	Complementation of fecA
fecA R	TCAGAACTTCAACGACCCCT
kps-Cm R	GTTCACAGAGACAACATGTATTTATATACGGGATCGAAAGGGATTCTTGCCAAAATTACGCCCCGCCCTG	Knockout of kpsCS
kps-Cm F	TCGCTGCGGTTGTTGGCTGATTGAGTAACCTGAAACGCCCATGAAAACCAGTTGATCGGCACGTAAGAGG
kps R	TTTTGTAACACTGACTGAATCGGC	Complementation of kpsCS
kps F	CACATTCTGGCTGTACGAAAAAGT
rssA 3 F	CGAAGCGACGGGAACAATTT	qPCR of housekeeping gene
rssA 3 R	GATGGATCTCTCCTGGCAGC
hcaT 2 F	CTTTTCCGCCGTTGTAGTGC
hcaT 2 R	ACCACTATCAACCACGGCAA
fecA qF	GACACCTACGGCAATCTGGTA	qPCR of target gene
fecA qR	CTGGACTGGAAATCGCTGTTC
fepA F	CTACAGTAAAGGTCAGGGCTG
fepA R	CAAGGAGATTGGTCTGGAGTTC
fhuE F	ATAGAGGTATAGCTGGCGTAGG
fhuE R	GTAAGTCAGCGTATCAACCCG
sitA F	CATATCCCGGCAGTCTTTAGC
sitA R	GCTCTATGTCGATTCCCTGAG
1TN	CTGACCCGGTCGAC	Binds slightly back from the transposon end
1OLIGOG	CAGACGTGTGCTCTTCCGATCGGGGGGGGGGG	Binds the C tail, adds part of the index primer binding site
2TNAnn	AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNTCGAGATGTGTATAAGAGACAG	Adds Illumina adapter and variable numbers of random (N) bases to increase read diversity among clusters in Illumina sequencing
2TNBnn	AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNTCGAGATGTGTATAAGAGACAG
2TNCnn	AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNTCGAGATGTGTATAAGAGACAG
2TNDnn	AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCTNNNNNNNTCGAGATGTGTATAAGAGACAG
BAR01	GATCGGAAGAGCACACGTCTGAACTCCAGTCACATCACGATCTCGTATGCCGTCTTCTGCTTG	Adds Illumina adapter and incorporates barcode to identify samples on multiplexed lanes
BAR02	GATCGGAAGAGCACACGTCTGAACTCCAGTCACCGATGTATCTCGTATGCCGTCTTCTGCTTG
BAR03	GATCGGAAGAGCACACGTCTGAACTCCAGTCACTTAGGCATCTCGTATGCCGTCTTCTGCTTG
BAR04	GATCGGAAGAGCACACGTCTGAACTCCAGTCACGCCAATATCTCGTATGCCGTCTTCTGCTTG
BAR05	GATCGGAAGAGCACACGTCTGAACTCCAGTCACACAGTGATCTCGTATGCCGTCTTCTGCTTG
BAR06	GATCGGAAGAGCACACGTCTGAACTCCAGTCACCAGATCATCTCGTATGCCGTCTTCTGCTTG
BAR07	GATCGGAAGAGCACACGTCTGAACTCCAGTCACTGACCAATCTCGTATGCCGTCTTCTGCTTG
BAR08	GATCGGAAGAGCACACGTCTGAACTCCAGTCACGATCAGATCTCGTATGCCGTCTTCTGCTTG
BAR09	GATCGGAAGAGCACACGTCTGAACTCCAGTCACACTTGAATCTCGTATGCCGTCTTCTGCTTG
BAR10	GATCGGAAGAGCACACGTCTGAACTCCAGTCACCTTGTAATCTCGTATGCCGTCTTCTGCTTG
Plasmids
pRE112	Source of chloramphenicol resistance gene for creating knockout strains
pKD46	Lambda Red recombination plasmid
pJET1.2	Cloning vector for creating complemented strains
pJG714	Tn5 delivery plasmid for creating transposon mutant libraries
pMO1	Complementation plasmid containing kpsSC genes
pMO2	Complementation plasmid containing fecABCDE genes

^aItalicized sequences are complementary to the chloramphenicol resistance gene from pRE112.

G. mellonella infections.

G. mellonella larvae were purchased from Best Bet, Inc. (Blackduck, MN), stored at 15°C in the dark, and used within 2 weeks. An overnight culture of bacteria was subcultured and grown to an absorbance of 1.0 at 600 nm and diluted in phosphate-buffered saline (PBS). The concentration of each inoculum was determined by serial dilution and colony counting after 24 h of growth on LB agar plates. Larvae without any evidence of melanization were selected and injected through the left hindmost proleg with 10 μl of the inoculum (containing between 10² and 10⁵ CFU) using a Hamilton 701RN syringe and a 30-gauge needle. Control larvae were injected with 10 μl of sterile PBS. The larvae were incubated at 37°C, and survival was monitored at the end of a 12-h period or at intervals over 72-h period.

Ethics statement.

Mouse experiments were performed in accordance with the recommendations found in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health (83). The protocol was reviewed and approved by the Institutional Animal Care and Use Committee of Brigham Young University (protocol no. 160302).

Mouse infections.

Lactating BALB/c mice between 9 and 12 weeks of age were used in all experiments. Lactating mothers 10 to 11 days postpartum were chosen for all experiments. Mice were infected based on the method described by Brouillette and Malouin (84) with slight modifications. Briefly, mice were anesthetized using 75 to 100 μl of a ketamine-xylazine solution. An overnight culture of bacteria was subcultured and grown to an absorbance of 1.0 at 600 nm and diluted in PBS. The concentration of each inoculum was determined by serial dilution and colony counting after 24 h growth on LB agar plates. A 50-μl volume of bacteria (250 CFU) in PBS was then injected directly through the teat canal into the ductal network of the L4 and R4 (fourth on the left and fourth on the right, respectively) mammary glands using a Hamilton model 1705TLL syringe and a 33-gauge needle with a beveled end. Pups were removed for 1 to 2 h after injections and then were returned. Control glands were inoculated with 50 μl of sterile PBS in both the L4 and R4 glands. Mice were monitored for 24 h, and tissue was subsequently harvested. Bacterial load in the mammary gland and spleen was determined by homogenizing the tissue in 1 ml of PBS and performing serial dilutions and colony counting after 24 h growth on LB agar plates.

M12 genome sequencing, assembly, and annotation.

Total DNA from strain M12 was isolated using a ZR Fungal/Bacterial DNA MiniPrep kit (Zymoresearch). The DNA was sheared and sequenced at the University of Southern California NextGen Sequencing Core. Illumina paired-end reads of 250 bp were generated on MiSeq, version 2, sequencer. Contigs were built from these reads using the Velvet short read sequence assembler (85). The contigs were then ordered with Mauve using the E. coli K-12 sequence as a reference (86). The partially assembled genome was annotated using a RAST online automated annotation service (87, 88).

Gene deletions and complementation.

Gene deletions of kpsCS and fecA were generated via lambda Red recombination in strain M12 carrying the plasmid pKD46 (89). Gene replacement constructs were created by PCR using pRE112 as a template to amplify the chloramphenicol resistance gene, cat, with 50 bp of homology to the target genomic sites. Bacteria expressing the recombinase enzymes were electroporated with 200 to 500 ng of purified DNA, and potential mutants were selected by plating on LB medium containing chloramphenicol. Potential mutants were then screened by PCR using primers flanking the recombination site and also with primers internal to the kpsCS or fecA genes. To complement the mutations, the fecABCDE or kpsCS genes were PCR amplified and cloned into the multicopy plasmid pJET2.1. The resulting plasmids were then used to transform the mutant strains.

Tn-seq screens.

The experimental workflow for the Tn-seq screens is outlined in Fig. S3 in the supplemental material. The M12 mutant library was created by biparental mating with the diaminopimelic acid (DAP) auxotroph E. coli MFDpir harboring plasmid pJG714, which contains a Tn5 transposon cassette in which the kanamycin resistance gene points out one side and the Salmonella P_trp promoter points out the other. Donor and recipient bacteria were spotted on LB agar supplemented with DAP for 2 h, after which the mixture was spread onto extra-large plates of LB agar containing kanamycin (50 μg/ml) without DAP. Separate colonies from 40 plates containing 10,000 independent colonies were then pooled to create a library of strain M12 consisting of ∼400,000 unique transposon insertion mutants. The library was frozen in 1-ml aliquots of LB broth with 15% glycerol for Tn-seq screens.

The transposon library was subcultured 1:10 into fresh LB broth and allowed to recover for 1 h before Tn-seq screens in LB broth, milk, G. mellonella, mammary gland, and spleens. For in vitro conditions, 4 × 10⁸ CFU were inoculated into 100 ml of unpasteurized cow's milk or LB broth in duplicates and grown for 12 h and then diluted 1:10 into fresh LB broth and grown for an additional 8 h. Bacteria were then harvested and frozen in LB broth–15% glycerol. For in vivo conditions, 5,000 CFU were injected into 2,000 larvae, and 10⁶ CFU were injected into the L4 and R5 mammary glands of six mice and incubated for 24 h. The larvae, mammary glands, and spleens were harvested, divided equally into biological duplicates, and homogenized in PBS. The homogenates were inoculated into LB broth containing kanamycin and grown for 8 h, harvested, and frozen in LB broth containing 15% glycerol.

DNA preparation and Illumina sequencing.

Cells from the input pool (LB broth) and output pools (milk, larvae, mammary glands, and spleen) were pelleted, and genomic DNA was extracted using a ZR Fungal/Bacterial DNA MiniPrep kit (Zymoresearch). Genomic DNA (150 ng) was digested using double-stranded DNA (dsDNA) Fragmentase (NEB) for exactly 12 min to create fragments that were between 500 bp and 3,000 bp. Genomic DNA fragments were isolated and cleaned using a QIAprep Spin Miniprep kit (Qiagen). To facilitate the PCR amplification of transposon insertion junctions, cytosine was added to the ends of the fragments using terminal transferase (NEB). C-tailed transposon junctions were PCR amplified (Phusion) for 15 cycles, and Illumina adapters with barcodes were added by an additional 20 cycles with an extension time of 20 s to generate products between 150 and 700 bp. The PCR products were column purified prior to sequencing. Primers used to amplify transposon junction sites and add Illumina sequencing adapters with barcodes are listed in Table 2. The 10 DNA samples with their unique barcodes were pooled with the same concentrations based on measurements from a NanoDrop spectrophotometer (ThermoFisher).

Tn-seq data analysis.

Data analysis was performed using the sequence alignment tool TnSeq Pipeline (90) and consisted of mapping the transposon back to the M12 reference genome using Bowtie2, read normalization, and the removal of reads located at the edges of genes to ensure that functional genes are not included in downstream analyses (91). Raw count tables for each gene generated by TnSeq Pipeline were analyzed by the DESeq package in R. The fold change (i.e., ratio of input/output read counts) values were log₂ transformed, and a P value was calculated using a negative binomial distribution performed by DESeq. Candidate genes identified to have a fitness cost under experimental conditions needed to pass two criteria: a log₂-fold change of −2 or less and a P value of 0.05 or less. Additionally, genes that were normalized for length that had few or no insertions were categorized as putative essential genes.

RNA isolation and qPCR.

Bacteria were grown overnight in LB broth and then subcultured to exponential phase (A₆₀₀ of 1.0). Washed bacterial cells were then transferred to either unpasteurized whole cow's milk or mouse milk or into LB broth at a concentration of 10⁸ CFU/ml and incubated for 30 min at 37°C. Milk samples were diluted 5-fold into cold PBS, pelleted, and resuspended in cold PBS. The samples were then diluted 5-fold in RNAlater (ThermoFisher) and stored at −80°C. RNA was isolated using an rBAC Mini total RNA kit (IBI Scientific) according to the manufacturer's instructions. The integrity of the isolated RNA was verified by gel electrophoresis. The RNA was used to make cDNA using a ProtoScript II First Strand cDNA synthesis kit (NEB). Quantitative PCR (qPCR) primers specific for each gene were designed to give 100- to 150-bp products (Table 2). Reaction mixtures for qPCR consisted of 2× Sybr Green master mix and High ROX master mix (Genesee Scientific) with 3 μM (each) forward and reverse primers, and reactions were performed using a StepOne real-time PCR system. The cycling conditions were as follows: 95°C for 15 min followed by 40 cycles of 95°C for 15 s and then 60°C for 1 min. A melt curve analysis was performed to confirm the specificity of the PCR amplification. The resulting threshold cycle (C_T) values were normalized as described by Vandesompele et al. (92) to the stably expressed genes rssA and hcaT. The fold change in expression levels of the iron genes in milk relative to those in the LB cultures was determined using the comparative ΔΔC_T values.

Capsule isolation and staining.

Capsule production was visualized as previously described (93). Briefly, bacteria were pelleted from 5 ml of saturated overnight cultures grown in LB broth. The bacterial cells were then resuspended in 1 ml of PBS and heated to 55°C for 1 h. The bacteria were removed by centrifugation, and the supernatants were treated with proteinase K for 1 h at 59°C. The samples were concentrated using 20-kDa-molecular-mass-cutoff Millipore filters, followed by electrophoresis on 10% SDS-polyacrylamide gels and staining with 0.125% alcian blue dye in 40% ethanol–5% acetic acid.