Low-dose inoculation of Escherichia coli achieves robust vaginal colonization and results in ascending infection accompanied by severe uterine inflammation in mice

Escherichia coli infection of the female reproductive tract is a significant cause of disease in humans and animals, but simple animal models are lacking. Here we report that vaginal inoculation of uropathogenic E. coli strains UTI89 and CFT073 in non-pregnant, estrogen-treated mice resulted in robust colonization of the vagina and uterine horns, whereas titers of the lab strain MG1655 were significantly lower. Non-estrogenized mice also became colonized, but there was more variation in titers. A dose of 104 colony-forming units (CFU) UTI89 was sufficient to result in colonization in all estrogenized mice, and we also observed bacterial transfer between inoculated and uninoculated estrogenized cage mates. UTI89 infection led to inflammation and leukocyte infiltration into the uterine horns as evidenced by tissue histology. Flow cytometry experiments revealed that neutrophil, monocyte and eosinophil populations were significantly increased in infected uterine horns. This model is a simple way to study host-pathogen interactions in E. coli vaginal colonization and uterine infection. There are immediate implications for investigators studying urinary tract infection using mouse models, as few E. coli are required to achieve reproductive colonization, resulting in an additional, underappreciated mucosal reservoir.


Introduction
Bacterial infections of the female reproductive tract (FRT) are common and can carry serious consequences. Escherichia coli is an opportunistic pathogen and frequent cause of human disease [1], including in the FRT [2]. Some women are vaginally colonized with E. coli in the apparent absence of symptoms, with colonization rates ranging from~5% [3,4]  inoculation, mice received two doses of 0.5 mg β-estradiol 17-valerate (Sigma, St. Louis, MO, USA) in 100 μl filter-sterilized sesame oil (Fisher, NH, USA) via intraperitoneal injection to synchronize the mice in estrus [23]. At the time of infection, 10 4 colony-forming units (CFU) of bacteria were resuspended in phosphate-buffered saline (PBS) and inoculated into the vaginas in 20 μl volumes. For mock infections (controls for histology and flow cytometry experiments), mice were estrogenized, then inoculated with 20 μl PBS and housed in cages separate from infected mice (except during transmission experiments where mock-infected mice were co-housed with infected mice). At experimental end points, animals were humanely euthanized by isoflurane inhalation followed by cervical dislocation. To enumerate bacterial CFUs, vaginal washes were collected using PBS as previously described [23], serially diluted and plated on MacConkey agar containing 25 μg/ml kanamycin or 20 μg/ml spectinomycin. At experimental end points, mice were humanely sacrificed and vaginas and uterine horns were harvested, weighed, homogenized and serially diluted in PBS, then plated on MacConkey agar containing 25 μg/ml kanamycin or 20 μg/ml spectinomycin for CFU enumeration.

Observations of the estrous cycle
To monitor the estrous cycle when testing the duration of estradiol treatment, vaginal washes were collected after 1, 2, 3, 6,8,10,13,15,17,20,22,25,27,29 and 31 days (a total of 15 times over the 31-day experiment) and wet mounts were prepared and observed at 20x magnification on a Nikon (Tokyo, Japan) Diaphot inverted microscope. Images were acquired using an AmScope (CA, USA) microscope digital camera (MU300, 3.1 MP). Representative images were assessed by four independent investigators and the consensus date on which each mouse came out of estrus was plotted; at least three out of four investigators agreed for each mouse.

Vaginal wash scoring
To assess immune cell infiltration into vaginal wash fluid, at indicated time points, 5 μl of vaginal wash was smeared onto a slide and stained with the Hema 3 staining kit (Fisher, NH, USA) after air drying. Immune cells and epithelial morphology were assessed on an Olympus (Tokyo, Japan) BX61 bright field microscope.

Histopathology
The reproductive tract (vagina, cervix, and uterine horns) was harvested as an intact structure, fixed overnight in 10% neutral-buffered formalin (Fisher, NH, USA), paraffin-embedded and sectioned. Slides were stained with hematoxylin and eosin, then imaged with an Olympus (Tokyo, Japan) BX61 bright field microscope. An observer blinded to the infection status of the samples scored the level of polymorphonuclear cells in the tissue based on the following rubric: 0 -absent, 1 -minimal, 2 -moderate, 3 -profound, 4 -severe (including immune cells crossing the epithelium into the lumen).

Flow cytometry
Replicates were conducted with five mice per group (1-2 mock-infected with PBS and 3-4 E. coli UTI89-infected). At 12 days post-infection, mice were humanely euthanized without perfusion and the vagina, cervix and uterine horns were collected and kept on ice in 1 ml RPMI buffer containing 1% fetal bovine serum, 1% penicillin/streptomycin, and 50 μM 2-mercaptoethanol. Single cell suspensions were generated and flow cytometry was performed as previously described [

Results
In this study, we sought to establish a simple mouse model of E. coli infection of the female reproductive tract. In C57BL/6 mice, after vaginal inoculation with 10 4 colony-forming units (CFU) of the clinical E. coli isolate UTI89, E. coli titers in vaginal washes rose and fell over time (Fig 1 and S1 Fig). Of 17 mice tested in two replicates, two had stably high titers over time, three never had detectable titers at any time point, and 12 had variable titers that were sometimes high, sometimes low and sometimes undetectable. The percentage of mice with any detectable titer was greatest at one day post-infection (dpi), with titers detected in 14 mice (76%), and decreased over time.
We reasoned that the heterogeneity we observed in E. coli vaginal colonization in . These prior studies did not report how long β-estradiol treatment is effective, so we first assessed the duration of the estrus phase in estrogenized mice for one month after treatment. C57BL/6 mice from Jackson or Envigo were given two doses of 0.5 mg β-estradiol 17-valerate by intraperitoneal injection and vaginal washes were collected over time and scored according to the criteria in [27] (Fig 2). Interestingly, we observed a vendor effect: three of five Envigo mice exited estrus by 13 days post-treatment, and by 31 days, only one mouse was still in estrus. In contrast, only one Jackson mouse exited estrus (by day 13) during the 31-day period. Thus, for subsequent experiments we used only estrogenized C57BL/6J (Jackson) mice ( Fig 3A).
We found that 24 hours after vaginal inoculation with 10 4 CFU of UTI89, estrogenized mice had vaginal titers clustered around 10 5 CFU/mg tissue, with a wider spread of uterine horn titers ranging from undetectable to 10 5 CFU/mg ( Fig 3B). We wondered whether colonization would be stable over time in estrogenized mice. Indeed, we detected consistently high vaginal wash titers over the course of 12 days (Fig 3C), between 10 7 and 10 8 CFU/ml. Some mice remained colonized for three weeks or more (Fig 3D), in keeping with the long duration of estrus in estrogenized Jackson mice.
The vagina is considered a reservoir for uropathogenic E. coli, which may ascend through the urethra to establish urinary tract infections [28]. Thus, we compared FRT infection with the well-characterized urinary tract infection clinical isolates UTI89 and CFT073 versus the lab strain MG1655, in each case using a 10 4 inoculum. Bacterial titers were enumerated at 12 dpi (Fig 3E-3G). Both UTI89 and CFT073 colonized the FRT robustly, with median titers of 10 8 CFU/ml vaginal wash,~10 5 CFU/mg vaginal tissue, and~10 3 to~10 4 CFU/mg uterine horn tissue; despite high bacterial titers, mice did not appear sick (no hunching, hair loss, lethargy, etc.). In contrast, MG1655 titers were statistically significantly lower throughout the FRT. Strikingly, 4 of 10 mice infected with MG1655 had no recoverable uterine horn CFUs, whereas bacteria were recovered from the uterine horns of all mice infected with UTI89 (P < 0.05, Fisher's exact test) or CFT073.
To test whether E. coli uterine infection reflects the extent to which the vagina becomes robustly colonized, we looked for significant correlations among titers of UTI89, CFT073 and MG1655 in vaginal washes, vaginal tissue and uterine horn tissue with Spearman correlation  for each strain individually, we found that UTI89 titers were not significantly correlated in any comparison (Fig 4A-4C, second panels). CFT073 titers were significantly correlated or trended toward significance for all comparisons (Fig 4A-4C, third panels), and only MG1655 vaginal wash and uterine horn titers were significantly correlated (Fig 4A-4C, fourth panels). These data suggest that bacterial strain-specific differences may impact the association between vaginal and uterine colonization in this model. We considered whether the acts of vaginal inoculation and/or washing may have introduced E. coli into the uterine horns. However, E. coli was not detectable in the uterine horns of mice sacrificed immediately after vaginal inoculation of 10 4 CFU of UTI89 ( Fig 5A). As well, similar levels of UTI89 were recovered from the uterine horns at seven dpi in all (four of four) mice that were never vaginally washed following inoculation (Fig 5B, compare to Fig 3G). Finally, we observed spontaneous E. coli transmission from infected to mock-infected (PBSinoculated) cage mates. In these experiments, three UTI89-infected mice were co-housed with two mock-infected mice (all estrogenized). At 12 dpi, vaginal and uterine horn titers were enumerated. Two experiments of this kind were performed with different housing conditions (see Methods). In both experiments, kanamycin-resistant UTI89 could be recovered from both vaginal and uterine horn homogenates of co-housed, mock-infected animals ( Fig 5C). Taken together these findings demonstrate that in estrogenized mice, ascending infection of the uterine horns can occur with low-level vaginal inoculation of E. coli or following transmission between cage mates, and does not require investigator intervention.
During the normal estrous cycle, white blood cells infiltrate the vaginal lumen after ovulation (metestrus and diestrus); keeping mice in a pseudo-estrus phase via administration of β-estradiol 17-valerate prevents this influx of white blood cells. Here, light microscopy of Hema 3-stained smears of vaginal washes collected at one, seven and 12 days post-UTI89 infection revealed white blood cells and parabasal epithelial cells (Fig 6A). A blinded assessment revealed significantly more white blood cells in infected mice compared to mock-infected (PBS-inoculated) controls, and higher numbers of parabasal cells, though not statistically significant (P = 0.11) (Fig 6B).
Since estrogenization suppresses vaginal leukocyte infiltration, we sought to further characterize the source of the inflammatory infiltrates in the FRT in response to E. coli infection. We first performed blinded scoring of tissue sections collected at 12 dpi from infected and mockinfected (PBS-inoculated) mice. We observed no evidence of vaginal histological inflammation (inflammatory score of zero) in three of three mock-infected mice and ten of 11 UTI89-infected mice (with one mouse having a score of two, defined by moderate infiltration of polymorphonuclear cells) (Fig 7A and 7B), suggesting that the white blood cells present in vaginal washes were not vaginal in origin. In the uterine horns, we observed polymorphonuclear leukocytes (PMNs) in the tissue of one of eight mock-infected mice, and in the tissue and/or lumen of 11 of 13 UTI89-infected mice (P < 0.01, Fisher's exact test) (Fig 7C and 7D). PMNs were also present in the uterine horn tissue and/or lumen in six of 11 MG1655-infected mice (Fig 7C and 7D), though only two of 11 had the highest score (defined by profound tissue inflammation plus immune cells crossing the epithelium into the lumen) compared to six of 13 UTI89-infected mice. Comparable levels of inflammation were seen in each horn of a given mouse. Because both horns were used for histology, bacterial titers of the uterine horns are not known for these experiments; however, vaginal washes were collected. UTI89 vaginal wash titers were not significantly correlated with uterine horn histology scores ( Fig 7E); although there was a wide spread of uterine histology scores, UTI89 vaginal wash titers were all clustered between 10 7 and 10 8 CFU/ml vaginal wash. In contrast, MG1655 vaginal wash titers and In two experiments (the first shown in black and the second in grey), a cage of five mice were estrogenized. Two were mock-infected with PBS and three were infected with 10 4 CFU of UTI89. At 12 dpi, UTI89 was detected in all infected mice, and was also detected in the vagina of four out of four mock-infected cage mates and in the uterine horns of three out of four mock-infected cage mates. All mice came from the Jackson Laboratory; experiments from panels A and B and replicate one in panel C were performed with static (non-ventilated) microisolator cages, and replicate two in panel C was performed with HEPA-filtered cages with airflow control. Zeros are plotted at the limit of detection and bars indicate median values.
https://doi.org/10.1371/journal.pone.0219941.g005 uterine histology scores suggest a different relationship in which higher bacterial numbers were seen alongside higher uterine inflammatory scores, but a positive correlation did not quite reach statistical significance (P = 0.0512) (Fig 7F).
To further characterize immune cell infiltration into the infected FRT, we performed flow cytometry to enumerate neutrophils, monocytes, macrophages, and eosinophils at 12 dpi in UTI89-infected mice relative to mock-infected (PBS-inoculated) mice (see S2 Fig for our gating strategy). In accordance with tissue histology, we did not observe higher levels of any of the above cell types in infected vaginal tissue compared to the mock-infected control group (Fig 8A). However, infected uterine tissues had significantly higher numbers of neutrophils, monocytes and eosinophils compared to mock-infected uterine horns (Fig 8B). Evaluation of these immune cells in cervical tissue did not reveal differences in the numbers of these immune cell subsets between mock-infected and E. coli-infected animals, suggesting that the cervix is not the source of these immune cells ( S3 Fig). Thus, ascending E. coli infection of the FRT elicits a strong uterine inflammatory response and severe immunopathology in the uterine horns, despite potential immunosuppressive effects of exogenous estrogen treatment.

Discussion
E. coli is a relatively common pathogen of the FRT, and ascending E. coli infection causes a variety of negative health outcomes [13,29]. Here we describe a model of ascending E. coli It is important to note that most of the experiments presented here employed treatment with exogenous estrogen (so-called "estrogenization") to keep mice in pseudo-estrus and  coli levels that we observed in non-estrogenized mice may be due to normal leukocyte infiltration during the estrous cycle. As well, the apparent lack of vaginal inflammation in estrogenized mice at 12 dpi, despite sustained, robust E. coli colonization, may be attributable to local immunosuppression due to estrogenization. Nonetheless, we observed abundant leukocytes in vaginal washes from infected, estrogenized mice as early as one dpi. Tissue histology and flow cytometry at 12 dpi suggested that the uterine horns, not the vagina, were the source of the inflammatory cells. Thus, the immunosuppressive effects of estrogen can be overcome, at least in the uterine horns. We note that uterine inflammation does not appear to be a general consequence of Gram-negative bacterial infection in estrogenized mice, as high-titer Prevotella bivia ascending infection resulted in neither localized (uterine) nor systemic inflammation by histological analysis and/or flow cytometry at two and seven dpi [33]. Our finding of uterine inflammation upon E. coli infection in mice is consistent with a previous report wherein surgical infusion of E. coli directly into the uteri of non-pregnant C57BL/6 mice caused severe inflammation [34], which was attributed to TLR4-driven signaling in both epithelial and stromal cells [35]. Our non-surgical model serves as a straightforward means of testing candidate virulence factors or vaccine antigens in the context of severe innate immune cell infiltration, which could yield findings relevant to E. coli infections at other body sites, such as the gut, bladder, and kidneys-or in the neonatal lung, blood and brain.
Our finding that MG1655 titers in the FRT were statistically significantly lower than titers of the uropathogenic strains UTI89 and CFT073 echoes findings in pregnant mice that a pathogenic E. coli strain caused more robust ascending uterine infection than the lab strain MG1655 and caused significantly greater upregulation of the neutrophil chemoattractants Cxcl1 and Cxcl2 in uterine tissue, although in that study titers were not determined [17]. Here we observed that for the two robust colonizers, UTI89 and CFT073, E. coli titers in vaginal washes and vaginal tissue (at 12 dpi) were relatively tightly clustered around~10 8 CFU/ml vaginal wash and 10 5 CFU/mg vaginal tissue, respectively. In contrast, uterine horn titers had a much wider range of~10 2 to~10 6 CFU/mg tissue. For both CFT073 and MG1655, uterine horn titers were significantly correlated with vaginal wash titers, and for MG1655, uterine horn histology scores were also significantly correlated with vaginal wash titers, suggesting that for these two strains, the ascending infection was a direct result of robustness of the vaginal infection. UTI89 did not display significant correlations between vaginal and uterine horn titers or uterine horn histology scores, and therefore it remains unknown why UTI89 uterine infection was so robust in some animals and more muted in others. Because our histology and flow cytometry experiments used the entire organ, the uterine horn titers are not known for these experiments and cannot be directly tested for correlations with immune cell infiltrates. Future studies should test whether the degree of uterine inflammation is correlated (positively or negatively) with UTI89 uterine horn titers, to better understand the pathophysiology of ascending E. coli uterine infection.
Although UTI89 and CFT073 are both uropathogenic E. coli strains belonging to the B2 clade, and are more similar to one another than they are to MG1655, they are genetically distinct strains and some key virulence factors are only found in UTI89 [36]; for example, UTI89 encodes the cytotoxic necrotizing factor CNF1, which is not found in CFT073 or MG1655 and which could therefore impact UTI89 colonization specifically. Others have reported heterogeneity in the virulence factor profiles of E. coli isolated from the FRT of pregnant and non-pregnant women [2,37,38]. However, little is known about the specific bacterial factors that may mediate colonization. We propose our model as a straightforward means of testing candidate antigens for their role in FRT infection and inflammation.
A few studies examining the beneficial role of Lactobacillus species as vaginal probiotics have tested the ability of Lactobacillus to prevent or clear vaginal E. coli infection in mice [39,40] and cynomolgus monkeys [41]. Our study highlights the importance of using multiple E. coli strains for such experiments, as we showed that strains differ in their ability to colonize the vagina. We note that mice harbor a vaginal microbiome that is distinct from the Lactobacillusdominated "healthy" microbiome seen in women [42]. In the present study, we did not assess whether or how the vaginal microbiota may have changed in response to estrogenization or E. coli infection, and we did not look for any bacteria other than E. coli in the uterine horns. Thus, whether and to what extent E. coli infection may disrupt existing microbial communities-or conversely, how the composition of the endogenous microbiome may influence susceptibility to ascending infection-remain open questions.
Mouse models of E. coli bladder infection typically use a 10 7 or 10 8 CFU dose (for example, [22]), while here we found that a vaginal dose of 10 4 CFU was sufficient to colonize all estrogenized animals. The observation that infected mice could apparently transmit E. coli to cohoused, uninfected mice provides further support for our conclusion that E. coli can very readily colonize the estrogenized FRT. Here we note that the second replicate of this experiment was performed after our mouse facility changed to a new housing system (see Methods). Although the overall conclusion of these experiments, that E. coli can be transferred between cage mates, held true, we did note lower overall levels of E. coli in uninfected cage mates following the change in housing. Whether these differences are due to aspects of the inoculation, such as trauma, fluid flow, anesthesia etc., and/or factors relating to the housing, such as bedding that is more dry and harboring fewer bacteria, remains to be elucidated.
Together with our observations that uropathogenic E. coli was an especially capable vaginal colonizer and that even non-estrogenized animals can carry E. coli vaginally for long periods, these findings suggest that urinary tract infection models may also result in significant E. coli FRT reservoirs, a phenomenon that may vary in intensity depending on housing conditions and other factors. There are multiple reported mouse models of chronic and/or recurrent bladder infection reported in the literature (for example, [22,43]). The likely outcome that E. coli in urine could result in vaginal colonization and ascending uterine infection in these models requires new consideration that an FRT reservoir may contribute to certain aspects of bladder pathophysiology in mice, as it does in humans. Importantly, elegant experiments previously showed that experimental bladder infection with UTI89 in pregnant mice caused intrauterine growth restriction (low gestational and birth weights), due to inflammation-mediated "crosstalk" between organ systems [44]. Whether cross-talk may be bi-directional-that is to say, whether uterine inflammation caused by ascending uterine E. coli infection could also affect UTI outcomes-is an open question. Another potential technical consideration is that "clean catch" procedures for urine collection do not exist for mouse experiments and thus vaginal colonization could potentially contribute to E. coli titers detected in urine.
A mouse surgical infection model was previously used to study the E. coli factors necessary for uterine colonization [34,35]. Our work shows that uterine E. coli infection can be studied without surgery, making future studies less technically challenging. As well, given the importance of E. coli uterine infection in causing pregnancy complications, a few studies have investigated vaginal E. coli inoculation in pregnant rodents [16,17,45], though without assessing bacterial titers, histology or immune cell infiltration. Performing vaginal infections in the context of pregnancy is essential, but also technically demanding and resource-intensive. Our model provides a less technically complicated, less invasive, and more natural means to investigate host-pathogen interactions in ascending E. coli uterine infections. We hope that this model may be used to discover bacterial virulence factors promoting ascension and inflammation, and to test therapeutic interventions. Those discoveries can then be validated in pregnancy models, which may lead to improved health outcomes.
Supporting information S1 Fig. Vaginal E. coli titers wax and wane in mice not treated with exogenous estrogen. Seventeen non-pregnant female Jackson C57BL/6 mice were vaginally inoculated with E. coli strain UTI89, and vaginal washes were collected to monitor infection status over 12 days. Each panel corresponds to one replicate of the experiment. Each point shows an individual mouse, with lines connecting the titers of a given mouse from one time point to the next. The percentage of mice with any detectable titer is shown in the box at the top of the graph. Zeros are plotted at the limit of detection. These experiments were conducted with static (non-ventilated) micro-isolator cages. Writing -review -editing: Amanda L. Lewis.