Uropathogenic Escherichia coli infection-induced epithelial trained immunity impacts urinary tract disease outcome

Previous urinary tract infections (UTIs) can predispose one to future infections; however, the underlying mechanisms affecting recurrence are poorly understood. We previously found that UTIs in mice cause differential bladder epithelial (urothelial) remodelling, depending on disease outcome, that impacts susceptibility to recurrent UTI. Here we compared urothelial stem cell (USC) lines isolated from mice with a history of either resolved or chronic uropathogenic Escherichia coli (UPEC) infection, elucidating evidence of molecular imprinting that involved epigenetic changes, including differences in chromatin accessibility, DNA methylation and histone modification. Epigenetic marks in USCs from chronically infected mice enhanced caspase-1-mediated cell death upon UPEC infection, promoting bacterial clearance. Increased Ptgs2os2 expression also occurred, potentially contributing to sustained cyclooxygenase-2 expression, bladder inflammation and mucosal wounding—responses associated with severe recurrent cystitis. Thus, UPEC infection acts as an epi-mutagen reprogramming the urothelial epigenome, leading to urothelial-intrinsic remodelling and training of the innate response to subsequent infection.

Previous urinary tract infections (UTIs) can predispose one to future infections; however, the underlying mechanisms affecting recurrence are poorly understood. We previously found that UTIs in mice cause differential bladder epithelial (urothelial) remodelling, depending on disease outcome, that impacts susceptibility to recurrent UTI. Here we compared urothelial stem cell (USC) lines isolated from mice with a history of either resolved or chronic uropathogenic Escherichia coli (UPEC) infection, elucidating evidence of molecular imprinting that involved epigenetic changes, including differences in chromatin accessibility, DNA methylation and histone modification. Epigenetic marks in USCs from chronically infected mice enhanced caspase-1-mediated cell death upon UPEC infection, promoting bacterial clearance. Increased Ptgs2os2 expression also occurred, potentially contributing to sustained cyclooxygenase-2 expression, bladder inflammation and mucosal wounding-responses associated with severe recurrent cystitis. Thus, UPEC infection acts as an epi-mutagen reprogramming the urothelial epigenome, leading to urothelial-intrinsic remodelling and training of the innate response to subsequent infection.
Urinary tract infections (UTIs) are one of the most common bacterial infections worldwide and are a substantial cause of morbidity in otherwise healthy females 1,2 . The high recurrence rate in susceptible individuals makes treatment challenging 3 , with one of the strongest risk factors for developing a UTI being a previous UTI 2 . The biological basis for recurrent UTI (rUTI) is poorly understood.
Without antibiotic treatment, acute UTIs in humans either self-resolve or develop into long-lasting chronic infections 4 . Infection of C3H/HeN mice with uropathogenic Escherichia coli (UPEC) recapitulates these two outcomes. Chronic cystitis in these mice is defined as persistent high titre bacteriuria (bacteria in urine) accompanied by chronic inflammation (sensitized mice), while resolution of infection defines resolved mice 5 . Analysis of bladders from resolved and sensitized mice reveals that infection leads to differential bladder remodelling that impacts susceptibility to rUTI. Resolved mice are resistant to rUTI, in part because they have an accelerated, transient Article https://doi.org/10.1038/s41564-023-01346-6 cells 12 , was used as a control. As both 5637 cells and our primary USCs are of basal cell origin, we seeded and cultured them on transwells for 2-3 weeks to compare cell differentiation phenotypes. Formation of intact differentiated urothelium, referred to herein as differentiated urothelium (or urothelia), by the primary USCs was confirmed by robust transepithelial electrical resistance (TER) (Fig. 1b), strong confocal staining of large hexagonal superficial facet cells for the terminal differentiation marker K20 (Fig. 1c), and the presence of cell junctions and uroplakin plaques on the tissue surface (Fig. 1d), similar to how superficial facet cells appear on the surface of intact bladder tissues 6 . Microscopic analysis of differentiated urothelial sections revealed a multilayer polarized epithelial (Ecad+) tissue layer ( Fig. 1e-h), with p63, K5 and K14 expression in the basal layer cells above the matrigel membrane (mostly K5+/K14+, a few K5+/K14−), p63 and weak Upk3a expression in some midlayer cells, indicative of intermediate cells, and strong K20 and Upk3a staining on the surface of the apical layer cells, indicating terminally differentiated superficial facet cells. Altogether, the distribution of these differentiation characteristics in basal, intermediate and superficial urothelial layers is consistent with that observed in mouse and human bladder tissue 13 . In contrast, none of these features were observed in 5637 cell transwell cultures, in which the cells were loosely packed together in layers 4-6 cells deep, but lacked evidence of polarization or cell junction formation (Extended Data Fig. 2a-d). Thus, primary USCs provide important advantages over a tumour cell line for studying the urothelium.

Differentiated urothelia mimic in vivo infection phenotypes
Using our murine model of rUTI 5,6 , we have shown that an initial UTI event results in long-lasting bladder remodelling, including structural and proteomic changes to the urothelium, that impact susceptibility to rUTI 14 . To investigate the role of USCs in bladder remodelling, we isolated bladder USCs from convalescent mice (4 weeks after the initiation of antibiotics), from both those with self-resolved infection (resolved) and those that developed chronic infection (sensitized), as well as from age-matched naïve mice as controls and established USC lines (n = 4 per infection history) (Fig. 2a,b). We propagated each of these USC lines for between 15 and 30 passages, then differentiated them on transwells and characterized each cell line by microscopy after 2-3 weeks of culture. Strikingly, we found that the urothelium derived from sensitized USCs recapitulated many of the morphological differences observed previously in vivo 6 even after many passages, including smaller surface cells and decreased expression of the terminal differentiation markers Upk3a and K20 when compared with differentiated urothelium derived from naïve and resolved USCs (Fig. 2c-e). Automated measurement of surface cell sizes showed that the apical cells of the sensitized differentiated urothelia are significantly smaller than those in naïve differentiated urothelia, whereas the resolved urothelia have an intermediate phenotype, consistent with in vivo data ( Fig. 2f-g) 6 . Collectively, these data indicate that transwell culture of USCs can recapitulate the morphological bladder epithelial remodelling phenotypes seen in vivo.

Epigenetic characteristics associate with disease history
Our data indicate that a previous infection results in USC-intrinsic changes that are heritable over many generations of cell culture. Therefore, we hypothesized that these changes are mediated by differential epigenetic modification in the USCs that associate with: (1) disease history and (2) genome-wide differences in chromatin accessibility. Chromatin accessibility was measured by Omni-ATAC-seq, a technique for sequencing regions of nuclear chromatin that are accessible to a transposase by sequencing 15 . We identified a total of 59,801, 63,195 and 82,030 highly reproducible accessible chromatin regions in two biological replicates each of the naïve, resolved and sensitized USC lines, respectively. Principal component analysis (PCA) of these USC lines separates sensitized USCs from other groups (Fig. 3a).
bladder TNFα/cyclooxygenase-2 (Cox-2) response, which promotes rapid elimination of infection and mucosal healing 6,7,8 . Sensitized mice are highly susceptible to rUTIs upon challenge, with robust sustained bladder TNFα and Cox-2 expression causing neutrophil transmigration across the bladder epithelium (urothelium) and mucosal wounding that promotes severe recurrent bacterial infections 6,7,8 . Thus, depending on disease history, the bladder tissue is differentially remodelled in a way that either increases or decreases susceptibility to rUTI.
These phenotypes led to our hypothesis that bladder mucosal remodelling is mediated in part by epigenetic changes in urothelial stem cells (USCs) that impact bladder mucosal defence against subsequent infections 6,7 . Thus, we isolated epithelial cells from the bladders of mice with different disease histories, established primary USC lines, propagated them in cell culture, and formed polarized and fully differentiated urothelium in vitro, which displayed morphological phenotypes that resembled the urothelial remodelling phenotypes observed in vivo 6 . We identified differences in chromatin accessibility, DNA methylation and histone modifications in the USC lines, which corresponded with differences in transcriptional responses affecting programmed cell death and cyclooxygenase-2 (Cox-2)-regulated inflammatory responses that impact rUTI susceptibility. Overall, our study provides epigenetic evidence of epithelial-intrinsic trained immunity caused by a mucosal bacterial infection, which alters the bladder mucosal response to subsequent infections, depending on the original disease outcome. This finding may explain the high prevalence of rUTI and have important therapeutic implications for chronic/recurrent bacterial infections in general.

Differentiated urothelial cells form urothelial barriers
To study urothelial-intrinsic changes that result from previous infection, we adapted a method for in vitro propagation of primary intestinal epithelial stem cells in three-dimensional (3D) culture 9,10 to culture primary USCs that were isolated from 8-week-old naïve C3H/HeN mice ( Fig. 1a and Extended Data Fig. 1a). Gene expression of p63, which encodes transformation-related protein 63 (p63), was measured because it is essential for the proliferative capacity of the stem cells 11 , while the expression of Axin2, which is a Wnt-target gene 10 , was measured to assess Wnt signalling activity in the 3D culture. Together, we can evaluate the stem cell pluripotency status by monitoring expression of these genes with RT-qPCR (Extended Data Fig. 1b,c). p63 and Axin2 gene expression remained high through the first 3 d of 3D culture in 50% percent conditioned media (CM), which is required to maintain pluripotency. Expression of p63 and Axin2 then decreased at 5 to 7 d post-infection, regardless of whether the cells were incubated in 50% or 5% CM at day 3 post-infection, showing that either extended culture or lower CM percentage can reduce Wnt signalling. However, uroplakin 3a (Upk3a), a surface protein expressed by differentiated urothelial cells, was only significantly increased in cells grown in 5% CM but not in 50% CM, as determined at 7 d post-infection (Extended Data Fig. 1d). After propagation of USCs in 50% CM over several passages and further culturing in 50% CM over 5 d, all cells stained positive for the epithelial cell marker E-cadherin and the basal urothelial cell marker keratin 5 (K5) but lacked Upk3a expression (Extended Data Fig. 1e). In contrast, culturing in 0% CM in 3D culture for 5 d after initial propagation resulted in the development of epithelial polarity with the formation of a central cavity (Extended Data Fig. 1e). Within the resulting cysts, cavity-facing cells differentiated into superficial facet-like cells (Upk3a+, K5−), whereas perimeter cells in contact with the matrigel matrix were basal cell-like (Upk3a-, K5+).
We next established a transwell culture system to differentiate USCs of juvenile C3H/HeN mice into polarized, stratified urothelial barriers (Fig. 1a). The 5637 human bladder carcinoma cell line (ATCC HTB-9), which has been used widely for the in vitro study of UPEC interactions with bladder cells and originates from basal bladder cancer Article https://doi.org/10.1038/s41564-023-01346-6 Among 2,880 differentially accessible regions (DARs) of chromatin identified between sensitized and resolved USCs, 925 regions are sensitized-accessible DARs (more accessible in sensitized than in resolved) and 1,955 regions are resolved-accessible DARs (more accessible in resolved than in sensitized) (Fig. 3b).
Gene Ontology (GO) pathway analysis on DARs using GREAT 16 revealed that genes associated with sensitized-accessible DARs are strongly enriched for many biological processes, including those involving programmed cell death (PCD), oxidative stress and immune response (Fig. 3c). Many of these DARs were in proximity to PCD pathway associated genes, including Casp1 and Gsdmc2/3 (Supplementary Table 1), which were previously found to be enriched in both whole-bladder RNA-seq and ex vivo urothelial proteomic comparisons of sensitized vs resolved mice 6,8 . Performing motif discovery analysis using HOMER 17 , we found that resolved-accessible DARs are highly enriched for transcription factor (TF) binding motifs of the AP-1 family members, which are known mediators of stress responses, often downstream of cytokine signalling (Extended Data Fig. 3a). In contrast, sensitized-accessible DARs are enriched not only for TF binding motifs of the AP-1 family, but also for those that regulate stem cell fate and tissue differentiation and development, including SOX family members EHF, Klf5 and RUNX2 (Extended Data Fig. 3b).
Chromatin remodelling is accomplished through two main mechanisms: DNA methylation and histone modifications. We performed   genome-wide profiling of DNA methylation and histone modification in our USC lines using WGBS 18 and CUT&RUN 19 , respectively. For CUT&RUN, we selected antibodies against histone modification marks: H3 lysine 4 trimethylation (H3K4Me3), H3 lysine 27 acetylation (H3K27Ac) and H3 lysine 27 trimethylation (H3K27Me3), which are associated with active promoters, active promoter/enhancers and polycomb repression, respectively 20 . Most of the differentially methylated regions (DMRs) were identified in the comparisons of sensitized USCs with either (1) naïve or (2) resolved USCs. We did not observe any global differences in DNA methylation between naïve, resolved and sensitized groups (Extended Data Fig. 4a,b). Sensitized-specific DMRs are displayed as a heat map together with the ATAC-seq and CUT&RUN peaks to visualize the epigenetic landscape across these DMRs (Fig. 3d). We found that sensitized-specific DMRs tend to be hypo-methylated (hypo-DMR) compared with naïve and resolved USCs (Fig. 3d)     transcription factor binding and potentially affecting nucleosome occupancy, together resulting in less accessible chromatin 21 . Concordantly, we found that hypo-DMRs in sensitized USCs have corresponding increased ATAC-seq, H3K4Me3 and H3K27Ac signals (Fig. 3d), suggesting that these regions are enriched in marks for active promoters and enhancers. Looking at the average signals in the genomic regions (+/− 5 kb) surrounding sensitized-specific hypo-DMRs, we saw similar patterns of hypo-methylation, increased chromatin accessibility and increased active histone marks compared with naïve and resolved USCs (Fig. 3e). However, the repressive histone modification, H3K27Me3, was not correlated with sensitized-specific DMRs (Extended Data Fig. 5a,b). The sensitized-specific hypo-DMRs are more associated with genic features such as promoters, exons and introns (Fig. 3f). A PCA plot of all DMRs separated sensitized USCs from naïve and resolved USCs (Fig. 3g). GO pathway analysis of sensitized-specific hypo-DMRs showed enrichment in immune response and cell-cell adhesion-related pathways (Fig. 3h). Thus, USCs maintain epigenetic memories of a previous UPEC infection that differ on the basis of the initial infection outcome and are mediated by differential DNA methylation and active histone modifications.

USC differentiation programmes associate with disease history
We then performed RNA-seq of naïve, resolved and sensitized USCs to investigate whether their altered epigenomes result in differential gene expression. We included juvenile naïve USCs ( Fig. 1) as comparators for age differences. Again, PCA of all DEGs separates sensitized USCs from other groups along principal component (PC) 1 (Fig. 4a). Sensitized USCs had 108 and 73 differentially expressed genes (DEGs) compared with naïve and resolved USCs, respectively (Extended Data Fig. 6a,b), of which 40 genes were common to both comparisons (Fig. 4b). The top 15 DEGs included the glutathione transferase genes Mgst1 and Mgst3 (Fig. 4b). Enriched pathways in sensitized compared to naïve and resolved USCs include those related to protection against reactive oxidative species (ROS), nuclear receptor signalling and stem cell pluripotency (Extended Data Fig. 6c,d). In contrast, no genes were differentially expressed between resolved and naïve USCs. Juvenile naïve USCs separated from the adult naïve USCs mainly along PC2 (Fig. 4a), but this comparison only revealed 8 significant DEGs (Extended Data Fig. 7a). A PCA biplot shows that genes including Znfx1 and Ly6e strongly influenced PC1, while genes including Kank1 and Krt1 strongly influenced PC2 (Extended Data Fig. 7b).
Despite the large number of epigenetic changes found in sensitized USCs, only a few DEGs were observed when USCs were cultured under stem cell-promoting conditions, suggesting that the epigenetic changes we found may have a greater impact on gene expression upon cell differentiation. Therefore, we performed RNA-seq of differentiated urothelia with or without UPEC infection. PCA of all DEGs showed that the transcriptional profiles of mock-infected differentiated urothelia segregated by infection history (Fig. 4c), indicating intrinsic differences in the differentiated urothelia due to previous infection history. Infection largely caused a uniform shift in the PCA plot for each cell line (Fig. 4c), probably reflecting a conserved transcriptional response to UPEC infection in each convalescent state (Extended Data Fig. 8a). Notably, expression of the Cox-2 encoding gene, Ptgs2, was more highly induced upon infection of sensitized and resolved urothelia compared with naïve (Extended Data Fig. 8b), in agreement with a recent in vivo study 7 . Differential gene expression between mock-infected sensitized and resolved differentiated urothelia were visualized in a volcano plot (Fig. 4d), and pathway analysis of DEGs was performed (Fig. 4e). Sensitized differentiated urothelia displayed activation of PTEN signalling, PPARɑ/RXRɑ and Glutathione-mediated detoxification pathways relative to resolved differentiated urothelia, independent of UPEC infection ( Fig. 4e and Extended Data Fig. 8c,d). Several of the DEGs were TF genes, including Klf4 and Klf5, (Extended Data Fig. 9), suggesting these may play a role in driving and maintaining unique epigenetic changes in sensitized urothelial cells compared with the other cell types.

Sensitized USC reprogramming reveals trained immunity signal
Based on our GO pathway analysis data implicating PCD pathways in sensitized-accessible DARs (Fig. 3c) and our in vivo observations that the remodelled sensitized urothelium is characterized by severe exfoliation and Cox-2 inflammation-dependent mucosal wounding during UPEC infection 6 , we specifically interrogated DEGs involved with PCD pathways. A heat map of gene expression shows that many genes associated with PCD pathways were differentially expressed in sensitized and resolved differentiated urothelia relative to each other and to adult naïve differentiated urothelia (Fig. 4f). While Casp1 (which was the most highly upregulated gene when comparing mock-infected sensitized urothelia to resolved (Supplementary Table 2)) and other pyroptosis-related genes (including Aim2, Gsdmc2 and Gsdmc3) were upregulated in sensitized differentiated urothelia, other pyroptosis-related genes (such as the Naips) as well as apoptosis and necroptosis-related genes were upregulated in resolved differentiated urothelia, suggesting that resolved and sensitized cells are predisposed to different PCD mechanisms during UPEC infection.
We next examined whether DNA methylation differences in the USCs correlated with RNA-seq fold changes in the differentiated urothelia between sensitized and resolved USCs, focusing on DMRs at promoter sites (Fig. 5a). Most genes, including Casp1 and Ptgs2os2 (a positive regulator of Cox-2 expression in cis and pro-inflammatory response regulator in trans), showed a negative correlation between relative gene expression and the level of DNA methylation at that gene's promoter site (Fig. 5a). Using the WashU Epigenome Browser, chromatin accessibility, DNA methylation and histone modification marks were visualized at the Casp1 and Ptgs2os2 loci (Fig. 5b,c). The Casp1 and Ptgs2os2 promoter loci of sensitized USCs are each relatively hypo-methylated while also being enriched in the active histone marks, H3K4Me3 and H3K27Ac, relative to naïve and resolved USCs (Fig. 5b,c and Extended Data Fig. 10a). Several TFs that were enriched in sensitized-accessible DARs (Extended Data Fig. 3a), such as members of the Sox and AP-1 families (Klf5, ETS and RUNX2), have predicted binding sites near the Casp1 promoter as indicated on the Epigenome Browser map (Extended Data Fig. 10b). Casp1 expression by RT-qPCR was approximately 1,000-fold higher in sensitized differentiated urothelia compared with resolved or naïve USCs, independent of UPEC infection (Fig. 5d). Concordantly, immunoblot staining showed detectable caspase-1 only in sensitized differentiated urothelia (Fig. 5e), in agreement with our previous ex vivo proteomics of convalescent sensitized mouse urothelium 8 .
In our previous studies, we found that the secreted pore-forming bacterial toxin α-hemolysin (HlyA), commonly produced by UPEC, induces caspase-1 and caspase-11(caspase-4 in humans)-dependent pyroptosis in human and mouse urothelial cells-a protective response that leads to exfoliation of infected cells 22 . We hypothesized that enhanced caspase-1 expression in the sensitized differentiated urothelium would lead to a more robust pyroptotic cell death response upon wild type (WT) (HlyA+) UPEC infection in vitro. A lactate dehydrogenase (LDH) cytotoxicity assay demonstrated that UPEC infection induced cell death in naïve, resolved and sensitized differentiated urothelia (Fig. 5f), but cell death was significantly greater in sensitized differentiated urothelia. In challenge infections using WT UTI89 and UTI89ΔhlyA strains in naïve, resolved and sensitized mice, we observed that ΔhlyA infection, which does not activate caspase-1-mediated pyroptotic cell death, showed significantly increased bacterial burdens in sensitized mice compared with WT mice, whereas no differences were observed in naïve and resolved mice (Fig. 5g). Furthermore, the incidence of recurrent chronic cystitis at 28 d post-infection was significantly increased in sensitized mice when infected with ΔhlyA compared with WT (Fig. 5h), indicating that caspase-1 overexpression

Fig. 5 | Increased caspase-1-mediated inflammatory cell death in sensitized USCs may protect sensitized mice from acute and chronic UPEC infection. a, For those
DMRs found within 1 kb of promoter regions, RNA-seq fold changes comparing sensitized vs resolved differentiated urothelia either with (y axis, dark purple dots) or without infection (y axis, grey dots) were plotted against the DNA methylation differences (x axis) between sensitized and resolved USCs. b,c, Differences in chromatin accessibility (ATAC-seq), DNA methylation (WGBS) and active histone modifications (H3K4me3 and H3K27ac) at the Casp1 (b) and Ptgs2os2 (c) loci in different USC lines were visualized as combined tracks using the WashU Epigenome Browser map. In WGBS data, the average % methylation at the Casp1 and Ptgs2os2 promoter sites (red box) are indicated; colour bars represent % methylation, grey backgrounds represent CpGs, and black lines indicate sequencing depth. CpGs within the Casp1 and Ptgs2os2 promoter regions have 8-25x and 18-23x read coverage, respectively. d, Gene expression of Casp1 in differentiated urothelia was measured by RT-qPCR (data from n = 4, 4, 4, 5, 4, 4 samples generated from adult naïve, resolved and sensitized differentiated urothelia that are then infected with either PBS or UTI89, respectively, are represented as mean ± s.d.). e, Protein expression of caspase-1 using two different cell lines (N2, R2, S2 and N3, R3, S3) was assessed by western blot, and N3, R3 and S3 are represented. f, Cell death of differentiated urothelia 4 h after UTI89 infection was measured by LDH assay. Data are mean ± s.d., obtained from n = 7, 10, 14, 8, 10, 4, 6 samples from generated from adult naïve, resolved and sensitized differentiated urothelia, 2-3 biologically independent cell lines, per condition, that are then infected with either PBS or UTI89, respectively; significance was determined with a one-way analysis of variance (ANOVA). g,h, Naïve, resolved and sensitized mice were challenged with 10 7 c.f.u. of WT UTI89 (HlyA+) or UTI89ΔhlyA. Data are combined from 2-3 independent experiments. g, Bladder bacterial burdens at 6 hpi (n = 10, 7, 19, 8, 14, 11 adult naïve, resolved and sensitized mice challenged with 10 7 c.f.u. of either UTI89 or UTI89ΔhlyA, respectively) examined over 2-3 independent experiments. Bars indicate median values and two-tailed Mann-Whitney U test was used to determine significance. h, Incidence of chronic cystitis at 28 dpi. Two-sided Fisher's exact test; P values are indicated when significant. Article https://doi.org/10.1038/s41564-023-01346-6 in sensitized urothelial cells is a protective response that helps to resolve challenge UPEC infection.

Discussion
Previous studies have shown that long-lasting bladder tissue remodelling occurs in response to UPEC infection, and this remodelling is accompanied by changes in susceptibility to subsequent infection, depending on previous infection outcomes [5][6][7] . We hypothesized that this altered susceptibility is mediated, at least in part, by the development of trained immunity at the bladder epithelial mucosa. In contrast with adaptive immunity, which encompasses antigen-specific responses by T and B lymphocytes, 'trained immunity' is characterized by antigen non-specific tissue adaptation to acute and chronic inflammation, sometimes in response to infection, and has been predominantly studied in professional innate immune cells such as macrophages, monocytes, dendritic cells and natural killer cells [23][24][25] .
Here we used a primary epithelial cell culture system 10 to elucidate the urothelial-intrinsic contribution to bladder mucosal remodelling as a consequence of a previous infection, discovering evidence for epigenetic reprogramming of the urothelial stem cells as a mechanism of trained immunity to subsequent urinary tract infection.
The bladder urothelium of previously infected mice is known to be resistant to intracellular colonization relative to age-matched naïve mice 6,7 . However, the mechanism for this intracellular colonization resistance differs between resolved and sensitized mice. In resolved mice, UPEC initially form intracellular bacterial communities in the superficial facet cells, similar to that in adult naïve mice, but they are rapidly shed within the first 6 h of infection via enhanced TNFα-mediated inflammation 7 . In contrast, intracellular bacterial communities do not form at all in sensitized urothelium in vivo, probably due to the small cell size and actin-gating of the incompletely differentiated superficial cells 6,26 . In this work, we have elucidated another facet of bladder colonization resistance in sensitized urothelium, where HlyA-mediated urothelial cell death and exfoliation further reduces early bladder colonization. This effect is probably a consequence of the increased baseline expression of caspase-1 and perhaps other inflammasome-associated factors, such as Gasdermins C2 and C3 that can act as terminal effectors of inflammatory cell lysis 27 , which were observed here in vitro and were previously described in ex vivo proteomics studies of mouse urothelia 8 . Aim2, which encodes a cytosolic innate immune sensor that can activate the caspase-1 inflammasome, was also more highly expressed in sensitized differentiated urothelia. In the skin of mice, imiquimod and the resulting inflammation enables the skin to have a more rapid Aim2-mediated response to a secondary inflammatory insult 28 , suggesting that epigenetic reprogramming of inflammasome components may be a common mechanism for priming inflammation sensors to prepare for secondary exposure at barrier tissue sites. Furthermore, in mice, maternal interleukin-6 (IL-6) produced in response to infection can induce epigenetic changes in fetal intestinal epithelial stem cells in utero 29 . In contrast to these previous studies, our work demonstrates a direct role of a mucosal bacterial infection in eliciting specific epigenetic changes to mucosal epithelial stem cells that alter the outcome of subsequent infections.
The protective caspase-1-mediated trained immunity in the sensitized bladder is often overcome by Cox-2-dependent inflammation that occurs in response to high bacterial burdens during the first 24 h of acute rUTI 6,8 . Cox-2 expression occurs mainly at the basal urothelial cell level, but its activity can elicit mucosal wounding through excessive recruitment of neutrophils, thereby transforming the colonization landscape in favour of extracellular colonization and growth of the bacteria. Thus, sensitized mice have competing protective and sensitizing responses to challenge infection, which typically manifest as an extreme bimodal distribution of infection burdens by 24 h post-infection 5 . Although both resolved and sensitized bladders have enhanced early Cox-2 responses in vivo, by 24 h post-infection, the bladder inflammatory response is only sustained in sensitized mice 7 and Cox-2 inhibition protects sensitized mice against severe recurrent cystitis 6,8 . Similarly, infection-induced Ptgs2 gene (encoding Cox-2) expression was enhanced in both sensitized and resolved differentiated urothelia relative to naïve. However, the expression of the Ptgs2os2 gene, which encodes LincRNA-Cox-2, a positive regulator of Ptgs2 expression and general inflammation 30 , differed between these cell lines, being increased in sensitized relative to resolved differentiated urothelia. Concordantly, we found differential epigenomic marks that were associated with increased accessibility of the Ptgs2os2 locus in the sensitized USCs, suggesting that Ptgs2os2 expression in differentiated urothelia is altered by epigenetic changes in the USCs. Thus, epigenetic changes to the Ptgs2os2 locus may play a role in promoting the sustained pro-inflammatory responses of the sensitized bladder, thus overcoming the protective function of increased Casp1 expression.
Changes in expression of DNA methyltransferases, which methylate CpG sites of DNA, have been implicated as a mechanism for remodelling the epigenome in response to acute UPEC infection 31 , potentially explaining how a previous infection, whether self-resolving or chronic, could alter the USC epigenome. However, chronic inflammation itself is also a potent inducer of epigenetic memory that could explain the differences in epigenetic marking between resolved and sensitized USCs. One model for this rewriting of the epigenome is the presence of so-called 'memory domains' where 'pioneer' TF binding to nucleosomes in response to stimuli opens up chromatin in stress-responsive loci and allows epigenetic writers to remodel the chromatin to allow it to remain open after the stimuli are removed 32 . Klf4, which was upregulated in sensitized differentiated urothelia compared with resolved urothelia, is a known pioneer TF 33 . Further, our motif discovery analyses support the hypothesis that epigenetic remodelling in the USCs occurs primarily at AP-1-associated DARs in response to a self-limiting acute infection (resolved-accessible DARs), but that severe acute infection leading to chronic infection and inflammation (sensitized-accessible DARs) induces epigenetic remodelling not only at AP-1-associated DARs, but also at additional TF-associated DARs, such as those with Klf and Sox family motif sites.
Our discovery of epithelial stem cell epigenetic reprogramming upon UPEC infection has implications for understanding the mechanism of epithelial-intrinsic trained immunity against not only UTIs, but also other types of infection or inflammatory disease. Further mechanistic studies may lead to novel therapies for a range of recurrent infections and inflammatory diseases. For example, therapeutic use of an inhibitor of histone demethylase LSD1, which is overexpressed in skin epithelial cancer, drives notable increases in H3K4 methylation in the cells, thus leading to both premature epidermal differentiation and the repression of squamous cell carcinoma 34 . Therefore, further investigation to identify which epigenetic factors, TFs or inflammatory mediators are directly responsible for establishing and maintaining these specific epigenetic memories in vivo would provide deeper mechanistic insights and shed light on potential therapeutic targets to prevent rUTIs and/or reverse the epigenetic imprinting that leads to increased susceptibility to recurrent disease.

Ethics statement
All animal experimentation was conducted according to the National Institutes of Health guidelines for the housing and care of laboratory animals. All experiments were performed in accordance with institutional regulations after review and approval by the Animal Studies Committee at Washington University School of Medicine in St Louis, Missouri.

Mouse infections
Female C3H/HeN mice (Envigo) were 7-8 weeks old ('juvenile') at the time of the initial infection. A total of 10 8 c.f.u. of UTI89 were inoculated into the bladder of C3H/HeN mice by transurethral catheterization 5,36 . C3H/HeN mice develop chronic cystitis in an infection dose-dependent manner and this inoculum results in chronic cystitis in ~50% of mice 6 . To monitor infection outcomes, urine was collected. Persistent bacteriuria (10 4 c.f.u. ml −1 ) is defined as a specific and sensitive cut-off for detecting chronic cystitis 5 . Chronic cystitis during initial infection was defined as persistent high bacteriuria (>10 4 c.f.u. ml −1 urine) at every timepoint urine was collected (1, 3, 7, 10, 14, 21 and 28 d post-infection), while resolution of cystitis was defined as urine bacterial titre dropping below this cut-off in at least one timepoint.
At 4 weeks post-infection, all mice were treated with trimethoprim and sulfamethoxazole in the drinking water for 10 d (54 and 270 μg ml −1 water, respectively) 6 . Urine was collected weekly to confirm clearance of bacteriuria. Four weeks after the initiation of antibiotics, naïve, resolved and sensitized mice were used to isolate primary USCs or used for secondary infection assay. For the secondary infection, mice were challenged with 10 7 c.f.u. of bacteria inoculated into the bladders, then humanely euthanized at 6 h post-infection, and bacterial burdens were determined to assess acute outcomes.

Cell line culture
Human bladder carcinoma epithelial cells, designated 5637 (ATCC HTB-9) cells, were cultured in RPMI-1640 medium containing 10% FBS at 37 °C in the presence of 5% CO 2 .

Primary USC isolation and culture
Bladder tissue from juvenile, convalescent naïve, resolved and sensitized mice were isolated, bisected and incubated in stripping solution at 4 °C overnight. The urothelial cells were scraped off from the bladder tissue, spun down at 4 °C at 300 g for 5 min, resuspended in fresh collagenase IV solution and incubated with rocking at 37 °C for 20 min. The cells were disaggregated by gentle pipetting, filtered with a 100 μm strainer, then washed with washing media. The cells were cultured in matrigel (BD Biosciences) with 50% L-WRN CM containing 10 mM Y-27632 and 10 mM SB431542 (R&D System) 9 . Media were changed every 2 d and cells were passaged every 3 d (1:2-3 split). USCs were used for experiments after 10 passages to remove any remaining non-stem urothelial cells.

Differentiated urothelium culture on transwell
USCs were washed in PBS with 0.5 mM EDTA, trypsinized in 0.05% Trypsin and 0.5 mM EDTA for 1 min at 37 °C, dissociated by vigorous pipetting, filtered through a 40 μm cell strainer and resuspended in washing media. Transwells (Corning Costar, 3413) were coated in PBS with 1:40 Matrigel for 30 min at 37 °C. Then 3-4 × 10 4 USCs were seeded on the transwell insert, and 100 μl and 600 μl 50% CM containing 10 mM Y-27632 were added to the apical and basolateral compartments of the transwell, respectively.

TER measurements
Resistance of the urothelial multilayers was assessed by TER measurement using an epithelial volt-ohm metre (World Precision Instruments). The average value of triplicate measurements was multiplied by the area of the transwell membrane (0.33 cm 2 ) to obtain a final value in ohm × cm 2 (ref. 37).

In vitro UPEC infection assay
When urothelium was fully differentiated (TER value >4,000 ohm × cm 2 ), cultures were washed 3 times in warm DMEM/F12 media and infected with UPEC strains at multiplicity of infection 10. Transwells were then incubated at 37 °C for 30 min, changed to media containing 100 μg ml −1 gentamicin to clear the extracellular bacteria and cultured for an extended time. After infection, apical and basolateral media were spun down at 2,000 g at 4 °C for 5 min and used for LDH assay (TaKaRa, MK401). Transwells were washed with sterile PBS, then used for various analyses.

Scanning electron microscopy (SEM)
Differentiated urothelia were washed 3 times in PBS, fixed in EM fixative (2% paraformaldehyde, 2.5% glutaraldehyde in 1× PBS) for 1 h on ice and washed 3 times in PBS. Samples were then post-fixed in 1.0% osmium tetroxide, dehydrated in increasing concentrations of ethanol, then dehydrated at 31.1 °C and 1,072 p.s.i. for 16 min in a critical point dryer 6 . Samples were mounted on carbon tape-coated stubs and sputter-coated with gold/palladium under argon 6 , then imaged on a Zeiss Crossbeam 540 FIB-SEM.

RNA isolation and RT-qPCR
RNAs were extracted from USCs or differentiated urothelia using RNAeasy Plus mini kit (Qiagen) and reverse-transcribed with iScript Reverse Transcription Supermix (BioRad). We used 1 μl 12.5 ng μl −1 ccomplementary DNA with intron-spanning primers specific to each gene, and iQ SYBR Green Supermix was used according to the manufacturer's instructions (BioRad). Sequences of the primers we used in this study are listed in Supplementary Table 4. Expression values were normalized to 18S, and relative expression compared to control was determined by the cycle threshold (ΔΔCt) method 38 . Each sample was run in triplicate, and average Ct values were calculated. Article https://doi.org/10.1038/s41564-023-01346-6

RNA-seq and data analysis
Illumina cDNA libraries were generated using a modified version of the RNAtag-seq protocol 39 . Briefly, 1 μg of total RNA was fragmented, depleted of genomic DNA, dephosphorylated and ligated to DNA adaptors carrying 5'-AN 8 -3' barcodes of known sequence with a 5' phosphate and a 3' blocking group. Barcoded RNAs were pooled and depleted of ribosomal RNA using the Ribo-Zero rRNA depletion kit (Illumina). cDNA libraries were generated by adding a second adaptor by template switching and PCR amplification with primers carrying Illumina P5 or P7 sequences, then the libraries were sequenced on the Illumina HiSeq 2500. Paired-end sequencing reads in a pool were demultiplexed on the basis of their associated barcode sequence using custom scripts (https://github.com/broadinstitute/split_merge_pl). Reads were then trimmed using cutadapt v1.6 and trimmed reads were aligned to the Mus musculus mm10 genome using tophat2 v2.0.11 and bowtie2 v2.2.2. Gene counts were conducted by HTSeq v0.6.0 and read counts were assigned to annotated transcripts using Salmon v0.8.2 7 .
Read normalization and differential expression were conducted with DESeq2 v1.14.0 40 . rlog transformations of DESeq-normalized reads were used for PCA plots. Fragments per kilobase of transcript per million mapped reads (FPKM) normalization of DEseq2 reads was used for z-score heat maps. TF expression was determined using DESeq2 FPKM-normalized values and a list of mouse TFs (n = 453) from HOCOMOCO v11 41 , a TF database of validated TF motifs. An adjusted P value cut-off of 0.05 was used and TF candidate expression was visualized using z-score heat maps. Statistically significant differences in gene expression were assessed by the Wald test, followed by multiple test correction using Benjamini-Hochberg false discovery rate (FDR), with adjusted P < 0.05 being considered significant. Pathway analyses were performed with ingenuity pathway analysis (IPA). Significance was determined by a right-tailed Fisher's exact test, with P adj < 0.05 being considered significantly enriched pathways.

ATAC-seq and data analysis
Single cells (1-2 × 10 5 ) of naïve, resolved and sensitized USCs were used for nuclei preparation, and 50,000 nuclei were counted and transferred into 25 μl of 2× TD buffer. Omni-ATAC-seq reaction mix (25 μl) including TDE1 enzyme was added to 25 μl of 50,000 nuclei in 2× TD buffer, then the samples were incubated at 37 °C for 30 min (tapped every 10 min during the incubation in a heat block). Transposed DNA fragments were immediately purified using a MinElute PCR purification kit (Qiagen). ATAC-seq libraries were amplified by PCR amplification (10-12 cycles) with an initial 5 min extension at 72 °C and purified using AMPure XP beads (Beckman Coulter). The purified libraries were eluted with 20 μl of nuclease-free water, quantified using Qubit dsDNA HS assay kit (ThermoFisher), and their size distribution checked with a 4200 TapeStation (High Sensitivity D1000 ScreenTape and Reagents). Paired-end ATAC-seq libraries were sequenced on an Illumina NextSeq 500 (~350 million reads).
To identify DARs, Diffbind v2. 10.0 was used on IDR < 0.05 ATAC peaks, and Benjamini-Hochberg FDR with a cut-off <0.05 was used for statistical significance. Significant DARs (FDR < 0.05) were used for generating volcano plots and heat maps. GREAT 16 (basal plus extension parameter) was used for GO pathway analysis. GREAT ranks results by binomial P value using a binomial test. Sensitized (FC > 1.5) and resolved-specific DARs (FC < −1.5) were separately analysed and the top 15 enriched pathways are shown in Fig. 3e-f.

WGBS and data analysis
Single cells (1-2 × 10 5 ) of USCs were treated with DNase I to remove trace DNA contamination from the Matrigel. Genomic DNA (gDNA) was prepared from the cells using DNeasy Blood & Tissue kit (Qiagen, 69504). Using 200 ng of gDNA and 0.4 ng lambda, DNA was bisulfite treated using EZ DNA Methylation-Direct kit (Zymo, D5020) and processed with xGen Methyl-Seq Library Prep kit (IDT, 10009824) to generate Illumina-compatible WGBS libraries. The libraries were sequenced on a NovaSeq S4 300XP (~300 million reads) by MGI institute.
WGBS analysis commands with specific parameters are detailed in the Code availability section. Briefly, fastqQC v0.11.8 was used to assess the quality of the raw reads. Subsequently, the paired-end reads were trimmed to remove adaptor sequences and low-quality reads with Cutadapt v1.18 and reassessed using FastqQC. The mouse reference genome mm10 was first bisulfite converted using Bismark v0.20.0. The paired-end reads were aligned to the mm10 bisulfite-converted genome and deduplicated using 'deduplicate_bismark'. DNA methylation levels were calculated using 'bismark_methylation_extractor' and displayed in a methylC format on the WashU Epigenome Browser 46 . Bisulfite conversion was estimated using the conversion rate of cytosine to thymine in the lambda reference genome.
DMRs were identified with DSS v2.43.2 47 using a two-group comparison for biological replicates and called using 'DMLtest' and 'call-DMR'. A PCA plot of CpG methylation within DMRs was generated using Deeptools v3.3.0. Biological replicates were combined by merging fastq files between replicates and reprocessing using the steps previously described. CpG density was visualized using a 5x coverage cut-off and ggplot2 v3.3.6.
Sensitized-specific DMRs were defined as the overlapping regions between naïve vs sensitized and resolved vs sensitized DMRs. The percent methylation for sensitized-specific DMRs was visualized using the R package 'ComplexHeatmap'. The DNA methylation over sensitized-specific hypo-DMRs were plotted using Deeptools and visualized using ggplot2. Overlapping regions between DMRs were identified and visualized using Intervene 48 'Venn' with default parameters. GREAT 16 analysis on sensitized hypo-DMRs was performed as described in ATAC analysis. The sensitized hypo-DMRs were also analysed for genomic annotation using UCSC (https://genome.ucsc.edu/cgi-bin/hgTables) to download GENCODE M25 (https://www.gencodegenes.org/mouse/ release_M25.html). The promoter was defined as 1 kb upstream of transcription start site. Genomic annotation priority was assigned in the following order: promoter, coding exon, 5' UTR, 3' UTR, intron and intergenic. DMRs were assigned to annotation if the DMR overlapped 20% of the annotation using BEDTools v2.27.1 intersect and were plotted using DNA methylation percent change between sensitized and resolved against the log 2 (FC) of associated genes between sensitized and resolved differentiated urothelia with or without infection.

CUT&RUN and data analysis
Single cells (0.2 × 10 6 ) of USCs were slightly crosslinked in 0.1% formaldehyde and fixed cell pellets were stored at −80 °C before use. H3K4Me3, H3K27Ac and H3K27Me3 CUT&RUN was performed using CUT&RUN assay kit (Cell Signaling, 86652), with a few modifications. Briefly, cells were attached to concanavalin A beads for each experiment. Cells were permeabilized with digitonin in the antibody binding buffer containing spermidine and protease inhibitors and then incubated with primary antibodies against H3K4Me3 (Cell Signaling, 9751, 1:50), H3K27Ac (Cell Signaling, 8173, 1:100) or H3K27Me3 (Cell Signaling, 9733, 1:50) at 4 °C overnight on a rotator. The beads-cells Article https://doi.org/10.1038/s41564-023-01346-6 mixture was washed 3 times with digitonin buffer, resuspended in 50 μl pAG-MNase and incubated at 4 °C for 1 h on a rotator. Samples were digested in PCR tubes containing 150 μl cold digitonin buffer and 3 μl CaCl 2 at 4 °C for 30 min in a thermal cycler. Then, beads were transferred back to the microcentrifuge tubes, 150 μl of 1× STOP buffer was added and tubes were incubated at 37 °C for 10 min. Placing tubes on a magnetic rack, supernatants were collected to a new tube. Crosslinks were reversed by adding 3 μl 10% SDS solution and 2 μl 20 mg ml −1 proteinase K, then samples were incubated at 65 °C for 2 h. DNA from enriched chromatin samples were purified using DNA spin columns (Zymo, D4013). The sequencing library was prepared with Ultra II DNA Library Prep kit (NEB, E7645) following the manufacturer's instructions, but reducing the anneal and extension time to 10 s during PCR enrichment of adaptor-ligated DNA.
A detailed list of commands and parameters can be found under Code availability. Briefly, fastqQC v0.11.9 was used to assess the read quality. Subsequently, the paired-end reads were trimmed with Cutadapt v1.9 and reassessed using fastqQC. Reads were then aligned using bowtie2 v2.3.4.1 49 . Mitochondrial reads were removed using samtools v1.9 and deduplicated using Picard v2.8.1 MarkDuplicates. Uniquely mapped reads were extracted using samtools view. Peaks were called using MACS2 v2.1.1.20160309 'callpeak': '-q 0.01' for narrow peaks H3K4Me3 and H3K27Ac, and '-q 0.05 --broad' for H3K27Me3. Encode-defined blacklisted regions were removed. For each histone modification, a consensus peak list was used to calculate the fraction of reads in peaks (FRIP). Reads then were converted to bigWig format using Deeptools and normalized using read coverage and FRIP score. The normalized biological replicates were combined using ucsc-bigwigmerge v377 and converted from bedGraph to bigWigs using kentUCS v334 and mm10 chromosome sizes from UCSC (http://hgdownload.cse. ucsc.edu/goldenPath/mm10/bigZips/mm10.chrom.sizes). To profile sensitized-specific DMR regions for other epigenetic modifications, the sensitized hypo-DMRs were overlapped with MASC2 narrow peaks for ATAC, H3K4Me3 and H3K27Ac, and broad peaks for H3K27Me3. The corresponding peak score was normalized using the FRIP scores and plotted using the R package 'ComplexHeatmap'. Using the normalized bigWig tracks, ATAC, H3K4Me3, H3K27Ac and H3K27Me3 signals were plotted over the sensitized-specific hypo-DMRs using Deeptools. The normalized bigWig signals were used for the casp1 heat map.

Statistics and reproducibility
For representing images of confocal, immunofluorescence staining and SEM, 3-4 different cell lines (among J1-5, N1-4, R1-4 and S1-4) were stained and imaged. For western blot images, two different cell lines were tested. Statistics for plots/graphs were analysed in GraphPad Prism v8.4.3. Exact P values are indicated when significant (P < 0.05) (GraphPad Prism did not provide exact P value when P < 0.0001).

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.