Animal Symposia and Workshops A-1 Bacterial Colonization Stimulates a Complex Physiological Response in the Immature Human Intestinal Epithelium

Bacterial Colonization Stimulates a Complex Physiological Response in the Immature Human Intestinal Epithelium. DAVID R. HILL, Sha Huang, Courtney Lynn, Disharee Mukherjee, Brooke Bons, Shrikar Thodla, Priya H. Dedia, Alana M. Chin, Yu-Hwai Tsai, Melinda S. Nagy, Thomas Schmidt,Seth Walk, Vincent B. Young, and Jason R. Spence. Department of Internal Medicine, Division of Gastroenterology, University of Michigan, Ann Arbor MI 48109; Department of Internal Medicine, Division of Infectious Disease, University of Michigan, Ann Arbor MI 48109; Department of Microbiology and Immunology, University of Michigan, Ann Arbor MI 48109; Department of Surgery, University of Michigan, Ann Arbor MI 48109; Depar tment of Cel l and Developmental Biology, University of Michigan, Ann Arbor MI 48109; and Department of Microbiology and Immunology, Montana State University, Bozeman, MT 59717. Email: hilldr@med.umich.edu


Introduction
The epithelium of the gastrointestinal (GI) tract represents a large surface area for host-microbe interaction and mediates the balance between tolerance of mutualistic organisms and the exclusion of potential pathogens (Peterson and Artis, 2014). This is accomplished, in part, through the formation of a tight physical epithelial barrier, in addition to epithelial secretion of antimicrobial peptides and mucus (Veereman-Wauters, 1996;Renz et al., 2011). Development and maturation of the epithelial barrier coincides with the first exposure of the GI tract to microorganisms and the establishment of a microbial community within the gut Koenig et al., 2011). Although microorganisms have long been appreciated as the primary drivers of the postnatal expansion of adaptive immunity Shaw et al., 2010;Hviid et al., 2011;Abrahamsson et al., 2014;Arrieta et al., 2015), and more recently as key stimuli in the development of digestion (Erkosar et al., 2015), metabolism (Cho et al., 2012), and neurocognitive function (Diaz Heijtz et al., 2011;Clarke et al., 2014;Borre et al., 2014;Desbonnet et al., 2014), it remains unclear how the human epithelial surface adapts to colonization and expansion of microorganisms within the immature GI tract.
Studies in gnotobiotic mice have improved our understanding of the importance of microbes in normal gut function since these mice exhibit profound developmental defects in the intestine (Round and Mazmanian, 2009;Gensollen et al., 2016;Bry et al., 1996;Hooper et al., 1999) including decreased epithelial turnover, impaired formation of microvilli (Abrams et al., 1963), and altered mucus glycosylation at the epithelial surface (Bry et al., 1996;Goto et al., 2014;Cash et al., 2006). However, evidence also suggests that the immature human intestine may differ significantly from the murine intestine, especially in the context of disease . For example, premature infants can develop necrotizing enterocolitis (NEC), an inflammatory disease with unknown causes. Recent reports suggest a multifactorial etiology by which immature intestinal barrier function predisposes the preterm infant to intestinal injury and inflammation following postpartum microbial colonization (Neu and Walker, 2011;Morrow et al., 2013;Greenwood et al., 2014;Hackam et al., 2013;Afrazi et al., 2014;Fusunyan et al., 2001;Nanthakumar et al., 2011). Rodent models of NEC have proven to be inadequate surrogates for studying human disease . Therefore, direct studies of host-microbial interactions in the immature human intestine will be important to understand the complex interactions during bacterial colonization that lead to a normal gut development or disease. Important ethical and practical considerations have limited research on the immature human intestine. For example, neonatal surgical specimens are often severely damaged by disease and not conducive for ex vivo studies. We and others have previously demonstrated that human pluripotent stem-cell-derived human intestinal organoids (HIOs) closely resemble immature intestinal tissue eLife digest Human newborns are exposed to large numbers of bacteria at birth. They must transition from the protective, sterile environment of the womb into the bacteria-rich world. The gut, in particular, must adapt as bacteria colonize it. Many of the first bacteria found in the newborn gut form the basis of the bacterial communities needed for a healthy intestine throughout life. In some premature infants, bacterial colonization of the intestine may trigger harmful inflammation and a serious illness called necrotizing enterocolitis.
It is not known exactly how the immature intestine first responds to bacteria. It is also unclear what goes wrong that causes illness in some premature infants. Learning more about how a healthy newborn intestine becomes colonized and responds to this colonization may help medical professionals to better care for normal infants and those who are at risk of intestinal disease. But it has been difficult to study because there is not much newborn intestinal tissue available for scientific research. One solution would be to grow tissue in the laboratory that is like tissue found in the newborn intestine and see how it adapts to bacteria. Now, Hill et al. use stem cells to grow a tissue in the laboratory that is very like immature newborn intestine and show that bacterial colonization helps it to mature. In the experiments, stem cells were grown into an intestine-like tissue. Analyses showed that this laboratory-grown tissue had the same patterns of gene expression as newborn intestines. Then, a type of bacteria called Escherichia coli that is normally found in the intestines of healthy babies was introduced to the intestine-like tissue. Hill et al. show that the initial contact with bacteria and changes in oxygen levels due to bacterial activity cause shifts in gene expression. These in turn stimulate the release of mucus and other protective responses.
A protein called NF-kB plays a central role in these normal bacteria-intestine interactions. Hill et al. show that using a drug to block NF-kB interferes with these processes. The experiments show that contact with bacteria encourages the immature intestine to protect itself from potential harm. More experiments like these may help scientists understand normal bacteria-intestine interactions in early life and how they may go wrong in disease. These studies might also help identify new treatments for babies with necrotizing enterocolitis. Finkbeiner et al., 2015;Watson et al., 2014;Forster et al., 2014;Dedhia et al., 2016;Aurora and Spence, 2016;Chin et al., 2017) and recent work has established gastrointestinal organoids as a powerful model of microbial pathogenesis at the mucosal interface (Leslie et al., 2015;McCracken et al., 2014;Forbester et al., 2015;Hill and Spence, 2017).
In the current work, we used HIOs as a model immature intestinal epithelium and a humanderived non-pathogenic strain of E. coli as a model intestinal colonizer to examine how host-microbe interactions affected intestinal maturation and function. Although the composition of the neonatal intestinal microbiome varies between individuals, organisms within the genera Escherichia are dominant early colonizers Bäckhed et al., 2015) and non-pathogenic E. coli are widely prevalent and highly abundant components of the neonatal stool microbiome Koenig et al., 2011;Bäckhed et al., 2015;Morrow et al., 2013). Microinjection of E. coli into the lumen of three-dimensional HIOs resulted in stable bacterial colonization in vitro, and using RNA-sequencing, we monitored the global transcriptional changes in response to colonization. We observed widespread, time-dependent transcriptional responses that are the result of both bacterial contact and luminal hypoxia resulting from bacterial colonization in the HIO. Bacterial association with the immature epithelium increased antimicrobial defenses and resulted in enhanced epithelial barrier function and integrity. We observed that NF-kB is a central downstream mediator of the transcriptional changes induced by both bacterial contact and hypoxia. We further probed the bacterial contact and hypoxia-dependent epithelial responses using experimental hypoxia and pharmacological NF-kB inhibition, which allowed us to delineate which of the transcriptional and functional responses of the immature epithelium were oxygen and/or NF-kB dependent. We found that NF-k B-dependent microbe-epithelial interactions were beneficial by enhancing barrier function and protecting the epithelium from damage by inflammatory cytokines. Collectively, these studies shed light on how microbial contact with the immature human intestinal epithelium can lead to modified function.

Pluripotent stem-cell-derived intestinal epithelium transcriptionally resembles the immature human intestinal epithelium
Previous work has demonstrated that stem-cell-derived human intestinal organoids resemble immature human duodenum Finkbeiner et al., 2015;Tsai et al., 2017). Moreover, transplantation into immunocompromised mice results in HIO maturation to an adult-like state Finkbeiner et al., 2015). These analyses compared HIOs consisting of epithelium and mesenchyme to whole-thickness human intestinal tissue, which also possessed cellular constituents lacking in HIOs such as neurons, blood vessels and immune cells . Thus, the extent to which the HIO epithelium resembles immature/fetal intestinal epithelium remained unclear. To address this gap and further characterize the HIO epithelium relative to fetal and adult duodenal epithelium, we isolated and cultured epithelium from HIOs grown entirely in vitro, from fetal duodenum, adult duodenum, or HIOs that had been transplanted into the kidney capsule of NSG immuno-deficient mice and matured for 10 weeks. These epithelium-only derived organoids were expanded in vitro in uniform tissue culture conditions for 4-5 passages and processed for RNA-sequencing (RNA-seq) (Figure 1-figure supplement 1). Comparison of global transcriptomes between all samples in addition to human embryonic stem cells (hESCs) used to generate HIOs ; E-MTAB-3158) revealed a clear hierarchy in which both in vitro grown HIO epithelium (p=5.06 Â 10 -9 ) and transplanted epithelium (p=7.79 Â 10 -14 ) shares a substantially greater degree of similarity to fetal small intestinal epithelium (Figure 1-figure sup plement 1A).
While unbiased clustering demonstrated that transplanted epithelium closely resembles fetal epithelium, we noted a shift toward the adult transcriptome that resulted in a relative increase in the correlation between transplanted HIO epithelium and adult duodenum-derived epithelium grown in vitro (Figure 1-figure supplement 1B, p=1.17 Â 10 -4 ). Principle component analysis (PCA) of this multi-dimensional gene expression dataset (Figure 1-figure supplement 1C) corroborated the correlation analysis, and indicated that PC1 was correlated with developmental stage (PC1, 27.75% cumulative variance) and PC2 was correlated with tissue maturation status (PC2, 21.49% cumulative variance); cumulatively, PC1 and PC2 accounted for 49.24% of the cumulative variance between samples, suggesting that developmental stage and tissue maturation status are major sources of the transcriptional variation between samples. HIO epithelium clustered with fetal epithelium along PC2, whereas transplanted HIO epithelium clustered with adult epithelium. We further used differential expression analysis to demonstrate that in vitro grown HIO epithelium is similar to the immature human intestine, whereas in vivo transplanted HIO epithelium is similar to the adult epithelium. To do this, we identified differentially expressed genes through two independent comparisons: (1) human fetal vs. adult epithelium; (2) HIO epithelium vs. transplanted HIO epithelium. Genes enriched in transplanted HIO epithelium relative to the HIO epithelium were compared to genes enriched in the adult duodenum relative to fetal duodenum (Figure 1-figure  supplement 1D). There was a highly significant correlation between log 2 -transformed expression ratios where transplanted HIOs and adult epithelium shared enriched genes while HIO and fetal epithelium shared enriched genes (p=2.6 Â 10 -28 ). This analysis supports previously published data indicating that the epithelium from HIOs grown in vitro recapitulates the gene expression signature of the immature duodenum and demonstrates that the HIO epithelium is capable of adopting a transcriptional signature that more strongly resembles adult duodenum following transplantation into mice.

HIOs can be stably associated with non-pathogenic E. coli
Given that the HIO epithelium recapitulates many of the features of the immature intestinal epithelium, we set out to evaluate the effect of bacterial colonization on the naïve HIO epithelium. Previous studies have established that pluripotent stem-cell-derived intestinal organoids can be injected with live viral (Finkbeiner et al., 2012) or bacterial pathogens (Leslie et al., 2015;Engevik et al., 2015;Forbester et al., 2015); however, it was not known if HIOs could be stably co-cultured with nonpathogenic microorganisms. We co-cultured HIOs with the non-motile human-derived Esherichia coli strain ECOR2 (Ochman and Selander, 1984). Whole genome sequencing and phylogentic analysis demonstrated that E. coli str. ECOR2 is closely related to other non-pathogenic human E. coli and only distantly related to pathogenic E. coli and Shigella isolates (Figure 1-figure supplement 3).
We developed a microinjection technique to introduce live E. coli into the HIO lumen in a manner that prevented contamination of the surrounding media ( Figure 1-figure supplement 2). HIOs microinjected with 10 5 live E. coli constitutively expressing GFP exhibit robust green fluorescence within 3 hr of microinjection ( Figure 1A and Video 1). Numerous E. coli localized to the luminal space at 48 hr post-microinjection and are present adjacent to the HIO epithelium, with some apparently residing in close opposition to the apical epithelial surface ( Figure 1B).
In order to determine the minimum number of colony-forming units (CFU) of E. coli required to establish short term colonization (24 hr), we microinjected increasing numbers of live E. coli suspended in PBS into single HIOs and collected and determined the number of bacteria in the luminal contents at 24 hr post-microinjection ( Figure 1C). Single HIOs can be stably colonized by as few as 5 CFU E. coli per HIO with 77.8% success (positive luminal culture and negative external media culture at 24 hr post-injection) and 100% success at !100 CFU per HIO ( Figure 1C). Increasing the number of CFU E. coli microinjected into each HIO at t = 0 did result in a significant increase in the mean luminal CFU per HIO at 24 hr post-microinjection at any dose (ANOVA p=0.37; Figure 1C). Thus, the 24 hr growth rate of E. coli within the HIO lumen was negatively correlated with the CFU injected (r 2 = 0.625, p=3.1 Â 10 -12 ; Figure 1C). Next, we examined the stability of HIO and E. coli co-cultures over time in vitro. HIOs were microinjected with 10 CFU E. coli and maintained for 24-72 hr ( Figure 1D). Rapid expansion of E. coli density within the HIO lumen was observed in the first 24 hr, with relatively stable bacterial density at 48-72 hr. A 6.25-fold increase in bacterial density was observed between 24 and 72 hr post-microinjection (p=0.036). Importantly, samples taken from the external HIO culture media were negative for E. coli growth.
Finally, we examined the stability of HIO cultures following E. coli microinjection ( Figure 1E). A total of 48 individual HIOs were microinjected with 10 4 CFU E. coli each. Controls were microinjected with sterile PBS alone. We found that external culture media was sterile in 100% of control HIOs throughout the entire experiment, and in 100% of E. coli injected HIOs on days 0-2 postmicroinjection. On days 3-9 post-microinjection some cultured media was positive for E. coli growth; however, 77.08% of E. coli injected HIOs were negative for E. coli in the external culture media throughout the timecourse. Additional control experiments were conducted to determine if the HIO growth media had any effect on E. coli growth. E.coli-inoculated HIO growth media showed that the media itself allowed for robust bacterial growth, and therefore the absence of E. coli growth in external media from HIO cultures could not be attributed to the media composition alone (Figure 1-figure supplement 3). Thus, the large majority of E. coli colonized HIOs remain stable for an extended period when cultured in vitro and without antibiotics.

Bacterial colonization elicits a broad-scale, time-dependent transcriptional response
Colonization of the immature gut by microbes is associated with functional maturation in both model systems (Kremer et al., 2013;Sommer et al., 2015;Broderick et al., 2014;Erkosar et al., 2015) and in human infants . To evaluate if exposing HIOs to E. coli led to maturation at the epithelial interface, we evaluated the transcriptional events following microinjection of live E. coli into the HIO lumen. PBS-injected HIOs (controls) and HIOs co-cultured with E. coli were collected for transcriptional analysis after 24, 48 and 96 hr (Figure 2). At 24 hr post-microinjection, a total of 2018 genes were differentially expressed (adjusted-FDR < 0.05), and the total number of differentially expressed genes was further increased at 48 and 96 hr post-microinjection relative to PBSinjected controls (Figure 2A). Principle component analysis demonstrated that global transcriptional activity in HIOs is significantly altered by exposure to E. coli, with the degree of transcriptional change relative to control HIOs increasing over time ( Figure 2B).
Gene set enrichment analysis (GSEA)  using the GO (Ashburner et al., 2000;Gene Ontology Consortium, 2015) and REACTOME (Croft et al., 2014;Fabregat et al., 2016) databases to evaluate RNA-seq expression data revealed coordinated changes in gene expression related to innate anti-microbial defense, epithelial barrier production, adaptation to low oxygen, and tissue maturation ( Figure 2C). Innate antimicrobial defense pathways, including genes related to NF-kB signaling, cytokine production, and Tolllike receptor (TLR) signaling were strongly upregulated at 24 hr post-microinjection and generally exhibited decreased activation at later time points. GSEA also revealed changes in gene expression consistent with reduced oxygen levels or hypoxia, including the induction of proangiogenesis signals. A number of pathways related to glycoprotein synthesis and modification, including O-linked mucins, glycosaminoglycans, and proteoglycans, were up-regulated in the initial stages of the transcriptional response Video 1. Animation of individual epifluorescent microscopy images from a human intestinal organoid (HIO) containing live GFP þ E. coli str. ECOR2. Images were captured at 10 min intervals over the course of 18 hr an coalated in sequential order. Representative of three independent experiments. See Figure 1A. DOI: https://doi.org/10.7554/eLife.29132.007 (Syndecans, integrins), exhibited a somewhat delayed onset (O-linked mucins), or exhibited consistent activation at all time points post-microinjection (Keratan sulfate and glycosaminoglycan biosynthesis). Finally, genes sets associated with a range of processes involved in tissue maturation and development followed a distinct late-onset pattern of expression. This included broad gene ontology terms for organ morphogenesis, developmental maturation, and regionalization as well as more specific processes such as differentiation of mesenchymal and muscle cells, and processes associated with the nervous system ( Figure 2C). We also made correlations between upregulated genes in the RNA-seq data ( Figure 2D) and protein factors present in the organoid culture media following E. coli microinjection ( Figure 2E). bdefensin 1 (DEFB1 (gene); BD-1 (protein)) and b-defensin 2 (DEFB4A (gene); BD-2 (protein)) exhibited distinct patterns of expression, with both DEFB1 and its protein product BD-1 stable at 24 hr after E. coli microinjection but relatively suppressed at later time points, and DEFB4A and BD-2 strongly induced at early time points and subsiding over time relative to PBS-injected controls. By contrast, inflammatory regulators IL-6 and IL-8 and the pro-angiogenesis factor VEGF were strongly induced at the transcriptional level within 24-48 hr of E. coli microinjection. Secretion of IL-6, IL-8, and VEGF increased over time, peaking at 5-9 days after E. coli association relative to PBS-injected controls ( Figure 2E). Taken together, this data demonstrates a broad-scale and time-dependent transcriptional response to E. coli association with distinct early-and late-phase patterns of gene expression and protein secretion.

Bacterial colonization results in a transient increase in epithelial proliferation and the maturation of enterocytes
While the transcriptional analysis demonstrated strong time-dependent changes in the cells that comprise the HIO following E. coli colonization, we hypothesized that exposure to bacteria may also alter the cellular behavior and/or composition of the HIO. Previous studies have demonstrated that bacterial colonization promotes epithelial proliferation in model organisms (Bates et al., 2006;Cheesman et al., 2011;Neal et al., 2013;Kremer et al., 2013;Ijssennagger et al., 2015). We examined epithelial proliferation in HIOs over a timecourse of 96 hr by treating HIOs with a single 2 hr exposure of 10 M EdU added to the culture media from 22 to 24 hr after microinjection with 10 4 CFU E. coli or PBS alone. HIOs were subsequently collected for immunohistochemistry at 24, 48, and 96 hr post-microinjection ( Figure 3). The number of proliferating epithelial cells (Edu nþ and E-cadherin nþ ) was elevated by as much as three-fold in E. coli-colonized HIOs relative to PBS-treated HIOs at 24 hr ( Figure 3A-B). However, at 48 hr post-microinjection, the proportion of EdU + epithelial cells was significantly decreased in E. coli colonized HIOs relative to control treated HIOs. This observation was supported by another proliferation marker, KI67 (Gerdes et al., 1984) ( Figure 3B), as well as RNA-seq data demonstrating an overall suppression of cell cycle genes in E. coli colonized HIOs relative to PBS-injected HIOs at 48 hr post-microinjection (Figure 3-figure supplement 1). By 96 hr post-microinjection the proportion of EdU+ epithelial cells was nearly identical in E. coli and PBS-treated HIOs ( Figure 3B). Collectively, these results suggest that E. coli colonization is associated with a rapid burst of epithelial proliferation, but that relatively few of the resulting daughter cells are retained subsequently within the epithelium.
The transcription factor Sox9 is expressed by progenitor cells in the murine intestinal epithelium (Bastide et al., 2007;Mori-Akiyama et al., 2007), and several epithelial subtypes are derived from a Sox9-expressing progenitor population in the mature intestinal epithelium (Bastide et al., 2007;  Furuyama et al., 2011). We examined SOX9 expression in HIOs following microinjection with E. coli or PBS alone over a 96 hr time course ( Figure 3C). In the PBS-treated HIOs, the majority of epithelial cells exhibited robust nuclear SOX9 expression at all time points examined. However, SOX9 expression was dramatically reduced in E. coli-colonized HIOs at 48-96 hr after microinjection and was notably distributed in nuclei farthest from the lumen and adjacent to the underlying mesenchyme, mirroring the altered distribution of EdU + nuclei seen in Figure 3B. This observation suggests that there is a reduction in the number of progenitor cells in the HIO epithelium following E. coli colonization and implies that other epithelial types may account for a greater proportion of the HIO epithelium at later time points post-colonization. We saw no appreciable staining for epithelial cells expressing goblet, Paneth, or enteroendocrine cell markers (MUC2, DEFA5, and CHGA, respectively; negative data not shown). However, expression of the small intestinal brush border enzyme dipeptidyl peptidase-4 (DPPIV) was found to be robustly expressed in the E. coli-colonized HIOs at 48 and 96 hr post-microinjection ( Figure 3D). DPPIV was not detected in any of the PBS-injected HIOs at any timepoint. Lysozyme (LYZ), an antimicrobial enzyme expressed by Paneth-like progenitors in the small intestinal crypts Bevins and Salzman (2011), was widely distributed throughout the epithelium of PBS-treated HIOs as we have previously described ) ( Figure 3D). However, in E. coli-colonized HIOs, LYZ expression was restricted to distinct clusters of epithelial cells and, notably, never overlapped with DPPIV staining ( Figure 3D). Given that bona fide Paneth Cell markers (i.e. DEFA5) were not observed in any HIOs, it is likely that the LYZ expression is marking a progenitor-like population of cells. Taken together, these experiments indicate that E. coli colonization induces a substantial but transient increase in the rate of epithelial proliferation followed by a reduction and redistribution of proliferating epithelial progenitors and differentiation of a population of cells expressing small intestinal enterocyte brush boarder enzymes over a period of 2-4 days.

E. coli colonization is associated with a reduction in luminal O 2
The mature intestinal epithelium is characterized by a steep oxygen gradient, ranging from 8% oxygen within the bowel wall to <2% oxygen in the lumen of the small intestine . Reduction of oxygen content in the intestinal lumen occurs during the immediate perinatal period , resulting in changes in epithelial physiology Kelly et al., 2015;Colgan et al., 2013;Zeitouni et al., 2016) that helps to shape the subsequent composition of the microbiota (Schmidt and Kao, 2014;Espey, 2013;Albenberg et al., 2014;Palmer et al., 2007;Koenig et al., 2011). Analysis of the global transcriptional response to E. coli association in the immature intestinal tissue revealed pronounced and coordinated changes in gene expression consistent with the onset of hypoxia ( Figure 2C-E). We therefore measured oxygen concentration in the lumen of control HIOs and following microinjection of live E. coli using a 50 m diameter fiberoptic optode ( Figure 4A-B). Baseline oxygen concentration in the organoid lumen was 8.04 ± 0.48%, which was significantly reduced relative to the external culture media (18.86 ± 0.37%, p=3.6 Â 10 -11 ). At 24 and 48 hr post-microinjection, luminal oxygen concentration was significantly reduced in E. coli-injected HIOs relative to PBS-injected HIOs (p=0.04 and p=5.2 Â 10 -05 , respectively) reaching concentrations as low as 1.67 ± 0.62% at 48 hr ( Figure 4A). E. coli injected HIOs were collected and CFU were enumerated from luminal contents at 24 and 48 hr post-microinjection. We observed a highly significant negative correlation between luminal CFU and luminal oxygen concentration where increased density of luminal bacteria was correlated with lower oxygen concentrations (r 2 = 0.842, p=6.86 Â 10 -5 ; Figure 4B). Finally, in order to assess relative oxygenation in the epithelium itself, we utilized a small molecule pimonidazole (PMDZ), which forms covalent conjugates with thiol groups on cytoplasmic proteins only under low-oxygen conditions (Arteel et al., 1998). Fluorescent immunochemistry demonstrated enhanced PMDZ uptake in E. coli associated HIO epithelium, and in HIOs grown in 1% O 2 as a positive control when compared to to PBS-injected  HIOs, or HIOs injected with heat killed E. coli at 48 hr post-microinjection ( Figure 4C). Thus, luminal and epithelial oxygen is reduced following microinjection of E. coli into the HIO, consistent with data in mice showing that the in vivo epithelium is in a similar low-oxygen state in normal physiological conditions (Schmidt and Kao, 2014;Kelly et al., 2015;Kim et al., 2017).

NF-kB integrates complex microbial and hypoxic stimuli
E. coli colonization elicits a robust transcriptional response in immature intestinal tissue (Figure 2) that is associated with the onset of luminal oxygen depletion and relative tissue hypoxia (Figure 4). We set out to determine whether we could assign portions of the transcriptional response to direct interaction with microbes or to the subsequent depletion of luminal oxygen. In the RNA-seq analysis (Figure 2), NF-kB signaling emerged as a major pathway involved in this complex host-microbe interaction, and NF-kB has been shown by others to act as a transcriptional mediator of both microbial contact and the response to tissue hypoxia (Rius et al., 2008;Gilmore, 2006;Wullaert et al., 2011). Gene Ontology and REACTOME pathway analysis showed that NF-kB signaling components are also highly up-regulated following microinjection of E. coli into HIOs ( Figure 2C and Figure 5figure supplement 1A). Thus, we assessed the role of NF-kB signaling in the microbial contact-associated transcriptional response and the hypoxia-associated response using the highly selective IKKb inhibitor SC-514 Litvak et al., 2009) to inhibit phosphorylation and activation of the transcription factor p65 ( Figure 5-figure supplement 1B). Another set of HIOs was simultaneously transferred to a hypoxic chamber and cultured in 1% O 2 with and without SC-514. At 24 hr post-treatment, HIOs were harvested for RNA isolation and RNA-seq. We devised an experimental scheme that allowed us to parse out the relative contributions of microbial contact and microbeassociated luminal hypoxia in the transcriptional response to association with live E. coli ( Figure 5A and Figure 5-figure supplement 1C). First, we identified a set of genes significantly up-regulated (log 2 FC > 0 and FDR-adjusted p-value < 0.05) by microinjection of either live E. coli or heat-inactivated E. coli (contact dependent genes). From this gene set, we identified a subset that was suppressed by the presence of NF-kB inhibitor SC-514 during association with either live or heatinactivated E. coli (log 2 FC < 0 and FDR-adjusted p-value < 0.05; Gene Set I, Figure 5B). Thus, Gene Set I represents the NF-kB dependent transcriptional response to live or dead E. coli. Genes induced by live or heat-inactivated E. coli but not suppressed by SC-514 were considered NF-kB independent (Gene Set III, Figure 5B). Likewise, we compared genes commonly up-regulated by association with live E. coli and those up-regulated under 1% O 2 culture conditions. A subset of genes induced by either live E. coli or 1% O 2 culture but suppressed by the presence of NF-kB inhibitor was identified as the NF-kB-dependent hypoxia-associated transcriptional response (Gene Set II, Figure 5B). Genes induced by live E. coli or hypoxia but not inhibited by the presence of NF-kB inhibitor were considered NF-kB independent transcriptional responses to microbe-associated hypoxia (Gene Set IV). Gene lists for each gene set are found in Supplementary file 1.
Following the identification of these four gene sets, we then applied over-representation analysis using the GO and REACTOME pathway databases to identify enriched pathways for each of the four gene sets, resulting in four clearly distinguishable patterns of gene pathway enrichment ( Figure 5C). Contact with either live or heat-inactivated E. coli is sufficient to promote expression of genes involved in maintaining epithelial barrier integrity and mucin production, an effect that is suppressed in the presence of NF-kB inhibitor. Additionally, key developmental pathways including epithelial morphogenesis, digestive tract development, and expression of digestive enzymes appear to be driven primarily by bacterial association and are largely NF-kB dependent. Robust innate and adaptive defense requires both bacterial contact and hypoxia, with some genes associated with antigen processing and cytokine signaling being NF-kB dependent (Gene Set II) and others associated with   D NF-kB-independent gene sets (Gene Sets III and IV). Genes associated with antimicrobial defensin peptides were enriched only in the hypoxia-asociated, NF-kB-independent gene set (Gene Set IV), suggesting that antimicrobial peptides are regulated by mechanisms that are distinct from other aspects of epithelial barrier integrity such as mucins and epithelial junctions (Gene Set I). TLR signaling components were is broadly enhanced by live E. coli and associated with both microbial contact and hypoxia were largely NF-kB independent (Gene Sets III and IV). There was a notable transcriptional signature suggesting metabolic and mitochondrial adaptation to bacteria that was independent of NF-kB and primarily driven by bacterial contact rather than hypoxia (Gene Set III).
To interrogate the transcriptional changes influenced by SC-514 exposure, we examined overrepresented genes sets from the GO and REACTOME databases in genes that were significantly upor down-regulated by treatment with SC-514 alone ( Figure 5-figure supplement 1C and D) . Notably, SC-514 alone does not appear to have a strong effect on the pathways identified in Figure 5C as key NF-kB-dependent responses to bacterial contact and/or hypoxia. In Figure 5-figure supplement 1E, we examined the degree of overlap between Gene Set I, Gene Set II, and the set of genes that are significantly down-regulated in PBS-injected HIOs treated with SC-514. This analysis demonstrates that the majority of genes in Set I and Set II are not significantly down-regulated in PBS-injected HIOs treated with SC-514. The most significant effects of SC-514 alone among Gene Set I and Gene Set II genes are related to metabolism, redox state, and ribosomal dynamics ( Figure 5-figure supplement 1F). Thus, the effect of SC-514 alone cannot account for the NF-kBdependent changes in innate and adaptive defense, epithelial barrier integrity, angiogenesis and hypoxia signaling, or intestinal development following bacterial contact and/or hypoxia during colonization.
Finally, we also examined the role of microbial contact and hypoxia in colonization-induced changes in AMP, cytokine, and growth factor secretion using ELISA ( Figure 5-figure supplement 2). Consistent with findings from the RNA-seq data, these results indicate that there are diverse responses to bacterial contact and hypoxia. We observed cases where cytokines were induced by either microbial contact or hypoxia alone (IL-6), other cases where hypoxia appeared to be the dominant stimuli (BD-1), and a third regulatory paradigm in which the response to live E. coli evidently results from the cumulative influence of bacterial contact and hypoxia (BD-2, IL-8, VEGF). Taken together, this analysis demonstrates that association of immature intestinal epithelium with live E. coli results in a complex interplay between microbial contact and microbe-associated hypoxiainduced gene expression and protein secretion.

Bacterial colonization promotes secretion of antimicrobial peptides
Antimicrobial peptides (AMPs) are key effectors for innate defense of epithelial surfaces (Muniz et al., 2012) and act to inhibit microbial growth through direct lysis of the bacterial cell wall and modulation of bacterial metabolism (Ganz, 2003;Bevins and Salzman, 2011;O'Neil and O'Neil, 2003;Vora et al., 2004;Brogden, 2005). Defensin gene expression is highly up-regulated Figure 5. NF-kB integrates complex microbial and hypoxic stimuli. (A) Analysis scheme for identifying genes sets representing the components of the transcriptional response to live E. coli that could be recapitulated with heat-inactivated E. coli (contact induced) or hypoxia (microbial-associated hypoxia induced) as well as the subsets of genes induced through NF-kB dependent signaling. HIOs were microinjected with PBS, 10 4 CFU E. coli or an equivalent concentration of heat-inactivated E. coli and cultured under standard cell culture conditions or hypoxic conditions (1% O 2 , 5% CO 2 , 94% N 2 ) with and without 10 mM SC-514. (B) Scatter plots with density overlay indicating the genes meeting the a priori criteria identified in panel A with an FDR-adjusted p-value of < 0.05 for the comparisons listed on the axes of the plot. (C) Bar plot of the proportion of genes in the input gene sets mapping to each pathway from the GO and REACTOME databases enrichment p-values for each of the gene sets identified in A. Pathways with enrichment p-values > 0.01 were excluded from the plot. Results represent N = 4-5 biological replicates per treatment condition, with each replicate consisting of 5-6 pooled and identically treated HIOs. DOI: https://doi.org/10.7554/eLife.29132.012 The following figure supplements are available for figure 5: following microinjection of E. coli into HIOs ( Figures 2D-E and and 4C). Using an annotated database of known AMPs (Wang et al., 2016) to query our RNA-seq datasets, we found that several AMPs are up-regulated in the immature intestinal epithelium following E. coli association ( Figure 6A). Among these, DEFB4A and DEFB4B, duplicate genes encoding the peptide human bdefensin 2 (Harder et al., 1997), were the most highly up-regulated; other AMPs induced by E. coli association included multi-functional peptides CCL20, CXCL2, CXCL1, CXCL6, CXCL3, REG3A , and LTF ( Figure 6A). Analysis of RNA-seq data from HIOs microinjected with live or heat-killed E .coli with and without NF-kB inhibitor or culture of HIOs under hypoxic conditions had indicated that defensin genes were enriched among the set of NF-kB-independent genes induced by hypoxia ( Figure 5C). We examined DEFB4A expression specifically ( Figure 6B) and found that relative to control treatment, microinjection of live E. coli resulted in a 7.38-fold increase in normalized DEFB4A expression. Consistent with the notion that DEFB4A expression is induced by hypoxia and is not dependent on NF-kB signaling, NF-kB inhibitor treated HIOs injected with E. coli still showed an~8-fold increase in gene expression and hypoxia-cultured HIOs showed a~5.5-fold PPBP  CXCL14  HAMP  CCL26  CCL25  LEAP2  CCL27  CXCL12  CAMP  DEFB1  HTN1  CXCL13  CCL21  HTN3  DEFB109P3  LYZ  SLPI  CXCL10  CCL11  DEFB124  LTF  CXCL3  REG3A  CXCL6  CXCL1  CXCL2  CCL20  DEFB4B Figure 7E). This result was confirmed with PAS/AB staining of HIOs microinjected with PBS, live or heat-inactivated E. coli, or cultured under hypoxic conditions for 24 hr, where bacterial contact promoted formation of a mucus layer while PBS microinjection or culture under hypoxic conditions did not ( Figure 7F). Taken together, these results indicate that association of the immature intestinal epithelium with E. coli promotes robust mucus secretion through an NF-kB-dependent mechanism and that hypoxia alone is not sufficient to recapitulate E. coli-induced mucus production.

NF-kB signaling is required for the maintenance of barrier integrity following bacterial colonization
Having established that the immature intestinal epithelium in HIOs (Figure 1-figure supplement 1) can be stably associated with non-pathogenic E. coli (Figure 1), resulting in broad changes in transcriptional activity ( Figure 2) and leading to elevated production of AMPs ( Figure 6) and epithelial mucus secretion (Figure 7), we hypothesized that these changes in gene and protein expression would have functional consequences for the immature epithelial barrier. RNA-seq analysis demonstrated broad up-regulation of transcription in genes involved in the formation of the adherens junction and other cell-cell interactions in HIOs after microinjection with live E. coli that was inhibited in the presence of NF-kB inhibitor SC-514 ( Figure 8A). We utilized a modified FITC-dextran permeability assay (Leslie et al., 2015) and real-time imaging of live HIO cultures to measure epithelial barrier function in HIOs microinjected with PBS, live E. coli, or live E. coli +SC-514 at 24 hr after microinjection ( Figure 8B). While HIOs microinjected with PBS or E. coli retained 94.1 0.3% of the FITC-dextran fluorescence over the 20-hr assay period, E. coli microinjected HIOs cultured in the presence of SC-514 retained only 45.5 ± 26.3% of the fluorescent signal (p=0.02; Figure 8B). We also measured the rate of bacterial translocation across the HIO epithelium, which resulted in contaminated culture media ( Figure 8C). HIOs microinjected with E. coli and treated with SC-514 were compared to E. coli microinjected HIOs treated with vehicle (DMSO controls) and PBS microinjected controls over 7 days in culture. HIOs associated with E. coli +SC-514 exhibited a rapid onset of bacterial translocation by days 2-3, with bacterial translocation detected in 96% of SC-514-treated HIOs by day 7 compared to 23% of HIOs microinjected with E. coli and cultured in DMSO (P=<2 Â 10 -16 ; Figure 8C). Therefore, blocking NF-kB signaling inhibited epithelial barrier maturation resulting in increased bacterial translocation during E. coli association with the immature epithelium.
Bacterial colonization promotes resilience of the epithelial barrier during cytokine challenge Finally, we assayed epithelial barrier function under circumstances recapitulating physiologic inflammation. TNFa and IFNg are key cytokines mediating innate and adaptive immune cell activity in the gut (Turner, 2009) during bacterial infection (Rhee et al., 2005;Emami et al., 2012) and in necrotizing enterocolitis Ford et al., 1996Ford et al., , 1997Halpern et al., 2003;Upperman et al., 2005). The combination of TNFa and IFNg has been previously demonstrated to induce barrier permeability in a dose-dependent manner in Transwell epithelial cultures (Wang et al., 2005;Wang et al., 2006). Thus, HIOs were microinjected with PBS or live E. coli and cultured for 24 hr and were subsequently microinjected with FITC-dextran and treated with PBS or a cocktail of TNFa H and E, AB, PAS, or PAS-AB and imaged under 100X light microscopy. (D) Confocal micrograph of HIO epithelium from a control HIO or an HIO microinjected with E. coli and cultured for 48 hr. Nuclei are stained blue with DAPI, and fluorescent antibody-labeled proteins E-cadherein and Mucin 5 AC are pseudocolored in white or red, respectively. UEA1 lectin is used to label the carbohydrate moiety Fuca1-2Gal-R, which is pseudo colored in green. 60X optical magnification. (E) Heatmap of normalized RNA-seq glycotransferase and mucin gene counts of HIOs associated with live or heatinactivated E. coli, E. coli + NF-kB inhibitor (SC-514) or HIOs cultured under hypoxic conditions for 24 hr. Results represent the mean of N = 4-5 biological replicates per treatment condition, with each replicate consisting of 5-6 pooled and identically treated HIOs. (F) PAS-AB staining of HIOs treated as indicated in the figure labels for 24 hr. 10X magnification. Histological and immunofluorescent images in panels B-D and F are representative of three or more independent experiments, each consisting of 5-10 HIOs per treatment group. DOI: https://doi.org/10.7554/eLife.29132.018 The following figure supplement is available for figure 7: and IFNg added to the external media to expose the basolateral epithelium ( Figure 8D). Loss of FITC-dextran fluorescence was observed using live-imaging and indicated that treatment with TNFa and IFNg alone resulted in a rapid and sustained decrease in luminal fluorescence relative to PBS or E. coli injected HIOs (p=5 Â 10 -4 , Figure 8D). However, HIOs associated with E. coli prior to addition of the TNFa and IFNg cocktail retained significantly more fluorescent signal relative to treatment with TNFa and IFNg alone (p=0.042, Figure 8D). We examined expression and distribution of the  Additional files

Supplementary files
. Supplementary file 1 List of differentially expressed genes and the Gene Set (I-IV) assignments used in the analyses presented in Figure 5.