Global host response to pathogenic E. coli

Previous studies from our group indicated that enteroaggregative E. coli formed unique adherence patterns to human intestinal enteroids that were dependent on intestinal segment and patient donor [35]. This adherence was also correlated to the presence of aggregative adherence fimbriae A gene (aafA) in infecting strains. Hypothesizing that the close association of E. coli in these aggregative morphotypes within the intestinal mucosa might correlate with the host response to the infection, we performed a large-scale transcriptome analysis of human intestinal enteroids across all four segments of the intestine (duodenum, jejunum, ileum, and colon) between at least three different patient donors with five different pathotypes of E. col (Fig 1A shows the general experimental scheme). In this regard, the duodenum, ileum and colon came from the same patient therefore allowing us to assess the effect of segment on the response, as well as compare the same segment between two different patients. In addition, we assessed responses to five different E. coli pathotypes. Each pathotype either causes a unique disease in the intestine or extra-intestinally. Such an approach allowed us to also determine how bacterial genetics within the same infecting species overlays on the other variables described above. The E. coli chosen for the study include a non-pathogenic control strain (E. coli: HS), enteroaggregative E. coli strain 042 (EAEC: a commonly used prototype strain that caused disease in a volunteer challenge study) [39], EAEC 042:ΔaafA (mutant strain lacking the major fimbrial gene aafA), enterohemorrhagic E. coli strain 8624 (EHEC prototype strain O157:H7, a Shiga toxin producing strain responsible for a food-borne disease in Oregon and Michigan [40], and extra-intestinal pathogenic E. coli (ExPEC, prototype strain CP9 that does not cause intestinal disease but is a major cause of bloodstream and urinary tract infections [41, 42]. These strains allow us to test the dependence of the host response on adherence (EAEC wild-type versus the aafA mutant), two related diarrheagenic strains with different disease pathologies (EAEC versus EHEC), intestinal infecting strains (EAEC and EHEC) versus an extraintestinal strain (ExPEC), and all pathogenic strains versus the non-pathogenic control (E. coli HS). A principal component analysis is an accepted approach to highlight global trends in large datasets and was performed here to examine the relationship of segment, patient, and strain to each other (Fig 1B). Interestingly, there were significant differences in the response between the most distal and proximal intestinal segments. In the duodenum, there was substantial variance in the baseline host response in uninfected samples between patients described by PC3 (Fig 1B, left panel, blue dots) that converged on a similar host response when bacteria, regardless of the pathotype or strain, were added (Fig 1B, left panel, all dots) described by PC1. In the duodenum from the PCA analysis the infection event drives the differential gene expression regardless of bacterial phenotype, with the bacterial phenotype, and patient identity, contributing to a lesser extent. The colonoids exhibited a less orchestrated response. The mock infected samples for three different patients showed a highly similar host response (Fig 1B, right panel, blue dots) that became highly scattered once infected, regardless of strain (Fig 1B, right panel, all dots). These associations were statistically significant (p < 0.05). This suggests that whereas the duodenum had a more similar response between patients when E. coli of any type were used, the colon has responses that were highly patient specific and varied dependent on the infecting organism. These trends were generally true for tissues that bordered these segments (i.e. the jejunum resembling more the duodenum and the ileum resembling the colon–S1A Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. RNA sequencing analysis of pathogenic E. coli infection of human intestinal enteroid monolayers (HIEMs).

(A) Experimental design and schematic representation for these experiments. (B) Principal component analysis of duodenum and colon monolayers demonstrating the variability observed in the samples (infected and mock controls). (C) Total number of significant genes identified for the EAEC infected group (BH adjusted p-value < 0.05). (D) Venn diagrams present the number of genes that are differentially expressed within the duodenum, jejunum, ileum and colon of EAEC infected samples compared to the uninfected control (E). Volcano plots of differential gene expression in each tissue for the EAEC infected samples compared to the mock infected controls. Red dots indicate genes that have an adjusted p-value of at least 0.05 with a linear fold change of at least 1.5.

https://doi.org/10.1371/journal.ppat.1008851.g001

We next assessed the general host response to infection, focusing mostly on EAEC. The duodenum, jejunum and colon demonstrated significant changes in gene expression, the duodenum having the highest of the three, and with the number of upregulated genes three times higher than the number of downregulated genes (Fig 1C). Remarkably, the ileum had little change in gene expression, either up or down, a trend observed for all E. coli pathotypes or strains assessed (S1B Fig, S1and S2 Appendices). Similar to EAEC response, the duodenum showed the greatest number of gene expression changes, with EHEC and ExPEC leading the way. Of note, EAEC induced the least number of gene changes among all the strains that were assessed, even less than the non-pathogenic control. Interestingly, the EAEC lacking the fimbrial gene aafA had over three times and over seven times the number of total gene expression changes in the duodenum and colon, respectively, than the wild-type strain (discussed below). A significant segment-dependent change in gene expression was observed with only 13 genes shared between the four segments (Fig 1D). For example, there were 306 different genes downregulated in each individual segment that were not shared between segments compared to just 75 shared by at least two segments (368 vs 152 for upregulated genes). Of over 13,000 profiled host transcripts, 560 were significantly altered in duodenum, 354 in jejunum, 32 in ileum and 257 in colon during EAEC infection (Fig 1E, S4 Fig and S2 Appendix), less than 10% of them overlapping between segments. Volcano plots are useful because they can show the dynamic range of the level of responses in the data set. As shown in Fig 1E, upregulated genes had a much larger range of expression than did the downregulated genes, with the duodenum have the greatest spread of all four segments and the ileum the least, a trend that was generally true for all E. coli pathotypes or strains (S2 Fig, S1 and S2 Appendices).

Gene expression signatures in segments and patients for E. coli pathotype infections

Hierarchical clustering can be used to determine how “related” gene responses between segments and pathotypes are to each other. This allows one to ask questions as to whether certain intestinal regions or pathogens behave like others, and so on. We used Pearson correlation coefficient analysis to examine the relationship, displayed in a dendrogram format, between various infection groups. As shown in Fig 2A, and consistent with data represented in Fig 1, the responses in the duodenum were highly consistent within each pathotype or strain between each of the three different patients examined. This means that in the duodenum, the response is less driven by the host and more by the infecting organism. In this regard, all the bacterial infections (HS, EAEC, EHEC, and ExPEC) were more similar to each other than mock, and, rather surprisingly, EAEC was more similar to HS than any others (including a strain with an identical background but lacking the fimbrial gene aafA–Fig 2A). This is in complete contrast with the responses observed in the colon; only the mock-infected colonoids were most related to each other amongst the three different patients. In fact, the responses were so divergent between the patients that there was no observable pattern that was significant, suggesting that the individual patient response to the organism was more important than the organism itself. When the breadth of up and down regulated genes are clustered into a heat map and compared between each bacterium across each segment, EAEC demonstrated the fewest changes amongst all assayed genes while EAEC lacking the fimbrial aafA gene demonstrated the most (Fig 2B). As was observed for EAEC, there were far more unique genes expressed in each individual segment than there were shared between segments for each of the E. coli pathotypes or strains, thus signifying the truly unique response of each of these tissues to these bacteria (S3 and S4 Figs). The notable exception to this was again EAEC lacking the fimbrial gene which produced the highest number of total up- and down-regulated genes shared in common between all four segments. A Gene Set Enrichment Analysis (GSEA) was performed on the differential expression gene sets for each infection. The most common pathways that were enriched are mostly involved in adaptive immune response, cellular differentiation, metabolism, phagosome activation, G-Protein Coupled Receptors, G-alpha signaling and mitosis (S5 Fig and S3 Appendix). One of the more striking observations was changes in apoptosis and/or cell death responses; every pathotype, and especially the EAEC lacking aafA, had a positive upregulation of cell death pathways. We confirmed that EAEC was directly altering cell responses by examining the thickness of the intestinal enteroid epithelium during a six-hour infection using Periodic acid-Schiff Stain/Alcian Blue (PAS-AB). Whereas the mock infected samples maintained an epithelial layer of close to 80 microns (Fig 2D), upon infection with EAEC this thickness decreased on average 4-fold, which we interpreted as caused mostly by Goblet cells releasing their mucus (Fig 2E), indicating that the epithelium was responding by attempting to increase barrier integrity. Note that no cell rounding, monolayer sloughing, or reduction of TER was observed at a similar time point for these infections [35]. The mechanistic underpinnings of these observations are currently unknown but might relate to the findings discussed below.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. RNA-Sequencing analysis reveals a pathogen-specific differential gene expression signature within different segments of intestine.

(A) Hierarchical clustering dendrogram of all gene expression changes using the Pearson correlation coefficient among mock and E. coli infected samples for all four segments of intestine (Duodenum, Ileum, Colon from patients 103, 104 & 109 and jejunum from patient 2, 3 & 11 –see Methods) (B) Samples were hierarchically clustered based on the union of differentially expressed genes across all comparisons (false discovery rate (FDR) < 0.05 and fold change (FC) exceeding 1.5x). Genes were converted to 1 for up-regulated genes and -1 for down-regulated genes and 0 if it a genes were not significant in a particular signature. Samples were clustered used the Euclidean distance. (C) Heatmap showing EAEC infection-specific regulation of Muc gene expression in HIEs. (D) PAS/AB staining of HIEMs infected with EAEC shows epithelial cell damage. The left panel shows the mock-infected epithelial cell layer whereas the right panel reveals epithelial thinning and damage caused by EAEC adherence (EAEC clusters denoted by arrows). (E) Quantification of epithelial thinning caused by EAEC infection. Scale bars indicate 10μm. Epithelial thickness was quantified from average of 12 images taken at 100x and with field size of 2560x1920 pixel. Unpaired t-test was carried out, error bars denote SEM and ** indicates significantly different (P<0.01), * indicates significantly different (P<0.05).

https://doi.org/10.1371/journal.ppat.1008851.g002

One of the groups of genes that had a significant up-regulation in expression across most of its paralog members were mucins. Mucins are apically localized proteins that often are glycosylated at one or more sites along the polypeptide chain. In the intestine, they are a major component of the mucus layer [43]. We assembled 10 individual genes encoding different mucins for which the number of reads in our gene expression analysis was above the statistical threshold for stringency and plotted their expression as a heat map across all four segments and four pathotypes (including the EAEC fimbrial mutant–Fig 2C). Remarkably, in 31 of the 40 (77%) possible scenarios across all four segments, EAEC infection led to an up-regulation of one or more mucin genes over the mock-infected control. In general, infection with any bacteria of any type overwhelmingly followed the same trend with the greatest responses being observed in the duodenum and jejunum (Fig 2C). Muc 1 and Muc 13 were upregulated under every condition (40/40). Nine of the ten genes were universally upregulated in the duodenum and jejunum, signifying a strong general mucin positive response in these enteroid segments. Only one gene, that of Muc6, was either unchanged or downregulated under these conditions. Somewhat unexpectedly, and in contrast to the duodenum and jejunum, the ileum and colon, although displaying an overall up regulation of Muc genes, showed eight of the ten genes that had either no change or a down modulation of the gene, with Muc12 being either unchanged or upregulated in 20 of 20 cases. These results suggest the enteroids undergo a strong general up regulation of most mucins to the presence of bacteria, regardless of the pathotype, but that there are significant segment-specific responses, especially at the most distal and proximal ends of the alimentary canal, with some mucins seemingly being purposely down modulated.

Adherence of EAEC to human intestinal enteroids

The finding that EAEC forms different adherence patterns on enteroids from each intestinal segment [35], the significant host response induced by EAEC in all four segments, and the generalized upregulation of many mucin genes in response to infection, prompted a closer scrutiny of this unique host-pathogen interaction. We used scanning electron microscopy (SEM) to examine the interaction of EAEC with enteroids from the duodenum, ileum, and colon. As shown in Fig 3A–3C, mock-infected cells showed a normal undulating topography with occasional droplets of mucus (white balls marked with green asterisks). EAEC-infected cells formed biofilm-like aggregates on the duodenum and colon and classic aggregative adherence on the ileum, findings consistent with previous publications [35]. Overall, the surface of the infected cell looked very similar to mock-infected cells with the notable exception that EAEC seemed to preferentially cluster around the droplets of mucus (Fig 3A–3C, Row 2). In fact, when images were taken in areas of the mucosal epithelium that lacked these mucus droplets, no bacteria were observed (Fig 3A–3C), Row 3). Quantification of this observation indicated that there was an average increase of 5–10 times in the level of bacteria associated with mucus droplets (Fig 3D). When the experiment was repeated with EAEC lacking the aafA gene, very few bacterial cells were observed, and those that were observed, bound to the surface by atypical adherence (Fig 3A–3C, Row 4)). Transmission electron microscopy demonstrated that many EAEC preferentially and closely associated with Goblet cells (Fig 3E). Finally, the binding of individual bacterial cells or aggregates of bacteria to the enteroid surface resulted, with time, in the large outgrowth of aggregated bacteria that grew upwards from the point of origin (Supplemental Video File 1). This behavior was not observed for the control non-pathogenic strain HS or EAEC that lacked the aafA.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. SEM and TEM images of EAEC adherence to human intestinal enteroid monolayers.

(A-C) shows SEM image of Mock, EAEC and EAEC: ΔaafA infected duodenal, ileal and colon differentiated HIE monolayers. Row one shows differentiated enteroid monolayers covered by secreted mucus droplets (denoted by the asterisk). Row two shows adherence of EAEC to mucus (yellow arrow) present on the duodenal, ileal and colon monolayers. Row three shows EAEC infected duodenal, ileal and colon monolayers but with no EAEC adherence and without the presence of mucus droplets. Row four shows EAEC: ΔaafA infected duodenal, ileal and colon monolayers. (D) Quantification of the EAEC aggregates in the presence and absence of mucus droplets on enteroid monolayers was quantified from average of 12 images taken at 1700x and error bars denote SEM and *** indicates significantly different (P<0.001). (E) TEM of colonoid monolayers showing vesiculation of Goblet Cell (GC) mucus and EAEC adherence to GC mucus (yellow arrow). Scale bars indicate 10μm.

https://doi.org/10.1371/journal.ppat.1008851.g003

To further understand the nature of this association, we performed immunofluorescence microscopy of colonoids infected with EAEC expressing the green fluorescent protein while simultaneously staining for MUC2, a goblet cell secreted mucin (Fig 4). MUC2 was not one of the genes that met the threshold for reads in the gene expression analysis but is a mucin previously reported to be associated with other diarrheagenic pathotypes such as EHEC [44, 45]. As shown in Fig 4A, mock-infected cells were effectively stained with DAPI, MUC2, and E-cadherin allowing for easy visualization of cell borders and nuclei. Interestingly, although EAEC associated with MUC2 (red), there was a sizable fraction that did not. This is further appreciated using a Z-stack analysis (Fig 4B) whereby approximately one-half of EAEC clustered directly above MUC2 but the remainder seemed to intimately associate with the cell surface in areas that lacked MUC2 staining. To further investigate this phenomenon and gain a better structural appreciation of this observation, we used “apical-out” colonoids reported by Co et al. [46]. Apical-out enteroids have their apical surface facing the media while their basolateral surface is directed inwards, a convenient feature that allows for a 3-dimensional analysis of the bacterial association with colonoids. As observed in Fig 4C, whereas there was definite clustering of EAEC around clusters of MUC2, there was also a significant association of EAEC with the epithelial surface that lacked MUC2, especially pockets of bacteria that wedged themselves either between the cells or in the folds of the apical-out structure.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Association of EAEC to intestinal mucus (MUC2) to TC202, colon line as revealed by immunofluorescence micrographs.

(A) Representative confocal micrographs of HIEMs stained for nuclei (DAPI, blue), MUC2 (Red), E-cadherin (Purple) and EAEC (GFP, green). HIEMs were either mock treated or infected with EAEC A2A:GFP or EAEC:ΔaafA at a MOI of 10 for three hours. (B) Volume projections of EAEC A2A:GFP infected HIEMs (C) EAEC infection on apical out colonoids. 3D reconstruction images show EAEC growing above goblet cells secreting MUC2. (Inset shows higher magnification). Scale bars indicate 10μm.

https://doi.org/10.1371/journal.ppat.1008851.g004

To determine the importance of mucus in contributing to the total level of EAEC binding to the surface of intestinal enteroids, we used two contrasting chemical-based approaches to either enhance (DAPT, N-[N-(3, 5-Difluorophenacetyl)-L-alanyl]-S-phenyl glycine t-butyl ester) or remove (NAC, N-acetyl glucosamine) the levels of mucus on the surface of colonoids (Fig 5A). As shown visually via Giemsa staining (Fig 5B), and quantified (Fig 5D), when the DAPT concentration was increased to 10 μM, so did the number of aggregated EAEC bound to the surface. In contrast, as the concentration of NAC was increased, a dose-dependent reduction in the number of aggregated EAEC on the enteroid surface was also observed (Fig 5C and 5E). When examined via the use of immunostaining for MUC2 as a proxy for the presence of mucus (Fig 5F), a marginal increase in MUC2 levels and fluorescent EAEC bound was observed after DAPT treatment whereas there was complete ablation of MUC2 and very little EAEC bound when enteroids were pre-treated with NAC, findings generally consistent with the Giemsa staining noted above. Thus, it seems as though a significant portion of EAEC binding to human colonoids is dependent on functional mucus (or surface carbohydrates that are sensitive to these agents), but that only some of that adherence can be directly attributed to MUC2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. : EAEC adherence is dependent upon host intestinal mucus.

(A) Experimental design explaining the chemical treatment performed to manipulate intestinal mucus and (B) showing increasing EAEC aggregative adherence to HIEMs with increased mucus production as induced by DAPT (a gamma secretase inhibitor). (C) EAEC clusters on HIEMs in the presence of N-Acetyl Cysteine (NAC), a mucolytic agent. (D-E) Quantification of EAEC aggregates in response to increasing doses of DAPT and NAC was quantified from average of 12 images taken at 100x and with field size of 2560x1920 pixel. Tukey’s multiple comparison test was performed, error bars denote SED and *** indicates significantly different (P<0.001). (F) shows representative confocal micrographs of HIEMs stained with fluorescent indicators for nuclei (DAPI, blue), MUC2 (Red), E-cadherin (Purple) and EAEC (GFP, green). HIEMs were infected with EAEC A2A:GFP that were treated with DAPT (10 μM) and NAC (3 mM). Representative of over 50 images. Scale bars indicate 10μm.

https://doi.org/10.1371/journal.ppat.1008851.g005

EAEC aggregation and association with human intestinal enteroids occurs through sulfated glycans

The finding that EAEC adherence was dependent on aggregative adherence fimbriae association with host mucus prompted a search for the glycan receptor that drives this process. We cloned DNA encoding amino acids 1–135 of aggregative adherence fimbriae A (AafA) separated by a 10-residue linker into pQE-30 and expressed the polypeptide as a hexa-his fusion protein, and purified AafA using Ni-His agarose affinity chromatography. As shown in S6A Fig, AafA purified largely as a single, full-length polypeptide that was greater than 95% pure as assessed by SDS-PAGE. To determine the glycan (s) that may interact with AafA, we screened the purified protein for binding to a synthetic glycan array comprising >500 glycans [47]. This microarray includes many synthetic glycan structures from glycoproteins and glycolipids, but not complex glycosaminoglycans (GAG). Interestingly, while no binding of AafA was observed at low concentrations (S6B Fig), there was some binding events above what is considered background (< 2,000 relative fluorescence units (RFU), background RFU that contributes to non-specific binding). While no clear binding specificity towards certain structures was observed, AafA seemed to show binding preference towards “negatively charged” glycan structures in this analysis (S6B and S6C Fig). Thus, a second array consisting of charged glycosaminoglycans (GAG), including oligomers from hyaluronic acid, chondroitin sulfates and heparan, were tested for binding to AafA. Strikingly, AafA showed strong and specific binding (10–50 times more binding) for compounds that consisted of oligomers of heparan. Heparan is a highly sulfated GAG polysaccharide sharing basic structures with heparan sulfates, the GAG chain attached to heparan sulfated proteoglycans (HSPGs–Fig 6A). Fig 6B demonstrates the core structure of HSPGs consisting of Glucuronic/iduronic acid (GlcA) and N-acetylyglucosamine (GlcNAc). To determine if HSPGs could be suitable ligands for AafA, we modeled the heparan sulfate analog N, O6-DISULFO-GLUCOSAMINE (SGN) onto the published NMR structure of AafA (Fig 6C) using UCSF Chimera. Indeed, a PATCHDOCK analysis predicted a potential site for the interaction of heparan sulfate with AafA based on electrostatic interactions [48–50], mediated by the residues H62, N61 and Q64 (Fig 6C, inset). To determine if indeed HSPGs represented the receptor for EAEC’s interaction with colonoids, we pretreated the monolayers for two hours with the enzyme heparinase. As shown in Fig 6D and 6E, whereas untreated colonoids retained the characteristic biofilm-like aggregation and binding of EAEC, treatment with heparinase substantially reduced association with EAEC. Similarly, adding soluble heparin to the infection completely abolished any association of EAEC with the colonoids (Fig 6F), presumably via competitive inhibition with HSPGs present on the HIE surface. The results were confirmed after being quantified by assessing the number of bacterial colony forming units (CFUs–S6D Fig) before and after treatment. No effect on bacterial viability was observed with this concentration of heparin. These data strongly suggest that EAEC interacts with human intestinal enteroids via the direct binding of aggregative adherence fimbriae to heparan sulfated proteoglycans.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. HSPG mediates adherence of EAEC to HIEMs.

(A) GAG microarray of AafA shows high affinity binding to heparan sulfate GAGs. (B) shows the core chemical structure of HSPG. (C) Docking of SGN molecule on AafA reveals electrostatic interactions mediated by the residues H62, N61 and Q64. (D-F) Removal of HSPG by pretreatment of HIEs with heparinase III or addition of Heparan Sodium Salt (HSS) to EAEC as a competitive inhibitor exhibited reduced EAEC adherence on HIEs. Representative images, scale bars indicate 10μm.

https://doi.org/10.1371/journal.ppat.1008851.g006

Bacterial strains

The bacterial strains used were: EAEC 042 (serotype 044:H18) [39], EAEC 042: ΔaafA, EAEC, E. coli: HS, EHEC 8624 [40], ExPEC CP9 (serotype O4:K53:H5) [41, 42] and EAEC A2A (a clinical isolate). The list and the sources of all strains used are provided in Table 1. Bacteria were grown in Tryptic Soy Broth at 37°C overnight and sub-cultured for at least 2 hours before every infection. All infections are performed at a multiplicity of infection (MOI) of 10 and a total infection time of 3 hours unless otherwise indicated.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Details of E. coli strains used in the study.

https://doi.org/10.1371/journal.ppat.1008851.t001

Generation of EAEC: GFP

EAEC A2A strain was genetically modified by insertion of a chloramphenicol resistance and superfolder green fluorescent protein (sfGFP) gene by classical P1 phage transduction [96].The construct was generated by the laboratory of Dr. Christophe Herman originally in E. coli K12 (MG1655). Modifications (sfGFP:Cat:FRT) were transformed into EAEC using P1 phage and integrated at the nfsA locus. Transformants were selected on Chloramphenicol plates. Selected transformants were assessed for fluorescence and confirmed using oligos for upstream and downstream of the nfsA site (FWD: GGTGGTTATTCTTCAGGTG and REV: CGCTCCAGGTATCCC).

Ethics statement

All human sample/tissue collection was performed in accordance with the Baylor College of Medicine Institutional Review Board regulations and were collected under protocol H-35094 with informed consent from all adult subjects participating in the study.

Human intestinal enteroid culture

Human Intestinal Enteroid Monolayers (HIEMs) were derived and cultured from 3D enteroid cultures of D103,D104, D109, J2, J3, J11, I103, I104, I109, TC202, C103, C104 and C109, as previously described [35, 97]. The apical-out colonoids cultures were generated as previously described [46].

Global gene expression

Infections were performed on enteroid monolayers and lines as described above after being differentiated for five days. After infection, host cells were lysed using Trizol and RNA extracted using previously published protocols [56].

Library preparation, construction and sequencing

mRNA was enriched using oligo (dT) beads, fragmented randomly in fragmentation buffer, followed by cDNA synthesis using random hexamers and reverse transcriptase. After first-strand synthesis, a custom second-strand synthesis buffer (Illumina) was added with dNTPs, RNase H and Escherichia coli polymerase I to generate the second strand by nick-translation. The final cDNA library was ready after a round of purification, terminal repair, A-tailing, ligation of sequencing adapters, size selection and PCR enrichment. All processing steps were performed by Novogene, USA.

RNA-Sequencing analysis, quality control, and results

Fastq files were first trimmed to remove adapters and base-pairs that did not meet quality standards using Trim Galore (Version 0.4.1) [98, 99]. Trimmed sequences are aligned against the human genome hg38 (GRCh38) genome using Hisat2 (version 2.1.0) and sorted using samtools (version 1.5) [100, 101]. Next, a count matrix was generated by mapping reads using featureCounts (version 1.6.0) [102], and only non-overlapping uniquely mapped features were counted.

Expression quantification and gene set enrichment analysis

Differential gene expression was determined using the R-packages edgeR, limma and voom following published guidelines [103, 104]. Briefly, the feature count matrix was filtered for the coding genes only, and then a cutoff of 1 count per million was used to remove genes with low expression. The filtered counts are normalized using the trimmed mean of M-values (TMM) approach. The R-package voom was then used to remove the dependence of the variance on the mean gene expression, before determining differential gene expression using the limma R-package. The Benjamini-Hochberg (BH) correction was applied to the p-values to account for multiple hypothesis testing (adjusted p-values). Genes are ranked by their log2 fold gene expression for use in gene set enrichment analysis (GSEA version 3.0). GSEA was performed using these ranked gene lists against the entire MSigDB database (version 6.1) [103, 104].

Principal component analysis (PCA)

PCA is a procedure that uses orthogonal transformations to reduce the number of variables. The principal components are sorted by decreasing amounts of explained variability in the data. Plotting the transcriptome profiles based on the first three principal components is used to reveal relationship between samples. Hierarchical clustering of samples was performed using the Pearson correlation as distance metric, based on used log2CPM transformed and filtered data, after correction for batch and patient variation.

Preparation of coverslips for HIE monolayer culture for SEM

12 mm diameters round #1.5 coverslips were modified with 3-(trimethoxysilyl) propyl methacrylate (TMSPMA) to create a surface favorable for collagen IV adsorption. To clean and activate the surface of the coverslips, they are treated with piranha solution (3 parts sulfuric acid to 1-part hydrogen peroxide) for 1 hour. After washing with distilled water and 95% ethanol, the coverslips are incubated overnight at room temperature with a 2% solution of TMSPMA in 95% ethanol pH 4.5. Next, they are washed with 95% ethanol, dried with a laboratory wipe, and baked at 90°C for 1 hour. To culture HIEs on top of the modified coverslips, temporary wells are formed from 3 mm thick sheets of silicone rubber. Circular wells with a 6 mm inner diameter and 10 mm outer diameter are created with hand punches. These wells are sterilized by autoclaving and applied to the coverslips with tweezers to form a watertight seal. A dilute collagen IV solution was added to these wells for 90 minutes at 37°C, and then HIEs are seeded into them.

Scanning electron microscopy (SEM) and transmission electron microscopy

HIEMs are plated either on TMSPMA coated 12 mm diameters round #1.5 coverslips (Fisher scientific, Cat # NC1129240) or 96-well Nunc cell culture plates (Corning, Cat# CLS3595) for SEM and TEM imaging respectively. After 4 days of differentiation, HIEMs are infected with EAEC A2A at a MOI of 10 for 3 hours. For SEM, after post-infection, HIEMs are fixed with glutaraldehyde and dehydrated gradually in a graded series of ETOH- 30%, 50%, 70%, 95%, and 2x 100% for 15–30 minutes. The specimens are then progressively substituted for ethanol, followed by complete evaporation of ethanol and coated with gold to be observed under SEM. For TEM, after infection, HIEMs are trypsinized, pelleted in PBS and fixed with glutaraldehyde and dehydrated in graded ethanol. The specimens are then progressively substituted for ethanol, followed by complete evaporation of ethanol and coated with gold to be observed under TEM.

Histology staining and Periodic Acid Schiff/ Alcian Blue staining

HIEMs cultured on Transwell and at 4 days of differentiation are infected with EAEC at a MOI of 10 and incubated for 3 hours at 37°C in the presence of 5% CO₂ in a humidified incubator. The cells are washed three times with PBS gently to remove any non-adherent bacteria. The cells are fixed with 4% paraformaldehyde (PFA), progressively dehydrated in ethanol in a graded series of 30%, 50%, 70%, 95%, and 100% for 15–30 minutes at room temperature, embedded in paraffin and serially sectioned to 5 μm sections. Periodic acid Schiff (PAS) and Alcian blue (AB, pH2.5) was used to investigate the presence of goblet cell secreted mucins in EAEC clusters. All goblet cell mucus are labeled in light blue, nuclei in red to pink and cytoplasm in pale pink. Paraffin embedded sections are deparaffinized, oxidized in periodic acid and rinsed in water. Later they are then stained with Schiff’s reagent, rinsed with water, followed by 1% Alcian Blue (AB) staining in 3% acetic acid at a pH 2.5 and rinsed with water. Sections are counterstained with hematoxylin, washed, dehydrated and mounted for visualization under light microscopy.

Immunofluorescence staining

To identify association of EAEC aggregative adherence to MUC2, GFP-labelled EAEC: A2A was used. HIEMs plated on 96 well iBidi μ plates were fixed with Clarks fixative and incubated for 10 minutes [45]. HIEMs were permeabilized and blocked with 5% bovine serum albumin (BSA) in 0.1% Triton X-100 in PBS for 30 min at RT. MUC2 and cell boundaries were detected using the following antibodies: MUC2 (1:200) (Abcam: ab 11197) and Alexa Fluor E-cadherin (1:200) (BD Bioscience 560062) after overnight incubation at 4°C. Nuclei were stained with 4′, 6′-diamidino-2-phenylindole (DAPI) (300 nM) for 5 minutes at RT. Orthogonal sections of the sample were captured using a Zeiss LSM 510 confocal microscope.

Giemsa wright staining

HIEMs cultured on chambered slides and at 4 days of differentiation were infected at a MOI of 10 and incubated for 3 hours at 37°C in the presence of 5% CO₂ in a humidified incubator. The cells were washed three times with PBS to remove any non-adherent bacteria. The cells were then fixed and stained using Hema fixative and solutions (Protocol, Cat#122–91) and imaged under oil immersion microscopy at 100x.

AafA expression and purification

AafA is the major subunit of the AAF/II pilus which is assembled by the FGL (long looped) chaperone/usher mechanism [105, 106]. During pilus polymerization, the flexible N-terminal extension (donor strand) of AafA binds to the hydrophobic groove of the adjacent subunit to complete the Ig-like fold, referred to as donor-strand exchange [107–110]. To generate a donor-strand complemented AafA expressed in the cytoplasm, the insert consisted of a 2-residue linker followed by residues 11–135 at the N-terminus, a ten-residue linker, and residues 1–10 at the C-terminus ligated into vector pQE-30 (Qiagen) using restriction sites BamHI and HindIII. The AafA sequence [108] was obtained from (GenBank: AAB82330.1), and the signal peptide was removed to express and retain the protein in the cytoplasm. Constructs with this rearrangement of residues and the incorporated linker allow for conformational flexibility between the N and C-terminus and have been shown to produce soluble and stable donor strand complemented AafA monomers [107–110]. To separate the His6 tag and also facilitate the flexibility, a short Gly-Ser linker was also incorporated at the N-terminus (atg cat cac cat cac cat cac ggt agt aat ttt tgt gat ata acg ata aca ccg gct aca aat cgt gat gtc aac gtt gac agg agc gca aat atc gac ctg agt ttt act att aga caa ccg caa cgc tgc gct gat gct ggt atg cga ata aaa gct tgg ggg gaa gcc aat cac ggt caa tta ctg ata aaa cct caa gga gga aat aaa tca gca gga ttc act ctg gcc tct cct agg ttt tct tac att ccg aat aat cca gca aac att atg aat gga ttt gtt ctt acg aat cct ggt gtt tat caa tta gga atg cag ggc tca att aca ccg gct atc ccg cta cga cca ggc cta tat gaa gta gta tta aat gct gag ctt gtg aca aat gat aac aag caa aat gca act gcg gta gca aaa act gcg acc agt act atc act gta gtg taa)His6-Gly-Ser-AafA was expressed in XL1-Blue Supercompetent Cells (New England Biolabs) which allows tight regulation of potentially toxic genes by laclq. Cells were induced with 1mM IPTG at an OD600 of 0.6 for 4 hours, harvested by centrifugation at 4000 x g for 20 minutes, and the pellet was stored overnight at -20°C. The next day the pellet was suspended in lysis buffer (100 mM NaH2PO4 10 mM Tris·Cl 8 M urea, pH 8.0) and French pressed. AafA was purified by Ni-NTA (Qiagen) according to manufacturer’s instructions.

Glycan microarray preparation and analysis

Oligosaccharide array composed of 560 defined glycans from Consortium for Functional Glycomics were from MicroArray Core Facility of the Scripps Research Institute (La Jolla, CA) [47]. Glycans were printed on Nexterion 3-D hydrogel coating (H) slides using a contact printer in replicates of six described previously (1). A second microarray consisting of 170 charged glycosaminoglycans (GAG), including oligomers from hyaluronic acid, chondroitin sulfates and heparan, were from Emory Comprehensive Glycomics Core (Atlanta, GA). Glycans were conjugated with bifunctional linker AEAB and printed on Nexterion 3-D hydrogel coating (H) slides using a contact printer Aushon 2470 (Quanterix, Billerica, MA) in replicates of four. His-tagged AafA was diluted in PBS buffer with 1% BSA and 0.05% Tween 20 and incubated on the microarray slides for 1 hour at room temperature. Slides were washed with PBS buffer with 0.05% Tween 20 and PBS buffer. Alexa Flour 647 anti-His-tag mouse mAb (MBL International, Woburn, MA) was diluted to 5 μg/ml and incubated on the slides for 1 hour. After washing and drying, slides were scanned with InnoScan 1100 AL scanner and data were processed using Mapix 8.2.5 software (Innopsys, Chicago, IL). Mean Relative Fluorescence Units (RFU) from replicate spots after subtracting local median background were averaged, and standard deviation (STDEV) were calculated and plotted as Error Bar in the histogram plot of glycan array result.

CFU counting

After infection, the plate was tipped over for gentle removal of the media (to prevent removal of the mucus layer). Cells were washed gently with RT PBS 3x to remove un-adhered bacteria and each time; the plate was tipped to remove the PBS. 100ul cold PBS was added to each well and cells were scraped off with a P1000 tip to recover adhered bacteria. This solution was serially diluted to 10^-7 in PBS and plated on LB agar. Colonies were counted the following day for each dilution and CFUs per monolayer calculated.

Statistical analysis

A statistical significance was determined by two-way ANOVA. Tukey’s multiple comparison tests are performed using GraphPad Prism version 7.0 Windows (Graph Pad Software, San Diego, CA, www.graphpad.com). Differences in the means are considered significant at *p≤0.05 with specific p values detailed in the figure legends. The model figures were created with BioRender.com.