Systems-level characterization of murine colitis

To create a tissue-level systems model of chronic colitis, we utilized the T-cell transfer (TCT) model of colitis, which allowed us to control the timing and severity of inflammation in mice [8]. In this model, naïve T cells isolated from wild-type (WT) mice are injected into Rag1 null mice, which lack an adaptive immune system. In the absence of regulatory T cells (Tregs), the naïve T cells produce an inflammatory response to resident microflora, resulting in colitis with high penetrance (80%–90%) and a latency of 3–15 weeks. Our initial experimental population included 20 Rag1 null mice injected with naïve T cells and, as a negative control, 8 Rag1 null mice injected with Tregs (S11 Fig). Mice were weighed biweekly and assessed for symptoms of inflammation, such as diarrhea and rectal prolapse. After animals began to show severe weight loss and diarrhea, they were killed randomly over a range of time points to capture a spectrum of inflammatory states, an important feature for subsequent mathematical modeling. After humane killing, colons from individual mice were subdivided into matched sections for histology, immunophenotyping by flow cytometry, and multiplexed, Luminex-based protein measurement (S1 Fig).

Naïve T-injected animals exhibited significant changes to the intestinal mucosa, including epithelial hyperplasia, expansion of the proliferative zone, loss of differentiation, and increased immune infiltrate (Fig 1A and S2 Fig). The majority (18/20) of the naïve T cell-injected animals showed some degree of crypt hyperplasia and immune cell infiltration, which were highly correlated to one another and with the amount of weight lost (Fig 1B–1D). The increased height of the colonic crypts was due to the dramatic increase in the number of epithelial cells per crypt (S2 Fig). Immunohistochemistry (S2 Fig) and immunophenotyping (S3 Fig) revealed that many immune cell types were present in increased numbers in the colons of animals with colitis. For example, in Rag1 null control animals, CD45+ cells made up only a small proportion of total colonic cells (0.1%–0.3%), and the majority of these cells were plasmacytoid dendritic cells (pDCs). In naïve T-injected animals, the percentage of CD45+ cells showed a strong correlation with crypt hyperplasia (Fig 1B), and this infiltrate was composed primarily of CD4+ T cells, macrophages, and neutrophils (S3 Fig), although the relative proportions of these cell types varied from animal to animal.

Download:

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

Fig 1. Phenotypic characterization of murine colitis.

(A) Representative histology of colons from control regulatory T cell (Treg)-injected animals and naïve T-injected animals with inflammation. The left panels show hematoxylin–eosin (HE)-stained colons; the center panels show immunohistochemistry for Ki-67, a marker of undifferentiated cells; and the right panels show immunohistochemistry for mucin-2, a marker for goblet cells. (B) Plot of average crypt height in distal colon versus the percentage of total CD45+ immune cells in the tissue, showing a strong positive relationship. R was determined by Spearman correlation. (C) Plot of average crypt height in distal colon versus weight loss score, showing a strong linear relationship. R was determined by Pearson correlation. (D) Plot of the percentage of total CD45+ immune cells in the colon versus weight loss score, showing a strong linear relationship. R was determined by Pearson correlation. (E) Scatter plots of crypt height profiles for the animals from panel A. Each dot represents a single crypt measurement and its location within the colon from distal to proximal colon. Focal and continuous inflammation classes were defined based on these profiles, with focal inflammation showing the variable phenotype (blue dots) and continuous inflammation showing consistent distal colon hyperplasia (red dots). The box at the bottom provides the immune infiltrate score at each location where crypts were measured, showing co-occurrence of hyperplasia and foci of immune infiltration. (F) Average crypt height profiles for each group, with the standard error about the mean indicated in shaded regions at points binned in 500-μm increments. N = 6, 7, and 8 for control, focal inflammation, and continuous inflammation, respectively. Underlying numerical values for panels B–F are provided in S1 Data.

https://doi.org/10.1371/journal.pbio.2002417.g001

In the course of phenotyping animals with colitis, we noted that the histological manifestations of colitis were not uniform throughout the colon but rather restricted to specific colonic regions. In order to quantitatively measure inflammation as a function of the geographical position in the colon, we measured the thickness of the mucosa every 5 crypts along the entire length of the colon for every mouse. These measurements were used to generate scatter plots that graph crypt height as a function of longitudinal location along the colon (Fig 1E). In addition to crypt height measurements, an immune infiltration score was used to identify and locate areas of inflammation. Foci of infiltrating immune cells were rarely found in control animals but were found throughout the colons of naïve T-injected animals. These infiltrates were more prevalent in continuously inflamed colons and were closely associated with areas of crypt hyperplasia (Fig 1E). After analyzing our entire experimental cohort, we found that we could divide experimental animals into 2 discrete classes based on the extent of colitis (Fig 1F). Continuously inflamed animals exhibited significant crypt hyperplasia (300–600 μm/crypt) throughout the distal colon, while focally inflamed animals were defined by variable crypt hyperplasia beginning at the proximal/distal junction and spreading distally towards the rectum. While animals that were killed at later time points had a higher prevalence of colitis, the presence and severity of colitis were most closely associated with weight loss and the time since initial weight loss (Fig 1C and 1D). In follow-up experiments, by using this weight loss metric to select the timing of drug treatments, we were able to ensure that we treated only animals with established continuous inflammation. Together, these data establish the quantitative and qualitative histological features of chronic colitis, defining crypt hyperplasia as a tissue-level feature of colitis in this model and identifying expansion of the proliferative zone and loss of differentiation as key cellular features of disease.

Having established quantitative phenotypic measures of colitis, we sought to identify the inter- and intracellular signals underlying inflammation. Fresh tissue from the distal colon was lysed for Luminex-based analysis of 37 cytokines, chemokines, and growth factors and 17 phosphoproteins marking key signaling pathways (S4–S6 Figs). These 54 “signals” represented all of the commercially available Luminex-based assays for which we were able to detect signal above background in the mouse colon. Many of the signals were significantly different in control animals and those receiving adoptive transfer of CD45RBhi T cells, and these changes were not simply due to the adoptive transfer, since animals that received CD45RBhi T cells did not exhibit increased signaling prior to the onset of colitis (S7 Fig).

We used principal component analysis (PCA) to cluster animals based on protein expression profiles and to identify key proteins driving this clustering. PCA showed a high degree of covariance between the proteins in the data set, and the majority of this variance was explained by the first principal component (PC1) (Fig 2A). Additionally, PC1 clustered each animal based on its degree of inflammation, with animals exhibiting continuous inflammation scoring positively on PC1, the focal inflammation animals scoring in the middle, and the control animals scoring negatively. Two of the naïve T cell-injected animals clustered with the negative control animals; however, these showed almost no hyperplasia or immune infiltration, representing the incomplete penetrance of this model (Fig 2A). Scores on PC1 were plotted against average crypt height and showed a strong linear correlation (R² = 0.80), confirming that this principal component is strongly associated with inflammation (Fig 2B). PC1 showed strong positive loading for chemokines involved in macrophage, neutrophil, and T-cell chemotaxis (macrophage inflammatory proteins [MIPs], KC [keratinocyte chemoattractant], monocyte chemoattractant protein 1 [MCP-1], and RANTES [regulated on activation, normal T cell expressed and secreted]), inflammatory cytokines (interleukin [IL]-1s and IL-6), and growth factors (leukemia inhibitory factor [LIF] and vascular endothelial growth factor [VEGF]) (Fig 2C). T helper cell 2 (Th2) cytokines (IL-31 and IL-33) did not show positive loading, suggesting polarization of the T-cell response towards a Th1/Th17-type response, which is typical of CD and the TCT model of IBD [9]. Intracellular signaling molecules, such as Mek and the insulin receptor, showed strong negative loading, suggesting that these pathways are down-regulated in inflamed colons. The main intracellular signaling pathways up-regulated in inflamed animals included NF-κB and mTOR. We were particularly interested in the mTOR pathway because it was activated at multiple levels, including Akt, p70 S6K, and S6RP. The activation of mTOR in the context of colitis was not specific to the TCT model; we found that mTOR was activated in multiple genetic models of IBD (S8 Fig) and in a subset of human IBD patients (S9 Fig).

Download:

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

Fig 2. Multivariate molecular analysis of murine colitis.

(A) Principal component analysis (PCA) based on Luminex measurements from distal colons identifies clusters of the control, focal inflammation, and continuous inflammation classes on principal component 1 (PC1). (B) Correlation between PCA and crypt height. For each animal, the score on PC1 was plotted against the average crypt height in the distal colon. This indicates that there is a strong correlation (R² = 0.8) between PC1 and epithelial measures of inflammation. (C) Loadings on PC1 represented as single dots for each protein and grouped and color-coded by pathway or category. Signals on the right are positively correlated with inflammation. (D) Immunohistochemistry for p-S6 ribosomal protein (p-S6RP), a readout of the mammalian target of rapamycin (mTOR) pathway. p-S6RP was strongly induced in inflamed animals and was found primarily in epithelial cells. Underlying numerical values for panels A and B are provided in S1 Data. MAPK, mitogen-activated protein kinase.

https://doi.org/10.1371/journal.pbio.2002417.g002

To determine whether mTOR signaling was generally up-regulated in the context of colitis or whether up-regulation in a specific cellular compartment accounted for the increased signal detected via Luminex, we performed immunohistochemistry for phosphorylated ribosomal protein S6 (p-S6) on colons from control and adoptive transfer mice. In control animals, we found high p-S6 signal only in intraepithelial lymphocytes, which is consistent with the well-characterized role for mTOR signaling in immune function (Fig 2D) [10]. In animals with colitis, by contrast, we detected strong p-S6 throughout the colonic epithelium (Fig 2D), suggesting that the increased mTOR signal that we detected derives primarily from ectopic activation in epithelial cells.

mTOR activation suppresses colonic differentiation to promote inflammation

In order to investigate a potential role for mTOR in driving colonic inflammation by affecting epithelial homeostasis, we tested the effects of the mTOR inhibitor rapamycin on colitis induced in the TCT model. To ensure experimental animals had severe inflammation, we utilized animals at late time points after adoptive transfer. We also used weight loss as a marker for onset of inflammation, beginning treatment on a mouse-by-mouse basis when an animal lost more than 1 gram of weight and did not regain the weight within 1 week. Inflamed animals were treated with rapamycin (5 mg/kg/day) for 2 weeks and then processed and analyzed as described above. Colons from rapamycin-treated animals exhibited a clear reduction in immune infiltrate, normalization of the proliferative zone at the bottom of the crypts, and restoration of goblet cell differentiation (Fig 3A and 3B). Quantification of crypt heights revealed a partial normalization of mucosal thickness and a significant decrease in average crypt height in rapamycin-treated animals (Fig 3C). Because rapamycin promotes expansion of Tregs in other systems [11], we explored whether the general reduction in inflammation following rapamycin treatment was associated with an increase in Tregs in the colonic lamina propria. While animals with colitis did have significant numbers of Tregs in the colon, those treated with rapamycin did not have greater numbers (Fig 3D)—in fact, they had fewer—indicating that rapamycin treatment did not suppress colitis in the TCT model by expanding the number of Tregs.

Download:

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

Fig 3. Mammalian target of rapamycin (mTOR) inhibition reduces inflammation.

(A) Representative histology of colons from vehicle (Veh.)-treated control animals, rapamycin (Rap.)-treated control animals, vehicle-treated and naïve T-injected animals, and rapamycin-treated and naïve T-injected animals. Individual panels show (from top to bottom) hematoxylin–eosin (HE)-stained colons, Ki-67, mucin-2, and Foxp3, a marker of regulatory T cells (Tregs). (B) Measurement of proliferative zone heights via Ki-67 staining. There is a significant reduction in crypt height for inflamed rapamycin-treated animals, as determined by Mann-Whitney test. *** represents p ≤ 0.001. N = 100, 60, 80, and 120 for control + vehicle, control + rapamycin, naïve T + vehicle, and naïve T + rapamycin, respectively. (C) Average crypt height profiles for each treatment group, with the standard error about the mean indicated in shaded regions at points binned in 500-μm increments. N = 4, 3, 6, and 8 for control + vehicle, control + rapamycin, naïve T + vehicle, and naïve T + rapamycin, respectively. (D) Quantification of Foxp3-positive Tregs. There is a significant reduction in Tregs in the lamina propria of all animals treated with rapamycin. Tregs were quantified as the number of Foxp3-positive cells in a random high-powered field (HPF). P-values were determined by Mann-Whitney test. *** represents p ≤ 0.001. N = 75, 50, 100, and 75 for control + vehicle, control + rapamycin, naïve T + vehicle, and naïve T + rapamycin, respectively. (E) Luminex-based measurements of activating phosphorylation sites on multiple proteins in the mTOR pathway. Measurements are in arbitrary relative fluorescence units (RFUs). Horizontal and vertical lines represent mean +/− standard error. Significance was determined by one-way ANOVA with Tukey-Kramer post-test. * represents p ≤ 0.05; ** represents p ≤ 0.01. (F) Projection of signaling data from vehicle- and rapamycin-treated samples onto a partial least squares regression (PLSR) model built from untreated samples. Original samples used to generate the model project into the gray oval (control), blue oval (focal inflammation), and red oval (continuous inflammation). Underlying numerical values for panels B–F are provided in S1 Data.

https://doi.org/10.1371/journal.pbio.2002417.g003

Analysis of phosphoproteins confirmed that signaling downstream of mTOR activity was strongly inhibited by rapamycin (Fig 3E), although phosphorylation of the upstream activator Akt appeared to be increased by rapamycin, presumably by relieving inhibition of the insulin-like growth factor receptor/insulin receptor substrate 1 (IRS1) axis [12]. We next sought to determine whether mTOR inhibition altered the local intracellular signaling network or whether it had a broad effect on the tissue-level inflammatory network (S4–S6 Figs). We first applied partial least squares regression (PLSR) to signaling data from our initial specimens in order to create a mathematical model linking dysregulated signaling to colitis, regressing to epithelial thickness. As with our original PCA model, the PLSR model was able to segregate animals with continuous inflammation from control animals and those with focal inflammation based on the accumulated signals that were measured in colonic lysates (Fig 3F). When the signaling data from animals treated with rapamycin were projected onto this PLSR model, 6 out of 8 rapamycin-treated animals clustered with the focally inflamed animals (Fig 3F). This reflects the fact that rapamycin treatment represses signaling directly downstream of mTOR and also the majority of the cytokines, chemokines, and growth factors associated with inflammation. Together, these results suggest that epithelial mTOR is a central regulator of intestinal inflammation in mouse models of IBD.

While long-term rapamycin treatment revealed a key role for mTOR in the maintenance of colitis, chronic inhibition does not provide mechanistic insight into the potential role that epithelial mTOR plays in this complex, tissue-level phenotype. In order to address this limitation, we treated inflamed animals acutely with rapamycin for 1, 4, 8, 24, or 48 hours. Within 1 hour of rapamycin treatment, goblet cells began to appear, and the proliferative zone began to normalize (Fig 4A and 4B). Within 24 hours, these cellular phenotypes were returned to nearly normal levels. By 48 hours, the inflammation-induced hyperplasia was reduced, reflecting the induced differentiation and decreased proliferation induced by mTOR inhibition (Fig 4C). Immune cell infiltrate changed more slowly following rapamycin treatment than did the epithelium, with a decrease in the proportion of macrophages and neutrophils only by 48 hours (Fig 4D). At that time point, the proportion of CD45+ cells that were T cells was significantly increased (Fig 4D). Cytokine and chemokine expression showed a variable kinetic response to rapamycin (S10 Fig). Some proteins decreased rapidly, within 1 hour of rapamycin treatment, to control levels, while others showed a graded decrease correlating to the goblet cell phenotype (Fig 4E). Some of the cytokines, like IL-6, even increased following rapamycin treatment (Fig 4E). These data are consistent with a kinetic model in which hyperactivated mTOR suppresses differentiation to promote colitis. Upon mTOR inhibition with rapamycin, colonic epithelial cells rapidly differentiate and cease to proliferate. Subsequently, the cytokine profile of the tissue changes, and, as a consequence, the complexion of the innate immune landscape returns to a state that resembles the normal colon.

Download:

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

Fig 4. Temporal control of cellular and molecular inflammatory phenotypes by rapamycin.

(A) Representative Alcian Blue/Periodic Acid Schiff (AB/PAS) stains for inflamed animals treated with vehicle (Veh.) or rapamycin (Rap.) for the indicated periods of time. Intracellular and extracellular mucins are stained dark purple and indicate the appearance of mature goblet cells. (B) Measurement of the proliferative zone before and after rapamycin. The proliferative zone is defined as the ratio of the highest phosphorylated histone H3 (PH3)-positive cell position in a crypt divided by the height of that crypt. For each animal, 12–25 crypts were measured, and bars represent the average proliferative zone for 2–3 animals. Significance was determined by one-way ANOVA with Tukey-Kramer post-test. * represents p ≤ 0.05; ** represents p ≤ 0.01. Bar graphs represent mean +/− standard error. (C) Average crypt height profiles for each treatment group, with the standard error about the mean indicated in shaded regions at points binned in 500-μm increments. N = at least 2 for each group. (D) Flow cytometric measurements of colonic immune infiltrate. Inflamed animals show an even mix of T cells, macrophages, and neutrophils, but rapamycin treatment leads to a loss of macrophages and neutrophils by 48 hours. Significance determined by ANOVA with Tukey-Kramer post-test. * represents p ≤ 0.05; ** represents p ≤ 0.01, both with respect to naïve T, vehicle control. N = 2–3 animals per condition. Bar graphs represent mean +/− standard error. (E) Representative cytokine expression patterns of response to acute rapamycin treatment. Macrophage colony-stimulating factor (M-CSF) drops immediately follow rapamycin exposure, while macrophage inflammatory protein 1-alpha (MIP-1α) declines slowly over 24 hours. Some proteins, like interleukin (IL)-6, transiently increase in expression following rapamycin treatment. N = 2–3 animals per condition. Bar graphs represent mean +/− standard error. Underlying numerical values for panels B–E are provided in S1 Data.

https://doi.org/10.1371/journal.pbio.2002417.g004

Rapamycin treatment does not affect the colonic microbiome

We next considered the possibility that rapamycin indirectly affected intestinal homeostasis by causing shifts in the composition of the commensal flora. To address this possibility, we profiled the fecal microbiome of control and inflamed animals, both before and after treatment with rapamycin (5 mg/kg/day for 2 weeks) (S11 Fig). Principal coordinate analysis (PCoA) of the beta diversity relationships between samples demonstrated that samples separated based on their inflammatory phenotype, but that rapamycin treatment did not shift the overall status of the microbiome, whether an animal had inflammation or not (S11 Fig). We also plotted the Bray-Curtis dissimilarity scores between all pairs of mice to see how individual mouse pre- and post-treatment scores compared with differences between animals. This analysis showed that the microbiome of control animals was most similar to other control animals, regardless of whether they received rapamycin (S11 Fig). The same was true for animals with inflammation. Although we did note that resolution of inflammation altered microbiome diversity and that the representation of certain bacteria, for example Lactobacilli, increased in inflamed mice after treatment, overall our data reveal that rapamycin itself does not alter microbial composition in this model.

Induction of differentiation reduces inflammation

In our initial hypothesis-generating experiment and in the subsequent perturbation experiments, we found an inverse correlation between epithelial differentiation and colonic inflammation. This observation led us to ask whether induction of differentiation independently of mTOR inhibition would have a similar effect on tissue homeostasis. To this end, we used the gamma-secretase inhibitor dibenzazepine (DBZ), which induces goblet cell differentiation by inhibiting Notch signaling [13]. Animals with colitis were treated with 10 μmol/kg DBZ or vehicle by daily intraperitoneal (IP) injection for 1 week. DBZ-treated animals exhibited complete conversion of the colonic epithelium to goblet cells, with high mucous levels in almost every cell in the crypt (Fig 5A). Consistent with rapamycin treatment experiments, induction of differentiation via Notch inhibition was accompanied by a reduction in macrophages and neutrophils (Fig 5B). Interestingly, there was no change in the proportion of T cells within the colon, suggesting that their recruitment and maintenance within the tissue is not affected by systemic inhibition of Notch. This result is consistent with our prior observation that rapamycin treatment rapidly induced differentiation prior to having any effect on the number of T cells in the colon (Fig 4D). The rescue of differentiation was also associated with a decrease in the inflammation-associated chemokine profile (Fig 5C and S12 and S13 Figs), particularly those associated with innate immune cell chemotaxis. When the DBZ- and vehicle-treated samples were projected onto a cytokine-specific PLSR model built from the baseline samples, DBZ-treated samples clustered with focally inflamed and negative control animals, just like the rapamycin-treated animals in our previous experiment (Fig 5D). These experiments demonstrate that 2 mechanistically independent methods of inducing differentiation in the colonic epithelium lead to a reduction in the expression of inflammatory cytokines and chemokines and subsequent reduction in inflammatory immune infiltrate. These observations provide strong evidence that the epithelium is a key player, beyond its role as a barrier, in the inflammatory process and that this connection is tightly linked to differentiation state.

Download:

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

Fig 5. Goblet cell differentiation leads to reduced cytokine expression and innate immune cell infiltration.

(A) Alcian Blue/Periodic Acid Schiff (AB/PAS) stain for mucin shows loss of goblet cells in inflamed animals that is restored by treatment with the Notch pathway inhibitor dibenzazepine (DBZ). (B) Immune infiltrate, as measured by flow cytometry expressed as a percentage of total cells, shows a reduction in macrophages and neutrophils following 1-week treatment with DBZ. Significance determined by unpaired t test, with p-values indicated above each cell type. N = 2–3 animals per condition. Bar graphs represent mean +/− standard error. (C) Inflammation-associated cytokine and chemokine expression was measured by Luminex and displayed as log2-fold changes versus control animals. These proteins represent the top 20 loadings from the partial least squares regression (PLSR) model described in Fig 5D. Data for individual animals are presented in S8 Fig. (D) DBZ- and vehicle (Veh.)-treated cytokine and chemokine measurements were projected onto a PLSR model built from non-drug-treated animals. DBZ-treated animals cluster with focal inflamed and noninflamed animals. Original samples used to generate the model project into the gray rectangle (control), blue rectangle (focal inflammation), and red rectangle (continuous inflammation). Underlying numerical values for panels B–D are provided in S1 Data. Avg, average; IFNγ, interferon gamma; IL, interleukin; LIF, leukemia inhibitory factor; LV1, latent variable 1; MCP-1, monocyte chemoattractant protein 1; MIG, mitokine induced by gamma interferon; MIP-1α/β, macrophage inflammatory protein 1-alpha/beta; M-CSF, macrophage colony-stimulating factor; RANTES, regulated on activation, normal T cell expressed and secreted; VEGF, vascular endothelial growth factor.

https://doi.org/10.1371/journal.pbio.2002417.g005

Enrichment of transit-amplifying cells in inflamed epithelium

Our studies indicate a clear role for the differentiation state of the epithelium in mediating cytokine and chemokine expression and innate immune recruitment in murine colitis. Nevertheless, the stage at which the epithelial cells were arrested remained unclear. In order to address this, we performed microarray analysis on 4 control and 6 inflamed colons (GEO accession GSE87317). Analysis of individual markers for different cell types suggested that there was a pronounced decrease of goblet cells and enteroendocrine cells and a moderate decrease of enterocytes and stem cells (Fig 6A and 6B), cellular phenotypes that we have confirmed by histologic analysis (Fig 1A and S2 Fig). Because transit amplifying cells (TACs) are defined functionally, the existing monovariate markers are not sufficient to assess their enrichment. In order to determine whether there was a relative increase in TACs in inflamed colon, we employed Gene Set Enrichment Analysis (GSEA) using published gene sets derived from epithelium enriched for various cell types. A stem cell gene set was derived from sorted Lgr5+ epithelial cells [14], while enterocyte and secretory progenitor gene sets were derived from various pharmacologic and genetic treatments [15]. Finally, a TAC signature was derived from regenerative epithelium composed of 97% TACs [16]. Using GSEA, we found de-enrichment for gene sets associated with secretory cells, enterocytes, and stem cells (Fig 6C and 6D). Conversely, we saw enrichment for genes associated with Atoh1 knockout (which lacks goblet cells) and the TAC signature. Importantly, several of the genes found in the TAC signature are chemokines, all of which were strongly induced transcriptionally in our microarray data set (Fig 6E). By contrast, there were no cytokines or chemokines present in any of the gene sets for stem or differentiated cell types. This observation confirms that cytokine expression is present within the epithelium itself and that this is enhanced in TACs, but not in stem cells or differentiated cells. Together, these analyses suggest that the suppressed differentiation and enhanced proliferation that is seen in colitis is the result of expansion of the TAC compartment and not the result of stem cell expansion.

Download:

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

Fig 6. Expansion of transit amplifying cells in animals with colitis.

(A) Markers of goblet cells, enterocytes (Ent.), enteroendocrine cells (EE), and stem cells measured by RNA microarray. Each gene was mean centered, and the means of inflamed and noninflamed groups are indicated in red and gray, respectively. Horizontal and vertical lines represent mean +/− standard error. Significance was determined by unpaired t test. * represents p ≤ 0.05, ** represents p ≤ 0.01, *** represents p ≤ 0.001. (B) Average fold change between inflamed and noninflamed groups for each of the genes in panel A. (C) Enrichment plots for experimentally derived gene sets for secretory progenitors (Secpro), stem cells, enterocytes (Ent), goblet cell-deficient epithelium (Atoh null), and transit-amplifying cell (TAC) epithelium. (D) Normalized enrichment scores for the gene sets indicated in panel C; false discovery rate (FDR) values are indicated above or below each bar. (E) Mean-centered expression values of the 4 chemokines found in the TAC gene set. The means of inflamed and noninflamed groups are indicated in red and gray, respectively. Horizontal and vertical lines represent mean +/− standard error. All genes showed a p < 0.001 in a t test, as indicated by ***. Underlying numerical values for panels A and C–E are provided in S1 Data.

https://doi.org/10.1371/journal.pbio.2002417.g006

Differentiation state of the epithelium directly controls cytokine expression

Although our in vivo results were consistent with a model in which undifferentiated epithelium supplies many of the cytokines needed to recruit inflammatory cells to the colonic lamina propria, these experiments did not rule out the possibility that these cytokine and chemokine changes are secondary effects originating from immune cells. In order to test whether the expression of proinflammatory signaling molecules is intrinsic to the epithelium, we measured the production of cytokines and chemokines as a function of epithelial differentiation state in an in vitro 3D organoid model. Crypts were isolated from WT mouse colons and developed into 3D organoids in matrigel following well-established protocols [17]. We then shifted the growth conditions to maintain a mixed cell population to enrich for proliferative stem cells or to differentiate into the enterocyte or goblet cell lineages (S14 Fig) [15]. Following confirmation of appropriate differentiation based on expression of cell type markers (Fig 7A), purified organoids were analyzed for expression of the suite of cytokines, chemokines, and growth factors that were measured in intact colons (S15 and S16 Figs). Relative to the stem cell-enriched and mixed population organoids, the majority of cytokines and chemokines were suppressed in organoids enriched for secretory and/or absorptive cells. For example, MCP-1 was reduced during differentiation in a lineage-independent manner, while VEGF and mitokine induced by gamma interferon (MIG) were reduced only when cells were differentiated into the goblet lineage (Fig 7B). Interestingly, MIP-1α, which was strongly induced in animals with inflammation (S5 Fig), was increased in expression when organoids were differentiated into the goblet lineage. Altogether, these data suggest that differentiation alters the cytokine and chemokine expression pattern of the colonic epithelium independent of immune cells.

Download:

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

Fig 7. Epithelial differentiation reduces inflammatory cytokine expression in colonic organoids.

(A) RNA markers of stem cells (Lgr5), enterocytes (Alpi), and goblet cells (Muc2) and enteroendocrine cells (Chga) were measured by quantitative polymerase chain reaction (qPCR) and normalized to standard culture condition organoid expression levels. Conditions for enrichment of stem cells or specific differentiated lineages can be found in the Materials and methods section. Horizontal and vertical lines represent mean +/− standard error. Significance was determined by ANOVA with Tukey-Kramer post-test. * represents p ≤ 0.05; *** represents p ≤ 0.001. (B) Representative cytokines, chemokines, and growth factors were measured by Luminex and represented in pg/ml. Horizontal and vertical lines represent mean +/− standard error. Significance was determined by ANOVA with Tukey-Kramer post-test. * represents p ≤ 0.05; ** represents p ≤ 0.01. (C) A partial least squares regression (PLSR) model built on cytokine, chemokine, and growth factor measurements from the mouse experiment described in Fig 1 with organoid protein expression data projected. Differentiated organoids project with focal inflammation and control groups, while proliferative, stem-like organoids are projected with the continuous inflammation group. Original samples used to generate the model project into the gray rectangle (control), blue rectangle (focal inflammation), and red rectangle (continuous inflammation). Underlying numerical values for panels A–C are provided in S1 Data. MCP-1, monocyte chemoattractant protein 1; MIG, mitokine induced by gamma interferon; MIP-1α, macrophage inflammatory protein 1-alpha.

https://doi.org/10.1371/journal.pbio.2002417.g007

To put the in vitro analysis of cytokine/chemokine expression into the context of our prior in vivo studies, we projected the organoid expression data onto the same PLSR model from Fig 5D, where expression of the cytokines and chemokines highlighted in Fig 5C specifies latent variable 1 (LV1). In this analysis, the more stem-like organoids were projected positively on LV1, clustering with the inflamed mouse colons (Fig 7C). Conversely, the absorptive- and secretory-differentiated organoids clustered with the focal and noninflamed colons. This analysis demonstrates that the patterns of cytokine and chemokine expression that correlate with loss of differentiation and immune cell infiltration in inflamed colons in vivo are largely recapitulated in an in vitro system that lacks both bacteria and immune cells. Together with the in vivo mouse experiments, this result indicates that the colonic epithelium plays an active role in inflammation that extends beyond its role in maintaining barrier integrity.

TCT

TCT was performed according to established methods [8]. WT and Rag1-deficient mice on the C57BL/6 background were used for TCT. These animals were purchased from the Jackson Laboratory (Bar Harbor, Maine, United States). Naïve T cells and Tregs were collected on the BD Aria sorter and injected at 400,000 and 200,000 cells/animal, respectively.

Ethics statement

All animal work was approved by the Institutional Care and Use Committees of Massachusetts General Hospital and Beth Israel Deaconess Medical Center under protocol numbers 2007N000058, 078–2014, and 080–2017. Approved protocols conformed to the USDA Animal Welfare Act, the PHS Policy on Humane Care and Use of Laboratory Animals, and the “ILAR Guide for the Care and Use of Laboratory Animals.”

Inhibitor treatments

Rapamycin was purchased from LC Laboratories (R-5000), and treatment was carried out for 1, 4, 8, 24, or 48 hours for the acute experiment and for 2 weeks for the extended experiment. Rapamycin was injected IP at 5 mg/kg in a vehicle of 5.2% Tween 80 and 5.2% PEG400 in water. DBZ was purchased from Sigma (SML0649-25MG) and used at 10 μmol/kg in water with 0.5% HPMC and 0.1% Tween 80.

Tissue processing and histology

Colons were removed and flushed with PBS, and one-fifth lengthwise was taken for flow cytometry. A matched fifth was cut into proximal and distal regions and lysed in 250 μl Bio-Plex lysis buffer with protease inhibitor, factors 1 and 2, and PMSF. The final piece was rolled and fixed overnight in 10% formalin for histology. For microarray tissue collection, one-fifth was snap frozen and processed as described below. This replaced flow cytometric measurement for 4 Treg and 6 inflamed animals from the initial study.

For crypt height measurements and gross histology, 5 μm sections of rolls were stained with hematoxylin and eosin (HE) according to standard protocols. Alcian Blue/Periodic Acid/Schiff was used from Leica (38016SS3A, 38016SS4A, and 38016SS4B), and goblet cells were visualized according to reagent instructions. Immunohistochemistry for phospho-Histone H3 S10 (Cell Signaling Technology 9701), liver fatty acid binding protein (Abcam ab7366), Ki-67 (Cell Signaling Technology 12202), chromogranin A (Abcam ab15160), and phospho-S6 ribosomal protein S235/236 (Cell Signaling Technology 4858) was carried out with citrate antigen retrieval and visualized using horseradish peroxidase reaction. For Foxp3 immunohistochemistry, antigen retrieval was performed in DAKO Target Retrieval Solution (#S1699) in a pressure cooker. Tissue sections were blocked with DAKO Serum-Free Protein Block (#X0909) and primary anti-E-cadherin (BD Medical Technology #BDB610181) and anti-FoxP3 (Abcam ab54501) antibodies were incubated overnight at 4°C in DAKO Antibody Diluent (#S3022). Following PBS washes, Alexa Fluor-labeled secondary antibodies (ThermoFisher) were incubated in DAKO Antibody Diluent for 1 hour at room temperature. Slides were mounted in ProLong mounting medium (ThermoFisher #P36962) and imaged on a Zeiss Axio Imager Z2.

Crypt height measurements were performed using an Olympus microscope or slide scanner software. Briefly, an arbitrary line was drawn from the crypt base to the top of the crypt every 5 crypts across the length of the colon. Crypt measurements in which multiple crypts were associated with a single line were deemed off axis and removed from subsequent analysis. For scatter plots, all measurements were evenly spaced across the distal and proximal regions separately using polyline measurements of the lengths of those sections. Averages were taken of all measurements in the distal and proximal regions separately. To calculate the proliferative zone, the distance from the crypt base to the highest PH3-positive cell or Ki-67-positive cell was divided by the total crypt height.

Flow cytometry

Tissue was homogenized in serum-free DMEM with 2 mg/ml collagenase type I C (VWR 234153-100MG) and incubated for 1 hour at 37°C. Following incubation, the sample was strained through a 45-μm filter and spun down for 5 minutes at 700 g. Cells were then stained with the following antibodies from BioLegend (1:300 in fluorescence-activated cell sorting [FACS] buffer) for 10 minutes: Alexa-488 CD326 (118210), BV-421 F4/80 (123131), BV-605 CD4 (100547), BV-510 cd11b (101245), Alexa-700 Ly6G (127622), APC CD25 (102012), PE/Cy7 cd11c (117317), APC/Cy7 CD45 (103116), and PE CD45RB (103308). Cells were analyzed on a 5-laser LSR II (Becton Dickson). Details of markers used to quantify individual cells types are presented in S2 Fig.

Protein measurements

Luminex-based protein measurements were carried out according to manufacturer’s instructions, as we have reported previously [34, 36, 37]. The following kits from Bio-Rad were used: Group I Cytokine Assay (M60-009RDPD), Group II 9-plex (MD0-00000EL), and Group III Th17 panel (171-FA001M). BioplexPro phosphoprotein measurements were carried out with the following groupings: 10-plex: Akt Ser473 (171-V50001M), c-Jun Ser63 (171-V50003M), CREB Ser133 (171-V50028M), Erk1/2 Thr202/Tyr204, Thr185/Tyr187 (171-V50006M), GSK-3 Ser21/Ser9 (171-V50007M), Jnk Thr183/Tyr185 (171-V50011M), Mek1 Ser217/Ser221 (171-V50012M), p38 MAPK Thr180/Tyr182 (171-V50014M), Stat3 Ser727 (171-V50021M), p90Rsk Ser380 (171-V50035M); single-plex: IR-1 Tyr1146 (171-V50031M), IRS-1 Ser636/Ser639 (171-V50030M), IκB Ser32/Ser36 (171-V50010M), p70S6K Thr389 (171-V50016M), S6RP Ser235/Ser236 (171-V50038M), and Atf-2 Thr71 (171-V50024M).

Colonic organoid culture

Crypt epithelial cells were isolated from mouse colons by incubating the tissue in 2 mg/ml type I collagenase (Invitrogen). Organoids were grown in matrigel in the presence of EGF, Noggin, R-spondin, and Wnt3a as previously described [17, 38]. During the first 4 days of culture, GSK3β inhibitor, CHIR99021 (3 μM, Stemgent), and histone deacetylase inhibitor and Notch agonist, VPA (1 mM, Sigma-Aldrich), were added to the medium to enrich for stem cells. Following this treatment, organoids were either maintained in this stem cell media, returned to standard growth media, or differentiated as previously described with the following minor modifications [17]. To increase enterocyte differentiation, organoids were grown in Wnt3a-deficient medium supplemented with VPA (1 mM) and the Wnt inhibitor IWP-2 (5μm, Stemgent). To promote goblet cell differentiation, Wnt3a-deficient medium was supplemented with the Notch inhibitor DAPT (10 μm, Stemgent) and IWP2. Following 4 additional days of treatment, organoids were harvested for RNA collection or lysed for Luminex analysis.

Microarray analysis and GSEA

RNA was harvested from snap-frozen colon tissue using the Qiagen RNAeasy microarray tissue mini kit (cat. 73304). Microarray was performed using Affymetrix mouse 430 2.0 GeneChip at the Dana Farber microarray core. Microarray data were deposited in the NCBI gene expression omnibus (accession number GSE87317).

The following gene sets were pulled and generated from published manuscripts: Atoh1 null, secretory progenitors (Sec-pro), and enterocytes [15]; Lgr5-positive cells [14]; and TACs [16]. All gene sets were used as published except the enterocyte gene set, which was generated using the raw data from Kim et al. Enterocyte-specific genes were defined as the transcripts that showed at least 2-fold increased expression compared to all other measured cell types (Lgr5, Atoh1 null, and Sec-pro) with an FDR < 0.05. Gene set enrichment analysis was run using the Broad Institute’s GSEA software [39].

Microbiome analysis

Mice were treated with rapamycin or vehicle daily for 2 weeks as described above. Fecal samples were harvested pretreatment and post-treatment, frozen until the end of the treatment period, and then homogenized in DNA/RNA Shield reagent (Zymo R1100-50) at 10% (v/v). DNA was extracted using the ZymoBIOMICS DNA Miniprep Kit (Zymo R2002). Generation of the 16s library, sequencing, read calling, and taxonomy assignment were performed by Zymo Research through their ZymoBIOMICS service. ZymoBIOMICS Microbial Community Standards (Zymo D6300) were used as positive controls. The library was sequenced on an Illumina MiSeq with >10% PhiX spike-in. Amplicon sequences were deduced from raw reads, and chimeric sequences were removed using the Dada2 pipeline [40]. Taxonomy assignment was performed using Qiime v1.9.1 [41]. PCoA and correlation analysis of Bray-Curtis dissimilarity scores were performed using MATLAB.