Trinitrobenzene sulfonic acid-induced intestinal injury in neonatal mice activates transcriptional networks similar to those seen in human necrotizing enterocolitis.

BACKGROUND: We have shown previously that enteral administration of 2, 4, 6-trinitrobenzene sulfonic acid in 10-d-old C57BL/6 pups produces an acute necrotizing enterocolitis with histopathological and inflammatory changes similar to human necrotizing enterocolitis (NEC). To determine whether murine neonatal 2, 4, 6-trinitrobenzene sulfonic acid (TNBS)-mediated intestinal injury could be used as a NEC model, we compared gene expression profiles of TNBS-mediated intestinal injury and NEC.
METHODS: Whole-genome microarray analysis was performed on proximal colon from control and TNBS-treated pups (n = 8/group). For comparison, we downloaded human microarray data of NEC (n = 5) and surgical control (n = 4) from a public database. Data were analyzed using the software programs Partek Genomics Suite and Ingenuity Pathway Analysis.
RESULTS: We detected extensive changes in gene expression in murine TNBS-mediated intestinal injury and human NEC. Using fold-change cut-offs of ±1.5, we identified 4,440 differentially-expressed genes (DEGs) in murine TNBS-mediated injury and 1,377 in NEC. Murine TNBS-mediated injury and NEC produced similar changes in expression of orthologous genes (r = 0.611, P < 0.001), and also activated nearly-identical biological processes and pathways. Lipopolysaccharide was top predicted upstream regulator in both the murine and human datasets.
CONCLUSION: Murine neonatal TNBS-mediated enterocolitis and human NEC activate nearly-identical biological processes, signaling pathways, and transcriptional networks.

INTRODUCTION

Necrotizing enterocolitis (NEC) continues to be a leading cause of mortality in premature neonates born prior to 32 weeks' gestation (1). The pathogenesis of NEC is complex and not well-understood; epidemiological studies associate NEC with risk factors ranging from chorioamnionitis, perinatal asphyxia, indomethacin therapy, formula feedings, human milk fortifiers, viral infections, feed thickeners, to severe anemia and/or consequent red blood cell transfusions (2). However, a unifying pathophysiological mechanism remains elusive.

In recent years, several clinical studies have shown that the incidence of NEC peaks in premature infants at a post-menstrual age (gestational age at birth + postnatal age) equivalent to 32 weeks' gestation (3–5). Considering this developmental predilection, and because clinical antecedents of NEC seem too diverse, we have argued that the pathoanatomy of NEC represents a generic injury response of the intestine during a certain developmental epoch, rather than mirroring specific causal mechanism(s) (6, 7). To investigate this hypothesis, we induced inflammatory intestinal injury in 10-day-old murine pups by using 2, 4, 6-trinitrobenzene sulfonic acid (TNBS) as a non-specific immunological insult that causes mucosal injury in the presence of the intestinal microflora (6). TNBS-induced murine neonatal enterocolitis was marked by monocyte/macrophage-rich infiltrates and extensive necrosis, and therefore, showed a strong histopathological resemblance to human NEC (6, 7).

In the present study, we used microarray technology to obtain a panoramic view of the gene networks in murine neonatal TNBS-mediated intestinal injury and human NEC. We integrated genetic, biological, and functional information to delineate differentially-expressed genes (DEGs), gene interactions and transcriptional networks. To determine whether this murine model could, at least in part, recapitulate human NEC, we also compared the murine and human datasets for dysregulated pathways/processes and their upstream regulators.

RESULTS

Microarray profiles of TNBS-mediated murine neonatal intestinal injury and human NEC

Administration of TNBS in C57BL/6 pups induced severe inflammatory changes and tissue necrosis in distal ileum and proximal colon. We used the first 0.5 cm of the proximal colon for microarray analysis. Principal component analysis (PCA) of these microarrays accounted for 62.7% of the variability in gene expression and revealed distinct clustering of gene expression in the TNBS-treated and control intestine (Figure 1A). Similarly, PCA of datasets from human NEC and uninflamed human neonatal intestine showed that the two groups also aggregated in distinct clusters (Figure 1B). In NEC, PCA accounted for 75.7% of the variability in gene expression.

Figure 1

Microarray profiles of TNBS-mediated murine neonatal intestinal injury and human NEC

(a) Principal component analysis (PCA) of microarray data from control (green) and TNBS-treated mice (red) showed distinct clustering of the two groups. N=8 pups/group. The graph is a scatter plot of the values of 3 principal components based on the correlation matrix of the total normalized array intensity data; each point represents one animal. Ellipsoids represent 95% confidence intervals of the clusters. X-, Y-, and Z-axis correspond to principal component 1 (PC1), PC2, and PC3; (b) PCA of datasets from uninflamed human neonatal intestine (green) and NEC (red) show that these two groups also aggregated in separate clusters. N=5 patients with NEC, 4 controls with uninflamed intestine.

Differentially-expressed genes (DEGs) in TNBS-mediated intestinal injury and human NEC

We next compared the signal intensities of genes in murine TNBS-mediated intestinal injury and control, and in human NEC vs. uninflamed neonatal intestine. Using fold-change cut-offs of ±1.5, p-value <0.05 (in analysis of variance), and a false discovery rate of 5%, we identified 4440 DEGs in murine TNBS-mediated injury out of a total of 39,601 probe sets. As shown in Figure 2A (top panel), 1648 genes were upregulated in TNBS-mediated intestinal injury, whereas 2792 genes were expressed at higher levels in control. Hierarchical clustering of these DEGs showed distinct expression profiles in the control and injury groups (bottom panel). The heat maps also highlighted that only a minority of DEGs were upregulated in intestinal injury.

Figure 2

Differentially-expressed genes (DEGs) in TNBS-mediated intestinal injury and human NEC

(a) DEGs in murine TNBS-mediated intestinal injury vs. control. Top: Scatterplot shows DEGs upregulated in intestinal injury (red) and control (green). Each gene is denoted by an X, the size of which is inversely proportional to the p-value. X- and Y-axis show signal intensities on a log2 scale. Data were analyzed by ANOVA. Grey marks indicates genes that were filtered out; Bottom: Hierarchical clustering of DEGs in TNBS-mediated injury vs. control highlights the distinct gene expression profiles of the two groups. Dendrogram above the heat map depicts hierarchical clustering of the samples (controls shown as green, injury samples red). Cluster distance is based on the average distance between all the pairs of objects in the two clusters. Dendrograms for DEGs are not depicted. In the heat maps, red shows increased expression, blue indicates decreased expression, whereas grey shows no change. Expression value intensities are illustrated by color with a range of −3 to +3 on a log scale; (b) DEGs in human NEC and uninflamed neonatal intestine. Top: Scatterplot shows genes upregulated in NEC (red) and uninflamed intestine (green); Bottom: Hierarchical clustering of DEGs shows distinct profiles of human NEC and uninflamed intestine. Settings similar to panel A.

In human NEC, we detected a similar pattern of gene expression. Using filters similar to the murine datasets, we identified 1377 DEGs out of a total of 33,297 probe sets. As shown in Figure 2B, 324 genes were upregulated in NEC, whereas 1053 genes were overexpressed in the uninflamed neonatal intestine. Hierarchical clustering confirmed distinct gene expression profiles in NEC and uninflamed intestine, and similar to TNBS-mediated injury, showed that only a small group of genes was upregulated in the NEC intestine.

We have listed DEGs in TNBS-mediated intestinal injury and NEC by functional categories in Table 1. To restrict this list to a smaller, more meaningful length, we chose thresholds ≥ ±2-fold with p<0.01. The top 10 upregulated and downregulated genes are listed in Table 2. A complete listing of the DEGs in TNBS-mediated intestinal injury and NEC is provided as Supplemental Tables S1 and S2, respectively.

Table 1

Differentially-expressed genes in TNBS-mediated murine neonatal intestinal injury and human NEC

TNBS-mediated murine intestinal injury			Human NEC
Gene Symbol (mouse)	Fold-Change (Intestinal injury vs. control)	p-value (Intestinal injury vs. control)	Gene Symbol (human)	Fold-Change (NEC vs. uninflamed intestine)	p-value (NEC vs. uninflamed intestine)
Leukocyte antigens
Ly6g	25.3817	2.03E-10	CD177	46.1533	1.06E-5
Ly6a	9.6089	2.09E-9	TREM1	24.328	8.65E-6
Ly6d	4.20968	2.00E-6	FPR2	18.3964	2.19E-4
Itgam	2.70284	4.57E-7	CSF3R	9.27637	1.32E-4
Itga5	2.01284	8.44E-4	FPR1	9.15378	1.32E-4
Cd14	2.23735	1.33E-4	C5AR1	9.07387	1.85E-4
			TLR2	7.46131	1.9E-4
			CD163	4.59848	1.03E-3
			CXCR1	4.92515	7.06E-4
			ITGAM	4.63906	6.22E-4
			CD14	4.05947	5.04E-4
			C5AR2	2.91767	1.82E-3
			TLR4	2.91104	5.57E-4
Calcium-binding proteins
S100a8	13.4153	5.30E-8	S100A12	28.1327	5.63E-5
S100a9	7.59614	3.90E-8	S100A9	14.4051	1.15E-4
S100a14	3.09206	1.76E-4	S100A8	11.1661	7.72E-5
S100a11	2.4841	1.85E-3	S100A11	2.01031	4.02E-4
S100a6	2.53026	4.32E-3	S100A14	−3.66239	7.91E-5
S100g	−8.39296	1.16E-9
Solute carriers
Slc7a11	7.18376	3.21E-8	SLC11A1	11.6304	1.91E-4
Slc17a4	−9.93905	1.07E-9	SLC16A6	7.52272	9.35E-4
Slc6a14	−9.1045	8.85E-6	SLC2A3	6.08879	3.11E-5
Slc17a5	−8.66322	1.31E-6	SLC25A37	3.77664	8.83E-4
Slc3a1	−7.91494	2.02E-5	SLC7A2	3.09536	6.95E-4
Slc30a10	−8.61152	3.13E-9	SLC5A1	−104.196	5.26E-7
Slc51a	−6.06685	5.78E-6	SLC5A12	−44.8826	3.72E-5
Slc34a2	−6.00088	6.84E-5	SLC2A2	−37.4188	3.38E-6
Slc40a1	−5.86264	4.44E-8	SLC15A1	−27.8306	4.85E-5
Slc5a8	−5.50916	1.37E-7	SLC44A4	−27.4277	1.14E-6
Slc10a5	−5.49049	4.03E-9	SLC6A19	−26.6032	1.43E-6
Slc6a19	−4.48791	1.91E-4	SLC7A9	−13.9373	6.63E-5
Slc15a1	−4.43097	1.72E-9
Slc7a8	−4.0919	2.89E-8
Cytokines, chemokines, and cognate receptors
Cxcl2	6.34927	1.25E-6	IL1A	41.982	2.75E-4
Il1b	3.92428	2.88E-5	IL1B	41.2998	1.59E-5
Ccl2	3.46452	6.03E-5	CXCL8	35.6662	8.81E-6
Il1rn	5.13162	5.77E-8	IL6	25.798	2.67E-5
Tnfrsf12a	3.13015	5.24E-8	IL1R2	24.1338	7.33E-4
Cxcl3	2.78899	9.78E-7	CCL3	13.7709	2.84E-4
Cxcl5	2.67758	5.26E-5	CCL3L3	12.3779	6.11E-4
Ccl7	2.90073	6.31E-5	IL1RN	12.3734	1.94E-3
Ccl3	2.71695	4.71E-4	CCL4L2	10.7223	8.41E-4
Il1r2	2.48661	9.60E-7	CXCL2	9.75128	1.89E-5
Il1r1	2.47906	5.11E-7	CCL4L2	9.1063	1.16E-3
Vip	4.50863	2.17E-5	CXCL1	7.75825	2.17E-4
Il19	2.43878	3.63E-4	FGF7	3.6188	9.70E-6
Il1a	2.3335	7.82E-4	ANGPT2	3.52358	1.94E-3
Cxcl1	2.33307	4.77E-5	CXCL3	2.83887	3.03E-4
Pf4	2.02541	4.38E-5	IL1R1	2.82136	1.50E-4
Fgf15	−5.60601	4.54E-7	CCL2	2.75979	1.62E-4
Tnfsf10	−2.08014	1.96E-6	CCL25	−15.5371	2.14E-6
Ccl21a	−4.07594	8.36E-6	CXCL14	−12.6101	5.75E-4
Ccl21b	−3.8607	1.41E-5
Enzyme inhibitors
Serpina3n	9.00078	3.30E-8	SERPINB2	55.8164	2.15E-4
Timp1	4.50872	1.44E-7	SERPINE1	9.21821	1.49E-3
Serpina3g	2.31019	3.88E-6	TIMP1	3.27933	1.65E-4
Serpine1	2.2178	1.42E-4
Enzymes
Mmp3	4.18699	2.78E-6	MMP8	18.5117	7.16E-4
Mmp10	3.30835	2.47E-5	ADAMTS9	3.33943	1.13E-3
Adam8	2.25568	1.22E-5	MMP14	2.47711	1.67E-3
Immune mediators
Stfa2	4.27652	1.38E-6	IRAK3	7.06643	1.03E-4
Gp49a	4.39035	1.97E-7	NLRC4	6.09349	6.13E-4
Thbs1	4.24043	1.22E-5	TNFAIP3	4.87626	1.01E-3
Stfa3	3.34135	4.61E-6	SIGLEC5	4.87278	1.89E-3
Stfa1	3.56203	1.27E-5	IL1RAP	3.83477	4.49E-4
Stfa2l1	12.0136	1.29E-7	THBS1	2.93907	1.54E-3
Saa3	3.18986	1.68E-8	NFKBIA	2.64241	8.93E-4
Transcription factors
Hif1a	2.55804	6.22E-6	CEBPB	3.48237	3.11E-4
Cebpd	4.16739	4.01E-8	HIF1A	2.29961	3.5E-3
Cebpb	2.03583	1.25E-3	RUNX1	2.31259	1.2E-3
Fos	2.03438	4.47E-7	CEBPG	−2.00683	5.5E-4
Runx1	1.98006	5.50E-7
Apolipoproteins
Apoa4	−10.143	3.98E-4	APOB	−93.8819	5.06E-7
Apob	−15.3669	1.15E-4	APOA4	−65.5193	1.46E-7
Apoa1	−5.47609	8.82E-4	APOC3	−25.3508	6.08E-6
			APOA1	−11.5383	8.7E-4
Cytochrome proteins
Cyp2c55	−56.9722	5.15E-11	CYP3A4	−144.116	3.45E-6
Cyp2d26	−18.5949	2.20E-13	CYP2B6	−27.1438	8.53E-6
Cyp2c65	−9.70339	2.07E-8	CYP3A5	−23.9293	5.91E-5
Cyp3a13	−6.83871	1.79E-6	CYP2C9	−12.0691	1.86E-4
Cyp2w1	−6.15221	7.43E-8
Cyp2c66	−5.87623	1.15E-8
Cyp2c68	−4.31912	8.58E-9

Table 2

Top 10 up- and downregulated genes in murine TNBS-mediated neonatal intestinal injury and human NEC

Murine TNBS-mediated intestinal injury: Top 10 upregulated genes
Serial No.	Gene	Gene name	Fold change	p-value
1.	Ly6g	lymphocyte antigen 6 complex, locus G	25.3817	2.03E-10
2.	S100a8	S100 calcium binding protein A8 (calgranulin A)	13.4153	5.30E-8
3.	Stfa2l1	stefin A2 like 1	12.0136	1.29E-7
4.	Ly6a	lymphocyte antigen 6 complex, locus A	9.6089	2.09E-9
5.	Gm12840	Predicted gene, 12840	9.01718	4.87E-11
6.	Serpina3n	serine (or cysteine) peptidase inhibitor, clade A, member 3N	9.00078	3.30E-8
7.	Gm3776	Predicted gene, 3776	8.30097	7.78E-8
8.	Akr1b8	aldo-keto reductase family 1, member B8	7.7098	9.22E-8
9.	S100a9	S100 calcium binding protein A9 (calgranulin B)	7.59614	3.90E-8
10.	Slc7a11	solute carrier family 7 member 11	7.18376	3.21E-8
Murine TNBS-mediated intestinal injury: Top 10 downregulated genes
1.	Cyp2c55	cytochrome P450, family 2, subfamily c, polypeptide 55	−56.9722	5.15E-11
2.	Car1	carbonic anhydrase 1	−32.9199	1.28E-9
3.	Gm5843	Predicted gene, 5843	−27.9322	8.64E-10
4.	Cubn	cubilin (intrinsic factor-cobalamin receptor)	−23.5509	1.60E-7
5.	1810065E05Rik	RIKEN cDNA 1810065E05 gene	−20.0201	5.38E-9
6.	Cyp2d26	cytochrome P450, family 2, subfamily d, polypeptide 26	−18.5949	2.20E-13
7.	Alpi	alkaline phosphatase, intestinal	−18.3635	1.78E-5
8.	Gm21885	predicted gene, 21885	−17.4586	4.32E-9
9.	Naaladl1	N-acetylated alpha-linked acidic dipeptidase-like 1	−15.9323	1.64E-5
10.	Apob	apolipoprotein B	−15.3669	1.15E-4
Human NEC: Top 10 upregulated genes
1.	SERPINB2	serpin family B member 2	55.8164	2.15E-4
2.	CD177	CD177 molecule	46.1533	1.06E-5
3.	IL1A	interleukin 1 alpha	41.982	2.76E-4
4.	IL1B	interleukin 1 beta	41.2998	1.59E-5
5.	CXCL8	C-X-C motif chemokine ligand 8	35.6662	8.81E-6
6.	AQP9	aquaporin 9	30.183	7.25E-5
7.	S100A12	S100 calcium binding protein A12	28.1327	5.63E-5
8.	IL6	interleukin 6	25.798	2.67E-5
9.	TREM1	triggering receptor expressed on myeloid cells 1	24.328	8.65E-6
10.	IL1R2	interleukin 1 receptor 2	24.1338	7.33E-4
Human NEC: Top 10 downregulated genes
1.	CYP3A4	cytochrome P450 family 3 subfamily A member 4	−144.116	3.45E-6
2.	SI	sucrase-isomaltase	−132.797	4.59E-6
3.	CLCA1	chloride channel accessory 1	−130.38	1.42E-4
4.	ALDOB	Aldolase B	−120.335	1.22E-7
5.	MUC13	mucin 13, cell surface associated	−104.353	2.13E-5
6.	MTTP	microsomal triglyceride transfer protein	−104.334	1.04E-7
7.	SLC5A1	solute carrier family 5 member 1	−104.196	5.26E-7
8.	APOB	apolipoprotein B	−93.8819	5.06E-7
9.	TM4SF20	transmembrane 4 L six family member 20	−86.7135	4.10E-7
10.	LCT	lactase	−83.0787	1.12E-5

To determine whether the transcriptional changes in murine TNBS-mediated intestinal injury model human NEC, we compared our murine and human datasets for the direction and intensity of change in gene expression. To circumvent the difficulties in comparing across species and microarray platforms, we extracted the data on the orthologous genes from our list of DEGs (Supplemental Table S3). As shown in Figure 3, these 484 genes showed a high degree of concordance in expression and were similarly up- or downregulated in TNBS-mediated injury and human NEC, respectively (Spearman's r = 0.611, p<0.001).

Figure 3

Murine TNBS-mediated intestinal injury and human NEC show similar changes in gene expression

Scatter-plot shows changes in the expression of orthologous genes in the human and murine intestine in NEC and TNBS-mediated intestinal injury, respectively. X-axis shows fold changes in human genes during NEC on a log2 scale, whereas the Y-axis shows changes in their murine orthologues during TNBS-mediated intestinal injury. Most data points were clustered in the top right (upregulated in both human NEC and murine intestinal injury) or bottom left quadrants (downregulated in both datasets) of the XY plane, indicating that the expression of these orthologous genes changed similarly in murine TNBS-mediated intestinal injury and human NEC.

Biological processes active in murine TNBS-mediated intestinal injury and human NEC

We next used the software program Partek Genomic Suite to map DEGs to Biological Process categories of Gene Ontology (GO) (8) and to rank these GO categories by enrichment score (computed as the negative natural log of the p-value). In both murine TNBS-mediated intestinal injury and NEC, the top biological processes were cellular processes (enrichment scores of 91.26 and 86.1, respectively) and single-organism processes (enrichment scores of 77.67 and 74.59, respectively). As shown in Figure 4a, the biological process activated in murine TNBS-mediated intestinal injury and human NEC were nearly identical.

Figure 4

Top biological processes in murine TNBS-mediated intestinal injury and human NEC

(a) Bar diagrams show the top biological processes activated in TNBS-mediated intestinal injury (left) and human NEC (right). The predicted GO categories were ranked by enrichment score [−log (p-value)]. Single-organism processes included various cellular, metabolic, developmental, and reproductive processes. Multi-organism processes included response to bacteria or other organisms; (b) Bar diagrams show immune system processes activated in TNBS-mediated injury (left) and human NEC (right); (c) Bar diagrams show biological processes leading to cell death in TNBS-mediated injury (left) and human NEC (right).

Histopathologically, inflammation and tissue necrosis are the hallmarks of NEC. Therefore, we focused on two GO Biological Process categories: the immune system and cell death. The top immune system processes identified in both TNBS-mediated injury and human NEC were leukocyte migration (enrichment scores of 15.88 and 19.39, respectively) and immune response (enrichment scores of 11.62 and 7.95, respectively; Figure 4b). The primary cellular process leading to cell death in both TNBS-mediated intestinal injury and human NEC was programmed cell death/apoptosis (enrichment scores of 14.26 and 3.61, respectively; Figure 4c).

Disease-related Pathways active in murine TNBS-mediated injury and human NEC

We next used the online software application Ingenuity Pathway Analysis (IPA) to map DEGs into disease-related GO pathways. Top canonical pathways activated in TNBS-mediated intestinal injury were similar to those see in human NEC (Table 3).

Table 3

Top canonical pathways activated in murine TNBS-mediated neonatal intestinal injury and human NEC

Murine TNBS-mediated neonatal intestinal injury
Name	Enrichment score	p-value	Overlap
LPS/IL-1 mediated inhibition of RXR function	12.665	2.16E-13	31.6% (61/193 genes)
Granulocyte adhesion and diapedesis	8.297	5.05E-9	29.5% (44/149 genes)
Agranulocyte adhesion and diapedesis	7.275	5.31E-8	27.5% (44/160 genes)
PXR/RXR activation	7.001	9.97E-8	40% (22/55 genes)
LXR/RXR activation	6.356	4.41E-7	29.9% (32/107 genes)

We also evaluated top inflammatory and cell death pathways in our murine and human datasets, utilizing the KEGG databases (9) with pathway enrichment tools within the software program Partek Pathway and in the Database for Annotation, Visualization and Integrated Discovery (DAVID) (10). Tumor necrosis factor (TNF)-activated signaling, hematopoietic cell signaling, cytokine-cytokine receptor interaction, nucleotide-binding oligomerization domain receptor (NOD-like receptor)-mediated signaling, and chemokine signaling pathways were enriched in both TNBS-mediated intestinal injury and human NEC (Table 4).

Table 4

Top inflammatory pathways activated in murine TNBS-mediated neonatal intestinal injury and human NEC

Murine TNBS-mediated neonatal intestinal injury
Name	Enrichment score	p-value	Overlap
TNF signaling	9.749	5.83E-5	27.1% (29/107 genes)
NOD-like receptor signaling	8.067	3.13E-4	31% (18/58 genes)
Hematopoietic cell signaling	7.263	7.01E-4	34.4% (22/82 genes)
Cytokine-cytokine receptor interaction	6.941	9.66E-4	20.1% (52/258 genes)
Chemokine signaling	3.221	4.11E-2	16.7% (32/192 genes)

To compare the mechanism(s) of mucosal inflammation and cell death in murine TNBS-mediated intestinal injury and human NEC, we evaluated the murine and human datasets for DEGs active in the top inflammatory pathways (TNF signaling, hematopoietic cell signaling, and cytokine-cytokine receptor interaction; Figure 5) and in promoting apoptosis. Several key mediators were common to both murine TNBS-mediated injury and human NEC (highlighted in orange): interleukin (IL)-1α, IL-1β, IL-1 receptor 1 (IL1R1), IL1R2, IL-6, TNF receptor superfamily member 1B (TNFRSF1B)/TNFR2, TNFα induced protein 3 (TNFAIP3)/A20, IL18R1, transforming growth factor-β3, TNF superfamily, member 10 (TNFSF10)/ TNF-related apoptosis-inducing ligand (TRAIL); chemokines CXC-motif ligand (CXCL)-1, CXCL2, CXCL3, CXCL5, CC-motif ligand (CCL)-2, CCL3, leukocyte receptors CD14, integrin alpha M (ITGAM)/CD11b, colony stimulating factor receptor (CSFR)-2b; the transcription factor CCAAT/Enhancer Binding Protein (C/EBP)-β, enzymes such as prostaglandin-endoperoxide synthase 2 (PTGS2) and alanyl aminopeptidase (ANPEP), and signaling mediators such as suppressor of cytokine signaling 3 (SOCS3), and nuclear factor of kappa light polypeptide gene enhancer in B-cells inhibitor, alpha (NFKBIA)/inhibitor of kappa B-alpha (IκBα). In the apoptotic pathways, IL-1α, IL1β, IL1R1, IκBα, and IL1R-associated kinase 3 (IRAK3) were enriched in both murine TNBS-mediated intestinal injury and human NEC (not depicted).

Figure 5

Top inflammatory pathways in murine TNBS-mediated injury and human NEC

Heat maps show the expression of DEGs involved in 3 top inflammatory pathways: TNF-activated signaling, hematopoietic cell signaling, and cytokine-cytokine receptor interaction in (a) murine TNBS-mediated injury and (b) human NEC. In the heat maps, expression values are shown on a log scale with a range of −3 to +3; red boxes show upregulation, blue show downregulation, and grey indicate no change. Dendrogram above the heat map depicts hierarchical clustering of the samples (controls shown as green, injury samples red). DEGs common to both murine and human datasets are highlighted in orange. Per convention, murine gene symbols are written with the first letter in upper case and the rest in lower case. Human gene symbols are capitalized.

Comparison analysis of murine TNBS-mediated injury and human NEC

We next used the web-based software application, Ingenuity Pathway Analysis, to compare murine and human datasets for canonical GO pathways, upstream transcriptional regulators, and disease processes typically mediated by the observed transcriptional networks. Heat maps developed by hierarchical clustering of DEGs showed that the top canonical pathways were enriched to similar degree in murine TNBS-mediated intestinal injury and human NEC (Figure 6a). In both datasets, LPS/IL-1 mediated inhibition of RXR function, granulocyte adhesion and diapedesis, activation of nuclear receptors, xenobiotic metabolism signaling (including activation of aryl hydrocarbon receptors), IL-12 and IL-17 signaling, and Toll-like receptor activation was enriched. Similarly, the predicted upstream regulators also showed remarkable similarity; lipopolysaccharide (LPS), TNF, IL-1β, interferon (IFN)-γ, and IL-6 were the top transcriptional regulators in both the murine and human datasets (Figure 6b). In both datasets, the top GO disease processes were inflammatory response, necrosis, apoptosis, and leukocyte migration (Figure 6c), which also happen to be the major histopathological findings in human NEC (11). In NEC lesions, existing evidence documents apoptosis mainly in intestinal epithelial cells and not in other cell lineages (12), and consistent with these data, we detected transcriptional networks that were generally consistent with apoptosis but not those typical of cell death in antigen-presenting cells or the connective tissue.

Figure 6

Comparison analysis of murine TNBS-mediated injury and human NEC show a high degree of congruity in the top-ranked canonical pathways, predicted upstream regulators, and disease processes

(a) Top canonical pathways in murine TNBS-mediated injury and human NEC were depicted in a heat map that was clustered hierarchically on two axes (pathways and functions vs. condition); enrichment scores [−log (p-value); range 0 to 16.9] were depicted in grayscale; (b) Heat map shows the top predicted upstream regulators (connected to DEGs through direct or indirect relationships) in murine and human datasets; (c) Heat map shows top disease processes in TNBS-mediated intestinal injury and human NEC.

LPS is predicted to activate similar transcriptional networks in both murine TNBS-mediated injury and human NEC

LPS was identified as a top upstream transcriptional regulator in both TNBS-mediated intestinal injury and human NEC. Interestingly, the predicted transcriptional networks activated by LPS in the murine and human datasets were nearly identical (Figure 7); in these networks, LPS was predicted to induce IL-1β, IL-6, TNF, and IFN-γ, which, in turn, activated the transcription factors nuclear factor-kappa B-1 (NF-κB1), signal transducer and activator of transcription (STAT)-3, and c-Jun. STAT1 and IκBα were predicted to dampen these inflammatory cascades. C/EBP-β activation was predicted with greater confidence in NEC than in murine intestinal injury. Tumor protein 53 (tp53)/p53 was predicted to limit TNBS-mediated injury but these interactions did not reach statistical significance in NEC. In contrast, glucocorticoid receptor (nuclear receptor subfamily 3, group C, member 1/NR3C1) was identified as a likely inhibitor in human NEC but not in murine intestinal injury.

Figure 7

LPS is predicted to activate similar transcriptional networks in both murine TNBS-mediated injury and human NEC

LPS-activated gene networks in (a) murine TNBS-mediated injury and (b) human NEC. Direct relationships were shown as solid arrows, whereas broken arrows show indirect relationships. Inset: prediction legend. IL = interleukin; TNF = tumor necrosis factor; IFNG = interferon-γ; NFKBIA = nuclear factor of kappa light polypeptide gene enhancer in B-cells inhibitor, alpha (NFKBIA)/inhibitor of kappa B-alpha (IκBα); NFKB1 = nuclear factor-kappa B-1; STAT = signal transducer and activator of transcription; JUN = Jun proto-oncogene, AP-1 transcription factor subunit; C/EBP-beta; TP53 = tumor protein 53; NR3C1 = nuclear receptor subfamily 3, group C, member 1/glucocorticoid receptor.

DISCUSSION

We present a detailed comparison of the transcriptional networks activated in murine neonatal TNBS-mediated intestinal injury and human NEC, and show a high degree of similarity between the two datasets in terms of DEGs and predicted biological processes, pathways, and the causal networks. These observations are consistent with our previous reports that TNBS-mediated intestinal injury in mouse pups resembled human NEC in both the cellular inflammatory response and cytokine/chemokine expression in the affected intestine (6, 7). In this context, the present study provides additional evidence suggesting that murine neonatal TNBS-mediated intestinal injury could be useful in modeling human NEC in preclinical studies.

We have previously described the use of TNBS to induce acute necrotizing enterocolitis in mouse pups (6). We postulated that the pathoanatomy of NEC represents a generic injury response of the developing intestine (6), and therefore, used TNBS as a non-specific, but predictable, immunological insult to induce bowel injury in pups. The temporal course and pathoanatomy of TNBS-induced intestinal injury in pups recapitulates several clinical characteristics of human NEC, and differs considerably from TNBS-induced colitis in adult mice (13) (Table 5). Our rationale for using TNBS to simulate NEC-like injury in mouse pups was as follows: (a) TNBS had been used successfully to induce colitis in adult mice to model inflammatory bowel disease (7); (b) TNBS administration in pups by gavage and enema produced inflammatory injury in the distal ileum and colon, and recapitulated the regional predilection of human NEC. The cellular inflammatory response was comprised of monocyte/macrophage-rich infiltrates and also resembled NEC (6); (c) TNBS did not affect germ-free mice, indicating that the mucosal injury occurred only in the presence of gut bacteria (similar to NEC) and was not due to chemical/corrosive action of TNBS (6); (d) TNBS-mediated enterocolitis can be used to investigate the temporal evolution of neonatal intestinal injury (from the time of instillation of TNBS), unlike the hypothermia-hypoxia model where individual animals develop bowel injury at different times during the NEC protocol (14, 15); (e) comparison of pups vs. adult mice treated with weight-normalized doses of TNBS can provide insights into the developmental aspects of NEC; (f) unlike NEC models involving splanchnic ischemia-reperfusion or the administration of platelet-activating factor where the animals need to be euthanized within a few hours (16, 17), the evolution of TNBS-enterocolitis can be observed over longer periods; (g) because all pups in a cohort are affected predictably, fewer animals are needed (compared to the hypoxia-hypothermia model where bowel injury may occur only in 40–70% mice); and (h) the use of 10-day-old pups allows enough time to test prophylactic interventions; it was convenient that the rodent intestine at birth resembles the preterm human intestine and takes 3 weeks to reach the structural/functional maturity of the term human neonate (18, 19).

Table 5

Comparison of TNBS-induced enterocolitis in mouse pups vs. TNBS-induced colitis in adult mice

	TNBS-induced enterocolitis in murine pups	TNBS-induced colitis in adult mice
Procedure	One-time administration of TNBS (2 identical doses of 50 mg/kg dissolved in 30% ethanol, administered by gavage and rectal instillation, respectively)	Usually induced by one-time rectal instillation of TNBS (0.5–6 mg/mouse, dissolved in 35–50% ethanol). Chronic colitis can be induced by repeated intrarectal applications of TNBS dissolved in ethanol, administered at weekly intervals
Mouse strains	C57BL/6 pups develop severe injury; strains such as 129SvEv are also highly-susceptible	C57BL/6 mice are relatively resistant to TNBS. Strains such as C3H/HeJ, SJL/J, and Balb/c are more sensitive
Regional predilection	Distal ileum, cecum, and proximal colon	Distal colon
Temporal evolution	Acute necrotizing enterocolitis evident at 12h. Injury peaks at 18–24h with >50% mortality. In survivors, mucosal injury starts to resolve at 48h. Histopathology shows necrosis, macrophage-rich infiltrates, and focal hemorrhages	Acute transmural inflammation within a few hours. Slower, subacute progression. Focal areas of necrosis, mononuclear infiltration, and basal cryptitis develops in 2–3 days. Mucosal inflammation persists for up to 2–3 weeks.
Cellular inflammatory response	Monocytes/macrophages (nearly 70%), some neutrophils (20–25%), but very few lymphocytes	Pleomorphic response; abundant CD4+ Th1 and Th17 lymphocytes
Cytokine expression in affected tissues	IL-1α, IL-1β, chemokines CXCL2, CXCL5, CCL3, CCL4	Interferon-γ, TNF, IL-2, and IL-12p70
Mechanism(s)	Unclear. Administration of ethanol or TNBS alone does not cause injury. Ethanol is believed to increase the permeability of the epithelial barrier, allowing TNBS to reach the lamina propria. Germ-free pups remain unaffected, indicating that the injury develops only in the presence of luminal microflora and is not due to chemical/corrosive action of TNBS	TNBS is believed to haptenize microbial- and self-proteins to render them immunogenic. Inflammatory response during first 5–6 days believed to represent a delayed-type hypersensitivity response

The number of DEGs in our murine samples was much larger than in the human datasets. We suspect that many genes in the human dataset did not rise to statistical significance because of clinical variability. In our mouse model, intestinal tissue was harvested from a pre-defined bowel region from all animals at one specific time-point, and was compared to controls that were identical in region, postnatal age, genetic background, diet, and environment with the exception of TNBS exposure. In contrast, patients with NEC underwent surgical resection at different gestational and postnatal ages, and at different time-points following onset of NEC. These NEC specimens were compared to control tissue samples that were not obviously inflamed but were also not entirely normal (those infants underwent abdominal surgery for intestinal obstruction or ileostomy repair), and were only approximately comparable in gestational and postnatal age to the NEC group. In addition to these sources of variability, some other clinical issues also merit consideration. With improved supportive care, it is often possible to delay surgery in some infants until after the acute phase has passed (20). In the interim, some of the early changes in gene expression in the affected bowel may subside and may no longer be detectable in surgically-resected samples. Finally, we must consider the possibility of variable RNA yield from surgical samples of NEC. During surgery for acute NEC, visual identification of necrotic bowel from areas of borderline viability can be a challenging task. Cognizant of the risks of short bowel syndrome, most surgeons take a conservative approach and limit resection to the most-obviously devitalized segments (21). Therefore, surgically-resected specimens of NEC often include patches that are completely necrotic with limited transcriptional activity.

The top immune system process in both TNBS-mediated intestinal injury and human NEC was leukocyte migration. Similarly, GO pathways TNF signaling, hematopoietic cell signaling, and cytokine-cytokine receptor interaction were highly enriched in both the murine model as well as human NEC. These findings are consistent with the prominent cellular inflammatory response seen in murine TNBS-mediated injury as well as human NEC (6, 7, 21). We identified several cytokines and chemokines in both murine and human datasets, including IL-1α, IL-1β, IL-6, CXCL1, CXCL2, CXCL5, CCL2, and CCL3. These findings are consistent with existing data from preclinical and clinical studies on cytokine expression in plasma and tissue samples (6, 22–24). Increased expression of the leukocyte receptors CD14 and CD11b can be explained by the monocyte/macrophage infiltration in NEC lesions (6, 7, 21, 25). We also detected increased NF-κB1 and IκB expression in our murine and human datasets, which is consistent with existing information that single nucleotide polymorphisms (SNPs) of NF-κB1 (g.-24519delATTG) and IκB (g.-1004A>G) are over-represented in patients with NEC. These SNPs have been implicated in the dysregulated inflammatory reaction seen in NEC (26).

The major mechanism of cell death in TNBS-mediated injury and human NEC was apoptosis. Epithelial cell apoptosis is an early event in NEC that may initiate the more generalized bowel necrosis seen in later stages of this disease (12). Several genes in the IL-1 signaling cascade were identified in apoptotic pathways in both the murine and human datasets. IL-1β can induce apoptosis in cultured enterocyte cell lines (27), but this relationship is likely to be more complex. Although usually seen as pro-apoptotic mediators, IL-1β and its cognate receptors can also play a protective role against detachment-induced cell death in villus enterocytes (28). Clearly, further study is warranted to elucidate the role of IL-1β and its cognate receptors in the context of NEC.

We identified LPS as the top upstream transcriptional regulator in both TNBS-mediated intestinal injury and human NEC. These findings are consistent with the increasingly well-documented role of Gram-negative bacteria in NEC pathogenesis. Infants who develop NEC display a microbial dysbiosis that antedates the onset of NEC, with over-representation of gammaproteobacteria (including Enterobacteriaceae and Pseudomonadaceae) (29, 30). The central role of bacteria in NEC pathogenesis is evident from the occurrence of NEC always after postnatal bacterial colonization and never in the sterile intra-uterine microenvironment prior to birth (31). Histopathologically, bacterial overgrowth and signatures of bacterial activity such as pneumatosis, the accumulation of gaseous products of bacterial fermentation in the bowel wall, are readily evident in NEC (11, 21). Clinically, the importance of Gram-negative bacteria in the pathogenesis of NEC is supported by evidence that enteral antibiotics such as aminoglycosides can protect against NEC and related mortality (32).

In conclusion, we have shown that murine neonatal TNBS-mediated intestinal injury activates transcriptional profiles that strong resemble human NEC. The major strength of our study is in its unbiased, systems biology-based approach that acknowledges the complexity of a natural disease such as NEC. However, there are also important limitations; software applications for gene annotation and gene ontology build models by querying the known literature and are therefore, limited to known interactions and are likely to miss hitherto-unknown regulatory networks. Given these limitations, all systems biology models are, to a certain extent, incomplete. There is a need for further corroboration of these findings in preclinical models and clinical studies, which can add data on covariates such as feeding experience, comorbidities, and microbial flora.

MATERIALS AND METHODS

Murine neonatal TNBS-mediated enterocolitis

Animal studies were approved by the Institutional Animal Care and Use Committee at University of South Florida. As described previously (6), TNBS enterocolitis was induced in 10-day-old C57BL/6 mice (n=11) by administering TNBS (2 equal doses of 50 mg/kg dissolved in 30% ethanol, w/v, by gavage and rectal instillation, respectively. After 24h, these animals were euthanized using CO2 inhalation followed by cervical dislocation. Control animals (n=10) received vehicle alone. Intestinal injury was confirmed by histopathological analysis of ileocolic region (6).

Tissue preparation and microarray analysis

We used a 0.5 cm segment of the proximal colon for analysis. RNA was extracted using TRIZOL reagent (Life Technologies, Gaithersburg, MD) and RNeasy mini kit (QIAGEN, Valencia, CA). Total RNA (100 ng) was amplified and labeled using the Ambion whole-transcript expression kit (Life Technologies) and Affymetrix WT sense target labeling kit (Affymetrix, Santa Clara, CA). Affymetrix GeneChip Mouse Transcriptome Assay 1.0 ST arrays were hybridized with 540 ng of labeled sense DNA, washed, and stained using the FS450 Fluidic Station (Affymetrix). Fluorescent signal intensities of each stained chip were captured by the GeneChip Scanner 3000 7G System (Affymetrix).

Gene-level expression data from surgically-resected human tissue samples of NEC

Microarray results from surgically-resected bowel affected by NEC and uninflamed neonatal intestine were downloaded from the public database Gene Expression Omnibus (GEO) at the National Center for Biotechnology Information (NCBI) (33). GEO Dataset files were downloaded from ftp://ftp.ncbi.nih.gov/pub/geo/DATA/SOFT/GDS/ and uncompressed from .gz compression format. These tissue samples (NEC, n=5; accession numbers GSM1661796, GSM1661797, GSM1661798, GSM1661799, GSM1133300; and uninflamed neonatal intestine, n=4; accession numbers GSM1133306, GSM1133307, GSM1133308, GSM1133309) have been described earlier (34). In this study, the authors obtained surgically-resected specimens of NEC (26.4–31.4 weeks' gestation) and uninflamed intestine (32.3–37.9 weeks' gestation, operated for intestinal obstruction or ileostomy closure). Standard Affymetrix protocols were followed for RNA extraction and hybridization of Human Gene 1.0 ST arrays (Affymetrix). Tissue specimens were validated by histopathology and mRNA expression by polymerase change reactions.

Microarray data analysis

Gene expression data were imported as .CEL files into the Partek Genomics Suite v6.6 (Partek, St Louis, MO). Raw data were processed by the Robust Multi-array Average method that involved quantile normalization, background correction, median polish probe set summarization, and log2 transformation to bring mean expression values to the same scale (35). Expression values from 2 control and 3 TNBS tissues were excluded in view of failure to pass quality control checks (36). Consequently, we used expression data from 8 control and 8 TNBS tissues in all downstream analyses. For comparison with human NEC, we uploaded .CEL datasets from 5 patients with NEC and 4 with uninflamed neonatal intestine. Average expression levels were distributed similarly across all samples.

Differentially-expressed genes (DEGs)

Gene expression between TNBS vs. control, and NEC vs. uninflamed neonatal intestine was compared by one-way analysis of variance on all probe sets processed by a robust multi-array procedure, combined with Fisher's least significant difference method for group-wise comparisons (37) and the Benjamini-Hochberg step-up procedure to identify false positives (38). To identify the DEGs, we set the thresholds for the false discovery rate at <0.05 and fold change ≥1.5. Hierarchical clustering analysis was performed on the DEGs using Partek default settings. Correlation statistics for murine and human orthologous genes were computed using the software program GraphPad Prism (version 7.00; GraphPad, La Jolla, CA).

Gene ontology and network analysis

The DEGs in murine and human samples were separately assigned to gene ontology (GO) categories using GO Enrichment analysis (Partek). The enriched GO biological processes were validated using the web-based tool DAVID (Database for Annotation, Visualization, and Integrated Discovery, National Cancer Institute, Frederick, MD) (10). To identify enriched pathways and transcriptional networks, the DEGs were further analyzed using the Pathway Enrichment module within Partek, which utilizes the KEGG databases (Kyoto Encyclopedia of Genes and Genomes) (9) to rank the DEGs by enrichment score [−log (p-value) derived from the contingency table; Fisher' Exact test].

The list of DEGs selected by the microarray analysis was then loaded separately in the web-based software application Ingenuity Pathway Analysis (QIAGEN, Redwood City, CA) to look for direct and indirect relationships at p<0.05 and experimental fold-change ≥ ±1.5. The data were computed to identify transcriptional networks associated with specific biological functions, diseases and signaling pathways (39). To estimate the degree of similarity between the GO pathways and biological processes activated in the murine and human datasets, we used the `comparison analysis' tool within the application (39).