Gut microbiota facilitates dietary heme-induced epithelial hyperproliferation by opening the mucus barrier in colon

Significance

Consumption of red meat is associated with increased colorectal cancer risk. We show that the gut microbiota is pivotal in this increased risk. Mice receiving a diet with heme, a proxy for red meat, show a damaged gut epithelium and a compensatory hyperproliferation that can lead to colon cancer. Mice receiving heme together with antibiotics do not show this damage and hyperproliferation. Our data indicate that microbial hydrogen sulfide opens the protective mucus barrier and exposes the epithelium to cytotoxic heme. Antibiotics block microbial sulfide production and thereby maintain the mucus barrier that prevents heme-induced hyperproliferation. Our study indicates that fecal trisulfide is a novel biomarker of mucus barrier integrity, which could be of relevance in human colon disease diagnostics.

Abstract

Colorectal cancer risk is associated with diets high in red meat. Heme, the pigment of red meat, induces cytotoxicity of colonic contents and elicits epithelial damage and compensatory hyperproliferation, leading to hyperplasia. Here we explore the possible causal role of the gut microbiota in heme-induced hyperproliferation. To this end, mice were fed a purified control or heme diet (0.5 μmol/g heme) with or without broad-spectrum antibiotics for 14 d. Heme-induced hyperproliferation was shown to depend on the presence of the gut microbiota, because hyperproliferation was completely eliminated by antibiotics, although heme-induced luminal cytotoxicity was sustained in these mice. Colon mucosa transcriptomics revealed that antibiotics block heme-induced differential expression of oncogenes, tumor suppressors, and cell turnover genes, implying that antibiotic treatment prevented the heme-dependent cytotoxic micelles to reach the epithelium. Our results indicate that this occurs because antibiotics reinforce the mucus barrier by eliminating sulfide-producing bacteria and mucin-degrading bacteria (e.g., Akkermansia). Sulfide potently reduces disulfide bonds and can drive mucin denaturation and microbial access to the mucus layer. This reduction results in formation of trisulfides that can be detected in vitro and in vivo. Therefore, trisulfides can serve as a novel marker of colonic mucolysis and thus as a proxy for mucus barrier reduction. In feces, antibiotics drastically decreased trisulfides but increased mucin polymers that can be lysed by sulfide. We conclude that the gut microbiota is required for heme-induced epithelial hyperproliferation and hyperplasia because of the capacity to reduce mucus barrier function.

Colorectal cancer, the second leading cause of cancer death in Western countries, is associated with diets high in red meat (1), whereas consumption of white meat does not have this association (2). Heme, the iron-porphyrin pigment, is present at much higher levels in red compared with white meat. Epidemiological studies show that heme intake is related to colon cancer risk (3, 4). Our previous studies show that when rodents consume heme, their colonic contents become more cytotoxic (5, 6). This increased cytotoxicity injures the colonic epithelial surface cells. To replace the injured surface cells, hyperproliferation from the stem cells in the crypts is initiated. Together with inhibition of apoptosis, this compensatory hyperproliferation leads to hyperplasia (6), which eventually can develop into colorectal cancer.

Dietary heme is poorly absorbed in the small intestine; ∼90% of dietary heme enters the colon (7). Besides the toxic effect of heme on the colonic mucosa, dietary heme affects the microbiota. The relationship between intestinal microbiota and colon cancer has long been suspected (8). In humans, a red meat diet increases Bacteroides spp. in feces (9). We recently showed that in mice, a heme diet changed the microbiota drastically, majorly increasing the Gram-negative bacteria (mainly Bacteroidetes, Proteobacteria, and Verrucomicrobia) (10). The gut microbiota can induce hyperproliferation via mechanisms occurring in the colon lumen, such as modulation of oxidative and cytotoxic stress or by influencing the mucus barrier. Oxidative stress induces the formation of peroxidized lipids, which react with heme to form the cytotoxic heme factor (CHF), thereby increasing cytotoxic stress (5, 11). In a time course study, we showed that there is a lag time in the formation of CHF and in the induction of hyperproliferation when mice are transferred from a control to heme diet (11). This lag time could be due to a time-dependent adaptation of the microbiota to the heme diet. Notably, heme does not increase cytotoxicity and epithelial hyperproliferation in the small intestine (12), indicating that formation of CHF only occurs in the colon where bacterial density is high. Moreover, these experiments suggested that CHF-induced hyperproliferation coincided with a reduced mucus barrier function (11), leading to enhanced contact of colonocytes with microbiota and toxic substances. In the present study, we investigate whether bacteria play a causal role in heme-induced cytotoxicity and hyperproliferation by using broad-spectrum antibiotics (Abx). Our results illustrate the crucial role of the gut microbiota in heme-induced hyperproliferation and indicate the involvement of microbial hydrogen sulfide formation in reduction of the polymeric mucin network to degrade the protective mucus barrier and expose the colon mucosa to CHF.

Results

Heme- and Abx-Induced Changes in the Colonic Lumen.

Mice were divided into four groups receiving either a control diet (C-group), a heme diet (H-group), a control diet with Abx treatment (CA-group), or a heme diet with Abx treatment (HA-group). After 2 wk of intervention the H- and HA-group had a lower body weight compared with their controls (Table 1). Fecal dry and wet weight was significantly increased in the HA-group.

Table 1.

Effects of heme and Abx on body weight and fecal parameters

Variable	Control	Heme	Control + Abx	Heme + Abx
Body weight (g)	27.7 ± 0.5a	24.9 ± 0.5b	27.2 ± 0.3a	24.8 ± 0.4b
Fecal wet weight (g/d)	0.49 ± 0.08a	0.60 ± 0.05a	0.62 ± 0.09a	1.20 ± 0.19b
Fecal dry weight (g/d)	0.12 ± 0.01a	0.11 ± 0.01a	0.12 ± 0.01a	0.19 ± 0.02b
TBARS (MDA equivalents, µmol/L)	12.70 ± 1.43a	59.84 ± 2.46b	11.06 ± 1.63a	46.06 ± 3.92c
Cytotoxicity (% lysis)	1.09 ± 0.45a	66.90 ± 10.45b	0.05 ± 0.04a	31.30 ± 8.98c

Data are represented as mean ± SEM (n = 9/group). Differences between the groups were tested by ANOVA with a Bonferroni post hoc test and different superscripts indicate significant differences (P < 0.05).

To confirm that Abx decreased the abundance of bacteria, quantitative PCR (qPCR) analyses with specific primers targeting total bacteria, Bacteroidetes, or Firmicutes were performed (Fig. 1A). Abx treatment significantly reduced the abundance of total bacteria and Firmicutes ∼100-fold and Bacteroidetes ∼1,000-fold. Moreover, the H-group had a significantly increased abundance of Bacteroidetes compared with the C-group, which corroborates previous observations (10). Abx thus drastically decreased the bacterial density, affecting both Gram-positive Firmicutes and Gram-negative Bacteroidetes. To study how this impacts the normal microbial modification of host compounds, we determined the fecal bile acid composition. No conjugated bile acids were detected in fecal water in the C- and H-group, where unconjugated and secondary bile acids were predominant (Fig. 1B). However, with Abx almost all bile acids were primary and conjugated with glycine or taurine (ratio about 3), showing that Abx blocked microbial bile deconjugation and dehydroxylation almost completely.

Fig. 1.

(A) Counts of total bacteria, Bacteroidetes, and Firmicutes measured by qPCR. Control bars are set at 100%, and other bars are relative to controls; mean ± SEM (n = 9/group). (B) Bile acid profiles determined by HPLC; mean ± SEM (n = 3/group). Letters indicate significant different groups (P < 0.05), ANOVA with Bonferroni post hoc test. Gly, glycine conjugated; Sec, secondary; Tau, taurine conjugated.

Heme increases oxidative and cytotoxic stress in the colon (10, 11). Reactive oxygen species (ROSs) induce the formation of lipid peroxides that react with heme to form CHF, thereby increasing the cytotoxicity of luminal contents (5, 11). We determined lipid peroxidation product levels by measuring thiobarbituric acid reactive substances (TBARS) in fecal water. TBARS were low in the C- and CA-group (Table 1) and increased significantly and to a similar extent in the H- and HA-group, implying that heme, both in presence and absence of Abx, induced ROS stress. Analogously, fecal water cytotoxicity (Table 1) was significantly increased in the H- and HA-group compared with their controls. Because Abx drastically reduced microbiota density (100- to 1,000-fold), but only slightly reduced cytotoxicity (2-fold) and TBARS (1.3-fold), it is unlikely that bacteria play a major role in the formation of TBARS and cytotoxicity.

Heme- and Abx-Induced Changes in the Colonic Mucosa.

Morphology analyses of H&E-stained colon tissue (Fig. 2A) confirmed the previously reported heme-induced increased crypt depth (H- vs. C-group). Abx treatment did not affect the colon morphology in the CA- vs. C-group but completely restored tissue morphology in the HA- vs. H-group. The crypt depth increase in the H-group did not result from inflammation because neutrophil and macrophage infiltration in the lamina propria was comparable to the C-group. Analogous to earlier reports (6), cell proliferation quantification using Ki67 staining (Fig. 2 B and C) shows that the heme diet strongly induced cell proliferation (H- vs. C-group), leading to expansion of the proliferative compartment and increased crypt depth. Abx treatment led to slightly reduced numbers of cells per crypt in the CA- vs. C-group, but did not significantly affect their labeling index or amount of proliferative cells. However, Abx treatment in the heme diet (HA- vs. H-group) completely suppressed heme-induced hyperproliferation and hyperplasia to levels observed in the C- and CA-groups (Fig. 2C). In conclusion, heme-induced hyperproliferation and hyperplasia in mouse colon only occurs in the presence of the gut microbiota.

Fig. 2.

(A) Histochemical H&E staining and (B) immunohistochemical Ki67 staining of colon of control and heme-fed mice. (C) Quantification of Ki67-positive cells per crypt, total number of cells per crypt, and labeling index (percentage of proliferative cells per crypt); mean ± SEM (n = 9/group). Letters indicate significant different groups (P < 0.05), ANOVA with Bonferroni post hoc test.

Abx Block the Heme-Induced Expression of Cell Cycle Genes.

Using whole genome transcriptomics we investigated whether the physiological changes were reflected in gene expression profiles. The heme diet (H- vs. C-group) led to 5,507 differentially expressed genes (q < 0.01), of which almost 90% (4,859) were not significantly affected in the HA vs. CA comparison (Fig. S1A). The 4,859 genes specific for the H-group were analyzed by GSEA, indicating that mainly cell cycle related processes were affected by heme (Fig. S1B). Moreover, mining of these genes for the involved transcription factors (Fig. S1C), revealed that Cdkn2a, Smarcb1, and the tumor suppressors Tp53 and Rb1 were inhibited, whereas oncogenes such as Myc and Foxm1, as well as cell cycle regulators E2f1 and Tbx2, were activated by heme. Importantly, these processes and transcription factors were not modulated in the HA-group compared with the CA-group. There were only 369 differentially expressed genes unique for the HA-group (Fig. S1A). Notably, none of the modulated processes identified in the HA-group related to the end points of our study. Because of the specific heme-Abx interaction, the Abx-mediated differential gene expression profiles and processes were substantially different in the heme diet background (Tables S1 and S2) compared with the control diet background (Tables S3 and S4). These observations indicate that heme-induced mucosal gene expression changes of cell cycle-related processes require the presence of the microbiota, which is in agreement with the microbiota requirement for the increased labeling index (Fig. 2C).

Fig. S1.

(A) Venn diagram showing the numbers of heme and heme plus Abx-specific regulated genes and of overlapping genes (q < 0.01). (B) GSEA showing positive enriched processes. Sources of gene sets are indicated in superscript and represent (1) gene ontology, (2) reactome, and (3) KEGG. Size represents the number of genes in the gene-set and NES is normalized enrichment score, which is the enrichment score for the gene set after it has been normalized to account for variations in gene set size. (C) Ingenuity analysis showing activated or inhibited transcription factors (z-score > 2 and P < 0.05) in both treatments specifically and in overlapping genes.

Table S1.

Fifty highest up-regulated genes (q < 0.01) in mouse colon scrapings for the comparison of heme Abx vs. heme

Gene symbol	Gene description	SI ± SEM				Fold change (heme + Abx vs. heme)	SI ± SEM				Fold change (control + Abx vs. control)
		Heme		Heme + Abx			Control		Control + Abx
		SI	SEM	SI	SEM		SI	SEM	SI	SEM
Gbp2	Guanylate binding protein 2	96	24	412	102	4.0	303	77	962	234	NS
Slc6a19	Solute carrier family 6 (neurotransmitter transporter), member 19	81	14	320	53	3.9	21	2	60	6	2.8
Enpep	Glutamyl aminopeptidase	145	10	557	41	3.8	209	21	265	18	NS
Slc5a4b	Solute carrier family 5 (neutral amino acid transporters, system A), member 4b	52	5	197	23	3.7	14	0	16	1	NS
Abcc2	ATP-binding cassette, subfamily C (CFTR/MRP), member 2	115	15	425	49	3.7	40	5	42	3	NS
Ace2	Angiotensin I converting enzyme (peptidyl-dipeptidase A) 2	67	7	239	32	3.5	93	6	165	4	1.8
Cyp3a25	Cytochrome P450, family 3, subfamily a, polypeptide 25	21	1	79	14	3.5	21	3	15	1	NS
Sis	Sucrase isomaltase (alpha-glucosidase)	29	3	105	23	3.3	117	19	147	16	NS
2010106E10Rik	RIKEN cdna 2010106E10 gene	60	7	191	13	3.2	70	5	82	3	NS
Trac	T-cell receptor alpha constant	97	14	274	27	2.8	127	10	320	24	2.5
1810030J14Rik	RIKEN cDNA 1810030J14 gene	666	76	1,664	82	2.5	2,451	249	2,909	158	NS
Cyp4f39	Cytochrome P450, family 4, subfamily f, polypeptide 39	38	4	92	6	2.5	23	2	23	1	NS
Dbp	D site albumin promoter binding protein	111	22	274	40	2.4	608	33	399	60	NS
Naaladl1	N-acetylated alpha-linked acidic dipeptidase-like 1	522	7	1,275	71	2.4	880	88	925	50	NS
Slc13a1	Solute carrier family 13 (sodium/sulfate symporters), member 1	141	22	336	25	2.4	786	120	974	71	NS
Rsad2	Radical S-adenosyl methionine domain containing 2	57	9	139	21	2.4	56	1	152	15	2.7
Npc1l1	NPC1-like 1	37	3	100	23	2.4	33	2	49	4	NS
Gcg	Glucagon	586	25	1,354	60	2.3	491	22	580	7	NS
Sstr1	Somatostatin receptor 1	112	8	255	8	2.3	234	19	226	6	NS
Gm10768	Predicted gene 10768	10	1	24	3	2.2	7	1	8	0	NS
Per3	Period homolog 3 (Drosophila)	81	10	174	19	2.1	286	14	198	27	NS
Cyp2f2	Cytochrome P450, family 2, subfamily f, polypeptide 2	107	5	223	13	2.1	426	32	218	13	−2.0
Ddx60	DEAD (Asp-Glu-Ala-Asp) box polypeptide 60	83	14	173	26	2.1	115	10	486	78	4.1
Gbp6	Guanylate binding protein 6	90	9	191	29	2.0	188	27	323	49	NS
Ccl5	Chemokine (C-C motif) ligand 5	99	5	203	20	2.0	163	7	267	18	1.6
Tgm3	Transglutaminase 3, E polypeptide	1,397	179	2,766	166	2.0	4,496	132	3,340	110	NS
Ang4	Angiogenin, ribonuclease A family, member 4	158	6	320	27	2.0	570	126	581	81	NS
Oasl2	2'-5′ Oligoadenylate synthetase-like 2	217	39	432	66	2.0	247	12	905	123	3.6
Cml5	Camello-like 5	16	0	31	2	2.0	20	1	19	1	NS
Rtp4	Receptor transporter protein 4	93	14	187	29	2.0	111	10	331	42	3.0
Slc37a2	Solute carrier family 37 (glycerol-3-phosphate transporter), member 2	329	41	641	58	2.0	933	67	830	76	NS
Apob	Apolipoprotein B	501	34	982	67	1.9	752	26	797	41	NS
Gm129	Predicted gene 129	50	5	98	11	1.9	160	15	104	13	NS
Slc6a20a	Solute carrier family 6 (neurotransmitter transporter), member 20A	98	5	190	6	1.9	234	12	228	10	NS
Rnf212	Ring finger protein 212	124	5	245	23	1.9	132	16	117	6	NS
Ces1d	Carboxylesterase 1D	343	21	654	25	1.9	335	32	284	8	NS
Slc10a2	Solute carrier family 10, member 2	169	20	321	25	1.9	376	63	536	46	NS
Gsdmc3	Gasdermin C3	1,006	92	1,906	101	1.9	2744	138	2,631	181	NS
Ces1g	Carboxylesterase 1G	1,947	44	3,710	177	1.9	757	138	766	47	NS
Edn1	Endothelin 1	212	43	380	20	1.9	495	22	267	32	−1.9
Cck	Cholecystokinin	47	2	88	5	1.9	44	1	42	2	NS
Gdpd2	Glycerophosphodiester phosphodiesterase domain containing 2	78	4	145	7	1.9	97	14	114	8	NS
Ifit3	IFN-induced protein with tetratricopeptide repeats 3	55	3	106	13	1.9	79	3	210	45	2.5
G6pc	Glucose-6-phosphatase, catalytic	19	2	37	5	1.9	14	1	16	1	NS
Slc7a8	Solute carrier family 7 (cationic amino acid transporter, y+ system), member 8	135	6	250	14	1.8	296	25	235	21	NS
Bmp3	Bone morphogenetic protein 3	270	8	498	21	1.8	564	49	395	21	−1.4
Trim30d	Tripartite motif-containing 30D	96	10	176	13	1.8	248	16	369	47	NS
Irf7	IFN regulatory factor 7	355	21	649	26	1.8	495	19	1,052	97	2.1
Slc35e3	Solute carrier family 35, member E3	156	8	285	14	1.8	167	12	106	5	−1.6
Ccl8	Chemokine (C-C motif) ligand 8	30	3	55	6	1.8	22	3	81	8	3.7

Table includes only genes with a signal intensity > 20 in at least one condition. n = 4 per group for Control, Control Abx, and Heme; n = 6 per group for Heme Abx. SI, signal intensity.

Table S2.

Fifty highest down-regulated genes (q < 0.01) in mouse colon scrapings for the comparison of heme Abx vs. heme

Gene symbol	Gene description	SI ± SEM				Fold change (heme + Abx vs. heme)	SI ± SEM				Fold change (control + Abx vs. control)
		Heme		Heme + Abx			Control		Control + Abx
		SI	SEM	SI	SEM		SI	SEM	SI	SEM
Reg3b	Regenerating islet-derived 3 beta	1,385	309	66	7	−19.5	566	297	84	19	NS
Reg3g	Regenerating islet-derived 3 gamma	1,115	170	118	10	−9.2	337	167	125	20	NS
Slpi	Secretory leukocyte peptidase inhibitor	511	58	167	14	−3.1	60	4	118	28	1.9
AI747448	Expressed sequence AI747448	570	94	225	49	−2.7	71	12	187	21	2.7
Capg	Capping protein (actin filament), gelsolin-like	724	32	272	13	−2.7	164	11	196	18	NS
Dhrs9	Dehydrogenase/reductase (SDR family) member 9	639	56	243	17	−2.6	227	15	190	29	NS
Trim29	Tripartite motif-containing 29	197	8	76	6	−2.6	29	1	38	4	NS
Spp1	Secreted phosphoprotein 1	51	5	20	1	−2.6	17	0	22	2	NS
Fabp1	Fatty acid binding protein 1, liver	606	99	249	22	−2.4	19	1	17	1	NS
Mfsd2a	Major facilitator superfamily domain containing 2A	138	20	60	6	−2.3	51	5	71	11	NS
Mcpt2	Mast cell protease 2	130	17	57	5	−2.3	26	2	33	3	NS
Nov	Nephroblastoma overexpressed gene	654	64	305	35	−2.2	695	79	598	16	NS
Lpcat4	Lysophosphatidylcholine acyltransferase 4	460	28	215	9	−2.1	244	6	281	13	NS
Nt5dc2	5′-nucleotidase domain containing 2	232	12	110	5	−2.1	100	3	112	9	NS
Mcpt1	Mast cell protease 1	65	9	31	3	−2.1	24	2	29	2	NS
Fer1l4	Fer-1-like 4 (C. Elegans)	83	9	41	2	−2.0	27	2	46	4	1.7
Il1rl1	Interleukin 1 receptor-like 1	63	5	31	1	−2.0	20	1	39	3	2.0
Tnfrsf8	Tumor necrosis factor receptor superfamily, member 8	66	6	34	1	−1.9	23	1	42	6	1.8
Duoxa2	Dual oxidase maturation factor 2	137	4	73	3	−1.9	177	27	151	22	NS
Fads3	Fatty acid desaturase 3	316	17	173	6	−1.8	162	6	150	12	NS
Cpn1	Carboxypeptidase N, polypeptide 1	402	22	226	18	−1.8	264	28	256	19	NS
Hist1h2ab	Histone cluster 1, h2ab	205	19	115	10	−1.8	68	5	117	12	1.7
E2f2	E2F transcription factor 2	191	6	109	5	−1.8	60	3	110	12	1.8
Rdh9	Retinol dehydrogenase 9	127	4	73	5	−1.8	103	6	78	3	NS
Pnpla1	Patatin-like phospholipase domain containing 1	74	9	42	3	−1.7	57	5	49	5	NS
Fkbp11	FK506 binding protein 11	69	2	40	1	−1.7	43	3	49	1	NS
4632434I11Rik	RIKEN cdna 4632434I11 gene	89	5	52	3	−1.7	38	1	57	5	1.5
Plau	Plasminogen activator, urokinase	68	7	40	4	−1.7	48	7	60	5	NS
Pla2g2a	Phospholipase A2, group IIA (platelets, synovial fluid)	103	7	60	3	−1.7	43	3	68	3	1.6
Spon2	Spondin 2, extracellular matrix protein	114	13	66	5	−1.7	71	3	78	7	NS
Tat	Tyrosine aminotransferase	191	21	111	6	−1.7	439	31	268	6	−1.6
Plvap	Plasmalemma vesicle associated protein	171	5	101	5	−1.7	114	17	111	10	NS
2310016C08Rik	RIKEN cdna 2310016C08 gene	108	16	62	4	−1.7	63	2	58	3	NS
Nrg1	Neuregulin 1	255	9	155	14	−1.7	83	10	199	14	2.4
Myl7	Myosin, light polypeptide 7, regulatory	55	3	33	2	−1.7	33	2	45	4	NS
Expi	Extracellular proteinase inhibitor	53	3	32	1	−1.7	43	3	33	2	NS
Nek2	NIMA (never in mitosis gene a)-related expressed kinase 2	288	13	177	11	−1.6	98	6	174	10	1.8
Duox2	Dual oxidase 2	391	25	237	6	−1.6	611	66	490	56	NS
Bmp8b	Bone morphogenetic protein 8b	152	7	93	5	−1.6	66	4	68	3	NS
Cdca3	Cell division cycle associated 3	465	18	285	13	−1.6	168	7	282	14	1.7
Ripk3	Receptor-interacting serine-threonine kinase 3	315	16	194	7	−1.6	166	6	237	4	1.4
Rbp4	Retinol binding protein 4, plasma	209	1	129	4	−1.6	178	18	189	13	NS
4930547N16Rik	RIKEN cdna 4930547N16 gene	144	6	90	6	−1.6	54	3	86	8	1.6
Aurka	Aurora kinase A	290	10	180	10	−1.6	111	5	178	19	1.6
Igsf10	Ig superfamily, member 10	142	7	88	3	−1.6	110	14	134	16	NS
Eef1g	Eukaryotic translation elongation factor 1 gamma	143	13	88	3	−1.6	76	7	108	10	1.4
Dlgap5	Discs, large (Drosophila) homolog-associated protein 5	171	5	107	6	−1.6	77	4	122	10	1.6
Cdc25c	Cell division cycle 25 homolog C (S. Pombe)	85	4	53	3	−1.6	34	1	54	4	1.6
Dkkl1	Dickkopf-like 1	34	1	21	1	−1.6	23	0	25	1	NS
Cenpa	Centromere protein A	393	16	247	11	−1.6	170	9	256	13	1.5

Table includes only genes with a signal intensity > 20 in at least one condition. n = 4 per group for Control, Control Abx, and Heme; n = 6 per group for Heme Abx.

Table S3.

Fifty highest up-regulated genes (q < 0.01) in mouse colon scrapings for the comparison of control Abx vs. control

Gene symbol	Gene description	SI ± SEM				Fold change (heme + Abx vs. heme)	SI ± SEM				Fold change (control + Abx vs. control)
		Heme		Heme + Abx			Control		Control + Abx
		SI	SEM	SI	SEM		SI	SEM	SI	SEM
Ighv1-85	Ig heavy variable 1–85	99	40	64	16	NS	20	8	135	62	5.7
Ighv8-12	Ig heavy variable V8-12	561	61	439	35	NS	196	41	1,111	246	5.6
Cxcl9	Chemokine (C-X-C motif) ligand 9	75	12	236	59	NS	62	18	326	47	5.6
Gm12250	Predicted gene 12250	32	6	81	22	NS	37	9	190	39	5.1
Ifi44	IFN-induced protein 44	70	7	124	21	NS	66	6	340	82	4.8
Oas3	2'-5′ oligoadenylate synthetase 3	50	5	97	14	NS	51	5	241	41	4.6
Ddx60	DEAD (Asp-Glu-Ala-Asp) box polypeptide 60	83	14	173	26	2.1	115	10	486	78	4.1
Usp18	Ubiquitin specific peptidase 18	67	9	89	16	NS	48	4	204	39	4.1
Ccl8	Chemokine (C-C motif) ligand 8	30	3	55	6	1.8	22	3	81	8	3.7
Gm5431	Predicted gene 5431	17	3	61	15	NS	49	11	185	48	3.7
Oas2	2'-5′ oligoadenylate synthetase 2	63	8	94	13	NS	56	3	215	42	3.6
Oasl2	2'-5′ oligoadenylate synthetase-like 2	217	39	432	66	2.0	247	12	905	123	3.6
Ifi47	IFN gamma inducible protein 47	69	14	128	28	NS	79	16	265	44	3.4
H2-DMa	Histocompatibility 2, class II, locus dma	112	8	175	17	NS	99	9	340	50	3.3
Gm9740	Predicted gene 9740	116	50	191	43	NS	76	8	251	37	3.3
Igkv1-117	Ig kappa chain variable 1–117	522	13	518	54	NS	355	57	1,201	259	3.2
Apol9b	Apolipoprotein L 9b	22	2	37	3	NS	21	4	66	11	3.2
Ifit1	IFN-induced protein with tetratricopeptide repeats 1	129	21	255	39	NS	188	13	610	94	3.1
Apol7c	Apolipoprotein L 7c	43	7	41	6	NS	29	3	89	7	3.1
Apol9a	Apolipoprotein L 9a	73	4	110	13	NS	68	7	213	34	3.1
H2-Q6	Histocompatibility 2, Q region locus 6	185	8	303	25	1.6	178	13	561	91	3.1
Ubd	Ubiquitin D	18	1	28	4	NS	17	1	58	15	3.0
H2-Aa	Histocompatibility 2, class II antigen A, alpha	525	29	791	55	NS	482	47	1,484	213	3.0
H2-Q9	Histocompatibility 2, Q region locus 9	104	11	160	16	NS	88	3	279	48	3.0
Igtp	IFN gamma induced gtpase	47	7	107	26	NS	71	13	222	44	3.0
Rtp4	Receptor transporter protein 4	93	14	187	29	2.0	111	10	331	42	3.0
Gm4951	Predicted gene 4951	28	2	54	10	NS	35	6	118	37	2.9
Slc6a19	Solute carrier family 6 (neurotransmitter transporter), member 19	81	14	320	53	3.9	21	2	60	6	2.8
H2-Ab1	Histocompatibility 2, class II antigen A, beta 1	698	30	965	62	NS	607	68	1,677	189	2.8
Herc6	Hect domain and RLD 6	69	8	135	19	NS	118	9	340	65	2.7
Cd74	CD74 antigen (invariant polypeptide of major histocompatibility complex, class II antigen-associated)	877	38	1160	94	NS	775	70	2,112	221	2.7
Gbp3	Guanylate binding protein 3	43	5	71	12	NS	56	9	157	31	2.7
AI747448	Expressed sequence AI747448	570	94	225	49	−2.7	71	12	187	21	2.7
Zbp1	Z-DNA binding protein 1	98	13	148	23	NS	142	21	395	76	2.7
Psmb8	Proteasome (prosome, macropain) subunit, beta type 8 (large multifunctional peptidase 7)	70	7	120	14	NS	88	11	249	47	2.7
Rsad2	Radical S-adenosyl methionine domain containing 2	57	9	139	21	2.4	56	1	152	15	2.7
Tap1	Transporter 1, ATP-binding cassette, subfamily B (MDR/TAP)	134	11	243	36	NS	161	18	439	72	2.6
Gm5571	Predicted gene 5571	429	102	439	71	NS	250	29	691	159	2.6
Dnase1l3	Dnase 1-like 3	59	3	104	9	1.7	53	2	138	5	2.6
Nlrc5	NLR family, CARD domain containing 5	45	3	68	7	NS	42	3	110	16	2.5
Trac	T-cell receptor alpha constant	97	14	274	27	2.8	127	10	320	24	2.5
Gbp8	Guanylate-binding protein 8	43	3	73	8	NS	47	6	120	17	2.5
Ifit3	IFN-induced protein with tetratricopeptide repeats 3	55	3	106	13	1.9	79	3	210	45	2.5
Ifit2	IFN-induced protein with tetratricopeptide repeats 2	28	1	39	4	NS	32	1	87	21	2.5
Upp1	Uridine phosphorylase 1	322	46	542	66	NS	518	53	1,280	148	2.5
Tfrc	Transferrin receptor	493	105	567	38	NS	288	2	708	30	2.4
Dhx58	DEXH (Asp-Glu-X-His) box polypeptide 58	81	7	111	10	NS	74	4	184	23	2.4
Nrg1	Neuregulin 1	255	9	155	14	−1.7	83	10	199	14	2.4
H2-DMb1	Histocompatibility 2, class II, locus Mb1	28	1	44	2	NS	30	5	73	12	2.4
Irgm2	Immunity-related gtpase family M member 2	146	22	343	68	NS	250	34	597	87	2.3

Table includes only genes with a signal intensity > 20 in at least one condition. n = 4 per group for Control, Control Abx, and Heme; n = 6 per group for Heme Abx.

Table S4.

Fifty highest down-regulated genes (q < 0.01) in mouse colon scrapings for the comparison of control Abx vs. control

Gene symbol	Gene description	SI ± SEM				Fold change (heme + Abx vs. heme)	Signal intensity (SI) ± SEM.				Fold change (control + Abx vs. control)
		Heme		Heme + Abx			Control		Control + Abx
		SI	SEM	SI	SEM		SI	SEM	SI	SEM
Des	Desmin	622	84	933	220	NS	2,474	559	674	195	−3.8
Cnn1	Calponin 1	629	95	955	216	NS	2,419	531	680	205	−3.7
Tacr2	Tachykinin receptor 2	87	9	128	23	NS	330	77	93	21	−3.4
Calb2	Calbindin 2	24	3	41	6	NS	102	23	29	5	−3.4
Synpo2	Synaptopodin 2	132	23	188	38	NS	512	123	143	27	−3.3
Actg2	Actin, gamma 2, smooth muscle, enteric	653	81	1,020	213	NS	2,343	483	748	219	−3.3
Pgm5	Phosphoglucomutase 5	286	42	445	89	NS	978	211	295	68	−3.2
Hspb8	Heat shock protein 8	78	11	111	18	NS	236	47	73	12	−3.1
Hspb7	Heat shock protein family, member 7 (cardiovascular)	51	4	65	8	NS	183	39	56	8	−3.1
Chrm2	Cholinergic receptor, muscarinic 2, cardiac	95	13	124	26	NS	280	56	93	27	−3.1
Kcnmb1	Potassium large conductance calcium-activated channel, subfamily M, beta member 1	88	8	108	15	NS	255	52	80	15	−3.1
Kcnip4	Kv channel interacting protein 4	21	3	32	6	NS	86	23	26	6	−3.1
Stmn2	Stathmin-like 2	50	5	89	16	NS	210	40	67	10	−3.1
Flnc	Filamin C, gamma	92	6	114	18	NS	285	63	88	14	−3.0
Hspb6	Heat shock protein, alpha-crystallin-related, B6	263	21	325	48	NS	784	166	255	57	−3.0
Myl9	Myosin, light polypeptide 9, regulatory	780	90	1,206	215	NS	2,647	502	885	181	−3.0
Tagln	Transgelin	319	40	457	89	NS	1,035	222	338	65	−2.9
Atp2b4	Atpase, Ca++ transporting, plasma membrane 4	125	15	198	39	NS	465	103	152	25	−2.9
Myh11	Myosin, heavy polypeptide 11, smooth muscle	573	83	886	183	NS	2,068	413	698	135	−2.9
Nmu	Neuromedin U	17	1	27	4	NS	73	13	26	6	−2.9
Chrna3	Cholinergic receptor, nicotinic, alpha polypeptide 3	41	4	70	11	NS	165	37	54	9	−2.9
Bves	Blood vessel epicardial substance	41	4	53	7	NS	125	27	42	6	−2.8
Grp	Gastrin releasing peptide	28	2	49	8	NS	125	28	43	8	−2.8
Cd59a	CD59a antigen	69	9	105	20	NS	231	55	76	10	−2.8
Popdc2	Popeye domain containing 2	40	4	56	9	NS	126	26	45	11	−2.8
Htr3a	5-hydroxytryptamine (serotonin) receptor 3A	20	1	26	2	NS	53	10	18	3	−2.8
Scube2	Signal peptide, CUB domain, EGF-like 2	37	4	48	8	NS	112	28	36	4	−2.7
Rit2	Ras-like without CAAX 2	21	1	28	3	NS	67	16	23	3	−2.7
Klk15	Kallikrein related-peptidase 15	23	2	31	4	NS	95	12	35	4	−2.7
Pcp4l1	Purkinje cell protein 4-like 1	103	9	152	27	NS	319	54	119	24	−2.7
Pdlim3	PDZ and LIM domain 3	394	42	522	88	NS	1,175	237	417	63	−2.7
Cpe	Carboxypeptidase E	100	10	139	16	NS	344	60	126	19	−2.7
Syt1	Synaptotagmin I	35	3	56	7	NS	132	27	47	8	−2.7
Cacnb2	Calcium channel, voltage-dependent, beta 2 subunit	44	3	65	10	NS	148	29	54	9	−2.7
Ppyr1	Pancreatic polypeptide receptor 1	24	2	32	6	NS	75	16	26	4	−2.7
Myom1	Myomesin 1	43	4	55	7	NS	133	31	45	3	−2.6
Rbpms2	RNA binding protein with multiple splicing 2	52	1	78	10	NS	144	30	52	8	−2.6
Snap25	Synaptosomal-associated protein 25	73	6	114	13	NS	259	45	98	15	−2.6
Cryab	Crystallin, alpha B	70	4	87	13	NS	218	44	83	15	−2.5
Penk	Preproenkephalin	70	9	110	17	NS	264	53	99	12	−2.5
Art3	ADP ribosyltransferase 3	49	6	67	9	NS	132	26	50	6	−2.5
Mir145	Microrna 145	100	18	131	22	NS	325	69	124	12	−2.5
Slc30a10	Solute carrier family 30, member 10	47	3	49	4	NS	256	34	105	17	−2.5
Cckar	Cholecystokinin A receptor	21	1	31	5	NS	68	13	27	4	−2.5
Slc2a4	Solute carrier family 2 (facilitated glucose transporter), member 4	42	5	69	9	NS	122	23	48	5	−2.5
Tmem90b	Transmembrane protein 90B	36	3	48	6	NS	112	25	42	4	−2.4
Myocd	Myocardin	65	7	79	9	NS	176	30	70	9	−2.4
Slc7a14	Solute carrier family 7 (cationic amino acid transporter, y+ system), member 14	30	2	46	7	NS	90	19	35	4	−2.4
Htr2b	5-hydroxytryptamine (serotonin) receptor 2B	14	1	22	3	NS	49	12	19	2	−2.4
Ckb	Creatine kinase, brain	525	31	486	70	NS	931	183	372	51	−2.4

Table includes only genes with a signal intensity > 20 in at least one condition. n = 4 per group for Control, Control Abx, and Heme; n = 6 per group for Heme Abx.

Abx Do Not Affect the Heme-Induced Antioxidant Response.

A set of 648 genes was significantly regulated in both the H- and HA-group compared with their controls (Fig. S1A). Of those shared genes, 599 were similarly regulated in both groups. Notably, this group of genes included the activation of several transcription factors, including the PPARs, involved in fatty acid metabolism, and Nrf2, involved in antioxidant response (Fig. S1 B and C). These data imply that oxidative stress and lipid peroxidation products induced the antioxidant response and the induction of PPAR target genes in the mucosa, irrespective of the addition of Abx to the heme diet.

Abx Block the Mucosal Sensing of Heme-Induced Luminal Cytotoxicity.

In the colon, the mucus preserves the epithelial barrier by protecting the surface epithelial cells through creating a diffusion barrier for toxic micelles (13). In the HA-group, substantial luminal cytotoxicity was present but did not induce hyperproliferation, indicating that CHF did not reach the surface epithelial cells. We investigated epithelial sensing of cytotoxicity using the marker genes survivin (Birc5), immediately early response 3 (Ier3), the necrosis facilitator Ripk3, and secretory leukocyte protease inhibitor (Slpi), which are up-regulated by cell injury as described earlier (11, 14). In the present study, heme up-regulated the expression of these cytotoxicity sensors only in the H-group but not in the HA-group (Fig. 3A). In addition, Fig. 3B shows that heme only injured surface cells, which corroborates our laser capture analysis (6). Taken together, these results indicate that the microbiota is required for the cytotoxic micelles to reach the surface epithelium, as the diffusion barrier of the mucus layer was maintained by Abx.

Fig. 3.

(A) Gene expression of injury markers Birc5, Ier3, Ripk3, and Slpi. (B) immunohistochemical colonic Slpi staining of control and heme-fed mice. (C) Gene expression of mucin genes 1–4 and Galnt 3 and 12. Expression levels of control is set at 1. Expression of other bars is relative to controls; mean ± SEM (n = 4 for C, H, and CA; n = 6 for HA). Letters indicate significant differences (P < 0.05), ANOVA with Bonferroni post hoc test.

An increased mucus barrier could be caused by increased mucin synthesis or by decreased mucin degradation. Expression levels of secreted Muc2 and cell-associated Muc4 were decreased in the HA- compared with the H-group (Fig. 3C). Moreover, the Kyoto Encyclopedia of Genes and Genomes pathway mucin type O-glycan biosynthesis was significantly repressed in the HA- compared with the H-group, according to GSEA analysis (q = 0.049). For instance, Galnt 3 and 12, involved in the first step of o-glycosylation, were down-regulated by Abx (Fig. 3C), indicating that mucus layer production is most probably decreased by Abx. Regarding mucus degradation, it is well established that the microbiota degrades mucins and uses their carbohydrates and amino acids as substrates for growth. In addition, sulfate-reducing bacteria (SRBs) use mucin-derived sulfate as electron acceptor in anaerobic respiration forming sulfides. As the Abx treatment drastically reduced the microbiota density, we hypothesized that Abx increased the mucus barrier by preventing microbial degradation of mucin. Notably, the abundance of the mucin-degrading Akkermansia muciniphila was eightfold increased by heme, whereas Abx reduced the abundance of this mucin degrader more than 1,000-fold (Fig. 4A). This result implies that the relative reduction of the Akkermansia population by Abx exceeded the overall effect of Abx on the total microbial load, supporting that Abx treatment could increase mucus barrier function by strongly reducing the levels of mucin degraders such as Akkermansia.

Fig. 4.

(A) Bacterial counts of A. muciniphila and SRB determined by qPCR. Controls are set at 100%, and other bars are relative to controls; mean ± SEM (n = 8–9/group). Letters indicate significant different groups (P < 0.05). (B) Rate and extent of S-S bond splitting and synthesis of trisulfide bonds; mean ± SD (n = 3–6/group). *Significant difference with thiol groups (P < 0.05). (C) Reaction scheme by which sulfide splits S-S bonds. (D) Concentrations of sulfides in fecal water; mean ± SEM (n = 8–9/group). (E) Excretion of fecal mucins, expressed as µmoles O-glycan per day (n = 6–9/group). (F) Western blot analysis of fecal mucin with or without DTT as reducing agent and with sulfide. Samples were stained with anti-Muc2 antibody. Each lane represents a pool of n = 9/group. Letters indicate significant different groups (P < 0.05), ANOVA with Bonferroni post hoc test.

Sulfide Reduces S-S Bonds, Leading to Mucolysis.

Mucus is high in intra- and intermolecular S-S bonds, stabilizing its polymeric, network-like structure. When S-S bonds are broken, mucin monomers dissociate and/or denaturate, leading to decreased viscosity and higher mucus accessibility for bacterial degradation (15). SRBs can use mucus-derived sulfate as oxidant in anaerobic respiration, generating sulfide. We hypothesized that sulfide could have a mucolytic effect by reducing the intermolecular S-S bonds, which contributes to an enhanced mucus barrier in the Abx groups due to the decreased abundance of SRBs (Fig. 4A) and the corresponding decrease of luminal sulfide production. We tested the reducing potency of several sulfur containing compounds on the model compound DTNB [5,5′-dithiobis-(2-nitrobenzoic acid)] containing a central S-S bond. Splitting of this S-S bond in DTNB leads to increased absorbance at 412 nm, and the assay allows the determination of the overall S-S splitting extent as well as the initial S-S splitting rate (Fig. 4B). Cysteine, glutathione, and N-acetylcysteine (NAC) were able to split the S-S bond with a similar rate between 24 and 31 µM/min, whereas the negative control Na₂SO₄ did not affect DTNB integrity. Importantly, sulfide gave a significantly higher rate of S-S bond splitting, 62.4 ± 5.9 µM/min, indicating that sulfide has a twofold more potent mucolytic effect compared with the amino acid thiols that have been shown to split S-S bonds and make mucins less viscous (16). Moreover, the overall extent of S-S bond splitting by sulfide was also twofold higher compared with the other amino acid thiols. Based on Ellman's mechanism (17), this indicates that a reactive persulfide anion originates from the splitting of the first S-S bond, which can subsequently target a second S-S bond, creating a trisulfide bond. We developed a method to quantitatively determine trisulfide bonds based on the difference between total bound sulfides and acid labile sulfides (18). Indeed, trisulfide bonds were generated when sulfide was used to reduce S-S bonds in DTNB (Fig. 4B), but not on amino acid thiol (NAC, GSH, or cysteine) treatment of DTNB. The theoretical (Fig. 4C) and measured ratio (Fig. 4B) of sulfide-dependent formation of thiol to trisulfide is 2, indicating that all sulfide reacts with DNTB to form trisulfide bonds.

To test whether similar redox reactions also occurred in vivo, total sulfides and trisulfides were determined in mice fecal water (Fig. 4D). Heme increased the levels of total bound sulfides, which is in agreement with literature showing that heme addition to the growth medium stimulates bacterial reduction of sulfate to sulfide (19). Concentrations of trisulfide in fecal water of mice not receiving Abx were much higher (with H > C) than those of mice receiving Abx, indicating that Abx suppressed sulfide-dependent splitting of S-S bonds and thus colonic mucolysis. In line with this, Abx drastically increased fecal excretion of mucin (Fig. 4E), because the slightly lower steady-state mucin synthesis (Fig. 3C) is not anymore balanced by its bacterial degradation. Fecal mucin was present as a high-molecular-weight form of Muc2 that could not penetrate the gel in SDS/PAGE analysis (Fig. 4F), whereas with DTT as a reducing agent most of the Muc2 appeared as a band of low MW in the gel. To corroborate the reducing effect of sulfide, Muc2 Western blotting was repeated for the CA- and HA-groups and showed that also sulfide lysed the polymeric Muc2 almost completely (Fig. 4F). Overall, these results indicate that in normal colon physiology, microbial sulfide opens the polymeric Muc2 network to bacterial degradation resulting in a lumen-to-surface permeability gradient through the mucus layer. Abx block these microbial processes and thereby increase the mucus barrier.

Discussion

This study shows that dietary heme changes the microbiota and increases ROS and cytotoxicity in the colonic lumen. In addition, heme injures the surface epithelium leading to compensatory hyperproliferation, hyperplasia, and differential expression of tumor suppressor and oncogenes, which increases colorectal cancer risk (20, 21). These luminal and epithelial effects of heme are similar to those detailed in our recent studies (6, 10, 11). Also the Abx effects on the microbiota replicate those shown by us in an earlier study (22). The crucial finding of the present study is that when the microbial abundance is drastically reduced by Abx, heme does not injure the surface epithelium and does not induce the carcinogenic changes in the crypts, mentioned above. The absence of carcinogenic changes is not due to the slightly lower levels of cytotoxicity and ROS in the HA-group, which can be explained by the higher luminal dilution factor because of the increased fecal wet weight. Recently, we reported that heme diets induce, in the colon lumen, covalent heme modification resulting in the very lipophilic and toxic CHF, which is solubilized in mixed micelles (11). Abx block the mucosal sensing of these cytotoxic micelles, as they prevent the heme-induced changes in epithelial histology and in up-regulation of injury markers such as Slpi. In contrast, Abx do not block mucosal sensing of luminal ROS, as the Nrf2-mediated antioxidant response was initiated and PPARs were activated by oxidized lipids in both heme diet groups. This differential mucosal sensing shows that, with Abx, the mucus layer is still permeable to small molecules, such as oxidized lipids, but no longer to larger micellar aggregates containing CHF. The absence of hyperproliferation in the HA-group also shows that mucosal exposure to ROS does not cause hyperproliferation. This result is in line with our previous observation that ROS is instantly formed after consumption of the heme diet, whereas there is a delay in the appearance of luminal cytotoxicity and the induction of hyperproliferation (11).

Our study implies that the colon microbiota facilitates heme-induced epithelial injury and hyperproliferation by opening the mucus barrier by the concerted action of hydrogen sulfide-producing and mucin-degrading bacteria. The principal steps of this hypothesis (Fig. 5) are (i) mucolysis by hydrogen sulfide to open the compact, protective mucus layer for (ii) further bacterial degradation, thereby (iii) allowing diffusion of luminal, cytotoxic micelles to the mucosal surface. Consequently, surface epithelial cells are less protected against luminal cytotoxicity, leading to induction of compensatory hyperproliferation. The diffusion barrier function of the mucus layer is illustrated by Muc2 KO mice, which display colitis and epithelial hyperproliferation, as well as spontaneous development of colorectal cancer (23). In addition, an in vitro study shows that apically applied mucin creates a diffusion barrier, preventing the contact between cytotoxic micelles and colonocytes (13). That microbiota increase the permeability of mucus barrier is illustrated by a study of recolonized vs. Abx-treated rats showing that bacteria colonizing the isolated colonic segment increase epithelial injury by luminally added toxic compounds (24). Unfortunately, there are no established methods to determine the permeability of the mucus layer in vivo. Although the mucus layer can be stained in appropriately fixed intestinal samples, this does not provide information about its permeability, as its thickness and permeability are not inversely related (15).

Fig. 5.

Proposed mechanism of how microbiota facilitates heme-induced compensatory hyperproliferation. (Upper) Processes when normal microbiota is present (i.e., without Abx) leading to compensatory hyperproliferation. (Lower) How Abx cause the mucus layer to be protective against cytotoxic micelles. R-S-S-R indicates native intra- and intermolecular disulfide bonds in the mucus that can be reduced by H₂S to thiols (R-S-H) and trisulfides (R-S-S-S-R).

Central to our hypothesis is that hydrogen sulfide can reduce and thus split S-S bonds, which opens the mucus layer. The compactness of this layer is determined by disulfide bond-stabilized polymer Muc2 network of C-terminal dimers and N-terminal trimers (15). Partial proteolysis by host proteases, which is visible in our denaturating gels of fecal mucin, does not change the structure of this network (15). However, splitting of S-S bonds dissolves it (i.e., mucolysis) resulting in reduced viscosity and increased permeability of the mucus layer (15). Typical S-S breaking agents are N-acetyl-cysteine (NAC), used in the treatment of cystic fibrosis, l-cysteine, and 2-mercaptoethanol, all known to decrease mucin viscosity in vivo and in vitro (16). Our DTNB results show that sulfide, compared with these thiols, is twofold more potent in breaking S-S bonds. The pK_a of hydrogen sulfide is about 1 unit lower than that of the thiols, implying that the concentration of the nucleophilic agent (i.e., the anion) in the splitting of S-S is higher with hydrogen sulfide. Moreover, as sulfide donates two electrons, it splits two S-S bonds, whereas thiols only split one. Elaborating on Ellman’s mechanism of S-S splitting (17), we reasoned that the highly nucleophlic persulfide, formed in the first reaction, generates a trisulfide bond in the second one. Our in vitro results show that trisulfide formation is indeed specific for S-S reduction by sulfide. Our fecal analysis shows that trisulfide is also formed in vivo and stimulated by dietary heme, probably because bacterial sulfate reduction is heme dependent (19). Abx strongly reduce overall bacterial abundance and suppress this trisulfide formation almost completely, supporting our mechanism that trisulfide is formed by bacterial sulfide. In line with this, Abx greatly increased fecal excretion of Muc2 in a high-molecular-weight polymeric form, as shown by nonreducing SDS/PAGE. Moreover, this polymeric Muc2 dissociates almost completely after reduction by DTT or sulfide, supporting the hypothesis that S-S bond splitting by sulfide opens the mucus barrier. This hypothesis is supported further by the recent finding that increasing the number of S-S bonds in the Muc2 network increases the mucus barrier in mouse colon (25).

Our mechanism of S-S bond splitting by sulfide is corroborated by a recent nutritional study by Devkota et al. (26), showing that monoassociation of germ-free mice with the sulfide-producing proteobacterium Bilophila wardsworthia, in the presence of taurocholate, results in breaking of the mucus barrier. The authors suggest that this is either due to sulfide (produced from taurine) or to unconjugated deoxycholate. However, the authors did not find barrier breaking in the presence of glycocholate, which is also metabolized to deoxycholate but does not generate sulfide. Therefore, we feel that their results can only be explained by the action of taurine-derived sulfide opening the mucus barrier via a mechanism analogous to what we propose here. Thus, it would be worthwhile to measure fecal trisulfides in that study.

Also in humans the colonic mucus layer functions as a barrier. As in mice, it prevents bacterial colonization of the epithelial surface and protects the surface cells from exposure to luminal toxic compounds (27). Three prevalent microbial profiles, so-called “enterotypes,” have been proposed to exist in human microbiota (28). Interestingly, for two of those enterotypes, mucin-degrading bacteria are identified as microbial drivers. One enterotype is rich in Prevotella and the co-occurring Desulfovibrio. Prevotella degrades mucin and Desulfovibrio may enhance the rate-limiting sulfatase step by hydrolyzing glycosyl-sulfate esters. The second mucin-degrading enterotype is rich in Ruminococcus and Akkermansia, both able to degrade mucins. We showed previously that dietary heme drastically increases the abundance of Prevotella and Akkermansia (10), which may be of relevance for these two enterotypes. The third enterotype is rich in Bacteroides using carbohydrates and proteins as substrates for fermentation (28). It would be of interest to see whether mucus barrier differences between different enterotypes exist or that diseases of the gut, such as colorectal cancer and IBD, are associated with mucin-degrading enterotypes.

Overall, we conclude that the microbiota facilitates the heme-induced hyperproliferation by opening the mucus barrier. Bacterial hydrogen sulfide can reduce the S-S bonds in polymeric mucin, thereby increasing the mucus layer permeability for mucin-degrading bacteria and for cytotoxic micelles. Consequently, epithelial surface cells are injured by the cytotoxic heme and compensatory hyperproliferation is initiated. This hyperproliferation might eventually lead to colorectal cancer (20). Our model and our results imply that fecal trisulfides can serve as a suitable marker of colonic mucolysis. Therefore, it would be of interest to measure levels of trisulfide in the human enterotypes, mentioned above, and in gut diseases in which the mucus barrier is compromised, such as irritable bowel syndrome (29).

Materials and Methods

Animal Handling and Design of the Study.

Experiments were approved by the Ethical Committee on Animal Testing of Wageningen University and were in accordance with national law. Eight-week-old male C57BL6/J mice (Harlan) were housed individually in a room with controlled temperature (20–24 °C) and relative humidity (55 ± 15%) and a 12-h light dark cycle. Mice were fed diets and demineralized water ad libitum. We designed our study as a 2 × 2 factorial experiment with heme and antibiotics as independent factors/treatments. Mice (n = 9/group) received either a “Westernized” control diet [40 energy% fat (mainly palm-oil), low calcium (30 µmol/g)] or this diet supplemented with 0.5 µmol/g heme for 14 d, as previously described (30). Broad-spectrum Abx, containing ampicillin (1 g/L), neomycin (1 g/L), and metronidazole (0.5 g/L), were administered in drinking water during the time of intervention. There were four experimental groups: control, heme, control plus Abx, and heme plus Abx. Feces were quantitatively collected during days 11–14, frozen at −20 °C, and subsequently freeze dried. After 14 d, the colon was excised, mesenteric fat was removed, and the colon was opened longitudinally, washed in PBS, and cut into three parts. The middle 1.5 cm of colon tissue was formalin-fixed and paraffin embedded for histology. The remaining proximal and distal parts were scraped, pooled per mouse, snap-frozen in liquid nitrogen, and stored at −80 °C until further analysis. Colonic contents were sampled for microbiota analysis. Chemicals were from Sigma-Aldrich, unless indicated otherwise.

Fecal Analyses.

Fecal water was prepared by reconstituting freeze-dried feces with double distilled water to obtain a physiological osmolarity of 300 mOsm/L, as described previously (5). Cytotoxicity of fecal water was quantified by potassium release from human erythrocytes after incubation (5) and validated with human colon carcinoma-derived Caco-2 cells (31). See SI Materials and Methods for TBARS, bile acids, and fecal mucin measurements.

Immunohistochemistry.

Histological H&E and immunohistochemical Ki67 (6) and Slpi (32) stainings were performed on paraffin-embedded colon sections as described previously. To quantify Ki67-positive colonocytes, 15 crypts per animal were counted. The number of Ki67-positive cells per crypt, total number of cells per crypt, and labeling index were determined.

RNA Isolation and Microarray Analysis.

RNA was isolated from colon scrapings and hybridized on Affymetrix GeneChip Mouse Gene 1.1 ST arrays (SI Materials and Methods). Genes satisfying the criterion of false discovery rate <1% (q < 0.01) were considered significantly expressed. Array data were submitted to the Gene Expression Omnibus with accession no. GSE40670.

Bacterial DNA Extraction and qPCR.

DNA was extracted from ∼0.1 g fresh fecal pellet from the colon using the method described by Salonen et al. (33). By qPCR, total bacteria were quantified using generic 16S rRNA primers and Bacteroidetes and Firmicutes using phylum-specific 16S rRNA primers (SI Materials and Methods).

Reduction of Disulfide (S-S) Bonds.

S-S splitting potency of sodium sulfide (Na₂S) and thiols was determined using DTNB as the model disulfide compound. Western blot analysis of fecal mucin was performed to determine whether DTT and sulfide reduces disulfide bonds in Muc2 (SI Materials and Methods).

Measurement of Trisulfide Bonds.

Trisulfides were calculated as the difference between total bound sulfides and acid-labile sulfides measured in individual fecal waters by GC-MS (SI Materials and Methods).

Statistics.

In vivo data are presented as mean ± SEM. Differences between groups were tested for main effects by two-way ANOVA. In vitro data are given as mean ± SD, and differences between groups were tested by one-way ANOVA with the Bonferroni multiple comparison test. P < 0.05 was considered significant.

SI Materials and Methods

TBARS.

To determine lipid peroxidation products in the gut lumen, TBARS in fecal water were quantified. The assay determines lipid peroxidation by quantifying the concentration of malondialdehyde (MDA) in fecal water. Briefly, fecal water was diluted fourfold with double-distilled water. To 100 µL of this dilution, 100 µL of 8.1% (wt/vol) SDS and 1 mL of 0.11 mol/L 2,6-di-tert-butyl-p-cresol, 0.5% TBA in 10% (vol/vol) acetic acid (pH 3.5) was added. To correct for background, TBA was omitted from the assay. TBARS were extracted, after heating for 75 min at 82 °C, with 1.2 mL l-butanol. The absorbance of the extracts was measured at 540 nm. The amount of TBARS was calculated as MDA equivalents using 1,1,3,3,-tetramethoxypropane as standard.

Bile acids were determined in fecal water of n = 3/group. Fecal waters were evaporated under nitrogen and resolubilized in 20% (vol/vol) acetonitrile in water. Samples were analyzed using reversed-phase HPLC combined with simultaneous amperometric detection of free and conjugated bile acids (34). Fecal mucins were quantified by means of their oligosaccharide side chains, as described in detail earlier (35), and expressed as µmoles O-glycans.

RNA Isolation.

Total RNA was isolated by using TRIzol reagent (Invitrogen) according to the manufacturer’s protocol. For microarray hybridization the isolated RNA was further column purified (SV total RNA isolation system Promega, Leiden, The Netherlands). RNA concentration was measured on a nanodrop ND-1000 UV-Vis spectrophotometer (Isogen) and analyzed on an Agilent 2100 bioanalyzer (Agilent Technologies) with 6000 Nano Chips, according to the manufacturer’s protocol. RNA was judged suitable for array hybridization only if samples exhibited intact bands corresponding to the 18S and 28S ribosomal RNA subunits and displayed no chromosomal peaks or RNA degradation products (RNA integrity number > 8.0).

Microarray Analysis.

The Ambion WT Expression kit (P/N 4411974; Life Technologies) in conjunction with the Affymetrix GeneChip WT Terminal Labeling kit (P/N 900671; Affymetrix) was used for the preparation of labeled cDNA from 100 ng of total RNA without rRNA reduction. Labeled samples were hybridized on Affymetrix GeneChip Mouse Gene 1.1 ST arrays, provided in plate format. Hybridization, washing, and scanning of the array plates was performed on an Affymetrix GeneTitan Instrument, according to the manufacturer’s recommendations. RNA labeling was performed in two rounds using a complete block design. Normalized expression estimates were obtained from the raw intensity values applying the robust multiarray analysis (RMA) preprocessing algorithm available in the Bioconductor library AffyPLM with default settings. Subsequently, an empirical Bayes method, called ComBat (36), was used to correct for the systematic error (batch effect) introduced during labeling. Probe sets were redefined according to Dai et al. (37). In this study, probes were reorganized based on the Entrez Gene database, build 37, version 1 (remapped CDF v14). Probe sets that satisfied the criterion of a false discovery rate <1% (q < 0.01) were considered significantly regulated and used for bioinformatics analysis by Ingenuity (Ingenuity Systems; www.ingenuity.com) and gene set enrichment analysis (GSEA; www.broad.mit.edu/gsea/). The number included in the array analysis were n = 4 per group for control, control plus Abx, and heme, and n = 6 for heme plus Abx. Genes with signal intensities below 20 in all mice from all treatments were considered not expressed and excluded from further analysis.

Bacterial DNA Extraction and qPCR.

Approximately 0.1 g fresh fecal pellet was subjected to mechanical and chemical lysis using 0.5 mL buffer [500 mM NaCl, 50 mM Tris⋅HCl (pH 8), 50 mM EDTA, and 4% (wt/vol) SDS] and 0.25 g of 0.1-mm zirconia beads and 3-mm glass beads. Nucleic acids were precipitated by addition of 130 μL, 10 M ammonium acetate, using one volume of isopropanol. Subsequently, DNA pellets were washed with 70% (vol/vol) ethanol. Further purification of DNA was performed using the QiaAmp DNA Mini Stool Kit (Qiagen). Finally, DNA was dissolved in 200 μL Tris/EDTA buffer, and its purity and quantity were checked spectrophotometrically (ND-1000; NanoDrop Technologies).

Total DNA was used for the quantitative estimation of total bacterial numbers per sample by real-time PCR amplification using generic 16S rRNA primers: forward Eub338F, ACTCCTACGGGAGGCAGCAG; and reverse Eub518R, ATTACCGCGGCTGCTGG. Bacteroidetes and Firmicutes populations were quantified by real-time amplification using phylum-specific 16S rRNA primers: Bacteroidetes forward Bact934F, GGARCATGTGGTTTAATTCGATGAT; reverse Bact1060R, AGCTGACGACAACCATGCAG and Firmicutes forward Firm934F, GGAGYATGTGGTTTAATTCGAAGCA; reverse Firm1060R, AGCTGACGACAACCATGCAC, respectively. Akkermansia muciniphila was quantified using genus-specific 16S rRNA primers S-St-Muc-1129-a-a-20 (AM1): CAGCACGTGAAGGTGGGGAC and S-St-Muc-1437-a-A-20 (AM2) CCTTGCGGTTGGCTTCAGAT. The sulfate-reducing capacity of the microbiota was estimated on basis of the gene-specific PCR that targets the dsrA gene that is associated with microbial sulfate respiration, using the following primers: forward RH1dsr, GCCGTTACTGTGACCAGCC and reverse RH3-dsr GGTGGAGCCGTGCATGTT. Bacterial counts are calculated and expressed in percentages relative to the control without Abx.

In Vitro Splitting of Disulfide (S-S) Bonds.

The potency of sodium sulfide (Na₂S), N-acetylcysteine (NAC), glutathione (GSH), and cysteine to split S-S bonds were determined using 5,5′-dithiobis-(2-nitrobenzoic acid) (DTNB) as model disulfide compound. Splitting of the S-S bond of DTNB can be quantified by measuring absorption at 412 nm (17). Analyses were performed at 20 °C in a Perkin-Elmer spectrophotometer using a 1.5-mL quartz cuvette containing 100 mM potassium phosphate buffer (pH 7.0) and 100 µM DTNB. The reaction was started by adding sulfide or the thiols (final concentration, 25 µM). Reaction rates were determined as the slope of progress curve immediately after mixing. The extent of the overall reaction was determined from the final, stable, absorption, and corrected for blank absorption, using the milimolar extinction coefficient of 14.15. Sodium sulfate (Na₂SO₄) was used as negative control.

Western Blot Analysis.

Homogenates of freeze-dried feces were made by adding 1 mL PBS containing protease inhibitors (CompleteMini; Roche Diagnostics) to 10 mg freeze-dried pooled feces (n = 9/group). Samples were incubated for 5 min at 100 °C with nonreducing sample buffer (without DTT), with reducing sample buffer (with DTT), or with nonreducing sample buffer with Na₂S. Samples were applied to SDS/PAGE [10% (wt/vol) gel] and transferred to polyvinylidene fluoride membranes under 350 mA for 4 h on ice. Twenty micrograms total protein was loaded. The blot was probed overnight at 4 °C with rabbit anti-mouse Muc2 antibody (38) (1:2,500) and then incubated with peroxidase-conjugated goat anti-rabbit antibody (1:5,000; Thermo Scientific). The signals were detected with the enhanced chemiluminescence detection system (Amersham ECL, GE Healthcare).

Measurement of Trisulfide Bonds.

From each individual fecal water 20 µL was added to 2-mL headspace vials containing 450 µL of Tris buffer (90 mM; pH 8.0) with or without 0.9 mM cysteine. Capped vials were incubated 30 min at 35 °C to facilitate cysteine-driven reduction of the central sulfur of the trisulfide bond to inorganic sulfide, as described earlier (18). Subsequently, 20 µL of 6 M HCL was added to the vials to force the sulfide as H₂S into the headspace. With a headspace sampler this H₂S was then measured on a GC-MS (TraceGC ulta; ThermoFinnigan) using a VF-1ms 30 × 0.25 column (Varian) with helium as carrier gas, and quantified with standard solutions of Na₂S × 9 H₂O treated identical to the samples. Trisulfide was calculated as the difference in H₂S between the vial containing cysteine (i.e., total bound sulfide) and the vial without it (i.e., acid-labile sulfide). To validate this method control samples containing dimethyltrisulfide were analyzed identically and showed a recovery >90%. It should be noted that our assay is not confounded by free inorganic sulfides (such as HS⁻ and S²⁻) in feces as these are removed during freeze drying. Samples from the in vitro incubations were analyzed identically. H₂S was only detected in the cysteine vials containing the samples from the Na₂S incubations, illustrating the specificity the trisulfide measurement.