Identification of EREG by whole-genome expression profiling in mouse models of CAC and sporadic CRC.
To uncover molecules specifically involved in the pathogenesis of CAC, we studied tumorigenesis in experimental models of CAC and sporadic CRC that mimic key characteristics of human CRC (12
). Whereas tumors in the Apcmin/+
model (where Apc
indicates adenomatous polyposis coli) of sporadic CRC typically exhibit exophytic, pedunculated polyps, flat and multifocal tumors are characteristically observed in the azoxymethane/dextran sodium sulfate (AOM/DSS) model of CAC by endoscopy (Figure ). To identify molecular mechanisms giving rise to such different growth patterns, we profiled gene expression in both models using high-density microarray analysis (HDMA). Tumor growth was monitored by repetitive mini-colonoscopy, and tissue was harvested when tumors reached a size covering between one-fourth and up to one-half of the colonic circumference, i.e., representing an endoscopic size score of 4 (15
Growth features of sporadic and colitis-associated tumors.
The cellular and molecular microenvironment of the intestinal wall has marked regional differences along the gastrointestinal (GI) tract; all tumors, including those from the Apcmin/+ model, were obtained from the distal part of the colon. To minimize stage-related differences in transcriptional profiles, all tumors were of the same size. Finally, to reduce the influence of different cellular compositions within the colonic wall, we used purified IEC from the distal colon as a control.
Whole-genome expression screening was performed with Affymetrix-GeneChips after purification of total RNA from tumors and respective control IEC. Hierarchical clustering as visualized by heat map analysis revealed groups of genes with similar expression patterns (Figure A). Expression profiles were then subjected to groupwise comparison, and advanced significance analysis between the tumors from both models revealed 924 probe sets (603 down, 321 up) with significance criteria of P < 0.05 after the Benjamini-Hochberg correction and at least a 2-fold change. The distribution of genes was visualized with a volcano plot (Figure B). The 25 genes with the greatest up or down differential expression showed a fold change (FC) ranging from 6 to 40 (Figure C). Among the genes with higher expression in the CAC model were several immune response–related genes such as IgI-V1, IgkC, or Arg1, which might be induced by immune cell infiltration. In addition, genes that have been related to tumor progression such as Malat1 and Klf5 were elevated in CAC. Among probe sets showing lower expression in the CAC model were several genes related to cell structure organization such as Pfn2, Actn1, and Mark1. Next, we performed Gene Ontology (GO) Enrichment Analysis using the list of differentially expressed upregulated genes, which identified GO terms such as cell proliferation, response to wounding, inflammatory response, and EGF binding in the CAC model compared with the sporadic CRC model (Figure D). In contrast, GO categories such as WNT receptor signaling pathway or TGF-β receptor signaling were not significantly enriched suggesting that no major differences exist in these pathways between the 2 models.
Differential gene expression studies in mouse models of sporadic and colitis-associated tumors identify EREG as a potential target.
To further investigate differences, we performed significance analysis between the microarrays from the CAC model and control epithelia as well as the sporadic CRC model and control epithelia. This analysis revealed a high number of probe sets with differential expression (P
< 0.05; FC > 2) (Figure E and Supplemental Table 1; supplemental material available online with this article; doi:
). By comparing these gene lists, we found overlapping gene expression patterns, as shown by Venn diagrams, and identified 387 probe sets that were exclusively upregulated in CAC but not in sporadic CRC (Figure F). We further refined our analysis by comparing the lists of probe sets “differentially upregulated in CAC model only” with the gene list “upregulated in CAC model versus sporadic CRC model” (Figure , B, E, and F). We obtained another list of genes, in which EREG, a member of the EGF family, ranked at position 3 of the most upregulated genes specific to CAC (Table ). Of note, EREG was also identified on the list of top induced probe sets when a direct comparison between both tumor entities was performed (Figure C). Thus, EREG was differentially expressed between CAC and controls, CAC and sporadic CRC, but not sporadic CRC and controls (Figure , B and E). Surprisingly, other EGF-like ligands were not differentially expressed in the comparative analysis between the 2 models (Supplemental Figure 1), suggesting selective differential regulation of the EREG gene. Subsequent validation experiments using a larger panel of mouse tumors and control tissue by quantitative PCR (qPCR) confirmed the observations from HDMA.
Gene list from differential gene expression analysis in experimental colon cancer models
Augmented EREG expression in UC patients with colitis-associated neoplasms.
To investigate the potential relevance of EREG for human CRC, we next studied tissue from different stages of tumor development including sporadic and colitis-associated high grade dysplasias (HGDS) and sporadic CRC as well as CAC (Figure , A and B). Strikingly, the expression of EREG, as assessed by immunohistochemistry, was highly significantly elevated in colitis-associated dysplasias and carcinoma as compared with colonic tissue from control patients (P < 0.01). In contrast, the expression of EREG in specimens from sporadic tumor development was lower, although still elevated at the average level compared with normal colonic tissue. Of note, EREG expression appeared to be more heterogenous in sporadic as compared with colitis-associated neoplasms. Consistently, quantitative analysis of EREG expressing cells revealed a significant increase in colitis-associated neoplasms (UC-associated HGDs and cancer) as compared with sporadic neoplasms (sporadic HGDs and cancer) (Figure C).
Human precursor lesions of colitis-associated tumors display significantly elevated EREG expression.
Additional differences between sporadic and UC-associated neoplasms were noted with regard to the cellular source of EREG production. In particular, we detected that EREG expression in colitis-associated tumors was dominated by tumor stromal cells, whereas EREG expression in sporadic tumors was often found in patches of epithelial cells that showed basolateral EREG enrichment. Taken together, these findings highlighted the potential functional relevance of EREG in UC-associated tumor lesions in humans.
EREG deficiency impairs colitis-associated tumor growth.
Since data using gene expression profiling and immunohistochemistry were consistent with a potential regulatory role of EREG in colitis-associated tumors, we next addressed the functional role of this growth factor in experimental tumorigenesis. In these studies, we took advantage of Ereg–/–
mice and performed experiments using the AOM/DSS model of colitis-associated tumor growth. Strikingly and in marked contrast to previous studies on sporadic intestinal tumorigenesis in the Apcmin/+
small intestinal model (16
mice were strongly protected from tumor development in the mouse AOM/DSS colonic model. This observation was verified by in vivo imaging of integrin αV
, macroscopic inspection, high-resolution endoscopy, and histopathological analysis of H&E-stained cross sections (Figure , A–E). In fact, endoscopic scoring demonstrated a significant difference in the tumor load between Ereg–/–
mice and WT controls (P
< 0.01). Of note, although we detected a difference in tumor numbers (Figure D), the striking difference in tumor load (Figure E) was particularly caused by significant differences in tumor sizes between WT and Ereg–/–
mice (Figure ). To analyze the kinetics of tumor development in both groups, we performed serial endoscopies in longitudinal studies directly addressing tumor growth over time. Whereas early neoplastic lesions were detected around days 15 to 20 in both groups, tumor load analysis showed an increasing difference between WT and Ereg–/–
mice at later time points (Figure , A and B). In fact, endoscopic analyses suggested that tumor growth is inhibited in Ereg–/–
mice at later time points, resulting in slower growth rates and a persistent majority of small tumors compared with the composition of tumor load in WT controls. In WT mice, almost one-fourth of all tumors were large (sizes 4 or 5) at day 80. In contrast, large tumors were not observed in Ereg–/–
animals by day 80 and very small tumors (size 1) accounted for more than 50% of all tumors in Ereg–/–
mice even at day 80 (Figure C). Thus, the strong reduction in tumor load over time in Ereg–/–
mice is due to a reduction of large tumors, suggesting a key role of EREG in tumor growth promotion rather than initiation.
Colitis-associated tumorigenesis in Ereg-deficient mice.
EREG promotes growth rather than initiation of colitis-associated tumors.
EREG regulates epithelial wound healing and promotes tumor growth.
As previous studies demonstrated a direct link between inflammatory activity and tumor growth in experimental colitis-associated tumorigenesis (17
), we next addressed EREG expression during DSS-induced colonic inflammation. Consistent with previous findings showing a strong temporary increase in intestinal EREG levels in acute colitis (19
), we noted repeated more than 10-fold upregulation of EREG expression during DSS cycles, mimicking flares of increased activity in IBD (Figure A). Interestingly, we observed a slight increase of average EREG expression over time, suggesting a marked and repeated EREG induction by flares in chronic intestinal inflammation. Although Ereg–/–
mice were more susceptible to acute DSS colitis than WT controls (16
), no significant differences in colonic inflammation between these mouse strains were found during chronic DSS colitis (Supplemental Figure 2). Thus, although EREG expression is induced by inflammation, this protein does not seem to importantly regulate activity of chronic intestinal inflammation.
EREG facilitates intestinal wound healing in vitro and in vivo.
To explore possible alternative functions of EREG, we performed studies on the role of this protein in epithelial wound healing. In initial studies, we used an in vitro assay for cell migration in which a mouse colonic epithelial cell line (CMT93) was grown to confluence before gaps of defined size were scratched into the cell layer. Subsequently, cells were cultured in the presence or absence of EREG followed by assessment of gap sizes. Strikingly, wound gap closure in the presence of EREG was significantly accelerated (about 50% of remaining gap size after 18 hours compared with untreated cells), suggesting a prominent role of EREG during reconstitution after epithelial injury (Figure , B and C). To study whether EREG also contributes to intestinal wound gap closure in vivo, we generated epithelial defects in the distal colon of Ereg–/–
mice and control animals by use of the biopsy forceps in mini-endoscopy. Wound healing was monitored by serial endoscopy (20
). Consistent with our data from in vitro experiments, intestinal wound healing was significantly delayed in Ereg–/–
mice compared with controls, indicating a major regulatory function of EREG in epithelial wound healing in vivo (Figure , D and E). Consistently, studies with heterozygous lacZ-Ereg
-reporter mice demonstrated the accumulation of EREG-producing cells around the wound bed in response to in vivo wounding in the distal colon (Figure F).
To directly demonstrate effects of EREG on epithelial cell proliferation, full-thickness rectosigmoidal pieces were incubated with mock or recombinant EREG and stained for Ki67 (Figure , A and B). EREG treatment led to a significant increase of epithelial cell proliferation (P < 0.05). To test whether EREG could directly increase the number of IECs and promote growth of colon tumors, we applied recombinant EREG to CMT93 cells and early colon tumors in vivo. Interestingly, we observed a significant increase in epithelial cell numbers after incubation in the presence of EREG, providing further evidence for an increase in proliferation mediated by EREG (P < 0.05). Of note, the number of cells showed a dose-dependent increase in the presence of EREG at 48 hours compared with control cells (Figure C). We next injected EREG or other EGF-like ligands into WT mice during AOM/DSS treatment and performed endoscopic tumor screening at day 30 (Figure D). We chose animals with similar tumor load and performed local injections of EREG, EGF, amphiregulin (AREG), or equal volumes of PBS. The injections were applied during endoscopy using a flexible syringe device near the base of similar sized small tumors every 3–4 days. Tumor growth was monitored by endoscopy, and mice with EREG injections displayed a significantly accelerated increase in tumor load compared with mice that had obtained PBS injections (P < 0.05). In contrast, other EGFR ligands such as EGF and AREG were less potent and did not result in a significant increase of tumor load as compared with the PBS control group (Figure , E and F).
EREG directly increases proliferation in IECs and promotes colon tumor growth.
In summary, these observations suggest that colitis-associated EREG expression promote not only intestinal wound healing, but also tumor growth.
Identification of tumor-associated fibroblasts as the key cellular source of EREG in colitis-associated neoplasms.
The cellular source of EREG has been assigned to epithelial and submucosal cells during acute inflammation (16
). To identify the cellular source of EREG in the AOM/DSS model of CAC, we took advantage of the LacZ reporter gene knockin of Ereg
-deficient mice and performed X-gal stainings in heterozygous animals. It was found that the majority of cells that stained positive for EREG in AOM/DSS tumors were stromal cells (Figure A). In addition, we detected some EREG-expressing epithelial cells. As fibroblasts display a frequent cell type in the tumor microenvironment that may produce regulatory molecules (21
), we next addressed the possibility that cancer-associated fibroblasts are the major cellular source of EREG production in AOM/DSS tumors.
Fibroblasts are the main producers of EREG in colitis-associated tumors.
To test this hypothesis, we performed consecutive stainings of cross sections from AOM/DSS-tumors of heterozygous lacZ-Ereg-reporter mice by using both X-gal staining and confocal immunohistochemistry. Here. we could show that over two-thirds of all EREG-expressing cells and virtually all EREG-expressing stromal cells colocalized with fibroblast markers including vimentin (VIM) and fibroblast-specific protein-1 (FSP1, also known as S100A4) (Figure , A and B). A similar expression pattern could be demonstrated for PDGFR-β (Figure A). In addition, approximately 60% of EREG-expressing stromal cells could be colocalized with α-SMA (Figure A), indicating an activated myofibroblastic phenotype in a substantial number of EREG-expressing cells in colitis-associated tumors. To verify whether tumor-associated fibroblasts represent also a major cellular source of EREG in CAC in humans, in an additional series of experiments, we stained samples from patients with UC-associated colorectal neoplasms with antibodies against fibroblast markers and EREG. In congruence with our observations from the mouse model of CAC, EREG-expressing stromal cells in human CAC demonstrated coexpression with the fibroblast marker VIM (Figure C), suggesting that EREG-producing tumor-associated fibroblasts are present in both murine and human colitis-associated neoplasms.
To further characterize EREG-producing cells in colitis-associated neoplasms, we next isolated fibroblasts from AOM/DSS tumors of heterozygous lacZ-Ereg-reporter mice. The purity of these fibroblast cultures was assessed by immunocytochemical analysis, which revealed a high number of cells expressing proteins related to tumor-associated fibroblasts including VIM (>90%), FSP1 (>90%), PDGFR-β (90%), versican (VCAN, also known as chondroitin sulfate proteoglycan 2 or NG2) (85%), and α-SMA (80%). Strikingly, EREG expression was seen in more than 75% of the cells in these fibroblast cultures, as demonstrated by X-gal stainings (Figure , D and E).
Whereas lacZ staining was useful to localize and count the number of EREG-expressing cells in murine cell culture, we took advantage of real-time qPCR to quantify the expression of EREG in tumor fibroblasts. Remarkably, these cells produced high amounts of EREG and a significant, more than 10-fold difference in EREG production was noted between tumor-associated fibroblasts and IEC (Figure F). Of note, the addition of recombinant EREG to tumor fibroblasts significantly increased EREG production, suggesting the presence of additional self-perpetuating autocrine and paracrine loops in these cells (Figure F).
In subsequent experiments, we aimed at the identification of signaling pathways inducing EREG expression in tumor-associated fibroblasts. It was found that tumor-associated fibroblasts produce high baseline levels of EREG, which could be further enhanced by in vitro stimulation with LPS or TNF-α (Figure G). As TNF-α and bacterial proteins such as LPS are thought to play a major role in IBDs (1
), these findings suggested that microbial products and proinflammatory molecules released during intestinal inflammation are important inducers of local EREG production by tumor-associated fibroblasts in colitis-associated neoplasms.
Adoptive transfer of EREG-producing tumor-associated fibroblasts promotes tumor growth in colitis-associated neoplasms in vivo.
In subsequent studies, we analyzed whether tumor fibroblast–derived EREG may affect growth of IEC. Accordingly, mouse IEC (cell line CMT93) were grown in the presence of conditioned medium from AOM/DSS tumor fibroblasts purified from WT and Ereg–/– mice, respectively. Strikingly, the conditioned medium obtained from the WT tumor fibroblasts resulted in a significant increase in the epithelial cell number after 48 hours, whereas the conditioned medium from Ereg–/– cells did not induce significant changes in cell numbers (Figure A). These findings indicated that fibroblast-derived EREG favors proliferation of IEC.
Adoptive transfer model: fibroblast-derived EREG promotes the growth of colitis-associated tumors.
In subsequent studies, we addressed the possibility that EREG derived from tumor fibroblasts augments growth of colitis-associated neoplasms in vivo. To further evaluate the in vivo relevance of fibroblast-derived EREG production for the growth of CAC in the AOM/DSS model, we intended to set up studies with adoptive transfer of tumor-derived fibroblasts. Accordingly, we developed an adoptive transfer system of fibroblasts via endoscopically guided local injections into small tumors (endoscopic tumor score 1). In a first set of experiments, we validated this injection technique and injected portions of 50 μl of a fibroblast suspension from heterozygous lacZ-Ereg-reporter mice (105 cells) into tumors of WT mice. At day 2 after injection, tumors were harvested and stained for X-gal to assess EREG expression. These studies demonstrated not only the presence of transferred fibroblasts surrounding the tumor tissue, but also showed EREG expression in transferred cells within the tumor stroma (Figure , B and C).
We next treated Ereg–/– mice with AOM and 1 cycle of DSS and performed endoscopic tumor screening at day 30 to ensure equal average tumor scores in all groups. We then injected Ereg–/– or WT fibroblasts from AOM/DSS-treated mice into small tumors of these Ereg–/– mice during endoscopy every 5 days. An additional group of WT mice did not receive tumor-associated fibroblasts and served as a control group. In consecutive studies, changes in tumor load were monitored endoscopically in all 3 groups of animals (Figure , D–F). AOM/DSS-treated Ereg–/– mice repetitively injected with Ereg–/– fibroblasts showed significantly slower tumor growth as compared with AOM/DSS treated Ereg+/+ (WT) mice without fibroblast injections (P < 0.05). However, Ereg-deficient mice given WT fibroblasts showed significantly augmented tumor growth as compared with Ereg-deficient mice given Ereg-deficient fibroblasts (Figure , E and F), suggesting that EREG from tumor-associated fibroblasts induces tumor growth in vivo. Furthermore, Ereg–/– animals repetitively treated with WT tumor fibroblasts showed no significant differences compared with AOM/DSS-treated WT control mice, suggesting that adoptive transfer of EREG-producing but not EREG-deficient tumor-associated fibroblasts is sufficient to abrogate the differences in tumor growth between WT and Ereg-deficient mice. In summary, these observations indicate that colitis-associated EREG expression by tumor fibroblasts can promote tumor growth in the AOM/DSS model in vivo.
A key role of ERK in EREG-mediated regulation of cell proliferation.
EREG belongs to the EGF-like family of growth factors that can signal through of ERBB receptors, which can activate several pathways potentially involved in tumor growth such as ERK (also known as ERK1/2, MAPK3/1, or p44/p42-MAPK), p38-MAPK, and STAT3 pathways (23
). Thus, we tested a number of downstream targets potentially involved in EREG signaling in our experimental model of CAC. Our data showed that EREG causes a pronounced activation of ERK in IEC in vitro (Figure A). In contrast, other major pathways related to tumor development including p38-MAPK and STAT3 showed little or no changes upon stimulation with EREG (data not shown).
EREG promotes the development of colitis-associated tumors via activation of ERK.
To functionally address the role of ERK as a potential downstream effector molecule for the EREG-induced promotion of tumor growth, we tested whether the ERK inhibitors U0126 and PD98059 could counteract EREG-mediated effects on IEC and tumors. Indeed, while EREG significantly enhanced the proliferation of colonic epithelial cells (P < 0.01), this effect was completely abrogated by ERK inhibitors (Figure B), suggesting that ERK signaling controls EREG-mediated effects on epithelial cell proliferation. Furthermore, we detected that high levels of ERK phosphorylation in the AOM/DSS model could be blocked by incubation with these specific inhibitors despite the presence of EREG. These findings are consistent with ERK acting as the downstream effector of EREG-mediated tumor growth promotion in vitro and in vivo (Figure C). To address the in vivo relevance of this observation, we performed repetitive local injections into the base of small tumors during endoscopy and monitored tumor growth serially (Figure D). Interestingly, EREG injections resulted in accelerated tumor growth, while specific ERK inhibition with U0126 significantly suppressed tumor development (Figure , E and F). Of note, the same ERK inhibitor was markedly less potent in the Apcmin/+ model in sporadic colorectal tumorigenesis, as this inhibitor caused no significant reduction of tumor load in the Apcmin/+ model (Figure , E and F).
Thus, our data shows that interference with ERK signaling can block EREG-mediated growth promotion of experimental CAC.