Physiological Genomics

Deconvoluting the intestine: molecular evidence for a major role of the mesenchyme in the modulation of signaling cross talk

Xing Li, Blair B. Madison, William Zacharias, Åsa Kolterud, David States, Deborah L. Gumucio


Reciprocal cross talk between the endodermally derived epithelium and the underlying mesenchyme is required for regional patterning and proper differentiation of the developing mammalian intestine. Though both epithelium and mesenchyme participate in patterning, the mesenchyme is thought to play a prominent role in the determination of the epithelial phenotype during development and in adult life. However, the molecular basis for this instructional dominance is unclear. In fact, surprisingly little is known about the cellular origins of many of the critical signaling molecules and the gene transcriptional events that they impact. Here, we profile genes that are expressed in the separate mesenchymal and epithelial compartments of the perinatal mouse intestine. The data indicate that the vast majority of soluble inhibitors and modulators of signaling pathways such as Hedgehog, Bmp, Wnt, Fgf, and Igf are expressed predominantly or exclusively by the mesenchyme, accounting for its ability to dominate instructional cross talk. We also catalog the most highly enriched transcription factors in both compartments. The results bolster previous evidence suggesting a major role for Hnf4γ and Hnf4α in the regulation of epithelial genes. Finally, we find that while epithelially enriched genes tend to be highly tissue restricted in their expression, mesenchymally enriched genes tend to be broadly expressed in multiple tissues. Thus, the unique tissue-specific signature that characterizes the intestinal epithelium is instructed and supported by a mesenchyme that itself expresses genes that are largely nontissue specific.

  • microarray
  • transcription factors
  • intestinal development
  • bmp pathway

the mammalian small intestine develops from a tube of endoderm wrapped by mesoderm. Throughout embryonic, fetal, and adult life, the endoderm and mesoderm are mutually dependent upon one another for instructive signals that sculpt the eventual morphological form of each layer and control region-specific gene expression (40). Elegant tissue recombination studies have underlined the extent to which this tissue's cross talk controls patterning. For example, in trans-species grafts of mouse and chick, the mesoderm holds the instructional information that controls whether the epithelium adopts villus projections as in the mouse or ridge-like structures characteristic of the chick (42). Likewise, dissociated ileal mesenchyme of the rat is able to instruct isolated rat colonic epithelium to express small intestinal enzymes (10), again reflecting the instructional power of the mesodermal layer. At specific time points during development, instructions also pass from endoderm to mesoderm, since ileal endoderm can induce colonic mesoderm to express homeobox genes characteristic of the ileal region (11). The accumulated data suggest that a mutually reinforcing program directs proper intestinal development and predict that a predominant role is played by the mesenchyme in instructing and maintaining intestinal form and homeostasis (28). This prediction is further supported by the finding that different myofibroblast primary cell lines isolated from the gut induce different epithelial morphologies when cocultured with an epithelial cell line such as CaCo2 (11).

Clearly, many complex signaling events regulate the development and regional patterning of the gut tube; signaling cross talk also controls homeostasis as well as reaction to injury or inflammation. Microarray studies can shed light in a global way on subsets of genes that are changed in disease states or after alterations of specific signaling pathways in animal models. However, in many cases, microarray studies are performed on whole intestinal tissue, and thus components of both mesenchyme and epithelium are sampled together. It becomes difficult to decipher direct from indirect effects of any given signal in the absence of spatial information regarding the location of signaling molecules and transduction machinery, information that is not trivial to acquire, especially for multifactorial signals.

In this study, we have taken the first step in unraveling some of this complexity by examining gene expression profiles in separated epithelium and mesenchyme in the perinatal mouse intestine. At embryonic day (E) 18.5 (Fig. 1A), intestinal villi are well-developed and most of the adult cell lineages are present (Paneth cells develop after birth). Villi are polarized, with the proliferative compartment restricted to the base of the villi (Fig. 1B), but crypts have not yet formed. Clean separation of the epithelium from the mesenchyme is possible at this stage (Fig. 1, C and D).

Fig. 1.

Appearance of the embryonic day (E) 18.5 intestine and characterization of separated epithelium and mesenchyme. A: this section of proximal small intestine, stained with hematoxylin and eosin, reveals well formed villi containing absorptive cells and goblet cells (arrows). B: staining with anti-Ki67 (dark red nuclei at bottom of villi) reveals polarization of the villi, with proliferative cells located at the villus base. Crypts have not yet formed. C: epithelial fraction after separation of mesenchyme. D: mesenchymal fraction after separation of epithelium. E: PCR verification of tissue separation. Primers and conditions for PCR are listed in Supplementary Table S1. Smooth muscle gamma actin (Actg2) and MAdCAM are expressed in mesenchyme (Mes), while Villin1 and Ihh are expressed in epithelium (Epi). Hprt is expressed in all fractions. SI, unseparated small intestine.

This microarray analysis of isolated epithelial and mesenchymal fractions yields a molecular confirmation for a major role for the mesenchyme in the modulation of a wide variety of signaling pathways (e.g., Hedgehog, Wnt, Bmp, Fgf, Igf) since, with very few exceptions, the soluble signal inhibitors and modulators of these pathways are expressed primarily in mesenchyme. We also identify all transcription factors that are expressed predominantly in one tissue compartment or the other and document substantial differences in the distribution of transcription factor subtypes (e.g., zinc finger, homeobox, etc.) in the two compartments. Finally an examination of the expression patterns of the most enriched epithelial and mesenchymal genes using electronically available expressed sequence tag (EST) tallies reveals that while many of the highly enriched epithelial genes are intestine specific, nearly all genes highly enriched in the intestinal mesenchyme are expressed in numerous additional tissues. This suggests that a complex and multifactorial instructional signal from the mesenchyme is responsible for the generation and maintenance of a tissue-specific epithelial signature.


Separation of epithelium and mesenchyme.

All animals used in this study were C57BL/6; all experimental protocols performed on animals were done with previous approval from the University of Michigan Committee on the Use and Care of Animals (protocol #7788). Tissue separation was done as previously described (30). In brief, E18.5 embryos from C57BL/6 females were removed from the yolk sac and amnion and placed into ice-cold PBS containing penicillin-streptomycin. The entire small intestine, from duodenum to cecum, was isolated, cut open longitudinally, placed into 1.0 ml of cold Cell Recovery Solution (BD Biosciences) in a 12-well plate and incubated for 9 h at 4°C. On ice, each plate was agitated by hand for 30–45 min until all epithelium sloughed off, as determined by microscopic examination. The mesenchyme was gently removed from the more delicate epithelial fragments with sterile forceps and rinsed in a dish of sterile PBS to remove any remaining epithelial cells. Mesenchyme from three embryos was pooled into a 15-ml conical tube then snap frozen in liquid nitrogen; six such pooled samples were prepared. The remaining epithelial tissue was washed twice with 10 ml of cold PBS and collected by gentle centrifugation (100 g for 7 min at 4°C). Intestinal epithelium from three embryos was pooled into a 15-ml conical tube then snap frozen in liquid nitrogen; six pooled samples were prepared. Each pool of tissue was subsequently homogenized in 1.0 ml of TRIzol (Invitrogen), on ice, and RNA prepared per the TRIzol protocol. Total RNA was further purified using the RNeasy Mini Kit (Qiagen), and quality was assessed by electrophoresis of 2.0 μg of RNA on a 1% agarose gel. To assess epithelial vs. mesenchymal purity of RNA, we performed RT-PCR for both epithelium-specific mRNAs (Vil1 and Ihh) and mesenchyme-specific mRNAs (Madcam1 and Actg2). Three epithelial and three mesenchymal RNA fractions exhibiting minimal contamination with mesenchymal and epithelial mRNAs, respectively, were selected from among all of the samples for Affymetrix microarray analysis. According to the manufacturer's instructions, labeled cRNA probes were synthesized from total purified RNA and hybridized to the Affymetrix GeneChip Mouse Genome 430 2.0. The University of Michigan Diabetes Center Microarray Core Facility performed cRNA synthesis, array hybridization, and array scanning, all according to standard Affymetrix protocols.


Microarray was performed using Affymetrix MOE 430.2 arrays, containing ∼43,000 probes to study >39,000 characterized genes or ESTs. Six microarray chips were used (three epithelium and three mesenchyme) for the analysis. After the hybridizations, microarrays were washed and scanned to generate the image files (.DAT files), which were then processed using MAS 5.0 software to produce .CEL files. The .CEL files were analyzed by RMA (robust multiarray average), which subtracted background, normalized expression data and calculated gene expression values (25). The logarithm base 2 (Log2) of the average expression value for each probe pair on the three epithelial chips was subtracted from the Log2 of the average expression value on the three mesenchymal chips. The difference was converted to a numerical fold difference [FD = 2Log2(mes)−Log2(epi)] or enrichment score (ES, absolute value of the FD). Two-tailed t-tests were carried out to identify differentially expressed genes according to P ≤ 0.05 and fold difference ≥2.0. Where multiple probe sets were available for a given gene, only the probe set with the highest enrichment score is listed in the associated data tables. All data are available at the National Center for Biotechnology Information (NCBI) GEO Database ( under the series number GSE6383.

Search for Hnf4 binding sites.

Promoter sequences (500 bp upstream of transcriptional start site) were downloaded from Ensembl ( for each of the 100 most enriched genes in the epithelium and each of the 100 most enriched genes in the mesenchyme. In addition, three groups of 100 genes with enrichment scores of 1.1–0.9 were compiled and promoter sequences (500 bp) were downloaded. These three groups differed in their average expression level: CntlH included 100 genes with average expression values >10.0 in epithelium and mesenchyme; CntlM included 100 genes with average expression levels between 8.0 and 10.0 in both compartments; CntlL consisted of 100 genes with average expression values <8.0 in epithelium and mesenchyme. The MatInspector tool of the Genomatix software suite ( was used to search for Hnf4 binding sites in these five groups of genes (3, 38). The default stringency settings in MatInspector were used in the search. MatInspector uses two consensus Hnf4 sequences based on functional studies in the literature (12): R31G87G59N16C100A100A100A100G100K50T37C67A61 (Hnf4.01) and S57R57G100G100T65C100M57A100A100A100G100G65T100C100 (Hnf4.02). The subscripted numbers indicate the conservation value assigned to each nucleotide. Bases with a conservation value >60 are considered to have a high information content.

RT-PCR verification.

RT-PCR was used for quality control on the original separated tissue samples and to verify results of the microarray. All RT-PCR was done using cDNA produced from isolated epithelial or mesenchymal RNA. Conditions for amplification (magnesium, DMSO addition) were individually optimized for each probe set. Unless otherwise indicated, 30 cycles of amplification were used. Primers and conditions used for all PCR studies are listed in Supplementary Table S1. (The online version of this article contains supplemental material.)

In situ hybridization.

In situ probes were synthesized from cDNA fragments obtained by rtPCR and cloned into PCR4 TOPO vectors (Invitrogen). The following templates were used: Bmp2 (NM_007553 nucleotides 220-1250), Bmp4 (NM_007554 nucleotides 1-1115), Bmp5 (NM_007555 nucleotides 75-950), Bmp7 (NM_007557 nucleotides 770-1780), Bmpr2 (AF003942 nucleotides 3410-2595), Bmpr1a (NM_009758 nucleotides 390-1450), and Acvr1 (NM_ 007349 nucleotides 135-1165). To ensure the specificity of the probes, both sense and antisense probes were generated from each template and tested in parallel.

Small intestines from E17.5 C57BL/6 embryos were removed and fixed overnight in 4% paraformaldehyde. The tissue was then dehydrated, processed and embedded in paraffin, sectioned at 5 μm, and baked for 2 h at 56°C. For in situ hybridization, sections were dewaxed, rehydrated, digested in a 20-μg/ml proteinase K solution for 10 min at 37°C, postfixed, and treated in acetic anhydride solution. After prehybridization in 4× SSC, 50% formamide at 37°C for 1 h, the sections were hybridized overnight at 60°C with various probes in 50% formamide, 10% dextran sulfate, 1 mg/ml yeast tRNA, 1× Denhardt, 0.2 M NaCl, 10 mM Tris·HCl pH 7.5, 10 mM phosphate buffer pH 6.8, and 5 mM EDTA pH 8.0. After hybridization, the slides were washed for 3 × 30 min in 1× SSC, 50% formamide, 0.1% Tween 20 at 65°C, and then for 2 × 30 min in TBST (50 mM Tris·HCl pH 7.5, 150 mM NaC, 0.1% Tween 20) at room temperature. Sections were then blocked for 1 h in TBST containing 20% heat-inactivated FCS and 2% blocking reagent (Roche) and incubated in blocking solution with alkaline phosphatase-conjugated antidigoxigenin antibody (Roche), 1:1,000 dilution at 4°C over night. After several washes in TBST the slides were equilibrated in staining buffer, 100 mM NaCl, 50 mM MgCl2, 100 mM Tris·HCl pH 9.5, 0.1% Tween 20. The color reaction was performed in 10% polyvinyl alcohol, 100 mM NaCl, 5 mM MgCl2, 100 mM Tris·HCl pH 9.5, 0.1% Tween 20, 100 mg/ml 4-nitro blue tetrazolium chloride, and 50 mg/ml 5-bromo-4-chloro-3-indolyl-phosphate for 10–20 h. The color reaction was stopped in PBS, and the slides were mounted with 70% glycerol-PBS.


Validation of clean tissue separation.

The entire small intestine (duodenum to cecum) was removed from E18.5 fetuses and separated into epithelial and mesenchymal fractions as shown in Fig. 1, C and D (30). To assess the efficacy of the separation, cDNA was prepared from each individual fraction and analyzed by RT-PCR using primers for genes known to be expressed exclusively in either the epithelial (Villin1, Ihh) or mesenchymal (Actg1, MAdCAM) compartment. Only fractions that appeared to be free of contamination from the other compartment were used for microarray analysis (Fig. 1E).

Six microarray chips (three mesenchymal mRNA and three epithelial mRNA) were hybridized using independently isolated samples. The difference in relative expression of each probe set was calculated as described in materials and methods and expressed as a numerical FD or ES (equal to the absolute value of the FD). In Fig. 2A, all data are plotted according to [−Log10(P value), y-axis] vs. [Log2(FD), x-axis]. The boundaries of FD ≥ 2.0 and P ≤ 0.05 are indicated in the figure; these boundaries demarcate genes that are enriched in the epithelium (Epi: 1,812 known genes; 111 unknown ESTs or cDNAs) and in the mesenchyme (Mes: 4,245 known genes; 417 unknown ESTs or cDNAs). Complete lists of epithelially and mesenchymally enriched genes, along with their average relative expression levels and ES are provided in Supplementary Tables S2 and S3, respectively.

Fig. 2.

Microarray data distribution. A: as detailed in the text, the limits of fold difference (FD) = 2.0 and P ≤ 0.05 demarcate the genes enriched in epithelium and mesenchyme. B: average expression values as calculated by robust multiarray average (described in materials and methods) for all probes are displayed for epithelium (y-axis) and mesenchyme (x-axis). Points above the diagonal are epithelially enriched genes; those below the diagonal are enriched in mesenchyme.

The average relative expression value in epithelium (y-axis) vs. mesenchyme (x-axis) for each of the queried probes in the array as determined from the RMA analysis is graphically depicted in Fig. 2B. The data are well spread over ∼11 relative expression units (25–215). We examined the epithelial and mesenchymal expression levels for several housekeeping genes that are known to be ubiquitously expressed: Hprt, Gapdh, Ppig (cyclophilin G), and Actb. In each case, the ES is close to 1.0 (Table 1). Thus, these genes will fall on the diagonal in Fig. 2B.

View this table:
Table 1.

Array data for known compartmentalized genes

Additionally, we queried the data for several genes known to be compartment specific, including structural genes, transcription factors and secreted factors: Vil1, Fabp1, Cdx1, Cdx2, and Ihh for the epithelium; Vim, Actg2, Foxf1, Gli1, and Bmp4 for the mesenchyme. The results (Table 1) show robust compartment-specific ES for the genes that are expressed at high levels (e.g., Villin1 in the epithelium; Vimentin in the mesenchyme). However, for genes expressed at lower levels or in few cells (e.g., Ihh in the epithelium), the ES are lower (Table 1). This is a consequence of the low positive expression value in one compartment coupled with a relatively constant background signal in the other compartment of 4–5 relative expression units. Thus, low enrichment values (e.g., for Ihh, ES = 2.8) can still be indicative of compartment-specific expression. Indeed, experimental data obtained by in situ hybridization, immunohistochemistry, and PCR have confirmed that both Ihh and Shh (ES = 1.9) are in fact exclusively epithelial (30, 39). Thus, the array results present an accurate prediction of epithelial vs. mesenchymal compartmentalization. Nevertheless, some genes that are expressed at low levels or in few cells will be missed at a cut-off value of ES = 2.0, even if they are (like Shh), truly compartment specific.

Functional attributes of genes expressed in mesenchyme and epithelium.

As a second check on the reliability of the array data, we tallied the Gene Ontology (GO) terms linked to genes with FD > 5.0 and P ≥ 0.05. GO terms provide a “dictionary” of labels that describe three aspects of each gene: the Biological Process term reflects the cellular processes to which each gene is linked (e.g., signaling, proteolysis); the Cellular Component term describes the cellular organelle with which the protein is associated (e.g., cytosol, nucleus), and the Molecular Function term denotes the function performed by the transcribed protein (e.g., transcription factor, kinase).

As shown in Fig. 3, the Biological Process terms differ strikingly between the two compartments in a way that reflects their known functions. Among epithelially enriched genes, nearly 50% have linked Biological Process terms that reflect a role in nutrient absorption and processing: metabolism (21%), transport (20%), and proteolysis (7%). In contrast, signal transduction (12%) and transcription (12%) are the most frequently linked Biological Process GO terms for the mesenchymally enriched genes. In concert with this, the most prominent Cellular Component GO term for the epithelium (membrane, 33%) reflects its role in absorption and transport, while the predominant Cellular Component term for the mesenchyme (“extracellular,” 33%) is in keeping with the involvement of the mesenchyme in extracellular matrix production and instructional signaling.

Fig. 3.

Biological Process Gene Ontology terms for epithelium (left) and mesenchyme (right). Each term is represented by a different color; the same term is represented by the same color in each compartment (i.e., green is signal transduction). The percent representation of each term is denoted schematically by the pie chart and provided numerically. All terms shared by <1% of the genes are grouped into the category called Other.

Tissue-specific signature held by the epithelium.

We next examined the tissue specificity of the 100 most-enriched genes in the epithelium and mesenchyme (see Supplementary Tables S2 and S3). These genes tend to be among the more highly expressed genes in their respective compartments. The tissue specificity of each gene was estimated by counting the number of different tissues from which ESTs for that gene have been isolated according to the EST ProfileViewer (Unigene, NCBI). This analysis shows that the epithelially enriched genes are generally expressed in few tissues outside of the intestine while mesenchymally enriched genes are broadly expressed in multiple tissues (Fig. 4). A total of 47 of the 100 most epithelially enriched genes are expressed in 10 or fewer tissues, while only five of the most mesenchymally enriched genes (Phox2, Adamdec1, Colec10, Cnn1, Sh3bgr) are this restricted.

Fig. 4.

Tissue specificity of genes enriched in epithelium and mesenchyme. The number of tissues in which the top 100 epithelially and mesenchymally enriched genes are expressed was determined from expressed sequence tag counts using the Unigene Expression Profile viewer at the National Center for Biotechnology Information, linked to the Unigene page. Each square (Epi) or triangle (Mes) indicates a separate gene. While 50% of epithelial genes are expressed in 10 or fewer tissues, only 5 of the mesenchymally enriched genes show this degree of tissue restriction.

Compartmentalized transcription factors in mouse intestine.

Transcription factors play critical roles in controlling signaling cascades. A recent study revealed that >1,000 such factors are expressed in various regions of the gut tube during development (6). From our epithelially and mesenchymally enriched gene sets (Supplementary Tables S2 and S3), we pinpointed all transcription factors that are enriched by at least 2.0-fold (P ≥ 0.05) in each compartment. These genes were initially selected if they were tagged with the Biological Process term “transcription,” the Cellular Component “nucleus,” and/or the Molecular Function term “DNA binding.” Genes were then searched individually to either confirm experimental evidence for function related to Pol II transcriptional activity or to identify specific domains (e.g., Krüppel-type zinc finger domains) that are highly suggestive of transcriptional function. A total of 76 real and putative transcription factors were found to be enriched in the epithelium (Supplementary Table S4), and 373 were identified as enriched in mesenchyme (Supplementary Table S5). These 449 factors that are differentially expressed between epithelium and mesenchyme represent ∼40% of the total number of transcription factors expressed in the developing gut, according to a recent study by Choi et al. (6). The Choi analysis included both stomach and small intestine, and the analysis was done at four time points during development (E11, E13, E15, and E17). Since our study only examines one time point (E18.5) and one gut region (small intestine), it is striking that so many factors are compartment enriched.

The list of highly enriched epithelial transcription factors includes many with known roles in gut development and/or disease (e.g., Hnf4α, Hnf4γ, Klf5, Ehf, Cdx1, Cdx2, Gata4, Gata5, Gata6, etc.). Nevertheless, among the 10 most highly enriched epithelial factors are several without previously described roles in the intestine. Tcfec, the third most epithelially enriched factor (ES = 18), is a bHLH-leucine zipper family member closely related to Mitf, Tfe3, and Tfeb, members of the micropthalmia class of transcription factors that play critical roles in eye and melanocyte development (41). Creb3l3 (Creb-H; ES = 11) is a bZip family member that is abundantly expressed in liver, where it appears to act as a suppressor of proliferation (5). In accord with this, it is downregulated in hepatomas (5). Similarly, Ehf (ESE-3; ES = 11), an Ets family member, is also modulated in cancer and has a postulated role in the differentiation of glandular epithelia (27). The robust expression of these factors in the intestinal epithelium suggests that they would be interesting targets for further investigation.

Figure 5A provides RT-PCR data that verify transcription factor assignments made by the microarray for the 16 most enriched factors in the epithelium. These factors were between 66- and 6-fold enriched in epithelium, according to the array data, and the PCR results are in accord with these data. A few transcription factors with lower ES were also examined by RT-PCR: longevity assurance homolog 6 (Lass6, ES = 3.3); RAR-related orphan receptor gamma (Rorc, ES = 3.4); E2F transcription factor 5 (E2F5, ES = 3.0); and hairy and enhancer of split 6 (Hes6, ES = 2.1). In each case, PCR evidence of epithelial predominance was also obtained, indicating that even when the ES is low, the array results accurately reflect compartmental enrichment.

Fig. 5.

Transcription factors enriched in the epithelium. A: PCR verification of epithelial enrichment for the top 16 most enriched genes in the epithelium (above the line under Npas2). A few genes of lower enrichment values were also tested (Lass6, Rorc, E2F5, and Hes6). B: distribution of the classes of transcription factors found in mesenchyme (white bars), compared with the distribution of classes in the entire genome (cross-hatched bars). Data for the genomic distribution is from Gray et al. (15). C: distribution of the classes of transcription factors found in epithelium (dark gray bars), compared with the distribution of classes in the entire genome (light gray bars). ZnF, zinc finger; HOX, homeodomain; HLH, helix loop helix; HMG, high mobility group; NHR, nuclear hormone receptor; FH, forkhead; ETS, Ets factor; Other, all other classes.

Nearly 400 transcription factors were found to be at least twofold enriched in the mesenchyme (Supplemental Table S5). This is almost five times the number of transcription factors found to be enriched in the epithelium. Perusal of the mesenchymal list reveals factors known to be expressed in hematopoietic cells (Fli1), enteric neurons (Phox2), smooth muscle cells (myocardin), or myofibroblasts (Foxf1). Thus, the heterogeneous cell composition of the mesenchymal compartment may account for the large number of transcription factors that are enriched there.

Of the 10 most highly enriched factors in this compartment, three (Ets1, Elk3, and Fli1) are members of the Ets transcription factor family. These factors have overlapping gene expression patterns in fibroblasts, endothelial cells, and hematopoietic cells. The active expression of all three of these Ets factors is in accord with an important mesenchymal participation in matrix remodeling and cell migration in the E18.5 intestine (2, 24, 33).

Enrichment of transcription factor subtypes.

Gray et al. (15) recently mined the mouse genome to identify all transcription factors and categorized each factor according to its type of DNA binding domain. We compared the frequency of each type of DNA binding factor in our collection of epithelial or mesenchymal transcription factors to the genomic frequency of each type as compiled in the Gray study. The classifications of each factor identified in the epithelium and mesenchyme are included in Supplementary Tables S4 and S5, and the compiled results of the distribution analysis are presented in Fig. 5, B and C. The zinc finger (ZnF) class of transcription factors is the predominant class in the mouse genome (15). Interestingly, the relative frequency of the various types of mesenchymally enriched transcription factors closely mirrors that of the entire mouse genome (Fig. 5B). However, the epithelially enriched factors exhibit a very different profile, with an under-representation of ZnF and HMG (high mobility group) factors and an overrepresentation of HLH (helix loop helix), bZip and NHR (nuclear hormone receptor) classes (Fig. 5C). The prominence of NHR factors in the epithelium is particularly striking. These proteins make up only 2% of the mesenchymally enriched transcription factors but represent 17% of the epithelially enriched factors (Fig. 5B).

Hnf4 binding sites in epithelial genes.

Hnf4α and Hnf4γ are the most highly enriched transcription factors in the epithelium. In fact, Supplementary Table S2 reveals that both Hnf4α and Hnf4γ rank very high (#7 and #22, respectively), in the total list of epithelially enriched molecules. Hnf4α has been shown to bind to a large number of promoters in the liver (34) and functional studies indicate that it is essential for epithelial differentiation in the liver(37) and the colon(14). If Hnf4 proteins are also key regulators of transcription in the small intestinal epithelium, then a preponderance of epithelially enriched genes should exhibit cis binding sites for Hnf4. We downloaded 500 bp of sequence upstream from the 100 most enriched epithelial genes and the 100 most enriched mesenchymal genes from Ensembl. In some cases, more than one promoter (more than one transcriptional start site or TSS) was identified for a single gene. Thus, a total of 114 and 123 sequences were downloaded and searched for the 100 epithelial and 100 mesenchymal genes, respectively. MatInspector from the Genomatix software suite ( was used to search for potential Hnf4 binding sites in these sequences. As outlined in materials and methods, this program searches for two variants of the consensus Hnf4 binding site, both verified by functional binding site selection and affinity studies (12). When a single site was identified by both motifs, it was counted only once. To establish a baseline for the frequency of finding these cis elements in a similar size set of random genes, we also examined three groups of 120 sequences from among the genes that showed no enrichment between epithelium and mesenchyme (ES = 1.0). The three groups included genes expressed at high, medium, and low relative expression values, as described in materials and methods.

Using the default match criteria of MatInspector, we found 61 sequences from 114 epithelially enriched genes to contain putative Hnf4 sites (listed in Supplementary Table S6). In contrast, Hnf4 sites were found in only 29 sequences among the 123 sequences from mesenchymally enriched genes. Among the four groups of 120 sequences with enrichment scores of 1.0, the number of Hnf4 binding sites ranged from 23 to 37, similar to the number seen in mesenchymally enriched genes and approximately half of the number seen in epithelially enriched genes (Fig. 6A). In addition, the actual number of Hnf4 binding sites found was 42 in mesenchymally enriched genes and 85 in the epithelial set. Control groups ranged from 29 to 56 binding sites, similar to the number seen in the mesenchymally enriched set and half of the number seen in the epithelial set. Increasing the stringency of the Hnf4 match does not alter these findings, since 20 of the binding sites in epithelial genes have a >90% match to the consensus sequences (Supplementary Table S6, see Matrix Similarity). However, only two sites in the mesenchyme match at this level (data not shown).

Fig. 6.

Distribution of Hnf4 binding sites. A: the total number of sequences analyzed for each group is shown in the first set of bars. Epithelium and mesenchyme are represented by the black and white bars, respectively. Three control groups of genes with enrichment score (ES) = 1.0 are represented by various shades of gray patterned bars. The 3 control groups are composed of genes expressed at low (L; average expression value <8.0), medium (M; average expression value = 8.0–10.0), and high (H; average expression value >10.0) relative expression values, respectively. The central set of bars display the number of sequences that were positive for Hnf4 binding sites. Some of these sequences have >1 binding site. The 3rd set of bars indicates the total number of Hnf4 binding sites found in all sequences of each type searched. B: the location of Hnf4 binding sites is graphically displayed for each of 5 100-bp “bins” across the 500 bp sequences analyzed. The epithelium and mesenchyme are represented by open diamonds and open triangles, respectively. The 3 control groups are represented by closed symbols. Note the dramatic difference between the epithelial group and all other groups in the first 100 bp.

We next examined the location of consensus Hnf4 binding sites within the 500 bp promoters of mesenchymally and epithelially enriched genes containing such sites. Interestingly, Hnf4 binding sites appeared to be strikingly scarce in the first 100 bp of promoter sequence of all three control groups (solid symbols in Fig. 6B) and randomly distributed over the 200- to 500-bp range of these sequences. Importantly, the mesenchymally enriched genes (open triangles in Fig. 6B) displayed a pattern very similar to those of the three control groups. In contrast, consensus Hnf4 binding sites were highly concentrated in the first 300 bp of the promoters of epithelial genes (Fig. 6B, open diamonds). The number of binding sites was particularly high in the region between 0 and −100 bp from the transcriptional start site, the same region in which such sites appear to be scarce in mesenchymal and control genes. Taken together, these results predict that Hnf4 regulates a large number of intestinal epithelial genes, as in liver and colon. These data are in accord with those of Stegmann et al. (44) who found Hnf4 sites in a large number of intestinal genes that are highly upregulated during development of the intestine from the early fetal (E13) to adult stage and/or are upregulated in adult villus tips compared with crypts.

Mesenchyme as a modulator of signal transduction.

To learn more about the direction of signaling cross talk in the intestine, we next examined the enrichment values for molecules that participate in several signaling pathways known to be important in intestinal development and homeostasis, including: Notch, Hedgehog (Hh), Wnt, Igf, Fgf, and Bmp. Genes enriched more than twofold in either compartment are listed in Tables 2 and 3.

View this table:
Table 2.

Compartmentalization of signaling pathway genes

View this table:
Table 3.

Compartmentalization Bmp pathway transcripts

Notch signaling in the epithelium is essential for the control of epithelial lineage allocation (32, 43, 46), and recent work has also implicated Notch signaling in the expansion of the crypt progenitor pool (13). The array data collected here are consistent with low levels of Notch expression in the epithelium (average expression values of 7.3–9.5), but it is interesting that all four of the Notch genes are actually more enriched in the mesenchyme (Table 2): Notch1 (ES = 3.1), Notch2 (ES = 6.2), Notch3 (ES = 4.8), and Notch4 (ES = 6.4). Thus, it is likely that important mesenchymal roles for this pathway also exist.

Current experimental data indicate that Hh signals originate in the epithelium and that the Hh signal transduction machinery is located exclusively in the mesenchyme (30). The microarray data are consistent with this (Table 2). Though enrichment values for Ihh and Shh in the epithelium are low, most likely because of their expression in few cells, significant mesenchymal enrichment is seen for the receptors Ptc1 (ES = 27.7) and Ptc2 (ES = 3.4), co-receptor Smo (ES = 4.9), and transcription factors Gli1 (ES = 6.3), Gli2 (ES = 3.4), and Gli3 (ES = 9.3). A single pan-inhibitor of the Hh signal has been described: Hh interacting protein (Hhip) (7). Like Ptc, this membrane-bound inhibitor binds Hh with high affinity and is also a direct target of Hh signaling. Hhip expression was earlier shown to be restricted to the intestinal mesenchyme (30); in agreement with this, the array data suggest a 25-fold enrichment of Hhip in the mesenchymal compartment.

Information concerning the spatial distribution of several Wnt signaling molecules in the developing and adult intestine has been previously published (17, 29, 31, 45). The microarray data for E18.5 fetal intestine largely corroborate those studies, with a few interesting differences (see Table 2). First, the noncanonical receptor, Fzd6, appears to be expressed in both compartments of the late fetal intestine but is somewhat more enriched in the mesenchyme (ES = 3.3). In contrast, in adult intestine, expression of this factor is clearly predominant in the epithelium (17). Fzd1, Fzd2, and Fzd7 are also strongly mesenchymally enriched in our E18.5 samples (ES = 8.0, 15.6, and 10.6, respectively). This correlates well with the adult findings for Fzd1 and Fzd2, which are both expressed in smooth muscle, but not for Fzd7, which is epithelial in the adult (17). Finally, the inhibitors Dkk-2 and Sfrp-2 are highly mesenchymally enriched in the E18.5 intestine (ES = 10 and 11.7, respectively) but were not detected in the adult intestine (17). These differences suggest a temporally dynamic patterning of Wnt pathway molecule expression.

It is striking that Wnt5a (ES = 15.3) and Fzd2 (ES = 15.6) exhibit the highest enrichment values for any Wnt family member and Frizzled receptor, respectively, and both are enriched in mesenchyme. Both of these proteins can function in the noncanonical Wnt signaling pathway, as can Fxd6, which is enriched 3.3-fold in the mesenchyme. These findings suggest that β-catenin-independent signaling may play an important role in this compartment.

As seen with the Hh signaling system, the array data indicate that the predominant site for manufacture of inhibitors of the Wnt pathway is the mesenchyme. Sfrp1, Sfrp2, Dkk2, and Dkk3 are all greatly enriched in this compartment (ES = 78.5, 11.7, 10, and 8, respectively). The relative expression values suggest that Sfrp4 and Sfrp5 are expressed at low to moderate levels in both epithelium and mesenchyme (Supplementary Tables S2 and S3), but strikingly, none of the Wnt inhibitors are preferentially enriched in the epithelial compartment.

In the Igf signaling system, several Igf binding proteins (Igfbp) act to inhibit or potentiate Igf activity, often in a context-specific manner (8). Like the Wnt and Hh inhibitors, Igfbps are prominently expressed in the mesenchyme and in fact are among the most highly expressed and highly mesenchymally enriched of any of the signaling molecules analyzed (Table 2). Enrichment values of >20-fold are seen for Igfbp2, Igfbp3, Igfbp4, Igfbp5, and Igfbp7. Messenger RNA for Igfbp3, the binding protein to which the majority of circulating Igf is bound(18), is >100-fold enriched in mesenchyme and has a very low relative expression value (5.7) in epithelium. This protein has been shown to possess both Igf-dependent and Igf-independent activities and has proapopototic activity in the absence of Igf (21). Igf1 as well as the receptor Igf1r (which binds both Igf1 and Igf2) are also enriched in mesenchyme.

Fgf signaling has been shown to play a role in anterior/posterior patterning of the early gut tube (9), but its roles in later gut development are unclear. By far, the most robustly expressed Fgf family member in the E18.5 intestine is Fgf13, which is enriched 47.6-fold in mesenchyme (Table 2). Many other Fgfs are expressed at low levels in both epithelium and mesenchyme (relative expression values of 7 or greater), including Fgf1, Fgf3–15, Fgf17, Fgf18, Fgf21, and Fgf22 (data not shown). However, only Fgf5, -7, and -15 are enriched in a particular compartment, with Fgf5 and Fgf15 enriched in epithelium (ES = 2.3 and 2.9, respectively) and Fgf7 enriched 2.2-fold in mesenchyme (Table 2).

Two findings suggest that Fgf signaling may be particularly important in mesenchyme. First, of the four Fgf receptors, Fgfr1 and Fgfr2 are highly expressed in mesenchyme and are enriched by 35.5- and 3.7-fold, respectively, in that compartment. Fgfr3 and Fgfr4 are expressed at much lower levels, likely in both epithelium and mesenchyme (data not shown). Second, several members of the Sprouty (Spry) and Sprouty-related (Spred) families are expressed predominantly in the mesenchyme. These proteins are intracellular inhibitors of Fgf signaling and Spry2 is a direct target of Fgf signals(4). As shown in Table 2, Spry1, 2, and 4 are enriched 5-, 7-, and 35-fold in mesenchyme, respectively; Spred1 and Spred2 are also mesenchymally enriched (ES = 13.6 and 3, respectively). Interestingly, Fgfbp1, a soluble enhancer of Fgf signals, is one of the few signaling modulators that we found to be enriched in epithelium (ES = 6.2). Fgfbp1 binds directly to several Fgfs and potentiates their activity (1).

Bmp pathway: modulation by numerous mesenchymal factors.

Recent investigations have suggested that Bmp signaling to the epithelium is important in patterning the crypt/villus axis and in pathological states such as juvenile polyposis syndrome or JPS (19, 20, 22). However, the expression of Bmp pathway molecules other than Bmp4 has not been systematically examined in the intestine; thus we mined the array data for compartmentalized Bmp pathway molecules (Table 3).

Bmp4 is highly expressed and enriched (ES = 23.8) in the mesenchyme, in agreement with previous reports(19, 20, 30). However, this is not the only Bmp family member that is enriched in this tissue layer: Bmp5 is 25-fold enriched, Bmp6 is 3.3-fold enriched, and Bmp2 is 2.2-fold enriched in mesenchyme. Bmp7 is the only family member showing epithelial enrichment (ES = 4.7). PCR validation of these results is provided in Fig. 7A. In situ hybridization studies further confirmed these expression patterns (Fig. 7D). Particularly noteworthy here is the pattern of Bmp5, which, like Bmp4, is highly expressed in stromal cells adjacent to the epithelium.

Fig. 7.

Distribution of Bmp signaling pathway molecules in epithelium and mesenchyme. A: PCR verification of microarray results for several Bmp ligands. Note that Bmp7 is the only ligand that is slightly enriched in the epithelium. B: PCR verification of array results for Bmp receptors and Smad signaling proteins. All are expressed in both epithelium and mesenchyme, with slight enrichment visible for some in the mesenchyme. C: PCR verification of the location of inhibitors and modulators of Bmp signaling. The majority of these proteins are expressed exclusively or predominantly in the mesenchyme. None are preferentially enriched in the epithelium. D: in situ hybridization on sections of intestine from E17.5 embryos, using antisense probes for various Bmp pathway molecules: Bmpr2, Acvr1, Bmpr1a, Bmp2, Bmp4, Bmp5, and Bmp7. All tissue sections are from the proximal small intestine, except for Bmpr1a, which is from the distal small intestine.

The array data suggest that the receptors and signal transducers (Smads) of the Bmp pathway are expressed in both epithelial and mesenchymal compartments (Table 3, and see the entire array database at with the series record number GSE6383). Bmpr2, the most robustly expressed of all of the receptors, is enriched fourfold in mesenchyme, though it is also expressed in the epithelium (Table 3, Fig. 7D). Two of the three type I receptors, Bmpr1a (Alk3) and Acvr1 (Alk2), are mesenchymally enriched (2.6-fold and 4.5-fold, respectively), while the third, Bmpr1b (Alk6) is expressed at much lower levels in both mesenchyme and epithelium and not enriched in one particular compartment (data not shown). In situ hybridization confirms mesenchymal enrichment as well as expression in the epithelium at the base of the villi for Bmpr2 and Acvr1 (Fig. 7D). Bmpr1a is expressed primarily in the mesenchyme in the proximal duodenum (data not shown) but is found in both compartments in the more distal small intestine (Fig. 7D). Of the Smads, Smad1, Smad4, Smad5, Smad6, and Smad7 are expressed, and all except Smad4 are slightly enriched in the mesenchyme (Table 3, Fig. 7). Smad4 exhibits similar, high relative expression values of 10–11 in both epithelium and mesenchyme. These data indicate that Bmp signals most likely impact both epithelial and mesenchymal compartments, though studies to date have primarily addressed epithelial signal transduction (20).

The Bmp pathway is particularly rich in molecular inhibitors and modulators and these can act as on/off switches for Bmp function (50). Strikingly, and consistent with findings for the Hh, Wnt, Fgf, and Igf families, the expressed soluble inhibitors and modifiers of the Bmp signaling pathway are located primarily or exclusively in the mesenchyme, including: three follistatin family members (follistatin, follistatin 5, and follistatin-like 1), chordin-like 1, twisted gastrulation, Tolloid (Bmp1), Tolloid-like-1, Bmper (Cv2), Gremlin1, and Crim1 (Fig. 7, Table 3). This finding places the major control of Bmp signaling activity squarely in the domain of the mesenchymal compartment.


In this study, we established an intestinal tissue catalog wherein hundreds of intestinal genes are categorized according to their likely location in the epithelium and/or mesenchyme. Because the technique of tissue separation is very effective and little contamination is detectable between compartments (Fig. 1, C–E), compartmentalization is clear and the data are verifiable by PCR. Within this dataset, all transcription factors that are enriched in epithelium or mesenchyme were identified. The expression of genes involved in several signaling pathways that mediate intertissue communication were also catalogued, providing a clearer picture of the contributions of each compartment in tissue cross talk. These data will provide a useful guide that will contribute to the future dissection of developmental and pathological processes in the intestine.

Since heterotypic grafting experiments have shown that the mesenchyme holds a well-demonstrated power to instruct the regional specificity of the epithelium, in both the embryo and the adult (28), we were particularly interested to investigate whether the transcriptome of the epithelium and/or mesenchyme could suggest a molecular basis for this. We found that the epithelial transcriptome is highly tissue restricted, as expected given its specialized functional roles. However, the underlying mesenchyme exhibits a rather ubiquitous-looking gene expression profile. Similarly, at the transcription factor level, the profile of transcription factor classes expressed by the epithelium differs markedly from the profile of the genome as a whole, while the mesenchymal transcription factor profile closely resembles that of the genome. It is possible that instructional signals from the mesenchyme are powered by the few more tissue-specific genes present in that tissue (e.g., genes such as Foxf1 and Nkx2.3). In support of this, the Nkx2.3 null mouse as well as the Foxf1+/−/Foxf2+/− double heterozygote exhibit defects in mesenchymal as well as epithelial patterning (35, 36, 48). Alternatively or additionally, the instructional ability of the mesenchyme might rely on the use of a large arsenal of soluble signaling proteins in a combinatorial manner; indeed we show here that the mesenchymal transcriptome contains such an arsenal. Across several signaling pathways, (Notch, Hh, Wnt, Igf, Fgf, and Bmp), a mesenchymal predominance is notable for several of the signaling molecules themselves (Table 2). But most striking for all of these pathways is the remarkable number of soluble inhibitors and modulators of signaling that appear to be expressed predominantly or exclusively in the mesenchyme. Thus, the mesenchyme has an enormous potential to activate or suppress these important instructional pathways.

We paid particular attention to the Bmp pathway in this study since this pathway is important in both development and disease, yet few of the multiple molecular participants in this pathway have been carefully studied. The array data revealed that the mesenchyme has the ability to control multiple aspects of Bmp-mediated intestinal patterning and homeostasis. Among the Bmp ligands, both Bmp4 and Bmp5 are highly enriched in the mesenchymal compartment. Bmp5 has been less widely studied than Bmp4, but a mouse mutation has been identified that has an intestinal phenotype (short ear, se). Se mice exhibit defective skeletal structures and show changes in the morphology of several soft tissues, including the intestine, where intestinal looping is altered (16). Among Bmp inhibitors and modulators, robust enrichment of a remarkably large variety of these molecules is observed in the mesenchyme (Table 3, Fig. 7). Given the fact that JPS (23), a relatively rare autosomal dominant syndrome involving hamartomatous polyps that causes a predisposition to gastrointestinal cancer, is known to involve alterations in Bmp pathway signaling, it will be important to test whether mutations in Bmp5 or any of these multiple Bmp pathway inhibitors can also give rise to this syndrome.

The distribution of Bmp receptors and Smads in epithelium and mesenchyme suggests that both compartments can receive and transduce Bmp signals. Particularly interesting is the distribution of Smad4 and Bmpr1a, both of which are transcribed in epithelium as well as mesenchyme; Bmpr1a is actually enriched in the mesenchyme, particularly in the proximal small intestine, though Smad4 is not. Inactivating mutations in these two genes account for 40% of the cases of JPS (23). Interestingly, there is still debate as to whether the epithelium or the mesenchyme is responsible for initiation of the JPS pathology. Early studies of patients with SMAD4 mutations revealed clonal genomic deletions only in the mesenchyme (26). Thus, the epithelial malignancy was postulated to be due to landscaping by the mesenchyme. Later work identified loss of both SMAD4 alleles in the epithelium, a finding more consistent with a tumor suppressor mechanism (49). But interestingly, biallelic SMAD4 loss was found in some stromal and pericryptal fibroblasts of the polyp, suggesting a possible clonal origin for the mutant epithelium as well as part of the stroma of the polyp. The authors speculated that epithelial cells that lose the ability to receive or process Bmp signals might give rise to mesenchymal cells. In this light, it is interesting that epithelial/mesenchymal transition in the kidney is regulated by the balance of Bmp (promoting epithelial) and Tgfβ (promoting mesenchymal) signals (47). Though functional corroboration is needed, the data presented here are consistent with the possibility that the mesenchyme controls this balance and thus may be the compartment that is primarily responsible for the emergence of JPS pathology.

In summary, the data in this study provide a starting place for decrypting gene expression in the wild-type perinatal small intestine. It is clear, however, that these patterns of gene expression are labile; they are both time sensitive and subject to change in disease or injury. Nevertheless, these data will be of great value for tracing signaling cross talk and may eventually lead to the development of molecular tools for the manipulation of these signaling pathways. Functionally, this global view of the different mesenchymal and epithelial transcriptomes reveals compartmentalization of many of the molecules that direct epithelial/mesenchymal cross talk. The data indicate that a complex but nonintestine-specific mesenchymal tissue secretes multiple soluble molecules that serve to support and direct an epithelial signature that is uniquely intestinal.


D. L. Gumucio acknowledges support from National Institute of Diabetes and Digestive and Kidney Diseases Grant P01-DK-062041 and the University of Michigan Center for Computational Medicine and Biology. W. Zacharias and Å. Kolterud are both Fellows in the Organogenesis Training Program (NIH-T32-HL-07505).


The authors are grateful to Dr. Linda Samuelson for helpful discussions and critical reading of the manuscript.

The authors acknowledge the excellent service of the Microarray Core Facility in the University of Michigan Diabetes Center. Tissue processing was facilitated by the Center for Organogenesis Morphology Core.

Current address for B. B. Madison: Dept. of Genetics, Univ. of Pennsylvania, Philadelphia, PA.


  • Article published online before print. See web site for date of publication (

    Address for reprint requests and other correspondence: D. L. Gumucio, Dept. of Cell and Developmental Biology, Univ. of Michigan Medical School, 109 Zena Pitcher, 2045 BSRB, Ann Arbor, MI 48109-2200 (e-mail: dgumucio{at}


View Abstract