Multiomic spatial atlas shows deleted in malignant brain tumors 1 (DMBT1) glycoprotein is lost in colonic dysplasia

Abstract

Colorectal cancer (CRC) is responsible for over 900,000 annual deaths worldwide. Emerging evidence supports pro‐carcinogenic bacteria in the colonic microbiome are at least promotional in CRC development and may be causal. We previously showed toxigenic C. difficile from human CRC‐associated bacterial biofilms accelerates tumorigenesis in Apc ^Min/+ mice, both in specific pathogen‐free mice and in gnotobiotic mice colonized with a defined consortium of bacteria. To further understand host–microbe interactions during colonic tumorigenesis, we combined single‐cell RNA‐sequencing (scRNA‐seq), spatial transcriptomics, and immunofluorescence to define the molecular spatial organization of colonic dysplasia in our consortium model with or without C. difficile. Our data show a striking bipartite regulation of Deleted in Malignant Brain Tumors 1 (DMBT1) in the inflamed versus dysplastic colon. From scRNA‐seq, differential gene expression analysis of normal absorptive colonocytes at 2 weeks postinoculation showed DMBT1 upregulated by C. difficile compared to colonocytes from mice without C. difficile exposure. In contrast, our spatial transcriptomic analysis showed DMBT1 dramatically downregulated in dysplastic foci compared with normal‐adjacent tissue. We further integrated our datasets to generate custom colonic dysplasia scores and ligand‐receptor mapping. Validation with immunofluorescence showed DMBT1 protein downregulated in dysplastic foci from three mouse models of colonic tumorigenesis and in adenomatous dysplasia from human samples. Finally, we used mouse and human organoids to implicate WNT signaling in the downregulation of DMBT1 mRNA and protein. Together, our data reveal cell type‐specific regulation of DMBT1, a potential mechanistic link between bacteria and colonic tumorigenesis. Published 2025. This article is a U.S. Government work and is in the public domain in the USA. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Introduction

Among cancer‐related deaths worldwide, colorectal cancer (CRC) is the second most common cause [1]. The survival rate for early‐stage CRC is 90% compared to 14% for stage IV [2, 3]. Modifiable risk factors for CRC are well defined, but the effect of bacteria on colorectal carcinogenesis is not well understood. The list of bacteria associated with CRC, including Fusobacterium nucleatum and pks⁺ Escherichia coli, is growing [4, 5, 6, 7]. Suggested mechanisms for bacteria‐mediated tumorigenesis include damaging DNA, activating Wingless (WNT) signaling, blunting antitumor immunity, and promoting biofilm formation [8]. Bacterial colonization of mouse models for interrogating mechanisms of CRC tumorigenesis has included enterotoxigenic Bacteroides fragilis, pks⁺ E. coli, F. nucleatum, and bacterial biofilms [4, 6, 9, 10, 11]. However, not all bacterial inflammation is sufficient to induce tumorigenesis [12].

Previously, we showed that C. difficile from human CRC accelerates tumorigenesis in the distal colon of genetically susceptible Apc ^Min/+ mice [13]. These mice contain a truncation mutation in Apc leading to multiple intestinal neoplasia (Min). They develop 40–50 small intestinal adenomas, but few if any colonic adenomas. The mechanisms by which C. difficile increases tumor formation in the distal colon appear to involve WNT signaling, reactive oxygen species, and IL‐17 [13]. Here, we identified an epithelium‐specific factor, the host protein Deleted in Malignant Brain Tumors 1 (DMBT1). DMBT1 is a glycoprotein involved in mucosal immune defense and epithelial differentiation. It represents a potential molecular link between bacteria and cancer because of its dual functions in innate immunity and regulation of epithelial maturation [14]. DMBT1 is a 340 kDa glycoprotein with 13 homologous scavenger receptor cysteine‐rich domains that bind host immune factors, including immunoglobulin and complement [15, 16, 17, 18, 19, 20, 21, 22]. DMBT1 binds lipopolysaccharide, enabling it to agglutinate and exclude bacteria from the epithelium [17, 23]. From bulk tissue experiments, DMBT1 mRNA and protein have variable expression in gastric and colon carcinomas [19, 24, 25, 26, 27]. It remains unknown how DMBT1 affects CRC tumorigenesis.

Here, we created a multiomic atlas to characterize C. difficile‐associated colorectal tumorigenesis. We identified the cell type‐specific differential regulation of mouse Dmbt1 in colonic dysplasia and demonstrated the generalizability of this phenotype in human CRC.

Materials and methods

Mice

Animal experiments were approved by animal care at Johns Hopkins University, protocol MO23M87. C57Bl/6 mice of Apc ^Min/+ (Δ716 Apc) genotype were bred germ‐free as previously described [13]. At age 8 weeks, males were inoculated by single gavage of the 29‐member bacterial consortium described previously along with 10⁴ spores of C. difficile strain Cd_3728T [13]. The control group was given a single gavage of the same consortium without C. difficile.

scRNA‐seq

Mouse colons were harvested and washed in cold phosphate‐buffered saline (PBS). The distal two‐thirds were agitated in chelation buffer at 4 °C followed by mechanical and enzymatic dissociation using cold‐activated protease/DNase. Single cells were encapsulated into aqueous droplets using lysis buffer and polyT‐oligo beads and inDrop [28]. Purification, cDNA library polymerization, and indexing were performed as described previously [13, 29]. Sequencing was performed using a NovaSeq6000 (Illumina, San Diego, CA, USA). Raw sequences were aligned to genome GRCm38.85 to generate count matrices using the dropEst pipeline [30]. Counts matrices were filtered using Dropkick [31]. Preprocessing, dimensionality reduction, clustering, and correlation visualization were performed in Python (v. 3.9) using Scanpy [32]. Differential gene expression and gene set enrichment analysis were performed in R (R v. 4.4.2, Vienna, Austria) using Scanpy and GSEApy, respectively [33].

Spatial transcriptomics

Mouse colons were harvested, fixed, and embedded using standard techniques. Tissue sections (5‐μm) were processed using the GeoMx RNA protocol. Standard markers provided in the kit included: CD45‐594, PanCK‐532, and SYTO‐13 (NanoString Technologies, Seattle, WA, USA; ThermoFisher Scientific, Waltham, MA, USA). Regions of interest were based on normal versus dysplastic morphology. Segments were created with an epithelium mask based on pan‐cytokeratin‐positive staining and a lamina propria mask based on pan‐cytokeratin‐negative/CD45‐positive areas. Segments were separate samples for sequencing. GeoMx Digital Spatial Profiling was performed using the manufacturer's protocols and Mouse Whole Transcriptome Atlas [34] (NanoString Technologies). Barcodes were sequenced using a NovaSeq6000, and sequencing files were compiled with bcl2fastq (Illumina), demultiplexed, and converted to Digital Count Conversion files by NanoString (GeoMx DnD pipeline v2.3.4) [35]. For each no‐template polymerase chain reaction (PCR) control, a minimum of 1,000 raw reads was required to pass filtering, and each probe was required to pass the Grubbs outlier test in at least 20% of segments (α = 0.01), which left 16,471 targets. We removed segments containing less than 5% of targets above thresholds, which left 172 segments. Expression was normalized using the 3^rd‐quartile method. Differential expression was analyzed with linear mixed‐effect models, accounting for inter‐mouse heterogeneity.

Immunofluorescence

Mouse colons from the azoxymethane/dextran sodium sulfate‐induced cancer model and the transgenic Lrig1 ^CreER/+ ;Apc ^fl/+ adenoma model were obtained from previously published work [36, 37]. The human colon tissue microarray was obtained from the University of Virginia Pathology Department Cooperative Human Tissue Network, approved through IRB‐exemption #070166. We used anti‐β‐catenin (ref. 12F7D1, 1:400; Vanderbilt Antibody Protein Resource, Nashville, TN, USA), anti‐Dmbt1 (ref. HPA040778, 1:400; Sigma‐Aldrich, Saint Louis, MO, USA), goat anti‐mouse, and donkey anti‐rabbit secondary antibodies (ThermoFisher Scientific).

Murine organoids

Lrig1 ^CreER/+ ;Apc ^fl/fl colonic organoids were treated with 1 μm (Z)‐4‐Hydroxytamoxifen (Sigma‐Aldrich) for 72 h in Mouse IntestiCult (STEMCELL Technologies, Vancouver, BC, Canada). Three days posttreatment, organoids were spilt using TrypLE (Gibco, ThermoFisher Scientific, Waltham, MA, USA) for 3 min at 37 °C and plated at 300 cells/μl in 30 μl Matrigel (Corning, Corning, NY, USA) domes [38]. After 7 days, organoids were harvested for RT‐qPCR, viability, and immunoblotting.

RT‐qPCR

Total RNA was extracted using 1 ml TRIzol (Invitrogen, ThermoFisher Scientific, Waltham, MA, USA) for 10 min. RNA clean‐up was performed with Direct‐zol MiniPrep (Zymo Research, Irvine, CA, USA). cDNA was generated with Verso cDNA (ThermoFisher Scientific). TaqMan Fast advanced master mix and probes (ThermoFisher Scientific) were used in a QuantStudio thermocycler (Applied Biosystems, Foster City, CA, USA) following the manufacturer's protocol. Primer probe sets used were: ACTB (Hs01060665_g1), NKD1 (Hs01548773_m1), AXIN2 (Hs00610344_m1), and DMBT1 (Hs01069306_m1).

Immunoblotting

Five 30‐μl domes per condition were combined in cell dissociation solution (2 mm EDTA in PBS) then washed in PBS. RIPA with 0.1% SDS was added to organoids. Samples were lysed for 30 min before sonication. BCA protein assays (ThermoFisher Scientific) were performed following the manufacturer's protocol. SDS‐PAGE using NuPage LDS sample buffer was run in 3%–8% Tris‐acetate gels (Invitrogen). Proteins were transferred to nitrocellulose membranes overnight at 4 °C. Membranes were blocked in 5% milk in TBS‐T (0.1% Tween‐20) and probed with anti‐DMBT1 primary antibody (ref. SAB2108362, 1:600; Sigma‐Aldrich) and anti‐rabbit HRP antibody (Invitrogen). Membranes were developed using Pierce ECL (ThermoFisher Scientific).

Human organoids

Normal colonic organoids were derived under approved IRB #182165. Organoids were spilt into single‐cell suspension using TrypLE (Gibco) for 3 min at 37 °C and plated at 700 cells/μl in 30 μl Matrigel domes. Organoids were cultured in Human IntestiCult (STEMCELL Technologies) medium until domes were full. Media was replaced with 80% DMEM/F12: 20% IntestiCult, including CHIR99021 (Tocri, Bristol, UK) diluted in DMSO (Fisher, ThermoFisher Scientific, Waltham, MA, USA) at 10 μm for 72 h. Organoids were washed in PBS, and RNA was extracted.

Statistical analysis and data availability

All statistical analyses were performed using Python (v. 3.9). For scRNA‐seq differential expression analysis, we used a Wilcoxon test with Benjamini–Hochberg correction in Scanpy. Gene set enrichment was performed with GSEApy using random permutation (n = 1,000) and Bonferroni correction. Ligand‐receptor inferences were based on Wilcoxon tests using CellPhoneDB [39]. Statistical significance thresholds were p ≤ 0.05 and false discovery rate (FDR) ≤0.05 for all tests. Visualizations were created with listed packages, seaborn for violin plots, and ggplot2 for volcano plots [40, 41]. Institution‐supported artificial intelligence based on GPT‐4 (OpenAI) was used to troubleshoot errors (aichat.app.vumc.org). Scripts were verified by experienced users. Codes are available upon request. Sequence files and datasets are available: Geo accession GSE270726 and GSE270745.

Results

C. difficile‐colonized Apc^{Min /+} mice had expansion of unique cell types and states

Our prior work demonstrated that C. difficile from human CRC‐associated biofilms exacerbates adenoma formation in the distal colon of Apc ^Min/+ mice [13]. To identify the mechanism(s) driving tumorigenesis, we performed two transcriptomic experiments on colonic tissue from gnotobiotic Apc ^Min/+ mice. Germ‐free Apc ^Min/+ mice were given a single oral gavage of a 29‐bacteria consortium plus toxigenic C. difficile as described previously (Figure 1A) [13]. Control mice were given the same consortium but without C. difficile. Two weeks after inoculation, colonic tissue was collected for scRNA‐seq or formalin‐fixed and paraffin‐embedded for spatial transcriptomics (Figure 1B). Dimensionality reduction illustrated the relative similarities of clusters and cell type assignments using Uniform Manifold Approximation and Projection (UMAP) (Figure 1C). Unsupervised clustering was also performed (supplementary material, Figure S1A). We applied Cellular Trajectory Reconstruction Analysis using gene Counts and Expression (CytoTRACE) to predict the stem‐like potential of clusters based on the number of different genes expressed per cell (supplementary material, Figure S1B) [42]. The score shows, in addition to the stem and transit amplifying compartment (TAC) cells, the absorptive precursor 2 (Abs_Pre2) cluster contained cells predicted to have stem‐like capabilities. Similarly, differentiated cell types, like goblet cell types and mature absorptive cells (Abs1), were predicted to have low stem‐like potential with lower CytoTRACE scores (supplementary material, Figure S1B and Table S1). To enable direct comparisons between C. difficile‐inoculated mice and controls, we merged similar clusters irrespective of inoculation. Progenitor, goblet, deep crypt secretory, and stem clusters comprised a greater proportion of cells from C. difficile‐inoculated mice compared to controls (supplementary material, Figure S1C–E). Agr2‐high goblet cells were present mostly in controls, whereas Clca1‐high goblet cells were present mostly in C. difficile‐inoculated mice. Canonical marker genes were used for making cell type assignments (supplementary material, Figure 2A). Pearson correlations between clusters further supported the similarities from dimensionality reduction (supplementary material, Figure 2B). The distribution of cells from each condition and mouse demonstrated the relative contribution of each condition and validated that no individual mouse skewed the clustering (Figure 1D,E). Based on cell cycle‐specific genes (supplementary material, Table S2), the UMAP was colored to show the proliferating cells (Figure 1F).

Figure 1.

Cell type identification and analysis of scRNA‐seq from colonic epithelial cells in C. difficile‐colonised Apc ^Min/+ mice demonstrated unique cell states. (A) Schematic of experimental design: 8‐week‐old germ‐free Apc ^Min/+ mice were gavaged with C. difficile‐containing bacterial consortium or control consortium as previously published [13], and mice were euthanized at 10 weeks. (B) Epithelial crypts were dissociated from the colonic tissue for scRNA‐seq; n = 3 mice/group. Spatial transcriptomics was performed on formalin‐fixed, paraffin‐embedded whole colon; separate experiment with n = 3 mice/group. (C) UMAP plot with cell type assignments for clusters based on canonical mouse gene markers (supplementary material, Figure S2). (D–F) UMAP plots colored by inoculation (C. difficile or control), mouse, or cell cycle phase. (G–K) UMAP plots colored by scaled gene expression (a.u. = arbitrary units) of canonical gene markers for different cell states: fetal reversion (Marcksl1), autophagy (Sqstm1), senescence (Cdkn1), intestinal plasticity (Prom1), and regenerative stemness (Anxa1).

Figure 2.

Differential gene expression analysis of absorptive colonocytes from scRNA‐seq indicated increased immune activity and upregulation of Dmbt1 expression in C. difficile‐inoculated mice. (A) Violin plot of Dmbt1 expression from C. difficile‐inoculated mice versus control within each merged cell type, scaled expression, arbitrary units (a.u.). (B) Volcano plot of differentially expressed genes from absorptive cells based on inoculation. Labeled genes were among the most differentially expressed. (C) We used computational gene set enrichment analysis to determine whether the predefined set of genes in Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways showed statistically significant concordance in absorptive cells. Listed pathways contain genes that were differentially upregulated in C. difficile‐inoculated versus control mice; FDR = false discovery rate. (D) Gene enrichment plot of absorptive cells for the KEGG cell adhesion molecules pathway showed the degree to which upregulated genes correlated with C. difficile‐inoculation (Pos) compared to control (Neg) and the ranked metric quantification of the signal‐to‐noise ratio. NES = normalized enrichment score.

To characterize the cell states present in these clusters, we colored the UMAP plot to show expression for Marcksl1, Sqstm1, Cdkn1a, Prom1, and Anxa1, which represent fetal/embryonic reversion, autophagy, senescence, dysplastic priming, and regenerative stemness, respectively (Figure 1G–K) [43, 44, 45, 46, 47]. Interestingly, a subset of absorptive cells (Abs3, cluster 21 in Figure 1C) from C. difficile‐inoculated mice was enriched with these marker genes. This pattern suggests, perhaps, that this cluster simultaneously is responding to injury and is susceptible to dysplastic transformation [48]. We investigated whether these cells exhibited signs of gastric metaplasia based on the findings of Chen et al, who identified a gastric metaplasia signature when characterizing sessile serrated lesions associated with bacteria‐induced injury [49]. The Abs3 cluster scored highest compared with other clusters (supplementary material, Figure S2C), but was driven by expression of Axna1 and Gsdmd without a strong influence from other genes (supplementary material, Figure S2D).

Differential gene expression analysis of absorptive colonocytes

To identify genes contributing to cell type and cell state differences, we performed differential gene expression analysis comparing the inoculation groups within each cell type (Figure 2A, and supplementary material, Figure S3A–H; Table S3). Dmbt1 was among the most upregulated genes in the C. difficile‐inoculated mice. We focused on Dmbt1 as a potential factor in bacteria‐induced dysplasia because of its dual role as a pathogen‐associated molecular pattern receptor and an inducer of epithelial differentiation [14]. Dmbt1 expression was increased in C. difficile‐inoculated mice compared to controls in all cell types but was most highly expressed in absorptive and progenitor cells (Figure 2A,B and supplementary material, Tables S3 and S4). Gene set analysis of absorptive cells showed significant enrichment of cell adhesion and immune response (Figure 2C,D and supplementary material, Table S5). These pathways are implicated in colonic dysplasia [49]. The gene set enrichment analysis for absorptive precursors and deep crypt secretory cells comparing C. difficile‐inoculated mice versus controls showed similarly enriched pathways (supplementary material, Figure S3A–D). For EEC, goblet, immune, progenitor, stem, and tuft cell types, there were no significantly enriched pathways (FDR ≤0.05) in C. difficile‐inoculated mice. Despite little change in Dmbt1 between C. difficile‐inoculated mice versus controls among the goblet cell subtypes, we wanted to understand the potential contribution of goblet cell types to early tumorigenic events. We compared Goblet Clca1‐Hi to Goblet Agr2‐Hi cells and identified similar immune response pathways, but also cellular senescence and viral carcinogenesis (supplementary material, Figure S3E–G). Clca1‐high goblet cells have significantly elevated levels of canonical goblet cell genes (Mxd1, Zg16, and Tff3) compared with Agr2‐high goblet cells (supplementary material, Figure S3H and supplementary material, Table S6) [50]. The high expression of Duox2 in the Clca1‐high goblet cells suggests their possible role as sentinel goblet cells [51].

DMBT1 was significantly downregulated in colonic dysplasia

The earliest observation of dysplastic foci in our prior studies led us to the specific timepoint of 2‐weeks postinoculation for this work [13]. Dysplastic cells were not identifiable from scRNA‐seq alone, because they were rare compared to the stem and proliferative cells. To identify transcriptional changes in C. difficile‐associated dysplasia, we performed spatial transcriptomics, with particular attention to dysplastic foci and normal adjacent tissue (supplementary material, Figure S4A,B). For this experiment, we repeated the mouse experiment in Figure 1 and compared C. difficile‐inoculated mice versus controls. Only mice who received the 29‐bacteria consortium plus C. difficile developed dysplastic foci, and the incidence of dysplasia was the same as we reported previously [13]. The epithelia and lamina propria were collected as separate segments within each region of interest (supplementary material, Figure S4A–C). Dimensionality reduction was performed on all segments and displayed by UMAP plots (Figure 3A,B). These plots further demonstrated the similarities between segments in an unbiased manner (Figure 3A,B and supplementary material, Table S7). Gene expression analysis comparing dysplasia to normal tissue revealed upregulation of tumor‐associated genes in lamina propria: Cxcl14, Igf1, and Mmp13 (Figure 3C and supplementary material, Table S8) [52, 53, 54]. Increased WNT signaling was evident by Nkd1 and Axin2 upregulation in dysplastic foci compared to normal epithelial tissue (Figure 3D). Dmbt1 was one of the most differentially expressed genes, but was downregulated in dysplastic foci compared with normal colonocytes (Figure 3D), which surprised us because of its upregulation in scRNA‐seq (Figure 2). Differential expression between C. difficile‐inoculated versus control mice showed upregulation of inflammatory genes, like Lcn2 from lamina propria segments (Figure 3E and supplementary material, Table S8), and reactive‐oxygen species‐producing genes, like Nos2 from epithelia (Figure 3F). The spatial transcriptomics data supported our scRNA‐seq data, but included a set of genes specifically dysregulated in dysplastic foci.

Figure 3.

Spatial transcriptomic analysis of C. difficile‐inoculated mice versus controls. (A and B) Dimensionality reduction with UMAP demonstrated segregation based on gene expression for all segments. Each point represents an individual segment, and the shape of each point represents the tissue in which the segment was chosen. The color of the point signifies the mouse of origin (A) or the inoculation received (B). (C and D) Volcano plots of differential gene expression between normal and dysplastic tissue for (C) lamina propria and (D) epithelium; top genes with high absolute log₂ fold‐changes are labeled. (E and F) Volcano plots of differential gene expression compared segments from (E) lamina propria and (F) epithelium of control mice versus C. difficile‐colonized mice.

Novel integration of spatial transcriptomics and scRNA‐seq revealed potential pathways to C. difficile‐associated dysplasia

To integrate our two transcriptomic datasets, we leveraged their individual strengths: the ability to identify dysplastic foci with spatial context versus higher resolution expression data in the scRNA‐seq dataset. We began with significantly upregulated genes (log₂ fold‐change >1, adj. p < 0.05) in dysplastic epithelia versus normal from spatial transcriptomics (Figure 3D and supplementary material, Figure S5A). We derived a dysplasia score from this gene set and applied it to the scRNA‐seq data and observed the highest scoring cells in the stem cell cluster (supplementary material, Figure 5B,C). We refined the list to include only 12 genes with increased expression in the stem cell cluster: Sox4, Axin2, Ifitm3, Hes6, Prox1, Tubb5, Phlda1, Jun, Smoc2, Slc12a2, Ascl2, and Hopx. This revised dysplasia score enhanced the signal‐noise ratio and was highest in cells from C. difficile‐inoculated mice (Figure 4A,B). Interestingly, enteroendocrine and tuft cells also had high dysplasia scores (supplementary material, Table S9).

Figure 5.

Dmbt1 expression is downregulated in multiple mouse models of colonic tumorigenesis. (A) Volcano plot showing Dmbt1 among the most differentially regulated genes expressed by differentiated colonocytes from mice given protumorigenic slurry (magenta) versus nontumorigenic slurry (blue). Genes with the highest absolute log₂ fold‐changes are labeled (supplementary material, Table S11). (B) Dmbt1 normalized gene expression from spatial transcriptomics demonstrated upregulation in epithelial cells from C. difficile‐inoculated mice but downregulation in dysplastic foci. Points are segments with bar and whiskers as mean and standard error of the mean. Asterisks (*) indicate a significant difference (p < 0.05 from Wilcoxon rank sum test with continuity correction) (supplementary material, Table S11). (C) Immunofluorescence microscopy shows Dmbt1 protein expression is reduced in dysplastic foci, as indicated by altered β‐catenin distribution (decreased membranous, increased cytoplasmic and nuclear localization) compared to adjacent normal tissue in three different mouse models of colonic neoplasia. Dmbt1 was reduced in 100% of dysplastic foci (n = 57) from 11 mice. (D) Organoid formation assay using Lrig1 ^CreER/+ ;Apc ^fl/fl organoids treated with 1 μm 4‐hydroxytamoxifen (4OH‐Tam) versus vehicle control at 7 days posttreatment, shown using phase contrast microscopy. (E) Cell viability based on luminescence (RLU = relative light units). (F) RT‐qPCR analysis of Dmbt1 expression in 4OH‐Tam‐treated Lrig1 ^CreER/+ ;Apc ^fl/fl organoids compared to vehicle control where relative quantification (RQ) is 2^−ΔΔCt, and Ct is the cycle threshold. Expression is quantified relative to the reference transcript Actb. These plots represent averages from 43 independent wells of organoids from two different animals. (G) Immunoblotting analysis of vehicle control versus 4OH‐Tam‐treated Lrig1 ^CreER/+ ;Apc ^fl/fl organoids.

Figure 4.

Integration of spatial and single‐cell transcriptomics revealed potential pathways to C. difficile‐associated dysplasia. (A) UMAP plot of scRNA‐seq dataset representing the 12‐gene dysplasia score (a.u. = arbitrary units) generated from the spatial transcriptomics dataset. (B) Violin plot showing dysplasia score distribution for each inoculation and cell type. (C) Scatterplot and Spearman correlation between scaled Dmbt1 gene expression and the dysplasia score for all epithelial cells from scRNA‐seq data. Shading around the trendlines indicates 95% confidence intervals. C. difficile Spearman = −0.3415, p‐value = 1.958e‐113; Control Spearman = −0.3003, p‐value = 3.155e‐90. (D) Heatmap showing scRNA‐seq expression for the 12 genes used to create the dysplasia score. Gastric precancer genes are shown with relatively no expression. (E and F) Heatmaps displaying the total number of significant ligand‐receptor interactions inferred by CellPhoneDB. (G) Venn diagrams illustrating the intersection of C. difficile‐specific interaction partners from scRNA‐seq and spatial transcriptomics datasets. (H) Chord diagrams illustrating the predicted interactions with Sema7a by cell type in C. difficile‐colonized mice and control mice. No significant Sema7a interactions were predicted in control mice.

In the scRNA‐seq dataset, we found higher dysplasia scores correlated with lower Dmbt1 expression (Figure 4C). The differential expression of these 12 genes is shown in Figure 4D with Sox4, Ifitm3, and Smoc2 among the most upregulated genes in stem cells from C. difficile‐inoculated mice. These genes are implicated in oncogenic signaling in colorectal cancer [55, 56, 57]. DMBT1 is a marker for gastric precancerous lesions and is upregulated during spasmolytic polypeptide‐expressing metaplasia (SPEM) [19, 58]. However, we did not observe expression of markers for SPEM and gastric dysplasia: Tacstd2, Aqp5, and Ceacam5 (Figure 4D). Dmbt1 appears to be upregulated by C. difficile but downregulated in dysplasia, representing an interesting confluence of our transcriptomic datasets.

To determine if inferred ligand‐receptor pairs from scRNA‐seq data could be validated with spatial information to identify novel mechanisms with higher confidence, we leveraged the high‐resolution scRNA‐seq data. Using CellPhoneDB, we compared the sum of interactions between C. difficile‐inoculated mice versus controls (Figure 4E,F) [39]. From these interactions, we identified 84 genes unique to the C. difficile‐inoculated mice that were not significant in controls (Figure 4G red circle and supplementary material, Table S10). For validation, we searched for these interaction partners in our spatial transcriptomics epithelial segments (supplementary material, Table S8). We identified Semaphorin 7a (Sema7a) and Tumor Necrosis Factor Receptor Superfamily, member 1b (Tnfrsf1b) as significant genes in C. difficile‐inoculated mice from both scRNA‐seq inferred ligand‐receptor pairs and spatial datasets (Figure 4G blue circle). The only interacting partners for Sema7a predicted from CellPhoneDB were integrin α1β1 and plexin C1 (Plxnc1). These interactions were present in C. difficile‐inoculated mice predictions but not controls (Figure 4H). This validation approach suggested Sema7a is important for the host response to C. difficile during tumorigenesis. Although we did not find significant differential expression of Sema7a‐binding partners Itgb1, Itga1, or Plxnc1 in the spatial transcriptomics dataset, our integration method has wide applicability for future work. The interaction of Tnfrsf1b with its ligand Tumor Necrosis Factor alpha is likely to involve the cellular immune response, and our ongoing studies investigate mechanistic contributions of the immune compartment.

DMBT1 expression was downregulated in multiple mouse models of colonic tumorigenesis

To understand and validate the dysplasia‐specific Dmbt1 downregulation phenotype, we performed post hoc analysis of scRNA‐seq data from our previously published work using protumorigenic slurry‐inoculated mice versus nontumorigenic slurry‐inoculated mice [13]. The protumorigenic slurry contained C. difficile and led to significantly increased Dmbt1 gene expression compared with nontumorigenic slurry (not containing C. difficile) in differentiated colonocytes (Figure 5A and supplementary material, Table S11). This upregulation of Dmbt1 from our C. difficile biofilm slurry model supports our findings from the 29‐bacteria consortium with C. difficile (Figure 2A). Looking at the spatial transcriptomics data in the consortium model to examine Dmbt1 expression, we found that dysplastic segments from C. difficile‐inoculated mice expressed less Dmbt1 than normal tissue segments from either group (Figure 5B). Interestingly, the expression of Dmbt1 was higher in normal segments from C. difficile‐inoculated mice compared with normal segments from control mice (Figure 5B and supplementary material, Figure S6A). Dmbt1 upregulation in nondysplastic tissue from C. difficile‐inoculated mice compared with controls is consistent with other gastrointestinal infections [23, 58].

We next investigated Dmbt1 protein expression using immunofluorescence in two additional mouse models of colonic tumorigenesis: the azoxymethane/dextran sodium sulfate‐induced colitis‐associated cancer model and transgenic Lrig1 ^CreER/+ ;Apc ^fl/+ mice [36, 37]. In each model, Dmbt1 was markedly reduced and inversely correlated with β‐catenin, used here to mark dysplasia with increased WNT signaling (Figure 5C). These data illustrate the downregulation of Dmbt1 as a common phenotype in murine colonic dysplasia, suggesting that reduced Dmbt1 expression may be a generalizable characteristic of colonic dysplasia, not unique to C. difficile‐associated tumorigenesis.

To ascertain the mechanism of Dmbt1 downregulation in dysplastic foci, we derived colonic organoids from Lrig1 ^CreER/+ ;Apc ^fl/fl mice. Apc loss of heterozygosity occurs concomitantly with dysplasia in Lrig1 ^CreER/+ ;Apc ^fl/+ mouse adenomas [37], so we hypothesized that biallelic truncation of Apc and subsequent WNT signaling activation would lead to Dmbt1 downregulation. We induced Apc recombination in colonic organoids using 4‐hydroxytamoxifen (4OH‐Tam) and showed increased organoid formation and viability compared to untreated controls (Figure 5D,E). Next, we showed 4OH‐Tam lowered both Dmbt1 expression using RT‐qPCR and Dmbt1 protein using immunoblotting (Figure 5F,G).

DMBT1 protein expression is lost in human colonic dysplasia

To determine if DMBT1 loss occurs in human colonic tumorigenesis, we examined adenomas from a published multiomic atlas of precancerous colon polyps [49]. These data show DMBT1 gene expression is upregulated in differentiated crypt‐top cells and downregulated in adenoma‐specific cells (Figure 6A–C and supplementary material, Table S12). While some of the adenoma‐specific cells have increased DMBT1 expression in cells from tubular adenomas (TA), the tubulo‐villous adenomas (TVA) with high‐grade dysplasia had the lowest expression of DMBT1 (supplementary material, Figure S6B–D and Table S12). When comparing the expression of DMBT1 in crypt‐top (CT) versus adenoma‐specific cells to canonical WNT signaling target genes (Figure 6C), we observed an inverse correlation similar to the mouse colonic dysplastic foci with increased β‐catenin staining (Figure 5C).

Figure 6.

DMBT1 expression was decreased in human colonic dysplasia. (A) UMAP plot of scRNA‐seq data from human colonic precancerous polyps with coloring indicating cell types, EEC = enteroendocrine. (B) Violin plot showing DMBT1 scaled expression (a.u. = arbitrary units) in different cell types. Asterisks (*) indicate significant difference from all other cell types (p < 0.05 Kruskal–Wallis with Dunn's test, supplementary material, Table S12). (C) Heatmap of DMBT1 mean expression alongside canonical WNT signaling target genes indicated an inverse relationship between WNT activity and DMBT1 expression. (D) DMBT1 immunostaining in well‐differentiated, low‐grade dysplastic crypts (arrowhead) adjacent to high‐grade dysplasia (arrow) within a human colon adenocarcinoma. (E) Quantification of DMBT1 immunofluorescence staining in a human tissue microarray with specimens spanning the progression from normal adjacent tissue to metastatic colorectal cancer. The average fluorescence (raw integrated density) for 3 or 4 tissue cores per patient sample is shown as a single point; n = 3–15 patients/group (*p < 0.05, **p < 0.01 from Kruskal–Wallis with Dunn's test). Blue lines are means, and whiskers are standard error of the mean. (F) RT‐qPCR analysis performed on normal human colonic organoids. Each point represents the mean of three technical replicates from seven independent wells pooled together. NKD1 and AXIN2 are canonical WNT signaling target genes. Relative quantification is equivalent to 2^−ΔΔCt, and Ct is the cycle threshold. Expression is quantified relative to the reference transcript ACTB, normalized to vehicle control, and scaled to min–max of 0 to 1 (*p < 0.05 from Wilcoxon Rank Sum Test). CHIR = CHIR99021.

To investigate DMBT1 protein expression, we performed immunofluorescence microscopy on a human tissue microarray containing diverticulosis, ulcerative colitis, normal‐adjacent to adenocarcinomas, tubular adenomas, tubulo‐villous adenomas, adenocarcinoma, and metastatic carcinoma (Figure 6D,E and supplementary material, Table S12). These data revealed that average DMBT1 protein levels were highest in tubular and tubulo‐villous adenomas, but lowest in metastatic adenocarcinomas (Figure 6E and supplementary material, Figure S7). To determine whether WNT signaling downregulated DMBT1, we added CHIR99021 to normal human colonic organoids (Figure 6F). CHIR99021 is a specific inhibitor of GSK3β and potent activator of WNT target genes, such as NKD1 and AXIN2. The results showed CHIR99021‐mediated activation of WNT is associated with decreased DMBT1. The downregulation of DMBT1 in human colonic dysplasia appears to be regulated, at least in part, by WNT signaling in mouse and human colon.

Discussion

Comprehensive profiling of unique tumor models has elucidated generalizable phenotypes and valuable insights into CRC biology [49, 59]. This approach is especially revealing in the context of bacteria‐associated tumorigenesis [10, 60, 61]. Building on our discovery that C. difficile exacerbates colonic adenoma formation in Apc ^Min/+ mice, we integrated scRNA‐seq and spatial transcriptomics to understand the molecular underpinnings of this phenotype. We reveal how the intersection of single‐cell and spatial datasets identified the dysplasia‐specific downregulation of DMBT1 in colorectal precancerous adenomas and adenocarcinomas in mouse models and, importantly, human tissues.

Meticulous mapping of cell types and states in the gastrointestinal tract has uncovered fundamental insights into inflammation and cancer [62, 63, 64, 65, 66, 67, 68]. We observe the unique Abs3 cluster representing a subset of Krt20‐positive absorptive colonocytes from C. difficile‐inoculated mice with high expression of Prom1 and Anxa1 (Figure 1J,K). We interpret these cells to be differentiated and reactive while coincidentally expressing markers of dysplastic and predysplastic colonocytes. We initially thought this Abs3 cluster might represent the origin of sessile serrated lesions that develop through a gastric metaplasia state [49]. However, we did not see a significant overlap with that gene signature (supplementary material, Figure S2C). To be clear, the gastric metaplasia state in colonic precancerous sessile serrated lesions is different from the intestinal metaplasia, or SPEM, occurring in the stomach [69]. Our mouse model hinges on the germline mutation of one Apc allele, which already initiates increased WNT signaling. This baseline WNT activation favors the development of conventional adenoma cells over sessile serrated cells during dysplastic transformation [49, 70].

DMBT1 functions in immune exclusion of pathogens and regulation of the cell proliferation‐differentiation axis. We observed it specifically downregulated in colonic dysplasia, suggesting the loss of DMBT1 could be a link between infection, inflammation, and tumorigenesis [14]. DMBT1 is transcriptionally regulated via NF‐κB signaling downstream of Toll‐Like Receptor 4 in infection and STAT3 signaling in Crohn's colitis [23, 71]. Other genes, such as BVES and SATB2, are downregulated in colitis‐associated dysplasia and recently have been reported as potential biomarkers [72, 73]. Although loss of DMBT1 in other cancers is caused by deletion at the 10q25.3–26.1 locus [74], DMBT1 is deleted in only 5%–17% of CRC (Genomic Data Commons, National Cancer Institute, https://portal.gdc.cancer.gov/genes/ENSG00000187908, accessed 29 November 2023). DMBT1 has not been shown to function as a traditional tumor suppressor via DNA repair, interference with secondary messenger signaling, or modification of oncogenes/oncoproteins. Studies with Dmbt1 ^−/− mice have led to mixed results with respect to colonic inflammation [75, 76], but Dmbt1‐deficient mice have not been used in conventional colonic adenoma/cancer models. DMBT1 slows proliferation of mouse embryonic stem cells and initiates their differentiation into single‐layered epithelia [77]. In renal tubules, DMBT1 promotes the columnarization and terminal differentiation of intercalated cells [78]. DMBT1 overexpression reduces the onset of ovarian cancer and decreases metastasis of gallbladder cancer through promoting terminal differentiation [79, 80]. These data suggest DMBT1 may inhibit tumorigenesis by driving differentiation. In human CRC, loss of DMBT1 may be associated with decreased cancer‐related survival [24, 25]. The increased DMBT1 in adenomas compared with controls (Figure 6) introduces the possibility that DMBT1 is indirectly regulated by WNT target genes. Our data identify decreased Dmbt1 in dysplastic crypts of three different mouse models of tumorigenesis. Therefore, we suspect DMBT1 downregulation might provide a selective advantage for dysplastic transformation or progression to adenocarcinoma. Downregulation of DMBT1 might support colonic tumorigenesis by allowing dysplastic cells to maintain a more undifferentiated state, retaining their proliferative capacity. It is also possible that loss of DMBT1 is a bystander effect of other mechanisms driving tumorigenesis. Future studies are needed to determine how this mucin‐like glycoprotein might function to limit CRC tumorigenesis.

Author contributions statement

EHG, CNH, DBL, MJS, KSL, CLS, RJC, JLD and NOM contributed to project conceptualization. EHG, SRK, MER, HML, CNH, AJS, XL, DBL, JLD and NOM contributed to methodology and experimental design. EHG, SRK, MER, HML, HK, CNH, HD, AJS, XL, JLD and NOM performed experiments and data collection. EHG, SRK, MER, AJS, MKW, QL, JLD and NOM performed data validation. EHG, SRK, MER, HK, CNH, JLD and NOM contributed to data visualization. EHG, SRK, MER, HK, CNH, AJS, XL, QL, KSL, CLS, RJC, JLD and NOM performed data analysis and interpretation of results. MJS, QL, KSL, CLS, RJC, JLD and NOM acquired funding. DBL, MJS, QL, KSL, CLS, RJC, JLD and NOM provided supervision and coordination of this work. EHG, JLD, and NOM drafted the article. EHG, DBL, MJS, KSL, CLS, RJC, JLD and NOM contributed to article editing. All authors approved the final version of the article.

Supporting information

Figure S1. Additional analysis for the scRNA‐seq dataset illustrated unsupervised clustering, CytoTRACE scores, and merged clusters by similar type

Figure S2. Cell type assignments, correlations, and metaplastic scoring for scRNA‐seq dataset

Figure S3. Gene set enrichment and differential gene expression analysis for absorptive precursors, deep crypt secretory cells, and goblet cell subtypes

Figure S4. Spatial transcriptomics of C. difficile‐colonized Apc ^Min/+ mice using the GeoMx Digital Spatial Profiler (DSP)

Figure S5. Dysplasia score genes from spatial transcriptomics

Figure S6. Additional transcriptomics analysis

Figure S7. Immunofluorescence and H&E microscopy images from the human tissue microarray

Table S1. Differential gene expression for scRNA‐seq Leiden clusters as cell types

Table S2. Mouse cell cycle genes

Table S3. Statistics for DMBT1 expression by cell type

Table S4. Differential gene expression for scRNA‐seq comparing C. difficile to control

Table S5. Gene set enrichment analysis for absorptive cells

Table S6. Differential gene expression analysis for Clca1 versus Agr2 goblet cells

Table S7. Source data and metadata for UMAP plots of spatial transcriptomics

Table S8. Differential gene expression analysis for spatial transcriptomics

Table S9. Dysplasia score statistics by cell type

Table S10. Ligand‐receptor pair inference source data

Table S11. Differential gene expression analysis and Dmbt1 spatial transcriptomics expression

Table S12. DMBT1 human scRNA‐seq statistics

Data availability statement

All data supporting the findings of this study are publicly available. These data are included in the published article, supplementary information files, and/or can be obtained at NCBI Gene Expression Omnibus (GEO) database (https://ncbi.nlm.nih.gov/geo/) under Accession Numbers GSE270726 and GSE270745.