Integrated scRNA-seq profiles of left-sided and right-sided colonic adenomas

To determine the single-cell profiles of colonic polyps in connection to their sidedness, we first compiled two single-cell transcriptomic datasets with 78 polyp samples from 44 patients (figure 1A), of which 46 samples were from right-sided colon (caecum, ascending colon and hepatic flexure) and the rest from left-sided colon (descending colon, sigmoid colon) (figure 1B). After quality control, a total of 75 559 cells were included, in which 30 274 cells and 45 285 cells were originated from left-sided and right-sided polyps, respectively. Highly variably expressed genes within these cells were used for principal component analysis (PCA) (online supplemental figure S1A). To mitigate batch effects, the scRNA-seq data were integrated by Harmony algorithm.9 Principal components corrected by Harmony were employed to generate the unified Uniform Manifold Approximation and Projection (UMAP) embedding space and perform graph-based clustering (online supplemental figure S1B). Cells were classified into three major cell types, including epithelial cells (EPCAM⁺),10 immune cells (PTPRC+)11 and mesenchymal stromal cells (COL1A1⁺)12 (online supplemental figure S1C). All major cell types originated from two datasets merged together without obvious bias (online supplemental figure S1D). These results indicated that integrated scRNA-seq datasets were suitable for further analysis.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Single-cell analysis reveals distinct cellular profiles in colonic conventional ADs based on location. (A) Graphic overview of two published single-sequencing datasets included in our study (created with BioRender.com). The numbers (n) of patients in each dataset and samples in colon segment are provided in the figure. (B) Bar graphs showing summary statistics for the number of cells, patients and samples subjects collected by colon segment (left) and their dataset and histology compositions (right). (C) Bar graphs showing the sidedness composition in ADs and SPs, respectively. (D) UMAP view about the distribution of epithelial cells in left-sided and right-sided ADs (left). Bar graphs displaying the sidedness composition of epithelial cells from ADs in each dataset (right). (E) UMAP view of nine epithelial cell clusters with annotation information (left). Dot plot visualisation of marker gene expression across the defined clusters (right). (F) Bar graphs showing the distribution of annotated cell clusters within left-sided and right-sided ADs. (G) Line chart displaying the sidedness prevalence for each cell type estimated by Ro/e score (top). Bar plot showing the relative proportion of cells from left and right side for each epithelial subset (bottom). (H) Milo analysis demonstrating the enrichment of each cell type in left-sided and right-sided ADs. ADs: Conventional adenomas; HS: HTAN Stanford dataset; HV: HTAN Vanderbilt dataset; SPs: Serrated polyps; UAMP: Uniform Manifold Approximation and Projection.

Colonic polyps were histologically categorised into conventional ADs and SPs. The distribution of epithelial cells was almost even across left-sided and right-sided ADs (47% vs 53%), while epithelial cells of SPs exhibited the enrichment in the right side (left vs right, 21% vs 79%) in line with previous research (figure 1C and online supplemental figure S1E).5 To decipher the alteration in polyps caused by colonic location only, SPs were excluded from further investigation. The remaining 55 792 epithelial cells from ADs were reclustered (figure 1D). Identified clusters were annotated to colonic cell types based on canonical marker genes, including mature colonocytes (SLC26A2, CA1, GUCA2A), mature goblet cells (mGCs) (MUC2, TFF3, SYTL2), stem cells (OLFM4, LGR5, ASCL2), transit amplifying cells (MKI67, TOP2A, PCNA), BEST4⁺ colonocytes (BEST4, CA7, OTOP2), enteroendocrine cells (CHGA, SCGN, MS4A8), tuft cells (PLCG2, LRMP, SH2D6) (figure 1E and online supplemental figure S1F). The composition of cell types was relatively homogenously distributed within left-sided and right-sided ADs. Of note, the right-sided ADs exhibited decreased mGCs, BEST4⁺ colonocytes and enteroendocrine cells, but increased proportions of stem cells and tuft cells (figure 1F). Altered cellular composition between left-sided and right-sided ADs was further validated by Ro/e analysis and Milo (figure 1G and H).13 14

Mature goblet cells of right-sided ADs are depleted and dysfunctional in mucin biosynthesis and antigen presentation

As mGCs showed the greatest variation according to ADs sidedness, goblet cell-associated fractions were subjected to clustering and UMAP visualisation (figure 2A). mGCs in right-sided ADs expressed a lower level of MUC2 (p<0.001) and reduced proportionally within the goblet cells cluster compared with that in left-sided ADs (p=0.007) (figure 2B and C). Since a fundamental function of mGCs is to synthesise and secrete mucin, mucin gene expression was examined in goblet cells of ADs. In right-sided ADs, both gel-forming mucins (MUC2, MUC5AC and MUC6) and membrane-associated mucins (MUC4, MUC12, MUC13 and MUC17) expression decreased, suggesting the deficiency of mucin synthesis in right-sided ADs (figure 2D). Apart from gene transcription, mucin biosynthesis involves glycosylation in the Golgi apparatus, oligomerisation and dense packing for storage in large granules before secretion.15 mGCs from right-sided ADs were defective in disulfide isomerase, glycosylation and secretion pathways (figure 2E), as evidenced by decreased expression of genes involved in protein transportation (ARF1 and TMED9) and glycosylation (GALNT6 and B3GNT7) compared with left-sided ADs (all p<0.001) (figure 2F). Sialylation of glycans on mucus is essential for its stability and integrity by protecting against bacterial degradation.16 Sialyltransferases expression in mGCs of right-sided ADs was largely reduced (ST6GALNAC1, ST6GALNAC6 and ST3GAL4) (all p<0.001) (figure 2G). Together, these results suggested that mucus as the intestinal barrier might be defective in right-sides ADs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

mGCs in right-sided ADs are reduced in number and impaired in mucin production and antigen presentation. (A) UMAP view of the total goblet cells. (B) MUC2 expression level displayed in the UMAP plot (left). Colours represent the log-normalised data. UMAP plot showing the distribution of cells from left-sided and right-sided ADs (middle). Expression level of MUC2 in iGC and mGC from left-sided and right-sided ADs (right) presented by violin plot. (C) Box plot showing the fraction of mGCs in left-sided (n=21) and right-sided ADs (n=25). Symbols represent ratio value of individual samples. (D) Dot plot displaying the mucin expression in in iGC and mGC from left-sided and right-sided ADs. (E) Heatmap showing the pathway activity involved in mucin glycosylation, transportation and secretion calculated by GSVA in iGC and GC from left-sided and right-sided ADs. (F) Expression level of GALNT6, B3GNT7, ARF1 and TMED9 in iGC and GC from left-sided and right-sided ADs presented by violin plot. (G) Expression level of gene involved in sialylation (ST6GALNAC1, ST6GALNAC6, ST3GAL4 and NANS) in iGC and GC from left-sided and right-sided ADs. (H) Dot plot displaying the major histocompatibility complex-related genes expression in iGC and mGC from left-sided and right-sided ADs. (I) Heatmap displaying the pathway activity involved in antigen processing and presentation calculated by GSVA in iGC and GC from left-sided and right-sided ADs. (J) Heatmap demonstrating the regulons enrichment in iGC and GC from left-sided and right-sided ADs via SCENIC analysis. (K) The developmental trajectory of the iGCs, goblet cells and stem cells, colour coded by the subpopulations (left) and location (right). (L) Representative images of anti-MUC2 (top) and anti-TFF3 (bottom) antibody-stained sections of left-sided (n=13) and right-sided ADs (n=11) (left). Violin plot further showed the MUC2 and TFF3 staining score. Scale bar=100 µm. (M) Representative images of AB/PAS-stained sections of left-sided (n=13) and right-sided ADs (n=11) (left). Bar graphs further showed the staining area and mucus thickness (right). Scale bar=100 µm. Box plots (C and M) showed the median (centre line), the upper and lower quantiles (box) and the range of the data (whiskers). All p values (B, C, F, G, L and M) were determined using an unpaired two-sided Wilcoxon rank-sum test. ADs: Conventional adenomas; GSVA: Gene Set Variation Analysis; iGC: immature goblet cells; mGC: mature goblet cells; SCENIC: single-cell regulatory network inference and clustering; UMAP: Uniform Manifold Approximation and Projection.

Aside from mucin biosynthesis, goblet cells could form goblet cell-associated antigen passages and transmit luminal substances to antigen-presenting cells to induce adaptive immune responses.17 To evaluate the antigen presentation capacity of goblet cell in ADs, genes relative to major histocompatibility complex (MHC) were determined. Intriguingly, there was a marked decrease in MHC-associated gene expression (B2M, HLA.A, HLA.B and HLA.C) in mGCs of right-sided ADs (figure 2H), together with attenuated antigen folding, assembly and presentation pathways (figure 2I).

Next, we determined the possible involvement of regulons in goblet cells in ADs using single-cell regulatory network inference and clustering analysis. Transcription factors ATOH1 and SPDEF governing goblet cell development showed higher regulator activity in mGCs of both left-sided and right-sided ADs without discernible difference (figure 2J). Notably, the regulator activity of KLF4 involved in terminal differentiation of goblet cells were reduced in mGCs of right-sided ADs (figure 2J).18 Instead, HNF4A and HNF4G activities were enhanced in mGCs with dysfunctional status in right-sided ADs (figure 2J). To some extent, this result was in line with previous findings that loss of HNF4A/4G with SMAD4 disruption promotes goblet cell identities.19 Considering the altered function status and regulon activity, we pooled the goblet cells with stem cells from left-sided and right-sided ADs together to explore the developmental process via pseudotime analysis. This analysis showed that the stem cells were at the beginning of the trajectory path, whereas mGCs were at a terminal state along the differentiation trajectory from progenitor to mature cell fates (figure 2K, left). In particular, mGCs from left-sided and right-sided ADs positioned at the distinct ends of the developmental trajectory, implying dissimilar differentiation process (figure 2K, right). Furthermore, we validated our findings in clinical ADs samples via immunohistochemical and Alcian blue (AB)/Periodic acid-Schiff (PAS) staining. Immunohistochemistry (IHC) confirmed diminished expression of goblet cell markers, namely MUC2 and TFF3 (p=0.008 and p=0.004), in right-sided ADs as compared with their left-sided counterparts (figure 2L). The sections of right-sided ADs showed smaller mucus vacuoles (p=0.008) and thinner inner mucus layer (p<0.001), as assessed by AB/PAS staining on the Carnoy’s fixed sample (figure 2M). Taken together, our results indicated the loss of goblet cells along with defective mucin biosynthesis and antigen presentation in right-sided ADs.

Biofilm formation and bacterial invasion in right-sided ADs lead to epithelial inflammation response

Given that mucus production by goblet cells hampering bacterial translocation and avoiding inflammatory pathologies, we hypothesised that defective mGCs might instigate inflammation in right-sided ADs. Supporting this notion, goblet cells proportion negatively correlated with inflammatory response caused by lipopolysaccharide (LPS) in colonic ADs (R=−0.3, p=0.04) (figure 3A). In line with this, most cell types were enriched in inflammatory response caused by LPS pathway in right-sided ADs (p<0.001) (figure 3B). These cells also underwent epithelial cell apoptotic process (p<0.001) (online supplemental figure S2A). Epithelial cell apoptotic process was positively associated with inflammatory response caused by LPS among ADs samples (R=0.59, p<0.001) and across different cell types (R=0.88, p=0.002) (online supplemental figure S2B and S2C). This implied that apoptosis in epithelial cells of ADs might be a consequence of inflammatory response mediated by bacterial translocation. To spatially visualise the bacteria-induced inflammation within ADs tissues, we performed spatial transcriptomics using the tissue section from five ADs samples, including two left-sided and three right-sided ADs samples. Single-sample Gene Set Enrichment Analysis (ssGSEA) analysis was employed to assess the enrichment of LPS-induced inflammatory response pathway within spots. Coloured spots based on enrichment score revealed increased LPS-induced inflammation predominantly in the interior of the right-sided ADs section with overall significantly higher enrichment (p<0.001) (figure 3C), suggesting that bacteria and their toxins that cross the intestinal barrier may instigate inflammation in right-sided ADs. Thus, we speculated that there was a higher bacterial burden within these tissues. Quantitative PCR (qPCR) targeting universal 16s ribosomal DNA (rDNA) indicated an increased abundance of bacteria in right-sided ADs (p=0.025) (figure 3D). Intriguingly, IHC staining of bacteria component LPS showed the bacterial aggregation as penetrating and occupying into the crypts of ADs (figure 3E). It appeared as dense communities of bacteria encased in a matrix in contact with epithelial cells, representing typical biofilm morphology. Biofilm presence was more prevalent among right-sided ADs compared with the left-sided counterpart (p=0.039) (figure 3E, top). The existence of biofilm in crypts correlated with increased bacterial tissue invasion in right-sided ADs (p=0.002) (figure 3E, bottom). Consistently, the spots around the crypts with biofilm exhibited high enrichment of LPS-induced inflammation and apoptosis pathways (p=0.035 and p=0.032). The decreased trend of the pathway enrichment was observed in the spots aways from the crypts (figure 3F). These results indicated that the cells spatially surrounding the crypts with biofilm occupation underwent the bacteria-induced inflammation in right-sided ADs. To further determine the alteration in the composition and relative abundance of gut microbiota, we sequenced the 16s rDNA V3–V4 regions from the tissues of left-sided and right-sided AD samples. PCA revealed that the bacterial composition in right-sided ADs was separated from left-sided counterpart (p=0.003) (figure 3G). At phylum levels, Proteobacteria displayed relative high abundance exclusively in right-sided AD samples, while Fusobacteria were rare (figure 3H). This was in line with previous research, suggesting the predominant presence of Proteobacteria and non-detectable Fusobacteria in biofilms of Familial adenomatous polyposis samples.7 Together, biofilm occupation in the crypts facilitated the bacterial translocation, leading to inflammatory response in right-sided ADs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Biofilm formation and bacterial invasion in right-sided ADs trigger inflammatory responses in epithelial cells. (A) Correlation plot presenting the inflammatory response by LPS score calculated by AddModuleScore with the proportion of goblet cell across the samples (n=46). 95% CI is indicated with grey colour. (R represents Pearson’s correlation and its coefficient of determination, p value is 0.04.) (B) Violin plot displaying the inflammatory response by LPS score calculated by AddModuleScore in individual cells from annotated cell type in left-sided and right-sided ADs. (C) Feature plot showed the inflammatory response by LPS score calculated by ssGSEA in the spatial transcriptomics. (D) Box plots illustrating the relative 16s level in left-sided (n=14) and right-sided ADs (n=12). (E) Representative images of immunocytochemistry staining against the LPS in the tissues sections of ADs. Bar graphs presenting the composition of biofilm formation in the left-sided and right-sided ADs with Fisher’s exact test for comparison (top). Box plots displaying LPS⁺ cell ratio in the left-sided and right-sided ADs. Box plots showed the median (centre line), the upper and lower quantiles (box) and the range of the data (whiskers). (F) Representative images of immunocytochemistry staining against the LPS (left). Feature plots and box plots showed the inflammatory response by LPS score calculated by ssGSEA in the indicated regions. (G) Principal component analysis revealed a significant shift in overall diversity of microbiota composition between left-sided and right-sided ADs. (H) Bar chart displaying the relative abundance of the top five bacteria at the phylum level in the individual samples of left-sided and right-sided ADs. The p values (B, D, E and F) were determined using an unpaired two-sided Wilcoxon rank-sum test. ADs, adenomas; LPS, lipopolysaccharide; ns, no significance.

Immune cell infiltration in ADs is modulated by bacteria in right-sided ADs

As immune cells are effectors of inflammation, we examined immune cell composition and functional status in ADs (figure 4A). Immune cells (PTPRC⁺) were increased in right-sided ADs (p=0.016) (figure 4B). We clustered immune cells into monocytes/macrophages (CD14, CSF1R, CSF3R and C1QB), T cells (CD3D, CD247, TRBC1 and TRAC), B cells (CD79A, CD19, PAX5 and IGHM) and mast cells (TPSB2, VWA5A, MS4A2 and CPA3) (figure 4C). Among them, T cells are predominant in ADs, demonstrating markedly increased in right-sided ADs compared with left-sided ADs (p=0.007) (figure 4D). In parallel, T cells in right-sided ADs were enriched in T cell extravasation signature, implying dependence on peripheral lymphocytes for replenishment (figure 4E). In right-sided ADs, T cells, mast cells and monocytes/macrophages all showed the enrichment in defence response to Gram-negative bacterium geneset, while B cells were enriched in antimicrobial humoral response and interleukin 1α production signature, in line with increased bacterial infiltration (figure 4E). Single-cell regulatory network inference and clustering analysis further revealed that inflammation-associated regulons were enriched in the immune cells of right-sided ADs (eg, RELA for T cells, STAT5A for mast cells, REL for B cells and IRF7 for monocytes/macrophages) (figure 4F). Collectively, these results indicated that immune cells were involved in inflammatory response against bacterial invasion in right-sided ADs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

The bacterial presence modulates the Immune-inflammatory response in right-sided ADs. (A) PTPRC expression level displayed in the UMAP plot (left). Colours represent the log-normalised data (right). (B) Bar graph (left), Milo analysis (middle) and box plots illustrating the immune cell infiltration in right-sided AD compared with left-sided counterpart. (C) UMAP view of the four clusters with annotation in immune cell cluster (left). Dot plot displaying the expression of defined markers for the cell types in immune cell cluster (right). (D) Box plot showing the fraction of immune cell subgroups in left-sided and right-sided ADs. (E) Heatmap illustrating the pathway activity involved in immune activation and response to bacteria calculated by GSVA in the immune cell subgroups from left-sided and right-sided ADs. (F) Heatmap showing the regulons enrichment in the immune cell subgroups from left-sided and right-sided ADs via SCENIC analysis. (G) UMAP plot of three subclusters (T cell-1, T cell-2 and T cell-3) in T cells (top). Bar chart displaying the composition of T cell subclusters (bottom). (H) The developmental trajectory of the T cell, colour coded by the subpopulations (left) and location (right). (I) Volcano plot showing the differential genes between T cell-1 and remaining cells (left). The expression level of LGALS3, LGALS4 and DEFA5 in the T cells from left-sided and right-sided ADs presented by Violin plot. (J) Increased LGALS3, LGALS4 and DEFA5 expression in human Jurkat T cells treated with LPS and PGN (1 and 10 μg/mL) for 24 hours (n=3). Data presented as the mean±SD. P values were determined using unpaired two-sided Wilcoxon rank-sum test (B and D) and Student’s t-test for two-group comparison (J). AD: Conventional adenomas; GSVA: Gene Set Variation Analysis; LPS: Llipopolysaccharides; ns: no significance; PGN: peptidoglycan; PTPRC: protein tyrosine phosphatase receptor type C; SCENIC: single-cell regulatory network inference and clustering; UMAP: Uniform Manifold Approximation and Projection.

As T cells were dominant component in ADs, these cells were reclustered and grouped into three subpopulations (T cell-1, T cell-2 and T cell-3) (figure 4G, upper). T cell-1 subset was elevated in right-sided ADs, accompanied by downregulated T cell-2 cluster (figure 4G, lower). We next explored the dynamic immune states and cell transitions in these cells by inferring the state trajectories using Monocle.20 T cell-2 and T cell-3 clusters were positioned at the beginning and the middle of the trajectory path, while T cell-1 cluster was located at terminal suggesting its differentiated status (figure 4H). Next, we compared differentially expressed genes (DEG) between T cell-1 cluster cells and remaining T cells. Intriguingly, LGALS3, LGALS4 and DEF5A gene expression were increased in T cells from T cell-1 cluster in right-sided ADs (figure 4I). LGALS3 and LGALS4 encode galectin-3 and galectin-4, respectively, which has been shown to interact with and agglutinate commensal and pathogenic bacteria.21 22 To validate the role of bacterial components on LGALS3 and LGALS4 expression in T cells, we performed bacterial LPS and peptidoglycan (PGN) treatment in Jurkat T cells. As shown in figure 4J, both LPS and PGN induced LGALS3 and LGALS4 expression in human Jurkat T cells in a dose-dependent manner (p<0.05) (figure 4J). These results suggested that bacterial existence in right-sided ADs might induce LGALS3 and LGALS4 expression in T cells exercising antibacterial activities.

S100A11 expression is induced by bacteria in right-sided ADs

Next, we focused on two clusters of epithelial cells lacking expression of canonical colonic epithelial cell markers in ADs (figure 5A and online supplemental figure S3A). These clusters are enriched in colon cancer signature compared with colonocytes and goblet cells (p<0.001), inferring they are neoplastic cells in ADs with malignant potential (online supplemental figure S3B). Thus, the cells in these two clusters were reclustered and visualised in UMAP plot. They were divided into five clusters that are uniformly distributed across the datasets and samples (figure 5B and online supplemental figure S3C). Intriguingly, subclusters 0 and 4 were mainly positioned in left-sided ADs and then annotated as Epi-L cluster. Subclusters 1, 2 and 3 were predominantly located in right-sided ADs and named as Epi-R cluster (figure 5B). Consistently, cells of Epi-L cluster and Epi-R cluster were enriched in left-sided and right-sided ADs, respectively, and confirmed by Milo analysis (figure 5C). To delineate the dissimilarity in Epi-L and Epi-R clusters, we performed DEG and pathway analysis (online supplemental figure S3D). Epi-R clusters are enriched in inflammation-related pathways and response to bacterium (figure 5D). Hallmark pathway analysis further showed the glycolysis and epithelial mesenchymal transition (EMT) pathway were enriched in Epi-R cluster (online supplemental figure S3E). Of note, the score of glycolysis and EMT pathway in Epi-R cluster were highest among other cell types in epithelial component of ADs (all p<0.001) (online supplemental figure S3F and S3G), further suggesting its increased malignant potential. Hence, the Epi-R cluster, predominantly distributed in right-sided ADs, was enriched in inflammation response and EMT process.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

Bacterial presence in right-sided ADs induces S100A11 expression. (A) UMAP plot showing the distribution of Epi-1 and Epi-2 in the total epithelial cells from ADs. (B) UMAP displaying the distribution of five subcluster (left), location (middle) and annotation (right) in the cells derived from the Epi-1 and Epi-2 clusters. (C) Bar graph (left) and Milo analysis (right) demonstrating the preferred distribution of Epi-L and Epi-R clusters in the left-sided and right-sided ADs, respectively. (D) Gene Ontology enrichment analysis showed increased pathway activity in inflammation and response to bacteria based on the upregulated genes in Epi-R. (E) Heatmap showing expression of the cluster-specific genes in Epi-L and Epi-R. (F) Intestinal adenoma organoids were constructed using Apc^Min/+ mice and further treated with LPS and PGN. (G and H) Increased S100A11 expression in adenoma organoids cells treated with LPS and PGN (1 and 10 μg/mL) (n=3) for 24 hours measured by reverse transcription-quantitative PCR (G) and for 48 hours detected by flow cytometry (H). (I) Diagram showing Apc^Min/+ mice treated with Escherichia coli with or without antibiotics. (J) Representative images of FISH with EUB338 probe and immunostaining against the S100A11 in sections of the colonic adenomas of Apc^Min/+ mice with or without EAEC and antibiotics treatment (left) and quantification of MFI for S100A11 staining (right) (n=4 for each group) (scale bar=50 µm). (K) Diagram displaying the Apc^Min/+ mice treated with DSS (top). The images of the harvested colon in these mice (n=5) (bottom). (L) Representative images of FISH with EUB338 probe and immunostaining against the S100A11 in sections of the colonic tumours of Apc^Min/+ mice treated with 2% DSS (scale bar=50 µm). (M) Bar graphs presenting the composition of biofilm formation in the tumours on the distal and proximal colon with Fisher’s exact test for comparison. (N) Box plot showing the quantification of MFI for S100A11 staining in the sections from DSS-treated Apc^Min/+ mice. (O) Box plot demonstrating the S100A11 staining MFI in the sections of proximal and distal colon tumours with or without biofilm existence. (P) Increased S100A11 expression level in the right-sided ADs (n=12) compared with left-sided counterpart (n=14) as well as left-sided (n=11) and right-sided colon (n=10) samples of our cohort. (P) Correlation plots showing relative 16s level with S100A11 expression in our in-house samples (blue: left-sided ADs, n=14; orange: right-sided ADs, n=12) (R represents Pearson’s correlation and its coefficient of determination). Data presented as the mean±SD (G, H, J, N, O and P). P values were determined using unpaired two-sided Student’s t-test for two-group comparison (G, H, J, N, O and P). ABX: antibiotics; DSS, dextran sulfate sodium; EAEC:enteroaggregative Escherichia coli; FISH: fluorescent in situ hybridisation; LPS: Llipopolysaccharides; MFI: mean fluorescence intensity; ns: no significance; PGN: peptidoglycan; UMAP: Uniform Manifold Approximation and Projection.

To further characterise the Epi-L and Epi-R clusters, we identified the distinct markers of these cells. The cells in Epi-R cluster specially expressed DUOX2, SORCS2, ANPEP, TMEM132C, S100A11 and ANXA2 (figure 5E). To determine the effect of bacteria on these genes, we constructed organoid models derived from intestinal ADs of Apc^Min/+ mice (figure 5F and online supplemental figure S4A and S4B). Treatment of these organoids with bacterial LPS and PGN upregulated DUOX2 (encoding dual oxidase 2) expression (online supplemental figure S4C), consistent with previous reports (p<0.01).23 Notably, S100A11 mRNA was also elevated with LPS and PGN treatment (p<0.05) (figure 5G). Induction of S100A11 by LPS and PGN was validated by flow cytometry analysis (p<0.05) (figure 5H). For further verification, biofilm-forming enteroaggregative Escherichia coli (EAEC) was administered to Apc^Min/+ mice with or without antibiotic cocktail treatment (figure 5I). EAEC-induced biofilm was visualised by fluorescence in situ hybridisation (FISH) using a bacterial 16s rRNA probe (EUB338). Co-staining with S100A11 revealed that its expression was upregulated in biofilm-positive AD crypts from EAEC-treated Apc^Min/+ mice, an observation abrogated in antibiotics-treated groups (all p<0.01) (figure 5J). To further verify that biofilm-induced S100A11 expression is sidedness-specific, we used Apc^Min/+ mice treated with 2% dextran sulfate sodium (DSS), which developed tumours along the proximal-distal axis of the colon (figure 5K). FISH detection using a 16s rRNA probe revealed that biofilm formation was more prevalent in tumours from proximal colon in mice (p=0.03) compared with those from distal colon (figure 5L and M), mirroring widespread occurrence of bacterial biofilm in human right-sided ADs. In line with this, S100A11 level was increased in the proximal colon tumours as compared with distal counterparts (p=0.009) (figure 5N). Of note, S100A11 expression was exclusively elevated in biofilm-residing tumours from the proximal colons (p=0.014) (figure 5O), suggesting that bacterial biofilm might mediate S100A11 expression in proximal colon tumours. Furthermore, to validate the increased S100A11 expression specifically in the human right-sided ADs, we integrated the Epi-L and Epi-R cluster cells with the epithelial cells from the colon tissue samples in HS and HV datasets for further analysis (online supplemental figure S5A). Cells in colon tissue exhibited the higher expression of canonical markers, while demonstrating markedly lower S100A11 expression compared with the Epi-R and Epi-L cluster cells (online supplemental figure S5B and S5C). In particular, no significant alteration was observed in S100A11 expression in cells from left-colon and right-colon tissues (online supplemental figure S5C). Consistently, tumour-specific elevation of S100A11 was demonstrated in the right-sided ADs in our cohort, as determined by reverse transcription-qPCR, ELISA and spatial transcriptomics (all p<0.01) (figure 5P, online supplemental figure S6A and S5D). Intriguingly, there was a significant correlation between bacterial 16s and S100A11 expression in right-sided ADs (r=0.671, p=0.016) but not in left-sided ADs (r=−0.083, p=0.778) (figure 5Q). Taken together, these results implied that the existence of bacteria and biofilm might induce tumour-specific S100A11 expression in right-sided ADs.

S100A11 suppresses antitumour immunity through MDSC recruitment

To probe the potential role of S100A11 in tumourigenesis, we overexpressed S100A11 in murine AD organoids (lower endogenous S100A11 expression), and accordingly performed S100A11 knockdown in CT26 colon carcinoma cell line (higher endogenous S100A11 expression) (online supplemental figure S6B, figure 6A, B and E). S100A11 overexpression had no significant effect on cell proliferation (figure 6C). Nevertheless, S100A11 overexpression promote the tumour growth of AD organoids in immunocompetent C57BL/6 hosts (p=0.002), implying that S100A11 might suppress antitumour immune responses to promote tumourigenesis (figure 6D). Intriguingly, ELISA results revealed that S100A11 levels in the overexpression allograft tumours could mimic the physiologically higher S100A11 levels in the human right-sided ADs (online supplemental figure S6A and S6C). Furthermore, immunosuppressive role of S100A11 was validated in CT26 allografts. We observed significantly slower tumour progression of S100A11-KD CT26 cells in wild-type BALB/c hosts (p=0.001), but not in immunodeficient hosts (figure 6F and G). Taken together, these results suggested that S100A11 promote tumour growth of colon AD and carcinoma through impairing antitumour immunity.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6

S100A11-induced MDSC recruitment impairs antitumour immunity. (A) Organoid cells derived from intestinal adenoma of Apc^Min/+ mice were further transfected with lentivirus for S100A11 OE. (B) Western blot analysis confirming the S100A11 overexpression in organoid cells. (C) Representative images of Ctrl and S100A11 OE organoid cells. Box plot displaying the quantification of organoid diameter from Ctrl and S100A11 OE group. (D) Tumour volume (left) and tumour weight (right) were increased in S100A11 OE organoids (n=4). (E) Western blot analysis confirming the S100A11 knockdown in CT26 (top). CT26 with or without S100A11 expression were inoculated into immunodeficient and immunocompetent mouse (bottom). (F) S100A11 knockdown did not affect tumour volume (left) and weight (right) in NSG mice (n=5). (G) S100A11 knockdown impaired tumour growth (left) and reduced tumour weight (right) in the BALB/c mice (n=5). (H) S100A11 OE enhanced MDSC infiltration in adenoma organoid allograft tissues (top) (n=5). S100A11 knockdown reduced the MDSC levels in CT26 allograft tissues (bottom) (n=5). (I) Activated and cytotoxic T cells levels were decreased in adenoma organoid allograft tissues with S100A11 OE (top) (n=4). Knockdown of S100A11 enhanced T cell activation and cytotoxicity (bottom) (n=5). (J) Conditioned medium from S100A11 OE organoid cells induced MDSC migration, which were suppressed by anti-S100A11 neutralising antibody (1 µg/mL) treatment (left, n=4). Anti-S100A11 neutralising antibody (1 µg/mL) suppressed MDSC migration in CT26-shNC (shRNA-Negative Control) cell conditioned medium without effects on S100A11 KD cells (right, n=4). (K) Anti-S100A11 neutralising antibody (1 µg/mL) treatment suppressed the growth of CT26-shNC allografts, but had no effect on S100A11 allografts in BALB/c mice (n=6). (L) Anti-S100A11 neutralising antibody (1 µg/mL) treatment reduced MDSC infiltration, but had no effect on S100A11 allografts in BALB/c mice (n=6). Data are presented as mean±SEM for growth curve (F, G and K) and mean±SD for others (D, F, G, H, I, J, K and L). P values were determined using two-tailed two-way analysis of variance for growth curve comparison (F, G and K), unpaired two-tailed Student’s t-test for two-group comparison (D, F, G, H, I, J, K and L). Ctrl, control; KD, knockdown; MDSC, myeloid-derived suppressor cells; ns, no significance; OE, overexpression.

Some S100 family members, such as S100A8 and S100A9, are secreted extracellularly and promote the chemotaxis of immune cells, such as MDSCs.24 We thus hypothesised that S100A11 might drive MDSC infiltration. Flow cytometry showed that MDSC infiltration were induced in AD organoid allografts with S100A11 overexpression (p=0.001) (figure 6H and online supplemental figure S 7). IHC staining results indicated that elevated MDSC level was observed in the colonic tumours from biofilm-forming EAEC-treated Apc^Min/+ mice in line with the increased S100A11 levels (p=0.008) (online supplemental figure S6D). Accordingly, knockdown of intrinsic S100A11 level led to decreased MDSC infiltration in AD organoid tumours and CT26 tumours (all p<0.01) (figure 6H and online supplemental figure S6F and S6E). Particularly, granulocytic MDSCs were reduced in CT26 S100A11-KD allograft tumours (p=0.002), while no significant change was found for monocytic MDSCs (online supplemental figure S6G). MDSC dampens antitumour immune response by suppressing T cell-mediated immune responses.25 In coincidence with this, activated and cytotoxic T cells were decreased in S100A11-OE organoid allografts (all p<0.05), while they were significantly elevated in CT26 allografts with S100A11-KD (all p<0.05) (figure 6I). To examine whether secreted S100A11 was chemotactic signal for MDSC recruitment, we performed in vitro transwell assays. Conditional medium from CT26 cells enhanced MDSC migration compared with that of organoid cells (p<0.001), which correlates with higher S100A11 expression and MDSC infiltration in CT26 allograft tumours (all p<0.05) (online supplemental figure S6J, S6C and S6H). Furthermore, conditional medium from AD organoid cells with S100A11 OE promoted MDSC migration (p< 0.05), an effect was blunted by S100A11 neutralising antibody (p=0.005) (figure 6J and online supplemental figure S6I). S100A11 neutralising antibody also abrogated MDSC migration in conditioned medium from CT26-shNC (p =0.007), but not in CT26-shS100A11 cells (figure 6J). Thus, S100A11 functions as chemotactic signal for MDSC recruitment. We next investigated the functional role of S100A11-mediated MDSC recruitment in vivo. Administration of S100A11 neutralising antibody inhibited growth of CT26-shNC allografts, as evidenced by reduced tumour volume (p<0.001) and weight (p=0.008), but had no effect on S100A11-KD tumours (figure 6K). Flow cytometry showed that S100A11 neutralisation antibody suppressed MDSC infiltration in controls but not in S100A11 knockdown tumours (figure 6L). This further supports the role of S100A11 in MDSC recruitment and facilitating tumour growth.

S100A11-RAGE axis mediates MDSC infiltration in colonic adenoma and carcinoma as therapeutic target in combination with anti-PD1 therapy

S100 family proteins are small calcium binding proteins with high structural homology.26 Several S100 members including S100A11 are ligands of the receptor for advanced glycation end products (RAGE).27 S100A8/A9 can bind to RAGE expressed on MDSC to promote intratumoral accumulation.24 We thus examined whether S100A11-RAGE axis mediated recruitment of MDSC in colonic AD and carcinoma. The RAGE antagonist FPS-ZM1 (5 mg/kg) impaired the growth of AD organoid allografts with S100A11 overexpression, as evidenced by reduced tumour volume (p=0.003) and weight (p=0.003) (figure 7A). Flow cytometry revealed FPS-ZM1 suppressed MDSC infiltration induced by S100A11 overexpression in AD organoid allografts (p=0.019) (figure 7B). In CT26 allografts, FPS-ZM1 inhibited the tumour growth of shNC allografts (p=0.037) without significant effects on S100A11-KD tumours (figure 7C). Consistently, reduced MDSC caused by FPS-ZM1 treatment was observed in shNC allografts (p=0.002) but not in S100A11-KD group (figure 7D). These results suggested that RAGE mediated the S100A11-induced MDSC recruitment in colonic AD and carcinoma. To further verify this, FPS-ZM1 were applied to in vitro transwell assays. FPS-ZM1 suppressed the enhanced migration of MDSC in conditioned medium from S100A11-OE AD organoids (p<0.001) (figure 7E). Similarly, the MDSC migration was abrogated in conditioned medium of CT26-shNC (p=0.003), but not in S100A11-KD group (figure 7E). These were consistent in vivo results, further supporting that S100A11-induced MDSC recruitment depends on RAGE.

• Download figure
• Open in new tab
• Download powerpoint

Figure 7

MDSC infiltration mediated by S100A11-RAGE axis emerges as a therapeutic target synergising with anti-PD1. (A) RAGE antagonist FPS-ZM1 impaired the tumour growth (left) and tumour volume (right) induced by S100A11 OE in organoid allograft models (4–5 mice per group). (B) RAGE antagonist FPS-ZM1 reduced the upregulated MDSC levels caused by S100A11 OE in organoid allograft models (4–5 mice per group). (C) RAGE antagonist FPS-ZM1 suppressed the growth of CT26-shNC allografts, but had no effect on S100A11 knockdown allografts in BALB/c mice (n =5). (D) RAGE antagonist FPS-ZM1 reduced the MDSC in CT26-shNC allografts without effects on S100A11 knockdown allografts in BALB/c mice (n =5). (E) Conditioned medium from S100A11 OE organoid cells induced MDSC migration, which were abrogated by FPS-ZM1 treatment (left, 1 μM, n=4). FPS-ZM1 treatment (1 μM) attenuated the MDSC migration in the CT26-shNC conditional medium without effects in S100A11 KD groups (right, 1 μM, n=4). (F) S100A11 knockdown synergised with anti-PD1 in suppressing the growth of CT26 allografts (n=6). (G) Activated and cytotoxic T cells were induced by combined S100A11 knockdown and anti-PD1, while MDSC infiltration was reduced by the combination of S100A11 knockdown plus anti-PD1 therapy in CT26 allografts (n=6). (H) FPS-ZM1 impaired the growth of CT26 allografts with anti-PD1 treatment (n=5). (I) Azeliragon synergised with anti-PD1 in suppressing the growth of CT26 allografts (n=5). (J) Activated and cytotoxic T cells were induced by combined Azeliragon and anti-PD1, while MDSC infiltration was reduced by the combination of Azeliragon plus anti-PD1 therapy in CT26 allografts (n=5). Data are presented as mean±SEM for growth curve (C, F, H and I) and mean± SD for others (A, B, C, D, E, F, G, I and J). P values were determined using two-tailed two-way ANOVA for growth curve comparison (C, F, H and I), two-tailed one-way ANOVA for multiple comparison (A, B, C, D, E, F, G, I and J). ANOVA, analysis of variance; KD: knockdown; MDSC, myeloid-derived suppressor cells; ns: no significance; OE: overexpression; PD1: programmed cell death protein 1; RAGE: receptor for advanced glycation end products.

As MDSCs are the major barrier for tumour immunotherapy, targeting S100A11-RAGE axis might be potential approach to inhibit MDSC infiltration and boost immunotherapy efficacy. Indeed, S100A11 knockdown plus anti-PD1 exerted synergistic effects in suppressing tumour growth as compared with single treatment, leading to tumour regression (p<0.01) and reduction in tumour weight (p<0.05) (figure 7F). Flow cytometry revealed S100A11 knockdown plus anti-PD1 synergised to reduce MDSC and promote activated and cytotoxic T-cell infiltration (figure 7G). We further examined whether targeting S100A11-RAGE axis using RAGE antagonist have synergistic effect in combination with anti-PD1. The results showed that FPS-ZM1 plus anti-PD1 synergistic suppressed tumour progression in mice (p<0.001) (figure 7H). Another RAGE antagonist Azeliragon is orally active and well-tolerated as a granted orphan drug for patient with glioblastoma. Oral administration of Azeliragon (4 mg/kg) suppressed tumour growth by ~50% in CT26 allografts without body weight reduction in hosts (figure 7I and online supplemental figure S6K), proving it as effective and safe medication for CRC. Importantly, Azeliragon plus anti-PD1 exerted synergistic effects as compared with single treatment, leading to tumour regression (p<0.05) and reduced tumour weight (p<0.05) (figure 7I). A combination of Azeliragon and anti-PD1 synergised to reduce MDSC while promoting cytotoxic T-cell infiltration (figure 7J). These results implied that targeting S100A11-RAGE axis with Azeliragon was promising in combination with anti-PD1 therapy.