Human gut M cells resemble dendritic cells and present gluten antigen

Abstract

Microfold (M) cells are rare intestinal epithelial cells that reside in the follicle-associated epithelium of Peyer’s patches¹. M cells transport luminal antigens to submucosal antigen-presenting cells^2,3. These insights primarily derive from transmission electron microscopy and studies using genetically modified mice^2,3,4. Here we establish an intestinal organoid model to study human M cells and reconstruct the differentiation trajectory of M cells through transcriptome profiling. The results indicate that as well as facilitating luminal antigen transport, human M cells also directly present antigens via the class II major histocompatibility complex (MHC-II). Notably, the related enterocytes only express MHC-II in chronic inflammatory states and do not express typical dendritic cell markers. Human M cells physiologically express a gene profile that resembles that of dendritic cells. Similar to dendritic cells, M cell development is induced by RANKL and CSF2 and requires the transcription factors SPIB and RUNX2. HLA-DQ2.5 M cells process and present gluten antigen as demonstrated in organoid–T cell co-culture assays. These findings suggest that M cells may have a central role in coeliac disease.

Main

M cells are rare epithelial cells that are involved in intestinal mucosal immunity^1,2,3. They have irregular microvilli on their apical surface^4,5 and transcytose luminal antigens to the immune cells that reside in their basally located ‘pocket’^2,3,4. Mouse M cells are derived from LGR5⁺ intestinal stem cells⁶, require RANKL⁷ and the transcription factors SPIB^6,8 and SOX8 (ref. ⁹), and express Gp2^{ (ref. 10}). The differentiation trajectory of human M cells has largely remained unknown. Mouse M cell markers are often not expressed in humans—for example, Ulex europaeus agglutinin-1 (UEA-1) uniquely recognizes mouse M cells¹¹, whereas human M cells display sialyl Lewis A antigen¹². Of note, the M cell pocket contains both dendritic cells and T cells¹³. M cells may directly interact with T cells through a dendritic cell-independent mechanism.

Human M cell organoids

We recently developed a culture protocol for human intestinal organoids¹⁴, which we further optimized as ‘M cell medium’ by adding RANKL, tumour necrosis factor (TNF) and retinoic acid, leading to the appearance of GP2⁺ cells (Fig. 1a–c). Consistent with observations in mice⁷, RANKL appeared to be essential for the development of M cell organoids (Fig. 1b,c). M cell organoids exhibited reduced numbers of proliferative buds (Extended Data Fig. 1a–d), as noted previously for mouse M cell organoids¹⁵. As previously reported¹⁶, lymphotoxin-β (LTB) enhanced the frequency of GP2⁺ cells but led to cell death in our culture (Extended Data Fig. 1e,f). LTB had little effect on absolute numbers of M cells and was thus excluded from the differentiation cocktail.

Fig. 1: A human intestinal organoid model containing M cells.

a, Schematic of M cell differentiation from human intestinal organoids. Representative cell types in the cultured organoids are indicated. Images were created in BioRender. van Es, J. (2025) https://BioRender.com/xh7haut. EEC, enteroendocrine cell; TA cell, transit-amplifying cell. b,c, Representative flow cytometry analysis (b) and quantification of GP2⁺ M cell percentage (c) in cultured M cell organoids. The effect of each M cell differentiation factor was evaluated by removing it from the M cell medium. c, For each tested factor, three different donors were tested. Each dot represents the mean value derived from three independent wells. Data are mean ± s.e.m. P values from one-way ANOVA with Dunnett’s test. Ctrl, control; RA, retinoic acid. d,e, Representative confocal image (d) and flow cytometry analysis (e) of SPIB-P2A-tdTomato reporter organoids cultured in M cell medium. Mature M cells are marked by tdTomato fluorescence and GP2 antibody staining (green, arrows in d). Three independent experiments were performed on cells from one donor with similar results. Scale bar, 50 µm. f, scRNA-seq analysis of 375 cells derived from human M cell organoids. Cell clusters are visualized in uniform manifold approximation and projection (UMAP) plots and coloured by cell type (left) or by duplicate plates (right, representing two biological replicates). Cell numbers are shown in brackets below each indicated cell type. g, Violin plots showing the expression levels of cell-type-specific markers for M cells and enterocytes in the M cell organoid-derived scRNA-seq dataset containing 375 cells. h, Representative TEM images of M cell organoids. M cells are identified by having fewer apical microvilli compared with neighbouring enterocytes. M cell organoids from two donors were tested with similar results (see also Extended Data Fig. 1p). Scale bars: 10 µm (left), 2 µm (middle and right).

Source Data

We next created SPIB reporter organoids (SPIB-P2A-tdTomato; Extended Data Fig. 1g). M cell medium efficiently generated SPIB-expressing (tdTomato⁺) cells (Extended Data Fig. 1h). Immunofluorescent staining and flow cytometry analysis revealed that SPIB expression was induced in more than 75% of cells, with approximately 3% co-expressing GP2 on their apical surface (Fig. 1d,e). These SPIB⁺GP2⁺ cells expressed other M cell markers such as SOX8 and CCL23 (ref. ¹⁶), as demonstrated by quantitative PCR (qPCR) analysis following fluorescence-activated cell sorting (FACS) (Extended Data Fig. 1i). Cells derived from M cell organoids were sorted by FACS for single-cell RNA-sequencing (scRNA-seq) analysis. We thereby identified three cell clusters corresponding to enterocytes, intermediate cells and M cells (Fig. 1f,g and Extended Data Fig. 1j,k).

To identify marker genes for primary M cells, we re-analysed the published scRNA-seq resource¹⁷, which reportedly contained around 400 M cells among more than 420,000 total cells. This dataset was integrated with three other scRNA-seq datasets that covered all other intestinal cell types (Extended Data Fig. 1l). We noted that more than half of the previously annotated M cells expressed GP2 but no other M cell markers (Extended Data Fig. 1l,m). These cells were marked by TFF2 and PGC and derived from the duodenum (Extended Data Fig. 1m,n). GP2 is known to mark human Brunner’s gland cells¹⁸. Immunohistochemistry (IHC) staining for TFF2 on human duodenum sections stained Brunner’s gland (Extended Data Fig. 1o). Exclusion of these cells defined a core set of primary human M cell markers, including GP2, SPIB, SOX8, SLC2A6, PTAFR and GABRP (Extended Data Fig. 1m), which were also highly expressed in organoid M cells (Fig. 1g). M cells exhibit far fewer and shorter apical microvilli compared with enterocytes^4,5. Transmission electron microscopy (TEM) imaging of organoid M cells visualized this difference (Fig. 1h and Extended Data Fig. 1p). We further established a two-dimensional (2D) air–liquid interface (ALI) culture based on M cell medium (Extended Data Fig. 1q–s). Apical microvillus structures were clearly visible with actin staining, and co-staining for GP2 confirmed that cells with fewer microvilli were M cells (Extended Data Fig. 1t). We conclude that our organoid model generates genuine human M cells.

ICAM2 marks M-lineage cells

GP2 is the only known surface marker for rare, fully mature M cells¹⁰. The primary M cell dataset suggested ICAM2 as a broader M cell marker (Fig. 2a). Indeed, ICAM2 was expressed in more than 90% of GP2⁺ organoid M cells (Fig. 2b), yet about 50% of GP2^− cells also expressed ICAM2 (Fig. 2b). SPIB⁺ M-lineage organoid cells could be subdivided into ICAM2^− and ICAM2⁺ cells (Fig. 2c,d). Thus, we could distinguish three stages within the M-lineage cells—SPIB⁺ICAM2^−GP2^− early M cells, SPIB⁺ICAM2⁺GP2^− immature M cells and SPIB⁺ICAM2⁺GP2⁺ mature M cells—in addition to SPIB^−ICAM2^−GP2^− non-M cells (Fig. 2d,e). These four populations were sorted for transcriptome profiling (Fig. 2d,f). SPIB^− non-M cells expressed enterocyte markers that decreased in abundance during M cell differentiation (Fig. 2g and Extended Data Fig. 2a). Immature M cell markers (TNFAIP2^{ (ref. 19}), SOX8 and CCL23) were enriched in ICAM2⁺ cells (Fig. 2g). M cell maturation coincided with appearance of previously reported human M cell markers and RANKL-responsive genes¹⁶ (Extended Data Fig. 2b). In total, 1,639 differentially expressed genes (DEGs) were identified (Fig. 2h). Gene ontology (GO) analysis identified enterocytes and M cells as the most relevant cell types on the basis of the downregulated and upregulated genes in M cells, respectively (Fig. 2i). Additionally, gene set enrichment analysis (GSEA) revealed nutrient metabolic pathways and brush border components in enterocytes, whereas M cells expressed genes related to pathogen infection and immune cell interaction (Extended Data Fig. 2c–e). Markers for other intestinal cell types were undetectable in M cell organoids (Extended Data Fig. 2f). In human Peyer’s patch tissues, immunofluorescent staining confirmed the presence of ICAM2⁺ M cells (Extended Data Fig. 3a). Some ICAM2⁺ cells exhibited features of mature M cells, yet most represented immature M cells with enterocyte-like brush borders. scRNA-seq analysis of primary M cells confirmed that early, immature and mature stages were distinguishable by SPIB, ICAM2 and GP2 expression (Extended Data Fig. 3b). GP2^− and GP2⁺ ICAM2-enriched subpopulations equally expressed immature M cell markers, such as TNFAIP2 and CCL23 (Extended Data Fig. 3b). Thus, organoid M cells faithfully recapitulated human M cell differentiation. Although GP2 uniquely marks mature M cells (Fig. 2g and Extended Data Fig. 3c,d), the transcriptome of ICAM2⁺GP2^− M cells strongly resembled that of GP2⁺ M cells (Extended Data Fig. 3c,d). Only nine genes were upregulated more than twofold in GP2⁺ cells (Extended Data Fig. 3c,d). Indeed, CLU is known to mark GP2⁺ M cells¹⁰. The minimal transcriptomic difference was also observed in primary M cells (Extended Data Fig. 3e,f). Of note, some GP2^− M cells also displayed the typical M cell microvillus morphology (Extended Data Fig. 1t, arrows).

Fig. 2: Differentiation trajectory of human M cells.

a, Violin plot showing the expression levels of ICAM2 across multiple human intestinal cell types from tissue-derived scRNA-seq datasets containing 15,543 cells. b, Representative flow cytometry analysis of ICAM2 and GP2 expression in M cell organoids. Three independent experiments were performed on two donors with similar results. c, Representative confocal images of ICAM2 antibody staining in SPIB-P2A-tdTomato reporter organoids cultured in M cell medium. M cells are marked by tdTomato fluorescence (red) and show positive staining of ICAM2 antibody (green, primarily detected on the basolateral cell surface of M cells). Three independent experiments were performed with cells from one donor with similar results. Scale bar, 50 µm. d, Representative flow cytometry analysis of ICAM2 and GP2 expression in SPIB-P2A-tdTomato reporter organoids cultured in M cell medium. Three independent experiments were performed on one donor with similar results. e, Schematic of spatial-temporal differentiation of human M cells, showing markers associated with M cells at different stages. Images were created in BioRender. van Es, J. (2025) https://BioRender.com/xh7haut. f, Principal components analysis plot showing transcriptomal changes during M cell differentiation. n = 2 biological replicates from one donor. Each dot represents one biological replicate for the FACS-sorted cell population. g, Expression levels (normalized counts) of cell-type-specific markers across the indicated cell populations. n = 2 biological replicates. Data are mean values. h, Heat map showing expression patterns of DEGs during M cell differentiation. n = 2 biological replicates. i, Volcano plot showing GO analysis based on downregulated and upregulated DEGs in M cells, which identifies the most relevant cell types as enterocytes and M cells, respectively. P values from one-sided Fisher’s exact test.

Source Data

ICAM1 and ICAM2 are co-expressed by primary (Fig. 2a and Extended Data Fig. 3g) and organoid (Extended Data Figs. 2b and 3h) M cells. ICAM2 is located basolaterally (Fig. 2c and Extended Data Fig. 3a). ICAM molecules can facilitate adhesion between T cells and antigen-presenting cells²⁰. We next analysed the role of ICAM2 in CD4⁺ T cell binding to M cell organoids. We genetically deleted ICAM2 (Extended Data Fig. 3i) and blocked ICAM1 with a neutralizing antibody, which resulted in decreased numbers of T cells attached to the M cell organoids (Extended Data Fig. 3j,k). Notably, ICAM2 knockout did not affect M cell differentiation (Extended Data Fig. 3l,m).

Human M cell development requires RUNX2

Expression of 63 transcription factors increased during M cell differentiation (Extended Data Fig. 4a), including the known M cell transcription factors SPIB, SOX8, EHF²¹, RELB, NFKB1 (nuclear factor (NF)-κB p105 subunit) and NFKB2 (NF-κB p100 subunit)²² (Extended Data Fig. 4a), whereas expression of 59 transcription factors decreased. ONECUT2 drives enterocyte differentiation in mice¹⁵. Its expression was markedly decreased during human M cell maturation (Extended Data Fig. 4b). Treatment with the ONECUT2 inhibitor CSRM617 resulted in more than twofold increase in the number of GP2⁺ M cells in organoids (Extended Data Fig. 4c,d). SPIB is described as a master regulator for MHC-II-expressing B cells²³ and dendritic cells²⁴. Spib deficiency depletes GP2⁺ M cells in mouse Peyer’s patch epithelium⁸. SPIB-knockout organoids revealed the same essential role of SPIB in human M cell development (Extended Data Fig. 4e,f).

RANKL signalling is an essential niche factor for dendritic cell activation²⁵. Several RANKL-induced M cell transcription factors, besides SPIB, are known from dendritic cell biology (that is, EHF, RELB and NFKB1/2)²⁶. We hypothesized that dendritic cells and M cells might share a RANKL-induced transcription regulatory network. One transcription factor expressed by M cells, RUNX2, attracted our interest (Extended Data Fig. 4a,b). Like SPIB, RUNX2 is a master regulator for dendritic cells²⁷ and its expression is induced by RANKL²⁸. Addition of CADD522, a RUNX2-specific inhibitor, resulted in dose-dependent loss of GP2⁺ M cells (Extended Data Fig. 4g–i). Similarly, RUNX2 knockout led to significant loss of GP2⁺ organoid M cells (Extended Data Fig. 4g–j).

CSF2 promotes human M cell maturation

We next interrogated the dataset for M cell-expressed receptors (Extended Data Fig. 4k). Consistent with the functional effects of RANKL, TNF and retinoic acid, their receptors (TNFRSF11A, TNFRSF1B and RARG, respectively) were expressed in M cells (Extended Data Fig. 4k). We performed FACS-based screening to identify potential niche factors that could promote M cell maturation (Extended Data Fig. 4l,m). Among 12 tested ligands, CSF2 resulted in a more than fourfold increase of GP2⁺ organoid M cells (Extended Data Fig. 4l–n). CSF2 is widely used for dendritic cell differentiation in vitro. The CSF2 receptor was expressed in primary human M cells, indicating its potential function in vivo (Extended Data Fig. 4o). Notably, expression of CSF2 receptors was undetectable by qPCR analysis in organoid-derived GP2⁺ mouse M cells (Extended Data Fig. 5a,b) and was minimal in a published mouse scRNA-seq dataset containing RANKL-induced M cells²⁹ (Extended Data Fig. 5c), demonstrating an interspecies difference.

Human M cells express dendritic cell-related genes

The shared transcription factors (SPIB, RUNX2, RELB and NF-κB) and signalling regulators (RANKL and CSF2) between M cells and dendritic cells suggested a shared gene regulatory network. This prompted us to search for additional dendritic cell genes in our RNA-seq dataset. A series of lymphoid dendritic cell marker genes was induced during M cell maturation (Fig. 3a). Intestinal lymphoid dendritic cells have been defined by signature genes including CD83, LAMP3, IL7R and FSCN1^{ (refs. 17,30}) and resemble conventional dendritic cells (cDCs) more than plasmacytoid dendritic cells (pDCs)³⁰. We integrated scRNA-seq datasets of primary human intestinal epithelial cells with various immune cell types (Extended Data Fig. 5d,e). Unsupervised, transcriptomic similarity-based clustering of individual epithelial and immune cells identified a unique cell cluster (cluster 9), in which both primary M cells and lymphoid dendritic cells resided (Extended Data Fig. 5d,e, with cell types annotated by the previous study¹⁷). Signature genes defining cluster 9 included the lymphoid dendritic cell markers (Extended Data Fig. 5f) and were expressed by both lymphoid dendritic cells and M cells (Fig. 3b and Extended Data Fig. 5f). Within cluster 9, M cells, but not lymphoid dendritic cells, specifically expressed conventional M cell markers including ICAM2, CCL23 and SOX8 (Extended Data Fig. 5f). Notably, although M cells clustered with lymphoid dendritic cells, a number of cDC genes (CD74, DC-SIGN and CD11B) and several pDC genes (CXCR3 and RUNX2) also appeared upon M cell differentiation (Extended Data Fig. 5g). Thus, M cells did not resemble a specific dendritic cell subtype, but rather exhibited broader expression of dendritic cell-related genes. Finally, GO analysis using all DEGs induced during M cell maturation identified activated dendritic cells and antigen-presenting cells as being most closely related (Fig. 3c).

Fig. 3: Human M cells express dendritic cell markers including MHC-II.

a, Heat map showing the expression patterns of a set of lymphoid dendritic cell markers during M cell differentiation. n = 2 biological replicates. b, Dot plot showing the expression levels of a set of lymphoid dendritic cell markers (cluster 9 markers) across multiple intestinal epithelial and immune cell types from tissue-derived scRNA-seq datasets. n = 20,210 single cells. Dot colour relates to normalized mean expression values and dot size indicates the fraction of expressing cells. c, Volcano plot showing GO analysis based on upregulated DEGs in M cells, which identifies the most relevant cell types as activated dendritic cells (top) or antigen-presenting cells (bottom). P values from one-sided Fisher’s exact test. d, Expression levels (normalized counts) of MHC-II-related genes across the indicated cell populations. n = 2 biological replicates. Data are mean values. e, Representative flow cytometry analysis of MHC-II and GP2 expression in cultured M cell organoids (wild type or CIITA-knockout (KO)) versus normal intestinal organoids (with or without IFNγ). Three independent experiments were performed with similar results in wild-type organoids derived from three donors and in two CIITA-knockout clonal organoid lines derived from one donor. Abs, antibodies; WT, wildtype. f, Representative confocal images of MHC-II antibody staining in SPIB-P2A-tdTomato reporter organoids cultured in M cell medium. M cells are marked by tdTomato fluorescence (red) and show positive staining of MHC-II antibody (green, detected in both cytoplasm and primarily on basolateral surface of M cells). Three independent experiments were performed on cells from one donor with similar results. Scale bar, 50 µm. g, Representative immuno-electron microscopy images showing MIIC structures in endosomes, identified by immunogold staining of MHC-II antibody in organoid M cells. Arrows indicate MHC-II vesicles in the endosome. M cell organoids from two donors were tested with similar results (see also Extended Data Fig. 8a). Scale bars, 200 nm.

Source Data

Human M cells express MHC-II

Dendritic cells take up exogenous antigens and process these into peptides, which are then presented by MHC-II complexes. FACS-sorted M cells exhibited distinct morphology from SPIB^− cells when cultured in 2D, with multiple protrusions resembling dendritic cells (Extended Data Fig. 5h). These protrusions facilitated phagocytosis of bacteria particles (Supplementary Video), which could be blocked by the phagocytosis inhibitor cytochalasin B (Extended Data Fig. 5i,j). Notably, expression of MHC-II genes and the invariant chain (CD74) increased during M cell maturation (Fig. 3d and Extended Data Fig. 5k). Flow cytometry analysis for MHC-II and GP2 confirmed expression of MHC-II proteins on GP2⁺ as well as GP2^− M cells (Fig. 3e). Notably, MHC-II proteins were expressed by SPIB⁺ M-lineage cells, with the staining signals primarily detected on the basolateral cell surface (Fig. 3f), whereas SPIB^− cells did not express MHC-II genes (Fig. 3d). Flow cytometry analysis confirmed that MHC-II⁺ M cells were more mature, with more than 98% of MHC-II⁺ cells co-expressing ICAM2 (Extended Data Fig. 5l). Consistently, FACS-sorted MHC-II⁺ cells expressed various markers of immature and mature M cells (Extended Data Fig. 5m).

Notably, intestinal organoids cultured without M cell-inducing factors did not express MHC-II in the absence of IFNγ (Fig. 3e), as also shown elsewhere¹⁶. By contrast, human organoid M cells spontaneously expressed MHC-II without IFNγ under homeostatic conditions (Fig. 3e). To test whether the IFNγ-independent expression of MHC-II in M cells required the transcription factor CIITA, we generated CIITA-knockout organoids (Extended Data Fig. 5n). These organoids contained similar numbers of M cells, yet MHC-II expression was absent (Fig. 3e). IRF1 is the classical upstream transcription factor that induces CIITA expression in the presence of IFNγ. Expression of IRF1 was not induced during M cell differentiation (Fig. 3d and Extended Data Fig. 3b), indicating that CIITA expression in M cells does not depend on IRF1, which further confirms that MHC-II expression in M cell organoids is independent of IFNγ signalling. Beyond IRF1, several IFNγ-independent transcription factors have been reported as potential upstream regulators of CIITA (for example, RFX5 (ref. ³¹) and NF-κB³²). These genes were induced by threefold to fivefold during M cell differentiation (Fig. 3d).

In primary intestine, MHC-II can be expressed by enterocytes³³. The expression of MHC-II in enterocytes crucially depends on IFNγ stimulation during inflammation^33,34. MHC-II expression is almost completely abolished in Ifngr1-knockout mouse enterocytes, even with pathogenic bacterial infection³⁴. In human intestinal tissues, MHC-II⁺ enterocytes were not uniformly detectable but showed different distributions between donors and tissue areas (Extended Data Fig. 6a), indicating that expression of MHC-II is not a spontaneous, intrinsic property, but is induced by exogenous factors such as IFNγ. Notably, unlike M cells, the transcriptome of MHC-II⁺ enterocytes did not resemble that of lymphoid dendritic cells (Extended Data Fig. 6b). In the scRNA-seq datasets, MHC-II⁺ enterocytes did not cluster with any dendritic cell subtype (Extended Data Fig. 6c). Transcriptomic comparison between MHC-II-expressing M cells and enterocytes identified 414 DEGs that were expressed at higher levels in M cells (Extended Data Fig. 6d,e). These DEGs were again highly co-related with blood dendritic cells, as revealed by GO analysis (Extended Data Fig. 6f).

It has remained controversial whether human M cells can directly present antigens to CD4⁺ T cells^5,35,36. We performed immunofluorescent staining of MHC-II on primary human Peyer’s patch tissues. MHC-II signals could be detected intracellularly and on the basolateral cell surface of M cells (Extended Data Fig. 6g,h). Of note, scRNA-seq analysis based on reliable M cell marker genes revealed that both ICAM2⁺GP2^− and ICAM2⁺GP2⁺ primary M cells expressed MHC-II genes and CD74 (Extended Data Fig. 6i). This analysis avoids limitations of IHC staining regarding the uncertainty of MHC-II signals between neighbouring cells^35,36. Furthermore, flow cytometry analysis of MHC-II in ICAM2⁺ or GP2⁺ primary M cells confirmed MHC-II protein expression in more than 97% of ICAM2⁺ and GP2⁺ M cells (Extended Data Fig. 6j,k). Although two previous M cell studies reached contradictory conclusions regarding MHC-II expression in M cells, both detected MHC-II signals in human follicle-associated epithelium (FAE) tissues^35,36. It should be noted that besides GP2⁺ M cells, FAE also contains many M-lineage cells that do not express GP2 (refs. ^9,19) and morphologically resemble non-M enterocytes. In M cell organoids, many GP2^− immature M cells were found to express MHC-II in the absence of IFNγ, in addition to GP2⁺ M cells (Fig. 3e). Therefore, we hypothesized that in primary tissue, MHC-II⁺ FAE cells (GP2^− normal microvilli) are likely to represent immature M cells that constitutively express MHC-II, rather than representing IFNγ-induced MHC-II⁺ enterocytes. Indeed, many ICAM2⁺ immature M cells carried normal microvillus structures in human FAE (Extended Data Fig. 3a), and a majority of these primary ICAM2⁺ M cells expressed MHC-II (Extended Data Fig. 6k). Of note, the presence of occasional MHC-II⁺ enterocytes (induced by IFNγ) in human FAE could not be formally excluded.

Notably, MHC-II expression was not observed in mouse M cells derived either from organoids (Extended Data Fig. 5b,c) or from primary FAE tissues (Extended Data Fig. 7a–c). Most dendritic cell markers identified in human M cells were not expressed in mouse (Extended Data Fig. 5c). GO analysis of mouse M cell markers identified terms related to conventional M cell functions, but no terms related to antigen presentation (Extended Data Fig. 7d). These differences may explain why an antigen-presenting function of M cells has not been observed in mouse experiments despite clear similarities between mouse and human M cells (Extended Data Fig. 7e).

A hallmark of antigen presentation via MHC-II molecules is the MHC-II compartment (MIIC), where antigen processing and loading occur³⁷. We assessed the existence of MIIC in M cells by immuno-electron microscopy imaging of MHC-II and CD63, two marker proteins for MIIC structures³⁷, and identified structures that represent early, multivesicular, intermediate and multilaminar MIICs in M cells and resemble the corresponding MIIC structures in dendritic cells^37,38 (Fig. 3g and Extended Data Fig. 8a). We also analysed MIIC in normal intestinal organoids. Consistent with undetectable levels of MHC-II in unstimulated normal intestinal organoids (Fig. 3e), no MHC-II⁺ vesicles were observed in the endosomes (Extended Data Fig. 8b). Upon IFNγ stimulation, normal intestinal organoids predominantly exhibited multivesicular MIIC structures, whereas tubular extensions and the multilaminar structures were rarely observed (Extended Data Fig. 8b). Given the proposed role of multilaminar MIICs in high-fidelity peptide exchange³⁹, these structural differences may underlie the limited capacity for complete antigen processing and presentation by IFNγ-induced enterocytes, despite the presence of surface MHC-II. Additionally, like dendritic cells, various pattern recognition receptors and key components of the antigen processing and presentation machinery were expressed in M cells, further supporting their potential antigen-presenting function (Extended Data Fig. 8c–e).

Human M cells present gluten antigen

Next, we established an organoid/T cell co-culture assay focused on T cell activation in coeliac disease (CeD), an auto-immune disorder that is triggered by gliadin-derived antigens and linked to HLA-DQ haplotypes (HLA-DQ2.5)⁴⁰. To our knowledge, the CeD model is the most comprehensively characterized antigen-presentation platform in the human intestine, with clearly defined HLA haplotypes, pathogenic antigen sequences and antigen-specific T cell receptor (TCR) sequences^41,42. We synthesized a representative 33-mer antigenic peptide originated from α2-gliadin (referred to as 33-mer peptide; with sequence LQLQPF(PQPELPY)₃PQPQPF) and generated transgenic reporter Jurkat T cell lines expressing the specific TCR: S16 (ref. ⁴²) or D2 (ref. ⁴¹) (Extended Data Fig. 9c).

Organoid lines were established from two HLA-DQ2.5 donors, which expressed MHC-II in M cells upon differentiation (Extended Data Fig. 9a, b). After adding the 33-mer peptide, we performed immunofluorescent staining using an antibody (DONQ52 antibody) that recognizes immunogenic gliadin peptide–HLA-DQ2.5 complexes⁴¹. Staining signals were detected in M cell organoids with the HLA-DQ2.5 haplotype, but not in non-HLA-DQ2.5 organoids or when the peptide was absent (Extended Data Fig. 9d). Given that MHC-II molecules are primarily located basolaterally (on the ‘outside’) in M cell organoids (Fig. 3f), the 33-mer peptide in culture medium could bind directly to HLA-DQ2.5 without intracellular processing. We therefore generated apical-out M cell organoids by removing the extracellular matrix, resulting in luminal localization of MHC-II (Fig. 4a). Loading the 33-mer peptide onto HLA-DQ2.5 would thus require endocytosis. Immunofluorescent staining using the DONQ52 antibody again detected gliadin peptide–HLA-DQ2.5 complexes in GP2⁺ M cells derived from HLA-DQ2.5 donors, but not in those from non-HLA-DQ2.5 donors (Fig. 4b). HLA-DQ2.5 M cells with limited phagocytosis also exhibited reduced peptide–HLA-DQ2.5 signals (Fig. 4b), confirming that the peptide was taken up by M cells and loaded onto HLA-DQ2.5.

Fig. 4: Human M cells present gluten antigen for T cell activation.

a, Schematic of normal versus flipped organoids. Images were created in BioRender. van Es, J. (2025) https://BioRender.com/xh7haut. BME, basement membrane extract. b, Representative confocal images of DONQ52 antibody staining in apical-out M cell organoids derived from HLA-DQ2.5 and non-HLA-DQ2.5 donors. The 33-mer peptide was added before staining. M cells are marked by GP2 antibody staining (green). DONQ52 antibody specifically recognizes the gliadin peptide–HLA-DQ2.5 complexes (red). In each condition, three independent experiments were performed on cells from two donors with similar results. Scale bars, 20 µm. c, Schematic of organoid–T cell co-culture assay. Images were created in BioRender. van Es, J. (2025) https://BioRender.com/xh7haut. d, Representative flow cytometry analysis and quantification of eGFP⁺ T cell percentage after co-culture with organoids treated with or without 33-mer peptide. Organoids (org.) from two HLA-DQ2.5 (DQ2.5) donors were tested. For each donor, n = 3 biological replicates are shown; each dot represents one biological replicate. Data are mean ± s.e.m. P values from two-tailed t-test. e, Activation of CD4⁺ T cells derived from patients with CeD, quantified by flow cytometry analysis of CD69⁺ cells after co-culture with MHC-II⁺ or MHC-II^− cells treated with or without 33-mer peptide. MHC-II⁺ and MHC-II^− cells were sorted by FACS from M cell organoids derived from HLA-DQ2.5 donor 2. Patient-derived T cells were isolated from peripheral blood mononuclear cells (PBMCs) from two HLA-DQ2.5 donors. For each PBMC donor, one T cell-only well and two independent co-culture wells are shown. Each dot represents one well. Data are mean values or mean ± s.e.m. P values from one-way ANOVA with Tukey’s test.

Source Data

To assess the effect of M cell-presented, CeD-associated epitopes on T cell activation, we co-cultured organoids with Jurkat T cells expressing the HLA-DQ2.5-restricted, gluten peptide-specific S16 or D2 TCRs. T cell activation was measured by eGFP fluorescence or luciferase activity induced by activated NFAT (Fig. 4c and Extended Data Fig. 9c). M cell organoids treated with the 33-mer peptide efficiently activated T cells compared with untreated organoids (Fig. 4d and Extended Data Fig. 9e). By contrast, the addition of 33-mer peptide to intestinal organoids without M cells did not induce T cell activation (Fig. 4d and Extended Data Fig. 9e). We validated this in a more physiologically relevant context by co-culturing the MHC-II⁺ and MHC-II^− cells derived from HLA-DQ2.5 M cell organoids with primary CD4⁺ T cells isolated from blood lymphocytes of patients with CeD with the HLA-DQ2.5 haplotype. Flow cytometry analysis for the T cell activation marker CD69 confirmed MHC-II-restricted, gliadin antigen-dependent T cell activation, mediated upon M cell antigen presentation (Fig. 4e).

Human M cells deamidate gliadin peptides

We next exposed apical-out M cell organoids to enzyme (pepsin, trypsin and tissue transglutaminase)-digested whole-gliadin proteins. DONQ52 antibody staining again detected gliadin peptide–HLA-DQ2.5 complexes in GP2⁺ M cells (Fig. 5a), indicating that gliadin proteins were taken up through the apical side of M cells, processed and loaded onto HLA-DQ2.5. In co-culture assays, these gliadin protein-treated M cell organoids activated antigen-specific T cells to a level similar to that of the 33-mer peptide-treated M cell organoids (Fig. 5b).

Fig. 5: Conversion of gliadin peptide to pathogenic CeD antigen by M cell-derived TGM2.

a, Representative confocal images of DONQ52 antibody staining (red) in apical-out M cell organoids treated with gliadin proteins that were pre-treated with pepsin, trypsin and TGM2. M cells are marked by GP2 antibody staining (green). Three independent experiments were performed on two HLA-DQ2.5 donors with similar results. Scale bar, 20 µm. b, Quantification of luciferase activity in reporter T cells co-cultured with M cell organoids treated with or without indicated peptides. n = 8 biological replicates on one donor. Data are mean ± s.e.m. P values from one-way ANOVA with Dunnett’s test. c, Amino acid sequences showing TGM2-mediated pathogenic conversions from glutamine (Q) to glutamic acid (E) in 33-mer peptide. d, Violin plot showing the expression levels of TGM2 in the M cell organoid-derived scRNA-seq dataset containing 375 cells. e, Quantification of luciferase activity in reporter T cells co-cultured with M cell organoids treated with or without 33-mer peptide (Q). n = 4 biological replicates from one donor. Data are mean ± s.e.m. P values from one-way ANOVA with Dunnett’s test. f, Representative confocal images of DONQ52 antibody staining (red) in apical-out M cell organoids treated with gliadin proteins that were digested with pepsin and trypsin. M cells are marked by GP2 antibody staining (green). Three independent experiments are performed on two HLA-DQ2.5 donors with similar results. Scale bar, 20 µm. g, Validation of TGM2 knockout by targeted genotyping. sgRNA, single guide RNA. h, Representative flow cytometry analysis of MHC-II and GP2 expression in TGM2-knockout M cell organoids. Three independent experiments were performed with cells from one donor with similar results. Ref., reference. i, Quantification of luciferase activity in reporter T cells co-cultured with TGM2-knockout M cell organoids treated with or without indicated peptides. n = 8 biological replicates from one donor. Data are mean ± s.e.m. P values from one-way ANOVA with Dunnett’s test.

Source Data

Efficient gliadin-derived peptide loading onto HLA-DQ2.5 requires deamidation: glutamine (Q) is converted to glutamic acid (E) in the original gliadin peptides⁴³ (Fig. 5c). The responsible enzyme, TGM2, has been reported to be predominantly expressed by intestinal stromal cells residing on the basolateral side of gut epithelium⁴⁴, and its expression can be upregulated by RANKL⁴⁵. Of note, M cells highly expressed TGM2 to levels more than 18-fold higher than in enterocytes (Fig. 5d). Thus, M cells may convert the original gliadin peptides into the pathogenic peptides. We added the 33-mer peptide (Q) without the Q-to-E conversion (Fig. 5c) to M cell organoids, which resulted in T cell activation (Fig. 5e). This T cell activation by 33-mer peptide (Q) could be blocked by ZED1227, a TGM2-specific inhibitor (Fig. 5e). DONQ52 antibody staining confirmed that enzyme (pepsin and trypsin, but without tissue transglutaminase)-digested gliadin proteins were processed and efficiently loaded onto HLA-DQ2.5 (Fig. 5f). This implied Q-to-E conversion of the gliadin protein-derived peptides by M cell-derived TGM2. To further validate this, we generated TGM2-knockout organoids (Fig. 5g). Absence of TGM2 protein in M cell organoids and the culture medium was validated by ELISA (Extended Data Fig. 9f). Knockout of TGM2 did not affect M cell differentiation and MHC-II expression (Fig. 5h). Thus, the 33-mer peptide (Q to E) and TGM2-treated gliadin proteins could still be presented by HLA-DQ2.5 on the TGM2-knockout organoids as read out in the reporter T cell activation assays (Fig. 5i). The 33-mer peptide (Q) and gliadin proteins (the latter not treated with TGM2) did not activate the reporter T cells (Fig. 5i). This was consistent with a previous report that found that D2 TCR activation depends on the presence of E residue in the gliadin-derived epitopes⁴¹. In addition, SPIB-knockout organoids did not activate the reporter T cells in co-culture assays when using 33-mer peptide (Q) and gliadin proteins without TGM2 treatment as antigens (Extended Data Fig. 9g), confirming the essential role of M cells and M cell-derived TGM2 in T cell activation. Notably, compared with M cells, IFNγ-induced MHC-II⁺ enterocytes exhibited significantly lower efficiency in activating reporter T cells in organoid–T cell co-culture assays with TGM2-treated gliadin proteins as antigens (Extended Data Fig. 9h). These observations implied that human M cells directly take up wheat-derived gliadin proteins from the gut lumen and process these into CeD-associated peptides for HLA-DQ2.5 loading, to subsequently activate T cells.

To generalize our findings on M cell-mediated antigen presentation, we sought another defined antigen-based approach. We used a previously described⁴⁶ human-specific antigen cocktail, presented by MHC-II molecules. Adding this antigen cocktail in co-cultured M cells and autologous primary naive CD4⁺ T cells resulted in MHC-II-restricted, antigen-dependent naive CD4⁺ T cell activation by M cells (Extended Data Fig. 10a). In addition to the antigen cocktail, we incorporated CytoStim, an antibody-based reagent that was previously used to demonstrate antigen-presenting function in human HSPCs⁴⁶. This approach reveals the intrinsic capacity to trigger activation and proliferation in a system that bypasses the constraints of peptide specificity and donor matching. Flow cytometry analysis of the co-cultured naive T cells demonstrated robust upregulation of CD69 and significant T cell proliferation, both occurring in an MHC-II-restricted, CytoStim-dependent manner (Extended Data Fig. 10b). Notably, the activation levels (as indicated by CD69 expression levels) of naive CD4⁺ T cells induced by CytoStim are higher than those induced by the antigen cocktail (Extended Data Fig. 10a). Although we observed a significant increase in CD69⁺ cells with the antigen cocktail, these CD69⁺ cells probably included T cells with suboptimal activation (Extended Data Fig. 10a). Together, these results provided functional evidence that M cell-mediated antigen presentation can induce naive T cell activation and proliferation.

Effective activation of naive T cells by professional antigen-presenting cells requires co-stimulatory molecules, particularly CD80 and CD86. Although the expression levels remained modest, M cell organoids showed increased expression of these co-stimulatory molecules compared with normal intestinal organoids (Extended Data Fig. 10c). Notably, these genes could be significantly induced in M cells by immunological stimuli, such as IFNγ (Extended Data Fig. 10d), similar to what has been reported in dendritic cells, suggesting that M cells may exhibit dendritic cell-like co-stimulatory potential. By contrast, IFNγ treatment only upregulated CD40 expression but did not induce the expression of CD80 and CD86 in normal intestinal organoids (Extended Data Fig. 10d). In line with these gene expression results, IFNγ-induced MHC-II⁺ enterocytes, when using CytoStim for co-culture assays, exhibited significantly lower efficiency than MHC-II⁺ M cells in inducing naive T cell proliferation (Extended Data Fig. 10e), again demonstrating the superior capacity of M cells to promote CD4⁺ T cell activation.

Discussion

Our findings reveal that human M cells share substantial similarities with dendritic cells and function as professional, yet non-haematopoietic, antigen-presenting cells. Mouse M cells do not express MHC-II. Thus, M cell-mediated antigen presentation may be unique to the human gut. Notably, the conventional role of M cells in inducing gut immune responses through transcytosis is well supported by studies in mouse and human systems. MHC-II-expressing M cells differ significantly from these previously described MHC-II⁺ enterocytes regarding their IFNγ independency and capacity to provide the full spectrum of signals required for efficient CD4⁺ T cell activation. Given the limited microbial exposure and low levels of IFNγ in the fetal and neonatal gut, M cells may act as early antigen-presenting cells in intestinal epithelium via constitutive MHC-II expression. This interaction could contribute to early mucosal immune education during intestinal development. M cells reside uniquely in gut-associated lymphoid tissues, where the most active epithelium–immune interactions occur. By contrast, enterocytes are not associated with lymphoid structures and, with their dense brush borders, are unlikely to sample antigens but rather are specialized to transport small sugars, amino acids, short peptides and lipids.

A previous study has reported that IFNγ-stimulated enterocytes can present CeD peptides for CD4⁺ T cell activation. However, the T cells and IFNγ-stimulated enterocytes involved in the co-culture assay are derived from humanized mouse models, and the transwell-based co-culture system does not support direct epithelial–immune cell interactions⁴⁷. M cells express TGM2 at much higher levels than enterocytes. Rahmani et al.⁴⁷ only use deamidated gliadin in their co-culture assays, which bypasses TGM2 requirement. In another recently published 2D CeD organoid model⁴⁸, Santos et al. also apply the modified gliadin peptide, rather than native gliadin protein. Of note, their organoid model does not contain M cells. We analysed the expression of 58 CeD-associated gene alleles identified by a genome-wide association study of 4,533 individuals with CeD⁴⁹. Eight of these genes—all of which are known to be expressed by immune cells—are highly expressed in human M cells. Among these eight genes, SOCS1, CLEC16A, ETS1 and RUNX3 are the most prominent in the genome-wide association study (Extended Data Fig. 10f). Furthermore, IL-7 and IL-15, whose receptors are highly expressed by M cells (Extended Data Fig. 4k), have been previously implicated in the pathogenesis of CeD^48,50. Future studies on CeD may investigate the potential roles of these genes in M cells. It is also unclear whether and how naive T cells would migrate to Peyer’s patches without previous priming, which typically occurs in gut-draining lymph nodes. One possibility is that cDCs initially prime the T cells, after which M cells further stimulate them to acquire pathogenic effector functions. Although the organoid platform offers experimental advantages over in vivo studies, it remains a reductionist system and the contribution of human M cells to gliadin-specific T cell activation in CeD, as experimentally tested in organoid system, awaits in vivo validation. Another limitation is that TGM2 was detectable in the culture medium at very low level. We cannot definitively demonstrate that gluten peptide deamidation occurs within M cells. Our findings propose a role of M cell-mediated antigen presentation in modulating intestinal immune responses, tolerance and immunological disorders.

Methods

Human specimens and organoid lines

Two terminal ileum tissues (donors H0514 and H0515-2) were obtained for immunofluorescent staining and flow cytometry analysis from Diakonessen Hospital Utrecht, with approval by the Medical Ethical Committee of the hospital and with informed consent of each patient.

The PBMCs from patients with CeD were collected from participants of the CeDNN study. CeDNN was approved by the Medical Ethical Committee of the University Medical Center Groningen, with METc no. 2013/440. Written consent was signed by all participants, their parents or legal representatives for participants aged under 16 for CeDNN. Normal PBMCs were purchased from Sanquin, with a signed material transfer agreement.

A total of five human intestinal organoid lines were established in our laboratory and recruited to this study. Three ileum tissues (donor 12339, 4280N/HLA-DQ2.5 #1 and 4403N/HLA-DQ2.5 #2) were obtained from the Netherlands Cancer Institute (NKI); one ileum tissue (donor N39) and one colon tissue (donor P11N) were obtained from the Diakonessen Hospital Utrecht, all with approval by the Medical Ethical Committee of the respective organizations and with informed consent of each patient.

This study was approved by the Ethical Committee of Hubrecht Institute and was in accordance with the Declaration of Helsinki and according to the Dutch law. The study complied with all relevant guidelines and regulations regarding research involving human participants.

Mice

Wild-type C57BL/6 mice were used in this study. Both male (n = 2) and female (n = 8) mice (more than 8 weeks of age) were included. No specific sample size was chosen, and neither randomization nor blinding were necessary, as no comparison was made. No specific ethical guidance was required. In this study, we only collected the intestinal Peyer’s patch tissues from euthanized mice for flow cytometry analysis and immunofluorescent staining, which was approved by the Institutional Review Board of Hubrecht Institute.

Organoid culture and M cell differentiation

Human intestinal organoid culture

Three-dimensional cultured human intestinal organoids were mixed with BME (R&D Systems, 3536-005-02), seeded as ~5 µl droplets, cultured in optimized culture of human intestinal organoids (OCHIO) medium and passaged weekly through mechanical dissociation as per the method previously described¹⁴.

M cell differentiation

To generate human M cell organoids, a differentiation cocktail including 100 ng ml^−1 RANKL (Peprotech, 310-01 C), 20 ng ml^−1 TNF (Peprotech, 300-01 A) and 0.1 µM retinoic acid (Sigma-Aldrich, R2625) was added into the OCHIO medium three days post-passaging. The differentiation process was typically achieved within 5–8 days.

Apical-out organoids

To release cultured organoids from BME, ice-cold Corning Cell Recovery Solution (Sigma-Aldrich, CLS354253) was used. Following the release, M cell organoids underwent a polarity change through an overnight suspension culture in M cell medium with 3 µM CHIR99021 (Tocris, 4423). Gliadin proteins or 33-mer peptides were then added for 5 h prior to immunostaining.

Two-dimensional ALI culture

Human intestinal organoids cultured in OCHIO medium were dissociated using 1 ml TripLE (TrypLE Express Enzyme, Thermo Fisher, 12605010) at 37 °C for 6–8 min, followed by gentle pipetting (20 times) to ensure thorough dissociation. The resulting cell suspension was passed through a 40-µm cell strainer and washed twice with 15 ml wash medium. The wash medium consisted of adDMEM/F12 (advanced Dulbecco’s Modified Eagle’s Medium/F12, Gibco, 12634028) supplemented with 100 U ml^−1 penicillin/streptomycin (Gibco, 15140122), 10 mM HEPES (Gibco, 15630056) and 1× Glutamax (Gibco, 35050038). Post-wash, cells were resuspended with OCHIO medium and seeded into the transwell inserts (Greiner, 662641) pre-coated with 10% BME. OCHIO medium was added to the apical (top) and basolateral (bottom) chambers, and cells were cultured for two days. After this period, unattached cells were removed, and the remaining attached cells were cultured for an additional two days until complete confluency was achieved. To induce M cell differentiation, the OCHIO medium was removed from both chambers, and M cell differentiation medium was added only to the bottom chamber to establish an ALI system. M cell differentiation was carried out for six to eight days.

Mouse intestinal organoid culture and differentiation

The protocol is similar to the human organoid cell culture and differentiation protocol with only minor modification. Instead of OCHIO medium, we used ENR medium for mouse organoid culture⁵¹. ENR medium consisted of adDMEM/F12 supplemented with 100 U ml^−1 penicillin/streptomycin, 10 mM HEPES, 1× Glutamax, 1× B-27 supplement (Thermo Fisher, 12587010), 2.5 mM N-acetyl-l-cysteine (Sigma-Aldrich, A9165), 0.25% (v/v) recombinant Noggin (U-Protein Express BV, Custom order), 50 ng ml^−1 EGF (Epidermal Growth Factor, Peprotech, AF-100-15), 10 mM NIC (Nicotinamide, Sigma-Aldrich, N0636) and 0.5% (v/v) R-spondin3 (U-Protein Express BV, Custom order).

Other immune factors and molecules used in this study

LTB (Lymphotoxin α2/β1, R&D system, 679-TX-010/CF) was tested at 100 ng ml^−1, the same concentration as previously reported¹⁶. Cytokines were all from Peprotech with the following catalogue numbers: IL-2, 200-02; IL-3, 200-03; IL-4, 200-04; IL-5, 200-05; IL-7, 200-07; IL-15, 200-15; IFNα, 300-02AA; IFNγ, 300-02; MIF, 300-69; CX3CL1, 300-31; CSF2, 300-03. In the cytokine screening experiment shown in Extended Data Fig. 4l,m, interleukins, interferons, CX3CL1, MIF and CSF2 were used at 10 ng ml^−1. Lipopolysaccharide (Sigma-Aldrich, L4391) was used at 100 ng ml^−1. The concentrations of these immune factors were previously tested by us in organoid systems^14,52, which are effective without significantly affecting organoid viability. Effect of CSF2 on M cell differentiation was further tested at higher concentrations (20 ng ml^−1 and 50 ng ml^−1) in organoids derived from unrelated donors. CADD522 (Selleck, S0790) was used at different titrated concentrations from 1 nM to 1 µM, and 100 nM of CADD522 was further used for validation in organoids derived from unrelated donors. CSRM617 (Sigma-Aldrich, SRL2608) was used at 50 µM. The TGM2 inhibitor ZED1227 (MedChemExpress, HY-19359) was used at 200 nM in co-culture assays and added 2 h before the peptide treatment.

Genetically modified organoids generated by CRISPR

Preparation of cells for electroporation: organoids were cultured in stem cell expansion medium⁵³ for 4–5 days and then dissociated into small cell clumps with 1 ml TripLE at 37 °C for 4 min, followed by gentle pipetting (20 times). The dissociated cell suspension was filtered through a 40-µm cell strainer, washed twice with cold Opti-MEM (Thermo Fisher, 31985070) and resuspended in Opti-MEM for electroporation. The expansion medium consisted of adDMEM/F12 supplemented with 100 U ml^−1 penicillin/streptomycin, 10 mM HEPES, 1× Glutamax, 1× B-27 supplement, 1.25 mM N-acetyl-l-cysteine, 10 mM NIC, 1% (v/v) recombinant Noggin, 0.5 nM WNT surrogate (U-Protein Express BV, Custom order), 50 ng ml^−1 EGF, 0.5 µM A83-01 (Tocris, 2939), 20% (v/v) RSPO1, 1 µM SB202190 (Sigma-Aldrich, S7067) and 1 µM PGE2 (Prostaglandin E2, Tocris, 2296).

sgRNAs targeting different genes were cloned into pSPgRNA vector (Addgene plasmid #47108) according to the previously described protocol⁵⁴. The sgRNA sequences used in this study are listed in Supplementary Table 1.

SPIB-P2A-tdTomato knock-in reporter organoids⁵² were generated using the CRISPR-HOT method as described^55,56.

SPIB⁵², CIITA, RUNX2 and ICAM2-knockout organoids were generated by CRISPR-mediated C-to-T base-editing⁵⁷. Four plasmids were co-transfected into organoid cells via electroporation: a C-to-T base editor (Addgene plasmid #112100 for NGG or #125615 for NGN), a plasmid encoding a gene-specific knockout sgRNA, and a two-plasmid transposon system⁵⁸ (designed to introduce hygromycin resistance for selection). Three days post-electroporation, live single cells were sorted by FACS, based on the negative staining of DAPI (Sigma-Aldrich, 10236276001). The sorted cells were then cultured in expansion medium supplemented with hygromycin to select successfully transfected cells. Single-cell-derived subclones were subsequently hand-picked and expanded. Successful knockout organoids were confirmed by targeted genotyping via Sanger sequencing.

TGM2-knockout organoids were generated by the conventional CRISPR knockout method⁵⁹. Four plasmids were co-transfected into organoid cells via electroporation: a frame selector plasmid (Addgene plasmid #66940), a plasmid encoding a TGM2-specific knockout sgRNA, and the two-plasmid transposon system. Three days post-electroporation, organoid cells were FACS-sorted based on the mCherry fluorescence. The sorted cells were then cultured in expansion medium supplemented with hygromycin. Single cell-derived subclones were subsequently hand-picked and expanded. Successful knockout organoids were confirmed by targeted genotyping via Sanger sequencing.

Immunostaining and imaging

Before fixation, organoids were released from BME using ice-cold Corning Cell Recovery Solution and primary tissues were dissected to remove the muscle layers. Then, primary tissues or organoids were fixed in 4% formaldehyde solution at 4 °C for 18 h or room temperature for 1 h, respectively, and washed three times with PBS (phosphate-buffered saline) and then embedded into OCT for cryosectioning. For immunofluorescence staining, the cryosections were blocked at room temperature for 1 h, using a blocking buffer containing 10% FBS (fetal bovine serum) in PBS-T (PBS with 0.1% Triton X-100). Primary antibodies, diluted in blocking buffer, were applied and incubated overnight at 4 °C. The sections were subsequently washed three times with PBS-T, followed by incubation with secondary antibodies overnight at 4 °C. After three additional washes with PBS-T, sections were mounted for imaging. For whole-mount immunostaining in transwell inserts, the same fixation and staining protocol were used as described above. However, instead of mounting, the transwell inserts were submerged in PBS-T, put in the glass-bottom well plate, and directly staged on the microscope for imaging. All the images were captured using a Leica SP8 confocal detection system integrated with a Leica DMi8 microscope. Leica LAS X software (v.3.5.7.23225) was used for image processing and export.

Primary antibodies used in this study: mouse anti-human GP2 (MBL, D277-3, clone 3G7-H9, diluted 1:200); mouse anti-human ICAM2 (Thermo Fisher, BMS109BT, clone CBR-IC2/2, diluted 1:200); mouse anti-human HLA-DP/DQ/DR (Thermo Fisher, MA1-25914, clone CR3/43, diluted 1:200); rat anti-mouse MHC-II (I-A/I-E) (Thermo Fisher, 14-5321-82, clone M5/114.15.2, diluted 1:200), UEA-1-Rhodamine (Vector, RL-1062-2, diluted 1:1,000); Phalloidin (Sigma, 65906-10NMOL, diluted 1:1,000); DONQ52 antibody (provided by Chugai Pharmaceutical, diluted 1:100).

IHC staining images using TFF2 and MHC-II antibodies in primary human intestinal tissues were obtained from the Human Protein Atlas⁶⁰ (https://proteinatlas.org), details regarding antibodies and patient numbers are included in the related figures.

Flow cytometry analysis

Organoids were released from BME and dissociated into single cells as described above. Post-dissociation, cells were incubated with primary antibody on ice for 30 min, followed by staining with DAPI or PI. Samples were analysed on a BD LSR Fortessa X20 equipped with four lasers (BD Bioscience) or a BD FACSMelody cell sorter equipped with three lasers (BD Bioscience). Cell sorting was performed on a BD FACS Influx cell sorter equipped with five lasers (BD Bioscience). BD FACS Diva software (v.8.0.1) and FlowJo (v.10.8.0) were used for data analysis and display. Sequential gating strategies for flow cytometry can be found in Supplementary Fig. 2.

Preparation of single-cell suspension from primary intestinal tissues: Human terminal ileum tissue was first dissected to remove the muscularis externa, following a previous protocol⁶¹. Regions containing human or mouse Peyer’s patches, which could be clearly identified under dissection microscopy, were enriched, minced into small pieces (~1 mm²), and digested in digestion buffer (adDMEM/F12 supplemented with 100 U ml^−1 penicillin/streptomycin, 10 mM HEPES, 1× GlutaMAX, 1 mg ml^−1 collagenase (Sigma-Aldrich, C9407), and 5 mM EDTA) for 30 min at 37 °C, with vigorous shaking every 5 min. The digested tissues were then filtered through a 100-µm cell strainer, washed twice with wash buffer (adDMEM/F12 supplemented with 100 U ml^−1 penicillin/streptomycin, 10 mM HEPES, 1× GlutaMAX, and 10% FBS), and dissociated into single cells for flow cytometry analysis, similar to the procedure used for organoids, as described above.

Antibodies used in FACS analysis: mouse anti-human GP2-AF488 (MBL, D277-A48, clone 3G7-H9); rat anti-mouse GP2-AF488 (MBL, D278-A48, clone 2F11-C3); mouse anti-human ICAM2-biotin (Thermo Fisher, BMS109BT, clone CBR-IC2/2); mouse anti-human HLA-DP/DQ/DR-BV421 (BD horizon, 564244, clone Tu39); mouse anti-human HLA-DP/DQ/DR-FITC (BD Pharmingen, 562008, clone Tu39); mouse anti-human CD4-AF488 (Biolegend, 317420, clone OKT4); mouse anti-human EpCAM-APC (Biolegend, 369810, clone CO17-1A); rat anti-mouse MHC-II (I-A/I-E)-BV421 (Biolegend, 107631, clone M5/114.15.2); rat anti-mouse CD45-PE (Biolegend, 103106, clone 30-F11); rat anti-mouse EpCAM-APC (Thermo Fisher, 17-5791-80, clone G8.8); mouse anti-human CD69-PE (Biolegend, 310905, clone FN50). Antibodies were diluted 1:100.

Sample preparation and immunogold labelling for electron microscopy imaging

For TEM imaging, organoids were fixed with 1.5% glutaraldehyde in 0.1 M cacodylate buffer for 24 h at 4 °C. Following fixation, the samples were washed with 0.1 M cacodylate buffer and post-fixed with 1% osmium tetroxide in the same buffer containing 1.5% potassium ferricyanide for 1 h in the dark at 4 °C. The samples were then dehydrated in ethanol, infiltrated with Epon resin for 2 days, and subsequently embedded in the same resin. Polymerization of the resin was carried out at 60 °C for 48 h. Ultrathin sections (70 nm) were obtained using a Leica Ultracut UCT ultramicrotome (Leica Microsystems) and mounted on Formvar-coated copper grids. The ultrathin sections were stained with 2% uranyl acetate in water, followed by lead citrate. Images were captured using a Tecnai T12 electron microscope equipped with an Eagle 4kX4k CCD camera (Thermo Fisher Scientific).

To prepare ultrathin sections for immunolabelling, M cell organoids or normal intestinal organoids treated with or without IFNγ (48 h) were chemically fixed with 4% paraformaldehyde in 0.1 M phosphate buffer at 4 °C. After fixation, the organoids were washed with 0.1 M phosphate buffer, embedded in 12% gelatin, and infused in 2.3 M sucrose for 24 h. The mounted gelatin blocks were frozen in liquid nitrogen. Thin sections were prepared using a cryo ultramicrotome (Leica EM Ultracut UC6/FC6, Leica Microsystems). Ultrathin cryosections were collected with 2% methylcellulose in 2.3 M sucrose.

For immunogold labelling, cryosections were incubated on drops of PBS for 60 min at 37 °C, followed by incubation in 50 mM glycine in PBS for 10 min and 1% BSA in PBS for 15 min at room temperature. Sections were then incubated with mouse anti-human CD63 antibody (BD Biosciences, 556019, clone H5C6, diluted 1:20) or mouse anti-human MHC-II antibody (Thermo Fisher, MA1-25914, clone CR3/43, diluted 1:20) diluted in 1% BSA in PBS for 60 min. After five washes with drops of 0.1% BSA in PBS, sections were incubated for 20 min with a bridging antibody, rabbit anti-mouse (DAKO, diluted 1:300). Following five additional washes with drops of 0.1% BSA in PBS, sections were incubated for 20 min with protein A-conjugated, 10 nm diameter colloidal gold particles (CMC Utrecht, diluted 1:50), diluted in 1% BSA in PBS. Sections were then fixed with 1% glutaraldehyde in PBS for 5 min, followed by washes with drops of PBS for 10 min, and two washes with distilled water. Grids were embedded in a thin layer of 1.8% methylcellulose (25 Ctp) containing 0.4% Uranyl acetate. A control for non-specific binding of the colloidal gold-conjugated antibody was included by omitting the primary antibody. Images were captured using the Tecnai T12 electron microscope equipped with the Eagle 4kX4k CCD camera.

Sample preparation and scRNA-seq analysis of M cell organoids

Differentiated M cell organoids were dissociated into single cells for FACS sorting as described above. Single DAPI^− live cells were sorted using a BD FACS Influx cell sorter. Sorted individual single cells were collected in 384 well plates containing ERCC spike-ins (Agilent), reverse transcription primers (Promega) and dNTPs (Promega) as previously described⁶². A total of 768 cells (2 plates) were collected. scRNA-seq was conducted according to the SORT-seq method⁶³. Sequencing libraries were prepared with TruSeq small RNA primers (Illumina) and sequenced on the Illumina NextSeq platform. Sequencing reads were aligned to the human GRCh38 genome to generate the gene expression matrix for subsequent analysis.

The scRNA-seq datasets were loaded into Seurat (v.3.1.4) objects in R Studio (v.3.6.3) for data integration, analysis and visualization⁶⁴, according to the standard pipeline. To create Seurat objects, data were filtered with nFeature_RNA values set to more than 1,000. In total of 375 cells fit this criterion. After creating the Seurat objects, data normalization was performed based on the LogNormalize method, with the scale.factor set to 10,000. Variable features were found by the vst method. Per plate, the two datasets were then anchored together with 2,000 integration features and 50 dimensions to find integration anchors before being clustered in accordance with the standard Seurat pipeline. Dimensional reduction was performed using the UMAP method based on the top 20 principal components. Cell clustering was based on the unsupervised shared-nearest neighbour (SNN) method with the resolution set to 0.7. A total of three cell clusters were identified and annotated, based on the well-known intestinal cell-type markers. Violin plots, heat maps and individual UMAP plots for the given genes were generated by the Seurat toolkit functions VlnPlot, DoHeatmap and DimPlot, respectively.

Re-analysis of primary human intestinal scRNA-seq datasets

Four published datasets were loaded into Seurat objects in R Studio for data integration and visualization. To create the Seurat objects, in GSE119969 (ref. ⁶⁵) and GSE125970 (ref. ⁶⁶) datasets, genes expressed in at least three cells and cells with nFeature_RNA > 200 were selected, in the GSE146799^{ (ref. 62}) dataset, genes expressed in at least three cells and cells with nFeature_RNA > 1,000 were selected. We directly used the Seurat object created by Elmentaite et al.¹⁷, but only selected cells from the adult and paediatric healthy donors in our analysis. To accelerate the speed of analysis, we removed the stem/TA cells and enterocytes from the Elmentaite et al. dataset. The other three datasets still contain sufficient numbers of these cells for analysis. Datasets were then anchored together with 2,000 integration features and 30 dimensions to find integration anchors before being clustered in accordance with the standard Seurat pipeline. Dimensional reduction was performed using UMAP method. Cell clustering was based on the unsupervised SNN method. After cell-type annotation, based on the well-known intestinal cell-type markers, we only used the cells with a clear cellular identity for further analysis and data display. Unsupervised sub-clustering of primary SPIB-expressing M cells was similarly done with the SNN method, by setting the resolution to 2, which identified six sub-clusters. The developmental stage of each sub-cluster was defined on the basis of the expression of well-known M cell markers: GP2 for mature M cells, ICAM2, TNFAIP2 and CCL23 for immature M cells, and SPIB for early M cells.

Integration of human intestinal epithelial and immune cells: the Seurat object of primary intestinal myeloid cells, created by Elmentaite et al.¹⁷, was included and integrated with the epithelial cell types derived from the Seurat object generated above, which includes enterocytes, Paneth cells, tuft cells, goblet cells and genuine M cells (excluding the Brunner’s gland cells). MHC-II⁺ enterocytes were derived from the IBD samples of Elmentaite et al. dataset. The datasets were anchored together based on 2,000 integration features and 30 dimensions. Dimensional reduction was performed using UMAP method and cell clustering was based on the unsupervised SNN method.

For transcriptomic comparisons of immature versus mature M cells or M cells versus MHC-II⁺ enterocytes, the raw counts data were analysed using the DEseq2 (v.1.26.0) R package⁶⁷ with default settings. Normalized counts of MHC-II genes after DEseq2 analysis were shown in the Extended Data Fig. 7c. DEGs were defined by |log₂ (fold change)| ≥2 with P value ≤ 0.001. GO analysis was performed by uploading these DEGs into the Enrichr software⁶⁸ to identify the most relevant cell types.

Re-analysis of scRNA-seq datasets of mouse intestinal tissue- or organoid-derived M cells

Sequencing data²⁹ of mouse intestinal organoids cultured with or without RANKL was retrieved from the GEO database (GSE92332). Analysis was performed in Python (v.3.11.9) using the Scanpy library and the recipe based on Wu et al.⁶⁹. In brief, cells with more than 5% mitochondrial genes were filtered out, and counts per cell were normalized and log₁₀-transformed. Dimensional reduction was performed by finding the 2,000 most highly expressed genes (excluding mitochondrial and ribosomal), scaling and principal components analysis (50 components) and computing of the neighbourhood graph (k-nearest neighbours (KNN) = 200).

Sequencing data of primary mouse intestinal cell types were also retrieved from the GEO database (GSE92332). Raw counts data of M cells and distal enterocytes (cell-type annotation same as the original study) were subjected to DEseq2 analysis with default settings. Normalized counts of MHC-II genes after DEseq2 analysis were shown in the Extended Data Fig. 7c. GO analysis was performed using the mouse M cell marker genes (revealed by the original study) and the Enrichr software to identify the most relevant functional pathways.

T cell binding assay in co-culture

CD4⁺ T cells were activated to induce LFA-1 expression by Dynabeads Human T-Activator CD3/CD28 for T Cell Expansion and Activation (Thermo Fisher, 11131D). M cell organoids were released from BME and co-cultured with activated CD4⁺ T cells for 6 h in suspension culture to facilitate T cell binding. After incubation, the organoids with bound T cells were gently collected into a 15-ml tube containing 13 ml of culture medium. Organoids were allowed to spontaneously settle by gravity for 3–4 min. The supernatant containing unbound T cells was then removed. Organoids were then washed twice using the same method to thoroughly remove the remaining unbound T cells, collected by centrifugation, and dissociated into single cells followed by flow cytometry analysis. The number of CD4⁺ T cells that still bind to organoids was normalized against the number of EpCAM⁺ epithelial cells. To inhibit T cell binding, ICAM2 knockout M cell organoids were used, and an ICAM1 neutralizing antibody⁷⁰ (10 µg ml^−1, Thermo Fisher, 15-0549-82) was further added 1 h prior to the T cells.

Phagocytosis of bacteria particles in 2D cultured M cells

FACS-sorted M cells were seeded on glass-bottom 96-well plates coated with Invasin⁷¹ to facilitate cell attachment. After overnight culture, fluorescent bacteria particles were prepared and added according to the manufacturer’s protocol (Thermo Fisher, P35361), followed by live imaging. For live imaging, plates were staged on the Leica SP8 confocal detection system fitted on a Leica DMi8 microscope (equipped with a CO₂ and temperature control system) to capture time-series images every 5 min for 2 h. Images were analysed and movie was generated by the IMARIS software (v.9.3). The phagocytosis inhibitor cytochalasin B (10 µM, Sigma-Aldrich, C6762) was added 3 h prior to the addition of bacteria particles.

Preparation of 33-mer peptides and wheat-derived gliadin protein

To prepare pepsin/trypsin-treated gliadin protein, 10 g of wheat-derived gliadin (Sigma-Aldrich, G3375) was dissolved in 100 ml of 0.2 N hydrochloric acid, followed by pepsin treatment (Sigma-Aldrich, P6887, 2 mg ml^−1) for 2 h at 37 °C, with stirring. pH of the mixture was adjusted to 7.4 using 2 M sodium hydroxide. Subsequent trypsin treatment (Sigma-Aldrich, T1416) was performed at 37 °C for 4 h, with vigorous stirring. The mixture was then heated at 100 °C for 30 min to stop enzyme activity, and freeze-dried into powder. Further TGM2 (tissue transglutaminase) treatment was performed by incubating 1.08 ml dissolved powder (prepared as 1.11 mg ml^−1 in PBS) with 120 µl TGM2 enzyme (Sigma-Aldrich, T5398, prepared as 1 mg ml^−1 in PBS containing 1 mM CaCl₂) for 2 h at 37 °C. The 33-mer peptides were synthesized by GenScript, with the sequence LQLQPF(PQPELPY)3PQPQPF and LQLQPF(PQPQLPY)3PQPQPF.

Reporter T cell lines with TCR overexpression

The human Jurkat T cell line derivative J76 reporter cell line (J76TPR⁷²), provided M. H. M. Heemskerk, was utilized in this study. The HLA-DQ2.5-glia-α2 specific TCR sequence S16 (TCRα-P2A-TCRβ) was cloned into a plasmid containing the PiggyBac Transposon system. J76TPR cells, which lack endogenous TCRα and TCRβ chains, were transfected with the S16 TCR and transposase plasmids through electroporation. The surface expression of the S16 TCR was evaluated by flow cytometry with the staining of an APC-conjugated anti-human TCR α/β antibody (Biolegend, 306718, clone IP26). Cells expressing the S16 TCR were FACS-sorted and cultured in IMDM (Iscove’s Modified Dulbecco’s Medium, Gibco, 12440061) supplemented with 10% heat-inactivated FBS, 100 μg ml^−1 streptomycin, and 100 U ml^−1 penicillin. The expression of the S16 TCR was routinely checked by flow cytometry every two weeks for use in subsequent experiments.

The human Jurkat T cell line derivative JRT3 cell line, with endogenous TCRα and TCRβ chains knockout, expressing NFAT-RE-Luc2 and HLA-DQ2.5-glia-α2 specific TCR D2, was provided by Chugai Pharmaceutical. These cells were cultured in RPMI1640 (Roswell Park Memorial Institute 1640 medium, Gibco, 11875093) supplemented with 10% heat-inactivated FBS, 100 μg ml^−1 streptomycin, 100 U ml^−1 penicillin, 1 x MEM non-essential amino acids, 1 mM sodium pyruvate, 0.2 mg ml^−1 hygromycin and 100 μg ml^−1 zeocin. Both cell lines were validated on the basis of gliadin-specific T cell activation, as indicated by the induced reporter gene expression, and were assessed and scored negative for mycoplasma contamination.

CeD PBMC isolation

Peripheral blood collected in EDTA tubes was processed for mononuclear cell isolation using BD Vacutainer Mononuclear Cell Preparation Tubes (CPT, BD Biosciences). A volume of 8 ml of whole blood was transferred into a CPT tube, followed by gentle inversion ten times to ensure proper mixing with the separation medium. Immediately prior to centrifugation, the tube was inverted an additional three times to homogenize the contents. Samples were centrifuged at room temperature for 15 min at a relative centrifugal force (RCF) of 1,500–1,800g. Following centrifugation, approximately 50% of the plasma layer was carefully aspirated without disturbing the underlying mononuclear cell layer, which appeared as a diffuse, cloudy interface. The entire cell layer was then collected using a disposable plastic Pasteur pipette and transferred into a sterile 15 ml conical tube. To wash the cells, 10 ml of PBS was added and the tube was gently inverted to mix. The suspension was centrifuged for 10 min at 300g at room temperature. During this step, a freezing medium was prepared, consisting of 80% heat FBS (Thermo Fisher) and 20% DMSO (Sigma-Aldrich). After centrifugation, the supernatant was discarded. The cells were resuspended in 0.5 ml of RPMI 1640 medium (Thermo Fisher) and transferred into a cryovial containing 0.5 ml of freezing medium, with gentle mixing to ensure homogeneity. The cryovials were then placed in a freezing container and stored at −80 °C for controlled-rate freezing.

Co-culture of reporter T cells with organoids and analysis of T cell activation

Organoids cultured in M cell medium or OCHIO medium (with or without IFNγ stimulation for 48 h) for 6–7 days were washed twice with wash medium (adDMEM/F12 supplemented with 100 U ml^−1 penicillin/streptomycin, 10 mM HEPES and 1× Glutamax) and broken into small pieces by mechanical shearing to ensure better antigen uptake. 10 µg ml^−1 of 33-mer peptides or 20 µg ml^−1 of gliadin proteins were added to the organoids for 5 h prior to co-culture. J76TPR cells expressing S16 TCR or JRT3 cells expressing D2 TCR were counted and co-cultured with antigen pre-treated organoids at a 1:1 ratio. The organoid and T cell co-cultures were maintained overnight in a co-culture medium consisting of 48.75% OCHIO medium, 48.75% T cell medium, and 2.5% BME. T cell activation in J76TPR was assessed by flow cytometry. Cells were stained with the APC-conjugated anti-human TCR α/β antibody. Then the eGFP fluorescence from the NFAT-RE-eGFP reporter construct in J76TPR cell line was analysed, as an indicator of T cell activation. T cell activation in JRT3 cells was assessed by measuring the luciferase activity using the Steady-Glo Luciferase Assay System (Promega, E2510) according to the manufacturer’s protocol. The luminescence intensity of each well was recorded for 1,000 ms using the Spark multimode microplate reader (TECAN Life Sciences). The induction of luciferase activity was quantified by calculating the ratio of the difference between the luminescence counts per second (cps) of the target wells and the average luminescence cps of the cell-free wells to the difference between the average luminescence cps of the T cell-only wells and the average luminescence cps of the cell-free wells.

Co-culture of primary human T cells with M cells and analysis of T cell activation

Organoids cultured in M cell medium or OCHIO medium (with or without IFNγ stimulation for 48 h) for 6–7 days were collected as a single-cell suspension as described above. Cells were then stained with antibodies and FACS-sorted into MHC-II⁺ and MHC-II^− populations for the following co-culture assays. All experiments were performed in flat bottom 96-well plates (Greiner, 655180) pre-coated overnight at 4 °C with 1% BME in PBS.

For antigen-dependent activation of autologous T cells, 20,000 FACS-sorted MHC-II⁺ and MHC-II^− organoid cells were plated and incubated overnight with a human MHC-II-restricted peptide pool (JPT Peptide Technologies, PM-CEFT-MHC-II-1; 5 μg ml^−1). The following day, 100,000 naive CD4⁺ T cells isolated from autologous PBMCs using MojoSort Human naive CD4⁺ T cell Isolation Kit (Biolegend, 480041) were added to the co-culture. After 24 h, T cell activation was analysed by flow cytometry following staining with DAPI, anti-CD4-AF488 (Biolegend, 317420), and anti-CD69-PE (Biolegend, 310905) FACS antibodies. The setting of the CD69⁺ gate was based on the negative controls.

For antigen-independent stimulation, FACS-sorted MHC-II⁺ and MHC-II^− cells were plated (20,000 per well) and co-cultured 3 h later with 100,000 naive CD4⁺ T cells from unrelated healthy donors. CytoStim (Miltenyi Biotec, 130-092-173) was added following the manufacturer’s protocol. After 24 h, CD69 upregulation was assessed as described above. For proliferation analysis, naive CD4⁺ T cells were pre-labelled with Cell Proliferation Dye eFluor 670 (Thermo Fisher, 65-0840-85) and analysed by flow cytometry after five days, following staining with anti-CD4-AF488 antibody.

To assess gluten-specific responses, MHC-II⁺ and MHC-II^− cells of organoids derived from HLA-DQ2.5 donor were FACS-sorted and plated as described above. The 33-mer gliadin peptides (10 μg ml^−1) were added for overnight incubation. Total CD4⁺ T cells were isolated from PBMCs of patients with CeD with the HLA-DQ2.5 haplotype, using the MojoSort Human CD4⁺ T cell Isolation Kit (Biolegend, 480009), and co-cultured at 100,000 cells per well. T cell activation was evaluated by flow cytometry following DAPI, CD4 and CD69 staining.

EdU labelling assay

Cultured organoids were incubated with EdU (1 mg ml^−1) for 2 h, followed by 1 h fixation with 4% formaldehyde solution at room temperature. EdU colour development was performed following manufacturer’s protocol (Thermo Fisher, C10640).

TGM2 ELISA

Cell lysates and culture media from M cell organoids were analysed by ELISA following the manufacturer’s protocol (Thermo Fisher, EH462RB). To prepare the samples, M cell organoids were collected and washed three times with PBS to thoroughly remove remaining BME, then lysed with lysis buffer (Thermo Fisher, 87787). Culture medium was pre-concentrated tenfold using Protein Concentrators (Thermo Fisher).

Bulk RNA sequencing of M cell organoids

For each M cell subpopulation, 2,000 DAPI^− live single cells (from one biological replicate) were FACS-sorted into the TRIzol reagent (Thermo Fisher, 15596026). Library preparation and sequencing were performed by Single Cell Discovery following the standard protocol. Sequencing reads were aligned to the human GRCh38 genome. Raw counts data was subjected to DEseq2 R package⁶⁷ and analysed with default settings. DEGs were defined by |log₂ (fold change) | ≥1 with P value ≤ 0.01. GO analysis was performed by uploading the DEGs into the Enrichr web software. GSEA was based on clusterProfiler (v.3.14.3) R package⁷³. log₂ (fold change) was calculated between the mature M cells and enterocytes. The KEGG gene sets (v.7.4) and all GO gene sets (v.7.4) were used as the reference.

RNA extraction and qPCR analysis

Following the manufacturer’s protocol, FACS-sorted cells were subjected to RNA isolation using a NucleoSpin RNA kit (Macherey-Nagel, 740955.50). Reverse transcription reactions were performed using GoScript reverse transcriptase kit (Promega, A5000). cDNA was subjected to qPCR analysis using iQ SYBR Green Supermix (Bio-Rad, 1708887) on a CFX Connect Real-Time PCR machine (Bio-Rad). For gene expression analysis, qPCR was performed with gene-specific qPCR primers. The Ct value of each gene was normalized to the housekeeping gene GAPDH (as the ΔC_t), and fold change between experimental groups was calculated with the 2^{−ΔΔCt} method. All qPCR primers used in this study are listed in Supplementary Table 2.

Quantification and statistics analysis

Data were presented as mean ± s.e.m. Two-tailed Student’s t-test was used when two experimental groups were compared. One-way ANOVA was used when multiple experimental groups were compared. All P values were calculated and all the graphs were generated using Graphpad PRISM (v.7.04). The statistical details for each experiment can be found in the figure legends. In volcano plots shown in (Extended Data Figs. 2c, 3c,e and 6d), DEseq2 used negative-binomial generalized linear models (NB-GLM) and tested coefficients (per gene) with two-sided Wald test to produce P values. In volcano plots shown in (Figs. 2i and 3c and Extended Data Figs. 6f and 7d), P values were computed from one-sided Fisher’s exact test, which is a proportion test that assumes a binomial distribution and independence for probability of any gene belonging to any test.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.