Hybrid identity and distinct methylation profiles of incomplete intestinal metaplasia in the stomach.

Introduction
Gastric intestinal metaplasia (GIM) refers to the transformation of gastric epithelium into an intestinal phenotype, primarily driven by chronic inflammation caused by Helicobacter pylori (HP) infection.1 2 GIM is widely recognised as a precancerous lesion that significantly increases the risk of developing intestinal-type gastric cancer (GC).3 Histologically, GIM is categorised into two subtypes: complete intestinal metaplasia (Com IM) and incomplete IM (Inc IM).4 These subtypes are traditionally distinguished by mucin expression profiles and morphological features, with Com IM resembling the small intestine (SI) and Inc IM resembling the colon.5 However, accumulating evidence suggests that Inc IM may not entirely reflect a colonic phenotype. Rather, it appears to represent a hybrid gastric-intestinal lineage.68 Importantly, many epidemiological studies have consistently shown that Inc IM carries a substantially higher risk for GC development compared with Com IM.9 Despite this clinical significance, relatively few studies have examined the molecular landscape of Inc IM in depth,1013 and none have specifically characterised the cellular phenotypes of Inc IM at single-cell resolution.
Murine models have demonstrated that oncogenic mutations in genes such as Apc, Kras, p53, Smad and Cdh1 in gastric epithelial cells can lead to GC.14 However, these models do not incorporate GIM as an intermediate step in Correa cascade, where intestinal-type GC typically arises through a sequence involving chronic gastritis, GIM, dysplasia and carcinoma. To date, no animal model has demonstrated that GC can arise directly from GIM stem cells. Transgenic mouse models expressing the intestine-specific homeobox genes Cdx1 or Cdx2 have demonstrated the development of classical GIM.15 16 However, in these models, Cdx1 or Cdx2 expression was induced throughout the entire gastric epithelium, a condition that does not accurately reflect the nature of human GIM development. Furthermore, these models did not exhibit progression to GC. This gap has fuelled ongoing debate about whether GIM progresses to GC.17
In this study, we employed an integrative approach combining spatial transcriptomics, organoid modelling and single-cell RNA sequencing (scRNA-seq) to delineate the molecular features of Inc IM and its association with GC. Our findings reveal that Inc IM harbours a hybrid gastric-intestinal lineage identity across both superficial differentiated cells and stem/progenitor cell populations, underpinned by widespread DNA hypermethylation at intergenic regions. Importantly, we demonstrate a strong phenotypic and molecular continuity between Inc IM and GC, providing compelling evidence that Inc IM serves as a true precursor to GC. These results not only define the molecular identity of Inc IM but also underscore its clinical significance, advocating for active surveillance and early therapeutic intervention to mitigate GC risk.

Methods
Detailed methods are provided in the online supplemental material.

Patient and public involvement
Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

Results

Spatially resolved transcriptomic analyses of GIM
To uncover the cellular and molecular phenotypes of GIM, we performed spatial transcriptomic analysis using the 10x Genomics Xenium platform with two tissue microarrays including samples of normal gastric mucosa, normal intestinal mucosa and GIM. The clinicopathological information of GC samples from which gastric tissues are derived is listed in online supplemental table 1. As the commercially available spatial transcriptomic panel lacks sufficient gene coverage to resolve the lineage and function of epithelial cells, we designed a custom panel comprising 293 genes enriched in epithelial cells from the stomach, SI and large intestine (LI) tissues (online supplemental table 2 and figure 1A). 29 cell clusters were initially identified, and epithelial clusters were reannotated into 21 clusters using histological evaluation and differential gene expression (DGE) profiling (online supplemental table 3). Our customised panel successfully resolved lineage-specific epithelial cell types across normal stomach, SI and LI tissues (figure 1B and online supplemental figure 1A,B). Com IM exhibited a cellular composition highly similar to that of normal SI, with only subtle differences in cell proportions, an expanded stem/progenitor population (figure 1B,C). An ambiguous lesion displaying features both Com and Inc IM, referred to as undetermined IM (Un IM), exhibited intestinal stem cell (ISC) clusters localised at the gland base, similar to SI and Com IM, but with distinct progenitor and surface cell populations (figure 1B,C). In contrast, Inc IM presented with a unique cellular architecture composed of distinct clusters annotated as Inc IM progenitor, Inc IM intermediate and Inc IM surface 1–3, while its basal gland regions resembled the antral glands (figure 1B,C and online supplemental figure 1B). Lineage-enriched gene expression analysis clearly demonstrated that Com IM closely resembled SI, while Inc IM exhibited a mixed expression profile incorporating markers from the stomach, SI and LI (figure 1D). Uniform Manifold Approximation and Projection (UMAP) further confirmed the intermediate state of Inc IM between gastric and intestinal lineages (figure 1E,F).

Early development of Inc IM within inflamed gastric mucosa
Notably, high-resolution spatial profiling enabled the identification of several discrete metaplastic foci within chronically inflamed gastric mucosa. One lesion located at the junction of the antral isthmus and basal gland regions was composed predominantly of Inc IM progenitor cells, indicative of early-stage Inc IM (figure 2A,B). A second lesion, also situated at a similar location, was enriched for intermediate enterocytes and intestinal progenitor cells, suggestive of early-stage Com IM (figure 2C,D). UMAP reveals that early Com IM, unlike Inc IM, appears disconnected from antral cells, suggesting a pronounced transition towards an intestinal phenotype in Com IM. Given that deep basal gland cells serve as stem cells in the human antrum,18 these histo-spatial findings implicate that GIM originates from the basal gland stem cells and expands via progenitor cell proliferation at the isthmus. In another case, Com IM glands were embedded within Inc IM-dominant mucosa, populated primarily by ISC and progenitor cells with Inc IM progenitors interspersed throughout (figure 2E,F). This cellular arrangement raises the possibility of bidirectional plasticity or interconversion between Com IM and Inc IM stem/progenitor populations. Transcriptomic profiling of early Inc IM legion revealed upregulation of intestine-enriched genes such as DMBT1, REG4, CDX2 and REG1B alongside downregulation of stomach-specific genes including TFF2, CLU, AXIN2 and GKN2 compared with adjacent gastric isthmus cells (figure 2H–J). Morphologically, the lesion appeared undifferentiated, with a thickened apical membrane (figure 2K). Among the highly upregulated genes, DMBT1 was further validated at the protein level along with the proliferation and progenitor markers, STMN1 and Ki-67 (figure 2L), consistent with previous findings where DBMT1 was suggested as an early GIM marker since its expression precedes the appearance of typical GIM morphology.19 To gain further insights into the developmental trajectory of GIM, we conducted trajectory and pseudotime analyses after excluding SI and LI samples from the cohort (figure 2M,O). These analyses revealed an overlapping region in which antral basal cells and Inc IM basal cell were closely located (figure 2M–O), suggesting that antral basal gland cells have the potential to give rise not only to gastric lineage cells but also to cells of the Inc IM lineage.

Mixed lineage commitment of Inc IM
To dissect the lineage complexity of GIM, we focused on two major epithelial compartments: surface-differentiated cells and stem/progenitor cells. Morphologically, surface cells in Com IM closely resemble those of the SI, whereas surface cells in Inc IM and Un IM exhibit a heterogeneous phenotype with mixed features of stomach, SI and LI (figure 3A). UMAP demonstrated that surface epithelial cells from the stomach, SI and LI formed distinct clusters, with Com IM clustering primarily with SI, while Inc IM and Un IM were positioned between the stomach and SI clusters (figure 3B). DGE analysis supported this observation, revealing coexpression of lineage-specific markers from the stomach, SI and LI in Inc IM (figure 3C and online supplemental figure 2A). Feature plots of representative lineage markers highlighted the mixed identity of Inc IM surface cells (figure 3D). While Inc IM expressed fewer SI markers overall compared with Com IM, REG4 was more highly expressed in Inc IM than in Com IM (figure 3C and online supplemental figure 2A). This increased expression was due to REG4 being restricted to goblet cells in the SI and Com IM, whereas in Inc IM, it was broadly expressed across most surface cells (online supplemental figure 2B). Genes involved in digestion and metabolism were expressed at comparable levels in Com IM and Inc IM. However, genes responsible for nutrient transport were significantly lower in Inc IM (figure 3E and online supplemental figure 2C), indicating Inc IM represents a functionally immature intestinal phenotype. Representative Xenium images show surface cells in Inc IM co-expressing both gastric and intestinal markers (online supplemental figure 3A). This hybrid identity was further confirmed by co-immunostaining for CLDN18 and FABP2, showing clear colocalisation of gastric and intestinal markers at the cell surface (online supplemental figure 3B,C).
Given that metaplasia entails phenotypic reprogramming initiated from stem or progenitor cells that clonally remodel entire glandular units,20 we next investigated the transcriptomic features of these cells (figure 3F). UMAP revealed a similar pattern to surface cells: Com IM clustered with SI, while Inc and Un IM localised between stomach and SI (figure 3G). Consistent with the findings from surface differentiated cells, stem/progenitor cells in Inc IM coexpressed markers from multiple lineages (figure 3H,I). These included the stomach-enriched transcription factor SOX2, the SI-associated markers DMBT1 and OLFM4, and LEFTY1, a gene enriched in the LI (figure 3J). Representative Xenium images display stem/progenitor cells in Inc IM coexpressing multiple lineage markers (online supplemental figure 3D). DMBT1 expression was validated at the protein level in both Com IM and Inc IM (online supplemental figure 3E). Interestingly, LEFTY1 expression was observed specifically in LI and Inc IM stem/progenitor populations, but not in Com IM or normal gastric tissue (onlinesupplemental figure 3F,G table 4). Histologically, Inc IM retains deep antral-type gland cells in the basal compartment (online supplemental figure 4A). Although assigned to the same clusters and sharing specific gastric and stem cell markers (online supplemental figure 4B–D), basal Inc IM glands were distinct from antral glands on UMAP (online supplemental figure 4B) and were characterised by high expression of OLFM4 and LEFTY1, and reduced expression of TFF2 (online supplemental figure 4C–F), suggesting that they are more likely of Inc IM origin. Additionally, representative images illustrating the mixed-lineage phenotype of Inc IM are presented in online supplemental figure 5. In summary, our results reveal that Com IM is transcriptionally and phenotypically analogous to SI. In contrast, Inc IM harbours distinct epithelial populations with transcriptional profiles indicative of mixed gastric, SI and LI lineages, both in surface and stem/progenitor compartments.

Generation of subtype-specific GIM organoids and their epigenetic landscape
Metaplasia is believed to emerge through transcriptional reprogramming that alters cellular and lineage identity, a process largely orchestrated by epigenetic modifications. To elucidate the epigenetic landscape underpinning each GIM subtype, we first established gastric organoid lines from gastrectomy specimens (figure 4A). From these primary organoids, we selected three representative organoid types: organoids derived from antral mucosa without IM, Com IM-predominant mucosa and Inc IM-predominant mucosa (figure 4B and online supplemental figure 6A–C). To obtain subtype-specific organoid lines, single spheres were isolated from each pool and subjected to serial passaging (figure 4C and online supplemental figure 6D). At protein levels, SOX2 and CDX2 serve as lineage-specific markers for gastric and intestinal epithelium, respectively.21 We confirmed their expression patterns in both tissue array samples (online supplemental figure 6E) and subtype-specific organoid lines (figure 4D). To further validate the lineage fidelity of each organoid type, we assessed the expression of stomach-enriched and intestine-enriched genes by RT-PCR before and after bone morphogenic protein (BMP) stimulation, a key regulator of gastrointestinal differentiation.22 Notably, BMP-induced upregulation of GKN1 and GKN2 was the most consistent hallmark of non-metaplastic gastric organoids, with expression levels increasing by several hundred-fold to thousand-fold (figure 4E). Expression of TFF2 and SOX2 remained higher in antral organoids (online supplemental figure 6F). Among the intestinal markers, ANPEP and ALDOB were most prominently induced by BMP stimulation in Com IM organoids (figure 4E), while CDX1, CDX2 and REG4 were preferentially expressed in both Inc IM and Com IM (onlinesupplemental figure 6F table 5). Previously, we reported that combined treatment with a MEK inhibitor and Pyrvinium selectively induced cell death in dysplastic human gastric organoids while arresting the growth of non-dysplastic lines,23 in which we did not discriminate GIM subtypes in non-dysplastic organoids. Here, we evaluated the same drug combination in Com IM and Inc IM organoids, observing comparable growth arrest or mild suppression in both GIM subtypes as well as in antral organoids (online supplemental figure 7A–D).
Next, we performed genome-wide DNA methylation profiling on organoids. Unsupervised hierarchical clustering demonstrated that Inc IM organoids aligned closely with antral organoids, while Com IM clustered with SI and LI counterparts (figure 4F). Inc IM organoids demonstrated extensive DNA hypermethylation, particularly in intergenic regions, a feature not observed in Com IM, which exhibited modest global methylation changes (figure 4G). Overall, Inc IM showed significantly higher DNA methylation levels than Com IM (online supplemental figure 8A), reaching levels comparable to those observed in SI and LI (online supplemental figure 8B–H). Locus-specific analysis further revealed that stomach-enriched loci such as GKN2 and CLDN18 remained hypomethylated in both the antrum and Inc IM (figure 4H and online supplemental figure 8I), while intestine-enriched loci including OLFM4 and FABP2 were hypermethylated in Inc IM but not in Com IM, SI or LI (figure 4I and online supplemental figure 8J). These patterns are consistent with transcriptomic findings and highlight the retained gastric identity of Inc IM despite its partial intestinalisation. We next examined the methylation status of known tumour suppressor genes (TSGs), often aberrantly methylated in GC.24 Intriguingly, Com IM organoids, but not Inc IM, exhibited elevated promoter methylation across a majority of these TSGs (figure 4I–K). Several genes demonstrated high methylation levels uniquely in Com IM (figure 4J), while others displayed similar methylation patterns in both Com IM and intestinal tissues, suggesting that promoter methylation of specific TSGs may represent a feature of intestinal lineage reprogramming (figure 4K). To further characterise chromatin dynamics, we conducted Assay for Transposase-Accessible Chromatin with sequencing on the organoid lines. Inc IM clustered most closely with antral organoids, whereas Com IM showed intermediate features and clustered with both antrum and Inc IM (figure 4L). Unlike its methylation profile, Com IM did not cluster with SI or LI, suggesting that chromatin accessibility changes may lag behind during metaplastic transformation. Together, Inc IM exhibits global intergenic hypermethylation, while retaining gastric epigenetic features. Conversely, Com IM adopts a more intestinal-like methylation profile along with promoter hypermethylation in TSGs.

Single-cell transcriptomic profiling of GIM organoids on BMP-induced differentiation
To further examine the lineage identity of GIM and to evaluate how closely subtype-specific gastric organoids mimic their in vivo counterparts, we performed scRNA-seq before and after BMP stimulation. Following BMP treatment, all organoid types exhibited characteristic thickening of the epithelial spheres and structural crowding, indicative of phenotypic and cellular transitions (figure 5A,B). UMAP projection and DGE analyses revealed lineage relationships that somehow differed from those inferred via spatial transcriptomic analysis. With broader transcriptomic coverage, Inc IM appeared more closely related to the antrum and despite strong histological resemblance and similarities in intestine-specific gene expression and methylation profiles, Com IM did not overlap with SI (figure 5C,D). This suggests that the phenotypic conversion observed in metaplasia does not require global transcriptional reprogramming but may instead be driven by selective regulation of lineage-specific genes.
Com IM organoids responded robustly to BMP stimulation, displaying significant upregulation of intestine-enriched genes (figure 5E and online supplemental figure 9A–C). In contrast, Inc IM organoids exhibited a modest increase in stomach-specific genes and a relatively attenuated induction of intestine-specific genes (figure 5E,F). Among the gastric markers, GKN1 and GKN2 were markedly upregulated, while intestinal markers including ALDOB, CDX2, CDH17 and KRT20 also showed significant but less pronounced increases (figure 5E,F). This differential response underscores the mixed lineage identity and partial differentiation capacity of Inc IM organoids. Consistent with the previous report,25 BMP signalling activation led to the downregulation of deep antral gland cell-associated genes, MUC6 and AQP5 (figure 5F) and stem/progenitor-associated markers (online supplemental figure 9D–F), indicating a shift away from a progenitor state towards luminal differentiation. Immunostaining confirmed coexpression of gastric (GKN1) and intestinal (REG4) markers in Inc IM organoids (figure 5G), and showed reduced expression of deep antral gland cell markers in BMP-treated antral organoids (figure 5H). Collectively, our data demonstrate that Inc IM maintains a transcriptional profile closely resembling gastric antrum, yet possesses the capacity to coexpress gastric and intestinal lineage markers on differentiation cues.

Cellular phenotypic association of Inc IM with GC
Asian Cancer Research Group defines GC into four molecular subtypes: MSI, microsatellite stable/epithelial–mesenchymal transition (MSS/EMT), MSS/TP53 negative (MSS/TP53⁻) and MSS/TP53 positive (MSS/TP53+). To investigate how Inc IM relates to GC and its molecular subtypes, we constructed an additional tissue microarray incorporating 21 GCs (figure 6A,B and online supplemental figure 10A, B). UMAP illustrated how each GC mapped in relation to normal stomach, intestines and GIM subtypes (figure 6C). Transcriptomic profiling revealed that many GCs coexpressed lineage-specific markers of the stomach, SI and LI (figure 6D), a pattern reminiscent of Inc IM. Projecting individual GCs onto the UMAP further revealed their phenotypic proximity to Inc IM (figure 6E). Of the 21 GCs, 17 shared cell populations with Inc IM, while four (GC1, GC5, GC8 and GC14) appeared unrelated to Inc IM phenotypes. Given the well-documented stem-like properties of GC, we examined the expression of stem/progenitor-related genes across normal mucosa, GIM and GC samples. GC exhibited elevated expression of DMBT1, HMGB2, SOX9, OLFM4 and STMN1 (figure 6F). Notably, DMBT1 was the only gene consistently expressed in both GIM subtypes and GC (figure 6F,G). Under inflammatory conditions, DMBT1 expression was induced in the antrum but not the corpus (figure 6G), potentially explaining the antral predilection of GIM. Protein-level validation confirmed elevated DMBT1 and STMN1 expression in GC samples (figure 6H).
To investigate the phenotypic relevance of Inc IM to GC in depth, we assessed cluster proportions of GC and performed UMAP, trajectory and weight-based similarity analyses. We defined ‘weights’ as the degree of transcriptional similarity between GC cells and epithelial clusters from the antrum and GIM subtypes. Integration of GC samples revealed new cell clusters not present in normal or GIM tissues, which we termed cancer-specific clusters (CSCs). Each tumour exhibited a distinct cellular makeup and varying degrees of association with Inc IM. For example, GC1 (MSI) was dominated by CSC1 (figure 7A–C), while the remaining MSI tumours showed highest similarity to Un IM, antrum and Inc IM, respectively (online supplemental figure 11A–C). GC6 (MSS/EMT) was enriched in ISC/progenitor cell cluster and was most closely aligned with Com IM (figure 7D–F). Conversely, GC5 was dominated by a colon stem/progenitor cell cluster (online supplemental figure 11D–F). GC12 (MSS/TP53⁻) was composed primarily of Inc IM progenitor cells (figure 7G–I). Among EBV-positive tumours (GC8 and GC9), GC8 was predominantly composed of CSC2, whereas GC9 was enriched for Inc IM and ISC/progenitor populations (online supplemental figure 11G–I). GC21 (MSS/TP53+) exhibited a dominant Inc IM basal cell phenotype (figure 7J–L), and most MSS/TP53+ tumours were primarily associated with GIM except GC17, a diffuse-type (signet ring cell) carcinoma (online supplemental figure 11K–N).
Overall, GC samples displayed a higher proportion of Inc IM-associated cell types (figure 7M). Although the cohort size limited statistical power, Inc IM-related cell types were more frequently observed in MSS/TP53− and MSS/TP53+ subtypes compared with MSI and MSS/EMT tumours (figure 7N,O). When categorising cases by dominant weights, 38% (8 of 21) of GCs were most closely associated with Inc IM, followed by antrum (24%), Un IM (19%) and Com IM (19%) (figure 7P). These findings indicate that in addition to Inc IM, both Un IM and Com IM cell types appear to be significantly associated with GCs; collectively, approximately three-quarters of the GCs analysed in this study are linked to GIM. In addition, to evaluate the prognostic significance of cellular phenotypes in GCs, we performed survival analyses using a large GC cohort.26 DGEs for each cell type were derived from spatial transcriptomic data, and the top five DGEs per cell type were used to define signature gene sets (online supplemental table 6). Kaplan-Meier analysis revealed distinct survival patterns associated with specific cellular phenotypes (figure 7Q). Interestingly, Inc IM-related cell types did not correlate with poor prognosis. In particular, GCs dominated by Inc IM base cells demonstrated a significantly favourable prognosis. This is in line with the observations that intestinal-type GCs generally have more favourable outcomes.27 28 Collectively, these results provide compelling evidence that a substantial subset of GCs may originate from Inc IM, potentially explaining the higher risk of GC observed in patients with Inc IM compared with those with Com IM.

Discussion
Although scRNA-seq of human gastric tissues has significantly advanced our understanding of the genetic landscape of GIM,12 13 29 30 it remains limited in its ability to distinguish between GIM subtypes. By leveraging spatial transcriptomics, we were able to dissect the molecular characteristics of each GIM subtype independently and identify distinct cell populations unique to Inc IM. This high-resolution approach allowed the analysis of both superficial differentiated cells and basal stem/progenitor populations within metaplastic glands. As expected, Com IM displayed a gene expression profile closely resembling that of the SI, consistent with its more differentiated and uniform intestinal phenotype, while Inc IM exhibited a hybrid transcriptomic signature incorporating features of gastric, SI and LI lineages. Notably, this mixed lineage profile was evident in both differentiated surface cells and basal stem/progenitor compartments of Inc IM, suggesting that the atypical phenotype of Inc IM may originate from multipotent progenitor cells with the capacity to adopt divergent epithelial fates. This parallels observations in Barrett’s oesophagus, where metaplastic glands can follow a gastric line of differentiation giving rise to glandular diversity associated with dysplasia progression.31 Our data offer a compelling explanation for the histological ambiguity and structural heterogeneity characteristic of Inc IM and also support the notion that Inc IM represents a metaplastic state that has failed to fully transition to a terminal intestinal phenotype, instead remaining in an intermediate phase between gastric and intestinal lineages.32
While several studies have reported the generation of GIM organoids,3335 it remains unclear whether they contain heterogeneous populations or which GIM subtypes they represent. To address this limitation, we developed subtype-specific metaplastic organoid lines. The establishment of these well-characterised, subtype-specific organoid lines representing Com IM and Inc IM provides a powerful platform for elucidating the genetic, epigenetic and phenotypic features unique to each metaplastic subtype. Moreover, they offer a valuable system for investigating the plasticity and potential interconversion between Com IM and Inc IM, further advancing our understanding of GIM heterogeneity.
By leveraging subtype-specific GIM organoids, we delineated distinct epigenetic landscapes underlying Com and Inc IM, respectively. DNA hypermethylation is a well-recognised molecular hallmark of GIM.10 Recent genome-wide methylation profiling of highly purified GIM crypts has revealed widespread hypermethylation, identifying an IM-specific epigenetic signature that is also detectable in a subset of GC.28 Our subtype-specific analysis revealed that Inc IM exhibits more extensive DNA hypermethylation than Com IM, primarily attributable to increased methylation at intergenic regions. In contrast, promoter hypermethylation of TSGs was confined to Com IM. Although the global methylation levels in Inc IM approached those of the SI, its overall methylation pattern remained more similar to that of antrum. Conversely, Com IM, despite exhibiting lower absolute methylation levels than both Inc IM and SI, displayed a methylation profile more closely aligned with SI, consistent with its more differentiated intestinal phenotype. These findings suggest that the success of intestinal lineage commitment during metaplasia is not solely dictated by the extent of DNA methylation but rather by locus-specific epigenetic reprogramming. While promoter hypermethylation of TSGs is a well-characterised driver of carcinogenesis, the functional implications of intergenic hypermethylation remain less well defined. Intergenic regions are enriched with enhancers and emerging evidence supports a regulatory role for methylation at these sites in modulating gene expression and influencing neoplastic progression.36 37 Our observation of pronounced intergenic hypermethylation in Inc IM raises the possibility that this epigenetic alteration may contribute to the elevated cancer risk associated with this subtype. However, further mechanistic studies are warranted to elucidate its potential role in gastric carcinogenesis.
Spasmolytic polypeptide-expressing metaplasia (SPEM) lineage cells within a pyloric gland metaplasia, another form of gastric metaplasia that develops in the corpus, are widely recognised as precursors to GC, with their underlying molecular mechanisms being increasingly delineated through murine models.3842 Supporting its relevance in human disease, Kumagai et al recently reported a case in which SPEM, gastric adenoma and gastric adenocarcinoma lesions all shared an identical KRAS mutation, suggesting a clonal origin from a founder SPEM lineage.43 In contrast, the role of GIM as a true precursor to GC remains less well defined. Many studies have sought to elucidate molecular links between GIM and GC.1011 4348 For instance, Gutierrez-Gonzalez et al described a case in which GIM adjacent to a dysplastic gastric lesion harboured identical mutations in the TSGs APC and TP53.47 Furthermore, large-scale cohort analyses by Huang et al identified genetic and epigenetic changes in GIM samples that may aid in stratifying patients by GC risk.10 11 Despite these findings, no study to date has demonstrated a consistent presence of canonical oncogenic mutations in GIM at the population level. This is consistent with the simple view that GIM represents a non-neoplastic, predysplastic condition that precedes the acquisition of significant genetic alterations. Given the limitations of purely genetic analyses in resolving the cellular origin of GC, we propose that phenotypic profiling based on lineage differentiation status may offer a complementary and potentially more informative approach. Through the application of organ-specific gene expression profiling, we identified that a substantial subset of GCs harbours cell populations transcriptionally similar to those of Inc IM. The representation of Inc IM-like cells in these tumours was markedly higher than that of either Com IM or antrum cell types. These findings are supported by recent work from Yue et al, which demonstrated strong transcriptomic concordance between cell populations within GC and those derived from GIM organoids.34 In addition, Hoft et al have identified a metaplastic subtype resembling Inc IM, arising from HP infection and autoimmune gastritis, which shared transcriptional features with GC.49 While not definitive proof of a direct precursor–product relationship, the phenotypic convergence between Inc IM and GC cells suggests that a subset of GCs may originate from Inc IM.
Our study has several limitations that warrant consideration. First, the gene panel used was limited to approximately 300 targets, which may have introduced selection bias during cell clustering and reduced our ability to capture the full transcriptomic complexity of the GIM. Recently released technologies now offer expanded gene coverage, which will likely overcome this constraint in future studies. Second, the Xenium platform relies on an in situ hybridisation-based methodology, which, although highly specific and spatially resolved, presents challenges in detecting highly expressed genes due to optical crowding. This limitation can result in signal spillover, whereby transcripts from highly expressing cells are erroneously detected in neighbouring, non-expressing cells. Consequently, certain rare and specialised cell populations such as Paneth cells, enteroendocrine cells and tuft cells were not reliably identified in our dataset. Third, we included one case classified as an indeterminate form of GIM (‘Un IM’), which exhibited overlapping morphological features of complete and incomplete subtypes. Although not an established pathological category, such ambiguous presentations are commonly encountered in clinical practice and reflect the recognised histological and molecular continuum of GIM. We believe inclusion of this case underscores the need for more objective criteria to define intermediate forms, which should be systematically validated in larger cohorts. Lastly, the number of GC cases included in our study was insufficient to achieve statistical significance for some observed associations with GIM subtypes. Nevertheless, we identified consistent and biologically meaningful trends that warrant further investigation in larger cohorts.
In summary, our findings define Inc IM as a distinct metaplastic phenotype marked by mixed gastric and intestinal lineage characteristics and a unique epigenetic landscape. The frequent presence of Inc IM-like cell populations in GCs strongly supports its identity as a high-risk precursor lesion. The establishment of subtype-specific Inc IM organoids offers a robust and physiologically relevant model for dissecting the molecular mechanism underlying GIM and its progression towards malignancy.

Supplementary material
10.1136/gutjnl-2025-335793online supplemental file 110.1136/gutjnl-2025-335793online supplemental file 210.1136/gutjnl-2025-335793online supplemental file 310.1136/gutjnl-2025-335793online supplemental file 4