Spatial analysis of hereditary diffuse gastric cancer reveals indolent phenotype of signet ring cell precursors

Abstract

Germline CDH1 loss-of-function mutations are causally linked to an increased lifetime risk of diffuse gastric cancer (DGC). Early, multifocal signet ring cell (SRC) lesions are ubiquitous among CDH1 variant carriers, yet only a subset of patients will develop advanced DGC. A multi-omics analysis was performed to establish the molecular phenotype of early SRC lesions and how they differ from advanced DGC using 20 samples from human total gastrectomy specimens of germline CDH1 variant carriers. Spatial transcriptomic analysis demonstrated reduced CDH1 gene expression and increased expression of ECM remodeling in SRC lesions compared to unaffected adjacent gastric epithelium. Single cell RNA sequencing revealed an SRC-enriched signature with markers REG1A, VIM, AQP5, PRR4, MUC6, and AGR2. Importantly, SRC lesions lacked alterations in known drivers of gastric cancer (TP53, ARID1A, KRAS) and activation of associated signal transduction pathways. Advanced DGC demonstrated E-cadherin re-expression, somatic TP53 and ERBB3 mutations, and upregulated CTNNA1, MYC, and MET expression when compared to SRC lesions.

Implications:

The marked differences in genomic and transcriptomic profile of SRC lesions and advanced DGC support the consideration of SRC lesions as precancers in patients with germline CDH1 mutations.

Introduction

Diffuse-type gastric adenocarcinoma is a lethal cancer with an infiltrating phenotype that evades early clinical detection (1,2). Sporadic diffuse-type gastric cancers frequently harbor somatic alterations in the CDH1 gene, which encodes the protein E-cadherin that is central to maintaining epithelial architecture, cell polarity, and cell-cell adhesion (3). Germline loss-of-function mutations in CDH1 increase an individual’s lifetime risk of diffuse gastric cancer (DGC) (4,5). To prevent the development of advanced gastric cancer, individuals with germline pathogenic or likely pathogenic (P/LP) CDH1 variants are advised to undergo prophylactic total gastrectomy as early as age 20 years (6,7). Interestingly, nearly all (80%–100%) gastrectomy explants from individuals undergoing prophylactic surgery harbor clinically occult, multifocal signet ring cell (SRC) lesions (8). Although SRC lesions are ubiquitous in these patients, the prevalence of clinically actionable, advanced gastric cancer in a large cohort of families with germline CDH1 variants was 14% (9). Furthermore, the lifetime risk of advanced gastric cancer in these individuals is estimated to range from 7% to 38%, depending on family history of gastric cancer (9). Thus, only a fraction of SRC lesions is likely to progress to advanced DGC in individuals with germline CDH1 variants.

Clinical and epidemiologic data indicate that SRC lesions may be non-obligate precursors of DGC in CDH1 variant carriers, however, their histopathologic designation as cancer drives current clinical management. Genetically engineered mouse models underscore that loss of CDH1 gene expression in gastric epithelia alone is not sufficient for development of advanced gastric cancer (10,11). However, the molecular alterations that impart SRC lesions with their uniquely indolent phenotype of infrequent progression to advanced cancer in humans are not known. Furthermore, the study of early gastric SRC lesions is challenging because of their microscopic size and subepithelial location in otherwise normal appearing stomachs. We conducted a spatially resolved transcriptomic analysis to address the incongruity between the ever-present SRC lesions described as invasive adenocarcinomas and the incomplete penetrance of advanced DGC among germline CDH1 variant carriers. Herein, we demonstrate that nascent SRC lesions are transcriptionally similar to native gastric epithelium, and that SRC lesions lack oncogenic changes and activation of gene pathways observed in advanced diffuse gastric cancers from individuals with germline CDH1 variants. These molecular data uphold clinical observations of the indolent nature of SRC and suggest a precancer designation for gastric SRC lesions in the setting of germline CDH1 P/LP variants.

Methods

Study cohort

Individuals with germline CDH1 P/LP variants were eligible for this analysis and enrolled in a prospective, natural history study of hereditary gastric cancer (NCT03030404), which was approved by the National Institutes of Health (NIH) Institutional Review Board. Informed written consent was obtained from all participants. All patients received multidisciplinary care at the NIH Clinical Research Center per institutional practice. Patient demographics including sex, race, CDH1 P/LP variant type, age at time of total gastrectomy, and final tumor stage are summarized in Supplementary Table S1. All gastrectomy specimens were examined by expert gastrointestinal pathologists.

Spatial transcriptomics with Nanostring GeoMx

The GeoMx Nanostring^™ platform for digital spatial profiling (DSP) using the whole transcriptome atlas (WTA) panel was used to analyze gastric tissue regions of interest (ROIs) for this study. Eleven gastric epithelial (EP) and 9 SRC lesion ROIs were captured from total gastrectomy samples from four patients with germline CDH1 P/LP variants. Median size of ROIs was 0.13mm² for EP (range 0.075–0.264) with a median cell count of 871 (range 440–1355), and 0.05mm² for SRC lesions (range 0.019–0.144) with a median cell count of 328 (range 154–825). In four patients with advanced DGC found on gastrectomy explants, 12 ROIs were selected from gastric epithelial (EP), 28 ROIs were selected from the superficial lamina propria, and 33 ROIs were selected from the submucosa and deeper. The median size of EP ROIs from DGC samples was 0.18mm² (range 0.105–0.326) with median cell count of 1208 (range 499–2172). The median size of DGC ROIs (superficial and deep combined) was 0.04mm² (range 0.012–0.171) with median cell count of 382 (range 102–1374). The formalin-fixed, paraffin-embedded (FFPE) tissue blocks were cut at 5μm thickness onto APEX treated slides to improve attachment. After incubating at 60°C overnight, deparaffinization, and antigen retrieval was performed on Leica BOND RX autostainer using HIER2 (EDTA-based antigen retrieval solution for 20min) and proteinase K (15min at 37°C, 1μg/ml). RNA probes (NanoString GeoMx WTA Human RNA probes catalog number 121401102) were added for hybridization at 37°C for 16h per protocol. Slides were then washed 2X with stringent wash (50% formamide in 2XSSC for 25min) followed by 2X wash with 2XSSC (2min). Slides were blocked with W buffer for 30min and incubated for 60min at RT with the antibody cocktail containing PanCK-532 RRID:AB_2935724 (1:40), CD45-AF594 RRID:AB_3169472 (1:40), aSMA-AF647(1:200) and Syto 13 (10mM, nuclear stain). Slides were washed with SSC buffer 2X and scanned on the GeoMx DSP machine for selection of ROIs, using FITC/525nm filter to detect nuclei stained with Syto 13 (150msec), Cy3/568 nm channel to detect PanCK (200msec), Texas Red/615 nm to detect CD45 (200msec) and Cy5/666nm to detect aSMA (200msec). Based on tissue morphology, ROIs for EP, SRC, and DGC were selected using polygons, with masking for panCK positive cells. The combined surface area of ROIs was 7699256.57mm² containing 57880 nuclei. Oligonucleotide barcodes from each ROI were UV-photocleaved, collected into 96-well plates, followed by library preparation and sequencing. Extracted mRNA barcodes were prepared and sequenced on a NextSeq2000 instrument using P2 and P3 flow cell kits.

Bioinformatics analysis (Nanostring GeoMx)

Fastq files for each ROI were converted to Digital Count Conversion (DCC) files using NanoString’s GeoMx NGS Pipeline v.2.3.3.10. DCC files. 18677 genes passed the recommended LOQ (level of quantification) of 2.0, which is the default setting for high confidence detection. Targets were further filtered to 10% higher than the LOQ of 2.0 for higher confidence level, therefore the analyzed targets were reduced to 11681 after quality control and count normalization. Unpaired t-tests were performed comparing normalized transcript counts amongst ROIs using GraphPad Prism Version 9.3.1 (GraphPad Software, Inc., San Diego, CA).

Gene set enrichment analysis (GSEA) was performed using version 4.3.2 (Broad Institute of Massachusetts Institute of Technology, https://www.gsea-msigdb.org/gsea/msigdb) utilizing from KEGG, Reactome, and Hallmark pathway databases. Gene sets with absolute normalized enrichment score (NES) greater than 1.5 and false discovery rate (FDR)-adjusted p value (q-value) <0.25 were considered significantly enriched. Default analysis parameters include weighted enrichment statistic, Signal2Noise ranking metric, and inclusion of publicly availably gene sets of 15–500 genes. Leading edge analysis (LEA) was performed within the GSEA software to identify individual gene contributions by aggregating related gene sets based on NES values, nominal p-value, and FDR-adjusted p-value (q-value) <0.25.

Targeted somatic sequencing of tumor specimens

Tumor DNA and RNA were extracted from FFPE tissue sections mounted on slides with macrodissection to enrich for tumor content using the QIAamp DNA FFPE Tissue Kit and RNeasy FFPE Tissue Kit (QIAGEN). 100ng of DNA was used to prepare libraries using the TruSight Oncology 500 Kit (TSO500, Illumina). Amplified pre-enriched libraries were hybridized to probes specific to 523 targeted genes. Enriched libraries were amplified, quantified, normalized, then sequenced as paired-end reads on NextSeq 550DX (Illumina). TSO500 local app was used for alignment, variant calling, and determination of tumor mutational burden (TMB) and microsatellite instability (MSI). Exome RNA sequencing libraries were prepared with 100ng tumor RNA using the Illumina RNA Prep, Tagmentation(L) with Enrichment Kit pairs with the whole exome probe panel. Final enriched libraries were sequenced on NovaSeq 6000 (Illumina). Custom in-house NGS pipeline was used to perform fastq file generation and alignment. Arriba fusion detection tool was used along with STAR aligner for fusion prediction. TSO500 assay plus exome RNA sequencing were designed to detect single nucleotide variations, small insertion/deletion (indel) events, copy number variations from 92 genes, TMB, MSI, and RNA fusions. Resulting vcf files were uploaded to QIAGEN Clinical Insight Interpret for variant filtering, annotation, classification, interpretation, and reporting. Pathogenicity was classified as pathogenic, likely pathogenic, and variants of uncertain significance while actionability was tiered as 1A, 1B, 2C, 2D, and 3 according to ACMG/AMP/CAP guidelines (12).

Histology and immunohistochemistry

CODEX multiplexed immunofluorescence tissue imaging was performed on FFPE gastric cancer tissue from total gastrectomy specimens of ten patients with germline CDH1 P/LP variants (Supplementary Table S1). SRC and DGC lesions were imaged for expression of 25 protein markers selected to interrogate tissue architecture (CD31, CD34, αSMA), cellular phenotypes (CD11c, CD15, CD163, CD79A), activation status (Granzyme B), cell segmentation (β actin, pancytokeratin, and NaKATPase) and genes of interested based on preliminary SRC transcriptomic data (Vimentin, Mucin-1, Claudin 7, E-cadherin) (Supplementary Table S2). Five mm sections were cut onto poly-L-Lysine coated coverslips. After incubating for 60min at 65°C, dewax was performed immersing the sections into xylene (2X 10min), followed by rehydration in ethanol gradient. Antigen retrieval was performed at 100°C for 20min using ER2 (EDTA buffer) in a pressure cooker. After blocking, tissues were incubated in antibody cocktail overnight at 4°C, followed by washes in PBS, fixation in 1.6% paraformaldehyde (10min), and methanol (5min). Cross-linking reagent was applied for 20min to covalently bind antibodies to tissues. Imaging of stained slides was performed on a Keyence BZ-X810 microscope using DAPI, FITC, CY3 and CY5 filters using a Nikon 20X objective with NA of 0.75. Number of tiles was selected to cover the SRC or DGC containing area of the tissue. Multiple layers (z stack n=9–14), separated by 1.5μm, were acquired to ensure every cell in the tissue was in focus.

Acquired raw images were processed using CODEX processor V1.7.0.6 without segmentation. Nuclear staining from cycle 2 was used as reference for focusing. During processing background subtraction, deconvolution, best focus selection (EDF), shading correction were applied. After conversion, individual stitched tiff images were converted to pyramidal tiff using bftools open-source conversion tool. Pyramidal tiff images were loaded into HALO image analysis platform (vs 3.4 IndicaLabs) and fused to create composite images for analysis. Area quantification FL algorithm was used to quantify expression of the different markers on manually selected annotation layers.

Single cell RNA-seq with 10X Chromium

High-viability, single-cell dissociation of fresh gastric tissue from total gastrectomy specimens from three patients (Supplementary Table S1) was collected for single cell RNA sequencing utilizing a protocol previously described (13). Known foci of SRC lesions from previous gastric mappings were harvested and utilized for downstream analysis. scRNA-seq was performed using the Illumina instrument. 10x Genomics Single Cell Chromium 3’mRNA libraries were sequenced with an average yield of ~440 million reads per library (~8.79 billion reads from 20 libraries). Overall sequencing quality was high in mRNA libraries with >95% of bases in barcode and unique molecular identifier (UMI) regions had Q30 (99.99% inferred base call accuracy) or above, whereas at least 91% of bases in the RNA reads had Q30 or above. Mean read count per cell was ~69000 across 20 libraries with average values of ~75% of the reads detected in cells and estimated cell count of 6628 per library.

Bioinformatics analysis with 10X Chromium

Initial processing of the scRNA-seq data involved generating fastq files from the raw RNA sequencing data, removal of the cells with extremely low number of UMI counts and generating raw count matrices using Cell Ranger v6.1.2 run with default parameter settings and the GRCh38-2020-A transcriptome reference. Median gene count (per cell) range was 581–1351 across 20 libraries along with a sequencing saturation range of 47%–74% and 39%–53% of reads mapped confidently to the transcriptome. Remainder of the scRNA-seq analysis was performed using the Seurat 4.1.0 package (14) of R (v4.1.1) including quality-based filtering, normalization of the expression data with the NormalizeData and SCTransform functions, and integration of the 20 samples with gene anchors (15) identified by the FindIntegrationAnchors function ran with the reciprocal PCA option of reduction, followed by clustering using the FindClusters function with the SCT_snn (graph.name) and 0.1 (resolution) options.

Filtering out low-quality cells, such as potential doublets/multiplets and cells with low gene counts resulted in 88282 cells with average of 1054 (SD 765) for the gene counts (nFeature_RNA) and 36% (30%) with respect to percent mitochondrial gene expression (percent.mt). In the filtering step, R package scDblFinder was used with default settings (v1.8.0), and isOutlier function of the R package scuttle (v1.4.0) for removing cells based on nmads (number of median absolute deviations) thresholds of 4 for high mitochondrial content (percent.mt), and 3 and 2.5 for low and high gene and UMI counts, respectively.

Cell type specific markers from the literature (16,17) were used for annotating cells through a mixture of manual annotation and by utilizing AddModuleScore function of Seurat along with clustering results. Differential expression analysis was performed on the log-normalized RNA expression assay using MAST (18), or “Model-based Analysis of Single-cell Transcriptomics,” whereas the enrichment profiles were generated by R packages WebGestaltR(v0.4.5) and enrichR(v3.2). Figures were generated using R packages ggpubr(v0.4.0.999), ggplot2(v3.4.2), and scCustomize(v0.7.0).

Data availability

The single cell RNA sequencing (10X Chromium) and spatial transcriptomics (NanoString GeoMx) data sets have been deposited in raw and processed formats following the Gene Expression Omnibus (GEO) guidelines under the GEO accession numbers GSE281274 and GSE289923, respectively.

RESULTS

Spatial transcriptomic analysis of gastric signet ring cell lesions

We examined early gastric SRC lesions, adjacent unaffected gastric epithelium, and advanced diffuse gastric cancers from patients with germline CDH1 P/LP variants. All patients had germline genetic analyses that showed loss of function frameshift, nonsense, splice site, missense, start-loss, duplication, or deletion mutations in the CDH1 gene (Fig. 1A, Supplementary Table S1). A multi-omics interrogation was applied to surgical specimens obtained from clinically indicated therapeutic total gastrectomy for gastric cancer or prophylactic total gastrectomy operations in patients with germline CDH1 mutations (Fig. 1B). Individuals who underwent gastrectomy for prevention of cancer had grossly normal appearing gastric mucosa on screening endoscopy and pathognomonic multifocal intramucosal SRC lesions located deep to the epithelial basement membrane within the superficial zone of the lamina propria (pT1a gastric cancer) (Fig. 1C) (19). SRC lesions appeared as horizontally dispersed, enlarged, round, mucin-filled cells with eccentrically placed nuclei imparting the characteristic signet ring appearance (Fig. 1D–E). Uninvolved, microscopically normal-appearing adjacent gastric epithelium was analyzed to establish gene expression of at-risk epithelial cells for comparisons. Histologic appearance and immunofluorescence imaging demonstrated nuclear (DAPI) and membranous pancytokeratin staining that supported identification of SRC lesions for digital selection of regions of interest (ROIs) for spatial transcriptomic analysis (Fig. 1F). Digital subtraction of leukocytes and fibroblasts/myocytes positive for CD45 and αSMA, respectively, was performed for precise acquisition of SRC lesion transcripts from selected ROIs in the RNA extraction step, allowing for in situ analysis of a purified population of germline CDH1-deficient SRCs lesions to date. The GeoMx Nanostring^™ platform for digital spatial profiling was applied using the whole transcriptome atlas panel, which resulted in 11681 detected genes out of ~18000 total genes for analysis. Uniform manifold approximation and projection (UMAP) analyses clustered SRC lesions with unaffected gastric EP, and separately from cell clusters derived from advanced DGCs (Fig. 1G).

Figure 1.

(A) Location of germline CDH1 P/LP variants in four patients with pT1aN0 (Stage IA) signet ring cell (SRC) lesions (top arrows) and four patients with pathologically confirmed advanced diffuse gastric cancer (DGC) (bottom arrows) utilized for GeoMx Nanostring^™ digital spatial profiling (DSP).

(B) Multi-omics study overview. Single-cell suspensions of gastric mucosa containing SRC lesions were subjected to single cell RNA sequencing. Co-detection by indexing (CODEX) multiplexed immunofluorescence tissue imaging and GeoMx Nanostring^™ digital spatial profiling was performed on formalin-fixed, paraffin-embedded (FFPE) gastric cancer tissue from total gastrectomy specimens. This figure panel was created with BioRender.com.

(C) Occult SRC lesions on upper endoscopy.

(D, E) H&E staining of an individual with Stage IA disease at (D) 100x magnification, scale bar measures 100 μm and (E) 200x magnification, scale bar measures 50 μm. Arrows indicate SRC lesions.

(F) Masked region of interest (ROI) selection and immunofluorescent imaging of SRC lesions.

(G) UMAP demonstrating transcriptomic clustering of adjacent unaffected gastric epithelium (EP) ROIs from pT1aN0 (n=11) gastrectomy specimens with SRC lesion (n=9) ROIs. Advanced DGC (n=61) ROIs form a distinct cluster.

(H) Volcano plot demonstrating extracellular matrix (ECM)-related genes with increased expression in EP or SRC (p< 0.05 and |log2 fold-change| ≥1).

(I) Differences in expression of select adherens junction genes in SRC ROIs (n = 9) compared to EP ROIs (n = 11).

(J) Heatmap demonstrating enrichment of select genes within the ECM proteoglycans gene set in SRC compared to EP from GSEA.

(K) Multiplex immunofluorescence imaging of Stage IA disease. Panel 1 H&E, panel 2 pancytokeratin in yellow, beta-catenin in red, SMA in blue, panel 3 nuclei in blue, panel 4 pancytokeratin in yellow, Ki67 in purple, panel 5 β-catenin in orange, and panel 6 E-cadherin in blue.

(L) No significant upregulation in expression of classical tumor proliferation markers in SRC compared to EP.

CDH1 expression loss paired with unique gene set enrichment in early gastric SRC

To identify transcriptional programs of nascent SRC, we analyzed differentially expressed genes within SRC lesion ROIs compared to gastric EP ROIs. Collectively, 1.9% (227/11681) of genes were differentially expressed in SRC versus EP (p<0.05, |log₂ fold-change|≥1), with 97 genes (42.7%, 97/227) upregulated in SRC compared to EP. Multiple upregulated differentially expressed genes were related to extracellular matrix (ECM) including COL3A1, BGN, LUM, SPARC, ELN and COL1A1 (Fig. 1H). Since the CDH1 gene product is a critical component of epithelial adherens junctions, we first examined the elements of this cell-cell adhesion complex, which consists of the E-cadherin transmembrane protein that is linked to the actin cytoskeleton via p120-catenin, β-catenin, and α-catenin. Normalized counts of CDH1 and CTNNA1, which encodes alpha-catenin, were significantly decreased in SRC compared to EP (Fig. 1I). Although CTNNB1, which encodes beta-catenin, was not decreased in SRC, a downstream target of β-catenin signaling, CCND1, which encodes Cyclin D1, was downregulated in SRC compared to EP (p=0.0001). RHOA, which encodes Ras homolog family member A GTPase, and for which gain-of-function mutations have been implicated in diffuse gastric cancer, exhibited lower expression in SRC compared to EP (p=0.01) (20).

Gene set enrichment analysis (GSEA) using the Reactome database revealed 58 significantly enriched gene sets in SRC compared to adjacent EP (q-value<0.25). Leading edge analysis (LEA) categorized nearly half (26/58) of the positively enriched pathways related to ECM re-organization (Supplementary Fig. 1A). Additional constitutive gene sets Degradation of the Extracellular Matrix (NES=2.09, q-value=0.001), Integrin Cell Surface Interactions (NES 2.04, q-value=0.001), ECM Proteoglycans (NES 2.18, q-value<0.0001), and Non-integrin Membrane ECM Interactions (NES 2.02, q-value=0.002) were enriched in SRC compared to EP (Fig. 1J). These data support that SRC migration into the subepithelial space coincides with loss of CDH1 expression and is facilitated by ECM remodeling.

Multiplex immunofluorescence imaging (CODEX) using 25 markers of gastric tissue bearing SRC was utilized to quantify protein expression (Fig. 1K, Supplementary Table S2). The area of E-cadherin expressing cells was markedly reduced in SRC compared to gastric EP (E-cadherin positive % area EP 38.1 vs SRC 20.9, p<0.001), whereas percent area expressing cytoplasm marker pancytokeratin was retained. Although there was no difference in CTNNB1 transcript levels in SRC compared to adjacent EP, β-catenin expression was markedly reduced in SRC lesions as detected by CODEX imaging (Fig. 1K). These results support previous observations that loss of E-cadherin expression is an inciting event that results in loss of cell polarity and is necessary for early SRC formation (21).

Early gastric SRC lack oncogenic activation

We have previously shown that SRC demonstrate clinically indolent and variable behavior in individuals with germline CDH1 variants who are under extended periods of cancer surveillance (22,23). Therefore, we hypothesized that alterations in classical oncogenic drivers observed in advanced DGC would be absent in SRC lesions (22). We observed no differences in expression of the following tumor suppressor genes and oncogenes in SRC lesions compared to EP: TP53 (p=0.78), SMAD4 (p=0.53), APC2 (p=0.42), and ARID1A (p=0.94) (Fig. 1L). KRAS, often activated in advanced gastric cancer, had significantly lower normalized counts in SRC compared to EP (p<0.001) (24). Additionally, VIM expression, associated with epithelial-mesenchymal transition, was elevated in SRC compared to EP (p<0.001) (25,26). To further explore potential changes in classical tumor suppressors and oncogenes, we performed somatic DNA sequencing via macrodissection of SRC lesions from gastrectomy explants, which revealed no TP53, KRAS, MYC, or ARID1A pathogenic gene alterations on Next Generation Sequencing (NGS). This suggests that early SRC lesions possess altered transcriptional programs (e.g., cell-cell adhesion and ECM remodeling) that provide the capacity to migrate yet lack genomic and transcriptomic hallmarks of advanced gastric cancers (27,28).

Single cell RNA sequencing reveals an SRC-enriched signature

To better understand gastric SRC lesions, we performed single cell RNA sequencing (scRNA-seq) of normal-appearing mucosa that contained occult SRC lesions. Because SRC lesions are not visible clinically, we leveraged our previously described method of systematic endoscopic gastric mapping of the fundus, body, incisura, and antrum for cancer surveillance in individuals with germline CDH1 P/LP variants (22,29,30). The anatomic location of previously biopsy-confirmed SRC was utilized in three patients prior to total gastrectomy (Supplementary Table S1). Macroscopically normal-appearing tissue was taken from these regions of the histologically-confirmed SRC lesions and processed to obtain single cell suspensions of gastric mucosa (13). Following quality control-based filtering of cells, 88282 cells with average counts of 1054 genes per cell (Fig. 2A) and 3940 transcripts per cell (Fig. 2B) were retained for the downstream scRNA-seq analysis. After integration of scRNA-seq data, we used graph-based clustering that identified 11 non-SRC clusters, each of which was assigned to a cell type using markers from the literature (17,31), including the Tabula Sapiens human cell atlas (See Methods, Fig. 2C–D). The resulting cell type annotations were utilized to make pairwise comparisons between the SRC cluster and the other gastric cell types (Supplementary Table S3). Notable immune cell markers by cell type included MS4A1 for B cells, CD8A for T cells, TNFRSF17 for plasma cells, and KLRD1 for NK cells. Parenchymal markers by cell type included PLVAP for endothelium, ATP4B, GKN1, and KCNE2 for parietal cells, PGA3 for chief cells, DCN, FN1, NOTCH3, and RGS5 for the cluster composed of stem cells, pericytes, and fibroblasts, CHGA for gastrin cells, KIT for mast cells, and GKN1, TFF1, and MUC5AC for pit cells.

Figure 2.

(A, B) Number of genes (A) and transcripts (B) projected onto the UMAP derived from single cell RNA sequencing of Stage IA gastric cancer mucosa.

(C) Dot plot depicting the average expression and detection rate (percentage of cells with any expression) levels of cell type specific markers. The color of each circle represents the average expression and the size represents the gene detection rate.

(D) UMAP projection of the annotated cell types: chief, parietal, plasma, pit, NK, endothelial, mast, B cell, T cell, signet ring cell (SRC), stem/pericyte/fibroblast, and gastrin.

(E) Enriched pathways derived from genes that are differentially expressed between the SRC and non-SRC populations. ECM-related gene sets that are downregulated (DR) in SRC compared to the stem cell/pericyte/fibroblast cluster include ECM-proteoglycans, cell-extracellular matrix interactions, and miRNA targets in ECM and membrane receptors.

Identification of SRC lesions in vivo could improve early cancer detection. Therefore, we explored scRNA-seq data for candidate SRC markers with low CDH1 expression and high expression of REG1A (a known SRC marker) to identify the SRC cell population cluster from adjacent EP-Chief cluster (Fig. 2D) (32–35). Within the SRC cluster, we demonstrate the evolution of early, intermediate, and late-SRC cell states defined by decreasing CDH1 expression and increasing REG1A expression (Fig. 2C–D). In addition to increased expression of REG1A, we found that additional SRC-enriched markers with elevated expression in the SRC cluster, especially at the EP/Chief-SRC boundary included VIM, AQP5, MUC6, PRR4, and AGR2. Both scRNA-seq and spatial transcriptomic results demonstrated that REG1A (p=0.003), VIM (p<0.001), and CD74 (p<0.001) were upregulated in SRC compared to gastric epithelium. VIM and CD74 were also upregulated, differentially expressed genes in SRC compared to adjacent unaffected epithelium (p<0.05, |log₂ fold-change|≥1). By computing an SRC-specific enrichment score from the gene set composed of REG1A and our candidate SRC markers (VIM, AQP5, PRR4, MUC6, and AGR2) using the AddModuleScore function (R package Seurat) and Pearson correlation analysis, we identified other genes whose expression levels were positively correlated (p<0.01) with SRC-specific enrichment score including SLPI, IL33, and CD74. Our pathway enrichment analysis utilizing genes differentially expressed between SRC and non-SRC populations (Supplementary Table S4) revealed pathways associated with ECM-proteoglycans, cell-extracellular matrix interactions, and miRNA targets in ECM and membrane receptors that were downregulated in SRC compared to the cluster composed of stem cells, pericytes, and fibroblasts, thereby supporting our findings derived from spatial transcriptomic data. Furthermore, ECM remodeling is a potential mechanism for early SRC lesion formation (Fig. 2E).

Advanced DGC is marked by oncogenic proliferation

We sought to understand the cellular changes associated with progression of early SRC lesions to more advanced (≥T2) tumor stages of DGC. We hypothesized that because SRC exhibited genomic and transcriptomic profiles akin to native gastric epithelium that advanced gastric cancers by comparison would show marked changes in transcriptional programs distinct from the quiescent phenotype of SRC lesions. Tumor-bearing tissue, pathologically staged pT2 to pT4, was obtained from four patients with germline CDH1 variants who underwent therapeutic gastrectomy for advanced DGC (Supplementary Table S1). One patient received systemic chemotherapy for invasive breast cancer prior to gastrectomy, and no patients received radiation therapy prior to surgery. DGC lesions were identified by gross mucosal abnormalities (i.e. ulceration) on endoscopy, (Fig. 3A) and demonstrated invasion by cancer cells through the submucosa and into the muscularis propria (T2), penetration of the subserosal connective tissue (T3), or invasion into the visceral peritoneum (T4). DGC cancer cells appeared smaller, had a higher nuclear-cytoplasm ratio, and contained less mucin than SRC lesions (Fig. 3B–C).

Figure 3.

(A) Advanced diffuse gastric cancer (DGC) lesion found on endoscopy (arrow).

(B, C) H&E of DGC at 200x magnification with 50 μm scale bars. (B) Cancer cells from advanced lesions exhibit high nuclear-cytoplasm ratio with no mucin and highly atypical nuclei with size variation (black arrows). (C) Advanced DGC cells infiltrate singly (black arrows) or in a pile formation (ellipse). Mitoses (red arrows) are frequently seen.

(D) Masked region of interest selection and immunofluorescent imaging of advanced DGC.

(E) Volcano plot showing representative genes with increased expression in SRC or DGC (p< 0.05 and |log2 fold-change| ≥1).

(F) Differences in expression of select adherens junction genes in SRC ROIs compared to DGC ROIs.

(G) Multiplex immunofluorescence imaging of advanced DGC lesions (arrows). Panel 1 H&E, panel 2 nuclei in blue, panel 3 pancytokeratin in yellow, Ki67 in purple, panel 4 E-cadherin in blue, panel 5 αSMA in blue, CD4 in yellow and CD8 in red.

(H) Heatmap demonstrating enrichment of select genes within the Mitotic G1 phase and G1 S transition gene set in DGC compared to SRC.

(I) Heatmap comparison of differentially expressed genes from RNA sequencing analysis between gastric epithelium (EP), SRC, and DGC ROIs.

Comparison of the spatial gene expression profiles of the SRC lesions and DGC ROIs (Fig. 3D) revealed that 3.9% of the genes (454/11681) were differentially expressed (p<0.05, |log₂ fold-change|≥1) of which 79 genes were upregulated in DGC compared to SRC (Fig. 3E). Top differentially expressed genes in this comparison included HMGA1, MYC, RETREG1, and ONECUT2, whereas the adherens junction complex gene set was enriched in DGC relative to SRC. Specifically, CDH1 (p<0.001), CTNNA1 (p=0.005), and CCND1 (p=0.02) expression was increased in DGC compared to SRC (Fig. 3F). Multiplex immunofluorescence imaging confirmed protein-level expression of transcriptomic findings (Fig. 3G). E-cadherin expression was retained in advanced DGC suggesting a temporal restriction of E-cadherin expression in SRC. This finding is consistent with the concept that loss of E-cadherin expression is often an epigenetic modification that facilitates not only altered cell adhesion but also promotes SRC formation and cell migration (36).

GSEA identified 241 gene sets in the Reactome database that were significantly enriched in DGC compared to SRC. LEA clustered the largest GSEA gene sets into Cell Cycle, Proliferation, and Replication (Supplemental Figure 1B). We found enrichment in gene sets involving G1 phase (Mitotic G1 phase and G1 S Transition, NES=1.71, q-value =0.061) (Fig. 3H), G1 to S phase transition (Signaling by Hedgehog, NES=1.65, q-value=0.081) and S phase (S Phase Signal, NES=1.69, q-value=0.065) in DGC compared to SRC lesions. DGC did not exhibit increased mRNA expression of genes shown to carry genomic alterations in sporadic advanced gastric cancer compared to SRC, such as TP53, SMAD4, APC2, KRAS, and ARID1A (11,27,37). However, somatic sequencing of DGC tumor DNA revealed pathogenic TP53 mutations in three of four tumors, and a likely pathogenic ERBB3 mutation in one (Supplementary Table S1). Differentially expressed genes among EP, SRC, and DGC demonstrated minimal inter-patient variation with similar clustering patterns of unaffected gastric EP and SRC lesions, compared to distinct clustering of DGC (Fig. 3I).

Although β-catenin (CTNNB1) was not differentially expressed in DGC by mRNA analysis, its activity is derived from post-translational modifications, several of which have been implicated in gastric cancer progression (38,39). Target gene transcripts of the Wnt pathway that were increased in DGC relative to SRC included MET (p=0.01), DKK1 (p=0.02), and CCND1 (p=0.02). GSEA also demonstrated enrichment of the β-Catenin Independent WNT Signaling (NES=1.87, q-value=0.021) in DGC compared to SRC lesions. MYC and MET expression was unchanged between EP and SRC, however, MYC was increased in DGC (p<0.001) relative to SRC lesions. The finding of dysregulated WNT signaling suggests that Wnt pathway activation is associated with transformation of SRC lesions to DGC.

Discussion

Here we apply spatial-omics techniques to elucidate the latent clinical behavior of early gastric SRC lesions that arise in individuals with germline CDH1 loss-of-function mutations. First, we show that E-cadherin expression is reduced or absent in the formation of intramucosal gastric SRC lesions (21). Specifically, we show gene set enrichment in ECM remodeling accompanies the loss of CDH1 gene expression in SRC lesions, which supports that misoriented cell division due to E-cadherin protein loss alone is insufficient to allow for SRC invasion of cell matrix. Moreover, we demonstrate that although SRC are termed carcinoma, these cells lack important and well-established hallmarks of advanced gastric carcinomas. Somatic mutations in cancer genes (e.g., TP53, ARID1A) that are common to advanced diffuse gastric cancers are notably absent in SRC lesions. Importantly, we demonstrate that the transition of SRC lesions to advanced stages of diffuse-type gastric cancer is marked by specific transcriptomic changes, such as MYC upregulation. Taken together these data help explain the extended cellular dormancy of SRC lesions in gastric mucosa, which is highlighted by their ubiquity despite mounting evidence of variable and incomplete gastric cancer penetrance in CDH1 variant carriers (22).

Previous studies have shown that loss of E-cadherin expression in gastric epithelia is a cancer-initiating event due to disruption of epithelial architecture and planar cell division, which can generate subepithelial, E-cadherin-negative SRC lesions (21,40,41). Monster and colleagues showed disrupted epithelial architecture in E-cadherin knock-out human organoids resulted in accumulation of individual detached cells with SRC morphology, which was increased with addition of a matrix protease (21). We provide corroborating data from human gastric tissue illustrating not only that SRC lesions exhibit reduced or absent E-cadherin expression, but also that SRC upregulate genes that promote ECM invasion and remodeling. Our results support that E-cadherin loss is necessary but insufficient for initiating diffuse-type gastric cancer. The findings here may also be applicable to CDH1-deficient lobular carcinoma of the breast, which is characterized by a discohesive, migratory phenotype (42,43). Loss-of-function variants in CDH1 are believed to be an early event in breast cancer carcinogenesis, observed in 50–60% of lobular carcinoma in situ (LCIS) and adjacent invasive lobular carcinoma (ILC) specimens (43,44). Somatic alterations in CDH1 and CTNNA1 have been found also in breast cancer leptomeningeal metastasis from ILC and non-ILC primary breast tumors (45). Multi-omics analyses of CDH1-driven ILC could similarly demonstrate mechanisms of breast precancer formation and disease progression.

The clinical management of SRC lesions found in CDH1 P/LP variant carriers is an area of clinical equipoise. Mounting clinical evidence supports that CDH1 variant carriers frequently harbor occult SRC lesions and can undergo extended periods of endoscopic surveillance as an organ-sparing approach to management of elevated hereditary diffuse gastric cancer risk (22,23,46). Herein, we address the clinical dilemma that SRC lesions, which are pathologically described as invasive adenocarcinomas, may not necessitate aggressive treatment. Through multi-omics analyses of human gastric tissue in germline CDH1 variant carriers, we show that gastric SRC lesions manifest a unique molecular phenotype that is distinct from their advanced cancer counterparts (47). Gastric SRC lesions lack important hallmarks of advanced diffuse-type gastric cancers, which may help explain why patients with known SRC do not invariably develop advanced gastric cancer. Previous molecular characterizations of advanced diffuse gastric cancer have consistently shown alterations in signal transduction by adhesion receptors and their downstream effectors (48). Notably absent from SRC lesions in our analysis were alterations in the same genes or pathways (i.e., TP53, ARID1A, RHOA) that would be expected in advanced gastric cancers (28). The human data presented herein mirror a genetically-engineered murine model in which Atpb4-Cre driven knockout of CDH1 in gastric epithelium generates intramucosal SRC lesions that fail to progress to DGC without additional TP53 tumor suppressor loss (10,11). To that end, we demonstrate that although SRC lesions in human samples do not harbor somatic TP53 mutations, we identified both TP53 and ERBB3 mutations in the advanced stage human tumors. We also demonstrate absence of E-cadherin expression at the protein-level in SRC lesions, with paradoxical E-cadherin re-expression in advanced DGC. This may be due to the plasticity of epigenetic modifications, specifically promoter hypermethylation, which is the most commonly reported “second-hit” in early gastric SRC, whereas advanced DGC frequently exhibit loss of heterozygosity or gain a second mutation such as TP53 (41). These data support that additional genomic alterations (beyond germline CDH1) are likely acquired with the progression of SRC to advanced gastric cancer. Additional large-scale studies are needed to investigate mechanisms of biallelic CDH1 inactivation and disease progression in germline CDH1 variant carriers (49). Furthermore, integration of molecular testing of SRC lesions found on surveillance endoscopic biopsy may help aid clinical guidance and identify lesions at higher risk of disease progression.

We performed scRNA-seq analysis of SRC lesions to identify potential cell-enriched markers of these precancers to inform gastric cancer interception strategies. We identified AQP5, REG1A, AGR2, and VIM as candidate biomarkers of SRC, with elevated REG1A expression in late-compared to early-state SRC lesions. REG1A appears to play a role in cell invasion, angiogenesis, and apoptosis in gastric cancer (33,34). Increased REG1A expression may act as an indicator of SRC lesion evolution and proclivity towards cell progression. Additional trajectory analyses with matched, intra-patient SRC and advanced DGC lesions are needed to further elucidate this finding. Wang et al studied AJCC Stage I early intestinal-type gastric cancers and demonstrated AQP5 as a marker of gastric cancer stem cells (50). Anterior gradient 2 (AGR2) may promote invasion of gastric cancer cells through coordinated effects on stromal cells (51). VIM is associated with poor prognosis in gastric cancer patients (52). To further differentiate SRC lesions from advanced lesions, we demonstrate that advanced DGC exhibited upregulation of MYC. These findings may be actionable because a metabolic switch could help identify SRC lesions that have initiated progression to an advanced stage.

Our analysis was limited by the rare nature of this hereditary cancer syndrome and limited number of occult, microscopic, gastric SRC lesions in human tissue samples. Moreover, there is a scarcity of germline CDH1 variant carriers presenting with advanced tumor-stage gastric cancers. This is likely due to the infrequent progression of SRC lesions to advanced disease and current consensus guidelines recommendation of prophylactic total gastrectomy at age 20 to 30 years upon diagnosis of SRC lesions on endoscopy (6). Furthermore, we were unable to perform trajectory analyses, assess inter- and intra-patient heterogeneity in stage-matched comparisons, and provide intra-patient comparisons of individual SRC foci. We did not explore the cell of origin of SRC lesions, although multiple prior studies have implicated the proliferative region of gastric epithelial glands (53,54).

Conclusion

We found the latent clinical behavior and incomplete cancer penetrance of gastric SRC lesions in germline CDH1 variant carriers is likely attributed to a precancer molecular phenotype. The data presented here provide biologic evidence supporting gastric SRC as non-obligate precursors of DGC in individuals with germline CDH1 variants. Based on this analysis and emerging cancer penetrance data, some patients with SRC may be at risk of overtreatment with prophylactic total gastrectomy owing to the unique, indolent SRC phenotype, and infrequent progression to advanced DGC (9,23). Additional investigation into the role of the tumor microenvironment of early diffuse-type gastric carcinogenesis is warranted, as is further study of biomarkers of SRC lesion progression to cancer.