Early Immune Changes Support Signet Ring Cell Dormancy in CDH1-Driven Hereditary Diffuse Gastric Carcinogenesis

This article is featured in Selected Articles from This Issue, p. 1249

Pathogenic germline CDH1 variants result in hereditary diffuse gastric cancer (HDGC) syndrome, a rare etiology of gastric adenocarcinoma, which is the fourth leading cause of cancer-related death worldwide (1–3). The study of diffuse-type gastric carcinogenesis in Western countries suffers from delayed presentation in sporadic gastric adenocarcinoma cases, as many patients are diagnosed with synchronous nodal or peritoneal metastases. In contrast to sporadic diffuse-type gastric adenocarcinoma, a growing number of patients with HDGC are identified relatively early in the disease course via genetic testing. Approximately 30% to 40% of pathogenic CDH1-variant carriers are expected to develop advanced gastric adenocarcinoma during their lifetime (4, 5). These high-risk patients can undergo a risk-reducing (prophylactic) total gastrectomy (RRTG) that removes all at-risk gastric mucosa and eliminates the chance of developing gastric adenocarcinoma (6).

Despite incomplete clinical penetrance of CDH1-associated advanced gastric cancer, nearly all patients with pathogenic germline CDH1 variants harbor occult, multifocal stage IA adenocarcinomas. These occult gastric carcinomas are characterized by intramucosal foci of signet ring cells (SRC) with various degrees of malignant potential (7–9). Tissue obtained from RRTG of patients with HDGC enables the investigation of the earliest steps of gastric carcinogenesis. The mechanisms responsible for both the transformation of SRCs from at-risk gastric mucosa as well as the relative dormancy of SRCs in HDGC are currently unknown, as are the subsequent steps that allow quiescent SRCs to progress into aggressive diffuse gastric carcinomas. To better understand the initiating steps of hereditary diffuse-type gastric carcinogenesis in HDGC, we compared the transcriptomic profiles of SRCs within the lamina propria to that of adjacent mucosa in non-SRC epithelium (NEP) using surgically explanted stomachs of patients with germline pathogenic CDH1 variants.

A review of prospectively acquired data identified individuals with pathogenic CDH1 variants who presented to our institution for evaluation. All patients provided written informed consent and were enrolled in a prospective study of hereditary gastric cancer (Clinical Trial Registration ID: NCT03030404). Patients were included in the analysis who were >18 years of age, had a loss-of-function CDH1 variant confirmed to be pathogenic based on criteria from the American College of Medical Genetics (10), demonstrated no clinical evidence of advanced gastric cancer, and underwent a risk-reducing total gastrectomy according to clinical management guidelines (11). In addition, there were no incidentally discovered advanced cancers on pathologic analysis. The study was approved by the NIH Institutional Review Board following ethical guidelines in accordance with the Declaration of Helsinki.

The workflow for capturing SRC-enriched areas in the lamina propria and adjacent regions of normal-appearing gastric mucosal epithelium is illustrated in Fig. 1A. At the time of operation, resected stomachs were formalin-fixed and paraffin-embedded (FFPE) in their entirety. Pathology was reviewed from each gastrectomy specimen to identify tissue sections positive for foci of intramucosal SRCs. Experienced gastrointestinal pathologists surveyed tissue blocks to identify and select areas with the highest concentration of SRC foci. Tissue slides were recut from the same FFPE blocks and plated onto glass PEN membranes (Thermo Fisher Scientific). All SRC foci were microscopically identified as regions of interest (ROI) based on morphology. An area of adjacent NEP of similar size was selected from each specimen. Excision of SRC ROIs and NEP tissue was performed using laser capture micro-dissection (LCM) on an Arcturus XT Laser Microdissection System Tissue using IR and UV laser and CapSure Macro LCM Cap. LCM was confirmed and performed independently by a second pathologist.

DNA/RNA was extracted using the AllPrep DNA/RNA FFPE Kit (catalog no. 80234, Qiagen) following the manufacturer's instructions. Sample preparation was performed using the NEB NEBnext Ultra Low Input Total RNA with rRNA Depletion kit. The average RNA extracted from SRC and NEP sections was 152.77ng (90.0–318.0) and 188.57ng (90.0–390.0), respectively. cDNA libraries were sequenced on the NovaSeq 6000 platform using paired-end sequencing with 100-bp read lengths. Reads were trimmed to remove sequencing adapters using Cutadapt (12;v1.18), and aligned to the human reference genome hg38 using STAR (ref. 13;v 2.7.0f) in two-pass mode. Library preparation and sequencing generated between 165 and 255 million reads per sample. More than 91% of the bases were above a quality score of Q30.

RNA expression levels were quantified using RSEM (14;v.1.3.1) with GENCODE annotation (v.30;15). Genes with fewer than 10 counts were removed prior to further analysis. Normalization and differential expression analysis was performed using DESeq2 (16;v1.32.0) in R (v4.2.2).

Principal component analysis (PCA) on the filtered data following variance stabilizing transformation was performed using mean scaling with default parameters. Differential expression was tested comparing the SRC to the NEP samples adjusting for the sample ID. Genes were declared differentially expressed at an absolute log₂ fold change of >1 and adjusted FDR q-value of 0.05.

Differentially expressed genes were used for downstream overrepresentation and gene-set enrichment pathway analyses using the ReactomePA, R/Bioconductor, and ClusterProfiler (17) packages. Reactome and KEGG pathways were used as the gene–pathway mapping databases, restricting to gene sets containing greater than 10 and fewer than 500. Genes were included into pathway analysis if log₂(FC) >1.2 and P_adj < 0.05. Significance threshold for differentially regulated pathways was set at P_adj < 0.001.

Differentially expressed gene sets between SRC and NEP samples were identified by running preranked GSEA (v4.3.2;18) on −log₁₀P values which were positive or negative depending on the direction of the change and using all gene sets in mSigDB (v2022.1.Hs; ref. 19).

The raw RNA-sequencing (RNA-seq) counts were normalized using transcripts per million (TPM) for use in cell deconvolution. CIBERSORTx (20) was used in absolute mode for deconvoluting 22 cell types using the LM22 reference dataset as distributed.

Because of the archival nature of the dataset, low RNA integrity number (RIN) scores were expected. We performed subsampling of the mapped read counts using the subSeq (ref. 21; v.1.22) package. This performs a random draw of mapped reads from a binomial distribution. We chose five probabilities for reads to be selected ranging from 100-fold to 1-fold reduction (or probability from 0.01 to 1). Differential expression tests followed by pathway analyses were rerun on these subsampled counts.

IHC was performed using a Leica Bond RX autostainer. Briefly, 5-μm–thick sections were cut from the same blocks as LCMs. Deparaffinization, rehydration, and 20 minutes of antigen retrieval in a citrate-based buffer were performed. Slides were incubated with either rat anti-human CD3ε (Bio-Rad, clone: CD3–12, RRID:AB_321245), CD4 (Leica, clone: 4B12, RRID:AB_563559), CD8 (Leica, clone: 4B11, RRID:AB_10555292), FOXP3 (Leica, clone: 236A/E7), CD56 (Leica, clone: CD564, RRID:AB_10554753), CD20 (Leica, clone: L6), MUC6 (Leica, clone: CLH5, RRID:AB_442114), PD1 (Roche, clone: NAT105), HLA-DR (Dako, clone: TAL.1B5, RRID:AB_2262753), mouse anti-human antibody that recognizes HLA-DPB1, HLA-DQB1, and HLA-DRB1 (HLA-D; Dako, clone: CR3/43, RRID:AB_629967) or appropriate isotype control. Detection was performed using the BOND Polymer Refine Kit (Leica, catalog no. DS9800, RRID:AB_2891238). Either Periodic Acid-Schiff (PAS) or eosin was used as a counter-stain. IHC images were quantified for cells positive for CD3ε and HLA-D using HALO Client (version 3.3.2541, Indica Labs). In HALO, ROIs were drawn at 20× magnification to encompass all SRCs, and an ROI of equivalent area was then drawn around NEP for each patient (size range 0.035–0.08 mm²). All ROIs were then analyzed using the CytoNuclear algorithm (version 2.0.9, Indica Labs) for CD3ε or HLA-D. The number of positive cells counted was divided by the area to calculate the cell frequency of each ROI. For each tissue section, one ROI of tumor and two immediately adjacent ROIs of normal gastric mucosa were manually selected by a board-certified pathologist (B. Gasmi) and reviewed by expert physician (J.L. Davis).

GraphPad Prism (version 9.4) was used to plot and analyze data. Unpaired two-tailed Student t test or two-way ANOVA was employed for statistical comparisons, with P < 0.05 used for significance threshold.

The data generated in this study are publicly available in dbGaP.

Twenty patients with pathogenic CDH1 variants from 16 kindreds who had undergone risk-reducing total gastrectomy (RRTG) in the absence of overt diffuse gastric cancer between January 2016 and May 2019 were retrospectively identified from a prospective study of hereditary gastric cancer syndromes (Clinical Trial Registration ID: NCT03030404). The majority of patients were White (19/20) and female (14/20) with an average age of 47-years-old at RRTG (Table 1). Ninety percent (18/20) of patients had a positive family history of gastric cancer. The genotype of CDH1 within the cohort included 13 different pathogenic CDH1 variants (Table 1).

On pathologic examination, all RRTG specimens contained occult SRC clusters in the lamina propria (stage IA gastric adenocarcinoma). SRC morphology was typical of HDGC (8), with large mucin-filled cytoplasm and eccentrically placed nuclei conferring a signet-ring appearance (Fig. 1A and B). Approximately 100–100,000 SRC and >10,000 NEP cells were microdissected per sample (Fig. 1C). Matched SRC and NEP samples were selected in 16 patients, whereas additional SRC samples only were obtained from 2 patients and NEP samples only were taken from 2 patients due to their matched samples having RNA concentrations below cutoff. After cDNA sequencing, reads were aligned and gene expression quantified for more than 15,000 genes. PCA demonstrated overall separate clustering between SRC and NEP samples, although there was some degree of overlap, possibly representing heterogeneity as epithelial precursors transform into SRCs (Fig. 1D).

Over 14,000 genes passed quality control and were included in downstream analysis (Supplementary Table S1). DESeq2 analysis identified 876 significantly differentially expressed genes (DEG) between SRC and NEP, with 564 upregulated and 312 downregulated in SRC (Supplementary Table S1). Genes responsible for the largest differences between NEP and SRC groups are shown in Fig. 1E. Several genes known to be involved in the process of malignant transformation were upregulated in SRC including proliferation (MKI67, multiple histone genes) and epithelial-to-mesenchymal transition (EMT; VIM, MMP2; Fig. 1F). Downregulated genes included those involved in parietal cell-specific function such as the H⁺/K⁺ ATPase (ATP4A) and intrinsic factor (CBLIF), potentially reflecting of the loss of normal epithelial cell differentiation upon carcinogenesis. Notably, although SRCs contain large stores of cytoplasmic mucin, several genes encoding mucins (e.g., MUC6) were also downregulated in SRC, possibly reflecting a loss of foveolar or mucus cells present in NEP. Interestingly, the levels of CDH1 expression were similar between NEP and SRC (P_adj = 0.172), consistent with the heterozygous variant causing a field defect of CDH1 expression in both SRC and at-risk NEP at the mRNA level, and indicating that further alterations are likely necessary for the transformation of E cadherin–deficient epithelial cells into early cancer cells. Other signaling molecules known to be involved in the progression or carcinogenesis of sporadic diffuse-type gastric adenocarcinoma were not differentially expressed in our patient dataset, including CTNNB1, ARID1A, FBXW7, and RHOA (Supplementary Table S1; refs. 22, 23).

Unexpectedly, we found a large number of immune cell–related genes that were upregulated in SRC, suggesting a potential role of the tumor–immune microenvironment (TiME) in either limiting or promoting the development of carcinogenesis (Fig. 1F). While CD45 (PTPRC) was upregulated in SRC, genes encoding CD11b (ITGAM) and CD11c (ITGAX) were not increased compared with NEP, implying no changes in the myeloid or dendritic cell populations. Instead, genes involved in T-cell lineage (CD3E, CD4), signaling (CD28, IL2RA, IL2RB, ITK, ZAP70), and trafficking (SELL, CCR7, S1PR1) were significantly upregulated in the mRNA from SRC compared with NEP. Interestingly, CD8A was not different between SRC and NEP (P_adj = 0.67). Several genes involved in T-cell activation (CD44, TNFRSF9) were expressed at equivalent levels in SRC compared with NEP; however, others (ICOS, TIGIT) were significantly elevated. Expression of the T-cell exhaustion marker HAVCR2 (encoding TIM3) was also elevated in SRC.

Next, Reactome and KEGG pathway analysis was performed to identify canonical oncogenic pathways whose gene sets were enriched in SRC. Unsupervised analysis of these databases failed to reveal any specific pathways related to oncogenesis whose genes were significantly enriched in SRCs compared with NEP. Instead, the pathways that were most enriched in upregulated genes included T-cell receptor (TCR) phosphorylation and transduction, ZAP70 translocation, CD28 costimulation, IL2 signaling, and programmed cell death protein 1 (PD-1) signaling (Fig. 1G and H; Supplementary Table S2). Subsampling analysis demonstrated similar pathways enriched in upregulated genes at all probabilities tested (Supplementary Fig. S1). Thus, the bulk transcriptomic profile of SRC foci appears strongly biased toward intracellular immune signaling compared with surrounding epithelium, suggesting an important interaction between the TiME and SRCs during the early steps of carcinogenesis.

As an additional in silico method, GSEA was performed. Using an FWER P-value ≤ 0.05 as a cutoff, 306 gene sets (out of >17,000) were found to be positively enriched and 54 negatively enriched. Consistent with our interpretation of DEGs, gene sets that included parenchymal stomach cell types were negatively enriched, including those specific for parietal and pit cells (Supplementary Fig. S2). Similar to the Reactome pathway analysis above, gene sets involving T-cell activation, TCR signaling, and lymphocyte costimulation, were among the top positively enriched (Supplementary Fig. S3A; Supplementary Table S3), while unlike the Reactome analysis, T-cell exhaustion and PD-1⁺ CD8⁺ T-cell gene sets were not significantly enriched (Supplementary Fig. S3B). In summary, analysis of the transcriptomes of the SRC regions of patients with pathogenic CDH1 variants show evidence for elevated lymphocyte signaling and dampened parenchymal differentiation gene signatures compared with NEP during the earliest phases of carcinogenesis.

Kumagai and colleagues recently reported that the composition of gastric tumor-infiltrating lymphocytes (TIL) was highly sensitive to the local environment set by the metabolic activity of cancer cells (24). To explore whether a similar mechanism was operative within SRC compared with NEP, we analyzed our GSEA data specifically looking for alterations in metabolic pathways that might potentially bias the TiME toward tolerance in gastric carcinogenesis. As observed within more advanced gastric tumors (25), gene sets involved in oxidative phosphorylation were found to be significantly downregulated (Supplementary Fig. S4A). However, unlike advanced gastric cancers (26), glycolysis was not significantly enriched in the SRC regions, indicating a limited Warburg effect (Supplementary Fig. S4B). Similarly, fatty acid oxidation was also not significantly enriched (Supplementary Fig. S4C). Other gene sets corresponding to signaling pathways known to be important in advanced gastric cancer, including cadherin binding, Wnt/β-catenin signaling, MYC targets, EMT, and p53 pathway, were not dysregulated (Supplementary Fig. S5). Thus, the immune changes that occur with early carcinogenesis described above appear to coincide with a decrease in oxidative metabolism and without concurrent increases in hallmark gene sets for glycolysis or carcinogenesis.

To confirm the presence of immune cell infiltration in SRC foci, we performed IHC on three representative patient samples, staining for CD3ε⁺ T cells and MHC II–expressing APCs. PAS was selected as a counterstain to positively identify SRCs. Both T cells and APCs were present in the SRC and NEP, but neither CD3ε expression nor HLA-D expression (including HLA-DP, -DQ, and -DR) were significantly different between NEP and SRC in this analysis (P = 0.672 and P = 0.723, respectively; Fig. 2A and B; Supplementary Fig. S6). Thus, although TCR-expressing cells and APCs could be seen within SRC carcinomas by IHC, their overall abundance did not appear to change compared with their presence in adjacent gastric mucosa by this technique. This observation suggests that the observed differences in DEGs and immune pathway enrichment from SRC foci transcriptomes likely results from changes in immune cell activity or a subset of these major leukocyte classes.

Next, we used CIBERSORTx, a cell deconvolution algorithm that can be used on human bulk RNA-seq data, to predict the relative frequency and major phenotypes of the immune cell compartment of SRCs with potentially greater sensitivity compared with the initial IHC analysis in Fig. 2A and B. (27). Twenty-one of the 22 immune subtypes were identified by CIBERSORTx in each NEP and SRC sample, with a notable absence of gamma-delta T cells in either region (Fig. 3A; Supplementary Fig. S7). Overall, CIBERSORTx revealed significantly more total immune cells within the SRC samples compared with the NEP samples (Fig. 3B), consistent with PTPRC upregulation in SRC (Fig. 1F). When comparing SRC regions to NEP regions, significant increases in the predicted frequency of Tregs as well as natural killer (NK) cells were noted (Fig. 3C; Supplementary Fig. S8). There were no significant changes in the overall predicted frequency of CD8⁺ T cells. There were also no changes in the subsets of CD4⁺ T cells including the follicular, memory/activated, or memory/resting subtypes of CD4⁺ T cells; however, this technique does not enable the relative quantification of the overall CD4⁺ T-cell lineage. In addition, no changes in the predicted frequency of granulocytes or myeloid cells, including both M1 and M2 macrophages, monocytes, and dendritic cells (Fig. 3D), were seen. Thus, in silico deconvolution of transcriptomic data predicted an elevation of NK cells and Tregs that infiltrate SRCs.

To evaluate these predictions on the protein level, we performed IHC on a number of cell markers. We observed SRCs and NEP regions in all slides by H&E, as well as MUC6 downregulation in SRCs as predicted by analysis of DEGs (Fig. 4). Similar to our DEG and GSEA data, we observed increases in total CD4⁺ cells and FOXP3⁺ cells that infiltrated SRC regions compared with NEP areas (Fig. 4). Consistent with CD4⁺ cell upregulation, HLA-DR expression was more intense in the SRC regions. The infiltration of CD8⁺ and PD-1⁺ cells was similar between the two regions. CD20⁺ cells (representing B cells) and CD56⁺ cells (representing NK cells) were absent from SRCs and NEP. Thus, while most major immune cells are comparable in frequency between SRC and NEP, the upregulation of HLA-DR⁺ and CD4⁺ cells suggests an enhancement of helper T cell–APC interactions in SRCs.

Diffuse-type gastric adenocarcinoma in Western countries most often presents at later stages compared with early-stage gastric adenocarcinoma seen in endemic countries with national screening programs (28). Consequently, the study of diffuse-type gastric adenocarcinoma carcinogenesis is challenging. HDGC affords a unique opportunity for the investigation of early gastric carcinogenesis, as patients with HDGC often present prior to the onset of advanced disease due to germline genetic testing for a pathogenic CDH1 variant and gastric cancer screening. Access to at-risk gastric mucosa bearing occult foci of SRC is possible due to current recommendations for surgical prophylaxis in patients bearing a pathogenic CDH1 variant (11). Thus, tissue obtained via RRTG offers a convenient approach to study the initiating events in diffuse-type gastric adenocarcinoma carcinogenesis within the context of a known pathogenic CDH1 genotype.

In this study, we analyzed transcriptomic differences in at-risk gastric epithelium to elucidate potential mechanisms underlying the nearly universal presence of SRC foci, but incomplete penetrance of advanced gastric cancer, in germline CDH1-variant carriers. Occult foci of SRC, which are pathognomonic of HDGC, exhibited increased expression of certain genes related to cell proliferation and EMT, as well as downregulation of oxidative phosphorylation pathways. However, there were no significant changes in oncogenic pathways associated with gastric carcinogenesis. The majority of upregulated transcriptional pathways in SRC samples were consistent with enhanced immune cell signaling compared with adjacent uninvolved gastric mucosa. Although there were increases in T-cell gene signatures by mRNA, there were no significant differences in the overall frequencies of CD3ε⁺ cells or MHC II⁺ APCs illustrated by IHC and CIBERSORTx. Interestingly, the only lymphocyte subsets increased in the SRC TiME consistently across all platforms were Tregs, both by CIBERSORTx and confirmed by IHC. CD4⁺ cells were upregulated in SRC by IHC, GSEA, and DEG analysis, but CD4⁺ T-cell subsets were not predicted to change in SRC regions by CIBERSORTx. This discrepancy may have been due to the signature matrix encompassing CD4⁺ T-cell subsets and not the overall CD4⁺ T-cell population. Meanwhile, HLA-DR stained more intensely within SRCs by IHC, but the quantities of APCs (DCs and macrophages) were predicted to be the same across sample type by CIBERSORTx. These differences between transcript and protein-level expression may have to do with posttranscriptional regulation of gene expression, but also highlight the limitation of bulk transcriptomic techniques to differentiate between overall cell frequency and an individual cell's quantitative gene expression. Overall, the finding across all modalities suggests an important role for CD4⁺ T cells and APCs within the TiME in the development of HDGC during the early stages of tumorigenesis.

Occult SRC discovered in asymptomatic patients with HDGC appear to embody an intermediate step between NEP and advanced gastric adenocarcinoma. Our observation that laser microdissected SRC and NEP regions had similar transcriptomic profiles with respect to cancer signaling pathways, and lack of upregulation of genes specifically implicated in sporadic diffuse-type gastric cancer (23, 29), is consistent with this interpretation. Furthermore, the underlying genetic differences between SRC due to germline CDH1 variants and sporadic overall diffuse-type gastric adenocarcinoma likely also contribute to the differences that we observed. For example, although 42% of sporadic diffuse-type gastric adenocarcinoma contain a somatic CDH1 mutation (23), only 11% of sporadic gastric adenocarcinoma with SRC features will have a CDH1 mutation (30). Meanwhile, in HDGC, the observation that the underlying gastric epithelial cells are heterozygous for the CDH1 pathogenic variant while SRCs are CDH1-null (E-cadherin deficient) suggests that the progenitor and its progeny may be more similar than different (31). Although advanced stages of hereditary and sporadic diffuse-type gastric adenocarcinoma may share similarities, our data are consistent with alternative etiologies (22, 23).

The upregulation of immune signatures in SRC foci supports that the loss of TiME control, rather than activation of classical oncogenic pathways, may be a major factor driving the progression of carcinogenesis in HDGC. Immune escape is a known hallmark of carcinogenesis (32), yet the composition of the TiME varies substantially with histopathologic stage in gastric cancer (33, 34). TILs and APCs can display variable penetration into advanced tumors relative to surrounding tissue. In sporadic diffuse-type gastric cancers, T-cell depth of penetration mediated by CCL2 correlates with degree of T-cell exhaustion (35). A prior study investigated the role of the adaptive immunity in tumor control in sporadic advanced gastric adenocarcinoma with SRCs, as the CD3⁺ T-cell infiltration correlated with (but did not independently predict) improved overall survival in treatment-naïve patients (36). Our data on quiescent HDGC foci, in contrast, demonstrated comparable levels of several antitumor immune cells including canonical CD8⁺ T cells, M1 macrophages, and dendritic cells between SRC and NEP. Combined with the known anticancer effect of the adaptive immune system, we hypothesize that this efficient penetration and increase of CD4 T cells contributes to SRC dormancy prior to progression into advanced stage gastric cancers.

In our patient population, regulatory T cells were the only CD4⁺ T-cell subtype predicted to be enriched in the SRC microenvironment. As a potently immunosuppressive lymphocyte, Tregs are enriched in many cancer types including advanced gastric adenocarcinoma (37). While their overall quantity is negatively prognostic, the role of Tregs in early-stage SRC tumor development is unknown and may be inverse to that of more advanced-stage sporadic gastric adenocarcinoma (38). Tregs comprise a low percentage of total CD4⁺ T cells in normal tissue, a fact that possibly explains their predicted increase in SRC foci by CIBERSORTx without an overall change in the CD3ε⁺ T-cell frequency by IHC. Tregs exert their suppressive function following classically described T-cell activation steps including TCR ligation, CD28 costimulation, and IL2 signaling (39–41). Furthermore, Tregs express surface markers including immune checkpoint molecules, with PD-1 and TNFR2 expression on Tregs in particular having prognostic significance in gastric adenocarcinoma (42, 43). Finally, inducible Tregs can differentiate from naïve CD4⁺ T cells, which are predicted to increase based on our data. Thus, we hypothesize that the mechanism of CD4⁺ T cell expansion in SRCs may also cause the counter-inflammatory expansion of Tregs, conceivably representing an important component of immune escape in early SRC formation.

Certain technical considerations impacted our experimental approach and limited the ability to evaluate the TiME. Unlike larger tumors, stage IA occult carcinomas in HDGC are typically less than 1 mm in diameter and multifocal, becoming visible only after fixation and careful histopathologic examination. Isolation of live immune cells for quantification would entail methods too coarse for accurate characterization of SRC-specific ROIs. Instead, we utilized bulk RNA-seq of FFPE and hematoxylin and eosin (H&E)-stained tissue as a technique which allowed for a more granular ROI selection, followed by CIBERSORTx to approximate immune frequencies. Importantly, some immune subsets are not accounted for by CIBERSORTx, including cytotoxic or exhausted CD8⁺ T cells. Other limitations of LCM included low RNA yields in four of our samples below analysis cutoff, as well as the inability to analyze individual SRCs that were scattered or isolated away from larger SRC clusters. Finally, these studies were correlative and did not provide a definitive mechanistic link between tumor dormancy and both TiME and oxidative metabolism. Future studies employing spatially resolved protein-level or transcriptomic techniques to further profile the in situ phenotype of the TiME of SRCs will be necessary to better elucidate the role of individual cells and immune cell subsets in the development of HDGC.

In summary, we describe an elevation of CD4⁺ T cells in SRCs to support a hypothesis of immune-mediated control of early-stage tumor growth, with Treg influx as a potential initiating step in immune evasion. As applied in other hereditary GI cancer syndromes (44–47), strategies for pharmacologic prophylaxis against the development of precancerous lesions may be possible by targeting the early activation of an antitumor immune response in HDGC.

J.C. Patterson reports grants from Cardiff Oncology, Inc during the conduct of the study; grants from Cardiff Oncology, Inc outside the submitted work; in addition, J.C. Patterson has a patent 9566280 issued, licensed, and with royalties paid from Cardiff Oncology, Inc, a patent for 63/077,157 pending, licensed, and with royalties paid from Cardiff Oncology, Inc, and a patent for 63/093,657 pending, licensed, and with royalties paid from Cardiff Oncology, Inc. B.A. Joughin reports a patent for US Patent Application 18/065,348 pending, licensed, and with royalties paid from Cardiff Oncology. M.B. Yaffe reports grants from NIEHS and grants from NCI during the conduct of the study; grants from Cardiff Oncology and personal fees from Applied Biomath outside the submitted work; in addition, M.B. Yaffe has a patent for U.S. Patent Application 18/065,348 pending, licensed, and with royalties paid from Cardiff Oncology; and is the Chief Academic Editor of the AAAS journal Science Signaling. No disclosures were reported by the other authors.

B.L. Green: Conceptualization, resources, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. L.A. Gamble: Conceptualization, resources, data curation, formal analysis, visualization, methodology, writing–original draft, writing–review and editing. L.P. Diggs: Conceptualization, resources, writing–review and editing. D. Nousome: Data curation, software, formal analysis, visualization, writing–review and editing. J.C. Patterson: Data curation, software, formal analysis, writing–review and editing. B.A. Joughin: Data curation, software, formal analysis, writing–review and editing. B. Gasmi: Formal analysis, investigation, visualization, methodology. S.C. Lux: Investigation, visualization, writing–review and editing. S.G. Samaranayake: Investigation, visualization, writing–review and editing. M. Miettinen: Validation, investigation, writing–review and editing. M. Quezado: Investigation, visualization, writing–review and editing. J.M. Hernandez: Supervision, methodology, project administration. M.B. Yaffe: Resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, project administration, writing–review and editing. J.L. Davis: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

This research was supported in part by the Intramural Research Program, NIH, NCI (to J.L. Davis), grant support from No Stomach For Cancer, Inc. (to J.L. Davis), and NIH grants R01-ES015339 (to M.B. Yaffe), R35-ES028374 (to M.B. Yaffe), and R01-CA226898 (to M.B. Yaffe). We thank the department of Molecular Histopathology Laboratory (MHL) at the Frederick National Laboratory for Cancer Research.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.