Stool Protein Mass Spectrometry Identifies Biomarkers for Early Detection of Diffuse-type Gastric Cancer

Gastric cancer (GC) is ranked fourth in prevalence and mortality among cancers worldwide (1). Cancer stage at diagnosis is a critical factor in patient prognosis as 5-year survival rates for localized disease are 70% compared with 6% for patients who present disseminated disease (2). Regular screening programs have proven effective in increasing GC patient survival by promoting early-stage disease detection. In one such case, biennial screening in South Korea for adults more than 40 years old has led to a substantial increase in the diagnosis of GC patients with early-stage disease from 39% in 2001 to 73% in 2016. This is in contrast to the United States, where routine screening is not practiced and only 25% of patients with GC present early-stage disease (3). GC is primarily diagnosed by endoscopy, which is arduous and resource-intensive and has inherent limitations, particularly in the detection of early-stage gastric cancers, which can be morphologically subtle. Therefore, alternative, noninvasive approaches to detect early-stage GC are needed.

GC can be histologically classified as diffuse, intestinal, or mixed subtypes (4). Although intestinal-type GC (IGC) has a better prognosis and decreasing incidence, diffuse-type GC (DGC) has a worse prognosis and rising incidence (5, 6). DGC does not classically undergo a stepwise development of metaplasia or preneoplastic lesions as seen in IGC, further limiting the ability for early detection through endoscopic screening (7). The spontaneous nature of DGC development makes studying the disease difficult; however, in cases of hereditary diffuse gastric cancer (HDGC), patients possess pathogenic germline mutations, predominantly in CDH1, which predispose them to develop DGC (8). HDGC is associated with a cumulative lifetime risk of developing GC between 33% and 83% (9), thereby necessitating routine surveillance and providing an opportunity to study DGC. Because early-stage DGC frequently seems macroscopically normal, regular endoscopic screening and extensive biopsies are required for disease surveillance. Thus, noninvasive approaches to early detection through profiling of tumor-associated biomarkers in bodily fluids are particularly attractive. Indeed, GC patient saliva, serum, and excreta have been shown to be variably enriched for tumor-associated DNA (10, 11), RNA (12–14), and proteins (15–18). Unfortunately, these studies have not resulted in clinically available biomarker assays for GC detection.

Stool is a promising source of GC biomarkers because it is an easily obtainable direct byproduct of the gastrointestinal system. We performed an unbiased protein mass spectrometry screen on stool obtained from genetically engineered mice with mixed-type GC and patients with HDGC to identify potential biomarkers for DGC detection. Because mass spectrometry of stool is not routinely performed, we validated our approach using a genetically engineered mouse model of GC, previously described by our group [referred to as Triple Conditional (TCON) mice] with tumorigenesis driven by oncogenic KrasG12D and loss of p53 and Cdh1 in gastric parietal cells along with a yellow fluorescent protein (YFP) reporter (19). Proteomic analysis of TCON mouse stool revealed a differential abundance of a panel of proteins that correlated with tumor development: actinin alpha 4 (ACTN4), N-acylsphingosine amidohydrolase 2 (ASAH2), dipeptidyl peptidase 4 (DPP4), and valosin-containing protein (VCP). Immunofluorescence analysis of TCON mouse stomachs with early-to-late-stage GC revealed a corresponding protein enrichment in the tissue. Proteomic analysis of stool obtained from patients with HDGC identified proteins that overlapped with the mouse panel, namely, ASAH2, DPP4, and VCP, as well as HDGC-specific proteins lactotransferrin (LTF) and tropomyosin-2 (TPM2) that were enriched compared with sex- and age-matched healthy individuals. ASAH2, which has not been previously associated with GC tumorigenesis, is a neutral ceramidase that catalyzes ceramide to sphingosine, which becomes phosphorylated to sphingosine-1-phosphate (S1P), a potent regulator of cell growth and survival (20). Pharmacologic targeting of ASAH2 reduced the viability of GC cell lines and GC patient-derived organoids (PDO) that were rescued by the addition of S1P. We also characterized the HDGC stool microbiome and identified microbial features, which correlate with patient disease status. Using an unbiased approach, we identified proteins enriched in the stool of GC mouse models and patients with HDGC that correlate with disease and may serve as biomarkers for the early detection of DGC.

Animal studies were approved by the Columbia University Animal Research Compliance. The TCON mouse model of mixed-type GC of our laboratory was previously described (19). TCON and littermate control mice were cohoused when necessary. Fresh stool samples obtained from mice were directly collected into microtubes. For flank tumor studies, TCON cells grown in DMEM, supplemented with penicillin, streptomycin, L-glutamine, and FBS (Gibco), were subcutaneously injected into the flank of mice at 1 × 10⁶ cells. The orthotopic mouse model was generated by injecting C57BL/6 mice with 1 × 10⁶ TCON cells transduced with luciferase lentivirus (Cellomics). For the generation of TCON tumor-derived cell lines, TCON mice stomachs were minced, digested in 2.5 mg/mL collagenase II and 0.5 mg/mL DNAse, and passed through a 40-µm strainer. Cells were grown in complete DMEM media at 37°C with 5% CO₂. The esophageal adenocarcinoma mouse model was previously established by another group (21). Briefly, it was generated by adding deoxycholic acid (Sigma) at 0.3% in the water of L2-IL1β mice for ad libitum consumption starting at 6 weeks of age. N-Methyl-N-nitrosourea (MNU; Spectrum) was added to this water at alternating weeks for five treatment cycles. The azoxymethane (AOM)-induced colorectal cancer model was generated by C57BL/6J mice (Jackson Laboratory) using a method that was previously described by another group (22). Briefly, 8-week-old mice were administered with AOM (Sigma-Aldrich, A5486) via intraperitoneal injection at a dosage of 10 mg/kg body weight. This treatment was repeated weekly for a total duration of 6 weeks, amounting to six injections per mouse. Mice weights were monitored regularly, and mice developed tumors around 20 weeks postinitial AOM treatment. The stool was collected as mice weights declined, indicative of growing tumors. Tumor growth was validated postmortem by examining histology slides of the intestinal tract.

Mice were fed a liquid diet (Bio-Serv) 48 hours prior to euthanasia with isoflurane. To collect gastric fluid, PBS (Corning) was flushed and collected through the stomach. To remove debris, the lavage was centrifuged at 5,000 × g for 5 minutes, and the supernatant was retained. This process was performed three times. The supernatant was processed using a DNA/protein isolation kit (Qiagen). Protein was quantified using DC Protein Assay (Bio-Rad) measured on a SpectraMax iD3. DNA was quantified using a DeNovix DS-11 Spectrophotometer.

The stool was dissociated in RIPA lysis buffer (25 mmol/L Tris pH = 7.4, 2.5 mmol/L EDTA, 1% Triton ×100, 0.1% SDS, 150 mmol/L NaCl) supplemented with 5% SDS and protease inhibitors (ThermoFisher) for 30 minutes at room temperature 20 to 25°C with intermittent vortexing. The solution was centrifuged for 10 minutes at 15,000 × g, and the supernatant was collected, quantified by DC Protein Assay (Bio-Rad) on a SpectraMax iD3, mixed with 6× Laemmli buffer, and boiled at 95°C for 5 minutes. Approximately 40 μg of lysate per sample was loaded onto precast SDS-PAGE gels (GeneScript), and electrophoresis was conducted for 70 minutes at 110 V. Gels were transferred to methanol-activated PVDF membranes for 70 minutes at 110 V. Membranes were blocked in 5% BSA in PBST for 1 hour at room temperature. Antibodies for ACTN4 (Proteintech Group, PTG #19096-1-AP), ASAH2 (PTG 27742-1-AP), DPP4 (PTG #29403-1-AP), LTF (PTG #10933-1-AP), TPM2 (PTG #11038-1-AP), VCP (PTG #10736-1-AP), YFP (Abcam, #ab13970), and α-tubulin (PTG #66031-1-Ig) were diluted 1:2,000 in 5% BSA in PBST and added to membranes for 1 hour at room temperature. Membranes were washed three times with PBST for 10 minutes, incubated with HRP-conjugated secondary antibody (PTG), and diluted 1:4,000 in 5% BSA in PBST for 1 hour at room temperature. Membranes were washed three times with PBST before incubation with ECL reagent (ThermoFisher) and imaged using the iBright 1500CL (ThermoFisher).

For dot blot assays, 10 µg of protein in 3 µL of lysate was spotted onto a nitrocellulose membrane and incubated at room temperature for 60 minutes. Membranes were blocked in 5% BSA in PBST for 1 hour at room temperature. Primary and secondary antibody protocols and imaging are the same as for Western blots.

The stool was collected from patients enrolled in the GI Disease and Endoscopy Registry at Massachusetts General Hospital. Patients were provided an at-home stool collection kit for use 2 days prior to the scheduled endoscopy. The stool was self-collected using a tube containing 200 proof ethanol with an enclosed scoop. Once returned, stool samples were centrifuged at 15,000 × g for 10 minutes and decanted, and the pellet was rinsed in ethanol twice and stored at −80°C. Study participants provided written informed consent. The study protocol was approved by the Mass General Brigham Institutional Review Board (Protocol #2015P000275). Studies were conducted in accordance with the Good Clinical Practice and Declaration of Helsinki.

Samples were lysed in a buffer of 8 mol/L urea, 100 mmol/L NaCl, 50 mL Tris-HCl (pH = 8), and protease inhibitors. Samples were sonicated with a Bioruptor (Diagenode), incubated on ice and centrifuged at 21,000 × g at 4°C. Protein was quantified using bicinchoninic acid (BCA) protein assay (ThermoFisher). Proteins were reduced with 5 mmol/L dithiothreitol for 30 minutes at 55°C and alkylated with 10 mmol/L iodoacetamide in the dark for 30 minutes. Proteins were diluted to 2 mol/L urea with Tris-HCl and digested with trypsin (Promega) at an enzyme:protein ratio of 1:25 for 12 hours at 37°C. The digestion was quenched with 1% trifluoroacetic acid. Samples were desalted using Pierce C18 Tips, 10 µL bed (ThermoFisher). Peptides were brought to 0.33 μg/μL in 0.1% formic acid.

Samples were analyzed with nanoLC/MS-MS. Peptides were separated using an UltiMate3000 (Dionex) HPLC system (ThermoFisher) with a 75 µm ID-fused capillary and packed with 2.4 µm ReproSil-Pur C18 beads to 20 cm. The HPLC gradient was 0% to 35% solvent B (A = 0.1% formic acid, B = 95% acetonitrile, 0.1% formic acid) for more than 20 minutes and from 45% to 95% solvent B in 40 minutes at a flow rate of 300 nL/min. The QExactive HF (ThermoFisher) mass spectrometer was configured following a data-dependent acquisition method. All solvents used in the analysis of MS samples were LC/MS grade. Full MS scans from 300 m/z to 1,500 m/z were analyzed in the Orbitrap at 120,000 FWHM resolution and 5E5 automatic gain control (AGC) target value, for a maximum injection time of 50 milliseconds. Ions were selected for MS2 analysis with an isolation window of 2 m/z for a maximum injection time of 50 milliseconds and a target AGC of 5E4. MS raw files were processed with Proteome Discoverer version 2.3, and MS spectra were searched against a target + reverse database with the SEQUEST search engine using Homo sapiens FASTA databases (reviewed, canonical entries, downloaded September 27, 2021). For global proteome samples, iBAQ quantification was performed on unique + razor peptides. Trypsin cleavage was specific with up to two missed cleavages allowed. Match between run parameters included a retention time alignment window of 20 minutes and a match time window of 0.7 mins. The false discovery rate was set to 0.01.

Stool supernatant was harvested after centrifugation at 15,000 × g for 10 minutes and reduced using 5 mmol/L dithiothreitol/25 mmol/L NH4HCO3 pH = 8.0 for 60 minutes at 52°C. The solution was stored at room temperature in 25 mmol/L iodoacetamide in the dark for 60 minutes. Protein digestion was performed with S-Trap™ Column (ProtiFi). Nano high-performance liquid chromatography analyses were performed using an Easy n-LC 1000 system (Thermo Fisher). A Q Exactive HF-X (Thermo Fisher) was used for MS analyses and operated with Xcalibur. Full MS scans were acquired on the Orbitrap from 350 m/z to 1,200 m/z at a resolution of 60,000 using an AGC value of 5 × 10⁵. The minimum threshold was 50,000 ion counts. MS-MS data were collected in centroid mode in the ion trap using an isolation width of 1.6 m/z units, a maximum injection time of 50 milliseconds, and an AGC value of 1 × 10⁴. Raw data files were analyzed using the Proteome Discoverer software (Thermo Fisher) against Mus musculus database.

Stomachs were dissected, rinsed in PBS, and incubated in 10% formalin overnight at room temperature before being rinsed with PBS, embedded in paraffin and sectioned by the Columbia University Molecular Pathology Shared Resource. Antigens were unmasked by sequential 5-minute incubations in the following solutions: 100% xylene, 100% EtOH, 95% EtOH, 80% EtOH, 70% EtOH, 50% EtOH, and 100% water. Slides were pressurized and heat-treated for 1 minute in an unmasking solution (Vector Labs, H3300). Slides were rinsed in PBST and then blocked in 5% BSA PBST for 1 hour at room temperature. The primary antibodies listed above were incubated per manufacturer recommendations in 5% BSA PBST for 1 hour at room temperature. Slides were rinsed five times with PBST and then incubated with fluorescent antibodies for 1 hour at room temperature in the dark. Slide covers were mounted using DAPI mounting gel (Electron Microscopy Sciences) incubated overnight at 4°C. Images were obtained using a Nikon TI2E AXR confocal microscope.

Human gastric adenocarcinoma specimens, namely, CCLF_UPGI_0008_T and CCLF_UPGI_0061_T, were obtained from consenting patients at the DFCI under institutional review board protocol DF-HCC 03-189. The organoid model generation process was done by the Cancer Cell Line Factory team at the Broad Institute. All procedures were carried out in accordance with an institutional review board-approved protocol. Patient tumor resections were immediately placed in a sterile conical tube containing DMEM media (Thermo Fisher) supplemented with 10% FBS (Sigma-Aldrich), 1% penicillin–streptomycin (Thermo Fisher), 10 μg/mL gentamicin, and 250 ng/mL fungizone. This media was maintained at a cold temperature 4°C during transport from the operating room to the research laboratory. Upon arrival, resections were transferred to a 15-mL conical tube containing 5 mL of DMEM media with 10% FBS, 1% penicillin–streptomycin, and digestion enzymes including regular collagenase (STEMCELL, #07912) and dispase (STEMCELL, #07913). The tube was subjected to rotation and incubated at 37°C for 1 hour. Subsequently, the cells were centrifuged at 1,000 rpm for 5 minutes. The resulting cell pellets were resuspended and subsequently embedded into Matrigel (Corning, #356231) following an established protocol (23). GC organoids were maintained and passaged using cold PBS and TrypLE Express (Thermo Fisher, #12604039) upon reaching 80% to 90% confluence. Tumor purity was verified by whole-exome sequencing.

Gastric epithelial adenocarcinoma cell lines AGS (female patient; RRID: CVCL_0139; CRL1739) and NCI-N87 (male patient; RRID: CVCL_1603; CRL5822) were obtained from ATCC, which performs short tandem repeat profiling on its cell lines. Cells were not authenticated again after purchase from ATCC. MKN45 gastric adenocarcinoma cell line (female patient; RRID: CVCL_0434; ACC409) was obtained from DSMZ Leibniz Institute, which performs short tandem repeat profiling of its cell lines. Cells were not authenticated again after purchase from ATCC. AGS, NCI-N87, and MKN45 cell lines were expanded and tested for mycoplasma using MycoAlert PLUS Mycoplasma Detection Kit (Lonza, LT07-318) upon receipt from ATCC and DSMZ before aliquoting in freezing media CELLBANKER 1 (Amsbio LLC, 11910) and stored in liquid nitrogen. Each cell line aliquot was thawed and used for experiments described herein after 1 week in culture, and cells were used for eight to ten passages. The 2D cell culture assays were maintained in RPMI media supplemented with 10% FBS and 1% penicillin, streptomycin, and glutamine (Thermo Fisher) and grown in an incubator at 5% CO₂ at 37°C. Small molecules C6 urea ceramide (C6-U-Cer), D-erythro-MAPP, ARN14988, ceramide, sphingosine, and S1P were solubilized as recommended by Cayman Chemicals. For viability assays, 96-well plates were seeded at 2,000 cells per well and allowed to recover overnight, and compounds were added and incubated at 5% CO₂ at 37°C for 72 hours. Viability was measured using Cell-Titer Glo viability reagent (Promega), and luminescence was measured using SpectraMax iD3.

PDOs were obtained from the Broad Institute and maintained in 50 µL Matrigel (Corning, #354234) in 24-well plates and cultured in 5% CO₂ and at 37°C. After Matrigel solidification, organoids were overlaid with DMEM/F12 (Gibco, #11320-033) and supplemented with 1X B27 (Gibco, #0080085SA) and 1× N2 (Gibco, #A13707-01); Wnt3A-, m-Noggin-, and R-spondin-conditioned media (10% total for all three); and 3% L-WRN-conditioned media (Sigma-Aldrich, #SCM105). Organoid media included the following additional growth factors, cofactors, and hormones: 50 ng/mL human epidermal growth factor (Sigma-Aldrich, #E9644), 1 mmol/L N-acetylcysteine (Sigma-Aldrich, #A9165), 20 mmol/L nicotinamide (Sigma-Aldrich, #N0636), 2 nmol/L human gastrin I (Sigma-Aldrich, #G9145), 10 mmol/L SB202190 (Sigma-Aldrich, #S7067), 10 ng/mL fibroblast growth factor-basic (PeproTech, #100-18B), 10 ng/mL FGF10 (PeproTech, #100-26), 1 µmol/L prostaglandin E2 (Tocris Bioscience, #2296), and 1 µmol/L A83-01 (R&D Systems, #2939). Organoids were passaged weekly, during which Matrigel was removed using Cell Recovery Solution (Corning, #354253) following the manufacturer’s instructions and split at a ratio of 1:4 to 1:8 with fresh Matrigel. PDOs were incubated with C6-U-Cer for 72 hours, and viability was measured using WST1 viability reagent and measured using SpectraMax iD3.

The stool proteome and microbiome data generated in this study can be requested from the corresponding author.

Few studies have investigated the use of stool proteins as biomarkers for GC in an unbiased manner; thus, we utilized our previously described transformed gastric parietal cell-specific KPC mouse model of GC, referred to as TCON, as a discovery platform (19). We hypothesized that stool biomarkers would originate from biomolecules shed into the stomach lumen either by tumor or nontumor cells affected by the presence of the tumor. To assess this, gastric lavage was collected from 5-week-old to 9-week-old TCON and littermate control mice and measured for total DNA and protein concentration (Fig. 1A). Substantially higher DNA and protein concentrations were observed in the gastric lavage obtained from TCON mice compared with littermate controls. The YFP reporter expressed in ATP4b-Cre-expressing transformed gastric parietal cells represents a surrogate for GC cell-derived proteins in the TCON mouse model. We probed for YFP in gastric lavage fluid from TCON mice by Western blot, revealing robust expression (Fig. 1B), which suggests that tumor cell-derived proteins are being released into the GI tract and possibly into the stool. To assess this, the stool lysates from TCON and littermate control mice were probed for YFP, revealing enrichment in TCON mouse stool but not littermate controls (Fig. 1C). The ability to detect tumor cell YFP in the stool of TCON mice suggests that other protein biomarkers may be present in the stool.

Previous stool mass spectrometry studies demonstrate that protein identification is variable based on the lysis buffer composition (24); thus, we utilized two buffer types that maximize protein discovery: 8 mol/L urea and 5% SDS. We selected 6-week-old TCON mice as our window for early detection because mice display YFP-positive tumors in the stomach without signs of metastasis (19) and cohoused them with littermate control mice to normalize their microbiome (25). Stool samples from 6-week-old TCON and littermate control mice were lysed and analyzed by mass spectrometry. A Venn diagram of identified stool proteins based on the lysis buffer used (Supplementary Fig. S1) suggests that certain proteins are enriched in the stool depending on the lysis buffer and the presence of GC tumors. Additionally, previously published mass spectrometry analysis of WT mouse stool proteins (26) was considered with our control stool proteins to validate our approach. A total of 765 proteins were identified by stool protein mass spectrometry from n = 2 littermate controls and n = 4 TCON mice. The 10 most enriched and depleted proteins (Fig. 2A) from the stool of TCON mice are listed as candidate biomarkers for GC detection. We focused on proteins enriched in TCON stool and validated them by probing stool from 7-week-old to 9-week-old mice with commercially available antibodies. Multiple proteins were found to be consistently enriched in stool from TCON mice compared with littermate controls: ACTN4, VCP, ASAH2, and DPP4 (Fig. 2B). This data identifies a panel of proteins that specifically correlates with the stool from the TCON mouse model of GC, which may serve as biomarkers for gastric cancer detection.

Next, we sought to understand the kinetics of stool biomarker protein expression in our TCON mouse model to evaluate its effectiveness in early detection. We conducted a weekly collection of stool from TCON and littermate control mice from 4 weeks to 8 weeks of age. As previously published (19), by 3 weeks of age, TCON mice have dysplastic lesions throughout the stomach. By 6 weeks of age, there is a substantial increase in lesion size with some localized invasion, whereas by 9 weeks of age, 100% of mice have invasive carcinomas with a median overall survival of 11 weeks. Immunoblot analysis of stool lysates identified biomarker protein expression as early as 4 weeks with increasing abundance during disease progression (Fig. 3A). Next, we assessed the expression of stool biomarker proteins in stomach tumors. Stomachs isolated from 6-week-old TCON and littermate control mice were sectioned and analyzed by immunofluorescence for biomarker protein expression along with YFP to indicate tumor cells (Fig. 3B; Supplementary Fig. S2). ACTN4 and DPP4 were identified as remarkably upregulated, whereas ASAH2 and VCP were noticeably enriched. The expression of ACTN4 and DPP4 was examined in the stomach tissue of TCON mice at 4 and 8 weeks of age, which mirrored previous observations that biomarker protein expression concentrated at the areas of YFP-positive tumor growth, in contrast to control tissue with general diffuse expression (Supplementary Fig. S3). Altogether, the data suggest that our stool biomarkers are upregulated in stomach tumors.

To further characterize the stool biomarkers, we examined their presence in the stool of multiple mouse models of gastrointestinal cancer. A commonly used experimental model of tumorigenesis is the subcutaneous injection of tumor cells into the flanks of mice. This model allows for convenient monitoring of tumor growth but has limited physiologic relevance because tumors lack the appropriate microenvironment. Subcutaneous implantation of TCON-derived GC cells into the flank of syngeneic C57BL/6 mice did not result in the biomarker expression of ACTN4 in the stool at early or late stages of tumor growth (Fig. 4A). In contrast, orthotopic injection of luciferase-expressing TCON GC cells into the stomachs of syngeneic C57BL/6 mice resulted in notable enrichment of ACTN4 and DPP4 in the stool after the tumor presence was confirmed by luminescence imaging (Fig. 4B). The difference in stool biomarker presence between subcutaneous and orthotopic GC cell implantation suggests a dependence on location for stool biomarker detection, such as, along the GI tract. Published studies have compared GC with Barrett’s esophagus/esophageal adenocarcinoma (EAC) and indicated a similar cell of origin (21) and molecular fingerprint (27). To explore whether our GC-correlated biomarkers also identify the presence of EAC, we probed for biomarker expression in stool from a previously established model of EAC generated by treating mice overexpressing IL1β in the esophageal epithelium with N-methyl-N-nitrosourea (MNU) and deoxycholate (DCA) administered in the drinking water (21). Immunoblot analysis of stool from EAC mice after MNU and DCA treatment revealed the enrichment of ACTN4 and DPP4 in the stool compared with stool from control mice without treatment (Fig. 4C). Colorectal cancer (CRC) is the only GI cancer with an FDA-approved stool test, potentially owing to reduced biomarker degradation at the posterior GI tract. Studies have suggested the role of ACTN4 and DPP4 in the tumorigenesis of CRC (28, 29); therefore, we wanted to examine the efficacy of these proteins as biomarkers for CRC. We analyzed the stool of mice treated with azoxymethane (AOM), a model frequently used to study spontaneous CRC (30). Mice developed tumors approximately 20 weeks after the initial AOM treatments, indicated by weight loss and posthumously confirmed by histology. Compared with untreated controls, we identified a substantial correlation of ACTN4 and DPP4 in the stool of AOM-driven CRC mouse models compared with sex- and age-matched untreated healthy controls (Fig. 4D). Collectively, these observations suggest that our stool biomarkers correlate with tumorigenesis along the GI tract and are potential biomarkers in multiple types of GI cancers.

Currently, the only FDA-approved use of stool in cancer diagnostics is for colorectal cancer screening. Previous studies of stool biomarkers for GC detection were targeted and limited in outcome (17, 18). We performed an unbiased stool mass spectrometry approach based on the conditions utilized in our mouse studies for biomarker discovery in the stool of patients with HDGC. The stool was collected from healthy donors (H) and patients with HDGC with CDH1 germline mutations stratified into an at-risk group (HDGC-AR) if GC was not detected by endoscopy or as confirmed GC group (HDGC-GC) if endoscopy confirmed the presence of GC at the time of stool collection. Donor information is summarized in Supplementary Table S1. The stools from our cohorts were analyzed by mass spectrometry, which identified 834 proteins in total. The 20 most up- and down-regulated proteins in the stool of patients with HDGC compared with healthy donor stool are listed (Table 1). Selected proteins were used to generate heat maps of stool proteins enriched in HDGC-GC and HDGC-AR groups (Supplementary Fig. S4A) and diminished stool proteins in the HDGC-GC group compared with healthy donors (Supplementary Fig. S4B). These data reveal distinct stool proteomes, which correlate with the donor tumor status. Additionally, a panel of proteins was identified that distinguishes the HDGC-GC group from the HDGC-AR group (Supplementary Fig. S4C) and may represent proteins specific to the presence of more established lesions. Validation of biomarkers enriched in the stool of patients with HDGC compared with healthy donors by immunoblot assay identified a panel of proteins associated with the presence of the disease: ASAH2, DPP4, VCP, lactotransferrin (LTF), and tropomyosin-2 (TPM2; Fig. 5A). Similar to our findings in TCON mice, we sought to determine if the stool biomarkers enriched in patients with HDGC are upregulated in the stomach. The primary tumor proteome of 84 DGC patients compared with healthy adjacent tissue has been previously characterized (31), and searching for our biomarkers in their data confirmed DPP4, LTF, and VCP as substantially upregulated (Fig. 5B). We next validated the HDGC stool biomarkers in a simplified version of an immunoblot analysis, commonly referred to as a dot blot assay, to assess the feasibility of rapid screening. Dot blot assay revealed a robust detection of ASAH2 and DPP4 in the stool of donors with endoscopy-confirmed GC tumors compared with healthy donors or HDGC donors without tumors (Fig. 5C). These data support the feasibility of stool protein diagnostics for HDGC and suggest potential candidate proteins for further study and translation to the clinic.

ASAH2 is a stool biomarker identified in our screen that has not been previously associated with GC tumorigenesis. ASAH2 is a neutral ceramidase that converts ceramide into sphingosine, after which it is phosphorylated by kinases into sphingosine-1-phosphate (S1P). S1P is a protumorigenic lipid secreted by many cancer cells and recognized by S1P receptors, which are G protein-coupled receptors that promote signaling through canonical growth pathways such as RAS, P13K, and STAT3 (20). Multiple tumor types have been demonstrated to exploit S1P signaling to promote tumorigenesis and metastasis, including liver (32), breast (33), colon (34), pancreatic (35), and GC (36). To test the role of ASAH2 in GC, we treated multiple GC cell lines with C6-U-Cer, a small molecule inhibitor of ASAH2 (Fig. 6A). GC cell lines were sensitive to ASAH2 inhibition with apparent complete toxicity at 5 µmol/L C6-U-Cer. There are five human ceramidases: one acid ceramidase (ASAH1), three alkaline ceramidases (ACER1-3), and one neutral ceramidase (ASAH2). To determine which ceramidases were critical for GC cell viability, we challenged the IGC cell line NCI-N87 and the DGC cell line MKN45 with inhibitors specific for neutral (C6-U-Cer), acidic (ARN14988), and alkaline (D-erythro-MAPP) ceramidases (Fig. 6B). Only the inhibition of neutral ceramidase ASAH2 induced cell death, suggesting that among the ceramidases, ASAH2 is critical for GC viability.

ASAH2 activity is downstream of ceramide and upstream of sphingosine in the S1P synthesis pathway; therefore, we hypothesized that the addition of downstream products of ASAH2 may rescue the antiproliferative effects of C6-U-Cer. GC cell lines MKN and NCI-N87 were treated with C6-U-Cer immediately followed by the addition of either ceramide, sphingosine, or S1P (Fig. 6C; Supplementary Fig. S5). Indeed, the addition of ceramide did not affect C6-U-Cer toxicity; however, the addition of downstream products sphingosine and S1P considerably restored GC cell viability. These data suggest that ASAH2 has a functional role in the survival of GC cell lines, possibly through S1P signaling. PDO more closely model an in vivo tumor compared with 2D cell lines; thus, we treated GC PDOs with C6-U-Cer and determined that GC PDOs are similarly sensitive to the ASAH2 inhibitor (Fig. 6D). PDO biopsy information (Supplementary Table S2) suggests that ASAH2 inhibition is effective against primary and peritoneal cavity metastasis tumors. These data support further investigation of ASAH2 inhibition as a targeted therapy for GC.

The microbiome plays a crucial role in human metabolism and consists of diverse symbionts that may influence the composition of proteins in the stool (37). In an exploratory analysis, we characterized the microbiome of patients with HDGC by 16S rRNA sequencing of patients with HDGC and healthy donor stool. Generalized examination of the bacterial composition was done by hierarchical clustering (Fig. 7A) and Shannon entropy measurements of species richness and evenness (Fig. 7B), which did not identify notable differences between the patient stool cohorts. However, comparisons of the relative abundance of bacterial families in the stool (Fig. 7C) identified a notable increase in members of the Streptococcaceae family among HDGC-GC patient stool (Fig. 7D), which aligns with observations in GC patients (38). Furthermore, analysis of bacterial species abundance between patient stool cohorts identified multiple remarkably differentially enriched species (Fig. 7E). Phascolarctobacterium faecium, Ruminococcus bromii, and Escherichia marmotae were considerably upregulated in HDGC-AR and HDGC-GC stool samples; meanwhile, Bacteroides xylanisolvens and Metalysinibacillus jejuensis are enriched in the stool of healthy donors. These data highlight several microbiome trends in HDGC and healthy donor stool and suggest that the stool microbiome may be a potential indicator of HDGC biomarkers.

Our study reports the identification of stool protein biomarkers in mouse models of GC and patients with HDGC through an unbiased proteomic mass spectrometry approach. The difficulty of studying biomarkers for DGC is the spontaneous nature of the disease and the advanced stage at which GC patients are diagnosed. Our GC mouse model serves as a valuable platform for discovery. The TCON mouse model of mixed-type GC reliably develops GC at established kinetics, which allowed us to optimize and validate stool mass spectrometry for the discovery of proteins in the stool that correlate with disease. We leveraged our mass spectrometry conditions to analyze the stool of patients with HDGC who reliably develop DGC and are routinely monitored for GC tumorigenesis. Immunoblot analysis validated protein candidates enriched in stool from TCON mice and patients with HDGC, confirming a correlation between disease and protein upregulation. Additionally, many biomarkers overlapped between TCON mice and patients with HDGC, suggesting a conserved feature of GC tumorigenesis in which proteins are enriched in the GI tract and ultimately end up in the stool. Our mouse and human data offer proof of principle for coupling preclinical systems and annotated patient cohorts to optimize detection strategies. For the first time, we examine the stool proteome in patients with HDGC and nominate several candidate proteins as biomarkers for disease detection.

The identification of proteins known to promote GC tumorigenesis such as ACTN4 (39), DPP4 (40), and VCP (41) validates our unbiased approach; however, a subset of proteins have not been previously associated with GC. These potentially represent proteins with undiscovered roles in GC tumorigenesis. Because of the prevalence of S1P in tumor biology, the identification of ASAH2 as a stool biomarker motivated the further characterization of its role in GC. Studies characterizing the protumorigenic functions of ASAH2 include protecting myeloid-derived suppressor cells from ferroptosis (42) and its importance in colorectal cancer tumorigenesis (43). Herein, we demonstrate that ASAH2 activity is important in the proliferation of GC cell lines and PDOs derived from a primary tumor and a peritoneal cavity metastasis. Further studies will determine if ASAH2 may be a therapeutic target for GC; nonetheless, its importance in GC cell lines in vitro is promising as validation for the proteins identified from our unbiased stool biomarker study.

We observed a subset of our stool protein biomarkers, originally discovered in GC, to correlate with tumorigenesis in multiple mouse models of GI cancer. ACTN4 and DPP4 have been implicated in the tumorigenesis of multiple tumor types, including GC, EAC, and CRC (44, 45); therefore, these proteins may act as stool biomarkers for multiple GI cancers. Another explanation is that the presence of the GI tumor induces nontumor cells along the GI tract to behave differently, resulting in changes in stool protein levels, leading to the diverse correlations with our GC biomarker panel with multiple GI tumor types. More studies need to be conducted to determine the organ(s) and cell type(s) responsible for mediating the stool biomarkers.

Mass spectrometry studies of the GC tumor proteome from patient biopsies have identified differentially expressed proteins and potential biomarkers (31, 46–48); however, our stool biomarkers are not among their most upregulated proteins. One explanation is that proteins enriched in GC tumors are abundantly expressed by healthy tissue throughout the GI tract, limiting its usefulness as a stool biomarker. Additionally, GC-specific proteins may be metabolized by the host or microbiome, preventing their accumulation in the GI tract and stool. Our approach avoids these and other unknown variables by quantifying protein species directly from stool. This is highlighted by our immunofluorescence images of mouse stomachs (Fig. 3B), which did not detect any statistical difference between control and TCON stomach expression levels of ASAH2 and VCP; however, stool protein levels show a robust correlation. The mechanisms regulating the accumulation and stability of our stool biomarker proteins through the digestive process have not yet been elucidated. We hypothesize the possibility that some biomarker proteins are protected in membrane-bound extracellular vesicles, which escort the proteins through the digestive system. Our future studies will aim to identify how the biomarker proteins survive degradative pressures in the GI tract.

Our exploratory analysis of the HDGC stool microbiome characterized microbial features, which distinguishes donors based on disease state. We hypothesize that these bacteria may influence the presence and stability of proteins in the GI tract, altering the stool protein composition. Interestingly, several bacterial species that correlated in the stool from non-GC donors are enriched in patients with cancer who respond to anti-PD1 therapy: P. faecium (49), B. xylanisolvens (50), and R. bromii (51). Additionally, we found E. marmotae, a potential enteric human pathogen (52), to be enriched in the stool of patients with GC-confirmed HDGC. The functional importance of these microbial features and whether the microbiome influences the availability of protein biomarkers remain topics for future studies; however, our data suggest that the microbial features themselves may be further studied as potential GC biomarkers.

Our study focuses on characterizing stool proteins upregulated in the context of GC; however, downregulated stool proteins may also serve as potential indicators of disease. Interestingly, many of the downregulated stool proteins we identified are digestive enzymes. It has been shown that loss of reference digestive function is associated with GC (53), which reflects the reduced abundance of digestive enzymes in our stool proteome findings. The stool proteins we identified to be downregulated in the context of GC may indicate a healthy state and remain a prospective area of study.

The stool protein biomarkers identified in this study, which correlate with disease in patients with HDGC, have potential clinical implications for the early detection of GC. However, we could not validate many of our top hits due to the lack of commercially available antibodies. The use of diagnostic mass spectrometry may avoid the problem of antibody availability and is a promising area of study. Nonetheless, assaying our validated panel of candidate biomarker proteins in stool samples suggests the possibility for high-throughput routine testing of individuals at high risk for GC to inform them when endoscopy is warranted. These populations include those infected with H. pylori, ethnicities, which are prone to developing GC, and families with a history of or individuals with known genetic predispositions to developing GC. Additionally, these stool biomarkers may have the potential to serve as indicators of patient response to tumor therapies, which are currently assessed by arduous tumor biopsies.

This study had a limited number of stool donors from Boston. Further studies are needed to expand our observations to more populations.

Through an unbiased mass spectrometry screen of stool, we demonstrate the potential of stool biomarkers as indicators of disease in patients with HDGC and GC mouse models. The overlap of biomarkers between mouse and human stool suggests conserved features of stool biomarker enrichment and supports the possibility of characterizing this phenomenon in the TCON mouse model.

D.A. Drew reports grants from the Entertainment Industry Foundation and Stand Up To Cancer (SU2C)/AACR during the conduct of the study. S.J. Klempner reports personal fees from Astellas, Amgen, Novartis, Pfizer, Merck, Sanofi-Aventis, Bristol Myers Squibb, AstraZeneca, Daiichi Sankyo, Taiho, and Exact Sciences, and personal fees from Mersana outside the submitted work; and serves as a Panel Member for NCCN Guidelines. D.C. Chung reports personal fees from Guardant Health, grants from Janssen, Iterative Scopes, and Immunovia, and personal fees from UpToDate outside the submitted work. A.T. Chan reports grants and personal fees from Pfizer and personal fees from Boehringer Ingelheim outside the submitted work. S. Ryeom reports grants from SU2C during the conduct of the study and grants from ImmPACT Bio outside the submitted work. No disclosures were reported by the other authors.

C.-L.C. Ho: Conceptualization, resources, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, writing–review and editing. M.B. Gilbert: Conceptualization, validation, investigation, methodology, writing–original draft, writing–review and editing. G. Urtecho: Validation, investigation, methodology, writing–review and editing. H. Lee: Validation, investigation, methodology, writing–review and editing. D.A. Drew: Resources, investigation, methodology, writing–review and editing. S.J. Klempner: Resources, investigation, writing–review and editing. J.S. Cho: Resources, investigation, methodology, writing–review and editing. T.J. Ryan: Investigation, methodology. N. Rustgi: Investigation, methodology. H. Lee: Resources, investigation, methodology. J. Lee: Resources, writing–review and editing. A. Caraballo: Resources, writing–review and editing. M.V. Magicheva-Gupta: Resources. C. Rios: Resources, methodology. A.E. Shin: Investigation. Y.-Y. Tseng: Resources, methodology, writing–review and editing. J.L. Davis: Conceptualization, resources, methodology, writing–review and editing. D.C. Chung: Conceptualization, resources, writing–review and editing. A.T. Chan: Resources, funding acquisition, writing–review and editing. H.H. Wang: Conceptualization, resources, supervision, funding acquisition, methodology, writing–review and editing. S. Ryeom: Conceptualization, resources, supervision, funding acquisition, methodology, writing–review and editing.

We thank Joy Hang Che for data analysis advice. We thank Yasmin Kadry and Nancy Shek for engaging in topic-related discussion. Heat map analysis was performed on Morpheus (https://software.broadinstitute.org/morpheus). Histology and immunofluorescence analysis of mouse stomachs were performed by the Molecular Pathology Shared Resource at the Herbert Irving Comprehensive Cancer Center at Columbia University Irving Medical Center. This research was supported by the Stand Up to Cancer Gastric Cancer Interception Research Team Grant (Grant Number: SU2C-AACR-DT-30-20) Award (A.T. Chan, D.A. Drew, S.J. Klempner, J. Lee, S. Ryeom). This research grant is administered by the American Association for Cancer Research, Scientific Partner of SU2C, and DeGregorio Family Foundation (S.J. Klempner). This manuscript was supported by the SKKU Excellence in Research Award Research Fund, Sungkyunkwan University, 2022 (J. Lee). D.A. Drew is supported by the National Institutes of Health award K01DK012. A.T. Chan is an American Cancer Society Clinical Research Professor supported by R35 CA253185. H.H. Wang acknowledges funding support from the NIH NCI (Grant no. 1R01CA272898). G. Urtecho was supported by the HHMI Hanna H. Gray Postdoctoral Fellowship (GT15182).