Pancreas

A fusion-based deep-learning algorithm predicts PDAC metastasis based on primary tumour CT images: a multinational study

Xue N, Sabroso-Lasa S, Merino X, et al. A fusion-based deep-learning algorithm predicts PDAC metastasis based on primary tumour CT images: a multinational study. Gut 2025; 74: 2024-2034. doi: 10.1136/gutjnl-2024-334237.

The majority of patients (52%) with pancreatic ductal adenocarcinoma (PDAC) have metastasis at diagnosis, making surgical resection unfeasible. Current imaging modalities (CT, MRI and PET-CT) have limitations in diagnosing small, inconspicuous metastasis particularly in metabolically active organs. These alongside clinical tools, biomarkers, and pathology, remain labour-intensive.

Xue et al., designed a Pancreatic Tumor Metastasis Prediction Deep-learning (PMPD) algorithm to predict distant metastasis based on PDAC contrast-enhanced CT scan (CECT) images with hope for it to be used as an adjunct by radiologists to improve the accuracy of diagnosis. This international multicentre study involved 661 patients with PDAC across four international datasets. They showed that the model could accurately predict the presence or absence of metastasis with area under the receiver operating curve (AUROC) up to 0.806, in external validation. In one dataset, the model identified 56% of metastases that were not visible on imaging. Accuracy was unaffected by metastasis location, tumor site and size, patient demographics and CT manufacturer. PMPD derived Metastasis Risk Score (MRS) prediction was the most potent prognostic factor outperforming resectability status and Carbohydrate Antigen 19-9 (CA 19.9). PMPD accurately predicted 65.6% of the metastasis that was discovered during surgery and not detected by radiographic imaging. 

This can help in individualizing therapy and conserve resources. It was also shown to have a potential role in predicting future early metastasis. Limitations include the lack of a true gold standard for metastasis detection, reliance on manual segmentation, dataset imbalances that may affect generalizability, and moderate predictive accuracy due to missing clinical variables.

Hepatology

Non-invasive risk-based surveillance of hepatocellular carcinoma in patients with metabolic dysfunction-associated steatotic liver disease

Lai J, Yang B, Lee H, et al. Non-invasive risk-based surveillance of hepatocellular carcinoma in patients with metabolic dysfunction-associated steatotic liver disease. Gut 2025; 74: 2050-2057. doi: 10.1136/gutjnl-2025-334981
 
Hepatocellular carcinoma (HCC) is an increasingly recognised complication of metabolic dysfunction–associated steatotic liver disease (MASLD), arising in both cirrhotic and non-cirrhotic livers. Current guidelines recommend HCC surveillance for individuals with an annual risk above 1%, typically those with cirrhosis. However, observational studies show that many MASLD-related HCC cases develop before cirrhosis.

Lai et al. highlighted the need for non-invasive tools to improve risk stratification and assessed the Fibrosis-4 (FIB-4) index and vibration-controlled transient elastography liver stiffness (LSM) for predicting HCC risk. In this multinational longitudinal study across 16 centres, 12,950 adults with MASLD and available FIB-4 and LSM results were followed for a median of 47.7 months, during which 109 participants (0.8%) developed HCC.

HCC incidence increased stepwise across FIB-4 categories: 0.07% annually for scores below the age-adjusted cut-off (<1.3 if <65 years; <2.0 if ≥65 years), 0.17% at intermediate levels, 0.77% for scores 2.67–3.25, and 1.18% for scores ≥3.25. For LSM, the annual incidence exceeded the surveillance threshold of 1% at values ≥20 kPa. Using a two-step approach, patients with elevated age-adjusted FIB-4 and LSM ≥15 kPa also reached an annual incidence above 1%. LSM performance was less consistent in individuals with BMI (body mass index) ≥30 kg/m². Women consistently had uniformly lower HCC incidence. 

The study concludes that FIB-4 ≥3.25 or LSM ≥20 kPa should prompt HCC surveillance in MASLD. Similarly, for a tiered strategy of elevated FIB-4 and LSM ≥15 kPa.

Although the study provided evidence-based cut-offs to refine HCC risk stratification. FIB-4 and LSM were assessed at a single time point, and the study did not evaluate whether serial measurements improve risk prediction or indicate when surveillance might be discontinued. The relatively short 47-month follow-up may underestimate long-term MASLD-related HCC risk and affect the proposed cut-offs. Further research, including local cost-effectiveness analyses, is needed before widespread adoption.

Intrahepatic donor microbiota-based metataxonomic signature detected in organ preservation solution enables prediction of short-term liver transplant outcomes

Lucas-Ruiz F, Vidal-Correoso D, Mateo S, et al. Intrahepatic donor microbiota-based metataxonomic signature detected in organ preservation solution enables prediction of short-term liver transplant outcomes. Gut 2025; 74: 2058-2069. doi: 10.1136/gutjnl-2025-335986. 

Biomarkers predicting graft survival have become a trend in transplant translational research, coinciding with rising interest in microbiome research, not only in various inflammatory or immune-related pathologies but also in transplant-related outcomes.  This study examines the current trends in microbiomes and biomarkers analysed in machine learning algorithms that predict liver transplant outcomes.  

The study was conducted at a transplant centre in Spain from July 2019 to July 2022, examining a microbiota-based metataxonomic signature in the organ preservation solution (OPS) of 110 liver transplant donors.  Then, the clinical outcomes of the graft recipients were followed up and linked to the microbiome footprints in OPS to predict the outcomes.   A further analysis of 29 additional transplant recipients and donors was conducted to enhance the robustness of the existing data.  The clinical outcomes of liver transplant in the monitoring include the common life-threatening complications predicting graft survival, such as early allograft dysfunction, acute graft rejection, hepatic artery thrombosis, biliary complications, and 1-year global survival.  The hypo- and hyper-abundant clusters of genes are linked to the microbial profile of OPS associated with liver transplant complications.  The hyperabundant genera correlate with lipid metabolism and immune activation, while the hypoabundant ones are associated with oxidative stress and vascular signalling.  It was found that certain microbiota genres in OPS can possibly predict the graft survival and other complications. The OPS-derived microbial metataxonomic structure is closely linked to the signature of liver/bile, such as Proteobacteria and certain genera, including Bacillus and Prevotella, which can be associated with adverse outcomes (p<0.05).  

The study highlights the necessity of further research into potential biomarkers linked to microbial DNA analysis, while machine learning algorithms facilitate the handling of large-scale datasets and the examination of multicentre studies.

Therapeutic inhibition of HBsAg and HBV cccDNA through a novel phased combination treatment: glycine and interferon-α

Lin C, Huang Y, Ran N, et al. Therapeutic inhibition of HBsAg and HBV cccDNA through a novel phased combination treatment: glycine and interferon-α. Gut 2025; 74: 2035-2049. doi:10.1136/gutjnl-2025-334813.

Chronic hepatitis B affects approximately 254 million people globally. A clinical cure remains elusive with clearance of the HBV viral reservoir covalently closed circular DNA (cccDNA) remaining a challenge. Glycine is a widely available, cheap compound used as a health supplement due to its role in multiple metabolic processes. This paper by Lin et al., identified glycine as both a novel non-invasive biomarker for hepatic injury (in the context of HBV (Hepatitis B virus) infection) and a therapeutic agent addressing cccDNA persistence through a unique mechanism.

Serum of 496 HBV-infected patients was analysed and the serum glycine was able to differentiate HBeAg (hepatitis B e antigen)-positive chronic hepatitis B (CHB) from healthy controls. Levels strongly correlated with ALT (Alanine transaminase) and liver inflammation. They infer this is due to glycine levels reflecting hepatocyte integrity, with intracellular glycine reserves leaking into the blood as hepatocytes are damaged. 

Treatment with glycine in a mouse model successfully reduced cccDNA levels by 50% by activating the mTORC1 (mechanistic Target of Rapamycin Complex 1) pathway. This induces hepatocyte proliferation thus dilutes the cccDNA reservoir during mitosis. Glycine administration demonstrated anti-inflammatory properties by blocking NF-kB (Nuclear Factor kappa-light-chain-enhancer of activated B cells) signalling thereby reducing hepatic necrosis. Lin et al., observed that high glycine levels can antagonise the efficacy of IFN-α (interferon alpha) treatment, which may indicate a population that is less likely to respond to this clinically. 

Lin et al., have demonstrated a novel adjuvant to achieve functional HBV cure, addressing cccDNA persistence through a proliferation-based dilution strategy rather than direct degradation or gene editing approaches. Understanding a proliferation-based strategy in the context of an oncogenic virus will be important to elucidate going forward.

Endoscopy

Cutting waste in endoscopy: a German multicentre study

Schmelzle M, Eisenmann S, Lamberth J, et al. Cutting waste in endoscopy: a multicentre observational study in the German healthcare system. Gut 2025; 74: 1989–1996. doi: 10.1136/gutjnl-2024-334891.

Endoscopy is a high-volume contributor to global greenhouse gas emissions with production of medical waste. Approximately 3kg of waste is generated per patient bed per day. Schmelzle et al., aimed to quantify waste and assess the influence of procedural setting from various endoscopic procedures and identify strategies to reduce waste.

This prospective, observational study was conducted over 4 weeks in two hospitals and two office units for non-emergency procedures.   Waste was recorded per patient in the endoscopy room. Waste from each procedure was collected in a single plastic bag and categorised into five groups. Waste from areas outside the endoscopy room was also collected and evaluated daily, then calculated per patient.

The study was conducted from July to October 2023, documenting 2889 procedures in 2275 patients. Hospitals accounted for 937 procedures (32%) in 790 patients (35%) while office units performed 1952 procedures (68%) in 1485 patients (65%).  A total of 2558kg of waste was collected during the study period. 73% of waste originated in the examination room with procedure-related waste contributing the highest proportion. On average, each endoscopic procedure generated 1119g of waste. This was compared with Germany’s 2023 average of 1019 g/day of waste per person annually. 

The study found significantly lower waste generation in office units compared with hospitals for several procedures. Combining procedures also reduced waste. Comparing reusable gowns with single-use gowns reported a 54% significant waste reduction. A recycling programme also reported a reduction in residual waste by 23% overall.  Limitations are the two hospitals and two office units where data was collected which may limit generalisability. The study focused on downstream waste and did not assess water or energy use, production emissions or travel.

This study highlights waste generated in gastrointestinal endoscopy may be preventable and waste-reduction strategies could be implemented without compromising patient care.

Colon

Extracellular vesicle-induced lipid dysregulation drives liver premetastatic niche formation in colorectal cancer

Cao J, Qin S, Li B, et al. Extracellular vesicle-induced lipid dysregulation drives liver premetastatic niche formation in colorectal cancer. Gut 2025; 74: 2012-2023. doi: 10.1136/gutjnl-2025-334851. 

This study asks how colorectal cancer (CRC) primes the liver before tumour cells arrive. It focuses on tumour-derived extracellular vesicles (EVs) and shows that EVs can reset hepatic lipid metabolism and immune signalling. This fits Paget’s “seed and soil” concept, where tumours prepare the “soil” by building hepatic premetastatic niches (PMNs) ahead of spread.

Cao et al., compared CT26 (colon tumour 26) CRC cells with a liver-tropic subline, CT26R2, generated by repeated in vivo selection. CT26R2-EVs showed stronger pro-metastatic activity than CT26-EVs. Mice received IV EVs for 4 weeks, then splenic injection of CRC cells. Prior CT26R2-EV exposure increased later liver metastasis. Rab27a (Ras-associated binding protein 27A) knockout, which reduces EV secretion, lowered tumour burden. 

CT26R2-EVs induced a fatty liver-like state. Mice developed higher liver weight and higher ALT (Alanine transaminase) and AST (Aspartate transferase). Histology and lipid staining showed steatosis, hepatocyte injury, and immune infiltration. This links to metabolic-associated fatty liver disease, which associates with several cancers. High-fat diet drives many cases, but this work suggests tumours can also induce a fatty liver-like niche. Fatty liver can enhance CRC liver metastasis by amplifying EV-driven oncogenic signalling and immunosuppression.

Kupffer cells (KCs) took up tumour EVs. EV exposure increased KC TNF (tumour necrosis factor) expression and raised serum TNFα. Free fatty acids (FFAs) can drive KC cytokine release. KC depletion reduced steatosis and metastasis. Pan-TNFα blockade lowered hepatic triglycerides, improved ALT and AST, and reduced tumour burden. Inhibiting fatty acid synthesis during EV exposure also reduced lipid accumulation and metastasis. This supports prior concepts that targeting EV biogenesis, KC activity, or TNF signalling can disrupt PMN formation.

These results provide new insights into the linkage between CRC and MASLD (Metabolic dysfunction-associated steatotic liver disease) and suggest that targeting CRC cell metabolism or liver lipid accumulation may offer a potential therapeutic approach to improve patient outcomes.

Reviewers

• Dr Sana Zehra Rajani, ST5 Gastroenterology, Milton Keynes University Hospital, Milton Keynes, UK
• Dr Sami Fouda, Royal Free London NHS Foundation Trust, London, UK
• Dr Jabed Ahmed, London North-West University Hospitals NHS Trust, London, UK
• Dr Khurum Hakeem, Gastroenterology Registrar, University Hospital of North Durham, Durham, UK
• Dr Htar Htar Hlaing, ST7 Gastroenterology, West Suffolk Hospital, West Suffolk NHS Foundation Trust, Bury Saint Edmunds, UK
• Dr Zillah Cargill, King’s College Hospital, King’s College Hospital NHS Trust, London, UK

BSG GUT Highlights is edited by Dr Philip J Smith, Honorary Consultant Luminal Gastroenterologist, Royal Liverpool Hospital, Liverpool University Hospitals NHS Foundation Trust, Mount Vernon, Liverpool, UK. Dr Smith is the Digital and Education Editor of GUT, Editor in chief of Frontline Gastroenterology and Associate Editor of BMJ Open Gastroenterology.

Funding

The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

Competing interests: None declared.

Provenance and peer review: Not commissioned; internally peer reviewed.