Society: AGA

Introduction: Shifts towards high-fat, low-fiber diets, may contribute to the rising incidence of esophageal adenocarcinoma (EAC). In mice high-fat diet promotes EAC, possibly through effects on the gut microbiome and the systemic bile acid pool. Reflux bile acids are thought to promote EAC, but the role of systemic bile acids is unknown. The aim of this study was to assess associations between systemic bile acids and stages of progression in Barrett’s esophagus (BE).
Methods: Subjects with and without BE were enrolled from three sites. Targeted bile acid profiling on serum collected on the day of endoscopy was performed by LC-MS. A subset (73%) of subjects completed a 12-month food frequency questionnaire. MANOVA was performed to assess global bile acid differences across groups. Principal components analyses were performed to identify drivers of global bile acid composition. Individual bile acid comparisons across groups were performed only for Kruskal-Wallis ANOVA p<0.05. Logistic regression analyses were performed to assess associations between logarithmically-transformed bile acid levels and neoplastic stages of BE.
Results: A total of 141 subjects were enrolled with bile acids profiled (49 non-BE; 92 BE: 44 no dysplasia (ND), 9 indefinite (IND), 17 low-grade dysplasia (LGD), 16 high-grade dysplasia (HGD), 6 EAC). Lower Healthy Eating Index score (HEI), older age, and higher BMI were associated with increased levels of several individual bile acids, with an inverse association with PPI use. Global bile acid pools were distinct comparing non-BE and BE subjects (p=0.002) and between non-BE and stages of BE neoplasia (p=0.004). Principal components analyses demonstrated that associations between bile acids and HGD/EAC were driven by the unconjugated primary bile acids cholic acid (CA) and chenodeoxycholic acid (CDCA) as well as forms of ursodeoxycholic acid (UDCA). There were clear shifts in individual bile acids with progression to HGD/EAC. (Fig 1) Increasing CA was associated with HGD/EAC in univariate analyses (Fig 2) and remained significant after adjusting for established EAC risk factors (vs. non-BE; aOR 2.03, 95%CI 1.11-3.71). Similar trends were seen after adjustment for HEI, fat, and fiber intake. CA and CDCA were highly correlated; the two combined were also associated with HGD/EAC (aOR 1.81, 95%CI 1.04-3.13). There were non-significant trends towards decreased levels of unconjugated and conjugated forms of UDCA associated with HGD/EAC.
Conclusions: Alterations in serum bile acids are independently associated with advanced neoplasia in BE and may contribute to neoplastic progression. Future studies should explore associated gut microbiome changes such as bacterial deconjugation via bile salt hydrolase, pro-neoplastic effects of CA and CDCA and protective effects of UDCA, and whether these bile acids represent viable therapeutic targets.

<b>Figure 1. </b>Heatmap of relative risk ratios from polytomous logistic regressions for associations of BE stages compared to controls for serum bile acid levels.

Figure 1. Heatmap of relative risk ratios from polytomous logistic regressions for associations of BE stages compared to controls for serum bile acid levels.

<b>Figure 2.</b> Cholic acid was significantly increased across stages of progression to EAC (p for trend=0.002).

Figure 2. Cholic acid was significantly increased across stages of progression to EAC (p for trend=0.002).

Introduction A confirmed diagnosis of low-grade dysplasia (LGD) on two endoscopies is currently considered a predictor for neoplastic progression in patients with Barrett’s esophagus (BE), and, therefore, ablation therapy is offered to these patients. However, recent studies show that aberrant expression of the p53 tumor suppressor protein might be a stronger risk factor for neoplastic progression than dysplasia status. The aim of this study was to evaluate the value of p53 expression compared to confirmed LGD in predicting neoplastic progression in BE patients.

Method Data was analyzed from a large prospective cohort study of 1031 BE patients (73% males, median age 61 years) with index diagnosis of non-dysplastic BE (NDBE), indefinite for dysplasia (IND), and LGD. Confirmed LGD was defined as a LGD diagnosis on 2 separate occasions, each LGD diagnosis was confirmed by two pathologists. P53 immunohistochemistry staining was performed on biopsies of 1545 endoscopies from 655 (64%) patients; both overexpression and loss of expression were considered aberrant. Neoplastic progression was defined as high-grade dysplasia (HGD) and/or esophageal adenocarcinoma (EAC). Cox regression analysis was used to determine neoplastic progression risk.

Results During a median follow-up of 6.2 (IQR 3.2-11.4) years, 73/1031 (7%) patients developed HGD/EAC. Neoplastic progression was found in 23/756 (3%) NDBE patients, 27/176 (15%) patients with a single diagnosis of LGD, and 23/99 (23%) patients with confirmed LGD. P53 expression was aberrant in 28/411 (7%) NDBE patients, 50/149 (34%) single LGD patients, and 51/95 (54%) confirmed LGD patients. Aberrant p53 expression was strongly associated with an increased risk of neoplastic progression after adjusting for age, gender, segment length, oesophagitis, and grade of dysplasia (HR 13.5, 95% CI 7.1-25.8). In NDBE patients, the absolute neoplastic progression risk increased from 2% with normal p53 expression to 50% with aberrant p53 expression. In patients with a single diagnosis of LGD, the progression risk increased from 8% to 37% when p53 expression was aberrant. Similar risks where found in patients with confirmed LGD, with a neoplastic progression risk of 6% with normal p53 expression and 47% with aberrant p53 expression.

Conclusion Aberrant p53 expression is a strong predictor for neoplastic progression in BE patients. The predictive value of aberrant p53 expression is independent of the grade of dysplasia. Aberrant P53 expression may be a better parameter than (confirmed) LGD to identify patients who need close surveillance or could benefit from ablation therapy.

Introduction: Clinical management of Barrett’s esophagus (BE) is challenging due to variability in the histologic grading of biopsies and uncertainty regarding patients’ risk of progression to esophageal adenocarcinoma (EAC). Objective risk stratification is needed to enable risk-aligned management to improve BE patient health outcomes. A tissue systems pathology test (TissueCypher, TSP-9) has been validated in multiple studies to predict the progression of BE to high-grade dysplasia (HGD) and EAC within 5 years. This study evaluated the predictive performance of the TSP-9 test versus clinicopathologic variables in a multi-center cohort to assess the effectiveness of the test in improving clinical management of BE.
Methods: Data from 5 published studies on the TSP-9 test was evaluated (n=699 patients with non-dysplastic BE (NDBE, n=567), indefinite for dysplasia (IND, n=50) and low-grade dysplasia (LGD, n=82)). 509 patients did not progress (median surveillance 6.7 years, IQR 5-9), 38 were diagnosed with HGD/EAC within 12 months (prevalent cases), and 152 patients progressed to HGD/EAC after 12 months (incident progressors, median 3.2 years IQR, 2-5). Age, sex, segment length, real-world pathology diagnosis (from health records), expert GI pathology review diagnoses, and TSP-9 risk results were collected. The predictive performance of the TSP-9 test and pathology diagnoses were compared, and TSP-9 was evaluated in clinically-relevant patient subsets.
Results: The sensitivity of the TSP-9 test in detecting progressors was 62.3% versus 28.3% for the expert diagnosis of LGD, while the real-world diagnosis did not provide significant risk stratification (Fig. 1A-C). The test detected 54.6% of incident progressors and 84.2% of prevalent cases. The TSP-9 high-risk class was a stronger predictor of progression than an expert diagnosis of LGD (HR=6.2 (95% CI 4.4-8.9) vs. HR=2.2 (95% CI 1.5-3.4), Fig. 1D). The improved sensitivity was due to the TSP-9 test’s ability to detect 57% of the progressors with expert diagnoses of NDBE (Fig. 2A). Patients with NDBE who scored TSP-9 high-risk progressed at a higher rate (3.2%/year) than patients with expert-confirmed LGD (2.3%/year). The TSP-9 test provided significant risk stratification in patients considered to be clinically high-risk (indefinite for dysplasia/LGD, male, long segment), and clinically low-risk patients (NDBE, female, short segment, Fig. 2).
Conclusions: The TSP-9 test demonstrated significantly improved risk stratification of BE patients versus expert pathology and clinical variables, and predicts the risk of prevalent and incident HGD/EAC independently of clinicopathologic variables. The test can 1] enhance detection of prevalent cases, 2] increase early detection of progressors at the NDBE stage, and 3] be used to guide risk-aligned management decisions to improve health outcomes for BE patients.

Introduction: Barrett’s Esophagus (BE) is characterised by the metaplastic replacement of squamous with columnar epithelium. However, BE is also an inflammatory condition and immune infiltration has been widely reported, little is known about the overall immune landscape at the cellular level, nor do we know much about the genes expressed. Furthermore, the role that immune cells play in progression to cancer is poorly understood. Here we use multiplex immunohistochemistry combined with laser-capture microdissection (LCM) RNAseq to understand how immune cells interact with each other in metaplasia in patients that develop dysplasia.

Methods: We set out to unveil spatial phenotyping of immune cells in the stromal compartments of non-dysplastic BE using a highly multiplexed imaging approach, termed co-detection by indexing (CODEX), with a panel of 56 antibody markers on Tissue Microarrays (TMA) that identify most cells within BE tissue. Our cohort consists of 26 BE patients (69 cores of non-dysplastic BE) who have never progressed to EAC and 6 BE patients (48 cores of non-dysplastic BE from patients that developed cancer). Only areas containing metaplasia were studied. The identification of cell types, spatial organization, and neighborhood-neighborhood interaction was performed using unsupervised, k-means clustering, and cross-type Ripley's K function respectively. The landscape of spatially resolved transcriptome of stroma was further profiled using LCM and RNAseq.

Results: Overall, our data identified twelve different types of immune cells in BE microenvironment. M1 macrophages and plasma cells were the most abundant immune cell type, followed by CD4+ and CD8+ T cells irrespective of disease progression. Furthermore, progressors showed a significant elevation of immune cell concentrations, in particular CD4+ Treg cells, Neutrophils, and Nature killer cells (NKs), and a reduced population of B cells and dendritic cells. Interestingly, an NK cell subset (CD11b^high, CD56^dim, Annexin A1^high, and P63^high) was exclusively observed in progressors. We also identified seven unique immune cell neighborhoods (CNs), and CNs enriched with plasma, NKs, CD4 cells, and DCs were spatially associated with progressive metaplasia. These findings were concordant with the RNAseq data showing a set of highly expressed genes in the stroma of the progressors that were associated with long-lived plasma cells (CD38^hi CD38+), neutrophils, and inflammation-associated unfolded protein responses.

Conclusions: We show for the first time that long-lived plasma cells at chronic inflammation sites may contribute to dysplasia progression. Together with a unique immune cell signature (CD4+ Treg, Neutrophils, NKs, and loss of DCs and B cells) in adjunct with CN/Cell interaction profiles may provide a novel strategy in developing models of cancer risk stratification in BE patients.

Introduction: Barrett’s esophagus (BE) progresses to esophageal adenocarcinoma (EAC) via the development of low (LGD) and then high grade dysplasia (HGD). The histological diagnosis of BE dysplasia is challenging due to subjectivity and confounding by inflammatory artifact. This leads to misclassification of dysplasia grade and substantial interobserver variability even amongst expert pathologists. Hence, there is a critical need to improve diagnostic capabilities for BE dysplasia grade determination, currently the only criteria for endoscopic therapy. We have previously developed an ensemble machine learning BE dysplasia deep learning model (BEDDLM), leveraging whole slide images (WSI) of BE biopsies, from one academic institution, combining a object detection (You Look Only Once” (YOLO)) and classifier (ResNet 101) models. We aimed to externally validate BEDDLM on BE WSIs from three academic medical centers.

Methods: We obtained non-dysplastic BE (NDBE), LGD, and HGD histology slides (hematoxylin and eosin stained) from two academic medical centers. All slides were reviewed by GI pathologists at the home academic institution, and by two additional expert GI study pathologists (CH and JL). The criterion standard was dysplasia grade as per agreement between at least two of three expert GI pathologists (2 study and 1 academic home institution). Slides were digitized using a high-resolution scanner (Aperio, Leica Biosystems, Buffalo Grove, IL). The object detection model identified regions of interest (ROI) providing first-pass dysplasia grade prediction, followed by feeding these ROIs into a ResNet101 classifier model. Dysplasia grade prediction was based on the highest grade of dysplasia agreement between the two models.

Results: We included 274 WSIs from 165 patients from two external academic medical centers in this analysis. Baseline patient characteristics are shown in Table 1. 18% of WSIs were from endoscopic resection specimens, which had not been used for BEDDLM development. Almost 50% of samples had a consensus diagnosis of NDBE with LGD and HGD being approximately 25% each. Results (Table 2) showed high sensitivity, specificity and F1 score (measure of precision and recall) for the diagnosis of NDBE and HGD, with more moderate results for LGD. Comparable results were found based on a 2 class model prediction (dysplasia vs. no dysplasia). Additional WSIs from external site 3 (with NDBE, LGD and HGD WSIs) are currently being analyzed.

Conclusion: We externally validated BEDDLM for the artificial intelligence powered prediction of BE dysplasia grade. This model, after additional optimization (particularly for LGD) and testing could be used in community and academic pathology practices to increase accuracy and decrease interobserver variability in BE dysplasia grade diagnosis.

Pallavi Patil

Appearances