Clinical trials in nonalcoholic steatohepatitis (NASH) require histologic scoring for assessment of inclusion criteria and endpoints. However, guidelines for scoring key features have led to variability in interpretation, impacting clinical trial outcomes. We developed an artificial intelligence (AI)-based measurement (AIM) tool for scoring NASH histology (AIM-NASH). AIM-NASH predictions for NASH Clinical Research Network (CRN) grades of necroinflammation and stages of fibrosis aligned with expert consensus scores and were reproducible. Continuous scores produced by AIM-NASH for key histological features of NASH correlated with mean pathologist scores and with noninvasive biomarkers and strongly predicted patient outcomes. In a retrospective analysis of the ATLAS trial, previously unmet pathological endpoints were met when scored by the AIM-NASH algorithm alone. Overall, these results suggest that AIM-NASH may assist pathologists in histologic review of NASH clinical trials, reducing inter-rater variability on trial outcomes and offering a more sensitive and reproducible measure of patient therapeutic response.

Introduction TGFβ is an immunoregulatory cytokine that can function as an oncogenic factor to mediate tumorigenesis, metastasis, and immune escape. Recent studies have shown that TGFβ can drive immune-excluded phenotype in tumor microenvironment (TME) by restricting CD8+ T cell infiltration; thus, tumors with an immune-excluded phenotype or high TGFβ levels may be particularly responsive to TGFβ blockade. Here we combine analysis of intra-tumoral and peripheral TGFβ levels with digital pathology-based immune-phenotyping, which might serve as an integrated approach to identify patients most likely to benefit from TGFβ blockade. Methods About 200 NSCLC FFPE tumor blocks with matching fresh frozen tumor and plasma were purchased from commercial sources. Quantitation of peripheral and intra-tumoral TGFβ (TGFβ1/2/3) levels was performed using a qualified assay for active and total forms. FFPE slides were stained for CD8+ T cells (SP57, Ventana) and scanned at 40x using the Aperio AT2 scanner by Covance. Machine learning (ML) models were developed using the digitized whole slide images to identify CD8+, CD8-, and non-lymphocytes, as well as cancer epithelium, cancer-associated stroma, and artifacts. Data-driven cutoffs were applied to model-generated human-interpretable features (HIFs) of CD8+ lymphocyte density to classify samples as desert, excluded, or inflamed. As an orthogonal approach to the CD8+ density-based cutoff method, all tissue and cell model predictions were used to train a graph neural network (GNN) to classify immune phenotypes. Results We found that TGFβ1 was the predominant isoform detected within the periphery while TGFβ1 and TGFβ2 were measurable in majority of tumor samples; TGFβ3 level was below the detection limit in most samples. Correlation analysis of TGFβ1 expression in plasma and tumor shows poor concordance in the matched samples, which could be due to the relatively lower level of intra-tumoral TGFβ and tumor heterogeneity. This discordance may explain why it is challenging to use plasma TGFβ to predict tumor TGFβ level. ML model quantification of CD8+, CD8-, and non-lymphocytes showed high concordance with the consensus score of 5 independent pathologists on a test set of held-out samples not used in training. This concordance was comparable to that of individual annotators to consensus. Immune-phenotyping results using both cutoff and GNN methods also showed moderate concordance with 5-way pathologist consensus. Conclusions We coupled a ligand binding assay for assessing TGFβ isoforms in both plasma and TME with an ML-powered digital pathology platform that can provide a standardized, scalable, and reproducible method to characterize cancer-immune phenotypes in TME. This integrative approach might identify potential biomarkers to select NSCLC patients that may benefit from TGFβ blockade. Citation Format: Robert Pomponio, Carrie Hendricks, Sarah Mitchell Bean, Hong Wang, Sergine Brutus, Charles Biddle-Snead, Archit Khosla, Adam Stanford-Moore, Cyrus Hedvat, Qi Tang, Roger Trullo, Brigitte Demers, Rui Wang. Quantification of TGFβ protein levels and digital pathology-based immune phenotyping reveal biomarkers for TGF-β blockade therapy patient selection in NSCLC [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5099.