Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/2023.10.25.563471

Machine Learning Based Identification and Characterization of Mast Cells in Eosinophilic Esophagitis

Zhang, S.·Caldwell, J. M.·Rochman, M.·Collins, M. H.·Rothenberg, M. E.
View original source ↗Export metadata ↓

Abstract

BackgroundEosinophilic esophagitis (EoE) is diagnosed and monitored using esophageal eosinophil levels; however, EoE also exhibits a marked, understudied esophageal mastocytosis. ObjectiveUsing machine learning, we localized and characterized esophageal mast cells to decipher their potential role in disease pathology. MethodsEsophageal biopsy samples (EoE, control) were stained for mast cells by anti-tryptase and imaged using immunofluorescence; high-resolution whole tissue images were digitally assembled. Machine learning software was trained to identify, enumerate, and characterize mast cells, designated Mast Cell-Artificial Intelligence (MC-AI). ResultsMC-AI enumerated cell counts with high accuracy. During active EoE, epithelial mast cells increased and lamina propria (LP) mast cells decreased. In controls and EoE remission patients, papillae had the highest mast cell density and negatively correlated with epithelial mast cell density. Mast cell density in the epithelium and papillae correlated with the degree of epithelial eosinophilic inflammation, basal zone hyperplasia, and LP fibrosis. MC-AI detected greater mast cell degranulation in the epithelium, papillae, and LP in EoE patients compared with control individuals. Mast cells were localized further from the basement membrane in active EoE than EoE remission and controls individuals but were closer than eosinophils to the basement membrane in active EoE. ConclusionUsing MC-AI, we identified a distinct population of homeostatic esophageal papillae mast cells; during active EoE, this population decreases, undergoes degranulation, negatively correlates with epithelial mast cell levels, and significantly correlates with distinct histologic features. Overall, MC-AI provides a means to understand the potential involvement of mast cells in EoE and other disorders. Clinical ImplicationWe have developed a methodology for identifying, enumerating, and characterizing mast cells using artificial intelligence; this has been applied to decipher eosinophilic esophagitis and provides a platform approach for other diseases. Capsule SummaryA machine learning protocol for identifying mast cells, designated Mast Cell-Artificial Intelligence, readily identified spatially distinct and dynamic populations of mast cells in EoE, providing a platform to better understand this cell type in EoE and other diseases.

Source connections

Explore related subjects

immunologyPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Zhang, S., Caldwell, J. M., Rochman, M., Collins, M. H., Rothenberg, M. E.. 2023-10-29. Machine Learning Based Identification and Characterization of Mast Cells in Eosinophilic Esophagitis. https://doi.org/10.1101/2023.10.25.563471

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Differential and Cooperative Immunomodulation by Saponin Adjuvants QS-21 and QS-7 in Human THP-1 Macrophages

QS-21, a triterpene glycoside saponin derived from Quillaja saponaria, is a potent immunostimulatory adjuvant incorporated in several FDA-approved vaccines. QS-7, a structurally related saponin fraction, has been reported to retain immunomodulatory activity with reduced hemolytic toxicity; however, its lower natural abundance has limited its systematic immunological characterization. To address this gap, we systematically compare responses to QS-21 and QS-7 in the THP-1 human macrophage model to resolve their distinct and cooperative contributions to antigen presenting cell (APC) activation. Following cytotoxicity profiling to establish safe and biologically relevant dosing, flow cytometry, proteomics, and transcriptomics were used to delineate the immunological programs engaged by each saponin alone and in combination. QS-7 alone induced minimal upregulation of cell surface activation markers and did not elicit a significant transcriptional signature, yet it drove dose-dependent secretion of a broad panel of pro-inflammatory cytokines and chemokines, revealing a dissociation between these two arms of macrophage activation. Strikingly, co-administration of QS-7 with QS-21 substantially potentiated macrophage activation across all parameters tested, amplifying surface marker upregulation, broadening cytokine and chemokine output to include IL-12p70 and IFN-{gamma}, and markedly expanding the pro-inflammatory transcriptional and pathway landscape induced by QS-21, while concurrently suppressing counter-regulatory lipid-sensing pathways (PPAR, RAR, FXR/RXR) that constrain inflammatory output. Collectively, these findings establish QS-7 not merely as an attenuated analog of QS-21, but as a functionally distinct immunomodulator that recalibrates both the magnitude and regulatory tone of QS-21-driven macrophage activation. These findings show that structurally distinct saponins elicit distinct and complementary macrophage responses, providing a foundation for structure-function studies. Together with synthetic biology and data-driven design, this knowledge can guide the development of new-to-nature saponin adjuvants.

immunology↗
bioRxiv2026

MAFB and MAF cooperatively orchestrate human anti-inflammatory macrophage identity via autocrine Interleukin-10

Macrophage lineage identity is governed by intricate transcriptional networks. Although the large MAF basic region-leucine zipper (bZIP) transcription factors MAFB and MAF are critical regulators of anti-inflammatory programs, their molecular coordination and non-redundant contributions within human macrophages remain poorly defined. Here, utilizing integrated proteomic, transcriptomic, and epigenomic profiling, we demonstrate that endogenous MAFB and MAF physically associate to form a stable heterodimeric complex on chromatin in human M-CSF-dependent monocyte-derived macrophages (M-Mac). Genome-wide CUT&Tag mapping reveals a hierarchical epigenomic architecture in which MAFB acts as the primary global anchor of M-Mac identity, whereas MAF provides localized transcriptional tuning. Mechanistically, both factors directly bind the IL10 locus, but functional IL-10 expression is primarily driven by MAFB. Neutralizing antibody assays reveal that endogenously produced IL-10 establishes an obligate autocrine feedback loop mandatory for both the initial acquisition and the maintenance of the M-Mac anti-inflammatory profile. Functionally, MAFB and MAF exert non-redundant control over core macrophage effectors: MAFB dictates apoptotic cell efferocytosis amd T cell activation, whereas MAF modulates phagocytosis, and acts as a brake suppressing acute pro-inflammatory cytokine/chemokine hyper-induction. Together, these findings establish a comprehensive molecular blueprint of how MAFB and MAF cooperatively fine-tune human monocyte-derived macrophage identity and functional plasticity.

immunology↗
bioRxiv2026

Peripheral Immune Cell Shifts Associated With Alzheimer's Disease Vary in Multi-Ancestry Cohorts

Background: Alzheimer's disease (AD) is heavily influenced by immune system dysregulation, yet the role of peripheral immune processes--and how these vary across ethnicities--remains understudied. Objective: To investigate ancestry-specific immune alterations in AD by profiling peripheral immune cell proportions using high-resolution deconvolution of whole blood RNA-sequencing data, characterizing disease associated cell types, and identifying enriched biological pathways. Methods: We analyzed whole blood RNA-seq data from non-Hispanic White (NHW), African American (AA), Caribbean Hispanic (CariHISP), and Peruvian (PERU) individuals, balanced between AD cases and cognitively normal controls. Immune cell proportions were estimated using CIBERSORTx and compared using linear models (propeller R package) adjusted for age, sex, and APOE genotypes in case-control models. Cell-type-specific differential expression was performed using CARseq, and pathway enrichment analyses performed using MSigDB gene sets. Results: Baseline immune composition differed substantially across cohorts for multiple innate and adaptive cell types. In a multi-ethnic AD versus control analysis, eight cell types reached FDR significance: AD was associated with decreased proportions naive B cells, CD56dim NK cells, B cells, intermediate monocyte I, and double-negative T cells, alongside significant expansion in neutrophils and non-classical monocytes. Interaction analyses revealed heterogeneous group-specific effects, with the broadest shifts in AA and CariHISP. Pathway enrichment indicated shared heme metabolism, interferon, and oxidative phosphorylation programs across groups, with distinct emphases: broad, strong heme metabolism enrichment and the highest oxidative phosphorylation signal (in myeloid populations) in AA; MYC targets in naive B cells and eosinophils in CariHISP; dendritic cell - and T follicular helper cell - centered interferon, inflammatory, and proliferative programs in NHW; and TGF-{beta} and apoptosis signatures, unique to PERU, alongside the broadest cell-type coverage. Conclusions: Peripheral immune dysregulation in AD is characterized by depletion of lymphoid and intermediate monocyte populations with concurrent neutrophil and non-classical monocyte expansion, and these patterns vary meaningfully by ancestry. These findings underscore the need for population-informed biomarkers and therapeutic strategies in AD.

immunology↗