A total of 24 (12 EoE, 12 Control) subjects were enrolled into this study with informed consent from all participants and/or guardians. All studies were conducted under the approval of the Institutional Review Board. EoE subjects were individuals with suspected EoE scheduled for esophagogastroduodenoscopy (EGD) for diagnostic confirmation. Control subjects were individuals undergoing EGD for investigative purposes or other non-inflammatory etiologies. All patients selected were undergoing an initial EGD and not yet prescribed therapeutics for EoE. All relevant data, including health information, laboratory results, pathology results, endoscopy scores, treatment, and questionnaire responses were recorded.

Tissue samples were obtained during clinically indicated EGD procedures with clinical assessment of Eosinophilic Esophagitis Endoscopic Reference Score (EREFS). Biopsies collected from endoscopically visible areas of active inflammation and healthy mucosa in subjects and controls, respectively. Definitive diagnosis of EoE was made using clinical, endoscopic, and histologic metrics per current guidelines (4), including symptoms of esophageal dysfunction, presence of ≥15 eos/hpf, and exclusion of other potential causes of esophageal eosinophilia. Controls were defined as patients who underwent an EGD for symptoms suggestive of EoE but not confirmed with endoscopic or histologic assessment. Samples were then stored in RNAlater (Qiagen) until further processing.

Active disease was defined as active inflammation based on endoscopy. Disease duration was categorized by years since diagnosis in 3 groups: <1, 1–5, >5 years. Eosinophilic Esophagitis Reference Score (EREFS) disease activity summary score was recorded at the time of biopsy (19). Eosinophilia grading was developed using eos/hpf histologic data: mild = 0–6 eos/hpf, moderate = 7–15 eos/hpf, severe = 15–30 eos/hpf, and profound = 31 + eos/hpf.

Biopsy samples were homogenized in RLT buffer (Qiagen). DNA, RNA, and protein were extracted using the AllPrep kit per manufacturer's protocol (Qiagen). RNA was assessed for quality using NanoDrop™ and stored at −80°C. 10 µl RNA (50 ng/µl) were plated on a barcoded 96-well plate provided by Thermo Fisher Scientific. Microarray analysis using Clariom™ S Assay for human samples was performed by Thermo Fisher Scientific. Plate normalization, hybridization, and processing are performed as part of the throughput at Thermo Fisher Scientific. Finalized data was distributed in the format of CEL files.

Population descriptors were expressed as frequency (number and percentage) and median ± IQR, as appropriate. Microarray analyses were performed using the Thermo Fisher Transcriptome Analysis Console (TAC). CEL files were imported to the TAC software and initial quality control evaluation was conducted. Samples passed quality control for three parameters, including labeling controls, hybridization controls, and positive vs. negative AUC thresholds. Principal Components Analysis (PCA) were used to determine if samples clustered based on disease status or other clinical metrics. Differential gene expression (DEG) was used to determine gene families and clusters that were differentially up-/down-regulated between EoE and Control groups, with significance defined as ≥2, ≤−2-fold change.

Analysis using ADVAITA Bioinformatics iPathwayGuide, which is an AI-based platform for predictive pathway analysis (ADVAITA Bioinformatics, Ann Arbor, MI, USA), was conducted in parallel. All significant DEGs were used as input to determine the significantly impacted pathways with pathway annotations being derived from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Significant pathways (false discovery rate adjusted p values <0.1) were presented as –log (p-value). The top significantly enriched Gene Ontology (GO) biological processes and cellular components are reported as -log10 (p-value). Significant DEGs for the SPRR and MUC family were input into g:Profiler (https://biit.cs.ut.ee/gprofiler/gost) for functional profiling with statistically significant related Reactome pathways reported as -log10 (p-value).

This study utilized tissue samples from 12 EoE and 12 Control subjects. The EoE group was 33.3% female compared to 83.3% female in the Control group (Table 1). The age distribution was comparable with a median age of 18 years in both EoE (range: 5–31 years) and Control (range: 5–49) groups. The study population overall was predominantly white (83.3%), non-Hispanic Latino (83.3%). None of the Control samples had evidence of disease activity on biopsy, whereas all the EoE subjects had evidence of active disease at the time of sampling. Half of the EoE subjects had a disease duration of greater than 5 years, 33.3% 1–5 years, and 16.7% less than 1 year. The median disease severity based on endoscopic reference score assessing EREFS in the EoE group was 3.5 (range: 1–6). Histological grading of the EoE biopsy samples revealed a median eosinophilia count of 31 eos/hpf (range: 15–92). The transcriptomic data was analyzed for the impact of these clinical features on the clustering of samples using principal component analysis (PCA). The samples strongly clustered by disease category, moderately clustered by disease duration, moderately clustered based on sex, and no defined clustering based on age category (Supplementary Figures S1, S2). Therefore, disease status better defines the clustering observed, with their being stratification given disease duration. Clustering based by sex may likely be explained by EoE being nearly 4 times more prevalent in males than females (20). We anticipate that clustering effects would become stronger given a larger sample population. Based on the appearance of these clusters, we further explored sample transcriptomics to better understand which gene families are driving the clusters.

Microarray analysis returned data on 21,448 genes, 1,059 of which were DEGs between EoE and Control samples. Scatterplot analysis of the DEGs revealed 806 (76.1%) were up-regulated and 253 (23.9%) were down-regulated (Figure 1A). The top 20 up-regulated (Figure 1B) and down-regulated genes (Figure 1C) represented a diverse repertoire of functions. The most differentially expressed up-regulated genes were JCHAIN (151.4-fold), CPA3 (92.8-fold), ANO1 (57.9-fold), and CDH26 (55.7-fold) which all had a greater than 50-fold change increase compared to Controls. Overall, the down-regulated genes were more differentially expressed with 3 genes having more than a 100-fold decrease compared to Controls: CRISP3 (−1411.9-fold), SPRR2E (−127.7-fold), and SPINK7 (−123.9-fold). Hierarchical clustering of up-regulated genes displayed close relationships between all but one of the Control samples, suggesting a similar transcriptomic profile. Similarly, one EoE sample clustered more tightly with the Control group than the EoE group (Figure 1D). Interestingly, hierarchical clustering of the down-regulated genes displayed 2 distinct subgroups of EoE patients, while the Control samples remained tightly clustered (Figure 1E). The Control sample that clustered with the EoE group in the up-regulated gene heatmap remained clustered with EoE samples in the down-regulated heatmap.

Our dataset confirmed several classic genetic markers of EoE. Among the top 20 up-regulated genes, 8 have been previously associated with EoE. Notably eotaxin-3 (CCL26), an eosinophil chemotactic chemokine, which has been extensively reported as a strong driver of EoE (13, 21) was represented in this group (Figure 2A). These up-regulated genes have also been reported to play a breadth of roles in EoE pathophysiology including: mast cells (CPA3), tissue remodeling (POSTN; CTSC), and ion channel function/inflammation (ANO1; ALOX15) (22, 23) (Figure 2A). The most highly down-regulated gene (CRISP3) has been described as impacting epithelial function in EoE patients (16). SPINK7, down-regulated by 132-fold in our dataset, has been implicated in regulating esophageal barrier maintenance and down-regulation may be associated with EoE pathogenesis (16). The downregulated genes have also been associated with cell adhesion and inflammation in EoE, such as DSG1 (14, 15) (Figure 2B). Several of the top pathways based on gene count were associated with barrier cell function and integrity (VEGFA-VEGFR2; Endothelin; EGF/EGFR) (Figure 2C). Tight junctions are a critical signaling component of barrier function and were also implicated in pathway analysis performed by ADVAITA iPathwayGuide (Figure 2D). The IL-18 signaling pathway, which has been linked to mast cell recruitment in food allergen induced EoE, was also represented in the pathway analysis (24) (Figure 2C). Autophagy was one of the top pathways to be implicated by the dysregulated transcriptome (Figure 2D). This is consistent with a previous study that found autophagy to be a potential drug target and critical for maintaining metabolic homeostasis within the esophageal epithelia following exposure to pro-inflammatory cytokines (25). Of interest, metabolic pathways was the top pathway predicted with a -log10 p-value of 2.78 (Figure 2D). Dysregulation of these predicted pathways is expected to impair functions largely related to intercellular signal transduction, catalytic activity, lipid metabolism, response to stimuli, and keratinocyte differentiation (Figure 2E). Pathologic activity is expected to interfere with protein binding, enzyme binding, and cadherin binding (Figure 2F). Defects in cadherin binding is another mechanism that may impair barrier function in EoE patients. A study conducted by Doshi et al. found that pediatric EoE patients had decreased membrane bound E-cadherin (26). Therefore, loss of E-cadherin is predicted to decrease barrier integrity by interfering with tight junction interactions. Our findings recapitulate the well-described etiology of EoE as a food allergen directed destruction of the esophageal barrier.

Previous work assessing esophagus-specific transcripts following whole exome sequencing described impaired regulation of epithelial differentiation and keratinization (27). Consistently, keratinocyte differentiation was one of the top GO biological functions predicted from our transcriptome profile in EoE patients (Figure 2E). Examination of genes involved in keratinocyte differentiation revealed a distinct enrichment of small proline-rich protein (SPRR) genes, as 7 out of 29 genes were significantly downregulated (Figure 3A). ADVAITA iPathwayGuide analysis reveals downstream implications of keratinocyte differentiation on epidermal cell differentiation that further affect epithelium development, cell differentiation, and anatomical structure development (Figure 3B). We next analyzed our whole-transcriptome dataset specifically for SPRR genes and found that all 10 SPRR genes were downregulated with 7 genes identified as significant (Figure 3C). GO biological analysis of the SPRR gene family predicts the involvement of these genes in keratinization, keratinocyte/epidermal/epithelial cell differentiation, skin/epidermis/epithelium development, and peptide cross-linking (Figure 3D). These biological functions are expected to occur within the cornified envelope (Figure 3E). Reactome pathway prediction further reinforces the implications genetic dysregulation has on formation of the cornified envelope, keratinization, and developmental biology (Figure 3F).

Mucin expression is a common biomarker for several diseases with interference in production impairing cellular integrity. Within the gastrointestinal tract, the amount of mucins generally increases from esophagus to rectum (28). Under normal physiologic states, the human esophagus produces MUC1, MUC4, MUC5B, and MUC20 (28). Evaluation of the MUC genes indicated that two genes were significantly downregulated: MUC22 (−36.63-fold) and MUC21 (−27.49-fold) (Figure 4A). MUC4 was the only gene in this family found to be significantly upregulated (8.08-fold) (Figure 4A). To date, these mucins have not been evaluated in the context of eosinophilic esophagitis. However, MUC4 is associated with normal squamous epithelium; therefore, an increase in this gene expression may be indicative of increased differentiation. However, increased MUC4 has also been observed other diseased esophagus states, including Barrett's esophagus and high-grade intraepithelial neoplasia (29). A glycoform of MUC21 with extended carbohydrate chains was found in suprabasal cells and serves as a marker for differentiation in squamous cell carcinoma (30), which suggests another possible marker for esophageal differentiation. Dysregulation of these mucin genes are expected to impact extracellular matrix constituents, particularly involving lubricant activity and structure (Figure 4B). GO cellular components largely predicts this dysregulated activity to occur within the Golgi lumen (Figure 4C), the site at which mucin dimers move from the ER for assembling of O-glycans. As the type and composition of mucins are indicative of tissue, physiologic state, and pathology, mucins contain a spectacle of O-linked oligosaccharides that vary between the types of mucins, tissues, and pathologic conditions and is predicted to vary between EoE patients and Controls (Figure 4D).

Previous work from the Human Protein Atlas evaluated esophagus-specific transcripts and found that transcripts were predicted to implicate keratinization and differentiation. This unique transcriptome profile was correlated with EoE biopsies lacking adequate tissue differentiation (27). Our work also found common genes related to keratinization and differentiation to be DEGs. Cornulin expression was found to be downregulated in our dataset comparing EoE biopsies to Controls (CRNN, −3.69-fold) (Figure 5A). Filaggrin (FLG, −16.63-fold) and desmoglein 1 (DSG-1, −132.09-fold) were also down-regulated in EoE patients (structural integrity of the epithelial barrier) (Figure 5A). Further, a member of the SPRR family, SPINK7 was decreased (−132.09-fold). Keratin-6C (KRT6C) was found to be downregulated at −2.35 (Figure 5A). This is consistent with our transcriptome profile for EoE patients as functional enrichment GO predicted impaired biological functions related to keratinization, differentiation, and epidermis/epithelium/tissue development (Figure 3D). Reactome analysis further predicted formation of the cornified envelope to be impacted (Figure 3F). Transcriptome analysis of the EDC genes reveals an overall downregulation of genes related to cornification, proper maturation, and terminal differentiation of squamous epithelial cells.