Many esophageal diseases can arise during development or throughout life. Therefore, well-characterized in vitro models and detailed methods are essential for studying human esophageal development, homeostasis and disease. Here, we (1) create an atlas of the cell types observed in the normal adult human esophagus; (2) establish an ancestrally diverse biobank of in vitro esophagus tissue to interrogate homeostasis and injury; and (3) benchmark in vitro models using the adult human esophagus atlas. We created a single-cell RNA sequencing reference atlas using fresh adult esophagus biopsies and a continuously expanding biobank of patient-derived in vitro cultures (n=55 lines). We identify and validate several transcriptionally distinct cell classes in the native human adult esophagus, with four populations belonging to the epithelial layer, including basal, epibasal, early differentiating and terminally differentiated luminal cells. Benchmarking in vitro esophagus cultures to the in vivo reference using single-cell RNA sequencing shows that the basal stem cells are robustly maintained in vitro, and the diversity of epithelial cell types in culture is dependent on cell density. We also demonstrate that cultures can be grown in 2D or as 3D organoids, and these methods can be employed for modeling the complete epithelial layers, thereby enabling in vitro modeling of the human adult esophagus.

The esophagus connects the upper pharynx with the stomach and is lined by a stratified squamous epithelium (Rosekrans et al., 2015). In early development, the esophagus and trachea develop from a common gut tube, which separates to give rise to the esophagus and the respiratory organs during the first trimester of the human fetal development (Rankin et al., 2016; Ernst et al., 2019). The esophagus is specified on the dorsal side of the common gut tube, with the lung being specified on the ventral side (Kuwahara et al., 2020). At this early time, the esophagus is organized as a pseudostratified epithelium that matures into a fully stratified epithelium (Ernst et al., 2019). Failure to achieve proper specification and separation can lead to different developmental defects, such as esophageal atresia or tracheoesophageal fistula (Houben and Curry, 2008).

The stratified epithelium of the esophagus is maintained throughout adulthood. Histological and more recent single-cell characterization of the esophagus has identified several distinct epithelial cell types/states (Madissoon et al., 2019; Busslinger et al., 2021). Understanding, describing and characterizing the normal/healthy esophagus during homeostasis is an essential first step to understanding injury and repair, disease states and disease progression. In vitro model systems are crucial for studying human health and disease. The adult esophagus is prone to many diseases, including eosinophilic esophagitis (Blevins et al., 2018), metaplasia (Barrett's esophagus) due to reflux and inflammation, esophageal squamous cancer (Kim et al., 2017) and esophageal adenocarcinoma (Saraggi et al., 2016). Genetic variation attributed to ancestry and racial background has been linked with esophageal disease despite risk factors being equal across populations (Spechler et al., 2002, 2011; Thrift and El-Serag, 2016; El-Serag et al., 2004). For example, there is a higher incidence of esophageal adenocarcinoma in the Caucasian/European American population compared with populations of African or African descent (El-Serag and Sonnenberg, 1997; Rastogi et al., 2008; Fleischer et al., 2008; El-Serag et al., 2014; Arnold et al., 2017; Then et al., 2020).

The goal of the current study was to leverage access to patient biopsy specimens to create an atlas of cell types within the healthy human esophagus and to use this atlas to benchmark in vitro models. Single-cell RNA-sequencing (scRNA-seq) data obtained from healthy esophagus biopsies identified four molecularly distinct epithelial cell types corresponding to different domains within the epithelium, which were validated by immunofluorescence. Within the basal cell compartment, where stem cells reside, we identified previously known (Busslinger et al., 2021) and new markers (i.e. COL17A1+, CAV1+, CAV2+). Adjacent to the basal cell domain was the suprabasal zone, which can be divided into three distinct domains based on protein markers: an LY6D+ epibasal domain, which corresponds to the domain with high proliferation [KI67 (MKI67)+], a KRT4+ mid differentiation zone, and a CRNN+ luminal zone of terminally differentiated cells.

To develop a robust method to maintain long-term cultures of the healthy and diseased human esophagus established in 2D cultures, this work also unifies different methods reported for the culture of rodent and human esophagus (DeWard et al., 2014; Busslinger et al., 2021; Madissoon et al., 2019; Mou et al., 2016; Liu et al., 2017; Yamamoto et al., 2016; Kasagi et al., 2018). We demonstrate that the method described herein is suited for 2D in vitro culture of the adult and developing human esophagus and esophagus tissue from healthy and diseased states. Using these methods, we have established a diverse growing biobank (currently n=55 cell lines) from individuals self-identified as European American, African American/Black, Asian, or of Hispanic descent. By comparing our in vivo reference atlas to the in vitro scRNA-seq data, we find that 2D in vitro-grown cultures possessed an abundance of cells transcriptionally similar to basal-epibasal cells. Our results demonstrate that cell density determines the balance of proliferation and differentiation within the culture. Basal cells maintained in 2D culture can also be used to grow 3D organoids. Taken together, this study provides detailed methods for long-term culture of the human esophagus in vitro, provides an atlas of the adult human esophagus, and uses scRNA-seq data as a roadmap to benchmark in vitro-cultured cells, providing a detailed foundation for pursuing human esophagus research.

Adult human esophagus epithelium contains molecularly distinct zones as defined by scRNA-seq and validated at the protein level

The esophagus is composed of stratified squamous epithelium, underlying stromal tissue (lamina propria), and smooth muscle (muscularis mucosae) (Rosekrans et al., 2015). To characterize the esophagus, we obtained normal/healthy squamous (NS) epithelial adult human biopsies (approximately 3mm2) (n=2 independent patients, with n=3-4 biopsies used for dissociation) and carried out enzymatic dissociation into single cells, which were captured using the 10x Chromium platform for subsequent sequencing. Louvain clustering defined seven transcriptionally distinct clusters (from 0 to 6) present in the biopsies (Fig. S1A-C, Table S1). Using well-established marker genes, we defined classes of cells as epithelial (CDH1+; Clusters 0-3) or stroma/lamina propria (VIM+; Clusters 4-6), which included immune cells (Clusters 4/5) (Fig. S1B-F). The contribution to each cluster was consistent across biological replicates (Fig. S1B), and classifications of each cluster were performed by cross-referencing the Human Protein Atlas (Fig. S1G) (Uhlen et al., 2005, 2015).

To characterize the epithelium further, epithelial clusters (0, 1, 2, 3) were computationally extracted and re-clustered, revealing four predicted sub-clusters (Fig. 1A, Table S2). Individual biopsies contributed to each cluster in similar proportions (Fig. 1B). Unsupervised clustering was used to plot the top five genes in an unbiased way (Fig. 1C, Table S2) and used to determine that epithelial clusters correspond to basal cells (Cluster 2), proliferative (Cluster 1), suprabasal (Cluster 0) and the differentiated/luminal zone (Cluster 3) (Fig. 1C, Fig. S2A). We observed that TP63, a marker canonically used to identify basal cells, was broadly expressed in basal, proliferative and suprabasal cell clusters (Fig. S2B). Protein staining validated this finding, which showed broad epithelial expression (Fig. S2C). scRNA-seq data identified genes that were highly enriched in the basal cell cluster (Cluster 2) (Fig. 1C,D) and exhibited specific high expression and localization to the basal cell layer by immunofluorescence; these included newly identified markers CAV1 and CAV2, alongside COL17A1, which has recently been identified as an esophagus stem cell marker (Busslinger et al., 2021) (Fig. 1C-E, Fig. S2). We additionally observed high COL17A1 expression in the adult mouse esophagus and human fetal basal cell zone of the esophagus (Fig. S2D,E). scRNA-seq identified LY6D as an enriched marker in the suprabasal/early squamous/differentiated cells and proliferative cluster (Clusters 0, 1) (Fig. 1D, Fig. S2). Supporting this, immunofluorescence showed that LY6D protein is most highly expressed in the cell layers just above the basal cell domain, which has been referred to as the epibasal/early differentiation domain (Zhang et al., 2021) (Fig. 1E, Fig. S2C), and which is also where the majority of KI67+ proliferative cells are observed (Fig. 1E). Within the suprabasal cluster (Cluster 0), we observed that KRT4 is expressed in a low-to-high gradient (Fig. 1D). At the protein level, KRT4 marks the mid-suprabasal epithelium above the LY6D+ epithelium and below CRNNHI differentiated luminal cells (Fig. 1E, Fig. S2C). Finally, Cluster 3 represents a CRNNHI zone that marks terminally differentiated luminal cells in the portion of the squamous epithelium near the lumen (Fig. 1D,E). Our data are consistent with recent publications characterizing adult human esophageal tissue using single-cell transcriptomics (Busslinger et al., 2021; Madissoon et al., 2019), and additionally provides identification of new basal cell markers (i.e. CAV1, CAV2), an epibasal marker (LY6D), and detailed cross-validation of mRNA signatures with protein markers to map different zones of the esophagus (Fig. 1F).

Fig. 1.

Identification of distinct molecular domains within the esophageal epithelium. (A) CDH1+ epithelial cells are sub-clustered from scRNA-seq data from normal matched tissue esophageal biopsies (n=2 biopsies/patients; Pt#1: 45-year-old male; Pt#2: 63-year-old female) from which a total of 7796 cells were analyzed after filtering, and 2651 genes per cell. Louvain clustering was used to predict clusters, which were visualized using UMAP. (B) Distribution of patient cells in each cluster. denotes the average contribution of both samples to the clusters. (C) Dot-plots of the top five genes expressed in each cluster and annotations for each cluster based on top genes and known genes (see also Table S2). (D) Feature plots of top marker genes expressed in each cluster, with CAV1 and COL17A1 for Cluster 2 (basal), KI67 for Cluster 1 (proliferative), LY6D and KRT4 for Cluster 0 (epibasal/suprabasal) and CRNN for Cluster 3 (luminal). (E) Representative immunofluorescence (IF) images in the adult human esophagus validating expression of genes identified by scRNA-seq. COL17A1, CAV1 and CAV2 (see Fig. S2) have the highest expression at the basal zone, KI67 marks proliferative cells at the basal-epibasal zone, LY6D expression is observed at low levels in the basal layer and high levels starting at the suprabasal layer (here referred to as epibasal), KRT4 is a mid differentiation marker and CRNN stains the luminal/cell layer of terminally differentiated cell types. Dotted line delineates the boundary between the lamina propria and the epithelial basal zone. (F) Summary schematic of the different epithelial zones of the adult esophagus with their corresponding markers identified by scRNA-seq and validated by immunofluorescence (validation of other markers is shown in Fig. S2, including higher-magnification images of LY6D). Created with BioRender.com. Images are representative of n=3 biological replicates. Immunofluorescence data was validated across multiple patient samples (summarized in Table S4). Scale bars: 100 μm.

Fig. 1.

Identification of distinct molecular domains within the esophageal epithelium. (A) CDH1+ epithelial cells are sub-clustered from scRNA-seq data from normal matched tissue esophageal biopsies (n=2 biopsies/patients; Pt#1: 45-year-old male; Pt#2: 63-year-old female) from which a total of 7796 cells were analyzed after filtering, and 2651 genes per cell. Louvain clustering was used to predict clusters, which were visualized using UMAP. (B) Distribution of patient cells in each cluster. denotes the average contribution of both samples to the clusters. (C) Dot-plots of the top five genes expressed in each cluster and annotations for each cluster based on top genes and known genes (see also Table S2). (D) Feature plots of top marker genes expressed in each cluster, with CAV1 and COL17A1 for Cluster 2 (basal), KI67 for Cluster 1 (proliferative), LY6D and KRT4 for Cluster 0 (epibasal/suprabasal) and CRNN for Cluster 3 (luminal). (E) Representative immunofluorescence (IF) images in the adult human esophagus validating expression of genes identified by scRNA-seq. COL17A1, CAV1 and CAV2 (see Fig. S2) have the highest expression at the basal zone, KI67 marks proliferative cells at the basal-epibasal zone, LY6D expression is observed at low levels in the basal layer and high levels starting at the suprabasal layer (here referred to as epibasal), KRT4 is a mid differentiation marker and CRNN stains the luminal/cell layer of terminally differentiated cell types. Dotted line delineates the boundary between the lamina propria and the epithelial basal zone. (F) Summary schematic of the different epithelial zones of the adult esophagus with their corresponding markers identified by scRNA-seq and validated by immunofluorescence (validation of other markers is shown in Fig. S2, including higher-magnification images of LY6D). Created with BioRender.com. Images are representative of n=3 biological replicates. Immunofluorescence data was validated across multiple patient samples (summarized in Table S4). Scale bars: 100 μm.

Rapid expansion and long-term maintenance of patient-derived esophageal basal cells in 2D cultures

Over the past several years, work has revealed that long-term culture of human epithelial tissues in vitro requires culture conditions that support epithelial stem cell self-renewal. Using this framework, 2D and 3D models of the human esophagus have been established from induced pluripotent stem cells and primary tissue (Trisno et al., 2018; Bailey et al., 2019; Yamamoto et al., 2016; Kasagi et al., 2018; Liu et al., 2017). Although 3D organoids have advantages, it has also been reported that esophagus organoids are particularly slow-growing and have been reported to have a low terminal passage, with no evidence of recovery after cryopreservation (Kasagi et al., 2018). 2D culture of stratified squamous epithelia, such as skin and esophagus, and other organs with basal stem cells, such as lung, have also been successful (Mou et al., 2016; DeWard et al., 2014). However, the establishment of 2D cultures and single-cell benchmarking to the native human esophagus have not been carried out. We, therefore, aimed to unify previous methods to achieve a long-term culture of the primary human esophagus and to compare in vitro-grown cells to the native in vivo esophagus.

To determine optimal conditions for primary esophagus cell expansion, we screened many combinations of media plus growth factors with a primary outcome measure of robust cell survival and expansion. Our most successful condition for long-term adult human esophageal cultures utilized sub-lethally irradiated feeder cells (3T3-J2i) (Suprynowicz et al., 2017) with the rho-kinase inhibitor (Y-27632) plus CHIR99021, EGF, and dual SMAD inhibition using inhibitors of TGFβ (A-8301) and BMP (noggin). Finally, we tested the addition of hydrocortisone based on previous studies culturing stratified skin epithelium in vitro (Vaughan et al., 1981). We refer to this culture condition as HYENAC (hydrocortisone, Y-27632, EGF, NOG, A-8301, CHIR99021) (see Materials and Methods for details).

To demonstrate the ability of HYENAC media to support long-term esophagus cultures, matched cells from biopsies used to generate the scRNA-seq data in Fig. 1 were plated on the day biopsies were obtained [day (D) 0]. All normal tissues used to generate 2D in vitro cell lines (Table 1) were either finely minced or enzymatically digested to produce single cells or small cell clusters before plating. Both single cells and cell clumps attached and expanded as small colonies, which eventually grew to confluence (Fig. 2A,B). Irradiated 3T3-J2i cells do not proliferate, whereas esophageal cells continue proliferating over time (Fig. 2C). We observed an average population doubling time in the 2D expansion system of 2.6 days (Fig. 2D). This doubling time is consistent with in vivo mouse esophagus studies showing that basal stem cells proliferate on average every 2-3 days (Piedrafita et al., 2020; Doupé et al., 2012). After three passages and 30-40 days in culture, we assessed in vitro cultures using scRNA-seq (n=3). We applied Louvain clustering to reveal five predicted clusters (Fig. 2E, Table S3), and plotted the proportion of total cells within each cluster (Fig. 2F). All clusters were enriched for the epithelial marker CDH1, with little non-epithelial (VIM+) expansion (average of biological replicates CDH1+ 96.6% versus VIM+ 5.4%; Fig. S3E) and CDH1 staining revealed a high percentage of epithelial cells (Fig. 2G-I); quantification of protein staining (percentage of CDH1+ cells, n=7 independent lines; Fig. 2I, top) demonstrated that ∼50-80% of cells were CDH1+. The proportion of CDH1+ cells did not change across passages (Fig. 2I, bottom). The top five most highly enriched genes were plotted (Fig. S3A), revealing that Cluster 4 contains a proliferative signature, and Clusters 1 and 2 expressed markers of basal-stem cells, including CAV1, CAV2 and COL17A1 and other known markers such as TP63, KRT15 and ITGB4 (DeWard et al., 2014; Busslinger et al., 2021) (Fig. 2J, Fig. S3B,C). We observed that Clusters 0 and 3 had cells that expressed low levels of markers typical of the epibasal, early and late differentiation zones (Fig. 2J).

Fig. 2.

Characterization of human esophageal biopsies grown in 2D in vitro at the single-cell level. (A) H&E of a representative biopsy of squamous epithelial cells of the esophagus. (B) Brightfield (BF) images of expansion of esophagus cell clusters/colonies for D3, D9 and D12. Scale bar: 200 µm. (C) Cell proliferation assay (see Materials and Methods) at 24, 48, 72 and 96 h showing proliferation of esophageal cells over time compared with sub-lethally irradiated 3T3-J2 mouse fibroblast cells (unpaired, two-tailed t-test; P=0.027). (D) Doubling life of esophageal 2D in vitro cells [n=4 patients (pt)]. We quantified cell numbers using two cell-counting approaches (by cell counter and by DAPI counts) and calculated the doubling time as previously described (Sherley et al., 1995). The average daily doubling time was 3.19 and 2.87, respectively (unpaired, two-tailed t-test; P=0.9722). (E) A total of 10,550 cells grown in vitro and 4269 genes/cell were analyzed using Louvain clustering and visualized by UMAP to predict five clusters. Clusters 1 and 2 express basal cell markers (see I), Clusters 0 and 3 express markers of early suprabasal cells, and Cluster 4 expresses proliferation markers (see Table S3). (F) Distribution plot of the average () number of cells contributing to each cluster per sample (n=3 biological replicates). (G,H) CDH1 expression across clusters (G), with validation of CDH1 (protein) expression by immunofluorescence (IF) (H), suggesting an enrichment of epithelial cell types using these methods. Scale bar: 100 μm. (I) Top: Quantification of the percentage of total cells that are CDH1+in vitro in multiple patient-derived cell lines [each number on the x-axis represents a unique patient sample (n=7 pt)]. Bottom: Epithelial cells do not significantly increase or decrease over passage number (n=3; unpaired, two-tailed t-test; P=0.1597). Error bars represent s.d. (J) Feature plots of genes identified in primary tissue differential epithelial zones for basal (CAV1, COL17A1), proliferative (KI67), epibasal (LY6D), suprabasal (KRT4) and luminal cells (CRNN). ns, not significant.

Fig. 2.

Characterization of human esophageal biopsies grown in 2D in vitro at the single-cell level. (A) H&E of a representative biopsy of squamous epithelial cells of the esophagus. (B) Brightfield (BF) images of expansion of esophagus cell clusters/colonies for D3, D9 and D12. Scale bar: 200 µm. (C) Cell proliferation assay (see Materials and Methods) at 24, 48, 72 and 96 h showing proliferation of esophageal cells over time compared with sub-lethally irradiated 3T3-J2 mouse fibroblast cells (unpaired, two-tailed t-test; P=0.027). (D) Doubling life of esophageal 2D in vitro cells [n=4 patients (pt)]. We quantified cell numbers using two cell-counting approaches (by cell counter and by DAPI counts) and calculated the doubling time as previously described (Sherley et al., 1995). The average daily doubling time was 3.19 and 2.87, respectively (unpaired, two-tailed t-test; P=0.9722). (E) A total of 10,550 cells grown in vitro and 4269 genes/cell were analyzed using Louvain clustering and visualized by UMAP to predict five clusters. Clusters 1 and 2 express basal cell markers (see I), Clusters 0 and 3 express markers of early suprabasal cells, and Cluster 4 expresses proliferation markers (see Table S3). (F) Distribution plot of the average () number of cells contributing to each cluster per sample (n=3 biological replicates). (G,H) CDH1 expression across clusters (G), with validation of CDH1 (protein) expression by immunofluorescence (IF) (H), suggesting an enrichment of epithelial cell types using these methods. Scale bar: 100 μm. (I) Top: Quantification of the percentage of total cells that are CDH1+in vitro in multiple patient-derived cell lines [each number on the x-axis represents a unique patient sample (n=7 pt)]. Bottom: Epithelial cells do not significantly increase or decrease over passage number (n=3; unpaired, two-tailed t-test; P=0.1597). Error bars represent s.d. (J) Feature plots of genes identified in primary tissue differential epithelial zones for basal (CAV1, COL17A1), proliferative (KI67), epibasal (LY6D), suprabasal (KRT4) and luminal cells (CRNN). ns, not significant.

Table 1.

Clinical characteristics of primary human esophageal samples for 2D or 3D in vitro culture

Clinical characteristics of primary human esophageal samples for 2D or 3D in vitro culture
Clinical characteristics of primary human esophageal samples for 2D or 3D in vitro culture

We observed that these culture conditions supported 2D in vitro culture of human fetal esophageal cells (samples used ranged from 54 to 100 post-conception days). Fetal tissue was mechanically and/or enzymatically digested and cultured as described for adult tissue (Fig. S4A). We observed expansion of both epithelial and mesenchymal cells (Fig. S4A). Immunofluorescence staining for CDH1, TP63 and KRT5 revealed the expansion of epithelial basal/epibasal colonies in vitro (Fig. S4B).

We then sought to determine whether esophagus cell lines could be cryopreserved using HYENAC in combination with standard protocols for freezing (Fig. S5A). We thawed four lines that were established and were in cryopreservation for different lengths of time, and from different age groups, sexes and races (Fig. S5B). All four samples were successfully thawed, expanded and passaged (Fig. S5D). In addition, for fetal samples, we observed the expansion of both epithelial and mesenchyme cell types after thawing (Fig. S5D), consistent with the matched fresh sample (Fig. S4).

2D in vitro-grown esophageal cells molecularly resemble basal cells from native tissue

To interrogate further how closely in vitro samples resemble in vivo esophageal tissue, we carried out three complementary but separate analyses using the scRNA-seq data and validated protein levels (Fig. 3, Fig. S6). First, we directly compared the single-cell transcriptomes of in vivo samples (described in Fig. 1) with in vitro data (described in Fig. 2) (Fig. S6A,B). Second, we integrated in vitro and in vivo data following batch correction to analyze the samples in one analysis (Fig. S6C-F). Third, we performed label transfer using Ingest (Wolf et al., 2017), using the in vivo uniform manifold approximation and projection (UMAP) embedding as a high-dimensional search space and projecting in vitro samples onto the in vivo map (Fig. 3A-F). To directly compare data from in vivo D0 (fresh biopsies) versus in vitro cultures, we obtained gene lists of the most statistically enriched genes in each cluster (from Figs 1 and 2) and examined gene overlap in the enrichment lists. The most highly similar clusters were proliferative (74% overlap) followed by the basal cell clusters (Fig. S6A,B). The in vivo basal cell cluster (Cluster 2) overlapped most closely with in vitro Clusters 1 and 2 with 39% and 33% overlap, respectively. The suprabasal cluster and differentiated in vivo clusters (Clusters 0 and 3) had the highest shared similarity to the differentiating in vitro clusters (Clusters 0 and 3) (Fig. S6A). With respect to the basal cell clusters (Cluster 2 in vivo, Cluster 1 and Cluster 2 in vitro), overlap included many basal cell markers that were common between clusters (i.e. CAV1, CAV2, ITGB4, COL17A1) (Fig. S6B). Next, we directly compared samples by integrating and batch correcting in vivo (epithelium only) and in vitro data with BBKNN (batch balanced k nearest neighbors) (Polański et al., 2020), followed by clustering (Fig. S6C-G). Integrated data generated four predicted clusters (Fig. S6D,E), which could be assigned to proliferative (Cluster 3), basal (Cluster 0), suprabasal (Cluster 1) and luminal (Cluster 2) cell types based on expressed genes (Fig. S6C-F). Both in vitro and in vivo cells contributed to each cluster (Fig. S6C); however, the distribution from in vitro or in vivo cells was not equal for all clusters. For example, only ∼4.5% of cells were designated as basal cells (Cluster 1) from the in vivo sample whereas ∼37.4% of cells were assigned to this cluster from the in vitro sample (Fig. S6G). Lastly, we re-clustered the in vivo data (entire data set, including stroma/immune) (Fig. 3A-C), and then used the Ingest function in Scanpy to project in vitro cells onto the in vivo UMAP embedding (Fig. 3D-F). This analysis revealed that most 2D in vitro cells mapped to the proliferative, basal and epibasal cell clusters of the in vivo search space, with far fewer cells mapping to the differentiated mid-luminal cells (Fig. 3E,F), as highlighted by distribution plots comparing the proportion of cells from in vivo tissue assigned to each cluster versus the projected proportion from in vitro cells (Fig. 3F). Taken together, the data suggest that 2D expanded esophageal cells closely resemble basal-epibasal cells observed in the human adult esophagus.

Fig. 3.

In vitro and in vivo esophagus share a high degree of molecular similarity. (A) Louvain clustering and UMAP visualization of in vivo samples. Blue dotted line highlights the epithelial (CDH1+) cluster and the yellow dotted line highlights VIM+ cells. (B) Feature plots of genes expressed in different cells within the esophagus, including CAV1, COL17A1, KI67, LY6D, KRT4 and CRNN. (C) Distribution of cells from each human sample to each cluster. (D) The Scanpy function Ingest was used to project 2D in vitro-grown cells onto the in vivo cell embedding. In vitro cells map to five clusters, with most cells mapping to in vivo basal (Cluster 2) and suprabasal (Cluster 0) clusters. (E) Feature plots showing expression of basal cell (CAV1, COL17A1), proliferation-associated (KI67), suprabasal marker (LY6D) and differentiated marker (KRT4, CRNN) genes. (F) Quantification of the proportion of cells from the in vivo sample in each cluster and the proportion of in vitro-cultured cells that map to each in vivo cluster, demonstrating that in vitro cultured cells maintain similar basal cell proportions (Cluster 2, green) but have a larger proportion of suprabasal-like cells (Cluster 0, orange). LP, lamina propria. (F) Immunofluorescence validation of COL17A1, CAV2 and KI67 expression co-expressed with TP63 and CDH1. The epibasal/suprabasal marker LY6D could not be detected. Scale bars: 2 mm (COL17A1, CAV2 and LY6D); 50 µm (KI67).

Fig. 3.

In vitro and in vivo esophagus share a high degree of molecular similarity. (A) Louvain clustering and UMAP visualization of in vivo samples. Blue dotted line highlights the epithelial (CDH1+) cluster and the yellow dotted line highlights VIM+ cells. (B) Feature plots of genes expressed in different cells within the esophagus, including CAV1, COL17A1, KI67, LY6D, KRT4 and CRNN. (C) Distribution of cells from each human sample to each cluster. (D) The Scanpy function Ingest was used to project 2D in vitro-grown cells onto the in vivo cell embedding. In vitro cells map to five clusters, with most cells mapping to in vivo basal (Cluster 2) and suprabasal (Cluster 0) clusters. (E) Feature plots showing expression of basal cell (CAV1, COL17A1), proliferation-associated (KI67), suprabasal marker (LY6D) and differentiated marker (KRT4, CRNN) genes. (F) Quantification of the proportion of cells from the in vivo sample in each cluster and the proportion of in vitro-cultured cells that map to each in vivo cluster, demonstrating that in vitro cultured cells maintain similar basal cell proportions (Cluster 2, green) but have a larger proportion of suprabasal-like cells (Cluster 0, orange). LP, lamina propria. (F) Immunofluorescence validation of COL17A1, CAV2 and KI67 expression co-expressed with TP63 and CDH1. The epibasal/suprabasal marker LY6D could not be detected. Scale bars: 2 mm (COL17A1, CAV2 and LY6D); 50 µm (KI67).

To validate these observations, we performed immunofluorescence on 2D cells for basal (COL17A1, CAV2), basal-epibasal (TP63), epibasal (LY6D), proliferative (KI67) and total epithelial (CDH1) markers. The 2D cells showed positive co-expression of basal (COL17A1, CAV2), basal-epibasal (TP63), epithelial (CDH1) markers (Fig. 3G, top) and were proliferation positive (KI67+) (Fig. 3G, bottom). Interestingly, even though the mRNA single-cell data suggested that two in vitro clusters (Clusters 0 and 3) (Fig. 2) express suprabasal genes, we did not observe LY6D expression in the 2D state, suggesting that the 2D expansion method enhances proliferative basal progenitor/stem cells of the esophagus.

Air–liquid interface, high-density culture, and 3D growth in Matrigel drive stratification

The media conditions presented above enriched basal stem cells in vitro; however, the human esophagus is composed of stratified epithelial layers that encompass distinct zones of differentiated cells in layers as they approach the lumen. Air–liquid interface (ALI) has been used previously (Blevins et al., 2018), whereby cells are seeded in Transwell plates and subsequently exposed to air in the upper chamber (5% CO2, 95% ambient air), leading to the formation of a stratified epithelium (Yamamoto et al., 2016). This assay is valuable because the Transwell format allows physiological measurements of the epithelial barrier (Kleuskens et al., 2021; Blevins et al., 2018). Using Transwells, we generated ALI cultures as previously described (Srinivasan et al., 2015) and measured the trans-epithelial electrical resistance (TEER; Ω•cm2; n=4) (Fig. 4A). We observed a significant increase in TEER by day 2 of seeding cells (from 0 to 80.2 Ω•cm2, P=0.001), and, after removing media from the apical chamber on D4, we observed a steady increase of TEER from D2 to D14 (D2 80.2 versus D14 142.4 Ω•cm2, P=0.1109) (Fig. 4A). The average TEER across all days in ALI was 126 Ω•cm2 (n=4). On D14, Transwells were fixed for histological and immunofluorescence staining for TP63 and KRT4 (Fig. 4B,C). We observed that by D14 Transwells had formed a multilayered tissue (Fig. 4B), with a basal layer of TP63+ cells at the bottom of the well and KRT4+ cells on top (Fig. 4B,C).

Fig. 4.

Esophageal stratification in vitro using air-liquid interface and cell density. (A) Schematic at the top depicts a Transwell with media in both chambers followed by removal of media from the upper chamber to create the ALI for differentiation (created with BioRender.com). Graph shows average TEER measurements for n=4 patient 2D in vitro-derived cultures that were seeded onto Transwells, and TEER was measured over time (Ω•cm2). (B,C) Histological (B) and immunofluorescence (IF) (C) characterization of D14 ALI cultures. (D) qPCR for proliferation (KI67), early and terminally differentiated (KRT4 and CRNN, respectively) and basal-suprabasal (TP63) mRNA markers in low versus high-seeded wells from four independent lines. (E) Immunofluorescence staining and quantification of protein expression of the early differentiation marker KRT4 in low versus high cell density assays. (F) Representative image of immunofluorescence for TP63 and KRT4 in primary tissue biopsy of the adult human esophagus. Boxed area is enlarged in the panels below. (G) Representative images of immunofluorescence for TP63 and KRT4 patient-derived in vitro esophagus primary 2D cell cultures, at low (2D) versus high (confocal z-stack maximum projections) density. (H) Quantification of cell types based on protein expression of n=3 independent patient primary tissue (from F and G) versus n=3 in vitro cells at low versus high density. The percentage of each cell type was determined by counting cells positive or negative for the respective markers (x-axis). (I) Representative images of immunofluorescence for KI67 and TP53 in primary tissue biopsy of the adult human esophagus. (J) Representative images of immunofluorescence at low and high density. At low density (top), nuclei identified by human-specific nuclear antigen (Hu-Nu; red) and co-stained with TP63 (green) are highly proliferative (marked by KI67) compared with the same cell line plated at higher density (bottom). (K) Quantification of KI67+/TP63+ cells at low and high confluence. Less-confluent cell colonies are highly proliferative compared with high density, confluent monolayers (n=3; unpaired, two-tailed t-test; P<0.001). All experiments were performed using at least n=3 biological replicates. Scale bars: 200 μm (F, top); 100 μm (B,E,G; F, bottom); 50 μm (C,I,J). Either unpaired, two-tailed t-test or multiple comparisons one-way ANOVA with Bonferroni Correction post-hoc analysis were used to compare the mean of groups. For in vitro cultures, normalization and percentages were calculated using double-positive cells for DAPI/Hu-Nu, to determine human cell percentages in vitro and exclude mouse feeder cells. Unpaired t-test was used to determine statistical significance. ns, not significant (P<0.05); **P≤0.01, ***P≤0.001, ****P≤0.0001). Error bars in A,D,E represent s.d. Dashed horizontal lines in H and K represent 0, and dashed vertical lines are visual divisions between in vivo versus low-density versus high-density cell population counts.

Fig. 4.

Esophageal stratification in vitro using air-liquid interface and cell density. (A) Schematic at the top depicts a Transwell with media in both chambers followed by removal of media from the upper chamber to create the ALI for differentiation (created with BioRender.com). Graph shows average TEER measurements for n=4 patient 2D in vitro-derived cultures that were seeded onto Transwells, and TEER was measured over time (Ω•cm2). (B,C) Histological (B) and immunofluorescence (IF) (C) characterization of D14 ALI cultures. (D) qPCR for proliferation (KI67), early and terminally differentiated (KRT4 and CRNN, respectively) and basal-suprabasal (TP63) mRNA markers in low versus high-seeded wells from four independent lines. (E) Immunofluorescence staining and quantification of protein expression of the early differentiation marker KRT4 in low versus high cell density assays. (F) Representative image of immunofluorescence for TP63 and KRT4 in primary tissue biopsy of the adult human esophagus. Boxed area is enlarged in the panels below. (G) Representative images of immunofluorescence for TP63 and KRT4 patient-derived in vitro esophagus primary 2D cell cultures, at low (2D) versus high (confocal z-stack maximum projections) density. (H) Quantification of cell types based on protein expression of n=3 independent patient primary tissue (from F and G) versus n=3 in vitro cells at low versus high density. The percentage of each cell type was determined by counting cells positive or negative for the respective markers (x-axis). (I) Representative images of immunofluorescence for KI67 and TP53 in primary tissue biopsy of the adult human esophagus. (J) Representative images of immunofluorescence at low and high density. At low density (top), nuclei identified by human-specific nuclear antigen (Hu-Nu; red) and co-stained with TP63 (green) are highly proliferative (marked by KI67) compared with the same cell line plated at higher density (bottom). (K) Quantification of KI67+/TP63+ cells at low and high confluence. Less-confluent cell colonies are highly proliferative compared with high density, confluent monolayers (n=3; unpaired, two-tailed t-test; P<0.001). All experiments were performed using at least n=3 biological replicates. Scale bars: 200 μm (F, top); 100 μm (B,E,G; F, bottom); 50 μm (C,I,J). Either unpaired, two-tailed t-test or multiple comparisons one-way ANOVA with Bonferroni Correction post-hoc analysis were used to compare the mean of groups. For in vitro cultures, normalization and percentages were calculated using double-positive cells for DAPI/Hu-Nu, to determine human cell percentages in vitro and exclude mouse feeder cells. Unpaired t-test was used to determine statistical significance. ns, not significant (P<0.05); **P≤0.01, ***P≤0.001, ****P≤0.0001). Error bars in A,D,E represent s.d. Dashed horizontal lines in H and K represent 0, and dashed vertical lines are visual divisions between in vivo versus low-density versus high-density cell population counts.

During our experiments, we observed that when cells became overly confluent they appeared to pile on top of each other. To investigate confluent/dense cultures in a reproducible manner, we varied the number of cells plated at D0 (Fig. 4D-K). Four independent cell lines (n=4) were plated on coverslips at 125,000 cells/well (low density) or 500,000 cells/well (high density) and were analyzed 5 days later (Fig. 4D). We observed that by D5, the high-density wells had lower expression of KI67, and increased expression of differentiation/stratification markers, such as KRT4 and CRNN, compared with the low-density plated wells (Fig. 4D). We did not observe a change in mRNA expression of the basal/epibasal marker TP63 (Fig. 4D). We fixed the cells and confirmed the expression of KRT4 by quantifying the pixel intensity of the protein in the high-density wells and in the low-density wells (where expression was absent) (Fig. 4E). These data suggest that cell density and confluency of cells lead to differentiation and stratification of cells, without the need for ALI in Transwells.

Given that TP63 and KRT4 are broad markers, with TP63 marking basal and epibasal cells, and KRT4 marking mid and late differentiating cells, we used this combination of markers to delineate basal/epibasal cells (TP63+ KRT4), mid differentiating cells (TP63+ KRT4+) and late differentiating cells (TP63 KRT4+). When we examined tissue biopsies, we observed that all populations of cells were distributed similarly (P>0.05) (Fig. 4G-I). Meanwhile, in the low-density in vitro cell population, we observed a statistically significant enrichment in TP63+ KRT4 cells compared with the in vivo tissue (86.04% low-density in vitro versus 30.99% in vivo; P<0.0001), with little presence of the other two populations (Fig. 4G-I). In the high-density culture, we observed a cellular distribution similar to that of the in vivo tissue: the TP63+ KRT4 cells had an average of 38.4% in the high-density in vitro versus 30.99% in the in vivo, the TP63+ KRT4+ double-positive cells had an average of 28.43% in the high-density in vitro versus 24.66% in the in vivo, and the differentiated TP63 KRT4+ cells had an average of 38.12% in the high-density in vitro versus 13.45% in vivo (Fig. 4G,H). Altogether, high-density culture conditions led to proportions of basal, early and late differentiating cells similar to in vivo tissue (Fig. 4H), without the need for ALI. We similarly quantified the proportion of TP63+ and TP63 cells that co-expressed KI67 (TP63+ KI67+; TP63+ KI67; TP63 KI67+) in culture and in vivo. We observed that the biopsy tissue contained TP63+ cells that were non-proliferative (TP63+ KI67; 38.42%), which was higher than the population of TP63+ KI67+ proliferative cells (16.83%), and proliferative TP63 cells were rarely observed (TP63 KI67+; 1.80%) (Fig. 4J-K). In low-density cultures, we observed significant enrichment of the basal-proliferative TP63+ KI67+ population, whereas the distribution of high-density cultured cells more closely reflected the in vivo distributions (Fig. 4K). Altogether, this suggests that cells maintained at low confluency are enriched for proliferative TP63+ cells, whereas cells at high confluency reflect the range of basal/epibasal and differentiated cells found in the stratified adult human tissue. Nonetheless, we had described TP63+ and KRT4+ as broad markers of basal-epibasal and suprabasal-luminal of the adult esophagus (Fig. 1). Therefore, we investigated how the high confluency in vitro protocol affected expression of COL17A1 (basal) and CRNN (luminal) markers (Fig. S7). Expression of COL17A1 was restricted to the base of the high-density culture, whereas CRNN-expressing cells were sparse and observed at the top of the stratified layers (Fig. S7A). Of note, we observed that during fixation and staining of the high-density cultures, there were populations of cells that were floating in the media and were washed off from the coverslips. Therefore, to test whether the process was causing loss of the CRNN+ population, we added a thin layer of HistoGel to the top of the coverslip before fixing cells. After immunofluorescence staining, we peeled off the HistoGel and observed flat-shaped CRNN+ cells attached to the HistoGel (Fig. S7D), suggesting that in the process of staining and fixing cells the luminal cells become detached and are sloughed off into the media, a phenomenon that is observed normally in the human esophagus.

During the process of generating the esophageal 2D cell lines, we observed that while passaging cells, increasing the media volume led to floating structures that resembled a 3D sphere (Fig. S8). After plating dissociated biopsies onto 6-well plates (9.8 cm2), wells were either given a low volume of media (2 ml) or a high-volume media (4 ml), and the media was changed every 2 days (Fig. S8A). We observed that low media volume led cells to attach to the tissue culture plate whereas high media volume led to the formation of organoids (Fig. S8B). Immunofluorescence of organoids after 30 days revealed an ‘inside out’ structure with the basal/epibasal marker TP63 localized on the inside of the organoid and the differentiation marker KRT4 localized on the outer side towards the media (Fig. S8C). This polarity has been demonstrated with intestinal organoids in suspension culture in the absence of matrix cues (Co et al., 2019; Capeling et al., 2022). This phenomenon was observed in multiple patient lines (Fig. S8D), and we found that the 3D floating organoids could be passaged (Fig. S8E). Lastly, established 2D cultures could be seeded into low-attachment plates and gave rise to organoids with a similar organization (Fig. S8E), and we observed no difference in 3D size area between organoids generated from 2D compared with those generated from primary tissue (Fig. S8E,F). These data show that 3D floating organoids can be generated from primary tissue, or from 2D cultured cells, and resemble the correct polarity observed in the human esophagus.

Finally, we sought to interrogate the potential of cryopreserved 2D cells to recover and form 3D organoids (Fig. 5A). We thawed cells and cultured them in either Matrigel, on low-attachment plates (referred to as suspension culture) or plated them into 2D culture on 3T3-J2i feeder cells (n=3 patient lines). We counted cell numbers at passage 1 and observed significant enrichment of cell number in all systems (Fig. 5B). We observed successful growth of all three conditions by D6 (Fig. 5C,D). When comparing Matrigel versus suspension culture over time, we observed the area/size of organoids to be greater in Matrigel compared with suspension (Fig. 5E). Brightfield images of suspension organoids revealed 3D structures, but we also observed far more debris and floating cells compared with Matrigel cultures (Fig. 5D). Immunofluorescence stains of D6 cultures revealed that suspension organoids are mostly COL17A1+ and have sporadic CRNN+ cells on the outside of the structure. By contrast, Matrigel organoids were CRNN at this early time point (Fig. 5F), with all cells expressing COL17A1+. We quantified KI67+ cells and observed that the presence of Matrigel significantly enhanced proliferative cells compared with suspension cultures (P<0.05) (Fig. 5G). Finally, we evaluated whether suspension and Matrigel organoids could be grown in longer-term cultures. We observed that Matrigel provided a favorable environment for continuous 3D growth in size compared with suspension (Fig. 5H). By D25, Matrigel organoids had grown significantly in size and possessed a complex 3D esophageal structure with the expression of cell-type markers from all esophageal zones (Fig. 5I,J). 3D Matrigel-grown organoids showed an inverse structure (basal progenitor-stem cells towards the outside versus suprabasal and differentiated cells towards the inside of the organoid) compared with suspension organoids. It was not possible to analyze suspension organoids at the D25 time point owing to their small size and inefficient growth in this condition. These results suggest that 2D cultured cells retain their ability to form 3D organoids that can be grown in suspension or in Matrigel.

Fig. 5.

2D basal progenitor-stem cells form entire esophageal epithelial sphere 3D organoids in vitro. (A) Schematic summary of the protocol for generation of basal-progenitor cells in 2D format, cryopreservation, and rescue for 2D/3D expansion. (B) Cell number quantifications of three patient (pt) lines thawed and rescued in either 2D, 3D (suspension versus Matrigel). Cells were counted when 2D wells reached 90-100% confluency and needed to passaging to avoid differentiation (time points vary by doubling rates per patient). (C) 2D expanded cells were passaged to either 3D (suspension versus Matrigel) or re-plated in 2D format (3T3-J2i). (D) By D6, growth was observed in all conditions under brightfield microscopy. (E) Quantification of the area of the sphere of 3D organoids of suspension versus Matrigel over time. (F) At D6, whole-mount immunofluorescence stains were performed on suspension versus Matrigel cultures and maximum projection confocal images for CRNN (luminal; arrowheads), COL17A1 (basal), KI67 (proliferation) and CDH1 (epithelial) are shown. (G) Quantification of the percentage of KI67+ cells in suspension versus Matrigel. (H) Area of 3D spheres over the span of 25 days was measured and quantified for sphere versus Matrigel. (I) Brightfield (BF) images and H&E stains on D25. (J) Matrigel spheres were collected and analyzed by immunofluorescence for the markers COL17A1 (basal), LY6D (epibasal-early), KRT4 (mid) and CRNN (late-luminal differentiation), visualizing the complete epithelial formation of esophageal organoids. Scale bars: 500 μm (D,I, left); 200 μm (I, middle); 100 μm (I, right; J, first three panels); 50 μm (J, last panel). In D,F, insets show enlargements of the boxed areas. Error bars in E,H represent s.d. In G, box limits represent the minimum and maximum values and horizontal line represents the mean. *P≤0.05 (unpaired, two-tailed t-test). Schematics in A and C created with BioRender.com.

Fig. 5.

2D basal progenitor-stem cells form entire esophageal epithelial sphere 3D organoids in vitro. (A) Schematic summary of the protocol for generation of basal-progenitor cells in 2D format, cryopreservation, and rescue for 2D/3D expansion. (B) Cell number quantifications of three patient (pt) lines thawed and rescued in either 2D, 3D (suspension versus Matrigel). Cells were counted when 2D wells reached 90-100% confluency and needed to passaging to avoid differentiation (time points vary by doubling rates per patient). (C) 2D expanded cells were passaged to either 3D (suspension versus Matrigel) or re-plated in 2D format (3T3-J2i). (D) By D6, growth was observed in all conditions under brightfield microscopy. (E) Quantification of the area of the sphere of 3D organoids of suspension versus Matrigel over time. (F) At D6, whole-mount immunofluorescence stains were performed on suspension versus Matrigel cultures and maximum projection confocal images for CRNN (luminal; arrowheads), COL17A1 (basal), KI67 (proliferation) and CDH1 (epithelial) are shown. (G) Quantification of the percentage of KI67+ cells in suspension versus Matrigel. (H) Area of 3D spheres over the span of 25 days was measured and quantified for sphere versus Matrigel. (I) Brightfield (BF) images and H&E stains on D25. (J) Matrigel spheres were collected and analyzed by immunofluorescence for the markers COL17A1 (basal), LY6D (epibasal-early), KRT4 (mid) and CRNN (late-luminal differentiation), visualizing the complete epithelial formation of esophageal organoids. Scale bars: 500 μm (D,I, left); 200 μm (I, middle); 100 μm (I, right; J, first three panels); 50 μm (J, last panel). In D,F, insets show enlargements of the boxed areas. Error bars in E,H represent s.d. In G, box limits represent the minimum and maximum values and horizontal line represents the mean. *P≤0.05 (unpaired, two-tailed t-test). Schematics in A and C created with BioRender.com.

Altogether, we have created a comprehensive map of the human adult epithelium at single-cell resolution with protein validation (Fig. 6). Furthermore, we have demonstrated and validated that 2D/3D organoids can be employed for modeling the complete epithelial layers, thereby enabling in vitro modeling of the human adult esophagus (Fig. 6).

Fig. 6.

Mapping the adult human esophagus in vivo and in vitro. Schematic representation describing and summarizing the findings of this study. We described a comprehensive map of different zones within the adult human epithelium at the single-cell level and validated it with protein markers. Markers for basal and three different zones in the suprabasal are described. Further, we show that human cells expanded in 2D are basal-proliferative and retain their ability to form an entire 3D esophagus either in suspension, Matrigel, ALI, or on a cover-slide without the need of ALI. Created with BioRender.com.

Fig. 6.

Mapping the adult human esophagus in vivo and in vitro. Schematic representation describing and summarizing the findings of this study. We described a comprehensive map of different zones within the adult human epithelium at the single-cell level and validated it with protein markers. Markers for basal and three different zones in the suprabasal are described. Further, we show that human cells expanded in 2D are basal-proliferative and retain their ability to form an entire 3D esophagus either in suspension, Matrigel, ALI, or on a cover-slide without the need of ALI. Created with BioRender.com.

Recent advances using scRNA-seq in the human esophagus have described epithelial subpopulations in more detail (Madissoon et al., 2019; Busslinger et al., 2021). Here, we add additional data sets to this emerging body of literature, and we also validated markers from all epithelial populations predicted by scRNA-seq in tissue sections, identifying unique markers of the basal cell layer (i.e. CAV1, CAV2), and the epibasal zone (i.e. LY6D), which has highest expression restricted to the second to fourth layers of cells. The LY6D/epibasal domain also corresponds to the domain with the highest amount of proliferation. In agreement with previous reports (Barbera et al., 2015), we find few KI67+ cells in the basal layer. Finally, consistent with previous work, we observed that KRT4 marks early differentiating luminal cells along with more terminally differentiated cells closest to the lumen marked by CRNN. The current work is also the first to describe in vitro-cultured esophageal cells using primary in vivo tissue as a benchmark. Of note, we found that the transcription factor TP63, which is known for its role in basal cell regulation in several tissues such as the skin, lungs, and esophagus in mice (Zhang et al., 2017, 2021; Domyan et al., 2011; Mou et al., 2016; Daniely et al., 2004), is not restricted to the basal zone in the human adult esophagus but is more broadly expressed throughout the basal and suprabasal zone, highlighting species-specific differences in gene/protein expression (Uhlen, 2005).

Although the focus of this work was to develop and describe robust in vitro systems to study the human esophagus, we also used these methods to develop a diverse biobank of human specimens. Our long-term motivations for developing this biorepository are to improve our understanding of how genetic differences and ancestry play a role in injury repair or disease. Multiple pediatric and adult esophageal diseases are known to have gender and racial disparities in their incidence and presentation (Hall, 2020; Abraham et al., 2016; Weiler et al., 2014). For example, a recent study interrogating patients with a history of gastroesophageal reflux disease (GERD), followed a cohort of both African Americans and European Americans and found that 8% of the European American cohort developed the pre-malignant condition Barrett's esophagus, whereas none of the African American cohort did (Alkaddour et al., 2015). These patient studies suggest that there are differences in how tissues respond to injury repair in the esophagus. Given that greater than 95% of the immortalized esophageal cell lines that have been used in the past are derived from populations of European descent (Rojas et al., 2020), enhancing the availability of diverse primary tissue will lead to an improved understanding of cell behavior, homeostasis and disease across the human population (Rojas et al., 2020; Pepejoy and Fullerton, 2016).

A key characteristic of esophageal basal stem cells is their potential to differentiate into a fully stratified tissue resembling the multilayered in vivo structure. Numerous studies have utilized ALI to drive in vitro esophagus cells to form a stratified and differentiated 3D tissue (Mou et al., 2016; Yamamoto et al., 2016). We also observed in the current study that these cells are capable of stratifying into an organized basal-to-luminal axis in ALI (Mou et al., 2016; Yamamoto et al., 2016). ALI is an important technical tool for studying the epithelial cell barrier function (Kleuskens et al., 2021; Blevins et al., 2018). Although ALI requires the use of Transwells, it is also of interest that a similar stratified tissue, possessing basal cells and differentiated cells, can be formed by solely modifying cell density in the well of a standard tissue culture plate. Contact-inhibition of epithelial cell proliferation is well understood to drive the differentiation (Pavel et al., 2018; Eagle and Levine, 1967). Such a system is straightforward and may enable scale-up for high-throughput studies or studies involving the simultaneous expansion of many patient-derived cell lines. The ability to generate human esophageal 2D monolayers and 3D structures (by either high-density suspension or Matrigel) coupled with the ability to cryopreserve and expand these cultures robustly, means that broad distribution and implementation of these cells for different kinds of experimental assays will enable and enhance research aimed at understanding esophageal development, homeostasis and disease. Furthermore, these systems will be important tools to advance our understanding of how ancestral genetics, environmental factors and carcinogens may differentially impact disease across patient populations.

Although only scratching the surface, with robust methods and benchmarking of in vitro cells to in vivo tissue using scRNA-seq and protein expression analysis, we have begun to develop a biobank of esophageal cell lines from across the diversity of the human population, and from a wide range of healthy and diseased states (Table 1). In vitro cellular models of the human esophagus can therefore be used to assess stem cell homeostasis as well as the effects of damaging and carcinogenic agents in patient-derived cells across the spectrum of diversity in the human population.

Sample collection

Histologically normal biopsies of the esophageal squamous epithelium were collected from consenting men and women who underwent upper endoscopy or surgical resection between 2017 and 2020 at the time of scheduled inflammatory bowel disease or Barrett's Esophagus screening or tumor surgical resection at the University of Michigan Health System (now Michigan Medicine). Samples were collected using protocols approved by the University of Michigan Institutional Review Board (IRB). Fresh samples were collected in cold HYENAC, and either used immediately for single-cell dissociation followed by scRNA-seq or were processed for culture; otherwise, biopsies were cryopreserved and stored at −80°C until use, at which point they were thawed and processed for generating new culture lines. To cryopreserve biopsies, tissue was minced into fine pieces and frozen using 1 ml/vial of HYENAC with 20% calf bovine serum (CBS) serum and 10% DMSO. Cryovials were placed in Mr. Frosty freezing containers (Fisher Scientific) at −80°C overnight and moved to liquid nitrogen tanks for long-term preservation. For culturing cell lines, we found that fresh biopsies could be processed immediately and grown at a 100% success rate. We observed that biopsy samples in HYENAC medium can be kept for 24 h at 4°C and can be grown successfully. Furthermore, biopsies can be shipped cold overnight (or cryopreserved with 10% DMSO), at which point viable cell lines could still be robustly established. For access to detailed protocols, see https://www.umichttml.org/protocols. Cell turnover was calculated using a previously described method (Sherley et al., 1995). All in vitro cultures were generated in the lab and tested negative for Mycoplasma. Cell proliferation was measured using Cell Proliferation Reagent WST-1 assay (Millipore Sigma, 5015944001).

Patient sample information

The patient's race was self-identified. For white non-Hispanics, we used the nomenclature European American; for Black, we used African American. Biological replicates utilized for scRNA-seq came from normal squamous biopsies. See Table S4 for patient Spence Lab biobank identification numbers and patient characteristics. Cell lines are available upon request.

Tissue processing and staining

Patient biopsy/tissues were fixed in 4% paraformaldehyde (Sigma-Aldrich) overnight, washed with PBS, and then dehydrated in an alcohol series: 30 min each in 25%, 50%, 75% methanol:PBS/0.05% Tween-20, followed by 100% methanol, 100% ethanol and 70% ethanol. Tissue was processed into paraffin using an automated tissue processor (Leica ASP300). Paraffin blocks were sectioned at 7 μm thickness, and immunohistochemical staining was performed as previously described (Spence et al., 2011). Briefly, slides were rehydrated in a series of Histo-Clear, 100% ethanol, 95% ethanol, 70% ethanol, 30% ethanol, and diH2O with two changes of 3 min each. Antigen retrieval was performed in 1× sodium citrate buffer in a vegetable steamer for 40 min. Following antigen retrieval, slides were washed in PBS and permeabilized for 10 min in 0.1% Triton X-100 in 1× PBS, then blocked for 45 min in 0.1% Tween-20, 5% normal donkey serum in PBS. Primary antibodies were diluted in blocking solution and applied overnight at 4°C. Slides were then washed three times in 1× PBS. Secondary antibodies and DAPI were diluted in blocking solution and applied for 60 min at room temperature. Slides were then washed three times in 1× PBS and coverside with ProLong Gold. For a detailed list of antibodies and conditions, see Table S5.

Complete growth media

The complete esophagus stem/progenitor expansion HYENAC media composition was as follows: (1) base media [advanced DMEM/F-12 (Invitrogen, 12634028), 2 mM GlutaMAX (Gibco, 35050061), 10 mM HEPES, 1× N-2 media supplement (Thermo Fisher, 17502001), 1× B-27 Supplement Minus Vitamin A (Thermo Fisher, 12587001), 1 mM N-acetyl-L-cysteine (Millipore Sigma, A9165)]; (2) growth factors and other additives (HYENAC): 900 nM hydrocortisone at (0.326ug/ml) (Sigma-Aldrich, H0888), 10 µM Y-27632 (Tocris; 125410), 100 ng/ml huEGF (R&D Systems, 236-EG-01M), 100 ng/ml noggin (recombinant human noggin; R&D Systems, 6057-ng), 500 nm A 83-01 (Tocris, 2939), 2.5 µM CHIR99021 (Tocris, 4423)]; (3) microbial prevention [100 µg/ml Primocin (InvivoGen, ant-pm-1)]; and (4) serum for support of the 3T3-J2i feeder cells [20% CBS (ATCC, 30-2030)]. For the first 48 h after biopsy collection we added 25 µg/ml gentamicin (Gibco, 15710064) and 2.5 µg/ml amphotericin (Gibco, 15290018) to minimize culture contamination. Cryopreservation medium was composed of complete HYENAC with 25 µg/ml gentamicin (Gibco, 15710064) and 2.5 µg/ml amphotericin (Gibco, 15290018) and the addition of 10% DMSO at the time of freezing.

Hematoxylin and Eosin

Hematoxylin and Eosin (H&E) staining was performed using Harris Modified Hematoxylin (Fisher Scientific) and Shandon Eosin Y (Thermo Scientific) according to the manufacturer's instructions.

Imaging and image processing

Fluorescently stained slides were imaged using a Nikon A-1 confocal microscope. Brightness and contrast adjustments were carried out using ImageJ (National Institute of Health, USA), and adjustments were made uniformly across images. For Fig. 4E, three images of DAPI and KRT4 were taken at the same exposure for each patient (n=3). Photos were uploaded to Fiji, and full image pixel intensity was quantified for DAPI and KRT4. Pixel intensity was normalized to their matched DAPI image and plotted as a ratio KRT4/DAPI.

For Fig. 5 3D growth, brightfield images (n=3-4 photos per well) over five time points (D5, D10, D14, D18, D25) were taken. The open-source image analysis software Fiji was used to measure the area of all organoids per image. For Fig. S7B,C, Fiji was utilized to determine basal versus luminal using orthogonal views and maximum intensity projections. z-stack images were opened in Fiji, then Image>Stacks>reslice followed by Image>stack>z-project were selected.

Schematics and diagrams

Schematics were created with BioRender.com and Adobe Illustrator.

Quantification and statistical analyses

Statistical analyses and plots were generated using Prism 8 software (GraphPad). For all statistical tests, a significance threshold of 0.05 was used. For each analysis, P-values are reported in the figures as: P>0.05, *P≤0.05, **P≤0.01, ***P≤0.001, ****P≤0.0001. Details of statistical tests are given in the figure legends. Except for scRNA-seq, three human tissue (HT) lines were used across experiments with two or three independent experiments and two or three technical replicates per experiment.

Tissue dissociation for scRNA-seq

To dissociate patient biopsies for scRNA-seq, tissue was placed in a Petri dish with ice-cold 1× HBSS (with Mg2+, Ca2+). To prevent adhesion of cells, all tubes and pipette tips were pre-washed with 1% bovine serum albumin (BSA) in 1× HBSS. The tissue was minced manually using spring-squeeze scissors before being transferred to a 15 ml conical flask containing 1% BSA in HBSS. Tubes were spun down at 500 g for 5 min at 10°C, after which excess HBSS was aspirated. Mix 1 from the Neural Tissue Dissociation Kit (Miltenyi, 130-092-628) containing dissociation enzymes and reagents was added and incubated at 10°C for 15 min. Mix 2 from the Neural Tissue Dissociation Kit was added, and the suspension was fluxed through P1000 pipette tips, interspersed with 10 min incubations at 10°C. Flux steps were repeated as needed until cell clumps were no longer visible under a stereomicroscope. Cells were filtered through a 1% BSA-coated 70 μm filter using 1× HBSS, spun down at 500 g for 5 min at 10°C, and resuspended in 500 μl 1× HBSS (with Mg2+, Ca2+). Then 1 ml of RBC Lysis Buffer (Roche, 11814389001) was added, and tubes were incubated on a rocker at 4°C for 15 min. Cells were spun down at 500 g for 5 min at 10°C, then washed twice in 2 ml 1% BSA, being spun down at 500 g for 5 min at 10°C each time. A hemocytometer was used to count cells, and samples were then spun down and resuspended to reach a concentration of 1000 cells/μl and kept on ice.

Single-cell library preparation

The 10x Chromium at the University of Michigan Advanced Genomics Core facility was then used to create single-cell droplets with a target of capturing 5000-10,000 cells. Single-cell libraries were prepared using the Chromium Next GEM Single Cell 3′ Library Construction Kit v3.1 according to the manufacturer's instructions.

Sequencing data processing and cluster identification

The University of Michigan Advanced Genomic Core Illumina Novaseq performed all scRNA-seq. Gene expression matrices were constructed from raw data by the 10x Genomic Ranger with the human reference genome (hg19). Single Cell Analysis for Python was utilized for analysis as previously described by Wolf et al. (2017). Filtering parameters for gene count range, unique molecular identifier (UMI) counts, and mitochondrial transcript fraction were implemented for each data set to verify high-quality input data. All tissue data sets were combined after organ-specific quality filtering had been performed for the remainder of the processing. Highly variable genes were removed, gene expression levels were log normalized, and effects of UMI count and mitochondrial transcript function variations were regressed via linear regression. z-transformation was then performed on gene expression values before samples were again separated by organ for downstream analysis. The UMAP algorithm (Becht et al., 2019; McInnes et al., 2018 preprint) was utilized alongside Louvain algorithm cluster identification within Scanpy with a resolution of 0.6 (Blondel et al., 2008) to perform a graph-based clustering of the top 10-11 principal components. A detailed protocol for tissue dissociation for scRNA-seq can be found at www.jasonspencelab.com/protocols.

Computational analysis of scRNA-seq

Overview

To visualize distinct cell populations within the scRNA-seq dataset, we employed the general workflow outlined by the Scanpy Python package (Wolf et al., 2017). This pipeline includes the following steps: filtering cells for quality control, log normalization of counts per cell, extraction of highly variable genes, regressing out specified variables, scaling, reducing dimensionality with principal component analysis (PCA), UMAP (Becht et al., 2019), and clustering by the Louvain algorithm (Blondel et al., 2008).

Sequencing data and processing FASTQ reads into gene expression matrices

All scRNA-seq was performed at the University of Michigan Advanced Genomics Core with an Illumina Novaseq 6000. The 10x Genomics Cell Ranger pipeline was used to process raw Illumina base calls (BCLs) into gene expression matrices. BCL files were demultiplexed to trim adaptor sequences and UMIs from reads. Each sample was then aligned to the human reference genome (hg19) to create a filtered feature bar code matrix that contains only the detectable genes for each sample.

Quality control

To ensure the quality of the data, all samples were filtered to remove cells expressing too few or too many genes (Figs 1-3, Figs S1-S4: <500, >7500, or a fraction of mitochondrial genes greater than 0.2).

Normalization and scaling

Data matrix read counts per cell were log-normalized, and highly variable genes were extracted. Using the simple linear regression functionality in Scanpy, the effects of total reads per cell and mitochondrial transcript fraction were removed. The output was then scaled by a z-transformation. Following these steps, the following analyses were achieved: Fig. S1, 9039 cells, 3897 genes; Fig. 1, Fig. S2 (extracted), 7796 cells, 2651 genes; Fig. 2, Fig. S3A-C, 10,550 cells, 4269 genes; Fig. S3E,F (HT239), 4617 cells, 4845 genes; Fig. S3E,F (HT344), 895 cells, 3034 genes; Fig. S3E,F (HT328), 4133 cells, 5195 genes; Fig. S3C-H, 7389 cells, 3413 genes; Fig. S4, 19,589 cells, 3486 genes.

Variable gene selection

Highly variable genes were selected by splitting genes into 20 equal-width bins based on log-normalized mean expression. Normalized variance-to-mean dispersion values were calculated for each bin. Genes with log-normalized mean expression levels between 0.125 and 3 and normalized dispersion values above 0.5 were considered highly variable and extracted for downstream analysis.

Batch correction

We noticed batch effects when clustering data owing to technical artifacts, such as data acquisition timing or dissociation protocol differences. To mitigate these effects, we used the Python package BBKNN (Polański et al., 2020). BBKNN was selected over other batch correction algorithms because of its compatibility with Scanpy and optimal scaling with large datasets. This tool was used in place of the nearest neighbor embedding functionality of Scanpy. BBKNN uses a modification of the k nearest neighbor algorithm by first splitting the dataset into batches defined by technical artifacts. For each cell, the nearest neighbors are then computed independently per batch rather than finding the nearest neighbors for each cell in the entire dataset. This helps form connections between similar cells in different batches without altering the PCA space. After completion of batch correction, cell clustering should no longer be driven by technical artifacts.

Dimension reduction and clustering

PCA was conducted on the filtered expression matrix as follows. Using the top principal components, a neighborhood graph was calculated for the nearest neighbors (Fig. S1, 16 principal components, 30 neighbors; Fig. 1, Fig. S2, nine principal components, 11 neighbors; Fig. 2, Fig. S3, 30 principal components, 16 neighbors; Fig. 3C-H, 11 principal components, 15 neighbors; Fig. S4, 16 principal components, 30 neighbors). BBKNN was implemented when necessary and calculated using the top 50 principal components with three neighbors per batch. The UMAP algorithm was then applied for visualization in two dimensions. Using the Louvain algorithm, clusters were identified with the following resolutions: Fig. S1, 0.3; Fig. 1, Fig. S2, 0.2; Fig. 2, Fig. S3, 0.3; Fig. 3, 0.4; Fig. S4, 0.3.

Cluster annotation

Each cluster's general cell identity was annotated using canonically expressed gene markers. Markers utilized included epithelium (CDH1), mesenchyme (VIM), neuronal (POSTN, S100B, STMN2, ELAVL4), endothelial (ESAM, CDH5, CD34, KDR), and immune (CD53, VAMP8, CD48, ITGB2).

Sub-clustering

After annotating clusters within the UMAP embedding, specific clusters of interest were identified for further sub-clustering and analysis. The corresponding cells were extracted from the original filtered but unnormalized data matrix to include 9039 cells and 3897 genes (Fig. 1A, Fig. S2). The extracted cell matrix underwent log normalization, variable gene extraction, linear regression, z-transformation, and dimension reduction to obtain a 2D UMAP embedding for visualization.

3D in vitro modeling

ALI

On D0, 200,000 cells were seeded into Transwells. HYENAC medium was kept in both chambers for 72 h after seeding. The medium on the upper chamber was removed 3 days post-seeding to create the ALI, as previously described. To measure TEER, the EVOM2 epithelial voltohmmeter was employed (World Precision Instruments). The medium was briefly added to the apical chamber for ALI cultures and then removed following measurements. On D14, samples in wells were either fixed, dehydrated, embedded into paraffin, sectioned and stained for H&E or fixed in 100% methanol for 20 min at −20°C followed by two washes of PBS and then stored at 4°C until immunofluorescence staining. On the day of the immunofluorescence procedure, wells were incubated with 10% neutral buffered formalin (20 min), washed with TBS solution for 5 min, incubated with 100% cold methanol for 5 min, washed with TBS for 20 min and incubated with a blocking solution (1% BSA, 5% donkey serum and 0.2% Triton X-100 in TBS) for 1 h. Primary antibodies (diluted in TBS solution with 1% BSA) were added, and plates were incubated overnight at 4°C in a humidifier chamber. Primary antibodies were used at a concentration of 1:500, except for Hu-Nu, which was used at 1:200. Following primary antibody incubation, samples were washed with Tris-buffered saline with 0.1% Tween 20 detergent (TBS-T) three times for 5 min each and then incubated with the respective fluorescence secondary antibodies (1:500) for 1 h. After a further three washes with TBS-T for 5 min each, samples were transferred to microscope slides as follows. A droplet of 5-10 µl of ProLong Gold was added to a microscope slide (trying to minimize bubble formation). The Transwell was carefully removed from the plate with a scalpel, and the transmembrane carefully detached from the Transwell with tweezers and placed onto a microscope slide, making sure the side with cells was in contact with the glass of the slide. Slides were then stored at 4°C or −20°C until ready for imaging by confocal microscopy.

Stratification using cell density

On D0, Siliconized Glass Circle Cover Slides [HR3-277 (12 mm), Hampton Research] were added into each well of a 24-well plate and plates were pre-coated with ∼18,000-20,000 cells of 3T3-J2i in 500 µl/well (DMEM+20% CBS). On D1, cells were trypsinized, counted, resuspended and plated at 50,000-125,000 cells/well (low density) or 500,000 cells/well (high density). The medium was changed every 2 days. On D5 or later (depending on the doubling life of each cell line), cells were washed with 500 µl/well of PBS and fixed with 500 µl/well 100% cold methanol for 20 min at −20°C. Subsequently, cells were briefly washed three times with 500 µl/well PBS and stored with 500 µl/well PBS at 4°C until ready for immunofluorescence staining as described above. See Fig. S7A for a schematic summary of the protocol.

Matrigel

On D0, 2D expanded cells were trypsinized, counted, resuspended and plated at 50 to 125,000 cells/well into Matrigel (356234, Matrigel Matrix, Corning). In short, cells were spun and embedded in Matrigel droplets, then incubated for 10-15 min to solidify the Matrigel. The media was changed every 2-3 days until sample collection and processing. For detailed information on Matrigel protocols, please refer to: https://www.corning.com/worldwide/en/products/life-sciences/products/surfaces/matrigel-matrix.html.

Suspension

Suspension plates were prepared as previously described (Capeling et al., 2022). On D0, 2D expanded cells were trypsinized, counted, resuspended and plated at 50,000-125,000 cells/well into suspension plates. The media was changed every 2-3 days until sample collection and processing.

Biological materials availability

This study generated a comprehensive, diverse biobank of the esophagus with matched tissues for disease study (Biobank n=55 samples for 2D or 3D in vitro cultures). For detailed information and availability of samples visit: https://docs.google.com/spreadsheets/d/1kUDDfv_7XEv-pq9Eq8DjLCCOL2K0Rsx5/edit#gid=1362394931.

We acknowledge and thank the University of Michigan Advanced Genomics Core for its expertise in operating the 10x Chromium single-cell capture platform and sequencing. We also thank the University of Michigan Microscopy Core for providing access to confocal microscopes and image analysis software and The University of Washington Laboratory of Developmental Biology staff. We extend a special thank you to Olivia I. Koues, Tricia Tamsen and Judith M. Opp. We acknowledge and thank the University of Michigan Histology core for their support in the histology processing of samples and a special thank you to Emma Snyder-White. We thank the Michigan Medicine Translational Tissue Modeling Laboratory (TTML) for providing growth medium and technical advice for human in vitro modeling. The Translational Tissue Modeling Laboratory is a University of Michigan-funded initiative (Center for Gastrointestinal Research, Office of the Dean, Comprehensive Cancer Center, Departments of Pathology, Pharmacology, and Internal Medicine) with support by the Endowment for Basic Sciences.

Author contributions

Conceptualization: D.F.-T., J.R.S.; Methodology: D.F.-T., M.A.H., M.K.D., J.R.S.; Software: J.H.W.; Validation: D.F.-T., C.J.Z., M.A.H., D.E.B., J.R.S.; Formal analysis: D.F.-T., J.H.W., C.J.Z., M.A.H., J.R.S.; Investigation: D.F.-T., M.A.H., A.W., J.R.S.; Resources: D.F.-T., M.A.H., M.K.D., A.W., K.K., C.L.M., M.S.B., A.J.C., D.J.T., S.H., Y.-H.T., A.J.M., T.W., S.P.H., D.K.T., J.L., P.D.R.H., J.S., J.R.S.; Data curation: D.F.-T., J.H.W., C.J.Z., J.R.S.; Writing - original draft: D.F.-T.; Writing - review & editing: D.F.-T., J.H.W., C.J.Z., M.A.H., M.K.D., J.R.S.; Visualization: D.F.-T., J.H.W., C.J.Z., E.M.H., A.J.M., J.R.S.; Supervision: D.F.-T., J.R.S.; Project administration: D.F.-T., J.R.S.; Funding acquisition: D.F.-T., J.R.S.

Funding

J.R.S. was supported by funding from the Chan Zuckerberg Initiative Seed Network, the University of Michigan Center for Gastrointestinal Research (UMCGR), and the National Institute of Diabetes and Digestive and Kidney Diseases (5P30DK034933). D.F.-T. was supported by the Center for Plasticity and Organ Design (CPOD) at the Medical School, University of Michigan (T32HD007505), the Michigan Institute for Clinical and Health Research (UL1TR002240, KL2TR002241, TL1TR002242) and the National Institute of Diabetes and Digestive and Kidney Diseases (1K99DK133804-01). Deposited in PMC for release after 12 months.

Data availability

Sequencing data used in this study are deposited at EMBL-EBI ArrayExpress under accession number E-MTAB-12266. Code used for single-cell analysis and data visualization is available at https://github.com/jason-spence-lab/Ferrer-Torres_Development_2022.

Abraham
,
A.
,
Lipka
,
S.
,
Hajar
,
R.
,
Krishnamachari
,
B.
,
Virdi
,
R.
,
Jacob
,
B.
,
Viswanathan
,
P.
and
Mustacchia
,
P.
(
2016
).
Erosive esophagitis in the obese: the effect of ethnicity and gender on its association
.
Gastroenterol. Res. Pract.
2016
,
7897390
.
Alkaddour
,
A.
,
McGaw
,
C.
,
Hritani
,
R.
,
Palacio
,
C.
,
Nakshabendi
,
R.
,
Munoz
,
J. C.
and
Vega
,
K. J.
(
2015
).
African American ethnicity is not associated with development of Barrett's Oesophagus after erosive Oesophagitis
.
Dig. Liver Dis.
47
,
853
-
856
.
Arnold
,
M.
,
Laversanne
,
M.
,
Brown
,
L. M.
,
Devesa
,
S. S.
and
Bray
,
F.
(
2017
).
Predicting the future burden of esophageal cancer by histological subtype: international trends in incidence up to 2030
.
Am. J. Gastroenterol.
112
,
1247
-
1255
.
Bailey
,
D. D.
,
Zhang
,
Y.
,
van Soldt
,
B. J.
,
Jiang
,
M.
,
Suresh
,
S.
,
Nakagawa
,
H.
,
Rustgi
,
A. K.
,
Aceves
,
S. S.
,
Cardoso
,
W. V.
and
Que
,
J.
(
2019
).
Use of HPSC-derived 3D organoids and mouse genetics to define the roles of YAP in the development of the esophagus
.
Development
146
,
dev178855
.
Barbera
,
M.
,
Pietro
,
M. D.
,
Walker
,
E.
,
Brierley
,
C.
,
MacRae
,
S.
,
Simons
,
B. D.
,
Jones
,
P. H.
,
Stingl
,
J.
and
Fitzgerald
,
R. C.
(
2015
).
The human squamous oesophagus has widespread capacity for clonal expansion from cells at diverse stages of differentiation
.
Gut
64
,
11
-
19
.
Becht
,
E.
,
McInnes
,
L.
,
Healy
,
J.
,
Dutertre
,
C. A.
,
Kwok
,
I. W. H.
,
Ng
,
L. G.
,
Ginhoux
,
F.
and
Newell
,
E. W.
(
2019
).
Dimensionality reduction for visualizing single-cell data using UMAP
.
Nat. Biotechnol.
37
,
38
-
47
.
Blevins
,
C. H.
,
Iyer
,
P. G.
,
Vela
,
M. F.
and
Katzka
,
D. A.
(
2018
).
The esophageal epithelial barrier in health and disease
.
Clin. Gastroenterol. Hepatol.
16
,
608
-
617
.
Blondel
,
V. D.
,
Guillaume
,
J. L.
,
Lambiotte
,
R.
and
Lefebvre
,
E.
(
2008
).
Fast unfolding of communities in large networks
.
J. Stat. Mech., Theory Exp.
2008
,
1
-
12
.
Busslinger
,
G. A.
,
Weusten
,
B. L. A.
,
Bogte
,
A.
,
Begthel
,
H.
,
Brosens
,
L. A. A.
and
Clevers
,
H.
(
2021
).
Human gastrointestinal epithelia of the esophagus, stomach, and duodenum resolved at single-cell resolution Ll Ll human gastrointestinal epithelia of the esophagus, stomach, and duodenum resolved at single-cell resolution
.
Cell Rep.
34
,
108819
.
Capeling
,
M. M.
,
Huang
,
S.
,
Childs
,
C. J.
,
Wu
,
J. H.
,
Tsai
,
Y. H.
,
Wu
,
A.
,
Garg
,
N.
,
Holloway
,
E. M.
,
Sundaram
,
N.
,
Bouffi
,
C.
et al.
(
2022
).
Suspension culture promotes serosal mesothelial development in human intestinal organoids
.
Cell Rep.
38
,
110379
.
Co
,
J. Y.
,
Margalef-Català
,
M.
,
Li
,
X.
,
Mah
,
A. T.
,
Kuo
,
C. J.
,
Monack
,
D. M.
and
Amieva
,
M. R.
(
2019
).
Controlling epithelial polarity: a human enteroid model for host-pathogen interactions
.
Cell Rep.
26
,
2509
-
2520
.
Daniely
,
Y.
,
Liao
,
G.
,
Dixon
,
D.
,
Ilona Linnoila
,
R.
,
Lori
,
A.
,
Randell
,
S. H.
,
Oren
,
M.
and
Jetten
,
A. M.
(
2004
).
Critical role of P63 in the development of a normal esophageal and tracheobronchial epithelium
.
Am. J. Physiol. Cell Physiol.
287
,
171
-
181
.
DeWard
,
A. D.
,
Cramer
,
J.
and
Lagasse
,
E.
(
2014
).
Cellular heterogeneity in the mouse esophagus implicates the presence of a nonquiescent epithelial stem cell population
.
Cell Reports
9
,
701
-
711
.
Domyan
,
E. T.
,
Ferretti
,
E.
,
Throckmorton
,
K.
,
Mishina
,
Y.
,
Nicolis
,
S. K.
and
Sun
,
X.
(
2011
).
Signaling through BMP receptors promotes respiratory identity in the foregut via repression of Sox2
.
Development
138
,
971
-
981
.
Doupé
,
D. P.
,
Alcolea
,
M. P.
,
Roshan
,
A.
,
Zhang
,
G.
,
Klein
,
A. M.
,
Simons
,
B. D.
and
Jones
,
P. H.
(
2012
).
A single progenitor population switches behavior to maintain and repair esophageal epithelium
.
Science
337
,
1091
-
1093
.
Eagle
,
H.
and
Levine
,
E. M.
(
1967
).
Growth regulatory effects of cellular interaction
.
Nature
72
,
1102
-
1106
.
El-Serag
,
H. B.
and
Sonnenberg
,
A.
(
1997
).
Associations between Di V Erent forms of gastro-oesophageal reflux disease
.
BMJ Journal
41
,
594
-
599
.
El-Serag
,
H. B.
,
Sweet
,
S.
,
Winchester
,
C. C.
and
Dent
,
J.
(
2014
).
Update on the epidemiology of gastro-oesophageal reflux disease: a systematic review
.
Gut
63
,
871
-
880
.
El-Serag
,
H. B.
,
Petersen
,
N. J.
,
Carter
,
J.
,
Graham
,
D. Y.
,
Richardson
,
P.
,
Genta
,
R. M.
and
Rabeneck
,
L.
(
2004
).
Gastroesophageal reflux among different racial groups in the United States
.
Gastroenterology
126
,
1692
-
1699
.
Ernst
,
L. M.
,
Ruchelli
,
E. D.
,
Carreon
,
C. K.
and
Huff
,
D. S.
(
2019
).
Color Atlas of Human Fetal and Neonatal Histology. Color Atlas of Human Fetal and Neonatal Histology
.
Cham
:
Springer International Publishing AG
.
Fleischer
,
D. E.
,
Overholt
,
B. F.
,
Sharma
,
V. K.
,
Reymunde
,
A.
,
Kimmey
,
M. B.
,
Chuttani
,
R.
,
Chang
,
K. J.
,
Lightdale
,
C. J.
,
Santiago
,
N.
,
Pleskow
,
D. K.
et al.
(
2008
).
Endoscopic ablation of Barrett's esophagus: a multicenter study with 2.5-year follow-up
.
Gastrointest. Endosc.
68
,
867
-
876
.
Hall
,
G. L.
(
2020
).
Patient-Centered Clinical Care for African Americans. Patient-Centered Clinical Care for African Americans
.
Springer International Publishing
.
Houben
,
C. H.
and
Curry
,
J. I.
(
2008
).
Current status of prenatal diagnosis, operative management and outcome of esophageal atresia/tracheo-esophageal fistula
.
Prenat. Diagn.
28
,
667
-
675
.
Kasagi
,
Y.
,
Chandramouleeswaran
,
P. M.
,
Whelan
,
K. A.
,
Tanaka
,
K.
,
Giroux
,
V.
,
Sharma
,
M.
,
Wang
,
J.
,
Benitez
,
A. J.
,
DeMarshall
,
M.
,
Tobias
,
J. W.
et al.
(
2018
).
The esophageal organoid system reveals functional interplay between notch and cytokines in reactive epithelial changes
.
Cmgh
5
,
333
-
352
.
Kim
,
J.
,
Bowlby
,
R.
,
Mungall
,
A. J.
,
Gordon Robertson
,
A.
,
Odze
,
R. D.
,
Cherniack
,
A. D.
,
Shih
,
J.
,
Pedamallu
,
C. S.
,
Cibulskis
,
C.
,
Dunford
,
A.
et al.
(
2017
).
Integrated genomic characterization of oesophageal carcinoma
.
Nature
541
,
169
-
175
.
Kleuskens
,
M. T. A.
,
Haasnoot
,
M. L.
,
Herpers
,
B. M.
,
van Ampting
,
M. T. J.
,
Bredenoord
,
A. J.
,
Garssen
,
J.
,
Redegeld
,
F. A.
and
van Esch
,
B. C. A. M.
(
2021
).
Butyrate and propionate restore interleukin 13-compromised esophageal epithelial barrier function
.
Allergy
77
,
1510
-
1521
.
Kuwahara
,
A.
,
Lewis
,
A. E.
,
Coombes
,
C.
,
Leung
,
F. S.
,
Percharde
,
M.
and
Bush
,
J. O.
(
2020
).
Delineating the early transcriptional specification of the mammalian trachea and esophagus
.
eLife
9
,
1
-
23
.
Liu
,
X.
,
Krawczyk
,
E.
,
Suprynowicz
,
F. A.
,
Palechor-Ceron
,
N.
,
Yuan
,
H.
,
Dakic
,
A.
,
Simic
,
V.
,
Zheng
,
Y.-L.
,
Sripadhan
,
P.
,
Chen
,
C.
et al.
(
2017
).
Conditional reprogramming and long-term expansion of normal and tumor cells from human biospecimens
.
Nat. Protoc.
12
,
439
-
451
.
Madissoon
,
E.
,
Wilbrey-Clark
,
A.
,
Miragaia
,
R. J.
,
Saeb-Parsy
,
K.
,
Mahbubani
,
K. T.
,
Georgakopoulos
,
N.
,
Harding
,
P.
,
Polanski
,
K.
,
Huang
,
N.
,
Nowicki-Osuch
,
K.
et al.
(
2019
).
ScRNA-Seq assessment of the human lung, spleen, and esophagus tissue stability after cold preservation
.
Genome Biol.
21
,
1
-
16
.
McInnes
,
L.
,
Healy
,
J.
and
Melville
,
J.
(
2018
).
UMAP: uniform manifold approximation and projection for dimension reduction
.
arXiv
.
Mou
,
H.
,
Vinarsky
,
V.
,
Tata
,
P. R.
,
Brazauskas
,
K.
,
Choi
,
S. H.
,
Crooke
,
A. K.
,
Zhang
,
B.
,
Solomon
,
G. M.
,
Turner
,
B.
,
Bihler
,
H.
et al.
(
2016
).
Dual SMAD signaling inhibition enables long-term expansion of diverse epithelial basal cells
.
Cell Stem Cell
19
,
217
-
231
.
Pavel
,
M.
,
Renna
,
M.
,
Park
,
S. J.
,
Menzies
,
F. M.
,
Ricketts
,
T.
,
Füllgrabe
,
J.
,
Ashkenazi
,
A.
,
Frake
,
R. A.
,
Lombarte
,
A. C.
,
Bento
,
C. F.
et al.
(
2018
).
Contact inhibition controls cell survival and proliferation via YAP/TAZ-autophagy axis
.
Nat. Commun.
9
,
2961
.
Pepejoy
,
A.
and
Fullerton
,
S.
(
2016
).
Genomics is failing on diversity
.
Nature
538
,
161
-
164
.
Piedrafita
,
G.
,
Kostiou
,
V.
,
Wabik
,
A.
,
Colom
,
B.
,
Fernandez-Antoran
,
D.
,
Herms
,
A.
,
Murai
,
K.
,
Hall
,
B. A.
and
Jones
,
P. H.
(
2020
).
A single-progenitor model as the unifying paradigm of epidermal and esophageal epithelial maintenance in mice
.
Nat. Commun.
11
,
1
-
15
.
Polański
,
K.
,
Young
,
M. D.
,
Miao
,
Z.
,
Meyer
,
K. B.
,
Teichmann
,
S. A.
and
Park
,
J. E.
(
2020
).
BBKNN: fast batch alignment of single cell transcriptomes
.
Bioinformatics
36
,
964
-
965
.
Rankin
,
S. A.
,
Han
,
L.
,
McCracken
,
K. W.
,
Kenny
,
A. P.
,
Anglin
,
C. T.
,
Grigg
,
E. A.
,
Crawford
,
C. M.
,
Wells
,
J. M.
,
Shannon
,
J. M.
and
Zorn
,
A. M.
(
2016
).
A retinoic acid-hedgehog cascade coordinates mesoderm-inducing signals and endoderm competence during lung specification
.
Cell Rep.
16
,
66
-
78
.
Rastogi
,
A.
,
Puli
,
S.
,
El-Serag
,
H. B.
,
Bansal
,
A.
,
Wani
,
S.
and
Sharma
,
P.
(
2008
).
Incidence of esophageal adenocarcinoma in patients with barrett's esophagus and high-grade dysplasia: a meta-analysis
.
Gastrointest. Endosc.
67
,
394
-
398
.
Rojas
,
A.
,
Araya
,
P.
,
Gonzalez
,
I.
and
Morales
,
E.
(
2020
).
Tumor microenvironments in organs
.
Adv. Exp. Med. Biol.
1226
,
23
-
35
.
Rosekrans
,
S. L.
,
Baan
,
B.
,
Muncan
,
V.
and
Van Den Brink
,
G. R.
(
2015
).
Esophageal development and epithelial homeostasis
.
Am. J. Physiol. Gastrointest. Liver Physiol.
309
,
G216
-
G228
.
Saraggi
,
D.
,
Fassan
,
M.
,
Bornschein
,
J.
,
Farinati
,
F.
,
Realdon
,
S.
,
Valeri
,
N.
and
Rugge
,
M.
(
2016
).
From barrett metaplasia to esophageal adenocarcinoma: the molecular background
.
Histol. Histopathol.
31
,
25
-
32
.
Sherley
,
J. L.
,
Stadler
,
P. B.
and
Scott Stadler
,
J.
(
1995
).
A quantitative method for the analysis of mammalian cell proliferation in culture in terms of dividing and non–dividing cells
.
Cell Prolif.
28
,
137
-
144
.
Spechler
,
S. J.
,
Jain
,
S. K.
,
Tendler
,
D. A.
and
Parker
,
R. A.
(
2002
).
Racial differences in the frequency of symptoms and complications of gastro-oesophageal reflux disease
.
Aliment. Pharmacol. Ther.
16
,
1795
-
1800
.
Spechler
,
S. J.
,
Sharma
,
P.
,
Souza
,
R. F.
,
Inadomi
,
J. M.
and
Shaheen
,
N. J.
(
2011
).
American gastroenterological association medical position statement on the management of Barrett's esophagus
.
Gastroenterology
140
,
1084
-
1091
.
Spence
,
J. R.
,
Mayhew
,
C. N.
,
Rankin
,
S. A.
,
Kuhar
,
M. F.
,
Vallance
,
J. E.
,
Tolle
,
K.
,
Hoskins
,
E. E.
,
Kalinichenko
,
V. V.
,
Wells
,
S. I.
,
Zorn
,
A. M.
et al.
(
2011
).
Directed differentiation of human pluripotent stem cells into intestinal tissue in vitro
.
Nature
470
,
105
-
110
.
Srinivasan
,
B.
,
Shuler
,
L.
and
Hickman
,
J. J.
(
2015
).
TEER Measurements in cells
.
J. Lab. Autom.
20
,
107
-
126
.
Suprynowicz
,
F. A.
,
Kamonjoh
,
C. M.
,
Krawczyk
,
E.
,
Agarwal
,
S.
,
Wellstein
,
A.
,
Agboke
,
F. A.
,
Choudhury
,
S.
,
Liu
,
X.
and
Schlegel
,
R.
(
2017
).
Conditional cell reprogramming involves noncanonical β-catenin activation and MTOR-mediated inactivation of Akt.S
.
PLoS ONE
12
,
1
-
23
.
Then
,
E. O.
,
Lopez
,
M.
,
Saleem
,
S.
,
Gayam
,
V.
,
Sunkara
,
T.
,
Culliford
,
A.
and
Gaduputi
,
V.
(
2020
).
Esophageal cancer: an updated surveillance epidemiology and end results database analysis
.
World J. Oncol.
11
,
55
-
64
.
Thrift
,
A. P.
and
El-Serag
,
H. B.
(
2016
).
Sex and racial disparity in incidence of esophageal adenocarcinoma: observations and explanations
.
Clin. Gastroenterol. Hepatol.
14
,
330
-
332
.
Trisno
,
S. L.
,
Philo
,
K. E. D.
,
McCracken
,
K. W.
,
Catá
,
E. M.
,
Ruiz-Torres
,
S.
,
Rankin
,
S. A.
,
Han
,
L.
,
Nasr
,
T.
,
Chaturvedi
,
P.
,
Rothenberg
,
M. E.
et al.
(
2018
).
Esophageal organoids from human pluripotent stem cells delineate Sox2 functions during esophageal specification
.
Cell Stem Cell
23
,
501
-
515.e7
.
Uhlen
,
M.
(
2005
).
A human protein atlas for normal and cancer tissues based on antibody proteomics
.
Mol. Cell. Proteomics
4
,
1920
-
1932
.
Uhlen
,
M.
,
Fagerberg
,
L.
,
Hallstrom
,
B. M.
,
Lindskog
,
C.
,
Oksvold
,
P.
,
Mardinoglu
,
A.
,
Sivertsson
,
A.
,
Kampf
,
C.
,
Sjöstedt
,
E.
,
Asplund
,
A.
et al.
(
2015
).
Tissue-based map of the human proteome
.
Science
347
,
1260419
.
Vaughan
,
F. L.
,
Kass
,
L. L.
and
Uzman
,
J. A.
(
1981
).
Requirement of hydrocortisone and insulin for extended proliferation and passage of rat keratinocytes
.
In Vitro
17
,
941
-
946
.
Weiler
,
T.
,
Mikhail
,
I.
,
Singal
,
A.
and
Sharma
,
H.
(
2014
).
Racial differences in the clinical presentation of pediatric eosinophilic esophagitis
.
J. Allergy Clin. Immunol. Pract.
2
,
320
-
325
.
Wolf
,
F. A.
,
Angerer
,
P.
and
Theis
,
F. J.
(
2017
).
SCANPY: large-scale single-cell gene expression data analysis
.
Genome Biol.
19
,
2926
-
2934
.
Yamamoto
,
Y.
,
Wang
,
X.
,
Bertrand
,
D.
,
Kern
,
F.
,
Zhang
,
T.
,
Duleba
,
M.
,
Srivastava
,
S.
,
Khor
,
C. C.
,
Hu
,
Y.
,
Wilson
,
L. H.
et al.
(
2016
).
Mutational spectrum of Barrett's stem cells suggests paths to initiation of a precancerous lesion
.
Nat. Commun.
7
,
10380
.
Zhang
,
Y.
,
Bailey
,
D.
,
Yang
,
P.
,
Kim
,
E.
and
Que
,
J.
(
2021
).
The development and stem cells of the esophagus
.
Development
148
,
dev193839
.
Zhang
,
Y.
,
Jiang
,
M.
,
Kim
,
E.
,
Lin
,
S.
,
Liu
,
K.
,
Lan
,
X.
and
Que
,
J.
(
2017
).
Development and stem cells of the esophagus
.
Semin. Cell Dev. Biol.
66
,
25
-
35
.
Zhang
,
Y.
,
Yang
,
Y.
,
Jiang
,
M.
,
Huang
,
S. X.
,
Zhang
,
W.
,
Al Alam
,
D.
,
Danopoulos
,
S.
,
Mori
,
M.
,
Chen
,
Y.-W.
,
Balasubramanian
,
R.
et al.
(
2018
).
3D modeling of esophageal development using human psc-derived basal progenitors reveals a critical role for notch signaling
.
Cell Stem Cell
23
,
516
-
529.e5
.

Competing interests

The authors declare no competing or financial interests.

Supplementary information