Epithelioids: Self-sustaining 3D epithelial cultures to study long-term processes

Studying long-term biological processes such as the colonization of aging epithelia by somatic mutant clones has been slowed by the lack of suitable culture systems. Here we describe epithelioids, a facile, cost-effective method of culturing multiple mouse and human epithelia. Esophageal epithelioids self-maintain without passaging for at least a year, recapitulating the 3D structure, cell dynamics, transcriptome, and genomic stability of the esophagus. Live imaging over 5 months showed epithelioids replicate in vivo cell dynamics. Epithelioids enable the study of cell competition and mutant selection in 3D epithelia, and how anti-cancer treatments modulate the competition between transformed and wild type cells. Epithelioids are a novel method with a wide range of applications in epithelial tissues, particularly the study of long term processes, that cannot be accessed using other culture models.


Introduction
The long-term culture of primary epithelial cells has been challenging. A wide range of tissues may be expanded using organoid cultures [1][2][3][4] . These have been used to model epithelial carcinogenesis and genetic disease to guide therapy [5][6][7][8][9] . However, organoids and similar models have a critical limitation. They continuously expand, and therefore do not attain the balance between proliferation and differentiation that maintains adult tissues in a steady state, and so require frequent passaging [10][11][12][13][14][15] . Classical organotypic cultures where epithelial cells are grown on permeable membrane inserts give excellent 3D epithelial histology but only last for a few weeks 16,17 . Older methods such as explant culture, where primary cells migrate out of tissue sample onto a plastic surface fail to replicate 3D tissue structure 18 . None of these systems retains both the architecture and the long-term self-maintaining capacity of adult epithelia.
These limitations have restricted the study of processes that require long-term, self-maintaining, epithelial cultures. In vivo, mutant clones expand over months to years, to multi millimeter sizes competing for the limited space available in the proliferative compartment of homeostatic epithelia [19][20][21][22][23][24] . Lineage tracing using transgenic mouse models can capture many of the features of mutant clonal dynamics in human epithelia but are slow and not suitable for screens to uncover the genes that regulate competitive fitness 19,[25][26][27][28][29][30] .
To overcome these limitations, we have developed 'epithelioids', suitable for studying processes that require long-term, centimeter scale epithelial cultures, by combining explant culture with organotypic protocols to generate 3D ex vivo epithelial reconstructions. This system avoids the stress of disaggregating tissue into a single cell suspension, allowing substantial in vitro tissue amplification, and uses a medium extensively tested for skin grafting in humans, at a lower cost than standard organoid protocols 3,15,31 . Moreover, we demonstrate that epithelioid cultures replicate tissue organization and gene expression and self-maintain for at least one year. This allows the study of long-term tissue and cellular processes in a 3D epithelial context.

Generation of epithelioid cultures from mouse and human epithelia
We began by culturing mouse esophageal epithelial explants on permeable membrane inserts in a standard keratinocyte medium supplemented with growth factors (cFAD, methods) (Fig.1a) 15 . Epithelial cells migrated out from the explant forming an expanding cellular 'halo' of keratinocytes within 48h, which grew at a constant velocity (0.52 mm 2 /h) until the cultures reached confluence (Extended Fig.1a-c

and Supplementary
Video 1). One week after plating, tissue explants can be removed. Cells from a single explant (of up to 2.5 mm 2 or 1/32th of the esophagus) covered a 4.5 cm 2 membrane in 15-18 days, a 182 fold increase in area.
Establishing cultures was highly efficient, with 95% of explants generating keratinocyte outgrowth (from multiple mice and researchers, see figure) (Fig.1b). The cultures comprised CDH1 + keratinocytes (Extended Fig.1e-g). Unlike primary keratinocytes plated on tissue culture plastic (Extended Fig.1h), epithelioids formed a 3D tissue-like structure with a basal layer of TRP63 + cells and suprabasal layers of KRT4 + differentiated keratinocytes ( Fig.1cd), recapitulating the architecture of esophageal epithelium. By using larger membranes, a single mouse esophagus could be amplified 332 x, yielding 264 cm 2 of cultured epithelium in 30 days (Extended Fig.1i-j).  To further amplify primary epithelial cultures, the standard approach is to use enzymatic digestion to prepare a single cell suspension which is then used to establish multiple new cultures. To avoid trypsinization-induced cell damage 32 , we instead opted to expand the cultures by repeating explant process ( Fig.1e-g). Confluent cultures were dissected into 16 pieces, each of which was placed on a permeable membrane insert. After 2-3 days cells colonized the new insert reaching confluence within 20 days (Methods, Fig.1e-f). The process was repeated over 4 consecutive passages and confluent stratified cultures with TRP63 + , ITGA6 + progenitors overlaid by KRT4 + differentiated cells continued to form (Fig.1h-i). Using this approach, a single mouse esophagus epithelium could potentially be amplified 12x10 6 fold in 100 days, obtaining up to 2.1x10 6 primary cultures (of 4.5 cm 2 each), corresponding to the area of two basketball courts (Fig.1h). At any stage, if required, cells can be trypsinized for downstream analysis, genetic modification and re-plating, cryopreservation or organoid generation (Fig.1h, Extended Data Figure 1k-m).

Fig. 1d) without contaminating fibroblasts or immune cells (Red and white cells in Extended
Long-term expansion of primary normal human esophagus cells has been challenging 33 . Using the same method developed for mouse, we were able to generate epithelioid cultures from human esophageal epithelium of transplant donors aged 36 to 76 years ( Fig.2a-b). Cultures maintained in cFAD expanded to confluence with high efficiency (Fig.2c). At confluence, the cultures had a basal layer of proliferating ITGA6 + keratinocytes with suprabasal layers of KRT4 + differentiated keratinocytes (Fig.2d-e). Thus, epithelioid cultures can amplify human esophageal samples from normal adult tissue and provide a robust platform for studying human esophageal biology. DAPI (blue), of an explant and the cells exiting from it to colonize the insert and form an epithelioid. (e) 3D image (e) and optical section with orthogonal views of the submucosa layer of the explant (f, scale bars=44 μm) and its newly formed epithelioid basal layer membrane (g, scale bars=41 μm). h, Primary Rosa26 mTmG mouse esophageal keratinocytes cultured in a plastic surface stained for TP63 (grey), KRT4 (green) and DAPI (blue  Schematic illustration of epithelioid generation from human esophagus. b, Age distribution of human donors expanded as epithelioids. Each dot represents a donor. c, Proportion of explants that form a cellular halo and contribute to epithelioid generation. Total number of explants plated per donor is indicated. n=480 explants from 2 donors. d-e, Rendered confocal z stack (d) and basal plane optical section with orthogonal views (e) of a typical human esophageal epithelioid stained for KRT4 (red), ITGA6 (white), KI67 (green) and DAPI (blue). Scale bars=38 μm for x-y and 15 μm for z. f, Illustration of mouse epithelioid generation from mouse oral mucosa, bladder urothelium and tracheal epithelium. g, Rendered confocal z stack (top) and basal layer optical section with orthogonal views (bottom) of a typical mouse oral mucosa epithelioid stained for WGA (white), TRP63 (green) and DAPI (blue). Scale bars=38 μm for x-y and 17 μm for z. h, Rendered confocal z stack (top) and basal plane section with orthogonal views (bottom) of a typical mouse bladder epithelioid stained for KRT20 (white), KRT5 (green), KRT14 (red) and DAPI (blue). Scale bars=28 μm for x-y and 11 μm for z. i, Rendered confocal z stack (top) and basal plane section with orthogonal views (bottom) of a typical mouse tracheal epithelioid stained for WGA (white) and DAPI (blue). Scale bars=38 μm. Source data in Supplementary Table 1.
We next applied the same protocol used for the esophagus to another mouse stratified squamous tissue (oral mucosa), a transitional epithelium (bladder urothelium) and a pseudostratified columnar epithelium (tracheal epithelium). In each case, the protocol generated confluent cultures that recapitulated the architecture of the original tissue ( Fig.2f-h). These results suggest that the epithelioid system can be extended to multiple types of epithelia.

Characterization of esophageal epithelioid cultures
We went on to characterize mouse esophageal epithelioid cultures in depth. Explants were plated on insert membranes and when a large cellular halo had formed around the explant (around 7 days after plating) they were removed. Once a confluent stratified culture was obtained (around 15-18 days after plating), we maintained the cultures in reduced growth factor media (mFAD, methods), refreshed 2-3 times a week (Fig.   3a). Under these conditions, epithelioids maintained the morphology of the in vivo tissue, with a basal layer of TRP63 + epithelial progenitor cells. Basal cells showed the typical polarization seen in vivo, expressing the hemidesmosome protein ITGA6 exclusively on the part of the membrane in contact with the basement membrane. Proliferation was restricted to the basal layer while suprabasal layers expressed KRT4 cells recapitulating in vivo features (Fig.3b-e). The proportion of basal cells in S-phase was 11%, comparable to that of the adult mouse esophagus (Extended Fig.2a-c). Cell tracing of proliferating cells and confocal live imaging showed that in epithelioids, similarly to the in vivo tissue, basal cells proliferate, and some exit the basal layer migrate into the suprabasal layers and are eventually shed (Extended Fig.2d-  RNA-sequencing analysis showed that gene expression in epithelioids highly correlated with the gene expression of epithelial cells directly isolated from mouse esophagus (Fig.3j). The expression of epithelial transcripts (Epcam, E-cadherin, Zo-1), (Extended Fig.2g), mRNAs characteristic of basal, differentiating and cycling cells, and of multiple esophageal cell fate regulators was not substantially altered by the culture conditions ( Fig.3k) 29,[34][35][36][37][38][39][40] , arguing that in vivo cell fate regulation may be preserved in epithelioids. Analysis of gene signatures of keratinocyte differentiation states 34 showed a correlation between in vivo and in vitro with the largest differences confined to late differentiation markers, reflecting decreased terminal differentiation in epithelioids (Extended Fig.2h). Air-liquid interface culture 16,41 , which enhances terminal differentiation, may be used to correct this discrepancy (Extended Fig.2i-k). Fig. 2: Characterization of mouse esophageal epithelioids. a-c, Proliferating cells that incorporated EdU after a 1h pulse in mouse esophageal epithelium and mouse esophageal epithelioids treated in minimal medium (mFAD) or complete medium (cFAD). a, protocol. b rendered confocal z stack (left) and basal plane section with orthogonal views (right) of a typical mouse esophageal epithelioid in mFAD stained for KRT4 (green), EdU(red) and DAPI (blue).

Extended Data
Scale bars=20 μm for x-y and 14 μm for z plane. c, Percentage of EdU + basal cells in the three conditions, each dot corresponds to an animal or an epithelioid, orange bars represent the mean values. d-f, In vitro proliferative cell tracing 48h after an EdU pulse. d, Protocol, EdU labels S phase cells, after 48 hours EdU + cells that have differentiated and left the basal layer are counted, revealing the rate of differentiation and stratification e, percentage of EdU + suprabasal cells versus total EdU + cells (mean±SD). f, rendered confocal z stack (left) and basal plane section with orthogonal views (right) of typical mouse esophageal epithelioid stained for KRT4 (green), EdU (red) and DAPI (blue). Scale bars=21 μm for x-y and 18 μm for z plane. g-h RNA sequencing comparing gene expression from mouse esophageal epithelium (in vivo) and esophageal epithelioids (in vitro). n=4 animals and 4 epithelioids from 4 different animals. g Epithelial genes and correlation between in vivo and in vitro samples in transcripts characterizing the different stages of differentiation 42 .
The orange line shows the linear regression between samples with the Pearson's coefficient and p-value of the correlation. i-k, Rendered confocal z stack (i), basal plane section with orthogonal views (j), suprabasal plane section with orthogonal views (k) of a typical mouse esophagus epithelioid exposed to air-liquid interface for 15 days, incubated for 1h with EdU and stained for KRT4 (green), EdU (red) and DAPI (blue). Scale bars=22 μm for x-y and 22 μm for z plane.
Source data in Supplementary Table 1.

Long-term epithelioids maintain tissue turnover and remain genomically stable
To model in vivo esophageal tissue, epithelioid cultures should be able to self-maintain in the long-term without passaging. We found epithelioids kept in mFAD refreshed twice a week remained in a steady-state for one year, with approximately constant levels of cell density and cell proliferation (

Long-term cell dynamics in epithelioids
We next analyzed long-term cell behavior in epithelioids by lineage tracing. Epithelioids were generated from R26-confetti mice, multicolor heritable cell labelling induced with adenoviral Cre recombinase, and the labelled cells placed in epithelioid culture for 6 months without passaging (methods). The cultures were imaged weekly using an Incucyte live imaging system (Essen Bioscience) ( Fig.4g-h). Initially, the culture is formed by a mixture of differentially colored individual cells. However, after 1 month, single-colored areas (SCA) started appearing, which are generated from a single cell or neighboring cells with the same color. We followed the behavior of 351 SCA from 9 cultures coming from 6 different animals starting 5 weeks after plating. SCA of sufficient size to be resolved at 5 weeks were tracked. We observed different patterns of evolution of SCA area. Most SCA became smaller (80 %), others grew and then shrunk (12 %), a minority remained constant in area (1 %) and finally some SCA grew progressively (7 %) (Extended Fig.3d-f).
Furthermore, from 8 weeks, the number of SCA declined, but the average size of the remaining SCA increased, so that the total labelled area remained approximately constant ( Fig.4i-k). These features are hallmarks of neutral drift, observed in clones labelled with a neutral reporter in squamous epithelia in vivo 11,44 . Two simple quantitative models of cell behavior in esophageal epithelium gave a good fit to the data (Methods, Fig.4i-k) 45 . As the behavior of SCA recapitulates that of neutral clones in vivo we conclude progenitor dynamics in epithelioids is similar to those in mouse esophagus in vivo 11,45 .
As animals age, they accumulate somatic mutations that may result in clonal expansions if they affect genes that regulate progenitor cell fate 46 . This process occurs at a low rate in the esophagus of ageing wild type mice, where mutations in genes such as Notch1 are seen in expanded clones in mice of 1 year of age 19,25 . To determine if somatic mutations impact the growth of SCA, 46 samples from 38 surviving SCA at the 9 month time-point were isolated by laser-capture microdissection. DNA was extracted and targeted sequencing performed for 192 genes implicated in driving clonal expansions in squamous epithelia and/or recurrently mutated in squamous cancer 19 . Median coverage was 106 fold. The estimated mutational burden was similar to that in age matched mouse oesophagus 19 , arguing the mutation rate is not substantially increased in epithelioid culture (Fig.4l). The low variant allele frequency (VAF) of most mutations, (71 % mutations had VAF<0.1, Fig.4m), indicates that these were unlikely to have altered SCA dynamics, as to do so a mutation must have a VAF close to 0.5 (at which level most cells in the SCA carry the mutation assuming they are diploid). This was the case for 26 % of SCA (Fig.4n), with 80 % of these sharing mutations with other SCA, suggesting that most mutations with large VAF were already present before labelling. Interestingly, the mutation shared by more SCA was a Notch1 frameshift mutation. This is consistent with the development of spontaneous Notch1 mutations that drive clonal expansions in ageing mice 19,25 . Therefore, most of SCA behavior can be explained by neutral drift and are not caused by the acquisition of a driver mutations in vitro.

Epithelioids as a tool to study neutral and non-neutral cell competition in stratified epithelia
The properties of epithelioids led us to speculate that they may be suitable for studying slow processes such as clonal competition in a tissue-like environment. We first investigated neutral competition between two populations of equal fitness. We established esophageal epithelioids from conditional R26-EYFP mice in which cells and their descendants express Enhanced Yellow Fluorescent Protein (EYFP) after genetic recombination by Cre recombinase 11 . Cells were infected with adenovirus encoding Cre achieving a 90±1 % recombination rate (Extended Fig.4a). RNAseq showed that the only transcript significantly altered by recombination was Rosa26 mRNA (5.17 fold change, adjusted p-value 1.52x10 -72 ) (Extended Fig.4b). Thus Cre-mediated loxP excision can be performed at high efficiency without altering overall gene expression. We then generated mixed epithelioid cultures with EYFP + recombined and unrecombined cells from the same esophagus and measured the proportion of each subpopulation over time (Extended Fig. 4c and Fig.5a-b).
The proportion of EYFP + cells remained constant over 2 months (Fig.5c). This recapitulates the neutral behavior of the same reporter allele in the esophagus in vivo 11 .
Next, we studied a non-neutral competition. We selected a conditional dominant negative mutant of Maml1 (DNM) that has strong advantage over wild type cells in the esophageal epithelium in vivo 47 . Epithelioids were generated from R26-DNM mice and infected with either null or Cre-encoding adenovirus to generate wildtype or DNM expressing keratinocytes from the same mice. These cells were mixed with EYFP expressing cells as described above and the proportions of cells were analyzed after 4 weeks (Fig.5d). DNM expressing cells outcompeted EYFP + cells, showing a significant increase in relative fitness relative to the uninduced cells from the DNM mice (Fig.5e-f). Confocal microscopy of day 15 cultures showed that ratio of suprabasal: basal cells was significantly lower for DNM expressing than non-expressing cells (Fig.5g). This is consistent with the behavior of DNM expressing clones in vivo, which gain a competitive advantage by progenitors generating fewer differentiating than progenitor daughters per average cell division 47 . We conclude that epithelioids are suitable platform for studying mutant cell competition.

Effect of chemotherapy and radiotherapy on the competition between transformed and nontransformed cells
Next, we turned to a more challenging application. Many cytotoxic cancer treatments cause substantial normal tissue damage alongside tumor cell killing. We hypothesized that longevity of epithelioids may allow  Fig.4d-e). The transformed cells were mixed with wild type cells from a Rosa26 nTnG mouse which express the Tomato protein targeted to the cell nucleus in epithelioid cultures. The transformed cells outcompeted the wild type cells (Fig.5i and m).
Next, we investigated the impact of anti-cancer treatments on wild type and transformed cell competition by exposing the cultures to intermittent doses of ionizing radiation (IR), epirubicin or 5-fluorouracyl (5FU) (Fig.5j-p and Extended Fig.4e), which are used to treat esophageal cancer 48 . All three treatments altered the competition between wild type and transformed cells over 10 weeks. 2 Gy IR exposure halted the expansion of transformed cells and induced aberrant large nuclei specifically in the transformed population ( Fig.5j and n and Extended Fig.4e). Conversely, epirubicin showed significant toxicity in both wild type and transformed cells, though the effect on cell fitness was more pronounced in transformed cells which were progressively depleted from the culture (Fig.5k and o and Extended Fig.4e). 5FU treatment initially inhibited the expansion of transformed cells, however, later transformed cells recovered and overtook wild type cells ( Fig.5l and p and Extended Fig.4e), consistent with the development of 5FU resistance published using other in vitro models 49 .
These results show the potential of epithelioids to study the differential effects of therapy on transformed and wild type cells competing in long-term continuous co-culture.

Discussion
The epithelioid system emerges as a facile and versatile method of generating 3D sheets of cultured epithelial tissue with multiple applications. This technique allows the production and long-term maintenance of large amounts of primary 3D epithelium from a small initial sample. It may be applied to human epithelia, allowing the amplification of small, precious patient tissue biopsies for the study genetic or other disorders in an organotypic context. Murine epithelioids can be generated from genetically manipulated mice, enabling a wide range of transgenic tools and sensors to be leveraged. Epithelioids are also amenable for live imaging, facilitated by its growth on a flat surface rather than as spheroids in suspension. Genetically manipulated cells may be followed by lineage tracing, paralleling in vivo studies but with substantial savings in time and cost. In combination of these properties enhance the analysis of multiple processes such as progenitor cell dynamics, epithelial differentiation, or cell-cell interactions in an organotypic context.
A particular advantage of epithelioids over other advanced cell culture methods is their ability to self-sustain for weeks to months without passaging, allowing processes that evolve slowly within tissues to be studied.

These include competition between mutant or transformed cells versus wild type cells, permitting
experimental validation of the selection of somatic mutations, and to define the effect of drug treatments on such competition, as shown above.
More broadly, potential applications of this system extend to studying how mutagenesis 19,21 , environmental exposures such as ionizing radiation 28 , physical damage such as wounding, ageing, long-term drug treatment, inflammation or metabolic alterations affect cellular states, and tissue function in a 3D epithelial context. We thank Esther Choolun, Tom Metcalf and Sanger RSF facility for technical support with animal research. We thank Claire Hardy, Calli Latimer, staff from the Cancer, Ageing, and Somatic Mutations programme support laboratory for technical support with sequencing and The Gurdon Institute core imaging facility for technical support with confocal microscopes.

Competing interests
The authors declare no competing interests.

Supplementary Information
Extended imaged by confocal live imaging. The z-projection of a full 3D stack is shown using a z-depth rainbow color scale (see Example of cells shedding from H2BGFP expressing esophageal epithelioids (see Methods) imaged by confocal live imaging. The z-projection of a full 3D stack is shown using a z-depth rainbow color scale (see Fig. 3

Animals
Multiple strains were used as a tissue source. C57/Bl6 mice were used as wild type, unless specified. In addition, we used the following genetically engineered mouse strains from the Jackson Laboratory:

Epithelioid generation and maintenance
Mice were euthanized, esophagus, trachea, bladder or oral mucosa were collected and the muscle layer removed with forceps, epithelium was cut in pieces (up to 32 for a mouse esophagus) and placed on top of a transparent ThinCert™ insert (Greiner Bio-One, 657641) with the epithelium facing upward and the submucosa facing the membrane. Where

Immunofluorescence
For whole-mount staining, the mouse esophagus was opened longitudinally, the muscle layer removed and the epithelium incubated for 3h in 5 mM EDTA-PBS at 37°C. The epithelium was peeled from submucosa and fixed in 4% paraformaldehyde in PBS for 30 min. For epithelioid staining, inserts were washed with PBS and fixed in 4% paraformaldehyde in PBS for 30 min. Then, tissue whole-mounts or membrane inserts were blocked for 1 h in blocking buffer (0.5% bovine serum albumin, 0.25% fish skin gelatin, 1% Triton X-100 and 10% donkey serum) in PHEM buffer (60 mM PIPES, 25 mM HEPES, 10 mM EGTA, and 4 mM MgSO4·7H2O). All reagents were purchased from Sigma Aldrich.
Tissues were incubated with primary antibodies (Supplementary table 1) overnight using blocking buffer, followed by 4 washes with 0.2 % Tween-20 in PHEM buffer of a minimum 15 min. When indicated EdU incorporation was detected with Click-iT chemistry kit following the manufacturer's instructions (Life technologies, 23227). Next, whole-mounts or inserts were incubated overnight with 1 μg/ml DAPI (Sigma Aldrich, D9542) and secondary antibodies (1:500) in blocking buffer. When indicated Alexa fluor 647-wheat germ agglutinin (WGA, Invitrogen W32466) was added 1:200 and Alexa fluor 647 anti-human/mouse CD49f (Biolegend, 313610) was added 1:250. Afterwards, samples were washed 4x15 min with 0.2% Tween-20 in PHEM buffer and mounted in Vectashield (Vector Laboratories, H-1000). Imaging was performed using a SP8 Leica confocal microscope with a 40 x objective with 1x digital zoom, optimal pinhole and line average, bidirectional scan, speed 400-600 hz, resolution 1024x1024. 3D stacks were generated including all the cell layers of the culture and where indicated, the basal layer plane was selected. Rendered confocal z-stacks were generated using Volocity 6.3 (Perkin Elmer) and Imaris 4.3 (Bitplane). Orthogonal views and individual planes were generated using Volocity 6.3 (Perkin Elmer) or Fiji ImageJ 58 .

Live imaging
IncuCyte live-cell imaging system (Essen Bioscience) was used for whole-well imaging once a day (mTmG growth curve experiments) or once a week (confetti lineage tracing experiments) using its 4 x objective. Images were analyzed using the Fiji Image J software 58 .
To evaluate cell division and cell differentiation, cells from Rosa26 M2rtTA /TetO-H2BGFP mice 59 were cultured in 6-well inserts. Confluent epithelioids were treated with doxycycline for 5 days to induce H2BGFP expression. Inserts were placed into custom-made holders to adapt them to a Leica SP8 confocal microscope stage. Cells were imaged using a HC PL FLUOTAR 40x/0.6 dry objective taking images of 512x512 resolution at 700 hz using a 1.28 x zoom, 0.5 AU pinhole, and a line average of 2. 32-plane z-stacks were obtained every 25 min for up to 40 h. Timelapses were analyzed using Fiji ImageJ software.

Organoid generation
To generate organoids, wild-type or C57BL/6 mouse esophageal epithelioids were washed in PBS and incubated with 0.05 % Trypsin-EDTA for 20 min at 37°C 5 % CO2. Cells were pelleted for 5 min at 350 g.
Trypsinized cells were then re-suspended in 7.5 mg/ml basement membrane matrix (Cultrex BME RGF type 2 (BME-2), Sigma Aldrich) supplemented with complete media and plated as 15 μl droplets in a 6-well plate. Once BME-2 was excised within 60 minutes of circulatory arrest and preserved in PBS until processing. Esophageal epithelium was peeled from the underlying muscle using forceps and most of the submucosa layer scraped using a scalpel. Then, the sample was cut in pieces (explants) which were placed on membrane inserts and cultured as described above for mouse esophageal epithelioid cultures. Immunostaining was performed as described above.

RNA sequencing
RNA was obtained from epithelioids 1 week after medium change to mFAD or from mouse esophageal epithelium  Table 1) were selected from scRNAseq data 42 .

EdU proliferation and lineage tracing
For in vivo proliferation analysis, 10 μg of EdU in PBS were administered by intraperitoneal injection 1 h before culling.
For in vitro proliferation analysis, epithelioids were incubated with 10 µM EdU for 1 h. For EdU in vitro lineage tracing, cells were incubated with EdU for 1 h, washed and kept in mFAD for an additional 48h. Esophageal epithelium wholemounts and cultures were fixed and stained with EdU-Click-iT kit and immunofluorescence as previously explained. EdU-positive basal cells were quantified from a minimum of 10 z-stack images using Fiji ImageJ.

Copy number analysis
DNA extraction was performed using the QIAMP DNA microkit (Qiagen, 56304) following the manufacturer's instructions. DNA from the ears of the same mice was extracted with the same method and used as germline controls.
Whole Genome Sequencing at low coverage was performed on either HiSeq 4000 machine (Illumina) to generate 150-bp paired-end reads. A modified version of QDNAseq (https://github.com/ccagc/QDNAseq/) was used to call changes in total copy number from the low coverage whole genome sequencing data 63 . QDNAseq was modified to include the correction of the coverage profile of the sample of interest by that of a matched control. Briefly, the procedure to call gains and losses is as follows: First sequencing reads were counted per 100 kb bins for both the sample of interest and the matched control. The bin-counts were then combined into coverage log ratio values to obtain what is commonly referred to as "logR". The calculation of logr is implemented similarly to how the Battenberg copy number caller calculates these values 64 : first the bin-counts from the sample of interest were divided by the control bin-counts to obtain the coverage ratio; the coverage ratio was then divided by the mean coverage ratio and finally the log2 was taken to obtain logr. The standard QDNAseq pipeline is then continued with first a correction of the logr for GC content correlated wave artefacts, segmentation and finally calling of gains and losses.
A post-hoc filtering step was subsequently applied to obtain robust copy number calls. We noticed that several regions were commonly called as altered due to coverage local inconsistencies in the matched controls. To identify the effect of these inconsistencies we applied the copy number pipeline in a run where each control was matched against all controls from the whole genome sequencing data described in 27 , expecting no alterations to be called. The run revealed common regions of false positive alterations on a number of chromosomes. Regions that were called in 3 or more different control-vs-control runs were subsequently masked from any analysis, i.e. copy number calls in these regions were not accepted.
Finally, after applying the masking we further filtered calls requiring a gain or loss to span at least 30Mb in size and that the alteration must constitute a gain or loss in at least 20 % of cells. To obtain an estimate of the percentage of sequenced cells that contained an alteration we applied the procedure that copy number caller ASCAT uses to find a tumor purity value 65 . ASCAT uses a grid search step where a range of purity and ploidy combinations are considered and ultimately a combination is picked by optimizing the amount of the genome that can be fit with an integer copy number value. We used this approach to estimate purity values only by fixing the ploidy at 2 and optimizing across a range of possible purity values.

SCA lineage tracing
Esophageal epithelioids from Rosa26 confetti/confetti animals were induced in vitro using adenovirus-Cre as specified above.
A week after induction, medium was changed to mFAD and whole wells were imaged in an Incucyte system as specified before. Images obtained were analyzed using Fiji ImageJ.

Quantitative SCA analysis and theoretical modelling
A least-squares minimization procedure was used to simultaneously fit the average area of SCA and the number of labelled SCA over time according to a single, equipotent progenitor model that describes proliferating cell behavior in esophageal epithelium in vivo 11,45 . The growth of the average SCA area was fitted to the theoretical linear expectation, i.e. a model of type ! + , while the decline in the total fraction of labelled SCA was described by an hyperbolic function of type ! /(1 + ′ ), with constrained parameters ( " = / ! , according to theory) 45 . Optimum parameter values . = / ! 0, ! 2 , ′ 3 4 were obtained by averaging goodness-of-fit values, measured as the sum of the squared residuals ( #,%&' − #,(%)*+ ) relative to the standard deviation of the observable ( ,#,%&' ), in both datasets. A zero-parameter fit followed for the total labelled area, modeled as constant ! ! given by the product of the average SCA size and the total number of surviving SCA at different time points, which is consistent with homeostasis. The first two time points were ignored in the fits to avoid initial stabilization-related effects.
For simulations of clonal dynamics, a 2D lattice implementation of a stochastic Moran process was adopted, where (clonogenic) progenitor cells were set to compete neutrally in a 200x200 squared (k=8 neighbors, default) or hexagonal (k=6 neighbors) grid with periodic boundary conditions 19,66 . A replacement rate Λ = 2 ′ was selected to meet the inferred SP-model kinetic conditions, being a scaling factor (no. cells/grid unit) used for tractable clone simulations, a parameter that was later regressed out before readout. Simulation results are shown overlaid on experimental data, with shaded areas reflecting 95 % plausible intervals given by limited sample sizes equivalent to those in the experimental data (at least 200 permutation-built subsets).
The code developed for the quantitative clonal analysis has been made publicly available and can be found at https://github.com/gp10/ClonalDeriv3D .

Single-colored area cutting and DNA sequencing
Single-colored areas were dissected using an LMD7 microscope (Leica Microsystems) and collected in separate tubes.
Samples were digested and DNA extracted using the QIAMP DNA microkit (Qiagen, 56304) following the manufacturer's instructions. DNA from the ears of the same mice was extracted with the same method and used as germline controls.
DNA sequencing was performed using a custom bait capture of 192 frequently mutated genes in cancer as in 19 , briefly samples were multiplexed and then sequenced using an Illumina HiSeq 2500 and paired-end 75-base pair (bp) reads.

Generation of p53* mutant transformed cells (p53*-TC)
Ahcre ERT -Trp53 flR245W-GFP/wt mice were induced to express the mutant p53 R245W allele and GFP reporter protein, by intraperitoneal injection of 80 mg/kg β-naphthoflavone (MP Biomedicals , 156738) and 0.25 mg Tamoxifen (Sigma Aldrich N3633) as previously described 28,30 . Once month later, mice were orally treated with the carcinogen diethylnitrosamine (DEN) (Sigma, catalog no. N0756) in sweetened drinking water (40 mg per 1,000 ml) for 24 h on 3 days a week (Monday, Wednesday and Friday), for two weeks to induce the formation of early lesions in the esophageal epithelium, followed by exposure to 10 doses of 2Gy of ionizing radiation using a whole-body caesium source irradiator. Mice were sacrificed and tumors were collected and cultured. After several rounds of expansion, cells were assessed for GFP expression (100% of the culture).

Cell competition assays
The indicated cell populations were trypsinized and mixed, 1:1. After one week in cFAD, when the cultures are fully confluent, medium is changed to mFAD and starting time point was collected. At the time points specified, cells were collected and analyzed by flow cytometry or fixed for microscopy as stated before. Where indicated cells were treated with 2 hour pulses of epirubicin 1µM, 5-fluorouracyl 5µM twice a week or irradiated with 2Gy (once a week using an Xstrahl CIX2 or RPS CellRad RSM-009 irradiators). Cell fitness over YFP + or nTnG cells was measured by quantifying fold increase of YFP-or p53*-TC cells respectively at the specified time point versus its proportion at the initial time point.

DATA AVAILABILITY
The sequencing data sets in this study are publicly available at the European Nucleotide archive (ENA) Accession numbers for RNAseq data on https://www.ebi.ac.uk/ena are as follows: