Transcriptional heterogeneity of stemness phenotypes in the ovarian epithelium

The ovarian surface epithelium (OSE) is a monolayer of epithelial cells surrounding the ovary that ruptures during each ovulation to allow release of the oocyte. This wound is quickly repaired, but mechanisms promoting repair are poorly understood. The contribution of tissue-resident stem cells in the homeostasis of several epithelial tissues is widely accepted, but their involvement in OSE is unclear. We show that traits associated with stem cells can be increased following exposure to the cytokine TGFB1, overexpression of the transcription factor Snai1, or deletion of Brca1. We find that stemness is often linked to mesenchymal-associated gene expression and higher activation of ERK signalling, but is not consistently dependent on their activation. Expression profiles of these populations are extremely context specific, suggesting that stemness may not be associated with a single, distinct population, but rather is a heterogeneous cell state that may emerge from diverse environmental cues. These findings support that the OSE may not require distinct stem cells for long-term maintenance, and may instead achieve this through transient dedifferentiation into a stem-like state. Using spheroids as a model, the authors report the molecular signatures of ovarian surface epithelial (OSE) cells exhibiting stemness phenotype under various conditions in vitro, and found that there is a large degree of heterogeneity in the transcriptional profiles of stem cells induced under different conditions. They suggest that maintenance of the OSE may not require a single stem cell population, but heterogeneous stem cells that can be induced transiently under diverse environmental cues.

I t is thought that stem cell populations are responsible for longterm maintenance of many adult tissues. The characterization of stem cells associated with epithelial tissue maintenance has been an active field of research for the past few decades. While several distinct stem cell populations have been functionally described, such as an LGR5 + population at the base of intestinal crypts 1 , it is unclear if all epithelial tissues are maintained by such defined populations. For example, it has been shown that following stem cell depletion, differentiated airway epithelial cells can dedifferentiate and become functional multipotent stem cells 2 . It is also unclear if stem cells are necessarily required to maintain epithelial tissues comprising a single cell type, as some baseline capacity for proliferation could maintain the entire tissue. In the mesothelium, for example, there have been reports of putative stem/progenitor cells for over two decades, but a welldefined stem cell population has yet to be identified 3 .
The ovarian surface epithelium (OSE) is a promising tissue for studying stemness dynamics in tissue maintenance. It is a monolayer of cells surrounding the ovary and, during each ovulation, this tissue is ruptured to facilitate release of an oocyte. Afterwards, the OSE layer is rapidly repaired [4][5][6][7] . Post-ovulatory wound repair is a poorly understood process despite ovulation being the primary non-hereditary risk factor for ovarian cancer.
Several putative OSE stem cell populations have been described, each defined by different cell surface markers, including ALDH1A1, LGR5, LY6A (Sca-1), and more [8][9][10][11] . However, relationships between the populations described in these studies are still unclear, with little known about co-expression or mutual exclusivity of markers. Further, it is unclear if these populations are static or can emerge from differentiated OSE in response to ovarian dynamics. We have previously shown that, like the mammary epithelium 12 , induction of an epithelial-tomesenchymal transition (EMT) can transiently promote features of stem cells (stemness) in differentiated OSE 10 . This is particularly relevant in the context of ovulation, as the EMT is thought to be an important component of wound repair and the EMT-promoting cytokine TGFB1 is present in follicular fluid, bathing adjacent OSE at ovulation 10,13 . It is also secreted by macrophages at the ovulatory wound and granulosa cells during follicular development 14,15 .
Here, we further demonstrate that features of stemness can be altered in OSE cells. Profiling gene expression of different populations with enhanced stemness, we demonstrate that some features are relatively common, including EMT-associated expression patterns and enhanced activity of ERK and NFkB signaling, but global expression profiles are widely variable and stemness is not exclusively dependent on these common features. Together, this work supports that OSE tissue maintenance may not require a distinct stem cell population, but can emerge in response to their environment.

Results
CD44 is a marker of EMT-associated stemness in OSE cells. We have previously demonstrated that mouse OSE (mOSE) cells undergo an EMT through canonical SMAD signaling when exposed to TGFB1 and acquire stem cell characteristics 10,16 . To confirm these findings, we first assessed the ability of TGFB1treated mOSE cells to form self-renewing spheroids in suspension culture. Treated cells formed over twice as many primary spheroids and, when dissociated and cultured, were more efficient at successfully generating secondary spheroids, confirming their capacity for self-renewal (Fig. 1a). Morphologically, mOSE spheres were large and compact, with no difference in their size regardless of whether they had been treated with TGFB1 (Fig. 1b). This suggests that the phenotype is not simply the result of increased proliferation rates, but rather reflects an increased ability for clonogenic growth. Consistent with this, we have previously shown that TGFB1 reduces the growth rate of mOSE 10 .
To validate this enhanced stemness in human cells, we also performed these experiments on primary cultures of human OSE (hOSE) cells. Since hOSE cultures have a low proliferation rate in vitro, we used methylcellulose-based suspension culture to immobilize the cells and minimize the impact of aggregation in our quantifications. While these conditions, along with the slower proliferation, resulted in smaller spheroids, TGFB1-treated hOSE cells formed 3 times as many spheroids as untreated cells (Fig. 1c).
To determine if this enhanced stemness is associated with the expression of previously reported markers of OSE stem cells, we measured their expression throughout 7 days of TGFB1 treatment in mOSE cells. Aldh1a1, Lgr5, and Nanog did not increase with TGFB1 treatment, and in some cases decreased over time ( Supplementary Fig. 1). While this does not preclude the possibility of these genes being valid markers of stem cell populations in vivo, these results suggest that their regulation is independent from TGFB1-associated stemness.
We next assessed the expression of a larger panel of markers from a commercial "Stem Cell Marker" qPCR array in order to identify putative markers that are associated with this stemness (Fig. 1d). This identified several highly upregulated markers following 7 days of TGFB1 treatment, including Ncam1 (14-fold), Cd44 (13-fold), and Ascl2 (6-fold) (Fig. 1d). CD44 has long been associated with stemness in mammary epithelial cells 17 and more recently in the oviductal epithelium 18 . We first validated CD44 RNA and protein levels throughout TGFB1 treatment and found that it increases after four days of treatment ( Fig. 1e; Supplementary Fig. 2). We then stained both control and TGFB1-treated spheroids for CD44 and Ki67 to assess the distribution of proliferation and this putative stemness phenotype within spheroids ( Fig. 1f; Supplementary Fig. 3). CD44 was detectable in many control spheroids but was expressed only in rare cells throughout the population. Cells comprising TGFB1-treated spheroids, however, ubiquitously expressed CD44. Proliferative cells were diffuse throughout the spheroids with no clear association with spatial location within the spheroid or relative to CD44 + cells. Both CD44 expression patterns are consistent with the possibility that CD44 expression enriches for stem-like cells capable of seeding clonogenic growth of spheroids. To test this, we sorted CD44 high cells from TGFB1-treated mOSE by fluorescence-activated cell sorting (FACS). When placed in suspension culture, CD44 high cells formed approximately 2.5 times as many spheroids compared to CD44 low cells (Fig. 1g).
Transcriptional profiling of mOSE stemness. We next sought to define a global profile of stemness, beyond a small number of markers. Spheroids themselves have been demonstrated to be enriched with stem/progenitor populations and challenge cells to exhibit stemness traits, including clonogenic growth and selfrenewal [19][20][21] . Since untreated mOSE cultures are capable of sphere formation, albeit at a lower frequency than TGFB1-treated mOSE cells, we reasoned that the transcriptional profile of these spheroids may represent an intrinsic stemness program, independent from exogenous factors. To compare this with TGFB1induced stemness, we performed RNA-seq on mOSE cells cultured as a monolayer or as spheroids, each with and without TGFB1 treatment.
Untreated mOSE cells cultured as spheroids exhibited striking differences from those cultured in a monolayer, with 4950 differentially expressed genes between the conditions (p < 0.05, absolute log fold change >0.5) ( Fig. 2a; Supplemental Data 1). Using an aggregate reference of GO terms, KEGG pathways, Reactome pathways, and MSigDB Hallmark gene sets from the Molecular Signatures Database (MSigDB) 22,23 , we used gene set enrichment analysis (GSEA) to identify biological features associated with these changes ( Fig. 2c; Supplemental Data 2). Spheroids were associated with decreased cell cycle, epithelial cell adhesion, and, interestingly, DNA repair. Along with these changes, spheroid culture activated expression of chemokine signaling and wound repair programs. We also note that CD44, which we had used as a selection marker for stemness in TGFB1treated mOSE, was also expressed over 4-fold higher in spheroids, whereas Aldh1a1, Lgr5, and Ly6a (Sca-1) were unchanged (Fig. 2b). We next used the PROGENy algorithm to infer changes in signaling pathway activity across these samples that may be contributing to these differences. Spheroids were associated with increased activity of many signaling pathways, with the largest increases in Hypoxia, NFkB, and MAPK signaling (Fig. 2d). While GSEA results suggest several EMT-related changes, TGFB1 and WNT signaling are interestingly reduced, suggesting that this EMT program may be activated through NFkB or ERK (Fig. 2d).
While TGFB1 signaling was decreased in untreated spheroids when compared to monolayer cultures, exogenous TGFB1 treatment of monolayers enhanced stemness, increasing the proportion of cells capable of forming self-renewing spheroids. While these results are seemingly contradictory, week-long exposure to exogenous TGFB1 may activate similar expression programs through secondary effects or signaling crosstalk 24 . We next assessed expression changes associated with TGFB1 This involved the activation of EMT-associated gene sets as expected, as well as a reduction in oxidative phosphorylation ( Fig. 3b; Supplemental Data 4). While TGFB1 signaling was the only pathway inferred to have significantly altered activity, the estimated activity of EGFR and MAPK was higher in TGFB1treated cells (p = 0.1 and 0.06, respectively) ( Fig. 3c). Consistent with this, the GO term "ERK1 and ERK2 signaling cascade" was significantly enriched in upregulated genes following TGFB1 treatment (Fig. 3b). This suggests that TGFB1 treatment initiates sequential or parallel signals similar to those present in spheroids. Consistent with this, untreated spheroids and TGFB1-treated mOSE monolayers have a significant overlap in expression changes relative to untreated mOSE cells cultured as a monolayer, sharing 270 upregulated genes and 293 downregulated genes (Fisher exact p = 3.0e-66 and 2.8e-117, respectively) (Fig. 3d). Conserved upregulated genes were strongly enriched for EMT-associated genes, NFkB signaling, and angiogenesis (Fig. 3e). Interestingly, very few gene sets were enriched in the conserved downregulated genes, with only interferon response and substrate adhesion genes being enriched (Fig. 3e). We then explored the gene expression patterns of TGFB1-treated spheres using a linear model with an interaction term to identify expression changes that were not simply the additive effects of TGFB1 and spheroid culture ( Supplementary Fig. 4a). We performed GSEA on genes ranked by the interaction coefficient of this model and found that TGFB1-treated spheres were associated with higher expression of genes associated with EMT, MAPK/ERK, and inflammatory pathways ( Supplementary Fig. 4b). They also had notable differences in the expression of metabolic genes, with an increase in oxidative phosphorylation and reduced cholesterol biosynthesis. Together, these findings suggest that various responses may contribute to OSE stemness, including increased MAPK/ERK signaling, activation of inflammatory response pathways such as NFkB and STAT, and EMT-associated gene expression.
Snail activation promotes a unique stemness program in mOSE cells. Signaling pathways are highly pleiotropic and it is unclear if TGFB1-enhanced stemness is activated from core EMT regulatory networks or alternative components regulated by TGFB1. The EMT transcription factor Snai1 (Snail) was upregulated in both TGFB1-treated mOSE cells and spheroids (Figs. 2a and 3a), and so to determine if EMT activation without exogenous cytokines could promote stemness, we derived mOSE cell lines with doxycycline-inducible Snail expression. Following Snail induction, cells had a higher sphere forming capacity than cells without doxycycline exposure, generating up to twice as many primary spheres and 3 times as many secondary spheres when passaged (Fig. 4a). Despite a higher sphere-forming efficiency, spheroid size was not different between control and Snailoverexpressing cells and the baseline proliferation rate of the cells was also unchanged (Supplementary Fig. 5a, b).
We next assessed the expression of the putative stem cell markers Cd44 and Sca-1, which are both increased with TGFB1 treatment, and found that Snail induction had no effect on their expression, suggesting that their validity as markers of stemness may be context specific. To determine if Snail induction activates similar expression patterns to TGFB1-treated mOSE and spheroids, including higher ERK and NFkB activity, we performed RNA-seq on these cells with and without doxycycline. Snail-induced changes were more modest than with TGFB1 treatment or in spheroid culture, with only 85 upregulated and 44 downregulated genes (p < 0.05, absolute log fold change >0.5; Fig. 4b; Supplemental Data 5). Interestingly, inferred pathway activity scores associated with EGFR, NFkB, MAPK, and TGFB1 were all unchanged following Snail induction (Fig. 4c). Further, no relevant gene sets associated with these pathways were enriched in the differentially expressed genes (Fig. 4d). The only gene sets associated with upregulated genes were largely related to cell morphology and extracellular matrix (ECM) remodeling ( Fig. 4d; Supplemental Data 6). Consistent with TGFB1-treated mOSE and spheroids, the MSigDB Hallmark "Interferon Alpha Response" was the only gene set enriched in the downregulated genes following Snail induction (Fig. 4d). We note that of the 85 upregulated genes following Snail induction, 13 are shared with those commonly regulated in TGFB1 treatment and spheroids (Fig. 4e). These genes largely represent components of the ECM, including Col18a1 and the metalloproteinases Mmp9 and Adamts4. As the conditions share no consistently activated downstream signal that could be induced by ECM changes, these findings suggest that expression programs associated with stemness phenotypes are heterogeneous. Given frequent enrichment of gene sets associated with a mesenchymal phenotype, stemness may consistently involve higher levels of these traits, which can emerge from variable expression patterns 25 .

BRCA1 loss promotes EMT-independent stemness in mOSE.
Spheroids were associated with higher expression of many genes that were not similarly induced by TGFB1 treatment. We noted that among spheroid culture-induced genes were several changes typically associated with ovarian cancer, including activation of the transcription factor Pax8, which is present in approximately 80% of ovarian tumors but not typically expressed in murine OSE 26 ; activation of Greb1, which promotes ovarian cancer growth 27 ; and loss of Brca1, which, along with Brca2, is mutated in approximately 22% of high-grade serous ovarian tumors. Interestingly, loss of BRCA1 has been associated with promoting dedifferentiation and activation of EMT expression patterns in mammary epithelial cells 28 .
As the association between BRCA1 loss and stemness in the OSE had not been assessed, we next derived a primary mOSE line from Brca1 tm1Brn mice harboring floxed Brca1 alleles. To determine if BRCA1 loss enhanced stemness in these cells, we  infected the cells with adenovirus containing either Cre recombinase (Ad-Cre) or GFP (Ad-GFP) as a control. Cre delivery, while not perfectly efficient, resulted in an approximately 60% reduction in BRCA1 levels across the population ( Supplementary Fig. 6). When placed in suspension culture, cells with reduced BRCA1 formed over 5 times as many primary spheres and 3 times as many secondary spheres than control mOSE cells, suggesting that BRCA1 loss also enhances stemness in mOSE cells (Fig. 5a). There was no difference in the size of the spheres and BRCA1 deletion modestly decreased the growth rate of cells in monolayer culture (p = 0.008, linear regression on cell counts from days 1-5) (Supplemental Fig. 7a, b).
To assess if BRCA1 loss enhances stemness phenotypes in vivo, we crossed the Brca1 tm1Brn mice with B6.129×1-Gt(ROSA) 26Sor tm1(EYFP)Cos /J mice to generate a Brca1 fl/fl YFP mouse line, allowing us to track Brca1-null cells following exposure to Ad-Cre. These mice were injected intrabursally (IB) with Ad-Cre or PBS, and injected intraperitoneally (IP) with bromodeoxyuridine (BrdU). Ovaries were collected after a 30-day chase period and assessed for retention of the BrdU label and activation of the YFP reporter. Ad-Cre injection IB in Brca1 fl/fl YFP mice showed successful activation of the YFP reporter, compared to the PBS injection (Fig. 5b). When combining the IB injections with an IP BrdU injection, Ad-Cre treatment increased the number of label-retaining OSE cells (Fig. 5c, d). Given the high frequency of ovulations of the mouse ovary over the 30-day chase period, this increased label retention following BRCA1 loss is consistent with BRCA1 deletion leading to expansion of quiescent stemlike cells in the OSE. Whether this is through expansion of an existing stem-like population or dedifferentiation of mOSE is unclear.
To determine if BRCA1 loss results in similar expression patterns to other conditions associated with stemness, we performed RNA-seq on mOSE cells isolated from these mice infected with Ad-Cre or Ad-GFP in vitro. BRCA1 loss resulted in a large shift in gene expression, with 1499 significantly upregulated genes and 1881 downregulated (p < 0.05, absolute log fold change >0.5; Fig. 6a; Supplemental Data 7). In mammary epithelium, the induction of EMT through Brca1 deletion was presumed to be due to loss of BRCA1-mediated repression on the promoter of the EMT transcription factor Twist1. In contrast, we found that Twist1 was approximately 8-fold lower in Brca1-null mOSE cells (Fig. 6a). There were also no EMT-associated gene sets enriched in upregulated genes. Rather, upregulated genes were largely enriched for gene sets associated with cell membrane transporters and downregulated genes were associated with cell cycle, oxidative phosphorylation, and DNA repair ( Fig. 6b; Supplemental Data 8). Brca1 deletion did not result in other features of stemness we observed in previous conditions, including activation of ERK and NFkB, and repression of interferon alpha response genes (Fig. 6c). Brca1-null cells were associated with reduced PI3K signaling and higher levels of estrogen signaling (Fig. 6c). Estrogen has been linked to EMT and stemness in other cell types 29 , but this is presumed to be through crosstalk, activating growth factor signaling pathways, which we do not see in Brca1-null OSE cells.
Comparing the expression profiles of each condition associated with stemness phenotypes in this study, we find minimal overlap in the specific genes activated or repressed in each (Fig. 6d). Ranking genes by the number of conditions they are activated or repressed in, we found that Adamts4 and Pnmal2 are the only genes upregulated in all four conditions (Fig. 6e). Adamts4 has been linked to stemness in uveal melanoma through modulating crosstalk between the cells and their adjacent ECM 30 . While this may be relevant here, conserved downstream signals promoting stemness remain elusive. We note high frequency of EMTassociated changes, including activation of Snail, various collagens, and repression of cytokeratins (Fig. 6e). Notably, however, specific EMT transcription factors and putative OSE stemness markers (Lgr5, Aldh1a1, Ly6a, Nanog, and Cd44) are only activated in 1-2 conditions, and are even repressed in some conditions (Fig. 6e).

Discussion
Several studies have reported putative stem cell populations in the OSE, but the relationships between these populations are unclear. The ability of differentiated epithelial cells to dedifferentiate and fulfil functional roles of stem cells has now been observed in several tissues, suggesting that static stem cell populations may not be required to maintain all tissues. In this study, we have further explored the ability of OSE cells to acquire features of stemness and have demonstrated it can be promoted by a variety of conditions. It may be expected that a common gene expression program would underlie the specific stemness phenotype we have assessed with these experiments, but we demonstrate that expression profiles are context specific.
While transcriptional responses were variable, several patterns were recurrent across multiple conditions. We observed that induction of an EMT with TGFB1 treatment or Snail overexpression could promote stemness in OSE, but expression profiles of spheroids naive to TGFB1 treatment also showed EMT activation, which has also been observed with ovarian cancer cells cultured as spheroids 31 . As spheroids have been shown to enrich for cells with stem cell properties, these findings suggest that intrinsic stemness-independent of exogenous treatments-may be associated with a more mesenchymal phenotype. The relationship between the EMT and stemness is well documented 12,32 , but there is growing evidence that stemness and EMT are not inextricably linked. For example, the EMT-promoted transcription factor PRRX1 suppresses stemness in breast cancer cells 33 . Further, transcriptional dynamics of the EMT have been shown to be highly context-specific, which explains why it does not consistently promote stemness 25 . Just as the EMT can occur without promoting stemness, we have shown that deletion of Brca1 promotes stemness in OSE without activating any EMTassociated expression, including Twist1 activation, which had been shown to drive stemness following Brca1 deletion in mammary epithelial cells 28 . Instead, Brca1 loss caused many changes in cell membrane transport and metabolic genes. While the mechanism of induced stemness following Brca1 loss is unclear, this provides strong evidence that stemness is not dependent on a mesenchymal expression profile and is perhaps as context specific as the EMT response 25 .
Several alterations in signaling pathway activity were also common across conditions. TGFB1 treatment and spheroids were associated with higher levels of ERK activity, which have both been linked to stemness in epithelial 34 and carcinoma cells 35 . In EGF-free media, paracrine/autocrine signaling is established, maintaining ERK activity in stem cell populations of intestinal organoids 35 . While these mechanisms may contribute to stemness in OSE spheroids or those treated with TGFB1, increased ERK activity was not enhanced following Snail overexpression or Brca1 loss. Similarly, a gene set comprising interferon alpha response genes was downregulated following spheroid culture, TGFB1 treatment, and Snail overexpression. While it is unlikely that interferon alpha itself was present, it is possible that various signaling pathways may affect common target genes. Consistent with this, disruption of type 1 interferon signaling promotes stemness in breast cancer cells 36 . None of these patterns, however, are consistent across all conditions, further supporting that mechanisms promoting stemness may vary considerably depending on environmental conditions (e.g., ovulatory wound repair, tissue expansion during folliculogenesis, natural cell turnover).
While we have relied heavily on in vitro models here, this has enabled us to explore the ability of OSE cells to acquire stemness following various experimental perturbations. This suggests that differentiated epithelial cells may be capable of self-regulating tissue maintenance in response to environmental cues, such as tissue damage. The expression profiles of this emergent stemness may be variable, depending on the specific properties of the cells' microenvironment. This model is particularly interesting because it is a stark contrast to how stem cells and differentiation hierarchies have been viewed for the last several decades. The OSE is a promising tissue to explore this further as it undergoes regular rupture and repair throughout reproductive cycles, and is a simple tissue comprising a single cell type, which may be the most likely to exhibit this behavior. Designing strategies to monitor stemness dynamics in vivo will be critical to understand these behaviors in a normal physiological context.

Methods
OSE cell isolation and culture. The isolation and culture of mOSE cells was done as previously described 10 , in accordance with the guidelines of the Canadian Council on Animal Care and under a protocol approved by the University of Ottawa Animal Care Committee. Briefly, ovaries from randomly cycling female mice (FVB/N, 6 weeks old) were collected and incubated in 0.25% Trypsin/PBS (Invitrogen) (37°C, 5% CO 2 , 30 min) to facilitate OSE removal. mOSE cells were isolated by centrifugation and plated onto tissue culture plates (Corning) in mOSE media [a-Minimum Essential Medium (Corning) supplemented with 4% FBS, 0.01 mg/mL insulin-transferrin-sodium-selenite solution (ITSS; Roche), and 2 µg/mL EGF (R&D Systems)]. hOSE cells were isolated and cultured as previously described 37 , with patient consent and under a protocol approved by the Ottawa Health Science Network Research Ethics Board (Protocol #1999540). Briefly, ovaries from 5 different women were collected during surgery for reasons other than ovarian pathology. Using a scalpel, hOSE cells were scraped from the ovarian surface and isolated by centrifugation in hOSE media (Wisent Bioproducts) supplemented with 10% FBS. All mouse and human OSE cells were passaged 2-3 times prior to experimental use and experiments were conducted with cells of a passage number less than 25.
Quantitative reverse transcription polymerase chain reaction (RT-PCR). The RNeasy Mini Kit (Qiagen) was used to extract RNA and the OneStep RT-PCR Kit (Qiagen) was used to synthesize cDNA. Quantitative PCR was done using the ABI 7500 FAST qRT-PCR machine (Applied Biosystems) using the Taqman gene expression (Life Technologies) and SsoFast gene expression (Bio-rad) assays utilizing Tbp as an endogenous control. Primer sequences are listed in Supplemental Table 1. RQ (relative quantity) was determined using the cycling threshold for the gene of interest in control or untreated samples compared to the cycling threshold in experimental samples, calculated using the Applied Biosystems 7500 FAST v2.3 software.   Western blot. M-PER mammalian protein extraction reagent (GE Healthcare) was used to extract protein from mOSE cells and run on NuPAGE 4-12% Bis-Tris gradient gels (Life Technologies). Polyvinylidene difluoride membranes were used to transfer protein samples. Membranes were blocked in 5% non-fat milk prior to antibody incubation. Antibody conditions are described in Supplemental being placed in spheroid culture and was replenished when plating cells in spheroid culture. Spheres were quantified using ImageJ using a pixel cutoff of >1000 pixels and a circularity limit of 0.5-1.0. For spheres cultured in methylcellulose, mOSE cells were placed in a 1:1 mixture of methylcellulose and stem cell media at 5 × 10 4 cells/mL in 24-well culture plates (Corning), and incubated at 37°C, 5% CO 2 for 28 days. Methylcellulose-embedded spheres were quantified using ImageJ using a pixel cutoff of >500 pixels and a circularity limit of 0.5-1.0. For each experiment, a minimum of 3 replicates were performed, each replicate was performed in three independent wells, spheres were counted in 4 fields per well, and the average count was reported.
Secondary sphere-forming assay. Primary free-floating mOSE spheres were collected and washed in PBS. Spheres were dissociated by first incubating in trypsin/PBS (Invitrogen) at 37°C for 10 min, then by passing cells through a 25 gauge needle to obtain a single cell suspension. Single cell suspension was verified using phase contrast microscopy. Cells were washed in PBS, counted using a hemocytometer, and plated in stem cell media at 5 × 10 4 cells/mL. Cells were incubated in non-adherent 24-well culture plates (Corning) at 37°C, 5% CO 2 for 14 days. Spheres were quantified using ImageJ using a pixel cutoff of >500 pixels and a circularity limit of 0. The libraries were normalized, denatured in 0.05 N NaOH, and then diluted to 200 pM and neutralized using HT1 buffer. ExAMP was added to the mix and the clustering was done on an Illumina cBot and the flowcell was run on a HiSeq 4000 for 2 × 100 cycles (paired-end mode) following the manufacturer's instructions. A phiX library was used as a control and mixed with libraries at 1% level. The Illumina control software was HCS HD 3.4.0.38 and the real-time analysis program was RTA v. 2.7.7. The program bcl2fastq2 v2.18 was then used to demultiplex samples and generate fastq reads.
RNA-seq processing and differential expression. Transcript quantification for each sample was performed using Kallisto (v0.45.0) 39 with the GRCm38 transcriptome reference and the -b 50 bootstrap option. The R package Sleuth (v0.30.0) 40 was then used to construct general linear models for the logtransformed expression of each gene across experimental conditions. Wald's test was used to test for significant variables for each gene and the resultant p-values were adjusted to q-values using the Benjamini-Hochberg false discovery rate method. Significant genes were defined as genes with a q-value <0.05. An effect size (beta coefficient of the regression model) cutoff of >0.5 or < −0.5 was also used for each data set. To compare TGFB1-treated spheres with monolayer conditions and the untreated spheroids, we used a regression model including an interaction term between spheroid culture and TGFB1 treatment to identify expression patterns associated with TGFB1-treated spheres that are not simply additive changes due to the two variables.
Reactome pathways, and Hallmark genesets were collected from the Molecular Signatures Database (MSigDB) 22,23 and used to query differential expression results ranked by fold change. All gene sets discussed in the manuscript have a significant enrichment (Benjamini-Hochberg adjust p-value <0.05). For pathway activity inference, we used the R package PROGENy (v1.9.6) 42 . Pathway activity was compared between experimental conditions using a simple linear model and pvalues were adjusted using the Benjamini-Hochberg false detection rate method.
CD44 cell sorting. mOSE cells were treated with TGFB1 (10 ng/mL, 2 days) prior to collecting cells for FACS. Cells (1 × 10 7 ) were trypsinized and a single-cell suspension was made using a 40 µm cell strainer. Cells were labeled and sorted as previously described 18 . Briefly, cells were resuspended in a flow buffer (4% FBS in PBS) and incubated with anti-CD44 conjugated to allophycocyanin (1:5000; eBioscience, San Diego, CA) for 15 min at 4°C. Unbound antibody was removed with washing buffer and the fraction of cells with surface protein labeled with CD44 antibody was determined using a MoFlo cell sorter (Dako Cytomation).
Statistics and reproducibility. Statistical analyses were conducted in R (v4.0.3). For all comparisons of means, data were assessed with linear models and two-sided Student's t tests. For experiments involving multiple comparisons, p-values were controlled using the Benjamini-Hochberg false discovery rate method. To ensure reproducibility, all experiments were conducted with independent biological replicates. The number of replicates are specified in figure legends. Individual values for data plots are included in the public GitHub repository at https://github. com/dpcook/ose_stemness.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Raw sequencing files have been deposited and are available along with processed transcript quantifications at GSE122875. Source data for all plots in main figures is provided with this paper.

Code availability
All code used to process data and generate figures is available in a public GitHub repository 43 . Code can be viewed at https://github.com/dpcook/ose_stemness.