Single-cell qPCR demonstrates that Repsox treatment changes cell fate from endoderm to neuroectoderm and disrupts epithelial-mesenchymal transition

A definitive endodermal cell lineage is a prerequisite for the efficient generation of mature endoderm derivatives that give rise to organs, such as the pancreas and liver. We previously reported that the induction of mesenchymal definitive endoderm cells depends on autocrine TGF-β signaling and that pharmacological blockage of TGF-β signaling by Repsox disrupts endoderm specification. The definitive endoderm arises from a primitive streak, which depends largely on TGF-β signaling. If the TGF-β pathway is blocked by Repsox, cell fate after the primitive streak induction is so-far unknown. We report here, that an induced primitive streak cell-population contained many T/SOX2 co-expressing cells, and subsequent inhibition of TGF-β signaling by Repsox promoted neuroectodermal cell fate, which was characterized using single-cell qPCR analysis and immunostaining. The process of epithelial-to-mesenchymal transition, which is inherent to the process of definitive endoderm differentiation, was also disrupted upon Repsox treatment. Our findings may provide a new approach to produce neural progenitors.


Introduction
Differentiation of human pluripotent stem cells (hPSCs) into definitive endoderm (DE) is the critical first step for generating visceral organs, such as liver, pancreas, gut, and lungs [1]. Most protocols for efficient production of DE cells employ exogenous Wnt and recombinant activin A to induce a primitive streak (PS) intermediate within 24 h, followed by continued TGF-β/ activin/nodal signaling for the subsequent 2-5 days. By systematically optimizing the differentiation protocol, Loh et al. were able to differentiate hPSCs into > 98% pure SOX17-expressing DE cells within 48 h [2,3]. In vertebrate embryos and during hPSC differentiation, activation of TGF-β/activin/nodal signaling by activin A is imperative for DE specification [4].
During vertebrate gastrulation, epiblast cells undergo an epithelial-to-mesenchymal transition (EMT) at the primitive streak. During the period of in-vitro endoderm differentiation, EMT also occurs with noticeable changes in cell morphology and upregulation of EMT-related genes [5]. We observed that endogenous TGF-β1 was largely secreted during endoderm specification, and pharmacological inhibition of TGF-β/activin/nodal signaling disturbed DE formation and EMT events. [6] Pluripotent epiblast cells can give rise to three germ layers (ectoderm, mesoderm, and endoderm), and neural tissues are traditionally considered to mainly originate from the ectoderm. The discovery of a bipotent neuro-mesodermal progenitor (NMp), which is considered to occur within the primitive streak-associated epiblast and is bipotential for the posterior neural plate and the paraxial mesoderm, however, challenges the traditional notion [7,8]. NMps, also referred to as axial stem cells, are thought to co-express the neural progenitor marker SOX2 and the early mesodermal marker brachyury (T) in the embryo [9]. Axial stem cells can give rise to neural lineages by persistent activation of SOX2 [10]. It is interesting that successful NMps can be induced from mouse epiblast stem cells (EpiSCs) when cultured in the presence of activin [11]. However, it remains unknown whether co-expressing T and SOX2 cells from hPSCs can be generated following PS induction by activin; moreover, cell fate changes due to TGF-β inhibition caused by Repsox after PS induction are not comprehensively understood.
Here, we report that numerous cells co-expressing T and SOX2 were observed following PS induction, and the subsequent efficient inhibition of TGF-β/activin/nodal signaling by Repsox promoted neuroectoderm formation, which can give rise to neural rosettes. Most DE-specific markers were not up-regulated in the presence of Repsox, and EMT events were also scarce. Based on these findings, we propose a model explaining the mechanism underlying the effects of Repsox.

Flow cytometry
Cells were separated individually using Accutase for 5-10 min and were re-suspended in 2% fetal bovine serum (buffer) to prevent unspecific antibody binding. Cells were labeled using CD325 (Thermofisher Scientific; cat. no. 17-3259-42) for 30 min on ice. Single cells were then fixed in 4% paraformaldehyde for 10 min and permeabilized in 0.1% wt/vol Saponin (Sigma; cat. no. S4521) in PBS for 45 min. Cells were then incubated with SOX17 (BD Biosciences; cat. no. 562205) in a FACS buffer (2% fetal bovine serum in PBS) for 30 min on ice. Control samples were stained using isotype-matched control antibodies. To detect the expression of T in activin-induced PS cultures, the fixed and permeabilized cells were incubated with human/ mouse brachyury Alexa Fluor1 488-conjugated antibodies (R&D Systems; cat. no., IC2085G) for 1 h at 37˚C. All cells were analyzed using an Accuri C6 device (BD Biosciences).

Quantitative reverse-transcription polymerase chain reaction (RT-qPCR)
Total RNA was isolated from samples in triplicate using TRIzol (Thermofisher Scientific; cat. no. 15596026), and 2 μg RNA was used for reverse-transcription with ReverTrace (TOYOBO, Osaka, Japan; cat. no. 34520B1). The produced cDNA was diluted for use as PCR template, and PCR reactions were performed using a SYBR1 Premix Ex Taq™ Kit (Takara Bio, Kusatsu, Japan; cat. no. RR420A) and a CFX96 Touch™ Real-Time PCR Detection System (Bio-Rad). Gene expression levels were normalized against the expression level of GAPDH. Primer sequences are shown in the supplementary file S1 Data.

Enzyme-linked immunosorbent assay (ELISA)
The concentration of TGF-β1 in the cell culture medium was determined using an ELISA kit (R&D Systems; cat. no. DB100B). Cell culture medium supernatant was assayed following the manufacturer's instructions. Optical density was measured at 450 nm using a microplate reader (Mithras2 LB943, Berthold Technologies), and chemokine concentrations were quantified using MikroWin 2010 software (version 5.14, Labsis Laborsysteme GmbH).

Cell migration assay
An artificial wound area in the differentiated and mostly confluent cell monolayer was created using a 200 μL pipet tip. The cultures were then rinsed twice using DMEM/F12 and were subsequently covered with fresh RPMI/B27-minus and activin A medium with or without Repsox. At 0 and 24 h, cells were assessed and photographed under an inverted phase contrast microscope (OLYMPUS IX51, Olympus Optical Co., Ltd).

Statistical analyses
Experiments were performed using three biological repeats, whenever possible. Data are shown as mean ± standard deviation (s.d.). The effects of treatments were tested using unpaired two-tailed Student's t-tests. Statistical significance is reported as P < 0.05.

Results
DE cells were produced at up to 80% confluency of the cultures in the presence of activin A, and an apparent EMT was observed. We detected a strong increase in TGF-β1 gene expression and a high level of TGF-β1 protein secretion into the medium after activin A induction. The addition of Repsox after PS induction led to a sharp decrease in TGF-β1 gene expression ( Fig  1A) and TGF-β1 concentrations (Fig 1B). The cells stained for immunofluorescence (IF) analysis showed a loss of the DE marker SOX17 and of the EMT-related protein N-cadherin (CDH2) in the presence of Repsox; however, the epithelial marker E-cadherin (CDH1) was maintained consistently. CDH1 and SOX17 expression levels appeared to be mutually exclusive (Fig 1C). The flow cytometry results indicated that about 33.7% of the SOX17+CD325 + cells were produced after activin A induction, and SOX17+CD325-cells were scarce, suggesting a mesenchymal status of the DE cells (Fig 1D). In the Repsox treatment, no SOX17 was detected, and CD325 was substantially downregulated, which was in line with the IF results.
A scratch assay (Fig 2A and 2B) was used to investigate the effects of Repsox on cell migration activity. Cells showed very limited migration activity (77 ± 20 μm/24 h) in the Repsox treatment population, indicating their loss of the mesenchymal phenotype. We investigated the inhibition effect of Repsox on endoderm induction using a qPCR (Fig 2C). The expression of the neuroectoderm marker genes PAX6 and SOX2 was significantly increased in the Repsox treatment after PS induction, as were the mesoderm-related markers GATA2/3, HAND1, and BMP4 [9,14,15]. In addition, NANOG was downregulated compared to hESCs (day 0). The mesenchymal markers CDH2, SNAI1, ZEB1, and VIM were also blocked by Repsox. The early and late endoderm markers SOX17, FOXA2, GATA4, and GATA6 were not substantially upregulated after Repsox treatment, nor were the mesoendoderm markers HHEX, EOMES, MIXL1, and WNT3.
The gene expression tests revealed that epithelial-mesenchymal-epithelial transition and acquisition of endoderm markers can be simultaneously disrupted by Repsox, an TGF-βR1 (ALK5) inhibitor, and differentiated cells seemed to express ectoderm and mesoderm markers instead. In order to characterize these cells, we performed single-cell qPCR on 81 cells to measure gene expression in individual cells treated with activin A and Repsox after PS induction. For data analysis, we used previously published single-cell qPCR data to produce relational networks of gene expression of those cells [6]. The cell population treated with Repsox was synchronized and homogeneous (Fig 3A). The expression of pluripotency gene NANOG in cell population treated with Repsox was obviously downregulated compared to hESCs. TGF-β signaling pathway inhibitor significantly blocked the acquisition of endoderm marker genes, particularly that of SOX17 (Fig 3B). After the Repsox treatment, the ectoderm marker genes PAX6 and SOX1 were up-regulated, as was SOX2, which was also up-regulated compared to hESCs (day 0; Fig 3B). The variation in gene expression of the hematopoiesis-related marker GATA2 [16] and of the mesoderm markers HAND1 [3,17,18] and HAND2 [16] was scarcely observed in single cells (S2 Data). For visualization, we produced scatter plots of individual cell expression of CDH1 against CDH2, SOX1, and of SOX2 against the DE marker gene SOX17 (Fig 3C). We showed that increasing ectoderm marker levels were associated with decreasing endoderm marker levels upon the addition of Repsox after PS induction, as SOX1 and SOX2 expression and SOX17 expression were mutually exclusive, and no EMT was observed as measured by the ratio of CDH1 and CDH2 expression. Both RT-qPCR and single-cell qPCR analysis revealed that TGF-β/activin/nodal signaling inhibition by Repsox after PS induction resulted in an upregulation of neuroepithelial markers such as SOX2, PAX6, and SOX1. The immunostaining results showed that many cells treated with Repsox expressed the early neuroectoderm markers PAX6 and SOX2 on day 3 (Fig 4A), which was scarcely detected in the DE population. After the primitive streak induction, numerous cells were observed to co-express SOX2 and T (83.4%; Fig 4A and 4B); however, T expression decreased rapidly on day 3 with and without the Repsox treatment. In order to further characterize the obtained early neuroectoderm cells, we continued culturing them in N2B27 basal medium with SB431542 and Dorsomorphin for eight days, and insulin was also added to improve the induction process [19]. Cells were then split and cultured in N2B27 basal medium with bFGF until neural rosettes were observed [19,20]. We observed the expression of the neuroepithelial marker PAX6 in the neural rosettes by immunostaining (Fig 4C). Neurospheres derived from these rosettes were further differentiated to neural cells, which were both TUJ1-positive and MAP2-positive (Fig 4D). The astrocyte marker GFAP can hardly be detected, since the neuroepithelium mainly gives rise to neurons upon differentiation and gliogenesis may be achieved after a long-time culture [21]. Based on the results above, the early neuroectoderm can be generated after a Repsox treatment following PS induction.

Discussion
Efforts have been made to efficiently differentiate hPSCs into DE cells [2,5,[22][23][24]. For nearly all of the methods aiming to do this, TGF-β/activin/nodal signaling activated by activin is essential for DE specification; however, it is insufficient for producing homogeneous DE cells even combined with the use of additional growth factors [25]. Single-cell qPCR is a powerful method that can be used to assess specific gene expression at single-cell resolution and reveal cellular heterogeneity, and this method has been widely applied in the biomedical field [26][27][28]. DEs were reported to arise from PS in vivo, as well as NMps [8,29]. Tsakiridis et al. reported the generation of NMp-like cell populations from mouse EpiSCs cultured in activin and FGF2 in vitro [11]. In our study, numerous T/SOX2-positive cells were detected in the activin A-induced primitive streak, the formation of which depended on endogenous Wnt with T as a specific marker [2]. We thus hypothesized that these NMp-like cells may be biopotential for neural and mesoderm differentiation.
Recent studies demonstrated that NMps can subsequently be differentiated towards neural fate with RA and a sonic hedgehog agonist in vitro [30,31]. Surprisingly, activin/nodal inhibition was found to efficiently promote neural conversion from hPSCs [32]. Based on these reports, we aimed to determine whether Repsox can steer T/SOX2-positive cells in an activininduced PS to be neural progenitors.
We previously reported that activin-induced DE formation was associated with a synchronous EMT mediated by autocrine TGF-β signaling, and blockage of TGF-β/activin/nodal signaling by Repsox during three days of differentiation inhibited EMT and DE formation [6].
Here, we investigated cell fate changes accompanying pharmacological inhibition of TGF-β signaling in activin A-induced PS cultures, particularly at single-cell resolution. We found that inhibition of TGF-β/activin/nodal signaling by Repsox after PS induction promoted neuroectodermal cell fates with the loss of mesenchymal characteristics. Our method largely differs from the classical derivation of PAX6 neural progenitor cells by dual SMAD inhibition [33]. We thus propose a model explaining the mechanism underlying the effects of Repsox treatments. The cell population after PS induction contained stem cells co-expressing SOX2 and T, which can give rise to both neuroectoderm and mesoderm. Repsox may facilitate the generation of neuroectoderm by the persistent activation of SOX2, which by itself was demonstrated to generate induced neural stem cells from mouse and human fibroblasts [34].
Taken together, our results demonstrate that the pharmacological inhibition of TGF-β/activin/nodal signaling in activin A-induced PS cultures leads to substantial cell fate changes from DE cells to neuroectoderm lineages with the disappearance of noticeable EMT events accompanying DE formation.