TGF-β modulates cell fate in human ES cell-derived foregut endoderm by inhibiting Wnt and BMP signaling

Summary

Genetic differences between pluripotent stem cell lines cause variable activity of extracellular signaling pathways, limiting reproducibility of directed differentiation protocols. Here we used human embryonic stem cells (hESCs) to interrogate how exogenous factors modulate endogenous signaling events during specification of foregut endoderm lineages. We find that transforming growth factor β1 (TGF-β1) activates a putative human OTX2/LHX1 gene regulatory network which promotes anterior fate by antagonizing endogenous Wnt signaling. In contrast to Porcupine inhibition, TGF-β1 effects cannot be reversed by exogenous Wnt ligands, suggesting that induction of SHISA proteins and intracellular accumulation of Fzd receptors render TGF-β1-treated cells refractory to Wnt signaling. Subsequently, TGF-β1-mediated inhibition of BMP and Wnt signaling suppresses liver fate and promotes pancreas fate. Furthermore, combined TGF-β1 treatment and Wnt inhibition during pancreatic specification reproducibly and robustly enhance INSULIN⁺ cell yield across hESC lines. This modification of widely used differentiation protocols will enhance pancreatic β cell yield for cell-based therapeutic applications.

Graphical abstract

Highlights

• •TGF-β1 or Wnt inhibition promotes pancreatic over liver fate in hESC differentiation
• •A TGF-β1-activated putative human OTX2/LHX1 GRN anteriorizes definitive endoderm
• •TGF-β1 inhibits endogenous Wnt and BMP signaling in foregut progenitors
• •Combined TGF-β1 and Wnt inhibition increase INSULIN⁺ cell yield across hESC lines

Serup and colleagues show that, in a widely used pancreas differentiation protocol, exogenous TGF-β1 anteriorizes hESC-derived definitive endoderm and then promotes pancreas fate at the expense of liver via inhibition of endogenous Wnt and BMP signaling in foregut progenitors. Combined TGF-β1 and Wnt inhibition additionally promote differentiation of pancreatic progenitors to INSULIN (INS)⁺ cells and reduce batch-to-batch variation between differentiations across hESC lines.

Introduction

Directed differentiation of human embryonic stem cells (hESCs) recapitulates signaling events governing cell lineage decisions, from germ layer specification during gastrulation, to later regionalization along anterior-posterior (A-P) and dorsoventral (D-V) axes. In vertebrates, a combination of Wnt and high Nodal signaling triggers formation of the definitive endoderm (DE) layer that emerges from the primitive streak. DE closure along the anterior and caudal intestinal portals forms the primitive gut (PG). The PG is patterned along the A-P and D-V axes to form foregut, midgut, and hindgut that are further patterned into distinct organ territories, including thyroid, lung, liver, pancreas, and gastro-intestinal tract (Arnold and Robertson, 2009; Grapin-Botton, 2005; Tam and Loebel, 2007). Posterior mesoderm, ectoderm, and endoderm fates are all promoted by high levels of Wnt, bone morphogenetic protein (BMP), and retinoic acid (RA) signaling (Bayha et al., 2009; Deimling and Drysdale, 2009; Spence et al., 2011; Stevens et al., 2017). Conversely, anterior fates arise when ligand-sequestering antagonists such as Dkk1, Sfrp, Noggin, and Chordin, secreted from anterior visceral endoderm and anterior mesendoderm (ME), protect anterior tissues from Wnt and BMP signaling (McLin et al., 2007; Rankin et al., 2011). Later, these repurposed signaling pathways act stage-dependently to promote distinct organ lineages such as liver and pancreas (Rankin et al., 2018). These arise from common multipotent ventral foregut progenitors influenced by fibroblast growth factor (FGF), BMP, and Wnt signals from adjacent mesoderm (Zaret, 2008). Dose-dependent FGF signaling from cardiogenic mesoderm and BMP signaling from septum transversum mesenchyme induce hepatic fate and suppress pancreatic fate (Deutsch et al., 2001; Jung et al., 1999; Rossi et al., 2001; Serls et al., 2005). Thus, most hESC pancreas differentiation protocols use BMP signaling inhibitors to reduce liver specification (Nostro et al., 2011; Pagliuca et al., 2014; Rezania et al., 2014). Conversely, transforming growth factor (TGF)-β signaling favors pancreatic over hepatic specification in hESC cultures (Loh et al., 2014). However, it is unclear how TGF-β and BMP signaling interact during these fate decisions.

How Wnt signaling acts on multipotent foregut progenitors to specify organ fate is not fully understood (Zaret, 2008). Ectopic Wnt signaling in Xenopus foregut progenitors inhibits development of foregut organ (pancreas, liver, and lung) buds, while repression causes liver and pancreas bud expansion (McLin et al., 2007). However, this can be interpreted as Wnt repression being required for specification of anterior endoderm, and so foregut progenitors. In zebrafish, loss of canonical Wnt2/Wnt2bb signaling blocks liver and swim bladder specification and pancreas and anterior intestines expand (Ober et al., 2006; Poulain and Ober, 2011). Studies in mouse and human systems indicate that non-canonical Wnt signaling promotes pancreas over liver fate, while contradictory results were found regarding the ability of canonical Wnt signaling to promote liver fate (Mahaddalkar et al., 2020; Rodriguez-Seguel et al., 2013). Thus, interaction between signals regulating segregation of liver and pancreas primordia remains poorly understood.

Similarly, we have limited understanding of how exogenous growth factors affect endogenous signaling pathways during hESC differentiation. Although varying endogenous signaling activity between hESC lines (Ortmann et al., 2020) is an obvious source of variation in differentiation outcomes, the role of endogenous signaling pathways has received scant attention. Generally, if detrimental to the desired differentiation outcome, endogenously active pathways are inhibited via trial and error while pathways whose activity is considered desirable are stimulated. Yet, crosstalk among signaling pathways occurs during embryonic development. It is therefore to be expected that variability of endogenous signaling, and addition of exogenous factors, will affect this crosstalk and the outcome of in vitro pluripotent stem cell (PSC) differentiation in a cell line-dependent manner.

Here we investigated how exogenous growth factors and inhibitors affect endogenous signaling during early pancreatic lineage specification in differentiating HUES4 and H1 hESC cultures. A developmental signaling pathway screen identified TGF-β1 and the Porcupine inhibitor IWP-L6 as potent inducers of pancreatic fate when present during development of foregut and early pancreatic endoderm (PE). We confirm the anteriorizing effect of Wnt inhibition and show that TGF-β1 anteriorizes endoderm by inhibiting the expression and/or function of Wnt signaling pathway components. Endogenous Wnt/β-catenin and BMP signaling promote liver and suppress pancreas fate in hESC-derived foregut progenitors, particularly in HUES4 cells, which are inherently more prone to liver differentiation than H1. Exogenous TGF-β1 suppresses liver fate by dual inhibition of BMP and Wnt signaling. Finally, we show that TGF-β1 treatment and Wnt inhibition during foregut and pancreas specification additively and robustly enhance INS⁺ cell yield across hESC lines.

Our work uncovers mechanisms that underlie an extensive crosstalk between developmental signaling pathways in foregut progenitors as they develop toward liver and pancreas during hESC differentiation. This can be exploited to enhance in vitro differentiation of pancreatic progenitors and, subsequently, β cells for cell-based therapeutic applications.

Results

TGF-β1 and Wnt inhibitors promote pancreatic progenitor marker expression

When subjecting a PDX1^EGFP/+ HUES4 reporter cell line (Ameri et al., 2017) (hereafter “HUES4”) to a control pancreas differentiation protocol (Rezania et al., 2014), we observed disappointingly low PDX1 expression at the end of stage 5 (S5, day [D]13), the beginning of endocrine differentiation (Figures 1A and 1B). Immunofluorescence (IF) analysis revealed efficient induction of SOX17⁺ FOXA2⁺ DE at the end of S1 (D3) with little to no expression of SOX1 (ectoderm) or BRACHYURY (mesoderm) (Figure S1A). As differentiation propensity varies between hESC lines (Osafune et al., 2008), we conducted parallel analyses on H1 hESCs, finding DE to be induced with comparable efficiency to HUES4 (Figure S1A). We therefore reasoned that subsequent A-P patterning and/or PE specification were suboptimal. To promote pancreatic differentiation, we added agonists or antagonists of selected signaling pathways active in A-P patterning and pancreas specification to differentiating (PDX1^EGFP/+) HUES4 cells from S2–S5 (D3–D13) and monitored GFP expression at the end of S5 (D13) (Figures 1A and 1B). Fluorescence-activated cell sorting (FACS) analysis revealed a significant increase in GFP⁺ cells when blocking Wnt secretion with the Porcupine inhibitor IWP-L6 or adding recombinant TGF-β1 or Activin A, while perturbing Notch signaling and relevant receptor tyrosine kinase pathways had minimal effects (Figure 1B). Concordantly, treatment with IWP-L6 or TGF-β1 during S2–S5 increased PDX1 protein expression at D13 by IF (Figure S1B).

Figure 1.

TGF-β1 and IWP-L6 anteriorize endoderm and promote pancreas over liver fate

(A) Overview of the basic differentiation protocol. Stages when screened compounds were added are indicated. DE, definitive endoderm; PGT, primitive gut tube; PP1, pancreatic progenitor 1; PP2, pancreatic progenitor 2; EP, endocrine precursor.

(B) FACS analyses of PDX1^EGFP/+ HUES4 cells showing percent GFP⁺ cells at D13 after treatment with the indicated factors during S2-S5. Mean ± SD (n = 3 from independent experiments). ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗∗p < 0.0001.

(C) IF staining for PDX1 (green) and NKX6-1 (red) in PDX1^EGFP/+ HUES4 cells at D13 treated with vehicle during S2-S5 or IWP-L6 during S2–S3, S4, or S5. Cells were counterstained with DAPI (white). Scale bar, 50 μm.

(D) Real-time qPCR analyses of PDX1, NKX6-1, OTX2, CDX2, and AFP expression, in PDX1^EGFP/+ HUES4 cells at indicated timepoints after treatment with indicated factors during S2–S3. Data are shown relative to undifferentiated PDX1^EGFP/+ HUES4 cells: mean ± SD (n = 3 from independent experiments). ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.0005, ^∗∗∗∗p < 0.0001.

See also Figure S1.

To better define when Wnt inhibition is required to promote pancreatic differentiation, we treated HUES4 cells with IWP-L6 during S2–S3, S4, or S5 or with vehicle control (S2–S5). Crucially, PDX1⁺ NKX6-1⁺ pancreatic progenitors were more prevalent at D13 after exposure to IWP-L6 during S2–S3 but not when added later (Figure 1C), consistent with prior work (Nostro et al., 2015). Furthermore, real-time qPCR gene expression analysis at the end of S3 (D7) for PDX1 and OTX2, a marker of anterior endoderm (and anterior ME and ectoderm), confirmed PDX1 upregulation only when IWP-L6 is added during S2–S3 and OTX2 upregulation when treated with TGF-β1 during S2 or when either TGF-β1 or IWP-L6 are added at S2–S3 (Figure S1C). PDX1 and NKX6-1 transcripts increased from D7 and D13, respectively, after IWP-L6 treatment, while the effect of TGF-β1 on PDX1 expression was only evident from D10 (Figure 1D). OTX2 (anterior) expression was significantly higher in both IWP-L6- and TGF-β1-treated samples than in controls at D5 and D7. Conversely, expression of the posterior marker CDX2 and the early liver lineage marker alpha fetoprotein (AFP) was suppressed from S2 onwards by both treatments (Figure 1D). These patterns of PDX1, OTX2, and AFP expression were confirmed at the protein level at D5–D13 after S2–S3 IWP-L6 treatment (Figure S1D). Together, these findings suggest that IWP-L6 inhibits secretion of posteriorizing Wnt proteins in our cultures, consistent with how Wnt proteins act in vivo (Loh et al., 2014; McLin et al., 2007).

Wnt inhibition or exogenous TGF-β1 anteriorizes endoderm

OTX2 and LHX1 are integral components of an anteriorizing gene regulatory network (GRN) in developing fish, frogs, and various mammals (Costello et al., 2015; Fossat et al., 2015; Ip et al., 2014; Sibbritt et al., 2018) and possibly humans given such broad conservation. IF revealed increased expression of both OTX2 and LHX1 in D5 HUES4 cells after Wnt inhibition with either IWP-L6 or the tankyrase inhibitor XAV-939 at S2 (Figures 2A and S2A). Notably, treatment with TGF-β1 alone or together with IWP-L6 increased OTX2/LHX1 expression to the same extent as IWP-L6 alone (Figure 2A). H1 gave similar results, with higher expression of OTX2/LHX1 in TGF-β1 and TGF-β1 + IWP-L6 conditions (Figure 2A). The homogenous D5 OTX2⁺ HUES4 population was FOXA2⁺ in all conditions; likewise, D5 OTX2⁺ H1 cells, especially prevalent after TGF-β1 and TGF-β1 + IWP-L6 treatment, were FOXA2⁺ consistent with an endodermal, foregut identity (Figure S2A). By D7 (S2–S3 treatment), OTX2 marked discrete clusters, distinct from emerging PDX1⁺ cells (Figure S2B) while expression of the foregut marker SOX2 was increased by all treatments in both lines (Figure S2C). PDX1⁺ cells were increased in D7 HUES4 differentiations after S2–S3 treatment with IWP-L6 or, more so with, XAV-939 and at D10 following TGF-β1, IWPL-6, or both as well as with XAV-939 treatment, in all cases at the expense of AFP⁺ cells abundant in control D7 and D10 cultures (Figures 2B and 2C). This pattern was recapitulated with H1 hESCs, the major difference being more prevalent PDX1⁺ cells in control H1, vs. HUES4, cultures (Figures 2B and 2C). The decrease in AFP⁺ cells with TGF-β1 and/or Wnt inhibition (XAV-939 giving a stronger effect than IWP-L6) was largely mirrored at D10 in both lines and was concomitant with the expanded OTX2⁺ anterior (pancreas-competent) population compared to controls after exposure to TGF-β1 with or without IWPL-6 (Figure S2D). Together, these results suggest that endogenous Wnt proteins posteriorize hESC-derived DE cultures while exogenous TGF-β1 has the opposite effect. While H1 appears more biased toward pancreatic differentiation than HUES4, exogenous TGF-β1 and/or inhibition of endogenous Wnt signaling promote foregut and pancreatic fate in both.

Figure 2.

Wnt inhibition or TGF-β1 anteriorizes endoderm and promotes pancreatic over hepatic differentiation across hESC lines

(A) IF staining for OTX2 (red) and LHX1 (green) in PDX1^EGFP/+ HUES4 and H1 cells at D5 after treatment with the indicated factors during S2. Cells were counterstained with DAPI (white). Scale bar, 50 μm.

(B and C) IF staining for PDX1 (green) and AFP (red) in PDX1^EGFP/+ HUES4 and H1 cells at D7 (B) or D10 (C) after treatment with the indicated factors during S2–S3. Cells were counterstained with DAPI (white). Scale bars, 50 μm.

See also Figure S2.

HUES4 hESCs are biased toward posterior endoderm differentiation

To determine how IWP-L6 and TGF-β1 affect hESC-derived DE cultures, we performed RNA sequencing (RNA-seq) analysis of HUES4 and H1 cells treated with IWP-L6, TGF-β1, or both at the end of S2 (D5) and S3 (D7) (Table S1). At D5, gene sets associated with fetal liver, liver, and intestinal lineages were significantly enriched in control vehicle-treated HUES4 vs. H1 cells (Figure 3A). Markers of liver (AFP, ALB, and PROX1; Figure 3B) and intestine (CTSE, GUCA2A, and CDX2; Figure 3C) were enriched in HUES4 cells, indicating HUES4 to be more inherently posterior endoderm biased. To investigate whether differences in endogenous Wnt and/or TGF-β signaling levels might underlie this bias, we assessed expression of Wnt pathway components in D5 control cultures. This revealed 3.07 log2FC higher expression of the classical non-canonical Wnt ligand WNT5A and a −5.92 log2FC for the Wnt antagonist SHISA3 in HUES4 vs. H1 cells (Figure 3D). Similar analysis of TGF-β signaling pathway targets uncovered no obvious differences between lines. Thus, the stronger tendency of HUES4 to differentiate into posterior endodermal fates such as liver and intestine might be ascribed to elevated endogenous Wnt signaling, likely through the non-canonical pathway during foregut and pancreas specification.

Figure 3.

TGF-β1 anteriorizes endoderm by antagonizing Wnt signaling via a conserved mechanism across hESC lines

(A–D) Tables showing p and FDR q-values (A) or log2FC and padj. values (B–D) for: the top 10 enriched gene sets (C8 cell types) (A), liver markers (B), intestinal markers (C), or genes encoding Wnt signaling pathway components (D) in PDX1^EGFP/+ HUES4 vs. H1 cells at D5.

(E) PCA showing clustering along PC1 and PC2 for PDX1^EGFP/+ HUES4 (n = 3 from independent experiments) and H1 (n = 2 from independent experiments) cells treated with indicated factors during S2 (D5) or S2–S3 (D7).

(F) Venn diagrams showing overlap of all regulated genes at D5 and D7 with an FDR <0.1 and fold change >1.5: PDX1^EGFP/+ HUES4 cells (left) and H1 cells (right).

(G) Heatmaps scaled by row to show patterns of regulated genes in PDX1^EGFP/+ HUES4 cells (n = 3 from independent experiments) (upper panels) and H1 cells (n = 2 from independent experiments) (lower panels) by k-means clustering analysis of RNA-seq data. Signature genes are indicated for each cluster.

(H) Normalized enrichment scores (NESs) from GSEA plots comparing gene expression data from each treatment of PDX1^EGFP/+ HUES4 cells at D5 (n = 3 from independent experiments) with the indicated gene sets. p values are indicated.

(I and J) GSEA plots comparing PDX1^EGFP/+ HUES4 D5 gene expression data, either for each treatment with gene sets for A-P pattern specification (I) or for TGF-β1 treatment with negative regulation of Wnt signaling (J) (n = 3 from independent experiments). Signature genes from leading and trailing edge analyses are shown in red and blue boxes, respectively. NES and p values are shown.

See also Figure S3.

TGF-β1 anteriorizes endoderm by antagonizing Wnt signaling

Principal-component analysis (PCA) of the RNA-seq data from D5- or D7-treated HUES4 and H1 cells revealed samples to cluster by cell line and treatment. Although the major separation was between cell lines, there was a clear transcriptional response to distinct treatments with similar trajectories of the clusters along principal component (PC) 1 and PC2 between lines (Table S2; Figure 3E). In cell line-specific PCA analyses, HUES4 showed a clearer separation by treatment than H1 although similar changes in gene expression were evident in both (Table S2, Figures S3A and S3B). DESeq2 analysis revealed substantial gene expression changes significant in both cell lines (adjusted p value <0.05) at D5 after treatment with IWP-L6 and/or TGF-β1 vs. vehicle. For both HUES4 and H1, a synergistic effect was seen with IWP-L6 + TGF-β1, indicating a greater number of regulated genes vs. the combined effects of separate IWP-L6 or TGF-β1 treatment (Figure 3F; Table S2). A similar total number of significantly changed genes were seen in D7-treated H1 cells as at D5. However, there were more transcriptional changes across all three treatment conditions in D7 compared to D5 HUES4 cells. There were also more TGF-β1-induced changes in HUES4 cells at D7, suggesting greater TGF-β1 sensitivity than H1 cells at D7. Notably, no synergistic effect was seen at D7 in the IWP-L6 + TGF-β1 condition for HUES4 or H1 (Figure 3F; Table S2).

K-means clustering revealed substantial overlap in gene regulation across conditions in both lines (Figures 3F and 3G; Table S2). Genes upregulated by TGF-β1 and TGF-β1 + IWP-L6 at D5 occupied clusters 1 and 6 for HUES4 with many being in H1 clusters 1 and 2. Strikingly, IWP-L6 markedly upregulated HUES4 cluster 1 genes, with cluster 6 genes weakly stimulated or unchanged (Figure 3G; Table S2). Similarly, for H1, cluster 1 was upregulated by IWP-L6 while cluster 2 was less upregulated if changed. Clusters 1 and 6 for HUES4 and clusters 1 and 2 for H1 included human orthologs of the anteriorizing Otx2/Lhx1 GRN downstream of Smad2/3 in Xenopus and mouse (Costello et al., 2015; Fossat et al., 2015; Sibbritt et al., 2018): EOMES, GSC, HHEX, LHX1, and OTX2 and downstream targets CER1, NOG, FZD5, and SHISA2. Similarly, genes downregulated by TGF-β1 and TGF-β1 + IWP-L6 at D5 (HUES4: clusters 2–5; H1: clusters 3–5) were generally downregulated by IWP-L6, albeit more moderately. These included posterior markers EVX1 and CDX2, as well as Wnt and BMP agonists and their target genes WNT11, RSPO3, LEF1, and RNF43 and BMP5/6, BAMBI, and ID2/4, respectively (Figure 3G; Table S2). Notably, and exclusive to D5 HUES4 but not H1, cluster 4 included liver markers ALB, FABP1, SERPINC1/F2, and APOB, downregulated by treatment with IWP-L6 and/or TGF-β1.

For both lines at D7, upregulation of OTX2/LHX1 GRN factors (HUES4: cluster 6; H1: cluster 2) and downregulation of posterior markers, Wnt and BMP agonists, and liver markers persisted. Furthermore, PDX1 was upregulated by IWP-L6 in HUES4 (cluster 1) and H1 (cluster 3) and moderately upregulated by TGF-β1 or TGF-β1 + IWP-L6.

Gene set enrichment analysis (GSEA) on IWP-L6- and/or TGF-β1-treated (at S2) D5 HUES4 cells was dominated by enrichment of Gene Ontology (GO) cell cycle categories (Figures S3C and S3D) while terms related to A-P pattern specification and negative regulation of Wnt signaling were also enriched (Figure 3H, Tables S3 and S4). Consistent with the former, TGF-β1 and/or IWP-L6 addition during S2 (D5) or S2–S3 (D7) upregulated proliferation markers at D5 (Figure S3E; Table S2), and FACS analysis of EdU incorporation revealed increased S-phase cells at D5 and D7 (Figure S3F). As increased proliferation of OTX2⁺ anterior endoderm prior to organ regionalization would yield a larger pool of pancreas-competent progenitors, we assessed OTX2⁺ cell proliferation by EdU incorporation. Both IWP-L6 and TGF-β1 at S2 significantly increased the percentage of EdU⁺ OTX2⁺ cells at D5 (Figures S3G and S3H). However, despite this, we failed to detect an increase in total cell numbers at D7 with any treatment during S2–S3 in either line (Figure S3I), suggesting that any increase in cell number due to higher proliferation is offset by increased cell death. Leading edge analysis of terms related to A-P pattern specification and negative regulation of Wnt signaling identified human orthologs of genes in the Otx2/Lhx1 GRN, OTX2, HHEX, LHX1, SIX3, and CER1 (Figure 3I) and Wnt pathway genes SHISA2, SFRP1/2, ROR2, TCF7L2, and KREMEN2 after TGF-β1 treatment (Figure 3J). Together, these data suggest that TGF-β1 anteriorizes endoderm by antagonizing Wnt signaling, potentially involving a putative human OTX2/LHX1 GRN and augmented anterior endodermal cell proliferation. Furthermore, IWP-L6 and/or TGF-β1 repress posterior endoderm differentiation earlier in the more posteriorly biased HUES4 line.

TGF-β1-induced gene expression changes suggest that a putative human OTX2/LHX1 GRN antagonizes Wnt signaling at multiple levels

To dissect how TGF-β1 antagonizes Wnt signaling, we analyzed expression of Wnt pathway agonists, antagonists, and target genes in response to TGF-β1 treatment. WNT11 was highly expressed and WNT3, WNT5A, and WNT5B more moderately so in control D5 and D7 HUES4 differentiations, while RSPO3 was the sole R-Spondin expressed (Figure S4A; Table S1). Notably, TGF-β1 significantly reduced expression of WNT11 and RSPO3 as well as target genes LEF1 and RNF43, while similar reductions after IWP-L6 treatment only reached significance for LEF1 and RSPO3 (Figure 4A). As expected, blocking TGF-β1 action with SB431542 rescued expression of RSPO3 and LEF1 at D5 but had no effect on IWP-L6-induced changes (Figure S4B). TGF-β1 stimulated expression of both soluble and intracellular Wnt antagonists SFRP1, SHISA2, and SHISA3 at D5 while SFRP1, SFRP2, SHISA2, and SHISA4 were induced at D7 (Figure 4B; Table S2). CER1 was also strongly induced at both time points but may not antagonize Wnt signaling in humans (see Discussion). Importantly, TGF-β1-treated D5 cells failed to activate Wnt target genes in response to exogenous Wnt3a/RSPO3, showing that Wnt signaling is blocked by simultaneous TGF-β1 treatment (Figure 4C). Furthermore, TGF-β1 also prevented Wnt3a-induced β-catenin stabilization (Figure 4D). Thus, SHISA2/3 induction was highly notable as the encoded proteins cell autonomously antagonize Wnt signaling by sequestering Frizzled (Fzd) receptors in the endoplasmic reticulum (ER) (Onishi and Zou, 2017; Yamamoto et al., 2005). We therefore analyzed Fzd mRNA and protein expression in control, IWP-L6-, and TGF-β1-treated cultures. The four highest expressed Fzd transcripts in D5 and D7 control HUES4 differentiations were FZD4, -5, -7, and -8 (RPKM >10; Figure S4A; Table S1). Consistent with mouse Fzd2, Fzd5, Fzd7, and Fzd8 being members of the Otx2/Lhx1 GRN (Costello et al., 2015; Sibbritt et al., 2018), IWP-L6 or TGF-β1 increased FZD2 and FZD5 expression significantly, with smaller, non-significant increases in FZD7 and FZD8 (Figure S4C; Table S2). To visualize FZD receptors and address whether TGF-β1-induced SHISA2/3 expression correlated with FZD subcellular localization, we permeabilized control and TGF-β1-treated cells and labeled them with a human-Fc-tagged version of the next-generation surrogate Wnt, DRPB-Fz7/8, which recognizes Fzd1, -2, -5, -7, and -8 (Dang et al., 2019; Miao et al., 2020). Confocal imaging showed prominent intracellular accumulation of FZD receptors in TGF-β1-treated samples vs. controls (Figure 4E), suggesting that SHISA2/3-mediated ER retention of FZD receptors may contribute to TGF-β1-induced Wnt signaling suppression.

Figure 4.

TGF-β1 activates an OTX2/LHX1 GRN that antagonizes Wnt signaling

(A) Expression of WNT11, RSPO3, LEF1, and RNF43 by RNA-seq in PDX1^EGFP/+ HUES4 cells treated with vehicle (Ctrl), IWP-L6, TGF-β1, or IWP-L6 + TGF-β1 (I + T) during S2 (D5) or S2–S3 (D7). Mean ± SEM (n = 3 from independent experiments). ^∗padj <0.05, ^∗∗padj <0.005, ^∗∗∗∗padj <0.0001.

(B) Differential expression of Wnt signaling antagonists by DESeq2 analysis of RNA-seq data from PDX1^EGFP/+ HUES4 cells treated with TGF-β1 or IWP-L6 + TGF-β1 during S2 (D5) or S2-S3 (D7). Log2FC relative to vehicle controls and padj. values are shown.

(C) Real-time qPCR analysis of LEF1, RNF43 and RSPO3 expression in D5 PDX1^EGFP/+ HUES4 cells in response to Wnt3a + RSPO3 stimulation during S2 in control (Ctrl) cells or cells treated with IWP-L6 or TGF-β1. Mean ± SD (n = 3 from independent experiments). ^∗p < 0.05.

(D) Western blot analysis for β-catenin and Tubulin (internal control) in D5 PDX1^EGFP/+ HUES4 cells treated with vehicle control, Wnt3a, TGF-β1 or Wnt3a + TGF-β1 during S2. Minimum, maximum, and mean (n = 3 from independent experiments).

(E) Confocal microscopy of D5 PDX1^EGFP/+ HUES4 cells stained for β-catenin (green) and Fzd receptors (red) and counterstained with DAPI (blue) after treatment with vehicle or TGF-β1 during S2. Scale bar, 10 μm.

(F) Real-time qPCR analysis of ROR2 expression at D5 in PDX1^EGFP/+ HUES4 cells treated with vehicle (Ctrl), IWP-L6, TGF-β1 or IWP-L6 + TGF-β1 (I + T) during S2. Mean ± SD (n = 3 from independent experiments). ^∗p < 0.05.

(G) Signal tracks of ATAC-seq data from Geusz et al. (2021) and SMAD2, EOMES, FOXA2 and OTX2 ChIP-seq data from Tsankov et al. (2015) at the SHISA2, SHISA3, SFRP1, and CER1 loci. Dashed lines denote regions with differential chromatin accessibility at the definitive endoderm (DE) stage, and TF binding at the mesendoderm (ME) and/or DE stages. ES, ES cells; GT, primitive gut tube; PP1, pancreatic progenitor 1.

See also Figure S4.

Examining Wnt co-receptor expression, we found a tendency to increased ROR2 mRNA in the RNA-seq data in response to IWP-L6 and TGF-β1 (Figure S4D) consistent with a 2-fold increase in ROR2 expression by real-time qPCR (Figure 4F). Conversely, LRP4 was downregulated (Figure S4D), while the more highly expressed LRP5/6 remained unchanged (Table S2). Overall, these changes may bias any residual Wnt signaling toward the non-canonical pathway.

In mice, fish, and frogs, genes encoding signaling antagonists such as Cer1, Sfrp1, and Shisa2 as well as Wnt receptors Fzd2, Fzd5, Fzd7, and Fzd8 are activated by Otx2 and Lhx1 in concert (Costello et al., 2015; Fossat et al., 2015; Sibbritt et al., 2018), while ligands such as Wnt8 and Wnt11 are repressed by Gsc with Otx2 (Seiliez et al., 2006; Yao and Kessler, 2001; Yasuoka et al., 2014). As these transcription factors (TFs) were strongly induced by TGF-β1 (Figure S4E), we assessed chromatin accessibility and TF binding for these Wnt pathway genes during hESC-to-endoderm differentiation (Geusz et al., 2021; Tsankov et al., 2015). We found enhanced chromatin accessibility and binding of EOMES, FOXA2, and OTX2 at two putative cis-acting regions ∼65 and ∼130 kb downstream of the SHISA2 gene at the DE stage (Figure 4G). Data for human LHX1 are unavailable, and, consistently, we were unable to source a chromatin immunoprecipitation (ChIP)-grade anti-LHX1 antibody. However, Otx2 and Lhx1 have been shown to bind the mouse Shisa2 gene (Costello et al., 2015). Similarly, two regions ∼30 and ∼117 kb downstream of SHISA3 and multiple regions in an ∼80 kb region upstream of SFRP1 showed enhanced chromatin accessibility at the DE stage. Additionally, EOMES, FOXA2, and OTX2 bound these regions. For CER1, we found enhanced chromatin accessibility and binding of SMAD2, EOMES, FOXA2, and OTX2 at the promoter region at both ME and DE stages (Figure 4G). Together, these findings suggest that TGF-β1 treatment at the DE stage acts via a conserved GRN downstream of OTX2, LHX1, and GSC. This network can suppress Wnt/β-catenin signaling through multiple mechanisms including ligand downregulation, induction of antagonists, and changes in co-receptor expression.

TGF-β1-induced inhibition of BMP and Wnt signaling promotes pancreas over liver specification

Suppression of AFP expression by IWP-L6 and TGF-β1 prompted us to further investigate pancreas vs. liver differentiation in our cultures. Besides AFP, other early liver-specific genes APOA1, FABP1, and TTR were reduced in expression at D7 (Figure 5A), and GSEA of the RNA-seq datasets showed both IWP-L6 and TGF-β1 to strongly reduce liver-specific gene expression (Figures 5B and S5A). Of all single-cell RNA-seq signatures following treatment at S2 (D5) or S2–S3 (D7), hepatocyte signatures were most profoundly reduced (Figure S5B, Tables S4 and S5). GSEA comparing our RNA-seq data with RNA-seq data from laser-capture microdissected dorsal pancreatic buds (DP) and hepatic cords (HC) of Carnegie stage 13 human embryos (Jennings et al., 2017) confirmed that expression patterns of genes induced or downregulated in IWP-L6-treated cells at D7 corresponded to genes found in dorsal pancreas (DP high) and HC (HC high) gene sets, respectively (Figures 5C and S5C; Table S6). Notably, IWP-L6 or TGF-β1 treatment at S2 alone (D3-D5) was sufficient to suppress D7 AFP expression, indicating that Wnt and TGF-β signaling already modulate liver-pancreas lineage segregation at the PG tube stage (Figure S5D). Together, our results suggest that canonical Wnt signaling promotes human liver specification, as in zebrafish (Ober et al., 2006; Poulain and Ober, 2011), while TGF-β signaling has an opposing effect.

Figure 5.

TGF-β1 promotes pancreas over liver fate by inhibiting BMP and Wnt signaling

(A) Expression of AFP, FABP1, APOA1, and TTR by RNA-seq in PDX1^EGFP/+ HUES4 cells treated with vehicle (Ctrl), IWP-L6, TGF-β1, or IWP-L6 + TGF-β1 (I + T) during S2 (D5) or S2-S3 (D7). Mean ± SEM (n = 3 from independent experiments). ^∗padj. < 0.05, ^∗∗padj. < 0.005, ^∗∗∗∗padj. < 0.0001.

(B) GSEA plots comparing PDX1^EGFP/+ HUES4 gene expression data from each (S2–S3) treatment at D7 with a gene set for liver-specific genes (Hsiao et al., 2001). Normalized enrichment score (NES) and p values are shown.

(C) GSEA plots comparing PDX1^EGFP/+ HUES4 gene expression data from each (S2-S3) treatment at D7 with genes highly expressed in human Carnegie stage (CS)13 dorsal pancreas (DP) versus hepatic cords (HC) and genes highly expressed in HC versus DP (Jennings et al., 2017).

(D and E) Expression of BMP ligands and secreted BMP antagonists (D) and BMP target genes (E) by RNA-seq in PDX1^EGFP/+ HUES4 cells treated with vehicle (Ctrl), IWP-L6, TGF-β1, or IWP-L6 + TGF-β1 (I + T) during S2 (D5) or S2–S3 (D7). Mean ± SEM (n = 3 from independent experiments). ^∗∗padj. < 0.005, ^∗∗∗padj. < 0.0005, ^∗∗∗∗padj. < 0.0001.

(F) IF staining for AFP (white) in PDX1^EGFP/+ HUES4 cells at D7 in response to Wnt3a, RSPO3, or Wnt3a + RSPO3 stimulation during S2–S3 in cells treated with vehicle, IWP-L6, or TGF-β1. Scale bar, 50 μm.

(G and H) Real-time qPCR analysis of AFP, FABP1 (G), LEF1, and RSPO3 (H) expression at D7 in response to Wnt3a, RSPO3, or Wnt3a + RSPO3 stimulation during S2–S3 in PDX1^EGFP/+ HUES4 cells treated with vehicle (Veh), IWP-L6, or TGF-β1. Mean ± SD (n = 3 from independent experiments). ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗p < 0.0005, ^∗∗∗∗p < 0.0001.

(I) Expression of WNT11, LEF1, RSPO3, and RNF43 by real-time qPCR in response to replacement of LDN193189 with BMP4 during S3 in D7 PDX1^EGFP/+ HUES4 cells treated with vehicle (Veh), IWP-L6, or TGF-β1. Mean ± SD (n = 3 from independent experiments). ^∗p < 0.05, ^∗∗p < 0.005.

(J) IF staining for AFP (white) in response to replacement of LDN193189 with BMP6 during S3 in D7 PDX1^EGFP/+ HUES4 cells treated with vehicle, IWP-L6, or TGF-β1. Scale bar, 50 μm.

(K) Western blot analysis for AFP and PDX1 with Vinculin as internal control in D7 PDX1^EGFP/+ HUES4 cells cultured in standard S3 medium (Ctrl), S3 medium without LDN193189 (w/o LDN) or with replacement of LDN193189 with either BMP4 or BMP6 during S3, and treated with vehicle (Veh), IWP-L6 (IWP), or TGF-β1 (TGF). Also shown are hESC (hES) and definitive endoderm (DE) stages.

(L) Real-time qPCR analysis of AFP and FABP1 expression in D7 PDX1^EGFP/+ HUES4 cells cultured in standard S3 medium (Ctrl), S3 medium without LDN193189 (w/o LDN) or with replacement of LDN193189 with BMP4 during S3, and treated with vehicle (Veh), IWP-L6 or TGF-β1 (TGF-β). Mean ± SD (n = 3 from independent experiments). ^∗p < 0.05, ^∗∗p < 0.005.

See also Figure S5.

To understand how TGF-β signaling inhibits liver specification, we examined our RNA-seq data for changes in BMP and Wnt signaling, known liver-promoting pathways. First, we found the classical BMP ligands BMP2, BMP4, BMP5, BMP6, and BMP7 to be expressed at moderate to high levels in controls (Figure S5E). TGF-β1 significantly suppressed expression of BMP5 and BMP6 and strongly induced the BMP antagonists CER1, FST, and NOG (Figures 5D and S4E). In contrast, IWP-L6 had no effect on these genes with the exception of BMP5. Consistently, target genes ID2, ID4, and BAMBI were downregulated by TGF-β1 but not by IWP-L6 (Figure 5E). Thus, TGF-β1-induced downregulation of liver markers may be mediated, at least partly, through BMP signaling suppression.

To test whether reduced Wnt ligand availability might underlie IWP-L6- and TGF-β1-mediated liver suppression, we co-treated cells with recombinant Wnt3a and/or RSPO3 during S2–S3 and assayed AFP and FABP1 expression. As expected, IWP-L6-mediated liver marker inhibition at D7 was fully reversed by co-treatment with Wnt3a with or without RSPO3 but not by RSPO3 alone (Figures 5F and 5G). However, Wnt3a, with or without RSPO3, did not prevent TGF-β1-induced inhibition of AFP and FABP1 expression, indicating that suppression of Wnt ligand expression cannot fully account for the effect of TGF-β1 on liver differentiation at D7. As also seen at D5, Wnt3a/RSPO3 treatment failed to induce Wnt target gene expression in the presence of TGF-β1 (Figure 5H), showing TGF-β1 to block Wnt signaling downstream of ligand availability. This may be due to upregulation of the antagonists SFRP1/2 and SHISA2/4 at D7 (Figure 4B).

Lastly, we asked whether BMP signaling influenced TGF-β1-mediated suppression of Wnt signaling and liver differentiation. Notably, BMP signaling increased expression of WNT11 and Wnt targets LEF1, RSPO3, and RNF43, and both IWP-L6- and TGF-β1-mediated suppression of these genes at D7 was partially rescued by replacing the BMP receptor inhibitor LDN193189 with BMP4 (Figure 5I). Strikingly, omitting LDN193189 and/or adding exogenous BMP4 or BMP6 counteracted IWP-L6- and TGF-β1-induced suppression of AFP and FABP1 expression at D7 (when PDX1 expression is not yet upregulated by TGF-β1), albeit the effect was most pronounced for IWP-L6 (Figures 5J–5L). These data suggest that TGF-β1-mediated BMP signaling suppression may contribute to reduced Wnt signaling by attenuating Wnt ligand expression.

TGF-β1 and Wnt inhibition increase efficiency and reduce batch-to-batch variation of INS⁺ cell differentiation

As TGF-β1 and/or Wnt inhibition during S2–S3 promote specification of pancreatic progenitors, we tested whether they influence their downstream differentiation into INS⁺ cells. HUES4 and H1 cells were treated with TGF-β1 and/or IWP-L6 during S2–S3 and differentiated toward immature β cells as per the control pancreas differentiation protocol used (Rezania et al., 2014) until D20 (S6) (Figure 1A). D11 cells were transferred from planar culture to an air-liquid interface by spotting onto 6-well filter inserts, forming cell clusters. At D20, clusters were analyzed by IF for INS and PDX1. In our hands, the unmodified (vehicle control) pancreatic differentiation protocol yielded obvious differences in INS⁺ cell numbers per cluster between replicate differentiations for both lines. Of three independent differentiations, for each line, one (differentiation C) yielded more abundant INS⁺ cells for all control-treated clusters than differentiations A and B (representative clusters, Figure 6; all analyzed clusters, Figure S6). Treatment with TGF-β1 and/or IWP-L6 yielded uniformly highly abundant INS⁺ cells in all replicate differentiation experiments. Therefore, clusters from distinct differentiations segregated into two classes: type 1 (differentiations A and B) where INS⁺ cells with all treatments appeared more prevalent than in control (H1 or HUES4) differentiations (left panels, Figures 6A, 6B, and S6) or type 2 (differentiation C) where INS⁺ cells in control clusters were more widespread than in type 1 control-treated clusters and in which there was a less obvious difference in frequency of INS⁺ cells compared to treated ones (right panels, Figures 6A, 6B, and S6). Colocalization of INS with the mature β cell marker PDX1 varied across all clusters and differentiations (Figures 6, S7A, and S7B). As PDX1 marks both mature β cells and pancreatic progenitors, normalization of INS⁺ area to PDX1⁺ area gives INS⁺ cell yield as a proportion of endocrine-competent pancreatic progenitors. For both hESC lines IWP-L6 and/or TGF-β1 yielded a significant increase in both INS⁺ cells and PDX1⁺ cells as a proportion of total (DAPI⁺) cells as well as INS⁺ cells as a proportion of PDX1⁺ pancreatic cells at D20 (Figures 7 and S7C). For most differentiations, TGF-β1 produced a greater increase in INS⁺ cells and PDX1⁺ cells than IWP-L6 while combined treatment gave an additive effect, further augmenting yield. When quantifications were segregated by differentiation, this pattern was enhanced in type 1 (differentiations A and B) clusters in which INS⁺ cell and PDX1⁺ cell yield was lower under control conditions than in differentiation C type 2 clusters (Figure 7B). Crucially, calculation of the coefficient of variation of INS⁺ and/or PDX1⁺ proportion per cluster for both lines combined showed that IWP-L6 dramatically reduced this metric across differentiations compared to control conditions with TGF-β1 reducing it even more while combined treatment further augmented this effect, reducing the coefficient of variation to a negligible value (Figure 7C). Thus, across differentiations and hESC lines, IWP-L6 and TGF-β1 treatment during S2–S3 potently and additively expands yield of INS⁺ cells and PDX1⁺ cells and yield of INS⁺ cells relative to PDX1⁺ cells, increasing pancreatic progenitor differentiation toward an INS⁺ β cell-like fate.

Figure 6.

TGF-β1 and Wnt inhibition during foregut and pancreatic specification robustly promote downstream INS⁺ cell differentiation

(A and B) IF staining for INSULIN (INS; red) and PDX1 (green) with DAPI counterstaining (white) of D20 (S6) PDX1^EGFP/+ HUES4 (A) or H1 (B) cells cultured at the air-liquid interface on 6-well filter inserts from D11 following treatment with vehicle, IWP-L6, TGF-β1, or IWP-L6 + TGF-β1 during S2–S3. Under control conditions, differentiations A and B for each cell line yielded INS⁺ cells with lower efficiency (type 1 cell clusters: left) than did differentiation C (type 2 cell clusters: right). Regions (i, ii) demarcated by dashed boxes in lower power images (upper rows) are shown at higher magnification below. Scale bar, 1 mm (upper panels) or 100 μm (lower panels).

See also Figure S6.

Figure 7.

TGF-β1 and IWP-L6 treatment during foregut and pancreatic specification additively and robustly increase INS⁺ cell differentiation efficiency across hESC lines

(A–C) Quantification of INSULIN (INS)⁺ area as a percentage of total cell cluster (DAPI⁺) area (upper row), PDX1⁺ area as a percentage of DAPI⁺ area (middle row) or INS⁺ area as a percentage of PDX1⁺ area (lower row) in D20 (S6) PDX1^EGFP/+ HUES4 or H1 cell clusters following treatment with vehicle (V), IWP-L6 (I), TGF-β1 (T), or IWP-L6 + TGF-β1 (I + T) during S2-S3. Data segregated by experiment/differentiation (differentiations A and B lower efficiency of INS⁺ cell differentiation under control conditions than C) and cell line (A), by cluster type with type 1 clusters (left) from differentiations A and B and type 2 clusters (right) from differentiation C for each cell line (B) or by treatment during S2–S3 with coefficient of variation (CoV %) for all pooled cell clusters for each treatment (right) (C). PDX1^EGFP/+ HUES4 cell cluster numbers analyzed for differentiations A, B (type 1 cell clusters), and C (type 2 cell clusters), respectively: Vehicle: n = 3, 2, 4; IWP-L6: n = 3, 3, 4; TGF-β1: n = 3, 2, 4; IWP-L6 + TGF-β1: n = 3, 3, 4 and for H1: Vehicle: n = 2, 2, 6; IWP-L6: n = 3, 2, 5; TGF-β1: n = 3, 3, 5; IWP-L6 + TGF-β1: n = 2, 3, 7 (all analyzed cell clusters shown in Figure S6). Mean ± SD, one-way ANOVA with Tukey’s multiple comparison test, ^∗p < 0.05, ^∗∗p < 0.005, ^∗∗∗∗p < 0.0001 (B–C).

See also Figure S7.

Discussion

Here we show that TGF-β signaling anteriorizes hESC-derived DE and then promotes pancreas fate over liver via inhibition of Wnt/β-catenin and BMP signaling in foregut progenitors.

TGF-β/Nodal-induced Smad2/3 signaling promotes anterior fate in mice by inducing expression of Eomes that directly activates Foxa2 and Lhx1, which, with Otx2, activates numerous anterior ME genes as well as negative regulators of BMP-, Nodal-, and Wnt/β-catenin signaling (Costello et al., 2015; Fossat et al., 2015; Ip et al., 2014; Sibbritt et al., 2018). We observed strong induction of EOMES, GSC, LHX1, and OTX2 and other anterior TFs after TGF-β1 treatment at S2, indicating anteriorization of the endoderm formed at S1. Many of these TFs were also induced by Wnt inhibition, as expected when blocking the posteriorizing effect of Wnt signaling. These findings are best explained by suppression of BMP, Nodal, and Wnt signaling at several levels including repression of ligand expression and activation of antagonists such as CER1, FST, LEFTY1/2, SFRP1, and SHISA2/3. These antagonists are known targets of the Otx2/Lhx1 GRN in mouse and Xenopus (Costello et al., 2015; Fossat et al., 2015; Sibbritt et al., 2018) and possibly human as indicated by our analysis of published ChIP sequencing (ChIP-seq) data (Tsankov et al., 2015). Our attempts to test this via ChIP-seq analysis of LHX1 in human cells were hampered by a lack of compatible antibodies. The mechanism underpinning suppression of BMP and Wnt ligand expression is unresolved, but GSC is a strong candidate as a direct repressor of ligand genes as it is induced by TGF-β1 in our cells and serves such a role in vivo, with OTX2 (Yasuoka et al., 2014). Co-binding of OTX2 with LHX1 or GSC likely determines activation or repression of OTX2 target genes, respectively, in Xenopus anterior development. SIX3, expressed in anterior neural plate, hESC-derived ME, and early mouse DE (Shim et al., 2020), may also be involved. It is induced by TGF-β1 in our cells and represses BMP and Wnt ligand expression in anterior neural plate (Gestri et al., 2005; Lagutin et al., 2003). ChIP-seq analyses of GSC and SIX3 in hESC-derived DE will test this.

While the identity of the endogenous posteriorizing Wnt ligand(s) is unconfirmed, we propose WNT11 as the key candidate as: 1) WNT11 is the highest expressed ligand in our cultures, 2) WNT11 and RSPO3 are strongly suppressed by TGF-β1, and 3) Wnt11 acts in Xenopus foregut patterning; Wnt11 activity must be suppressed by Sfrp5 in anterior endoderm to maintain anterior foregut endoderm identity (Li et al., 2008; McLin et al., 2007).

Liver markers were suppressed by both TGF-β1 and IWP-L6/XAV-939 from D5 onwards. Conversely, pancreatic differentiation was promoted, but only from D10 onwards by TGF-β1. Analysis of endogenous signaling activity (D5 and D7) confirmed suppression of Wnt signaling by IWP-L6 and suppression of both Wnt and BMP signaling by TGF-β1. TGF-β and BMP pathways often cross-repress each other, via mechanisms sometimes involving Smad4 sequestration (Candia et al., 1997; Galvin et al., 2010). Consistent with our work, a reciprocal relationship between TGF-β and BMP in liver vs. pancreas induction has been reported in hESCs (Loh et al., 2014), but no mechanism ascribed. Our RNA-seq data suggest that TGF-β1 may suppress BMP signaling in foregut progenitors by repressing BMP6 expression and inducing BMP antagonists (NOG, FST, and CER1). While BMP signaling promotes liver development in vivo (Chung et al., 2008; Rossi et al., 2001), a mechanism involving BMP signaling suppression may appear surprising as the BMP receptor inhibitor LDN193189 is present in S3 medium. However, as LDN193189 is only added at S3, it may leave time for endogenously produced BMPs to act at S2. Furthermore, LDN193189 only inhibits BMPR1B ∼50% at the concentration used (Sanvitale et al., 2013).

Notably, Wnt inhibition promoted pancreas differentiation, while suppressing liver differentiation, and TGF-β1 also suppressed Wnt signaling at D7. In zebrafish, Wnt2 and Wnt2bb act via Fzd5 to activate Wnt/β-catenin signaling and promote liver development (Ober et al., 2006; Poulain and Ober, 2011). Whether Wnt/β-catenin signaling promotes mammalian liver development is less clear. Non-canonical Wnt signaling was reported to favor pancreas over liver fate, but no evidence of Wnt/β-catenin signaling promoting liver fate was found in this study (Rodriguez-Seguel et al., 2013). However, Wnt/β-catenin signaling promotes liver fate in hESC-derived foregut progenitors (Mahaddalkar et al., 2020); concordantly, we could rescue liver fate in IWP-L6-treated cultures with Wnt3a. In our hESC cultures, the two most highly expressed Wnt genes are WNT5A and WNT11, thus both candidate liver inducers, likely augmented by RSPO3. Like Wnt5a, Wnt11 is often considered a non-canonical ligand, but both can also activate the canonical pathway (Li et al., 2008; Mikels and Nusse, 2006; Tao et al., 2005). While a putative liver-specifying Wnt ligand in vivo remains unidentified, candidates include Wnt2, Wnt2b, and Wnt5a, all expressed in mesoderm proximal to the developing liver (Han et al., 2020; McMahon and McMahon, 1989; Monkley et al., 1996; Rodriguez-Seguel et al., 2013; Zakin et al., 1998). Mouse Wnt2/Wnt2b compound mutants exhibit normal liver and pancreas development (Goss et al., 2009), but additional Wnt ligands (Wnt5a) have been suggested to specify liver in mammals (Poulain and Ober, 2011). Notably, the canonical co-receptor Lrp5 is enriched in liver progenitors compared to pancreas, further suggesting that Wnt/β-catenin signaling may promote liver fate. A contributing factor to TGF-β1-mediated suppression of liver fate may be the upregulation of ROR2 expression. In mouse foregut, Ror2 is expressed in pancreas, but not liver, progenitors and, as mentioned, non-canonical signaling favors pancreas over liver fate (Rodriguez-Seguel et al., 2013).

As also seen at D5, TGF-β1 suppressed Wnt activity at D7 even with exogenous Wnt ligand present, possibly by stimulating SFRP1/2 and SHISA2/4 expression at D7. CER1 is also strongly induced, but human CER1 (∼69% identical to mouse Cer1) may not antagonize Wnt signaling (Belo et al., 2000; Piccolo et al., 1999). Importantly, omitting LDN193189 from S3 medium, or replacing it with BMP ligand, reactivated WNT11 and Wnt target gene expression and partially restored liver differentiation in the presence of TGF-β1. This suggests that BMP signaling acts upstream of Wnt signaling and that TGF-β1-induced suppression of Wnt signaling could be mediated, at least partly, by inhibition of BMP-stimulated Wnt activity. Consistently, Bmp2b acts prior to Wnt2bb and Wnt2 in zebrafish liver development (Chung et al., 2008; Poulain and Ober, 2011). Our results show that BMP- and Wnt/β-catenin signaling coordinate induction of liver lineage in human progenitors and suggest mechanisms for TGF-β1-mediated suppression of both pathways that promote pancreatic over liver fate.

While the unmodified pancreatic differentiation protocol used (Rezania et al., 2014) induced DE with comparable efficiency in both hESC lines, HUES4 showed lower pancreatic differentiation bias, with a stronger tendency to differentiate into posterior endodermal fates (liver, intestine). Higher levels of endogenous non-canonical Wnt signaling in HUES4 cells during pancreatic specification might underlie this bias, underscoring how differences in levels of endogenous signaling activity impact differentiation propensity and thus outcome in a given protocol. This has received scant attention to date and will require further work to decipher. Regardless of inherent differentiation bias, exogenous TGF-β1 and Wnt inhibition function similarly in the two lines. Transcriptional response to treatments was, however, larger in HUES4, consistent with stronger effects needed to overcome their higher “baseline” Wnt signaling activity than H1.

Pancreatic differentiation propensity has not, to our knowledge, been directly compared between unmodified HUES4 and H1 lines. While it is conceivable that PDX1 haploinsufficiency in the PDX1^EGFP/+ HUES4 reporter line might impair responsiveness to pancreatic differentiation protocols, no impairment was reported in the ability of a PDX1-heterozygous hiPSC line to differentiate into PE using a modified version of the differentiation protocol used here (Wang et al., 2019). Concordantly, INS⁺ cell yields were comparable between PDX1^EGFP/+ HUES4 and PDX1-WT H1 lines indicating minor if any impairment in pancreatic and later INS⁺ cell differentiation ability in PDX1^EGFP/+ HUES4 cells.

Strikingly, exogenous TGF-β1 and Wnt inhibition during foregut and pancreas specification not only stimulate differentiation of pancreatic progenitors but additively and reproducibly promote their differentiation into INS⁺ cells. Wnt inhibition increased D20 INS⁺ cell yield and reduced batch-to-batch variation and TGF-β1 more so, and the combination even further augmented both effects. This is consistent with TGF-β signaling being upstream of Wnt signaling in governing pancreatic and INS⁺ cell differentiation but also points to TGF-β1 regulation of Wnt-independent targets such as components of the BMP signaling pathway. Changing effects of combined IWP-L6 and TGF-β1 treatment vs. either alone throughout the pancreas differentiation protocol (for example, synergism at D5 but not at D7 and additive effects by D20) imply varying targets dependent upon exact cellular context and stage of differentiation. Identifying these targets will provide new avenues for further refining directed differentiation protocols.

Experimental procedures

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Philip A. Seymour (philip.seymour@sund.ku.dk).

Materials availability

This study did not generate new unique reagents.

Data and code availability

RNA-seq data generated in this study are available at the ArrayExpress database under accession number E-MTAB-10715. Code for differential expression analysis is available at https://github.com/kristianHdL/FunaEtAl2024.

Cell lines and culture conditions

The human PSC lines H1 (WA01, WiCell; RRID: CVCL_9771) and PDX1^EGFP/+ HUES4 obtained from our facility (Ameri et al., 2017) were maintained in DEF-CS 500 culture media (Takara Biosciences #Y30010) following the manufacturer’s instructions with daily media change and passaged every 2–4 days with TrypLE express enzyme (1X, no phenol red) (GIBCO/Thermo Fisher Scientific # 12604039). All cells were cultured in a humidified 37°C, 5% CO₂ incubator.

Differentiation of hESCs in chemically defined conditions

PDX1^EGFP/+ HUES4 and H1 hESCs were differentiated to pancreatic progenitor cells and, ultimately, immature β cells via a previously described protocol (Rezania et al., 2014) with minor modifications. 150,000 cells/cm² were seeded in 6- or 24-well multi-well plates or in 8-well Ibidi chamber slides after single-cell suspension for 3–5 min at 37°C in TrypLE express enzyme. After 48 h, at D0 (∼90% confluency), cells were washed once in DPBS without Mg²⁺ and Ca²⁺ (Thermo Fisher Scientific) before addition of S1 basal media (MCDB 131 medium, no glutamine: Gibco/Thermo Fisher Scientific # 10372019) supplemented with sodium bicarbonate (1.5 g/L, Sigma-Aldrich #S6297), glucose (10 mM, Sigma-Aldrich #G8769), GlutaMAX (1x, Gibco/Thermo Fisher Scientific # 35050038), BSA (0.5%, Proliant Biologicals # 68700), CHIR-99021 (3 μM, Axon Medchem # 2435), and recombinant human (rh) Activin A (100 ng/mL, PeproTech # 120-14E). S1 cells (D3) were washed once in DPBS without Mg²⁺ and Ca²⁺ and further supplemented for 2 days with S2 media including sodium bicarbonate (1.5 g/L), glucose (10 mM), GlutaMAX (1x), BSA (0.5%), L-ascorbic acid (0.25 mM, Sigma-Aldrich # A4403), and rhFGF7 (KGF) (50 ng/mL, PeproTech # 100-19). During S3-S4, the media was supplemented with sodium bicarbonate (2.5 g/L), glucose (10 mM), GlutaMAX (1x), BSA (2%), L-ascorbic acid (0.25 mM), KGF (S3: 50 ng/mL; S4: 2 ng/mL), RA (S3: 1 μM; S4: 0.1 μM, Sigma-Aldrich #R2625), SANT-1 (0.25 μM, Selleckchem #S7092), TPB (S3: 200 nM; S4: 100 nM, Sigma-Aldrich # 565740), LDN193189 (S3: 100 nM; S4: 200 nM, Axon Medchem # 1509), and ITS-X (1:200, Gibco/Thermo Scientific # 51500056). During S5–S6, the media was supplemented with sodium bicarbonate (1.5 g/L), glucose (20 mM), GlutaMAX (1x), BSA (2%), RA (S5 only: 0.05 μM), SANT-1 (S5 only: 0.25 mM), LDN193189 (100 nM), T3 (1 μM, Sigma-Aldrich #T6397), ALK5i-II (10 μM, Millipore # 616452), zinc sulfate (10 μM, Sigma-Aldrich #Z0251), heparin (10 μM, Sigma-Aldrich #H3149), ITS-X (1:200), and 100 nM gamma secretase inhibitor XX (S6 only: 100 nM, Sigma-Aldrich # 565789). In the initial screen, the media was further supplemented with the following factors during S2–S5: rhActivin A (100 ng/mL), rhTGF-β1 (10 ng/mL, PeproTech # 100-21), SB-505124 (10 μM, Sigma-Aldrich #S4696), DAPT (10 μM, Selleckchem #S2215), CHIR-99021 (3 μM), IWP-L6 (5 μM, Axon Medchem # 2212), PD98059 (1 μM, Selleckchem #S1177), rhHGF (50 ng/mL, PeproTech # 100-39), rhEGF-L7 (50 ng/mL, PeproTech # 100-61), rhWnt3a (100 ng/mL, R&D Systems # 5036-WN-010), or rhEGF (50 ng/mL, PeproTech # 100-15). In the following experiments, IWP-L6 (5 μM) and/or rhTGF-β1 (10 ng/mL) were added during S2–S3 unless otherwise indicated. When applicable, SB431542 (10 μM, Selleckchem #S1067), XAV-939 (1 μM, Sigma-Aldrich #X3004), rhBMP4/6 (50 ng/mL, PeproTech: BMP4: # 120-05ET; BMP6: # 120-06), rhWnt3a (100 ng/mL, R&D Systems), or rhRSPO3 (100 ng/mL, PeproTech # 120-44) were added during S2–S3 according to the experimental setup. For air-liquid interface cultures, D11 planar culture cells were treated for 4 h with Y-27632 (10 μM, VWR # 688000) and detached using TrypLE express enzyme. Following cell counting (NucleoCounter NC-202), cells were spun down and resuspended in S5 medium (+10 μM Y-27632) at 0.5 x 10⁵ cells/μL. 7-10 × 10 μL drops were then seeded per 6-well (3 μm pore size) Transwell filter insert (polycarbonate: Thermo Scientific/Nunc or polyester: Corning). 1.5 mL of S5 medium (+10 μM Y-27632) was added under each insert.

Public datasets

The ATAC-seq data (GSE149148; Geusz et al., 2021) and ChIP-seq data (GSE61475; Tsankov et al., 2015) used in this study were downloaded from Gene Expression Omnibus (GEO) as processed data in bigwig and bed formats.

Statistical analysis and reproducibility

All statistics were performed using GraphPad Prism 8 and 9 software (GraphPad). Datasets with two groups having equal variances were analyzed by a two-tailed Student’s t test. For data with unequal variances, two-tailed Welch’s t tests were applied. Comparison of three or more groups was performed by one-way analysis of variance (ANOVA) followed by either Dunnett’s test using control samples as reference or Tukey’s test for comparison of all means unless otherwise stated in figure legends. Statistics for differential expression of RNA-seq are provided as adjusted p (padj.) values from the DESeq2 analysis. p values are displayed in the figures, and sample sizes are provided in the figure legends. Statistical significance is defined as p < 0.05 for qRT-PCR and GSEA data as well as FACS and IF image quantifications, while the DESeq2 analysis used a padj. cutoff set to 0.1 (default).

Published: June 27, 2024

Supplemental information

Table S3. Enriched Gene Ontology (GO) terms

GO-terms enriched in upregulated genes at D5.

GO-terms enriched in downregulated genes at D5.

GO-terms enriched in upregulated genes at D7.

GO-terms enriched in downregulated genes at D7.

Table S4. Enriched genes from A-P pattern specification, cell cycle and liver-specific gene datasets

Genes enriched at D5 and D7 in the GO_ANTERIOR_POSTERIOR_PATTERN_SPECIFICATION dataset.

Genes enriched at D5 and D7 in the CHANG_CYCLING_GENES dataset.

Genes enriched at D5 and D7 in the HSIAO liver genes dataset.

Table S6. Enriched genes vs. human Carnegie Stage 13 dorsal pancreatic bud and hepatic cord gene datasets

Genes enriched at D5 by GSEA against human Carnegie Stage 13 (CS13) dorsal pancreatic buds (DP) vs hepatic cords (HC) and HC vs DP.

Genes enriched at D7 by GSEA against human Carnegie Stage 13 (CS13) dorsal pancreatic buds (DP) vs hepatic cords (HC) and HC vs DP.