Pluripotent stem cell differentiation reveals distinct developmental pathways regulating lung- versus thyroid-lineage specification

The in vitro differentiation of pluripotent stem cells (PSCs) into lineages that are otherwise difficult to access in vivo provides a novel source of cells for basic developmental studies, disease modeling and drug development. As with some primary cell types, transfer of these PSC-derived lineages into 3D culture systems at key developmental stages of differentiation has produced so-called ‘organoids’, 3D structures that begin to resemble the structural and cellular diversity of in vivo organs (Lancaster and Knoblich, 2014). Most published attempts to derive these differentiated cell types or structures from PSCs rely on in vitro recapitulation of known in vivo embryonic developmental signals; however, this approach can be problematic when the pathways regulating in vivo development of a particular tissue have not been established or appear to be poorly evolutionarily conserved across species. These hurdles are particularly apparent in prior attempts to generate lung epithelia from PSCs (Green et al., 2011; Hawkins and Kotton, 2015; Longmire et al., 2012; Mou et al., 2012). As the lung is an organ that emerged late in evolutionary time compared with other endodermally derived lineages, limited model systems based on embryos of lower species, most of which lack lungs, are available to study its developmental biology; therefore, reductionist mammalian in vitro model systems may help to examine the roles of individual germ layers or lineages in lung organogenesis. In particular, defining the minimal signaling pathways that specify a small group of progenitors in the anterior foregut endoderm into lung epithelial lineage, as marked by the onset of expression of Nkx2-1, has remained elusive.

In seminal in vitro work, Snoeck and colleagues used the Wnt signaling stimulator CHIR99021 (CHIR), together with FGF10, FGF7, BMP4, EGF and retinoic acid (RA), to direct the differentiation of PSCs into lung epithelial cells from anterior foregut endoderm (Green et al., 2011). This cocktail results in the acquisition of human lung cell fate and induction of NKX2-1 (Green et al., 2011; Huang et al., 2014). It differs significantly, however, from the growth factors employed in mouse models by us (Longmire et al., 2012) and others (Mou et al., 2012) to induce lung fate from mouse PSCs in culture, or from primary mouse foregut endoderm in explant models (Serls et al., 2005). A particularly dramatic and perplexing additional difference between species includes the observation that, in mouse PSC models, both lung and thyroid lineages, the two tissue types known to emerge via Nkx2-1⁺ endodermal progenitors, tend to emerge together during in vitro-directed differentiation (Longmire et al., 2012), whereas in human PSC models lung epithelia without contaminating thyroid lineages can be generated (Dye et al., 2015; Huang et al., 2014). Thus, although stimulation of Wnt, BMP and FGF signaling pathways has been a unifying, common theme in most prior reports of lung differentiation (Green et al., 2011; Huang et al., 2014; Longmire et al., 2012; Mou et al., 2012), an unsettled controversy remains regarding whether human lung lineage specification fundamentally differs from mouse, requiring different developmental signals, such as different FGF ligands (e.g. ligands of FGFR2IIIb, such as FGF10 or FGF7), compared with lower species such as mice, which have been claimed to require the more broadly active FGF ligands FGF1 or FGF2 (Serls et al., 2005). For example, FGF1 or FGF2, via ligation of FGFR4, have been reported to be necessary and sufficient in mouse foregut endodermal explant models as an inducer of lung fate (Serls et al., 2005); however, mice deficient in FGF2 or FGFR4 develop lungs normally in vivo (Guzy et al., 2015; Weinstein et al., 1998; Zhou et al., 1998). Mice deficient in FGF10 or FGFR2IIIb display lung agenesis (De et al., 2000) and instead form a trachea-like structure. Specification of respiratory progenitors has occurred in FGF10-null embryos, however, as it has been shown that the mutant tracheal endoderm can be induced to form Sftpc-expressing organoids in vitro (Hyatt et al., 2004). This suggests that these FGF signals may act post-specification in branching morphogenesis and formation of primary lung buds. In vivo models of Xenopus and mouse lung development have also demonstrated the necessity of BMP signaling (Domyan et al., 2011; Rankin et al., 2016) and Wnt signaling (Goss et al., 2009; Harris-Johnson et al., 2009) for normal early lung development, causing further uncertainty as to whether these are the minimal signals required for lung specification or whether coincident FGF or other signaling is also necessary (Serls et al., 2005).

Further complicating matters are recent reports using the human PSC model system that employ widely varying multifactorial cocktails to induce lung fate (Dye et al., 2015; Green et al., 2011; Huang et al., 2014; Mou et al., 2012; Rankin et al., 2016; Wong et al., 2012), obscuring the possibility of distinguishing the minimal essential factors that act intrinsically on developing endoderm to specify lung cell fate. For example, combinations of Wnt/CHIR, BMP4, RA, SHH, FGF2, FGF4, FGF7, FGF10 or FGF18 have all been employed in vitro to induce lung fate in human PSC model systems in these varying reports. Only one previous report has addressed the key pathways required for lung specification across species, including frogs, mice and humans (Rankin et al., 2016).

Since the minimal pathways regulating in vivo lung lineage specification as well as their evolutionary conservation remain controversial, we employ a reverse approach, using PSC in vitro model systems to identify the key signaling pathways regulating lung lineage specification from foregut endoderm. In contrast to most previous claims, these minimal pathways appear to be evolutionarily conserved between murine and human species, and are similar to those recently found to regulate early lung specification in Xenopus and mice (Rankin et al., 2016). Our model systems suggest that FGF signaling, which was previously thought to be required for lung-lineage specification (Longmire et al., 2012; Roszell et al., 2009; Serls et al., 2005), appears to be dispensable, consistent with the findings reported for human PSC in vitro differentiation by Snoeck and colleagues (Huang et al., 2014). Of the many candidate signaling pathways previously proposed to regulate lung specification, Wnt+BMP signaling (in the presence of RA) appears to be necessary and sufficient to specify lung progenitors from foregut endoderm, whereas FGF+BMP signaling promotes specification to thyroid, the only other endodermal organ domain known to express Nkx2-1. Importantly, lung or thyroid progenitor pools can be isolated by cell sorting for functional 3D culture maturation once specified from PSCs, demonstrating that the progenitors specified with these minimal factors are competent to produce lung or thyroid epithelial spheres, respectively, for basic developmental studies or future regenerative medicine.

To dissect the minimal factors needed to induce lung fate from anterior foregut endoderm, we used a previously developed in vitro model system in which mouse embryonic stem cells (ESCs) are differentiated into anterior foregut endoderm prior to differentiation into putative lung or thyroid cell fates (Kurmann et al., 2015; Longmire et al., 2012). The transcription factor Nkx2-1 is selectively expressed in lung and thyroid epithelial lineages within definitive endoderm, and is the first known marker to be expressed upon lung- or thyroid-lineage specification, making it an ideal reporter for tracking the commitment of endodermal precursors into lung and thyroid epithelial cell fates (Kurmann et al., 2015; Longmire et al., 2012). We have previously used mouse ESCs carrying fluorochrome reporters (GFP or mCherry) targeted to the endogenous Nkx2-1 locus to demonstrate that PSC-derived foregut endodermal cultures exposed to Wnt3a, BMP4 and FGF2 (in the presence of base media containing RA) upregulate both lung- and thyroid-lineage marker transcripts (Longmire et al., 2012). Cultures exposed to only BMP4 and FGF2, however, upregulate thyroid markers and show no robust expression of mature lung markers (Kurmann et al., 2015).

Using the mouse ESC line carrying an mCherry reporter targeted to the Nkx2-1 3′ untranslated region (hereafter Nkx2-1^mCherry) (Bilodeau et al., 2014; Kurmann et al., 2015), we employed our previously published differentiation protocol (Longmire et al., 2012; Kurmann et al., 2015) to derive anterior foregut endoderm over 6 days in serum-free culture, then tested the capacity of different combinations of Wnt3a, BMP4 and FGF2 to induce Nkx2-1⁺ from days 6-14 (Fig. 1A). Combinations of BMP4+FGF2, Wnt3a+BMP4, BMP4 alone or all three factors induced Nkx2-1^mCherry+ cells (Fig. 1B), whereas Wnt3a alone did not (Fig. S1). As previously published, FGF2 alone induced predominantly neuroectodermal Nkx2-1⁺ cells without evidence of lung-specific or thyroid-specific transcript expression (Longmire et al., 2012; Kurmann et al., 2015; and data not shown).

To test whether Nkx2-1⁺ cells produced under each condition contained lung or thyroid progenitors, we sorted each Nkx2-1^mCherry+ population on day 14 and replated Cherry⁺ cells in 2D cultures containing a serum-free FGF-supplemented medium (Fig. 1A,B) that we have previously shown to promote proliferation and outgrowth of Nkx2-1⁺ cells that co-express transcripts associated with lung and thyroid maturation (Longmire et al., 2012; Kurmann et al., 2015). Nkx2-1^mCherry+ cells specified with FGF2+BMP4 were competent to express markers of thyroid lineage differentiation by day 25 [Tg, Tpo, Tshr and Slc5a5 (Nis); Fig. 1D] and did not upregulate lung differentiation markers (Fig. 1D). In contrast, the sorted Nkx2-1^mCherry+ population specified with Wnt3a+BMP4 upregulated markers of airway and alveolar lung epithelial differentiation by day 25 (Sftpc, Sftpb, Scgb1a1 and Aqp5). Wnt3a+BMP4 induced Nkx2-1⁺ cells that were competent to differentiate into cells expressing ∼100,000-fold increased levels of Sftpc compared with our previously published Wnt3a+BMP4+FGF2 conditions. Although we have previously established that FGF2 is required for and promotes thyroid fate in developing foregut endoderm (Kurmann et al., 2015), these results unexpectedly implied that FGF2 may inhibit lung specification from endoderm in the mouse ESC model system during this narrow developmental window. Thus, we propose the foregut endodermal lineage specification model shown in Fig. 1E, in which lung and thyroid lineages specify distinctly in response to simple signals consisting of two exogenously induced signaling pathways (BMP4+Wnt3a versus BMP4+FGF2). We next sought to further test this model by assessing whether each distinct Nkx2-1⁺ endodermal population contained bona fide progenitors competent to form functional lung versus thyroid cells in 3D culture.

To test whether the putative lung versus thyroid Nkx2-1⁺ endodermal progenitors specified in each condition displayed competence to form functional epithelial structures, we sorted each PSC-derived Nkx2-1^mCherry+ population to purity on day 14 for replating in 2D versus 3D cultures without any supporting mesenchyme (Fig. 2). Each condition, respectively, generated a day 14 yield of 239±22 versus 931±67 Nkx2-1⁺ cells per starting day 0 ‘input’ ESC (Fig. 1C), although these yields and specification efficiencies were not adjusted for day 6 definitive endoderm frequencies. We have previously demonstrated that, following thyroid-lineage specification, 3D culture outgrowth conditions augment subsequent thyroid epithelial gene expression and allow spontaneous self-organization of Nkx2-1⁺ cells into epithelial monolayered structures that resemble primary thyroid follicular epithelium. Importantly, these PSC-derived thyroid follicles are capable of sequestering thyroglobulin and initiating thyroid hormone biogenesis both in vitro and after transplantation in vivo (Kurmann et al., 2015; Dame et al., 2017). We verified that Nkx2-1⁺ cells specified with BMP4+FGF2 formed thyroid follicular epithelia (Fig. 2D), but lacked lung epithelial differentiation capacity in 3D culture. In marked contrast, sorted Nkx2-1^mCherry+ cells specified with Wnt3a+BMP4 lacked thyroid competence (Fig. 1D), but exhibited lung differentiation potential, including the capacity to organize into monolayered epithelial spheres expressing a broad diversity of airway and alveolar epithelial genes (Nkx2-1, Scgb1a1, Scgb3a2, Foxj1, p63, Sftpa, Sftpb, Sftpc, Sftpd and Aqp5; Fig. 2A-E). 3D culture of the Wnt3a+BMP4-specified sorted Nkx2-1⁺ cells significantly augmented lung (but not thyroid) epithelial gene expression, while suppressing mesenchymal marker expression, compared with 2D culture conditions (Fig. 2E,F). These results were specific to the mCherry⁺ population as Nkx2-1^mCherry− cells sorted on day 14 for replating exhibited significantly lower sphere-forming capacity (Fig. 2A) and little, if any, detectable expression of lung marker genes (Fig. S2).

We sought to determine whether recognizable proximal and distal lung epithelial phenotypes were emerging in the 3D cultured outgrowths of sorted Nkx2-1⁺ cells. The day 14 sorted Nkx2-1⁺ cells replated for 3D outgrowth formed at least two morphologically distinguishable colonies after 3 to 5 days: larger spheroid-shaped colonies and smaller irregular colonies (Fig. 2A). Phase-contrast microscopy, immunofluorescence microscopy and Hematoxylin and Eosin staining of the spheroid-type colony revealed the presence of secretory SCGB1A1⁺ cells, as well as beating, multiciliated, tubulin⁺ epithelial cells facing the inner lumina of these spheres, suggestive of differentiation of proximal airway cell types (Fig. 2B,C; Movie 1). In contrast, we observed pro-SFTPC protein expression in cells within irregular smaller clusters of cells, suggestive of distal epithelial differentiation (Fig. 3C,E).

The two distinct colony morphologies we observed arising from Nkx2-1⁺ sorted cells appeared to resemble the spherical airway versus irregular alveolar colony morphologies previously reported to arise from primary mouse airway and alveolar lung epithelia, respectively (Bilodeau et al., 2014; Lee et al., 2014). To further test the hypothesis that irregular-type PSC-derived colonies corresponded to distal lung cell types, we engineered a lentiviral reporter of distal lung epithelial gene expression composed of a previously published 3.7 kb human SFTPC proximal promoter element (Glasser et al., 1991) driving expression of a GFP reporter cDNA (hereafter SftpcGFP; Fig. 3A). We have previously published the faithfulness and specificity of a SftpcdsRed version of this lentiviral vector to identify murine cells that express Sftpc (Longmire et al., 2012). To test faithfulness of this lineage-specific lentiviral reporter in the mouse ESC in vitro differentiation system, we first tested the SftpcGFP vector in our 2D ESC system. We sorted Nkx2-1^mCherry+ versus Nkx2-1^mCherry− cells on day 14 of ESC differentiation and on day 16-18 transduced each replated and sorted population with SftpcGFP lentivirus (Fig. 3A,B). GFP⁺ cells were observed arising from the Nkx2-1^mCherry+ sorted population as early as 24 h after transduction and appeared as distinct cell clusters that expanded over time in 2D culture (Fig. 3C). Few if any GFP⁺ cells emerged from the sorted Nkx2-1^mCherry− population (Fig. S2). On day 30, we sorted SftpcGFP⁺ cells (representing 1-8% of the total outgrowth cells), and found these cells were enriched in expression of alveolar epithelial transcripts (Sftpc, Abca3 and Lamp3) compared with presorted or SftpcGFP⁻ cells (Fig 3D, Fig. S2), whereas expression of the proximal secretory lung marker Scgb1a1 was lower in the GFP⁺ compared with the GFP⁻ population (Fig. 3D). We therefore concluded the SftpcGFP lentiviral vector was a faithful reporter of endogenous distal alveolar epithelial gene expression. Next, we embedded the Nkx2-1⁺ cells infected with SftpcGFP lentivirus in Matrigel drops on differentiation day 18 to allow for 3D organoid formation (Fig. 3C). We detected SftpcGFP⁺ cells only in the smaller irregularly shaped colonies, whereas the larger spherical colonies containing beating cilia were rarely found to contain SftpcGFP⁺ cells. Taken together, these results suggested that the sorted day 14 Nkx2-1⁺ population specified with Wnt3a+BMP4 contains lung progenitors competent to form proximal airway and distal lung epithelia.

Having shown that distinct lung-competent versus thyroid-competent Nkx2-1⁺ progenitors could be generated from PSCs using distinct growth factor combinations, we next sought to profile the genetic programs of each progenitor population on day 14 of differentiation. Global transcriptomic profiles of the four sorted populations were obtained by microarray analysis (mCherry⁺ versus mCherry^– sorted samples prepared in ‘lung’ versus ‘thyroid’ media; Fig. 4A,B). Nkx2-1 and the neighboring locus encoding the Nkx2-1 associated non-coding intergenic RNA [NANCI/LL18 (Herriges et al., 2014), also known as E030019B13Rik] were each in the top three genes differentially expressed between mCherry⁺ and mCherry^– populations in either medium [ranked by either fold-change (FC) or P-value adjusted for false discovery rates (FDR)]. To identify differentially expressed genes that distinguish the four populations, we used moderated t-tests to establish the effect of specification media, the effect of Nkx2-1^mCherry expression status and an ‘interaction effect’ between Nkx2-1^mCherry status and medium condition (Fig. 4C, see Table S1 and supplementary Materials and Methods). We found that 1315 significantly differentially expressed genes, grouped into nine clusters (Fig. 4D), distinguished the four populations (log₂ FC>2 between any two populations with interaction effect FDR<0.25). To assess whether this analysis revealed gene clusters known to be associated with either thyroid or lung, we searched for markers of lung or thyroid epithelia, as well as markers of lateral plate mesoderm or developing lung mesenchyme (Fig. 4C,D) (Grindley et al., 1997; Motoyama et al., 1998; Rankin et al., 2016; Sato et al., 2008; Zhang et al., 2013). We found that gene cluster 5 was uniquely enriched in the Nkx2-1^mCherry+ cells induced in ‘thyroid media’ (Fig. 4D). This cluster contained the core thyroid transcription factors Pax8, Foxe1 and Hhex, as well as the thyroid specific marker Tg. Furthermore, cluster 5 contained a variety of additional transcripts previously shown to be enriched in developing or adult thyroid epithelial cells, including Prlr (Fagman et al., 2011), Cd36 and genes known to be involved in thyroid hormone biogenesis, such as Iyd, Slc16a2 (also known as Mct8; Schwartz and Stevenson, 2007) and Duoxa2 (Grasberger et al., 2012) (Fig. 4E). Sftpb was also enriched in this cluster, a finding in keeping with the reported expression of this transcript in primary, as well as ESC-derived, thyroid epithelial cells (www.gtexportal.org/home/gene/SFTPB and Dame et al., 2017).

Genes known to be expressed in developing lung epithelial cells were found in cluster 2, which was enriched in the Nkx2-1^mCherry+ population induced in ‘lung media’ (Fig. 4D). This gene cluster included the transcription factors Foxa1 (Fagman et al., 2011; Rossi et al., 1999), Irx2 and Lef1, a known factor involved in Wnt signaling and lung development (Xie et al., 2014). Lung epithelial-selective differentiation markers Sftpc and Cldn18 (Schlingmann et al., 2015), and the lung-specific cytokine Cxcl15 (also known as ‘lungkine’; Rossi et al., 1999) were found in this cluster, as were other genes previously described as expressed in the lung: Gprc5a, Shh, Lama3, Lgi3 and Wif1 (Fagman et al., 2011; Xu et al., 2011) (Fig. 4E). Interestingly, some genes observed in cluster 1 (unique to Nkx2-1^mCherry− cells induced in ‘lung medium’) suggested potential enrichment of putative lung mesenchyme (Foxf1, Gli1, Wnt2 and Tbx4; Motoyama et al., 1998; Rankin et al., 2016; Zhang et al., 2013) or non-lung foregut endodermal derivatives, such as esophagus and liver (Pitx1/2 and Ttr). Cluster 6 comprised genes enriched in Nkx2-1^mCherry+ cells induced in both ‘thyroid’ and ‘lung’ media, and contained Nkx2-1, non-specific epithelial transcripts, such as Cdh1 and Itga6, as well as semaphorins and the surfactant homeostasis regulator Gpr116 (Yang et al., 2013). Taken together, our global transcriptomic analyses indicated that the genes found in each cluster were consistent with BMP4+Wnt3a and BMP4+FGF2 specifying early lung and early thyroid epithelial identity.

Confirming our microarray findings by RT-qPCR (Fig. 5), we found that day 14 putative primordial thyroid progenitors induced by FGF2+BMP4 expressed transcript markers of thyroid lineage specification, Hhex, Pax8 and Foxe1, and both putative lung and thyroid lineages expressed Cdh1, Nkx2-1 and Epcam. Additional transcripts that have been previously shown to distinguish early thyroid from early lung lineages in vivo in mice (Fagman et al., 2011) were also enriched in each respective PSC-derived lineage: Cd44 and Prlr in the putative thyroid population; and Slc15a2 and Foxa2 in the putative lung population. Shh was enriched in the lung and absent in the thyroid progenitors, a finding in keeping with recent work demonstrating Shh signaling is required for lung but is dispensable for thyroid lineage specification in vivo (Rankin et al., 2016). Furthermore, the Wnt signaling target gene Axin2 was enriched in PSC-derived lung progenitors and absent in thyroid progenitors. Even when thyroid progenitors were formed in the presence of Wnt3a, these cells appeared to be unresponsive to Wnt as Axin2 remained suppressed (Fig. 5). These results are consistent with the necessity of Wnt signaling in primordial Nkx2-1⁺ lung endodermal cells but its dispensability in thyroid cells (Kurmann et al., 2015).

Our in vitro ESC model system unexpectedly suggested that FGF2 reduces expression of all lung markers while increasing early and mature thyroid markers (Fig. 1). This suggested FGF signaling might be required to specify Nkx2-1⁺ endodermal cells with thyroid, but not lung, competence. To test this hypothesis, we performed experiments in mouse foregut explants. We have previously shown that chemical inhibitors of BMP and FGF signaling blocked Nkx2-1 and Pax8 expression in the thyroid domain in developing Xenopus and mouse foregut explants (Kurmann et al., 2015). We used the same approach here to explore the effect of BMP or FGF chemical inhibition on Nkx2-1 expression in lung and thyroid fields in mouse embryo foreguts and explant cultures (Fig. 6). Co-staining of FOXA2 and NKX2-1 revealed that NKX2-1 was first detected at 6-8 ss (somite stage) in forebrain (Fig. 6A) and was only weakly detected at 10 ss in the thyroid (not shown), with strong thyroid detection but no lung detection appreciated by 12 ss (∼E8.5; Fig. 6A). By E9, both thyroid and lung Nkx2-1⁺ domains were detectable, with lung being FOXA2⁺ and thyroid FOXA2⁻, consistent with previous reports (Fagman et al., 2011). To model the sequence of thyroid- and lung-lineage specification from endoderm, we developed a mouse foregut explant culture system, starting with foreguts harvested at 6-10 ss of development (prior to any endodermal Nkx2-1 expression) followed by a 2-day explant culture period during which both thyroid and lung Nkx2-1⁺ domains have been specified, as detected by NKX2-1 immunostaining (Fig. 6B, left panel). In the presence of the BMP inhibitor DMH1, both lung and thyroid specification was completely blocked and no NKX2-1 signal was detected within the ECAD⁺/SOX2⁺ developing foreguts (Fig. 6B, right panel). When 6-8 ss embryos were incubated for 48 h in three different inhibitors of FGF signaling (PD173074, BGJ398 or PD161570), we observed diminished NKX2-1 immunostaining in the thyroid but not the lung domain, in contrast to vehicle control-exposed embryos (Fig. 6C). Despite the presence of each FGF inhibitor, NKX2-1 expression in the lung domain was preserved; the thyroid domain, however, was smaller than in vehicle-exposed controls, and there was a reduced distance between the lung and thyroid domains. These results suggested FGF inhibition specifically perturbed thyroid, but not lung, primordium formation. We considered the possibility that FGF signaling to the endoderm at an earlier developmental stage might still be required for lung-lineage specification. Thus, we repeated our FGF blockade experiments, initiating blockade at the presomitic stage of endodermal development and continuing FGF inhibition through the stage of expected lung-lineage specification. In embryo explants cultured in the presence of each of the three FGF inhibitors, we observed no detectable NKX2-1 expression in the region of expected foregut thyroid, demonstrating that thyroid lineage specification was completely blocked; however, expression of NKX2-1 in the foregut endodermal lung domain was preserved (Fig. 6D). Taken together, these results suggest that FGF signaling is dispensable for lung lineage specification but required during a narrow developmental window for thyroid specification.

Next we sought to determine whether endodermal Nkx2-1⁺ foregut progenitors specified in the absence of FGF signaling, but in the presence of Wnt and BMP inducers, were bona fide lung progenitors competent to bud, branch and differentiate into lung epithelial cells. Our goal was to isolate mouse anterior endoderm, free of the presence of contaminating mesoderm, in order to determine whether stimulation of Wnt and BMP signaling intrinsically in definitive endoderm, despite the presence of FGF blockade, is sufficient to induce lung progenitors competent to branch and differentiate. We developed an embryonic explant culture system where anterior endoderm isolated at the E7.5 early head-fold stage from SftpcGFP transgenic mice could be patterned first into lung-competent endoderm by exposure to RA, as recently published (Rankin et al., 2016), followed by exposure to CHIR plus BMP4 during a 5-day culture period when the FGF inhibitor BGJ398 was always present (Fig. 6E). To determine whether lung-competent progenitors had been specified from endoderm by CHIR+BMP4 in the presence of FGF inhibition during this 5-day culture period, we then recombined the resulting endoderm for 6 days of culture with E12 mouse lung mesenchyme, a tissue we have previously demonstrated is able to induce branching and distal lung epithelial differentiation in developing lung epithelial cells (Shannon, 1994; Shannon et al., 1998). The endoderm rudiments cultured in the presence of CHIR, BMP4 and BGJ398 up to the time of recombinant culture, did not express the SftpcGFP reporter or exhibit any branching at the onset of recombinant culture, but subsequently underwent branching and upregulated expression of SftpcGFP after 6 days of recombination with E12 lung mesenchyme (Fig. 6E). These results indicate that primary definitive endoderm stimulated with RA followed by CHIR and BMP4 in the presence of FGF inhibition indeed contained lung progenitors competent to branch and differentiate into SftpcGFP⁺ lung epithelial cells. In order to confirm the requirement for canonical Wnt signaling in providing lung competence to the developing endoderm explant, we incubated the RA-treated endoderm with specification medium, either supplemented with CHIR or with control medium without CHIR, before recombining it with the lung mesenchyme. In comparison with CHIR-treated endoderm, which branched extensively and induced expression of the SftpcGFP reporter, endoderm never exposed to Wnt activation did not induce expression of the reporter. Although exogenous BMP addition or withdrawal in this model had no impact on lung specification (data not shown), the necessity of endogenous BMP signaling in the explant model was confirmed in separate experiments where addition of DMH1 blocked SftpcGFP expression (Fig. 6B; data not shown).

Next we questioned whether the same minimal signaling pathways are also required to specify lung-competent NKX2-1-expressing endoderm in human PSCs differentiating in vitro. A combination of CHIR, FGF10, KGF, BMP4 and retinoic acid (hereafter CFKBRA) has been demonstrated to induce NKX2-1 expression in anterior foregut cells derived in vitro from human ESCs and induced PSCs (iPSCs) (Green et al., 2011; Huang et al., 2014; Kurmann et al., 2015; Hawkins et al., 2017) (Fig 7A, Fig. S3). First, using human PSC lines engineered to carry a GFP reporter targeted to the NKX2-1 locus (hereafter NKX2-1^GFP) (Hawkins et al., 2017), we verified that CFKBRA treatment of PSC-derived foregut endodermal cells generated NKX2-1⁺ cells by days 14-15, and further differentiation in 3D Matrigel cultures gave rise to EPCAM⁺/NKX2-1⁺ spheres of epithelial cells with detectable SFTPC and SFTPB expression by RT-qPCR (Fig 7A-D, Fig. S3). We screened for the presence of any cells within the NKX2-1⁺ endodermal population generated with CFKBRA that might express thyroid markers (PAX8, FOXE1, HHEX or TG) by analyzing our published single-cell RNA-Seq dataset, profiling 153 individual iPSC-derived cells on differentiation day 15 (iPS17 and BU3 iPSC lines). Consistent with our findings in mouse Nkx2-1⁺ lung progenitors, we found the majority of human NKX2-1⁺ cells expressed high levels of FOXA2 and SHH, but few (if any) cells expressed thyroid markers PAX8, HHEX or TG, and no cells expressed FOXE1 (Fig 7E, Fig. S4), as expected in the absence of FGF2. Next, we tested the minimal exogenous factors required to induce lung-competent endoderm in human cells by withdrawing each of the five factors in CFKBRA one at a time from our lung-specification medium. Consistent with a prior report by Snoeck and colleagues (Huang et al., 2014), withdrawal of either KGF or FGF10 did not adversely impact the efficiency of specification of NKX2-1⁺ population by day 15 of differentiation (RUES2 or C17 PSC lines; Fig 7B, Figs S3, S4). However, withdrawal of CHIR produced virtually no NKX2-1⁺ cells (<1%) and withdrawal of BMP4 or RA had variable effects on the percentage of NKX2-1⁺, depending on the ESC or iPSC clone tested (Fig. 7B and data not shown). These findings are consistent with our mouse ESC model, as our base medium in all mouse ESC experiments contains RA. They are also in keeping with recent work by Zorn and colleagues demonstrating that RA is required to pattern foregut endoderm to become Wnt and BMP responsive, and therefore lung competent, in Xenopus, mouse and human development (Rankin et al., 2016). We found Wnt3a could not substitute for CHIR in our human PSC system, consistent with the previously reported low-level response of human cells to in vitro treatment with recombinant Wnt (Fuerer and Nusse, 2010).

We next tested whether a minimal three-factor putative ‘lung’ medium containing only CHIR, BMP4 and RA (CBRA) versus a modified ‘thyroid’ protocol involving medium containing FGF2+BMP4 but no CHIR could generate NKX2-1⁺ cells when added to human PSC-derived foregut-staged cells (Fig. S4D, Fig. S5). We observed that each specification medium (CBRA versus FGF2+BMP4) generated NKX2-1⁺ cells by day 15 of differentiation (Fig. 7C-H, Fig. S5), but lung versus thyroid competence was distinct in each condition. For example, we observed the three-factor CBRA media on average induced 52% of cells (range 24-95%) to express NKX2-1^GFP by day 16 (Fig. 7C,D; equivalent to a yield of 7.2±2.2 NKX2-1⁺ cells per day 0 ‘input’ iPSC); these sorted GFP⁺ cells expressed FOXA2 and SHH, and exhibited the competence, upon replating in 3D matrigel cultures, to give rise to 3D epithelial spheres expressing the distal alveolar epithelial markers SFTPB and SFTPC, and lamellar bodies (Fig. 7F-H), without any detectable expression of thyroid marker genes. This experiment was repeated in four iPSC lines (C17, BU3, RC204 and T4; Fig. 7D) and one ESC line (RUES2) with varying NKX2-1 differentiation efficiencies but similar results, indicating lung but not thyroid competence in response to CBRA (Fig. 7H). Comparing sorted BU3 NKX2-1^GFP+ progenitors induced head to head with either CFKBRA versus CBRA, we found no difference in their competence to subsequently upregulate SFTPC by day 31 of differentiation (Fig. S4D). In marked contrast, BU3 NKX2-1⁺ progenitors differentiated by day 16 in response to FGF2+BMP4 expressed low levels of SHH on day 16 and no detectable competence to upregulate SFTPC, but did express PAX8 and did display competence to upregulate thyroid-selective markers such as TSHR (Fig. 7H; Kurmann et al., 2015; Hawkins et al., 2017). These findings are consistent with our previously published observations that markers of thyroid lineage (PAX8, TG, NIS or TPO; studied in RUES2, C17 and BU3 lines) are expressed in outgrowths of human PSC-derived foregut endoderm after exposure to FGF2+BMP4 (Kurmann et al., 2015).

We considered the possibility that CHIR, which stimulates canonical Wnt signaling by stabilizing β-catenin via GSK3β inhibition, can potentially have off-target effects. Hence, we sought to further test the hypothesis that canonical Wnt signaling via activated β-catenin is required for human lung lineage specification in our human PSC endoderm model system. To block β-catenin interactions with potential effector-binding partners, we repeated our human lung differentiation protocol using CHIR, BMP4 and RA to differentiate RUES2 cells during the putative lung specification stage, but additionally included inhibitors of β-catenin-mediated interactions with either p300 (IQ1) or CBP (CREB-binding protein; ICG001), two distinct downstream effector branches of canonical Wnt signaling (Eguchi et al., 2005; Moheimani et al., 2015; Sasaki and Kahn, 2014) (Fig. 7I-K). Although cells treated with CHIR during lung endoderm specification showed induction of NKX2-1 in more than 60% of cells, addition of the ICG001 inhibitor significantly blocked induction of NKX2-1, whereas treatment with IQ1 did not. Similar results were also observed in C17 iPSCs (data not shown). These results suggest that CHIR in the presence of BMP4 induces lung lineage specification in the human model system via canonical Wnt signaling that may be transduced, in part, through the interaction of β-catenin with CBP (Fig. 7K). Thus, our results indicate that, as in our mouse explant and mouse PSC model systems, Wnt and BMP signaling promote lung-lineage specification from human endoderm.

In contrast to in vivo organogenesis, where differentiation proceeds in a highly orchestrated process, in vitro-directed differentiation of PSCs is often fraught with varying levels of contamination with undesired lineages and heterogeneity. Sorting strategies designed to isolate PSC-derived tissue-committed progenitor subsets can help to decrease this heterogeneity (Holtzinger et al., 2015). In the past, we have used Nkx2-1 reporters to help isolate lung and thyroid lineages derived from PSCs in culture (Longmire et al., 2012; Kurmann et al., 2015); the lack of complete specificity of endodermal Nkx2-1, however, has resulted in the derivation of mixed populations of lung and thyroid lineages, an expected result given the known in vivo expression of Nkx2-1 at the time of lineage specification of these two tissues. In previous reports where human PSCs were differentiated into lung lineages without contaminating thyroid cells, it was unclear whether this was due to inherent differences in the responses of human cells compared with mouse cells or to subtle differences in the complex growth factor-supplemented media being used (Huang et al., 2014; Dye et al., 2015). Here we have used in vitro PSC model systems to dissect the minimal pathways required for lung- versus thyroid-lineage specification, in order to resolve these controversies and to provide progenitors capable of giving rise to functional epithelial spheres that are not hampered by mixed thyroid and lung populations. Our results demonstrate that surprisingly simple media supplemented with only two or three growth factors are required to produce each distinct lineage, and the signaling pathways that regulate their lineage specification is evolutionarily conserved across species from mice to humans.

Our results demonstrate that, in foregut endoderm, either in embryos or in PSC-based models, the combination of Wnt+BMP signaling (in the presence of RA) promotes lung- rather than thyroid-lineage specification, whereas the combination of FGF+BMP signaling promotes thyroid specification. RA, which is included in our base medium throughout endodermal and lung-lineage specification, is also required for successful lung NKX2-1⁺ specification in our human model. However, as recently published, RA signaling appears to be dispensable at the moment of lung specification and is instead required at an earlier developmental stage during foregut endodermal patterning, in order to render the foregut competent to subsequently respond to Wnt and BMP signals (Rankin et al., 2016). Hence, we do not revisit this recently established role for RA signaling in lung development.

We were surprised to find that lung-lineage specification in our models did not require the addition of any FGFs. Indeed inhibition of FGF signaling did not appear to dampen lung-lineage specification in mouse foregut endoderm, implying that FGF signaling is dispensable at this developmental stage, although the inhibitor experiments were not repeated for the human ESC/iPSC model system. How can we reconcile our results with the previously published observation of Serls et al. (Serls et al., 2005) that FGF1 or FGF2 added to mouse foregut endoderm cultures appears to induce Nkx2-1 as well as Sftpc, whereas in our work FGF2 appears to inhibit lung specification in favor of thyroid specification? First, it should be emphasized that both Nkx2-1 and thyroglobulin were found to be induced by FGFs added to mouse endoderm in Serls et al.’s publication, raising the possibility that thyroid lineage specification was occurring, consistent with our results. In the years that have passed since the 2005 Serls et al. report, it has been established that canonical Wnt signaling (induced in vivo by Wnt2/2b secreted from lateral plate mesoderm) is required for lung-lineage specification (Goss et al., 2009). As Serls et al. only added FGFs to endoderm cultures that showed induction of Nkx2-1 and Sftpc, it appears likely that some mesodermal cells were likely present in their model. Contaminating mesenchymal cells might be able to respond to FGFs by secreting factors required for lung-lineage specification, such as Wnt. Our results suggesting that FGF signaling is dispensable for lung-lineage specification from foregut endoderm when Wnt signaling is present fit other published observations, such as: (1) genetic deletion of FGF ligands or FGF receptors in mice has not been shown to completely abrogate respiratory field specification (Weinstein et al., 1998; Zhou et al., 1998; De et al., 2000; Guzy et al., 2015); and (2) exogenous addition of FGF ligands to developing foregut Xenopus or mouse embryos is not required for lung-lineage specification (Rankin et al., 2016). FGF7 and FGF10, when added to stimulants of Wnt and BMP signaling in our model, did not significantly dampen lung-lineage specification or induce thyroid-lineage specification. This further emphasizes the different receptors that the various FGF ligands are known to bind to induce signaling, with broad ligation of FGF receptors occurring in response to FGF2, whereas FGF7 and FGF10 more specifically ligate the FGFR2IIIb form of the receptor.

Our finding that canonical Wnt signaling and BMP signaling are required for lung-lineage specification is consistent with genetic mouse models where deletion of Wnt 2/2b (Goss et al., 2009) or β-catenin (Harris-Johnson et al., 2009) results in lung agenesis, and deletion of BMP receptors in late foregut endoderm markedly reduces the Nkx2-1⁺ respiratory field (Domyan et al., 2011). Our mouse explant model where BMP signaling could be inhibited at an earlier endodermal developmental stage suggests that early BMP signaling is absolutely required for lung specification and is consistent with recently published findings in the Xenopus endodermal model system (Rankin et al., 2016).

Most importantly, our results suggest that the minimal signaling pathways that regulate cell fate decisions in lung and thyroid development are evolutionarily conserved between diverse species from mice to humans. Use of the mouse PSC model system, in concert with prior findings in Xenopus, was able to delineate the minimal signaling pathways (Wnt and BMP) required for lung specification; these pathways also appear to regulate human specification in the PSC model system. Although perturbations of human foregut in vivo at the time of lineage specification (3-4 weeks gestation) are not possible, our results imply that these same pathways may also be active in humans in vivo and serve as a strong argument in support of employing multi-species model systems, including the in vitro mouse PSC model for the discovery of human developmental mechanisms.

Our results do not necessarily indicate that lung and thyroid lineages are specified from the same foregut endodermal precursors. We cannot exclude the possibility that the various ligands being studied selectively enhance survival or proliferation of each distinct lung or thyroid progenitor pool rather than solely regulating their lineage specification. Single-cell profiling, limiting-dilution outgrowth cultures or clonal lineage-tracing studies in the future may help to define the heterogeneity of the Nkx2-1⁺ progenitor pool as well as the differentiation repertoire of individual foregut cells.

Overall, our results help to establish the minimal evolutionarily conserved pathways required for generating lung or thyroid progenitors from mouse or human PSCs, and thus provide precise control in regulating cell fate decisions of PSCs in cultures. These findings help to resolve long-standing controversies in the field of lung development and should now facilitate more reproducible derivation of an inexhaustible source of progenitors for basic developmental studies, lung or thyroid disease modeling, and testing of regenerative drug- or cell-based therapies.

The Nkx2-1^mCherry ESC line was generated and employed for in vitro differentiation as previously described (Bilodeau et al., 2014; Kurmann et al., 2015) (see supplementary Materials and Methods). Definitive endoderm induction was performed for 5 days to generate embryoid bodies (EBs) in serum-free medium as previously published with 50 ng/ml activin A (R&D Systems, 338-AC) added to the base medium from days 2.5 to 5 (Longmire et al., 2012; Kurmann et al., 2015). For anteriorization of endoderm, on day 5 (120 total hours of differentiation) EBs were plated onto P100 Petri dishes in Nog/SB media: cSFDM supplemented with 100 ng/ml rmNoggin (R&D Systems, 1967-NG) and 10 μM SB431542 (Sigma, S4317), as previously described (Longmire et al., 2012). For Nkx2-1⁺ endoderm induction, EBs were plated on gelatin-coated 6-well plates at the equivalent density of 200,000 cells/well, or cells in single-cell suspension obtained by trypsinization at 100,000 cells/well, in specification media: cSFDM supplemented with the factors stated in the text and detailed in the supplementary Materials and Methods. Nkx2-1^mCherry+ cells were sorted and replated onto gelatin-coated 24-well plates on day 12-14 at a density of 5×10⁴ cells/well. Further differentiation and maturation of Nkx2-1⁺ sorted cells was performed in either 2D or 3D culture conditions, as indicated in the text and detailed in the supplementary Materials and Methods.

The RUES2 human embryonic stem cell line (a gift from Dr Ali H. Brivanlou, The Rockefeller University, New York, USA) and previously published human iPSC lines [BU3 (Kurmann et al., 2015) or C17 (Crane et al., 2015)] were maintained and differentiated as detailed in the supplementary Materials and Methods.

The previously published SftpcdsRed lentiviral reporter vector (Longmire et al., 2012) was recloned as SftpcGFP as detailed in the supplementary Materials and Methods. Three-dimensional organoids were harvested by incubating with Cell Recovery Solution (Corning, 354253) for 1 h at 4°C, fixed with fresh 4% paraformaldehyde for 30 min at room temperature, and embedded in either 2% low melting gel agarose (Lonza SeaKem LE agarose, 50.000) or Richard-Allan Scientific HistoGel Specimen Processing Gel (ThermoFisher Scientific, HG-4000-012) according to the manufacturer's instructions, prior to embedding in paraffin. Sections were deparaffinized, rehydrated and stained for Hematoxylin and Eosin (AML Labs) using standard methods. For immunofluorescent staining, sections were incubated in antigen retrieval solution (Dako, S-1699) at 95°C for 20 min, cooled in same solution to room temperature for 30 min, and permeabilized/blocked with 0.25% Triton X-100 (Sigma, T-8787) and 4% normal donkey serum (NDS, Sigma, D9663) for 1 h at room temperature. Sections were incubated with primary antibodies in 4% NDS overnight at 4°C, washed and incubated with the corresponding Alexa Fluorophore-conjugated secondary antibodies for 30-60 min at room temperature. Nuclei were counterstained with DAPI (Invitrogen, 1:10,000) or Hoechst dye (ThermoFisher, 1:500). Fluorescent images were captured on a Nikon Eclipse Ni upright microscope using Nikon Elements D4.00 software or a Zeiss confocal microscope. Antibody information and sources, as well as methods of immunostaining for flow cytometry are detailed in the supplementary Materials and Methods.

All studies involving mice were approved by the Institutional Animal Care and Use Committee of Cincinnati Children's Hospital. Foregut explant or whole-embryo cultures prepared from developing mice were all performed as we have previously published (Kurmann et al., 2015) and are detailed in the supplementary Materials and Methods.

After reverse transcription, 40 cycles of PCR were performed to quantify gene expression, calculated based on 18S normalized expression levels represented as fold-change mRNA expression [2^−(ΔΔCt)] compared with undifferentiated (day 0) PSCs. Undetected genes after 40 cycles of PCR were arbitrarily assigned a Ct value of 40 to allow quantitative calculation of fold-change. Further details and primers used for all PCR reactions are provided in the supplementary Materials and Methods.

Three biological replicates for each indicated condition were processed on day 14 of differentiation for sorting of mESC-derived cells based on expression of the Nkx2-1^mCherry reporter by flow cytometry. RNA extracts from each sample were processed for analysis of gene expression by Affymetrix GeneChip Mouse Gene 2.0 ST arrays. Detailed analyses and statistical methods for normalization, principle component analysis, identification of differentially expressed genes by ANOVA, and unsupervised hierarchical clustering can be found in the supplementary Materials and Methods and Table S1, which lists all genes, samples, calculated fold-changes in expression and FDR-adjusted P-values. Raw data .cel files can be downloaded from the Gene Expression Omnibus database (accession number GSE92916).

To screen for expression of thyroid markers in individual day 15 cells prepared without the use of FGF2, we re-analyzed our previously published dataset (Hawkins et al., 2017) of unsorted BU3 cells and NKX2-1^GFP+ sorted C17 cells as follows: single-cell sequencing reads from 178 cells were aligned to the human genome (GRCh38) and quantified using STAR (Dobin et al., 2013). From these, 25 cells were discarded owing to having either abnormally high mitochondrial gene counts or abnormally low aligned reads, leaving 153 cells for further analysis (82 BU3, 71 C17). Statistical methods applied for normalization, clustering and heatmap generation are detailed in the supplementary Materials and Methods.

Data are presented as sample means and standard deviations (s.d.) with sample numbers stated specifically within the text or figure legends. Differences between groups were analyzed using unpaired two-tailed Student's t-test, one-way or two-way analysis of variance (ANOVA) with Tukey's multiple-comparison post hoc test as stated in the text or figure legends; P<0.05 was used to indicate significant differences between groups.

We thank the members of the Kotton lab and KIWI group for insightful discussions. We thank Adam Gower and Eduard Drizik of the Boston University CTSI Microarray and Sequencing Resource Core for Affymetrix array processing and bioinformatics support (CTSA grant UL1-TR001430). We thank Anne Hinds of the Boston University Pulmonary Center for histology technical support and Dylan C. Thomas of the CReM for laboratory technical support. We thank Brian R. Tilton of the Boston University Flow Cytometry Core Facility for technical assistance.