In order to test the capacity of pluripotent stem cells to undergo directed differentiation to DE, we used a 2-stage serum-free culture protocol, developed by Keller and colleagues (6
), to recapitulate the early stages of endodermal differentiation that occur in the gastrulating embryo (Figure A). To establish the differentiation kinetics of pluripotent stem cells, we first used a well-characterized control 129/Ola ES cell line that features reporter transgenes, GFP and hCD4, targeted to brachyury (T) and Foxa2 loci, respectively (13
). As previously published (13
), this cell line demonstrated that nodal-activin signaling directed the differentiation of pluripotent stem cells into cells, reminiscent of the embryo’s anterior primitive streak (defined by the phenotype T+
), followed by differentiation of these intermediates into DE (defined by the phenotype T–
; ref. 14
) within 6 days in culture (Figure , B–D). After this first differentiation stage, the resulting DE progenitors underwent lineage specification (stage 2 hepatic-inducing media; ref. 14
and Figure E) into cells expressing the initial transcriptional regulators or markers of primordial liver (albumin [Alb
]) and also expressing low levels of lung or thyroid (thyroid transcription factor 1 [Ttf1
]) and pancreatic and duodenal homeobox 1 (Pdx1
). The waves of gene expression that define the kinetics of this differentiation sequence were evident either when assessing cells purified at intermediate stages of differentiation by flow cytometry or in unfractionated populations maintained without any cell sorting (Figure E). However, purification of T+
anterior primitive streak–like cells on day 4 resulted in more sustained overall expression of the endodermal marker, Foxa2, from day 7–15 of the culture protocol, presumably due to decreased heterogeneity in the cultured progeny of sorted cells.
Kinetics of differentiation of ES cells into DE.
Clone-to-clone variability in capacity of ES and iPS cell lines to undergo directed differentiation to DE in vitro.
Based on the differentiation kinetics of ES cells, we selected day 5 of in vitro differentiation as an optimal time when the majority of cells in each culture have differentiated into DE but have not yet undergone lineage specification to liver, thyroid, or lung. Hence, we sought to compare the capacity of iPS cells versus ES cells to undergo directed differentiation to DE over this 5-day period. We selected 4 iPS cell lines for initial testing: we previously generated the ST5 and ST8 cell lines from postnatal tail-tip fibroblasts from a Sox2-GFP knockin mouse using a doxycycline-inducible single lentiviral stem cell cassette vector (Tet-STEMCCA; ref. 25
); an additional Oct4-GFP iPS cell line was generated with this vector from tail-tip fibroblasts taken from an Oct4-GFP knockin mouse (26
); and a well-characterized 2D4 cell line was generated previously from Nanog-GFP knockin mice using 4 retroviral reprogramming vectors (12
). Importantly, all 4 iPS cell lines tested in vivo were able to efficiently form all germ layers, including DE, in teratoma assays, in mouse chimeras after blastocyst transplantation, and (for ST8 and 2D4 lines) in second generation mice generated after germ line transmission (refs. 12, 25, and Supplemental Figure 1; supplemental material available online with this article; doi:
). Comparing these 4 iPS cell lines to 2 ES cell lines, we found that all 6 cell lines responded to the 5-day culture protocol by downregulating the expression of pluripotent transcriptional regulators (e.g., Rex1
, and Sox2
; Figure ) and upregulating the set of essential endodermal master transcriptional regulators, such as Foxa2
, and Gata6
(Figure ). Transcriptional regulators that are selectively active in other lineages, such as Pax6
for neuroectoderm and Sox7
for extraembryonic endoderm, were not upregulated over this 5-day period (Figure and Supplemental Figure 2A), further suggesting that directed differentiation preferentially to the DE germ layer was accomplished in all 4 iPS cell lines. However, we noted marked clone-to-clone variability in the magnitude of this endodermal response to activin across all cell lines tested.
Gene expression changes in nonisogenic ES and iPS cells undergoing directed differentiation to DE over 5 days.
We considered whether the observed variability in endodermal differentiation capacity between all tested ES and iPS cell lines might be due to (a) differences in genotypes between the 2 ES lines and 3 out of the 4 iPS cell lines, (b) low levels of variable leaky expression of the integrated reprogramming transgenes in the iPS cell clones (25
), (c) effects of haploinsufficiency of the different loci targeted to make the different knockin reporter lines (Nanog-GFP, Sox2-GFP, or Oct4-GFP), (d) heterogeneity of cell populations produced without the use of cell sorting, or (e) inherent biological differences in the epigenetic states of each cell line. Hence, to control for genotype or knockin effects, we tested the strain-matched Sox2-GFP ES cell line used to make the mice from which the ST5 and ST8 iPS cell lines were derived (26
). We found that Sox2-GFP ES cells responded to the entire 15-day endoderm differentiation protocol with slightly slower differentiation kinetics compared with those of 129/Ola ES cells of a different genetic background (Supplemental Figure 2B). Differentiation in parallel with these strain-matched Sox2-GFP ES cells versus ST5 and ST8 iPS cell lines revealed highly similar differentiation kinetics quantified by percentages of cells expressing established surface markers, CXCR4 and ckit, as well as a recently described endoderm-specific cell surface marker, ENDM1 (ref. 28
and Figure , A and B). Although ST8 iPS cells appeared to differentiate into endoderm slightly faster than ES or ST5 cells (Figure B; 2-way ANOVA, P
= 0.04), there was no statistically significant difference in the overall peak endodermal differentiation efficiency of each cell line, quantified by the percentage of cells reaching similar ENDM1 expression by day 6 (Figure B; ANOVA, P
= 0.08). All 3 cell lines showed the capacity to robustly proliferate in these conditions, although growth kinetics were slightly better for the parental ES cell line in 7 out of 8 repeated experiments (Figure B).
Comparison of strain-matched ES and iPS cell capacity to undergo directed differentiation to DE, followed by hepatic lineage specification.
Next we compared the capacity of both ES and iPS cell–derived putative multipotent DE progenitors to undergo further lineage specification in response to inductive signals (stage 2 differentiation; Figure C). Reminiscent of the sequence of differentiation observed in developing embryos, after stimulation of the putative ES and iPS cell–derived endodermal progenitors with a defined serum-free media supplemented with lineage specifying growth factors, including BMP4, FGF2, and HGF designed to favor hepatic lineage specification, we observed sequential induction of the early liver marker genes, α-fetoprotein (Afp
) and α-1 antitrypsin (Aat
), followed by induction of Alb
mRNA and Alb protein expression in both ES and iPS cell lines (Figure , C, D, and F). After 19 days of differentiation, the resulting cells also displayed glycogen storage capacity (Figure E). As expected for a protocol favoring directed differentiation to hepatic lineages, only low-level lineage specification to other nonhepatic endodermal lineages was detectable in all 3 cell lines, evidenced by late and transient expression of Ttf1
, thyroid stimulating hormone receptor (Tshr
), intestinal fatty acid binding protein (Ifabp
), and Pdx1
. In this protocol, there was no induction of additional pancreatic lineage markers, such as pancreas transcription factor 1 subunit α (Ptf1a
) (Supplemental Figure 3A and data not shown). When each cell line was exposed to an established 3-stage culture protocol (29
) designed to favor pancreatic lineage specification via inhibition of Shh and supplementation of FGF10 and retinoic acid, all 3 clones displayed similar early pancreatic lineage specification, indicated by induction of Hnf6, Pdx1, and Ptf1α (Supplemental Figure 3). Taken together, these waves of gene expression during differentiation to endoderm-derived lineages further supported the definitive endodermal capacity of the day 5 cells derived from each cell line in vitro. Furthermore, to demonstrate in vivo functional potential to form endoderm, unsorted day 5 iPS cell–derived putative endodermal progenitors were transplanted beneath the kidney capsules of SCID mice. These cells displayed robust capacity to form endodermal epithelia expressing nuclear Foxa2 protein (Supplemental Figure 4A), confirming the in vivo functional potential of iPS cell–derivatives following in vitro directed differentiation.
In contrast to the favorable growth kinetics we observed during differentiation of ST5 and ST8 cells, we found that additional syngeneic iPS cell lines (SEF4 and SEF11) that exhibit high-level reprogramming transgene overexpression driven by a constitutively active EF1α promoter (EF1α-STEMCCA; refs. 25, 27) did not robustly form endoderm in this 5-day differentiation protocol (Supplemental Figure 2C). SEF4 and SEF11 iPS cells, which showed more than 50-fold leak of reprogramming transgenes compared with that of ST5 and ST8 cells, failed to increase their cell numbers over 5 days of directed differentiation, findings in keeping with our prior work documenting the adverse effects of reprogramming transgene overexpression on the endodermal developmental capacity of iPS cells (27
Sox2-GFP downregulation distinguishes ES and iPS cell–derived endoderm from nonendoderm.
Although directed differentiation of the ES and iPS cell clones over 5 days into DE appeared to be efficient, heterogeneity of the cells at each time point was evident, based on (a) the residual presence of some ckit–, CXCR4–, or ENDM1– cells (Figure A); (b) the presence of some cells failing to express the endoderm transcriptional regulator, Foxa2, by immunostaining (Supplemental Figure 5); (c) detectable expression of mesodermal genes, Myf5 and Gata1 (Supplemental Figure 2A); and (d) the residual presence of cells on day 5 with nonendodermal or pluripotent potential, as reflected by the capacity of iPS cell–derived day 5 cells to form some nonendodermal lineages, such as mesodermal (smooth muscle actin+) and neuroectodermal (Tuj1+) cells in vivo after kidney capsule transplantation (Supplemental Figure 4A).
We evaluated potential strategies for distinguishing and purifying ES and iPS cell–derived DE progenitors from other cells present on day 5 of differentiation. Based on the differentiation kinetics of control ES cells (Figures and ), putative endodermal progenitors derived from ES/iPS cells by day 5 should be identifiable based on the surface phenotype ckit+/CXCR4+/ENDM1+. Analysis of the kinetics of expression of the Sox2-GFP knockin reporter also revealed residual Sox2 locus activity but at consistently lower intensity (one-half–log drop in fluorescence), as ES or iPS cells differentiated into ckit+/CXCR4+/ENDM1+ cells (Figure A). Indeed, quantitative RT-PCR (qRT-PCR) analysis of sorted day 5 populations confirmed that putative endodermal cells could be distinguished from other cells using a Sox2-GFPdim/ckit+ sort algorithm, since endodermal marker genes were expressed preferentially in this population (Figure A). “Contaminating” cells expressing residual levels of Nanog and Rex1 localized to the Sox2bright/ckit– population outside this sort gate. Most importantly, decreasing the heterogeneity of the day 5 cell population by cell sorting, produced ES and iPS cell–derived endodermal cells expressing highly similar levels of endodermal master transcriptional regulators (Figure B). Overall, 77% ± 7.59% (average ± SEM), 18% ± 7.32% (average ± SEM) (2-tailed t test, P = 0.005) of cells in the sorted Sox2-GFPdim/ckit+ populations expressed a putative endodermal phenotype, defined as coexpression of the endodermal markers ENDM1 and CXCR4; whereas only 18% ± 13% of Sox2-GFPbright/ckit– cells coexpressed ENDM1 and CXCR4 (P = 0.005).
Methodology for purification of ES/iPS cell–derived endoderm.
To evaluate the in vivo differentiation potential of each sorted population derived from each ES and iPS cell clone, we performed kidney capsule transplantations in 36 SCID mice (Supplemental Figure 4). Four weeks after transplanting identical numbers of day 5 Sox2-GFPdim
cells, day 5 Sox2-GFPbright
cells, or day 18 hepatic differentiated cells (also sorted on day 5 Sox2-GFPdim
cells; data not shown), we found all transplanted cells typically gave rise to very small tumors localized to kidney capsules (0.09 ± 0.21 cm2
= 4 recipients per group). There was no statistically significant difference among groups in tumor size resulting from each differentiated, sorted cell population from each ES and iPS cell clone (ANOVA, P
= 0.16). In contrast, an identical number of control undifferentiated stem cells (sorted day 0 ES Sox2-GFPbright
cells) required recipient harvest at the 4-week end point of the study, due to abdominal distension, resulting from rapid overgrowth of the expected large 1.73 cm2
teratoma (Supplemental Figure 4B), a size consistent with our prior experiments using day 0 ES cells (refs. 25, 27, and data not shown). As has been published by others (6
), these findings suggest that endodermal directed differentiation of pluripotent stem cells reduces their tumorigenicity after transplantation, compared with that of undifferentiated stem cell transplants. Histological scoring of each ES and iPS cell–derived tumor revealed that endodermal epithelium was the predominant differentiated tissue type arising from each population sorted after 5 days of activin stimulation; however, no sorted population was completely depleted of mesodermal and ectodermal structures (Supplemental Figure 4D). Overall, the tumors arising from the day 5 endoderm-enriched sorted transplants were more well differentiated than the immature large teratomas that were found to arise from undifferentiated ES or iPSST5 and ST8 cell transplants, whose histology predominantly consisted of immature neural rosettes and other ectodermal keratinized derivatives in addition to endoderm and mesoderm (ref. 25
and Supplemental Figure 4B). Furthermore, in comparison with tumors arising from sorted day 5 Sox2-GFPbright
cells or from day 0 cells, the tumors arising from the sorted day 5 Sox2-GFPdim
population were relatively depleted of ectodermal skin-like keratinized epithelia (Supplemental Figure 4D). None of the 36 recipients showed any malignant features in the benign growths arising from the sorted transplants.
Kinetics of global gene expression during endodermal differentiation of ES and iPS cells mimics that of E8.25 mouse DE in the developing embryo.
We next compared the changes in the global gene expression programs of ES cells versus iPS cells during directed differentiation into DE. Microarray analyses were performed on transcriptomes prepared from 18 samples, representing undifferentiated (day 0) ES, ST5, and ST8 iPS cells and differentiated (day 5) sorted Sox2-GFPdim/ckit+ cells from each cell line. Principal components analysis across all genes measured on the array indicated that, in the undifferentiated state, the ST5 and ST8 iPS cell transcriptomes were highly similar to each other but slightly different from their parental ES cell line (Figure A). Differentiation over 5 days was responsible for the vast majority of the variability in gene expression across all samples (first principal component PoV = 73.3%); however, there was some variability in gene expression between different cell lines during differentiation (second principal component PoV = 10.4%; Figure A).
Microarray analysis of global gene expression in ES and iPS cells before (day 0) and after (day 5) endodermal differentiation.
In order to interrogate the kinetics of global gene expression of each cell line during directed differentiation to endoderm, we used 2-way ANOVA of all 18 samples to identify (a) genes that are differentially expressed between day 0 and 5 of differentiation (time effect); (b) genes that are differentially expressed between ES, ST5, and ST8 cell lines (cell-type effect); and (c) gene expression differences during differentiation that are modulated by the cell line type (interaction effect of time and cell type). Endodermal differentiation from day 0 to day 5 was associated with a very large number of gene expression changes (approximately 8,000 out of approximately 29,000 probe sets were significantly associated with the time effect at false discovery rate [FDR] < 0.001). Importantly, all master endodermal transcriptional regulators (Foxa2, Gata4, Gata6, Sox17) that were differentially expressed by qRT-PCR analysis between day 0 and day 5 (Figure B) were also found to be differentially expressed by this global gene expression analysis. Next, we designated the top 1,000 of these transcripts (ranked by time effect, FDR-adjusted P value) as a putative “1,000-gene endoderm kinetic” signature and performed cluster analysis to compare this differentiation kinetic among each cell line (Figure B).
When studying the directed differentiation of pluripotent stem cells, an important issue is determining how closely a putative lineage generated in vitro mimics the phenotype of its authentic counterpart that is specified during normal development in the embryo. Hence, we sought to establish whether the 1,000-gene endoderm kinetic established in our in vitro model overlapped with the authentic global gene kinetic of DE development in the mouse embryo in vivo (hereafter referred to as embryonic DE). In order to establish the global gene kinetic of embryonic DE, we prepared RNA extracts from embryonic DE cells purified by flow cytometry from E8.25 mouse embryos based on an established ENDM1+
/side scatter low algorithm (ref. 28
and Figure C). We compared the transcriptomes of these embryonic DE cells with those of undifferentiated ES cells and found 2,715 differentially expressed transcripts at the significance level of FDR < 0.001. We found this embryonic DE kinetic signature overlapped with more than 50% of the in vitro 1,000-gene endoderm kinetic (Figure C). These results indicate that ES and iPS cell–derived DE resembles but is not identical to E8.25 embryonic DE. Moreover, when the 2,715 genes that define embryonic DE were used to generate an unsupervised clustering dendogram of all 18 ES and iPS cell–derived samples, ES and iPS cell–derived endoderm clearly clustered together and were distinct from the transcriptome programs of undifferentiated ES and iPS cells (Figure D). This cluster analysis also demonstrated that endoderm derived from the ST8 iPS cell clone appeared more similar to endoderm derived from the parental ES cells than that from the ST5 iPS cell clone.
Endodermal differentiation accentuates differences in expression levels of imprinted genes between ES and iPS cells.
To evaluate potential differences in the gene expression programs of ES cells versus iPS cells, we first compared cells in both the differentiated and undifferentiated states. We found that 111 transcripts (including mRNA, microRNA [miRNA], and small nucleolar RNA) were differentially expressed across the 3 cell lines regardless of differentiation state (cell-type effect FDR < 0.001; Figure A). Clustering analysis illustrated that the majority of these transcripts distinguished the 2 iPS cell clones from their parental ES cells. Remarkably, we found 36 of these differentially expressed transcripts were encoded by the Dlk1-Dio3
–imprinted gene cluster on mouse chromosome 12qF1 (Figure , A and B, and Table ). For example, 2 of the top 3 most differentially expressed genes between iPS and ES cells were maternally expressed 3 (Gtl2
, also known as Meg3
) and maternally expressed 8 (Rian
, also known as Meg8
), which appeared to have low to undetectable expression levels in both ST5 and ST8 cell lines compared with those in ES cells (qRT-PCR; Figure C). Both these genes are noncoding RNA members of the Dlk1-Dio3
–imprinted cluster that are typically monoallelically expressed from only the maternally inherited allele along with 5 other noncoding RNAs (30
). Of 11 remaining transcripts that distinguished iPS cells from ES cells with more than 4-fold differential expression (fold-change cut-off set to ensure all genes met Figure ’s FDR < 0.001 cutoff; Table ), 10 were miRNAs encoded by the Dlk1-Dio3
gene cluster, and all 10 appeared to be silenced in both iPS cell clones. Overall, 63 members of this gene cluster had known probe IDs on our microarray platform, and 36 of these were differentially expressed between ES and iPS cell lines with FDR < 0.001 (Fisher’s exact test for enrichment, P
= 1.5 × 10–63
). These results suggested aberrant silencing of many maternally expressed members of this imprinted gene cluster in both iPS cell lines in both differentiation states.
Analysis of cell-type effects between ES and iPS cell samples, regardless of differentiation stage, reveals aberrant silencing of genes encoded by the Dlk1-Dio3–imprinted gene cluster on chromosome 12qF1.
Top differentially expressed transcripts in iPS cells versus ES cells with ±4 fold change
Next we focused on gene expression differences that might distinguish ES cells from iPS cells during endodermal differentiation. Analyzing the interaction of time effect and cell type, we found that 105 transcripts were differentially expressed (FDR < 0.01; Supplemental Figure 6), indicating that differences in expression levels of these genes emerged between the 3 cell lines during the 5 days of directed differentiation. The top-most differentially expressed of all genes was the imprinted maternally expressed gene Gtl2 (P value interaction of time and cell type = 3.59 × 10–5). By qRT-PCR analysis, we confirmed that endodermal differentiation exacerbated the difference in Gtl2 expression levels between the 3 cell lines, as Gtl2 was upregulated in ES cells during endodermal differentiation and hepatic lineage specification but remained silenced in both ST5 and ST8 iPS clones (Figure , C and D). In contrast, a paternally expressed gene, Dlk1, in this cluster was not silenced in iPS cells and was upregulated more in differentiating iPS cells than in ES cells (Figure C), suggesting that silencing of the maternally inherited genes was due to aberrant imprinting of the cluster rather than global silencing of both alleles of this genomic region. These findings also demonstrated that silencing of the transcripts normally expressed from the maternally inherited allele was not simply due to the differentiation state of the iPS cells.
Parental origin-specific expression of imprinted genes is typically regulated by differential DNA methylation of paternal and maternal alleles in the germline. Imprinting of the Dlk1-Dio3
gene cluster is regulated by differentially methylated regions (DMRs), including a key intergenic region (IG-DMR) located between the Dlk1
genes (refs. 31
, and Figure B). We found that approximately 50% of IG-DMR CpG islands were methylated in the parental ES cells or tail-tip fibroblasts prior to reprogramming, as would be expected for germ line imprinted regions (Figure E). In contrast, close to 100% of IG-DMR CpG islands were methylated in ST5 and ST8 cells, both before and after endodermal differentiation. Aberrant methylation of DNA CpG islands was not evident at other loci in iPS cells, such as the Oct4 proximal promoter region, which was appropriately reprogrammed to an exclusively unmethylated state in both iPS cell lines and was amenable to developmentally appropriate initiation of CpG methylation during subsequent endodermal differentiation (Supplemental Figure 7). In addition, CpG islands around the transcriptional start site of the key endodermal master regulator, Foxa2, remained unmethylated in fibroblasts, ES cells, and iPS cell lines both before and after endodermal differentiation (Supplemental Figure 7), indicating that, in contrast to Gtl2
gene regulation, epigenetic mechanisms distinct from CpG methylation are responsible for regulating expression of Foxa2 early in development. Taken together our findings support recent reports (20
) suggesting that iPS cells exhibit aberrant imprinting of the Dlk1-Dio3
gene cluster in an exclusively paternal pattern, with resultant silencing of maternally expressed genes and overexpression of the paternally expressed imprinted gene, Dlk1
We speculated that gene expression differences between ES and iPS cell lines might be particularly important if these differences were associated with an altered capacity to undergo directed differentiation or lineage specification to desired target cell lineages, such as endoderm. Indeed deletion of maternally inherited Gtl2 in mice is known to result in early postnatal defects in at least 2 endodermally derived epithelia, such as pulmonary alveolar hypoplasia and hepatocellular necrosis (32
). Thus, we sought to determine whether silencing of Gtl2 due to aberrant imprinting of the Dlk1-Dio3
gene cluster in iPS cells might be associated with altered endodermal differentiation capacity compared with ES cells. Although we had found no detectable difference among ST5 cells, ST8 iPS cells, and ES cells in terms of upregulation of early endodermal markers (Figures and ), we did note that significant differential expression of BMP4 emerged during endoderm differentiation (3.5-fold higher expression in iPS cells than ES cells; interaction of time and cell type, P
= 0.008). Since BMP4 is one predicted target of miRNA-380, encoded in the Dlk1-Dio3
cluster and aberrantly silenced in iPS cells (Table ), differential upregulation of BMP4 between iPS and ES cells during endoderm differentiation would be expected. Since higher expression levels of BMP4 and Dlk1 in endodermal precursors may potentially impact their capacity to undergo liver lineage specification and differentiation (14
), we quantified hepatic lineage specification across all 3 cell lines and noted the induction of Afp, α-1 antitrypsin, and albumin in ST5 and ST8 iPS clones to be greater than that in their parental ES cells in 4 out of 4 repeated experiments (Figure , C and D). Thus, the aberrant imprinting of the Dlk1-Dio3
gene cluster in ST5 and ST8 iPS cell lines did not appear to be associated with any detectable decrement in the capacity of those lines to undergo directed differentiation to DE or early hepatic lineage specification in vitro. Only after further hepatic differentiation to day 19 did 1 out of the 2 aberrantly imprinted clones (ST8) show a statistically significant, yet subtle, functional decrement in glycogen storage capacity relative to that of ES cells (P
= 0.03; Figure E).