H1 hESCs (WiCell) were maintained in feeder free conditions as previously described [17].

Once hESCs reached an average colony size of 1 mm in diameter, DE induction media was added for 4 days with media change every day. After 4 days media was replaced with pancreatic progenitor (PP) media for 2 days with media change every day. After 2 days, all-Trans Retinoic acid was added to the PP media for 2 additional days with media change every day. Media was then replaced with maturation media. After 2 days DAPT was added to maturation media. Cells were maintained in this media for 1 week with media change every day. Media formulations are found in table S1.

On day 0 of the protocol, several wells were treated with Accutase and starting cell density was estimated using a hemocytometer. 24 hours after initial DE media exposure, cell death was quantified by counting floating cells in the media and normalized with respect to the starting cell density. Additionally, the remaining attached cells were harvested with Accutase, stained with propidium iodide in PBS at a concentration of 10 ug/ml and the number of dead cells (PI positive) was quantified by flow cytometry. For quantification of cell number throughout the entire protocol, cells were exposed to alamar blue at day 0 according to manufacturer's instructions for quantification of cell number. This procedure was repeated at the end of each stage of differentiation (days 4,8,15), and cell number was calculated as described in the product manual, using day 0 values as a control for each of the stages.

qPCR was performed as previously described [11]. A list of the primers used can be found in the table S2. ΔCt values were calculated by subtracting the respective Ct value for GAPDH from the Ct value of the marker(s) of interest. ΔΔCt values were calculating by subtracting the ΔCt values for undifferentiated cells for the marker of interest from the ΔCt value for the same marker in each group. Relative expression was found by calculating 2^−ΔΔCt.

Flow cytometry was performed as previously described [11]. As a control for non-specific staining, cells were incubated in secondary antibody only. Cells were analyzed using an Accuri C6 flow cytometer. Antibodies and concentrations can be found in the table S3. For cell cycle analysis, cells were harvested and dissociated with Accutase, rinsed, centrifuged and resuspended in ice-cold 70% ethanol and fixed overnight in −20°C. Cells were rinsed and suspended in DNA staining buffer (PBS+0.1% Triton-X+0.2 mg/mL DNAse-free RNAse+0.01 mg/mL/1 million cells propidum iodide) for 25 minutes at RT. Stained cells were then directly analyzed on an Accuri C6 flow cytometer, the output of which being a histogram of the DNA content for the cellular population in each sample. To accurately determine the fractions of the cellular population in each phase of the cell cycle from these data, Modfit LT was applied to the DNA histogram data. Modfit identifies the G1 and G2 peaks of DNA histograms acquired by flow cytometry and fits established cell cycle models to these peaks in addition to the S phase “peak”. The area under the curve is calculated via this model, thereby obtaining relative proportions of each cell cycle phase within the population.

Cells were fixed using 4% parafolmaldehyde for 15 minutes at room temperature and permeabilized using 0.25% Triton X-100 (TX) for 15 minutes. Blocking was performed in 10% donkey serum in 0.05% TX for 30 minutes followed by primary antibody incubation which was performed overnight at 4°C in blocking buffer with primary antbodies. Secondary antibody incubation was performed for one hour at room temperature in the dark with appropriate antibodies diluted in blocking buffer. Nuclear staining was performed by incubation with Hoescht stain in PBS for 5 minutes. Pictures were taken using Olympus IX81 inverted microscope and Metamorph imaging software.

Differentiation results are presented as averages of 6 separate independent experiments. Error bars represent SEM. Kriskal-Wallis test was used to determine statistical significant difference between the DE induction treatments. Additional Mann–Whitney U tests were used for post-hoc comparison with Bonferroni correction of the α.

A multi-stage directed differentiation protocol was used to induce the hESCs to pancreatic lineage (Fig. 1). The first step was to induce DE through multiple alternate pathways, which was achieved by exposure to Activin in combination with one of the four other growth factors and molecules that modulate alternate pathways for DE induction.

While Activin alone can induce DE, it is typically combined with different molecules to increase the efficiency of induction. In pancreatic differentiation studies, DE is most commonly achieved by combination of Activin A with WNT3A[13], BMP4[9], PI3K inhibitor[18] or FGF2[19].

After 4 days of DE induction Activin A and other inducers were removed and all groups were exposed to the same subsequent signals as follows (Fig. 1): for PP induction Cyclopamine was added alone for two days and in combination with retinoic acid for two additional days; cells were then exposed to nicotinamide alone for 2 days and nicotinamide and DAPT for up to one week for the maturation stage.

Morphological examination of the matured cells exposed to alternate DE induction pathways revealed heterogeneous populations of cells in all conditions (Fig. 2A) containing groups of cobblestone like cells indicative of endoderm morphology; however, PI3KI cells appeared to be larger than other groups. To determine if this was attributed to cell confluence, and to analyze the system more thoroughly, we studied proliferation, apoptosis and dynamics of cell cycle under different induction conditions. Cell death at DE stage was comparable for all conditions except PI3K inhibition, which elicited high cell death (Fig. 2B). This is evident in Fig. 2C which shows a drop in cell number in PI3KI-DE. However, cell cycle dynamics (Fig. 2D–E) confirm a proliferative population, with similar dynamics between PI3KI and WNT3A conditions. Analysis of the cell cycle clearly indicates a maturing population of cells, transitioning from a dominant S phase to a dominant G1 phase, representative of mature cells [20]. As expected, undifferentiated hESCs (time = 0) have a short G1 phase as exhibited by a low sub-population (∼27%) in the phase, with subsequent increase in G1 residence time with differentiation (Fig. 2D). While the residence times of the S and G2/M phases are not expected to significantly change with differentiation, the fraction of the population in these phases decreases to compensate for the increased G1 phase (Fig. 2E, F). Compared to the PI3KI and WNT3A conditions, the kinetics of this transition, from dominant S to dominant G1, is very slow for the BMP4 and FGF2 conditions during initial DE induction, as exhibited by a small fraction of the population in G1 up until day 2 (Fig. 2D). The G1 population then increases until it reaches a level comparable to the WNT3A and PI3KI conditions at the end of DE induction (day 4). This is reflected in the proliferation data (Fig. 2C) showing an almost identical behavior between WNT3A, BMP4, and FGF2 at DE.

In order to confirm differentiation after DE induction, immunofluorescence (IF) and flow cytometry for SOX17 and FOXA2 was performed for all groups after 4 days of treatment. Both transcription factors were found to be expressed in all groups (Fig. 1A–D) with yield of FOXA2 positive cells ranging from 40–90%. qPCR was performed to examine expression of stage specific markers CXCR4, SOX17, FOXA2 and CER. As illustrated in Fig. 3A, upregulation of these markers was obtained under all differentiation conditions, with PI3KI consistently eliciting the highest upregulation, achieving close to 50 fold increase in CXCR4, 400 fold increase in SOX17, 10 fold increase in FOXA2 and 500 fold increase in CER.

Upon pancreatic induction, all the induction conditions show expression of PP marker PDX1 by IF (Fig. 1A–D). This was further confirmed by qPCR for PDX1, which showed that with the exception of BMP4, all other conditions strongly expressed PDX1 (Fig. 3B). A notable increase of other PP markers was also observed, particularly ISL1. BMP4 treated cells, however, consistently showed either comparable or lower upregulation of PP markers than the other groups. BMP4 treated cells additionally showed downregulation of PAX6.

At the last stage of differentiation, IF and flow cytometry confirmed expression of C-peptide for all groups with yields ranging from 9–24% (Fig. 1A–D). Detailed gene expression for mature β cell markers (Fig. 3C) revealed the highest INS mRNA upregulation under WNT3A and FGF2 conditions, both of them achieving over 10,000 fold increase compared to undifferentiated cells, with no statistical difference between them. While BMP4 condition showed the lowest (11 fold) upregulation of INS, it was the highest in upregulation of GLUC mRNA (Fig. 3C). It is noteworthy that there was considerable variability in the results from the different experiments, as observed by the error bars, attributed to population heterogeneity and variable response to global inductive cues [21], [22]. However, despite the variability, all experiments consistently showed a similar trend where highest insulin upregulation for every experiment was obtained in FGF2 of WNT3A treated cells, while BMP4 consistently lead to insignificant insulin upregulation. Results from individual experiments can be found in figure S1.

The above analysis clearly indicates that the initial pathway of endoderm induction plays a crucial role in subsequent maturation of the cells towards pancreatic lineage.

Research over the last decade has established the advantage of directed differentiation of hESCs following the sequence of in vivo development. Hence, there is an increased emphasis on aligning the in vitro differentiation dynamics to in vivo organogenesis events. Accordingly, we analyzed the alternate pathways of endoderm induction and subsequent maturation in the light of differentiation dynamics.

Pancreatic development can be broadly divided into 7 stages, each characterized by specific transcription factors. The first stage is primitive gut endoderm (PGE), from which pancreas, lung, thyroid, thymus, parathyroid, and liver are derived[23], followed by prospective pancreatic endoderm (PPE) containing prospective ductal, endocrine and exocrine pancreatic cells. The next step is the pancreatic progenitor (PP) stage marked by the transient expression of PTF1 and is followed by appearance of NGN3 expressing early endocrine progenitors (EEP). EEP develop into endocrine progenitors (EP), from which all the islet cell types develop, including α, β, γ, δ and ε cells. From here disappearance of NGN3 expression marks emergence of immature β- cells which mature into functional, INS expressing β-cells[24]. To draw a parallel to our 3-stage differentiation protocol, we combined specific developmental stages as follows: DE stage includes endoderm, PGE, and PPE; PP stage includes PP and EEP induction; and the maturation stage includes EP induction, immature β- cells and β- cell maturation. Fig. 4A illustrates a qualitative measure of the expression patterns of stage specific transcription factors across different stages of development as gathered from literature [23], [24]. Fig. 4 (B–E) presents parallel transcription factor dynamics for hESC differentiation under different DE induction conditions as observed in a representative sample with INS expression closest to the mean. For the purpose of this study, we defined presence of a marker as 10 fold or higher upregulation as observed by qPCR in order to account for experimental error. Overall the FGF2, WNT and PI3KI conditions were found to exhibit similar trends as in vivo development, only with some minor differences. For example, PTF1 is known to be an early and transient marker of pancreatic commitment, preceding PDX1 expression. While both FGF2 and WNT conditions show a gradual increase in PTF1, under PI3KI conditions PTF1 comes up very late even though PDX1 expression is detected much earlier, even at the DE stage. On the other hand PAX6, which is expressed early in α cells and later in the entire islet [25], and has been suggested to be a key component of glucagon secretion [26], is prominent in PI3KI conditions from an early stage (DE) and increases with maturation. BMP4 condition, however, was found to be an outlier, as it did not align with either in vivo sequences or any of the other conditions.

Our next goal was to determine which of the remaining conditions are more suitable to drive pancreatic maturation. One way to assess this is to find representative TFs that show coherent expression dynamics. To address this question we scrutinized each of the pathways individually through K-means clustering of each induction condition.

As shown in Fig. 6A, SOX17, FOXA2, HLXB9 were co-regulated under WNT3A, FGF2 and PI3KI conditions. These markers indicate the DE and dorsal pancreatic endoderm. This combination of SOX17, FOXA2 and HLXB9 was repeated in all the above induction conditions, indicating that each of these treatments is efficient for activating the primary DE transcriptional machinery and that at the later stage transcriptional activation is different. These markers were consistently expressed through all the differentiation stages. In addition, PI3KI and FGF2 clusters also contained ISL1. However, no other coherent cluster was obtained for the PI3KI condition, indicating lower alignment with developmental dynamics towards the later stages of maturation.

Additional clusters containing many later markers were obtained for WNT3A and FGF2 as shown in Fig. 6B. One among these was PTF1 and ISL1 under WNT3A which arise in the pancreatic precursor cells during the early bud-stage. Other late markers such as PAX6, PDX1, MAFA, GLUC, INS, NGN3, HNF6, and NKX2.2, which are expressed in the NGN3+ cells maturing to the β-cell stage [29], were also identified under WNT3A treatment. These later markers show continuous rise in expression across the stages. Therefore, it reinforces the observation that early WNT3A induced cells were found to closely shadow the in vivo embryonic transcriptional dynamics. For FGF2, clusters containing small number of late markers were identified as shown in Fig. 6B. Two groups were identified, one containing PTF1 and INS and the other containing HNF6, PAX6 and MAFA. However, FGF2 contained far less coherent markers at the later stages than WNT3A.

The above analysis indicates that modulation of Activin with WNT and FGF2 are likely routes to pancreatic β-cells, although WNT pathway is identified to be the most suitable because of the co-regulation of important markers during each stage of the differentiation process. Furthermore, this comparison reveals that even though the expression of DE and PP markers are quite similar for all the induction conditions at the end of DE stage, they deviate significantly upon maturation. This is suggestive of cellular ‘memory’ of pathway of initial induction even after phenotypic maturation.

The above results establish that different pathways of endoderm induction of hESCs have a significant influence on the cells' subsequent mature phenotype and functionality. Another way of looking at it is that efficiency of endoderm commitment, as analyzed by current markers, is not indicative of an efficient pancreatic maturation. The next question thus is whether any of the early or intermediate stages can reveal the potential for cellular maturation to islet cell types.

We addressed this by performing partial least squares regression (PLSR) analysis on the mean TF expression data to identify which early TFs, if any, were predictors of INS expression. Here we are seeking the TFs that showed the most significant correlation to INS expression over all the time points of the differentiation trajectory. The correlation of each of the TFs with INS for each induction condition is represented in Fig. 7 as associated regression coefficients. It is found that most of the PP markers show high degree of correlation to INS expression while there is no significant dependence on the DE markers analyzed. None of the early DE markers analyzed show a positive correlation to INS across all the induction conditions. The intermediate PP stage markers, including PTF1, PDX1, HNF6, NKX2.2, NKX6.1, and NGN3, are better predictors of INS. Also, WNT3A and FGF2 conditions gave positive coefficients with most of the PP and mature markers indicating that these conditions are optimal for INS expression. It is also observed that under BMP4 and PI3KI, the markers NKX6.1, PTF1 and NGN3 gave strong positive correlations indicating that these markers are in fact strongly associated with INS even under low INS upregulation. In addition, we analyzed the expression of PP markers after DE induction and found high expression of HLXB9, PTF1 and ISL1 at this early stage under some of the conditions for the selected sample. Interestingly, PTF1 resulted in high upregulation under FGF2 and WNT3A, which resulted in highest INS upregulation. This observation, combined with the fact that PTF1 expression is highly correlated to INS expression under many conditions from PLSR, suggests that analysis of PTF1 expression after DE induction could be used as a determinant of pancreatic potential.