Reconstructing human pancreatic differentiation by mapping specific cell populations during development

Information remains scarce on human development compared to animal models. Here, we reconstructed human fetal pancreatic differentiation using cell surface markers. We demonstrate that at 7weeks of development, the glycoprotein 2 (GP2) marks a multipotent cell population that will differentiate into the acinar, ductal or endocrine lineages. Development towards the acinar lineage is paralleled by an increase in GP2 expression. Conversely, a subset of the GP2+ population undergoes endocrine differentiation by down-regulating GP2 and CD142 and turning on NEUROG3, a marker of endocrine differentiation. Endocrine maturation progresses by up-regulating SUSD2 and lowering ECAD levels. Finally, in vitro differentiation of pancreatic endocrine cells derived from human pluripotent stem cells mimics key in vivo events. Our work paves the way to extend our understanding of the origin of mature human pancreatic cell types and how such lineage decisions are regulated. DOI: http://dx.doi.org/10.7554/eLife.27564.001


Introduction
Intensive efforts are currently dedicated towards the development of cell replacement therapies using cell types derived from human pluripotent stem cells (hPSC). Human insulin-producing beta cells represent a paradigm for this type of objective. These cells have a major physiological function, regulating circulating glucose levels by producing and secreting insulin. In patients suffering from type one diabetes, these cells are destroyed by an autoimmune mechanism, and would thus need to be replaced (Benthuysen et al., 2016). Beta cell replacement holds immense promises for diabetic patients and current strategies have reached major milestones Russ et al., 2015;Rezania et al., 2014). However, it is well accepted that a more detailed understanding of beta cell development in human is required to generate unlimited functional human beta cells (Johnson, 2016).
The adult pancreas is composed of acinar cells that excrete enzymes into the duodenum through a ductal tree, and of endocrine cells (approximately 1% of the total pancreatic cells) that are clustered together forming the islets of Langerhans. The endocrine cells secrete hormones such as insulin (beta cells), glucagon (alpha cells) somatostatin (delta cells), pancreatic polypeptide (gamma cells) and ghrelin (epsilon cells). The pancreas develops from the primitive gut tube that evaginates into a dorsal and a ventral anlage (Pan and Wright, 2011;Jennings et al., 2015). Multipotent epithelial pancreatic progenitors co-expressing the transcription factors PDX1 and NKX6-1 (Nelson et al., 2007;Cebola et al., 2015) proliferate upon signals (such as FGF10) from the adjacent mesenchyme (Bhushan et al., 2001) and subsequently differentiate into the acinar, ductal and endocrine lineages. Endocrine commitment is initially marked by the expression of a basic helix-loop-helix transcription factor, NEUROG3, and followed by the expression of the mature endocrine markers (Gradwohl et al., 2000;Gu et al., 2002).
Due to the difficulties associated with procuring staged human fetal tissues and the limited tools for their analysis, few data on human development is available and the majority of knowledge on tissue development derives from animal models. This also applies to the pancreas  where only a limited number of studies have been performed on human pancreatic development. Theses studies demonstrate similarities but also differences between rodent and human pancreatic development Jennings et al., 2015;Nair and Hebrok, 2015).
Knowledge on human pancreatic development remains limited. More information exists concerning human hematopoietic cell differentiation thanks to the characterization and use of cell surface antigens that enabled to identify, quantify and purify hematopoietic stem cells and progenitors at different stages of their development (Eaves, 2015). By mirroring the hematopoietic field, we developed here an approach where cell surface markers are used to recapitulate the hierarchical sequence of human pancreatic development. Specifically, we characterized the expression levels of specific markers at different stages of human pancreatic development corresponding to 7 to 12 weeks of development (WD). First we purified human pancreatic epithelial cells by selecting cells positive for the transmembrane glycoprotein EPCAM and by excluding CD45 + hematopoietic and CD31 + endothelial cells. Next we segregated pancreatic epithelial cells into four populations based on the GP2 and CDH1 (ECAD) expression levels. We observed that the expression levels of GP2 and ECAD correlate with acinar, ductal and endocrine functions. By using the additional cell surface markers CD142 and SUSD2 we further refined endocrine cell differentiation. Finally, our development model also applies to the in vitro differentiation of hPSCs into pancreatic endocrine cells.
Taken together our work provides a novel approach to study human fetal pancreas development and bridges the path between in vivo and in vitro differentiation of human pancreatic endocrine cells.

EPCAM expression is restricted to the epithelium in the human fetal pancreas
We tested if EPCAM can be used to purify human fetal pancreatic epithelial fraction enriched in pancreatic progenitors. CD31 and CD45 were used to exclude endothelial and hematopoietic cells respectively. Using this combination of antibodies on human fetal pancreatic cells (9.7WD), we detected three distinct fractions: the CD45 + /CD31 + fraction (the endothelial/hematopoietic cells), the CD45 -CD31 -EPCAMfraction, and the CD45 -CD31 -EPCAM + fraction ( Figure 1A). To unveil which fraction contained the pancreatic progenitors, we assayed the expression of PDX1 and NKX6-1. Immunohistochemistry analysis of human fetal pancreatic sections showed that EPCAM + cells expressed PDX1 ( Figure 1B). FACS analysis demonstrated that PDX1 and NKX6-1 were coexpressed in the CD45 -CD31 -EPCAM + fraction ( Figure 1C-E). RT-qPCR analysis on sorted fractions confirmed that PDX1 and NKX6-1 expressions were restricted to the CD45 -CD31 -EPCAM + fraction ( Figure 1F) whereas the CD45 -CD31 -EPCAMfraction did not express pancreatic markers and most likely represent the mesenchymal pancreatic fraction (later referred as population M) ( Figure 1A). These results suggest that a combination of the cell surface markers EPCAM, CD45, and CD31 can be used to purify the human fetal pancreatic epithelial fraction.

Acinar and endocrine functions segregate within the GP2 and ECAD populations
To characterize the four epithelial populations described above, we sorted the GP2 hi , GP2 + , GP2and E low populations and performed global transcriptomic analyses combined with RT-qPCR analyses at 9 and 11WD. As a non-epithelial control, we included the CD45 -CD31 -EPCAMfraction (population M) ( Figure 1A). Due to limitations in cell numbers, the GP2 hi population was only sorted at 11WD. Principal component analysis (PCA) on the sorted populations at 9WD revealed three clusters. PC1 separated the epithelial populations (GP2 hi , GP2 + , GP2and E low ) from the mesenchymal fraction (M), while PC2 segregated the E low population from the GP2 hi , GP2 + and GP2populations ( Figure 3A). Gene Set Enrichment Analysis (GSEA) using Gene Ontology database indicated that digestion was the most represented biological process in the GP2 hi and GP2 + populations while endocrine functions (insulin and peptide secretion, hormone secretion) were enriched in the E low population ( Figure 3B . In contrast, the E low population contained 91 (at 9WD) and 34 (at 11WD) differentially expressed genes (p<0.05) that were also enriched (98% and 100% respectively) in the adult endocrine cells (alpha, beta, delta, epsilon or gamma cells) ( Figure 3C We next generated heatmaps based on Gene Ontology lists and selected acinar, ductal and endocrine genes. By RT-qPCR analyses we confirmed that acinar markers such as CEL, CELA3A and CTRC were enriched in the GP2 + population at 9WD and in the GP2 hi population at 11WD ( Figure 4B).
Finally, to define the cell population where endocrine cells first differentiate, we followed the expression of the endocrine progenitor marker NEUROG3. NEUROG3 was first detected at 8.4WD in the GP2population ( Figure 4C) prior to the detection of the E low population ( Figure 2C) and next (9 to 13WD), found enriched in the E low population ( Figure 4C). Thus, our data suggest that the first human pancreatic endocrine progenitors differentiate in the GP2population and mature while decreasing ECAD levels in the E low population. . GP2 and ECAD expression in the human fetal pancreatic epithelium. GP2 and ECAD expressions were assayed by flow cytometry during development. (A) FACS plots display the expression at 9.4WD of CD45 and CD31 against EPCAM (left plot) and GP2 and ECAD gated on CD45 -CD31 -EPCAM + (right plot). n = 4 (B) Cell frequencies of the GP2 hi (GP2 hi ECAD + ), GP2 + (GP2 + ECAD + ), GP2 -(GP2 -ECAD + ) and E low (GP2 -ECAD low ) populations at 9.4WD. n = 4 (mean ±SEM) (C) GP2 and ECAD expressions on fetal pancreases at 7-12WD gated on CD45 -CD31 -EPCAM + cells. 7WD Figure 2 continued on next page CD142 and SUSD2 reveal heterogeneity within the GP2and E low populations during development The GP2population displayed both duct and endocrine progenitor cells markers suggesting it contains more than one cell type. We therefore sought for additional discriminant markers by scrutinizing our transcriptomic data. We observed two cell surface markers, CD142 and SUSD2 with opposite expression patterns: the E low population expressed lower CD142 and higher SUSD2 mRNA levels than the GP2 + and GP2populations ( Figure 5A). FACS analyses showed that at 9.4WD, the GP2 hi and GP2 + populations were uniformly CD142 + SUSD2 -, while the GP2and E low populations were further divided into three subsets: CD142 + SUSD2 -, CD142 -SUSD2and CD142 -SUSD2 + ( Figure 5B). Noteworthy, the CD142 -SUSD2 + subset was scarce in the GP2population (6%), but represented 40% of the E low population at 9.4WD ( Figure 5B). Accordingly, our data reflects the heterogeneity within the GP2and E low populations that is further resolved using CD142 and SUSD2 antibodies.
Next we examined the CD142 and SUSD2 expression patterns in the GP2and in the E low populations during development. At 7WD the majority of the GP2population was CD142 + SUSD2 -(95 ± 2%). This frequency gradually decreased to 62 ± 2% at 9.4WD while the frequency of the CD142 -SUSD2subset increased from 5 ± 2% at 7WD to 43% at 11.3WD p<0.05 ( Figure 5C,D). We detected the first SUSD2 + cells in the GP2 -CD142subset at a low frequency (1%) at 8.4WD ( Figure 5C). The first E low population was detected at 8.6WD and was divided into CD142 + SUSD2 -, CD142 -SUSD2and CD142 -SUSD2 + subsets ( As development progressed the frequency of the E low CD142 -SUSD2 + subset decreased (from 48 ± 7% at 8.6WD to 10 ± 1% at 12WD; p<0.05), while the frequency of the E low CD142 -SUSD2subset increased (from 30 ± 4% at 8.6WD to 79 ± 3% at 12WD; p<0.05) ( Figure 5B,C,E). Our data thus indicate that the first GP2cells are CD142 + . They progress in their differentiation program by down-regulating CD142, then up-regulating SUSD2 and finally decreasing ECAD levels.
Endocrine progenitors develop in the GP2 -CD142 -SUSD2subset and mature within the E low SUSD2 + subset We analyzed the expression pattern of endocrine markers from 8.6WD to 13WD in the three E low subsets. The E low CD142 + SUSD2subset did not express endocrine markers such as CHGA and NEU-ROG3 ( Figure 6-figure supplement 1). At 8.6WD, CHGA, NEUROD1, NKX2-2 and INS were solely detected in the E low CD142 -SUSD2 + subset ( Figure 6A). Interestingly, at 10-12WD, CHGA, NEU-ROD1 and NKX2-2 expressions were detected both in the E low CD142 -SUSD2 + and the E low CD142 --SUSD2subsets while INS was exclusively detected in the E low CD142 -SUSD2subset ( Figure 6B).
Next we assessed NEUROG3 expression in the different subsets. The first NEUROG3 + were detected in the GP2 -CD142subset at 8.4WD before SUSD2 up-regulation ( Figure 6C). From 8.6WD, when the E low population was first detected NEUROG3 was enriched in the E low CD142 -SUSD2 + subset ( Figure 6C). Our results indicate that endocrine progenitors first appear  in the GP2 -CD142 -SUSD2subset and next mature within the E low CD142 -SUSD2 + subset. INS is first detected in the E low CD142 -SUSD2 + subset and later on in the E low CD142 -SUSD2subset.

Human pluripotent stem cells differentiation into pancreatic endocrine cells mimics human fetal endocrine cell development
To determine if similar cell populations were present in pancreatic endocrine cells derived from three hPSCs (SA121 hESC, AD2.1 iPSC, AD3.1 iPSC), these cells were differentiated to the corresponding stage (stage 5) where significant endocrine induction occurs ( Figure 7A). CD142 and ECAD characterized three distinct populations: ECAD + CD142 + , ECAD + CD142and ECAD low CD142as it is the case in the human fetal pancreas (9.4WD) ( Figure 7B,C). Similar to the human fetal pancreas, SUSD2 + cells were enriched in the ECAD low CD142population ( Figure 7B,C). Moreover, NEUROG3, NEUROD1 and NKX2-2 were mainly expressed in the ECAD low CD142 -SUSD2 + population as observed in the human fetal pancreas at 8.6WD (Figures 6A,C and 7D). Finally, from stage 2-5 the temporal expression pattern of CD142 and ECAD was reminiscent of the ones occurring in the human fetal pancreas with a sharp decrease of CD142 + SUSD2subsets (from 95 ± 2 of CD142 + -SUSD2at stage 2 to 23 ± 4 of CD142 + SUSD2subset; p<0.05) (5D and 7E), as was also the case for SUSD2 expression (Figures 5E and 7F) in the three hPSC lines. To conclude we demonstrate that pancreatic endocrine cells derived from human pluripotent stem cells appear to go through the same intermediate developmental stages as observed during in vivo development.

Discussion
In this study, we reconstruct human fetal pancreatic differentiation by combining a specific combination of cell surface markers. Although a large set of data is available concerning development in rodent models, there is only limited knowledge on human development. Importantly, while rodent and human pancreatic development share many similarities, they differ on several aspects Nair and Hebrok, 2015). For example, the global shape and the way islet cells cluster are different between rodent and human pancreas (Brissova et al., 2005). Moreover, the expression pattern of major transcription factors crucial for proper pancreas development such   as NEUROG3, NKX2-2 and PDX1 differ between rodent and human (Villasenor et al., 2008;Salisbury et al., 2014;Jennings et al., 2013;Sussel et al., 1998;Ohlsson et al., 1993;Heimberg et al., 2000). As an example, NEUROG3 is expressed in two waves during rodent fetal pancreatic development (Villasenor et al., 2008) while only a single-phase is observed in human (Jennings et al., 2013). It can be speculated that the number of differences between rodents and humans are under-estimated and more will be discovered by studying human pancreatic development. This knowledge will be essential to better design protocols to direct pluripotent stem cells towards functional insulin-producing pancreatic beta cells Rezania et al., 2014;Russ et al., 2015).
Major progress has recently been made in the field of hPSC differentiation and a number of different cell types can be efficiently generated either in vivo following transplantation into immuneincompetent mice or in vitro under controlled conditions. However, the cells generated in vitro from hPSC do not seem fully differentiated and in some cases, display a fetal phenotype rather than an adult one. This is the case for hepatocytes (Baxter et al., 2015), cardiomyocytes (Karakikes et al., 2015), neurons (Playne and Connor, 2017) and pancreatic beta cells (Hrvatin et al., 2014). These partial successes could be in part due to the limited knowledge on human cell development. Here, we specifically designed new approaches to dissect in great detail pancreatic differentiation in human. We based our study on human fetal pancreases from 7 to 13WD. This developmental period corresponds to E12.5-E17 in the mouse when pancreatic progenitors first proliferate and next develop into differentiated cells (Gittes, 2009;Jennings et al., 2015). At 6-7WD, the human pancreatic epithelium is mainly composed of PDX1 + NKX6-1 + pancreatic progenitors (Cebola et al., 2015;Riedel et al., 2012) while endocrine cells are extremely rare (Polak et al., 2000;Castaing et al., 2001). During the following weeks, proliferating epithelial pancreatic progenitors differentiate into endocrine, acinar and duct cells (Jennings et al., 2013;Capito et al., 2013).
The fetal pancreas is a compound organ composed of epithelial, mesenchymal, endothelial and hematopoietic cells. We excluded endothelial and hematopoietic cells using CD31 and CD45 antibodies and used the transmembrane glycoprotein EPCAM as a marker to segregate the fetal pancreatic epithelium from the mesenchyme. Previous data indicated that EPCAM is expressed during human fetal life (18-20WD) in the ductal pancreatic epithelium and in developing islet-like cells but also in the adult human pancreas in duct and islet cells (Cirulli et al., 1998). Based on this previous report we demonstrated that EPCAM also marks the fetal pancreatic epithelium at earlier developmental stages, between 7 and 12WD. Moreover, we demonstrated that the EPCAM + compartment contains PDX1 + NKX6-1 + double-positive cells, a hallmark of pancreatic progenitors (Jennings et al., 2013;Cebola et al., 2015). Our data also indicate that 45% of the EPCAM + compartment expresses lower levels of PDX1 and NKX6-1. It would be interesting to determine if these cells are upstream progenitors of EPCAM + PDX1 + NKX6-1 + cells as recently suggested (Ameri et al., 2017). Then we further segregated the fetal pancreatic epithelium into four distinct populations using the cell surface markers GP2 and ECAD. GP2 is a glycoprotein that is highly enriched in the acinar cells of the adult pancreas (Hoops and Rindler, 1991;Yu et al., 2004). Limited knowledge is available on GP2 expression during pancreatic development. Very recently, using a model of pluripotent stem cells differentiation into pancreatic endocrine cells, GP2 was identified as a novel cell surface marker of human pancreatic progenitors (Ameri et al., 2017). Our ex vivo data further support this claim. First, we observed that at 7WD, nearly all pancreatic epithelial cells are GP2 + . Moreover, as development proceeds, the frequency of GP2 + cells decreases. Finally, transcriptomic profiling performed at different developmental stages strongly suggests that the GP2 + population can differentiate into endocrine and exocrine cells. Collectively, these data indicate that in the early human fetal pancreas, GP2 is indeed a cell surface marker of a multipotent cell population. Although the GP2 + cell population is multipotent, it is not yet certain if the GP2 + population contains multipotent GP2 + progenitor that can differentiate into the three lineages (endocrine, acinar and ductal) or represent a mixture of GP2 + progenitors already committed to a specific lineage. This could be addressed by utilizing a single cell culture approach. However, our attempts so far to culture so few cells sorted from human fetal pancreas have failed. Discovering new and efficient culture conditions such as co-culture on feeder layers could alleviate this issue (Trott et al., 2017).
In animal models, cell adhesion molecules have emerged as key regulators of embryonic morphogenesis and this topic has been extensively studied using pancreatic development as a model organ (Semb, 2004). As an example, during mouse and chick development, Pdx1 + Nkx6.1 + progenitors express high levels of ECAD. NEUROG3 + endocrine progenitors will develop from such pancreatic progenitors while lowering their ECAD level during their delamination from the ductal tree to develop into pancreatic endocrine cells (Gouzi et al., 2011). Information remains scarcer on the regulation of ECAD levels during human development. Our results indicate that at 7WD, GP2 + pancreatic progenitors express ECAD at high levels. At 8.6WD, ECAD low cells appear, their frequency increasing while development progresses. Interestingly, NEUROG3-expressing cells first appear at 8.4WD in the ECAD + population and are found later on in the ECAD low population. Whether it is linked to their delamination remains to be demonstrated. This step is followed (at 8.6WD) by the expression of endocrine markers such as CHGA, NEUROD1 and INS. Of importance, endocrine cells expressed low but significant levels of ECAD. This perfectly fits with mouse data that indicate that during development ECAD function is necessary for proper aggregation of endocrine cells after delamination (Dahl et al., 1996). Taken together, our data demonstrate that ECAD levels are tightly regulated during specific steps of human pancreatic development.
We further refined specific pancreatic cell populations by using CD142 and SUSD2 found in our transcriptomic analysis as additional cell surface markers. CD142 has been proposed as a marker of pancreatic endodermal cells that also labels additional cell types (Kelly et al., 2011). SUSD2 was previously used as a marker to enrich NEUROG3 + from hPSC derived pancreatic cells and the human fetal pancreas (Liu et al., 2014). Here, with our set of markers we reconstructed human pancreatic cell differentiation ( Figure 7G). The full combination of markers was required for this reconstruction as none of the markers was specific to the different subsets (Figure 7-figure supplement 1). Our data indicate that GP2 + CD142 + pancreatic progenitors can either give rise to GP2 hi CD142 + acinar cells or enter the endocrine pathway and express NEUROG3 by turning off GP2 and CD142. Endocrine maturation further progresses by up-regulating SUSD2 and decreasing ECAD level ( Figure 7G). The first INS + cells were detected at 8.6WD in the E low population as SUSD2 + and later on as SUSD2 -. Recently, differences in gene expression and functionality were observed between fetal, neonatal and adult beta cells (Hrvatin et al., 2014;Jermendy et al., 2011;Blum et al., 2012). Moreover, heterogeneity between adult beta cells was also recently described (Dorrell et al., 2016;Bader et al., 2016). Whether INS + cells positive for SUSD2 represent a first wave of beta cells that could be poly-hormonal remains to be tested. Moreover, comparative analyses between INS + cells from E low SUSD2 + and E low SUSD2subsets will define the differences in their transcriptional factor networks.
FACS-based approaches have been used since the eighties in the hematopoietic field to dissect hematopoiesis (Spangrude et al., 1988). However, it has rarely been used in the field of pancreatic development and even less frequently with human fetal pancreases. In the pancreas, the majority of the cell sorting approaches was performed by analyzing fluorescent signals from tagged proteins derived from transgenic mice (Gu et al., 2004;Miyatsuka et al., 2014). While highly informative, this strategy cannot be used to dissect human pancreatic development. Antibodies against cell , E low GP2 -CD142 -SUSD2 + (E low SUSD2 + ) and E low GP2 -CD142 -SUSD2 -(E low SUSD2 -) subsets at 8.6 and 10-12WD. (C) 8.6, in the GP2 -CD142 + SUSD2 -(GP2 -CD142 + ), GP2 -CD142 -SUSD2 -(GP2 -CD142 -), E low GP2 -CD142 -SUSD2 + (E low SUSD2 + ) and E low GP2 -CD142 -SUSD2 -(E low SUSD2 -) subsets. (D, E) Single cell RT-qPCR Figure 6 continued on next page surface markers were used in a limited number of studies on pancreas development and mainly in rodents. It was found that a combination of CD49f and CD133 antibodies can be used to enrich fraction in mouse pancreatic progenitors expressing NEUROG3-expressing cells with some data on human fetal pancreas. However, the enrichment in NEUROG3 on human fetal pancreas was limited using this combination of markers (Sugiyama et al., 2007). More recently, SUSD2 was used as a marker to enrich NEUROG3 + cells in hPSC -derived pancreatic cells and in the human fetal pancreas (Liu et al., 2014). Our data confirm this point. However our single cell qPCR indicate that SUSD2 does not mark all, but only a subset of NEUROG3 + that co-expressed NEUROD1 and NKX2-2. Thus, by using a combination of antibodies against cell surface markers, we demonstrate that cell populations highly enriched in specific functions can be sorted from the human fetal pancreas at different stages of development, allowing the reconstruction of the differentiation program. Our model constitutes a key advancement in understanding human fetal pancreas development by mapping out the pattern of differentiation of the three main pancreatic lineages. We further refined our work for the endocrine pathway by describing for the first time, discrete stages of human pancreatic endocrine cell differentiation and showed that our development model also applies to the in vitro differentiation of hPSCs into pancreatic endocrine cells. To the best of our knowledge, this type of sideby-side comparison that demonstrates that in vitro hPSC differentiation mimics in vivo events has rarely been done.
In conclusion, we provide a novel way of approaching human pancreatic differentiation. Our work will be useful to fill the limited knowledge on human pancreas development. It should also pave the way for developing new cell therapies for diabetic patients.

Pancreatic dissection and cell suspension preparation
All experiments on human fetal pancreas were performed at INSERM Paris, France. Human fetal pancreases were isolated from surgical abortion done by suction aspiration between 7 to 13 weeks of development (Castaing et al., 2001;Capito et al., 2013;Scharfmann et al., 2014) in compliance with the French bioethics legislation and the guidelines of our institution. Approval was obtained from Agence de Biomedecine, the French competent authority along with maternal written consent. Ages were determined on the basis of time since the last menstrual period and hand and foot morphology. Fetal pancreases were micro dissected with forceps under a binocular magnifying lens, rinsed with Hanks Balanced Salt Solution (HBSS) from Gibco to remove contaminating blood cells and gently disrupted using forceps. Afterwards, pancreases were incubated for 5 min in collagenase V (0.5 mg/ml) (Sigma Aldrich) in HBSS in the presence of calcium and magnesium. Cells were rinsed in HBSS and then incubated for 5 min in trypsin (0.05%) (Gibco). Finally, cells were rinsed in HBSS supplemented with 20% Fetal calf serum (from Eurobio). Pluripotent stem cells were washed with PBS without Ca2 + and Mg2 + (Invitrogen, FRANCE) and incubated with TrypLE select for 1-3 min.

Maintenance and differentiation of human pluripotent stem cell lines
A human ESC (SA121) (Heins et al., 2004) obtained from Takara and two iPSC (SB Ad2.1 and SB3.1) (van de Bunt et al., 2016) obtained from the StemBANCC consortium were applied for this study. These cell lines had been confirmed to be pluripotent by evaluation of pluripotency marker expression, tri-lineage differentiation and karyotyping and tested negative for mycoplasma contamination. All three lines were cultured in mTeSR1 medium (StemCell Technologies) on hESC-qualified matrigel (Corning). Cells were passaged every 3-4 days or when confluent by dissociating to a single cell solution using TrypLE select (ThermoFisher). Single cells were seeded onto freshly coated Matrigel tissue culture flask in mTeSR1 containing of 5 uM Tiger (Rock inhibitor, Sigma-Aldrich) and medium was replenished daily. For differentiation of the hESC/hiPSC lines cells were dissociated to a single cell solution using TrypLE select and resuspended mTeSR1 with 5 uM Tiger. Cells were seeded at a concentration of 0.35 Â 10 6 cells/cm 2 onto growth-factor reduced matrigel CellBIND surfaces (Corning). Cells were incubated for 24 hr before the start of the differentiation. Differentiation to the pancreatic lineage was conducted as described in a previously published protocol  with the following modifications: CHIR99201 (Axon Medchem) was applied at 3 uM and 0.3 uM concentration for the first and second day, respectively of the definitive endoderm differentiation instead of MCX-928 used in the original protocol. MCDB131 medium (Life technologies) was used as basal medium throughout the differentiation instead of BLAR medium. Cells were not dissociated and incubated as clusters on air-liquid filters during stage 5; instead cells were kept in 2D cultures throughout the differentiation.

Flow cytometry
Following dissociation, cells were incubated with antibodies for 20 min in FACS medium (HBSS +2% FCS), then rinsed in FACS medium and re-suspended in FACS medium with Propidium Iodide (1/ 4000) (Sigma Aldrich) or DAPI solution (0,1 ug/ml) (BD Biosciences) to label dead cells. For intra-cellular staining cells were fixed for 5 min in 3% PFA and then rinsed in DPBS from GIBCO. Then cells were permeabilized in DPBS +3% BSA +0.3% Triton and incubated with antibodies overnight at 4˚C, rinsed and re-suspended in DPBS. For each antibody, optimal dilution was determined by titration.
For human pluripotent stem cell differentiation cells were sorted from three independent differentiations of the ADSB3.1 iPSC line. Cells were lysed in RP1 lysis buffer from the NucleoSpin RNA/ Protein purification kit (Macherey-Nagel) and stored at À80˚C until purification of RNA. RNAs were extracted using NucleoSpin RNA/protein purification kit. RNA was subsequently converted to cDNA using iScript cDNA synthesis kit (Biorad) according to manufactures instructions. RT-qPCR was performed on a MX3005P qPCR system (Agilent Genomics) using a fast-2-step protocol (first 95˚C for 1 min, then 40 cycles of 95˚C for 10 s, 60˚C for 25 s). The following TagMan primers were used: NEU-ROG3 (Hs01875204_s1), NEUROD1 (Hs01922995_s1), NKX2-2 (Hs00159616_m1), bActin (Hs01060665_g1) and HPRT (Hs99999909_m1).

Global transcriptomic analysis
Total RNA was obtained from Trizol-preserved (Invitrogen, France) sorted cells by chloroform extraction, quantified and assessed for quality by Agilent-2100 Bioanalyzer (Agilent, Santa Clara, CA). cDNAs were prepared and amplified using Ovation Pico WTA System V2 kit (NuGEN Technologies, San Carlos, CA, USA) and hybridized onto GeneChip Human Gene 2.0 ST Array (Affymetrix,Santa Clara, CA, USA). Quality Control and normalization (RMA) were conducted using Bioconductor Project software (https://www.bioconductor.org), which provides log2 transformed expression values. Data were extracted from raw CEL files using a custom GeneChip library file (CDF file) provided by http://brainarray.mbni.med.umich.edu/CustomCDF (Dai et al., 2005).

Transcriptomic statistical analysis
Two-sample comparisons were made using two-tailed Student t-test. Transcripts for which any p-value was above 0.05 were filtered out, leaving 6444 transcripts for further analyses. We compared expression of these genes in each cell population using Gene Set Enrichment Analysis (GSEA) software and the 'Biological Process' database of the Gene Ontology Consortium (Subramanian et al., 2005). Biological processes with FDR q-value <0.05 were considered significant.
Cut off to define the 'specific enriched genes': Each population (GP2 hi , GP2 + , GP2 -, E low and M) was compared to the other populations using twotailed Student t-test. Genes were considered as enriched in a specific population when overexpressed (p-value<0,05 and Fold Change > 2) in a population compared to any of the others. The list of the 'specific enriched genes' is displayed in supplementary file 1b and c.
Expression pattern of the enriched fetal genes in the adult pancreas: We use data from the single-cell RNA-seq (Segerstolpe et al., 2016). The normalized data are available on ArrayExpress (http://www.ebi.ac.uk/arrayexpress/ experiments/E-MTAB-5060). Next, we generated heatmaps displaying the expression of the fetal 'specific enriched genes' in the adult acinar, alpha, beta, epsilon, gamma and ductal cells.
Heatmaps were generated by the 'heatmap2' function from gplots R package (https://cran.r-project.org/web/packages/gplots/) on standardized log2 expression values, with Pearson correlation as the distance function. d'Excellence consortium Revive and to the Departement Hospitalo-Universitaire (DHU) Autoimmune and Hormonal disease. The research leading to the results on the hiPSC has received support from the Innovative Medicines Initiative Joint Undertaking under grant agreement n˚115439, resources of which are composed of financial contribution from the European Union's Seventh Framework Programme (FP7/2007(FP7/ -2013 and EFPIA companies' in kind contribution. This publication reflects only the authors' views and neither the IMI JU nor EFPIA nor the European Commission are liable for any use that may be made of the information contained therein. Work on the hESC line was done without the support from StemBANCC. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Major datasets
The following dataset was generated: Author (

Methods supplementary information Determination of the developmental stage
Weeks of development (WD) were determined based on the length and the anatomy of the foot and the hand. We also used the Human Embryo Resource website (https://embryology. med.unsw.edu.au/embryology/index.php/Embryonic_Development). Then, the precise age was calculated with the length of the foot and the following formula: Age = 4.11 X Foot Length + 5.60 (Manjunatha et al., 2012) from the Egyptian Journal of Forensic Sciences).
The variability between individual fetal pancreases at the same WD is displayed in Figures 2D, 5D-E and 7E-F. qPCR All qPCR are from three independent experiments derived from the sorting of three independent fetal pancreases.

Global transcriptomic analysis
We performed a global transcriptomic analyses on three independent pancreases at 9WD and three at 11WD.