Elimination of Tumorigenic Stem Cells from Differentiated Progeny and Selection of Definitive Endoderm Reveals a Pdx1⁺ Foregut Endoderm Stem Cell Lineage

Abstract

Embryonic stem cell (ESC) derivatives offer promise for generating clinically useful tissues for transplantation, yet the specter of producing tumors in patients remains a significant concern. We have developed a simple method that eliminates the tumorigenic potential from differentiated ESC cultures of murine and human origin while purifying lineage-restricted, definitive endoderm-committed cells. A three-stage scheme utilizing magnetic bead sorting and specific antibodies to remove undifferentiated ESCs and extraembryonic endoderm cells, followed by positive selection of definitive endoderm cells on the basis of epithelial cell adhesion molecule (EpCAM) expression, was used to isolate a population of EpCAM⁺SSEA1⁻SSEA3⁻ cells. Sorted cells do not form teratomas after transplantation into immunodeficient mice, but display gene and protein expression profiles indicative of definitive endoderm cells. Sorted cells could be subsequently expanded in vitro and further differentiated to express key pancreas specification proteins. In vivo transplantation of sorted cells resulted in small, benign tissues that uniformly express PDX1. These studies describe a straightforward method without genetic manipulation that eliminates the risk of teratoma formation from ESC differentiated derivatives. Significantly, enriched populations isolated by this method appear to be lineage-restricted definitive endoderm cells with limited proliferation capacity.

Introduction

Optimistic views of regenerative medicine have envisioned the use of stem cells as eventually curative for many types of human disorders, including ischemic heart disease, Parkinsonism, diabetes and many other degenerative or genetically deficient disease conditions. The initiation of the first FDA-approved clinical trial using human ESC-derived cells [1] has raised the anxiety of many scientists who fear that should this trial fail, it could endanger public acceptance and respectability of the entire field of embryonic stem cell research. Potential tumorigenicity of donor cells is a major concern. Adding to this unease is the recent report that a child with ataxia telangiectasia developed multifocal tumors of the brain and spinal cord 4 years after treatment with human neural stem cells [2] originating from at least two donors, even though the cells were relatively freshly derived from chromosomally normal fetuses.

Numerous published reports have examined the potential of differentiated cell types derived from mouse and human ESCs to repair non-human target organs in intact animals. Recent studies, however, have yielded both encouragement and caution, with restoration of function evident to some degree in many cases, but co-existing in others with the troubling finding that the grafts contained evidence of cancerous growth [3,4,5,6,7,8,9].

Although ESCs offer great promise in regenerative medicine in terms of both the variety of tissues obtainable and expansion potential of precursors, their slowly emergent clinical status reflects, in part, an inherent risk that arises due to the lack of understanding of the hazard of formation of teratomas arising from differentiated ESC-derivative populations. In view of the magnitude of this risk, selection schemes have been devised using genetically modified cell lines to either remove undifferentiated ES cells or purify ESC-derived populations that have been directed into specific lineages [10,11,12,13,4,14,15]. However, few, if any, of these experiments have exhaustively tested such cell populations for their tumor potential in vivo. In addition, despite genetic selection, the resulting populations still contained some undifferentiated ESCs [16] or produced obvious tumor growths in animal transplant tests [17,4,7]. An explanation for these observations may lie in the less than ideal stringency of a single selection step.

We desired to address the concern of residual tumorigenicity in one instance of high priority in the field of regenerative medicine: the derivation of differentiated pancreatic β cells from ESCs. Aiming to achieve efficient elimination of tumorigenicity of heterogeneous differentiated ESC-derived populations, we devised a 3-step selective process that appears to completely remove all tumorigenic cells from concurrently enriched definitive endoderm cells, which include pancreatic precursors and progeny, while simultaneously eliminating another troublesome contamination: extraembryonic visceral endoderm (VYS). Commonly generated during typical ESC differentiation cultures during which post-gastrulation germ layer derivatives are formed, VYS has overlapping genetic expression profiles with definitive endoderm including, for example, alpha fetoprotein, FoxA2, Sox17, Hnf4α, and Hnf1β [18,19], that can confound careful gene expression analyses and potentially lead to erroneous conclusions.

Effective methods for enriching DE committed cells would be of great value and have been pursued. To this end, several studies have taken advantage of the possible differential expression of CXCR4 in definitive versus visceral endoderm, and studied sorted CXCR4⁺ cells [20,21,22]. However, CXCR4 may not be an ideal marker to use for DE enrichment for the following reasons. First, some studies have shown that CXCR4 is also highly expressed in visceral endoderm [18], and in many neuronal and mesodermal cells at early stages [23]. CXCR4 is expressed in many non-endodermal cell populations including neural, vascular, cardiac and skeletal muscle satellite cells, and lymphopoietic, myelopoietic and hematopoietic stem cells (reviewed in [24]). Furthermore, CXCR4 expression appears to be downregulated in definitive endoderm after e8.5 of mouse embryonic development [23], thereby potentially allowing some highly desired endodermal cell types in ESC-derived cultures to escape selection. Lastly, selective techniques would ideally be able to simultaneously remove all residual cell populations that could give rise to teratomas, however, as CXCR4 is also expressed in undifferentiated ESCs (our unpublished observations and [25]) this is not likely to be the case in CXCR4⁺ selected cell populations.

We have based our selection scheme on classic markers that have been extensively characterized and found to be exclusively expressed on only certain cell types in the mouse embryo. The Stage-Specific Embryonic Antigens (SSEA), first described in 1978 by Knowles and Solter, are well known highly specific markers (SSEA1 and SSEA3) for undifferentiated mouse ESCs [26] or VYS endoderm cells [27], respectively. In addition, the cell surface marker Epithelial Cell Adhesion Molecule (EpCAM) has been recently described as exclusively expressed on definitive endoderm cells after e9.5 [18], is known to be expressed in fetal pancreas [28], and has a role in the morphological development of the pancreas [29]. Prior to e9.5, EpCAM is also expressed on inner cell mass (ICM) and VYS cells as well as ESCs, which are derived from ICM cells. We therefore reasoned that if we initially removed ESCs and VYS cells present in ESC-differentiated cultures by sequential negative selection steps, using Magnetic Activated Cell Sorting (MACS) employing anti-SSEA1 and anti-SSEA3 antibodies, we could then positively select from the remaining SSEA1⁻/SSEA3⁻ population only those cells expressing EpCAM, which should represent highly purified definitive endoderm cells. In this paper, we present the results of a unique 3-stage selection scheme to purify definitive endoderm cells, the in vitro characterization of selected cells and results of in vivo growth demonstrating their inability to form teratomas or tumor growths. In addition, we present surprising results that suggest a novel way to further differentiate and/or select from purified definitive endoderm cells only those precursor cells expressing PDX1, the hallmark for posterior foregut/pancreatic-committed gut endoderm.

Results and Discussion

Rationale for MACS selection strategy of differentiated ESC cultures

As definitive endoderm and pancreatic progenitors would be useful starting materials for generating replacement therapies for diabetes, we focused on isolating these cells from differentiated ESC cultures. In previous experiments, we determined that PDX1⁺ progenitor cells developed and remained present over an extended period in ESC cultures, thus facilitating their detection and removal from differentiated populations [30]. We began by testing antibody to EpCAM, restricted to definitive endoderm by E9 in the developing embryo, for its ability to include cells expressing PDX1, the first diagnostic marker for regional pancreatic specification. Virtually all PDX1⁺ cells in murine EB₇₊₂₁ cultures (7 days EB culture followed by 21 days post-plating) co-expressed EpCAM (Fig 1a). Notably, even small numbers of PDX1⁺ cells were typically surrounded by a broader field of EpCAM⁺ cells (Fig 1b). These results suggested that sorting for EpCAM expression could isolate definitive endoderm-committed cells that included PDX1⁺ foregut/pancreatic progenitors.

Fig 1.

Protein and transcript expression profiles of differentiated ESC mass cultures prior to sorting at EB₇₊₂₁. (a) Nuclear PDX1 and EpCAM are co-expressed in many areas of the unsorted culture; (b) in some regions, not all EpCAM⁺ cells express PDX1, yet the majority of PDX1⁺ cells do express EpCAM. (c) EpCAM is not present on cells that express K14, a marker of basal cells of stratified squamous epithelia which are common cell types in differentiated ESCs (top), or in a human keratinocyte stem cell culture (bottom); (d–g) EpCAM is not co-expressed with neural or mesenchymal markers (Vimentin, striated Myosin, β-Tubulin, Nestin); Scale bars, 50μm. (h) Temporal expression patterns of key mesendoderm and definitive endoderm genes indicate cultures are beyond gastrulation stage at the time of sorting, since T (brachyury) and goosecoid (Gsc) transcripts are nearly absent by EB₇₊₂₁, whereas Foxa2 and Sox17 remain highly expressed. Pdx1 transcripts are also more abundant in post-EB cultures. *, p ≤ 0.05

To determine the specificity of EpCAM staining in differentiating ESC cultures, we further examined cells for co-expression of EpCAM and markers of other non-endoderm cell types. No cells co-staining for EpCAM and cytokeratin 14 (CK14), a marker of ectodermally-derived basal epidermal cells prevalent in non-specific differentiated ESC cultures, were observed either in differentiating ESC cultures or in cells of an established keratinocyte stem cell line (Fig 1c). Likewise, EpCAM was not co-expressed with mesoderm or neuronal markers including vimentin, myosin, β-tubulin and nestin (Fig 1d–g), results that agree with those previously obtained using quantitative RT-PCR (QPCR) to analyze embryonic sites of EpCAM expression [31]. However, in accordance with other published reports [32,33], we observed abundant co-staining for EpCAM and either SSEA-1 or SSEA-3/4, which mark undifferentiated murine ESCs or VYS cells, respectively, in differentiating mouse ESC cultures. For example, in FACS analyses of early cultures at EB₇₊₈ we found that ~60% of EpCAM⁺ cells expressed SSEA1, and ~20–25% of EpCAM⁺ cells co-expressed SSEA3. Thus, if cultures were sorted for EpCAM alone, the majority of cells isolated at this stage would represent undifferentiated, highly tumorigenic ESCs or contaminating primitive endoderm VYS cells that express many genes in common with definitive endoderm.

We therefore devised a multistep isolation strategy that first involved removal of both SSEA1⁺ (undifferentiated ESC), and SSEA3⁺ (VYS) cells, followed by positive selection for EpCAM⁺ cells (Fig 2a). This protocol has been repeated over 50 times; typically, about 10% of the starting material is selected and viability of sorted cells is generally greater than 90%. To ascertain whether MACS sorting enriches for EpCAM⁺ cells, we examined cells before and after MACS sorting by flow cytometry. Whereas less than 50% of the pre-sort population at EB₇₊₂₁ is EpCAM⁺ (which includes ESC and VYS cells), virtually the entire population is EpCAM⁺ after sorting (Fig 2b). Oct4 expression, characteristic of undifferentiated ESCs, appreciably diminishes during cell differentiation and is further significantly reduced by sorting (Fig 2c). Transcript levels of pluripotency markers Nanog and Sox2 were also decreased after sorting at 3 weeks differentiation (20-and 25-fold, respectively, data not shown). Antibody staining revealed that there was very limited or no expression of OCT4, Nanog, or SOX2 protein following sorting (Fig S1). To determine if the protocol successfully depleted VYS cells, pre- and post-sorted cells were stained for SSEA4, an additional murine marker of this lineage. No SSEA4-stained cells were observed in post-sorted cells (Fig 2d). Overall, these results demonstrate that MACS sorting results in a highly purified EpCAM⁺ population that is significantly depleted of ES and VYS cells.

Fig 2.

Isolation of EpCAM⁺SSEA1⁻SSEA3⁻ cells from mass differentiated ESC cultures by MACS sorting results in highly enriched EpCAM⁺ population diminished in undifferentiated ESC and extraembryonic endoderm transcripts. (a) Flow diagram of three-step sorting strategy. In the first sort, SSEA1 antibody removes undifferentiated ESCs expressing SSEA1; next, extra-embryonic endoderm is depleted using SSEA3 antibody. A third sort selects for remaining EpCAM⁺ cells. (b) FACS analysis of MACS-sorted cells indicates a nearly homogeneous population (96%) of EpCAM⁺ cells from an initial population that is approximately 50% EpCAM⁺. (c) Transcript levels of Oct4, indicative of undifferentiated ESCs, decline gradually during in vitro differentiation and are further significantly reduced by MACS sorting. *, p ≤ 0.05. (d) SSEA4, indicative of extraembryonic endoderm, was found on a subset of cells in unsorted EB₇₊₁₃ cultures, but not on similar stage cells sorted to select the EpCAM⁺SSEA1⁻SSEA3⁻ population and then cultured for two days. Scale bar, 50 μm.

Tumorigenicity of MACS sorted cells

To determine the teratoma-forming potential of sorted EpCAM⁺ SSEA1⁻ SSEA3⁻ cells in vivo, we injected freshly-sorted cells into subcutaneous abdominal sites in NOD/SCID mice (Fig 3a). Undifferentiated ESCs, unsorted populations of differentiated ESCs, and column-discard EpCAM⁻ cell populations were used as controls. As expected, large, rapidly-growing teratomas arose uniformly after injection of undifferentiated ESCs or unsorted cells of the same differentiation stage, and in a few cases from the EpCAM⁻ population as well, generally reaching >15 mm by 3–5 weeks post-inoculation. The percent of inocula that resulted in tumor growths exceeding 3 mm during 6 weeks following injection of different populations is shown in (Fig 3b). Strikingly, tumors never formed in animals following injection of freshly-sorted EpCAM⁺ cells, even after up to 23 weeks (Fig 3b, c).

Fig 3.

Tumorigenicity of EpCAM⁺ SSEA1⁻SSEA3⁻cells. (a) ESCs were differentiated to the EB₇₊₂₁ stage, selected for EpCAM⁺SSEA1⁻SSEA3⁻ cells according to the strategy in Fig 2a, and then sorted cells were immediately injected into subcutaneous (SC) sites in immunodeficient NOD/SCID mice and allowed to remain for 6–24 weeks. Photograph shows a representative mouse in which the unsorted, differentiated population grew into a teratoma by 6 weeks, but the sorted cell population did not. (b) Bar graph indicating the percent of inocula forming tumors at 6 weeks (or up to 24 weeks for EpCAM⁺ sorted cells) after SC transplantation of 10⁶ cells of either undifferentiated ESCs, sorted EpCAM⁺ cells, sorted EpCAM⁻ cells, or unsorted cells previously differentiated to the EB₇₊₂₁ stage (Unsorted EB₇₊₂₁) into recipient mice. Sorted EpCAM⁺ vs. Undifferentiated ESC, p<0.0001; Sorted EpCAM⁺ vs. Unsorted EB₇₊₂₁, p<0.0001; Sorted EpCAM⁺ vs. Sorted EpCAM⁻, p= 0.0549; Sorted EpCAM⁻ vs. Undifferentiated ESC, p<0.0008; Sorted EpCAM⁻ vs. Unsorted EB₇₊₂₁, p= 0.0152; Undifferentiated ESC vs. unsorted EB₇₊₂₁, p= 0.44; overall, p<0.0001. (c) Kaplan-Meier curve indicating the proportion of animals free of tumors, defined as presence of a nodule 3 mm or larger in diameter, at various times after transplantation of 1 × 10⁶ cells SC. Sorted EpCAM⁺ vs. Undifferentiated ESC, p< 0.0001; Sorted EpCAM⁺ vs. Unsorted EB₇₊₂₁, p<0.0001; Sorted EpCAM+ vs. sorted EpCAM⁻, p= 0.0117; Sorted EpCAM⁻ vs. Undifferentiated ESC, p<0.0001; Sorted EpCAM⁻ vs. Unsorted EB₇₊₂₁, p= 0.0091; Undifferentiated ESC vs. unsorted EB₇₊₂₁, p= 0.0025; overall p<0.0001. N differs between (b) and (c) in two categories because some animals did not have their injection sites measured serially.

An ongoing challenge in regenerative biology is the difficulty in controlling the heterogeneous nature of differentiated progenies that are typically derived from more primitive, pluripotent stem cells. In the case of ESCs, not only do differentiated cultures most frequently contain multiple lineages, but they can also contain residual undifferentiated ESCs [9]. Although it is generally believed that extended in vitro differentiation results in decreased teratoma formation, previous studies have found that ESCs differentiated for even extended periods of time are still able to form teratomas [9]. Furthermore, cultures differentiated using specific growth factors into predominantly definitive endoderm, including pancreatic progenitors, nonetheless could form teratomas when transplanted into the epididymal fat pad [5]. A promising approach to overcome both the problems of heterogeneity and tumorigenicity is to transplant cells after purification by sorting [34]. Although genetic-based selection techniques have been very effective at eliminating teratoma formation [7], genetic modification of stem cells introduces additional complexities and risks. In contrast, using MACS sorting of naturally occurring markers, we have been able to purify a population of cells from genetically unaltered ESCs that achieves two important aims: removal of all teratoma-forming potential and selection of definitive endoderm-committed cells, as further documented below.

Transcript and protein analysis of unsorted versus EpCAM⁺ sorted cells

To characterize the enrichment achieved by the MACS sorting protocol, we began by examining the differentiation stage attained in unsorted cultures at 1–3 weeks after EB plating. Brachyury (T), Goosecoid (Gsc), Sox17, Foxa2, and Pdx1 represent transcripts expressed during gastrulation or early periods of endoderm development. RT-PCR analysis revealed that T and Gsc levels peak during EB formation and subsequently decline in monolayer culture after EB plating. Conversely, Foxa2 and Sox17 transcripts increase after plating and are maintained at high levels thereafter, and Pdx1 levels increase postplating (Fig 1h). This temporal sequence is reminiscent of a pattern of endoderm gene expression observed during embryogenesis: T, initiated during early mesendoderm commitment, is not expressed in visceral endoderm and is downregulated as cell fate becomes restricted to definitive endoderm; Gsc expression begins at e6.5 in the node and early definitive endoderm, while Sox17 transcripts are present in both extraembryonic and definitive endoderm. Sox17 expression begins in the earliest e7.5 foregut and gradually strengthens in mid- and hind-gut regions before disappearing from the gut after e9.0 [35]. Foxa2, found in both primitive and definitive endoderm, is generally maintained in definitive endoderm-derived tissues throughout adult stages, including the pancreas [36]. Pdx1 begins to be expressed at e8.5 and persists in pancreatic β and some duodenal cells throughout life. The pattern observed suggests that endoderm present in later ESC cultures is most comparable to post-gastrulation embryonic endoderm in which gut regionalization has begun.

To evaluate the capacity of sorting for PDX1enrichment, the frequency of PDX1⁺ cells was determined by flow cytometric analysis in sorted and unsorted populations at EB₇₊₂₁. The percentage of PDX1⁺ cells can be highly variable in spontaneously differentiated cultures before sorting, although enrichment generally occurs to the same extent, approximately 2–5 fold (Fig 4a). PDX1 expression may vary in presort populations because spontaneous differentiation is dependant upon relatively rare, variable events. In a complex multistep process, it should be anticipated that, occasionally, differentiation to the desired lineage may be poor. In addition, PDX1⁺ cells are very efficiently included in the sorted EpCAM⁺ population, as FACS analysis of pre- and post-sorted PDX1⁺ cells indicates that virtually all PDX1⁺ cells are also EpCAM positive (Fig 4b).

Fig 4.

Analysis of sorted cells. (a) FACS determination of the percentage PDX1⁺ cells in presort and post-sort populations in 4 independent experiments. (b) FACS dot plots of data from experiment 1 (Fig 4a) showing co-expression of PDX1 and EpCAM in pre-and post-sorted cells, demonstrating efficient recovery of PDX1⁺ cells in the EpCAM⁺ sorted population. (c) Levels of gene transcripts characteristic of different embryonic germ layers and ESCs in sorted cells compared to unsorted cells, showing enrichment of endoderm markers relative to markers of other germ layers. *, p ≤ 0.05. (d) Phase contrast image of cells 24 hours after sorting (upper), or confluent cultures 4 days after sorting (lower), grown in medium containing 10% SR and FGF10 50 ng/ml. (e) Cells at 4 days post-sorting virtually all co-stain for Sox17 and Foxa2, consistent with a culture composed solely of definitive endoderm cells. Scale bars, 100 μm (d), 50 μm (e).

To determine whether MACS sorting simultaneously eliminates cells of other germ layers while enriching for definitive endoderm, we analyzed levels of transcripts characteristic of mesodermal and neural lineages and compared them to endoderm marker transcripts (Fig 4c). Vimentin expression, indicative of mesenchyme, was reduced 20 fold in sorted cells, while Nkx2.5 transcripts, specific for cardiac muscle, and neural/mesenchymal markers Sox1, β-3tubulin and Nestin decreased 2 to 5-fold. Conversely, endoderm-specific markers including Surfactant Protein C (SftpC), expressed in lung progenitors, and Transthyretin (Ttr) and Tyrosine aminotransferase (Tat), markers of hepatic development, increased by 3 to 5-fold. We also observed a 500 fold increase in the hindgut marker, Caudal box 1 (Cdx1). Genes characteristically expressed in the distal foregut including Prox1, specific for the liver/pancreas region [37], as well as pancreatic markers Pdx1, YY, Ngn3, and Insulin (Ins1), were also expressed at higher levels (4–10 fold) in selected EpCAM⁺ cells compared with the pre-sort population. Taken together, these results support the notion that positive sorting for EpCAM, together with removal of SSEA1- and SSEA3-expressing cells, enriches for endoderm-restricted cells.

Freshly-sorted EpCAM⁺ cells were morphologically homogenous and could be expanded in MES medium minus LIF supplemented with 10% SR and FGF10, the latter routinely included as it has been previously shown to promote proliferation of embryonic pancreatic precursors [38]. Confluent monolayers formed after 2–5 days (Fig 4d), indicating both survival and prompt cell division in culture following the sorting process. Nearly all sorted cells co-stained for FOXA2 and SOX17 (Fig 4e), key proteins expected in early definitive endoderm.

Differentiation potential of sorted EpCAM⁺ cells

In vitro differentiation capabilities of sorted cells were examined by switching cells to a series of serum-free differentiation media (SFDM; Fig 5a). Dramatic changes in morphology occurred rapidly in confluent monolayers following coating with Matrigel and renewal with SFDM/ITS including nicotinamide, Exendin-4 and FGF10. By 24 hours, cells were clearly aggregating (Fig 5b) and by 3–7 days became organized into tubular/cystic structures (Fig 5c) that stained intensely for EpCAM (Fig 5e). Abundant cysts evident 8–10 days after Matrigel addition (Fig 5d) were comprised of predominately simple epithelium (Fig 5d inset). PDX1 expression was particularly notable in all such structures (Fig 5f, g), which were also strongly EpCAM⁺ (Fig 5g).

Fig 5.

Differentiation of sorted EpCAM⁺ cells in vitro. (a) Differentiation scheme: sorted cells were grown to confluency in medium containing 10% Serum Replacement (SR) and FGF10, then coated with Matrigel and switched to serum free differentiation medium (SFDM) containing ITS (SFDM/ITS) FGF10, Nicotinamide, and Exendin 4 for 8–10 days (see Materials and Methods for details and doses). They were then grown for 7 additional days in SFDM as above except that B27 was substituted for ITS (SFDM/B27) and HGF and BTC were added. (b) Phase contrast image of cells beginning to aggregate 24 hours after Matrigel addition. (c) Phase contrast image of cultures 3 days after Matrigel addition, showing formation of tubular/cystic structures. (d) Numerous cystic structures (outlined in white stippling) developed after 8–10 days in SFDM. (inset) H&E stained tissue section of structures at 10 days in SFDM medium, primarily composed of simple cuboidal epithelium. (e) Cells incorporated into tubules are strongly EPCAM⁺. (f) Cystic structures, observed as a hollow ball in a single confocal slice (inset shows stacked image), are entirely PDX1⁺ and (g) EpCAM⁺ after 10 days in SFDM. Three-dimensional structures express pancreatic progenitor proteins after 8 days of differentiation in SFDM including (h) Hnf4α, (i) Hnf6, (j) Prox1 and (k) Sox9. Scale bars, 100 μm (b,c,d), 50 μm (d inset, e–k). (l) Gene transcript levels of sorted cells after differentiation in SFDM for 8 days compared to freshly-sorted cells demonstrate stable expression of pancreatic endoderm markers and an increase in Pdx1, Pax4 and Nkx2.2 transcripts. Pax4 was undetected in post-sort d0 cells; a Ct value of 40 was used in calculating the fold change.

Compared to freshly sorted EpCAM⁺ cells, the gene and protein expression profile of cystic structures suggested further differentiation into prospective foregut or pancreatic bud-like epithelium. Protein markers of committed pancreatic bud epithelium including HNF4α, HNF6, PROX1 and SOX9 were highly expressed in 8d cystic structures (Fig 5h–k), in addition to PDX1. Transcript levels of these genes were quite abundant in cystic stages, as gauged by low ΔCt values (ΔCt ≤ 5), as well as in post-sort d0 cultures (Fig 5l). Although transcript levels of genes did not change significantly after sorting, except for increased Pdx1, Pax4 and Nkx2.2 (Fig 5l), the amount of protein detected increased greatly. Very few cells in early monolayers were positive by immunofluorescence staining for any of these markers (Fig S2), indicating that differentiation over this interval may have involved changes in gene translation rather than transcription; alternatively, changes in ubiquitination or protein degradation could have occurred. In contrast to high levels of pancreatic progenitor transcripts, endocrine specification transcripts, such as Ngn3, Nkx6., and MafA were low in abundance (ΔCt ≥ 10) and did not increase substantially under SFDM/ITS conditions (Fig 5l). Interestingly, MafB transcripts were easily detected (ΔCt = 6) and Pax4 and Nkx2.2 transcripts increased significantly following culture in SFDM/ITS.

After an additional 7 days in SFDM/B27 containing HGF and BTC, gene expression analysis showed sustained enrichment of pancreas progenitor transcripts (Pdx1, Prox1, Sox9), whereas the expression of markers of other, non-pancreatic endoderm regions including thyroid (Nkx2.1 and Thyroid stimulating hormone receptor [TshR]), liver (Tat), and hindgut (Cdx1 and Hoxb13) remained very low or declined (Fig S3). Transcript and expression levels of EpCAM remained high throughout the entire differentiation period (not shown). Expression of the early lung marker Nkx2.1 decreased, while expression of the differentiated lung marker SftpC increased (Fig S3). Hence, further differentiated cultures showed sustained maintenance of high levels of pancreatic bud epithelium markers and reduced anterior foregut, liver and hindgut marker expression, indicating a potential shift in the profile of regional endoderm expression towards pancreatic lineages. However, despite their progenitor phenotype, differentiated sorted cells did not increase their endocrine-specific gene expression under the in vitro differentiation conditions tested. Although purified EpCAM⁺ cells expressed high levels of Pdx1, Prox1, Sox9, Hnf6, and Hnf4α, they lacked significant expression of Ptf1a, a transcription factor considered to be essential for in vivo pancreas specification. Indeed, this set of markers could indicate a posterior foregut progenitor lineage that has potential for both liver and pancreas. Protocols to differentiate these cells into insulin⁺ cells have yet to be devised. Achieving more complex levels of differentiation may require more robust growth factor differentiation protocols than those employed here. Additionally, development of pancreatic epithelia in isolation in vitro may require signals from surrounding cells, including mesenchyme. Recombining the purified population with other cell types prior to transplantation is another method to further investigate the potential of these cells to achieve pancreatic cell fates. Clearly, close monitoring would be essential to ensure that additional culture manipulations or introduced cell interactions did not result in altered tumorigenic status of the sorted cells.

Although numerous different growth factor strategies have been developed to derive β-cell-like cells from ESCs, thus far they have yielded relatively small numbers of pancreatic endocrine cells in vitro [39,40,41,42,43,44,7]. Interestingly, superior results to date have been obtained by injecting heterogeneous populations of endoderm-committed, though not necessarily endocrine-producing cells, into epididymal fat pads [5], suggesting a better strategy may be to obtain discrete populations of pancreatic progenitors that can be further differentiated in vivo. Our system may contribute to defining those signals that complete the final differentiation steps to mature β-cells from progenitor cells.

Development of non-tumorigenic foregut endoderm cells in vivo

Tumors formed after injection of either undifferentiated ESC or differentiated, unsorted cells were typical teratomas, including neural, mesodermal and gut structures (Fig 6a). No similar teratomas ever developed from EpCAM⁺ sorted cells. However, in approximately 40% of animals injected directly with fresh-sorted cells having no culture interval, small benign nodules arose that never exceeded 3 mm, even after 6 months or more of residence in vivo. All were similar, consisting solely of glandular epithelia lined by simple epithelial cells (Fig 6b,c). Some cells lining the cysts contained mucin, characteristic of intestinal epithelial cells and/or ductal cells of the pancreatico-biliary system (Fig 6d). Remarkably, the majority of cells lining the cysts expressed PDX1 in addition to EpCAM by immunohistochemistry (Fig 6e,f), with no insulin or glucagon detected (not shown), thereby closely resembling e13.5 pancreatic ducts that show similar co-expression of PDX1 and EpCAM (Fig 6g). A gene expression analysis of 5 independent nodules for a panel of endoderm and pancreatic genes showed that all nodules expressed high transcript levels (low ΔCt values) of YY, a marker specific for definitive foregut endoderm in the early embryo [45], and Pdx1, the first pancreas transcription factor, that were significantly enriched over levels present in the fresh-sorted inoculum cells (Fig 6h). Moreover, the nodules showed strong similarities in their overall expression pattern (Fig 6i), including low or undetectable levels of Oct4, Ngn3, Ptf1a, Cdx1, and Ins1 transcripts, and modest amounts of Sox2 and Prox1.

Fig 6.

In vivo differentiation of EpCAM⁺ cells. Freshly-sorted cells were injected SC into NOD/SCID mice and resulting growths were examined by histological staining, immunohistochemistry, and QPCR. (a) H&E stained tissue section of typical teratoma formed after transplantation of either ESCs, unsorted differentiated EB₇₊₂₁ cells, or sorted EpCAM⁻ cells. (b) H&E stained tissue section of a 1mm nodule formed after transplantation of EpCAM⁺ cells, composed of numerous cystic/glandular structures. (c) Higher magnification of area in 6b, showing simple cuboidal epithelium lined cysts. (d) Mucicarmine staining of a similar section, showing stained intracellular vacuoles (arrow) indicating mucin containing goblet cells such as those found in intestine, bronchial or pancreatico-biliary ductal cells. (e, f) Immunohistochemical analysis of two different nodules derived from inoculation of EpCAM⁺ cells reveals that virtually all EpCAM⁺ (green) epithelial cells lining the cysts co-stain for PDX1 (red). (g) Immunohistochemical staining of embryonic day 13.5 mouse pancreas sections illustrates the similarity of normal embryonic pancreatic tissue to the cystic/glandular structures derived from EpCAM⁺ cells, with corresponding co-staining of PDX1 (red) and EpCAM (green) in most cells. Scale bars, 500 μm (a), 200 μm (b), 50 μm (c–g). (h) Gene transcript levels of five individual primary nodules formed after SC injection of EpCAM⁺ cells. All nodules contain high levels (low ΔCt) of PDX1 and YY transcripts, signature genes for posterior foregut cells, and are highly enriched above post sort levels. (i) Fold change of transcript expression of all nodules relative to the mean. 19.00 was used as the Ct cutoff value. Nodule 2 did not yield enough material for analysis of all genes.

Morphologically identical cystic nodules were derived from three different ESC lines (R1 [46], D3, and YC5 [47]) following injection of fresh-sorted cells, suggesting a generalized basis exists for this phenomenon. We considered that a unique class of PDX1⁺ precursor cells might exist among cells of the sorted population having extensive capacity for cell division, but only a limited potential for growth in vivo. We tested this hypothesis by culturing cells in vitro from nodules generated from fresh-sorted cells in an effort to establish a PDX1⁺ cell line.

Candidate foregut stem cell line established in vitro from nodule-derived cells

A small nodule developed from MACS-sorted cells injected 6 months previously was minced and explanted into tissue culture. The outgrowth cells exhibited stem cell-like morphology (Fig 7a), displaying a high nuclear-to-cytoplasmic ratio and forming compact colonies, but unlike ESCs, were uniformly PDX1⁺/OCT4⁻ (Fig 7b). These cells have been in continuous culture for over seven months without signs of senescence and were readily cloned (not shown); a selected clonal population was uniformly HNF4α⁺ and K19/K7⁺ (Fig 7c,d) as well as PDX1⁺, EpCAM⁺ and DBA⁺ positive (not shown), an expression pattern consistent with that expected of a posterior foregut population. On the other hand, definitive endoderm markers, FOXA2 and SOX17, were not expressed in all cells and were not always co-expressed (Fig 7e) in either cloned or early uncloned cells. This pattern, in contrast to uniform co-expression of these markers in cells cultured from the time of sorting, suggests a phenotypic shift (or selection) in vivo, resulting in a more mature committed endoderm fate in nodule-derived tissue, since SOX17 levels progressively decrease along the gut tube during early development and the gene is not expressed in the gut after E9 [35], whereas FOXA17 remains expressed. QPCR analysis showed a gene expression profile compatible with that of an endoderm-restricted population, depleted of hallmark ESC, neural and mesodermal gene transcripts and enriched for endoderm genes (Fig 7f), compared to ESC levels. Regional endoderm specification is suggested by the high relative expression levels of mid/foregut genes, in particular Pdx1 and Cdx2, in contrast to low expression of more anterior endoderm genes, including Nkx2.1, SftpC and Pax8; and posterior endoderm genes, such as Cdx1, and Hoxb13. We measured the total amount of endoderm gene transcripts present in cells by determining Δ Ct values (low Δ Ct values indicate high expression) for a panel of pancreas- and foregut-restricted gene transcripts, which revealed that genes most highly expressed included Gata4, Pdx1, Cdx2, Foxa2, Hnf4a, Sox17, Sox9, Hex, Hnf1β and Hlxb9 (Fig. 7g). On the other hand, very low or undetectable levels of Ngn3, Nkx6.1, Nkx2.2, Ptf1a, MafA, Amylase, and Ins1 (dCT ≥ 12) were expressed, thus placing the cells in a differentiation hierarchy evocative of early posterior foregut endoderm. Clearly, testing the full range of differentiation capabilities of these cells is of paramount importance.

Fig 7.

Analysis of cells derived from an EpCAM⁺ primary nodule. (a) Phase contrast image of cells that grew out of nodule. Nodule was removed 23 weeks after sorted EpCAM⁺ cells were inoculated into immunodeficient mice, minced and plated onto irradiated fibroblasts. Cells resemble undifferentiated ESCs in morphology (inset), but the growth pattern differs in that they tend to form lacunae in the center of spread-out colonies and do not multilayer. (b) Immunohistochemical analysis demonstrated cells are PDX1⁺ and Oct4⁻, unlike ESCs. Cells also stained for other endoderm markers including Hnf4α (c), ductal cytokeratins K19 and K7 (d), and Foxa2/Sox17 (e). (f) Transcript analysis of ESC, neural, mesoderm and endoderm-restricted genes shows that compared to ESCs, the expression of pluripotency genes (Oct4, Sox2, Nanog) is significantly decreased whereas the expression of foregut endoderm-restricted genes (Pdx1, Cdx2,) is enhanced. As expected, Sox17 and Foxa2 transcripts are expressed at higher levels than in ESCs. The expression of neural (Sox1), mesoderm (Nkx2.5, Meox1), primitive streak (T, Gsc) and hindgut genes (Hoxc4, Hoxb13, Cdx1) are either not detected or show reduced expression in these cells compared to ESCs. Liver markers (Afp, Tat) are also more abundantly expressed in these cells than in ESCs, consistent with being a posterior foregut-restricted cell population. (g) QPCR analysis of EpCAM⁺ primary nodule-derived cells, or candidate FGSCs, of a panel of genes whose expression characterizes various stages of pancreas development Transcript levels of the highest transcribed genes are presented; other genes were less abundant or not detected. (h) Immunohistochemical analysis of small secondary nodule derived from candidate foregut stem cells that had been inoculated into immunodeficient mice shows re-establishment of the cystic/glandular structures which co-stained for PDX1 and EpCAM, similar to the parent or primary nodule. (i) Secondary EpCAM⁺ cysts express the H2k^b haplotype of the injected cells. Host NOD/SCID mice are H2K^d. Scale bars, 50 μm (a–e) and 100 μm (h, i).

One hallmark of a stem cell line is the ability to be propagated while maintaining a stable progenitor phenotype. To address this, we evaluated the expression levels of 23 genes, including Oct4, endoderm, pancreatic, and liver genes over a 7 week period of approximately 10 passages (Fig S4). Stable expression of definitive endoderm and posterior foregut endoderm markers indicates derivation of a liver-pancreas restricted posterior foregut stem cell-like population (Foregut Stem Cell or FGSC). Unlike previous efforts to derive endoderm-committed proliferative cells from ESCs [48,7], these FGSCs appear to be non-tumorigenic: when cells were injected subcutaneously into immunodeficient animals, none (0/9) formed large tumors, while again, small benign nodules developed in some animals. Secondary nodules were morphologically indistinguishable from primary nodules, exhibiting multiple small cystic structures composed of PDX1⁺/EpCAM⁺ cells (Fig 7h). Figure 7i demonstrates that secondary nodules derived from candidate FGSC grown in NOD/SCID (H-2K^d) mice expressed an MHC Class I antigen characteristic of the originating ESC line (H-2K^b) but not the host mouse.

Our working hypothesis is that nodules originate from rare cells present in the sorted population that are relatively undifferentiated but endoderm-biased. We postulate that this rare population is instrumental to the formation and detection of in vivo growth of FGSCs revealed through primary nodules, and furthermore, that FGSCs represent an intermediate stage between pluripotent ESCs and fully pancreas committed cells. It is possible that extended (6 months) time in vivo enhanced further specification or stabilization of PDX1 expression within some sorted cells forming the nodule. Despite uncertainty over their origin, FGSC cultures represent prime material to investigate steps necessary to become pancreas, liver and possibly stomach or duodenal cells.

Application of MACS selection strategy to HuESCs

We began to apply a similar MACS selection strategy to human ESC-derived differentiated cell populations by modifying the selection strategy; SSEA antibodies were switched to account for the differing SSEA antigen expression profile of human ESCs and VYS cells compared to their mouse counterparts (Fig. S5a). PDX1⁺ cells in human unsorted differentiated cultures were also EpCAM⁺ (Fig S5b), suggesting the scheme was likely to result in endoderm progenitor enrichment. Similar to mouse ESC cultures, a minority of presort cells expressed PDX 1or EpCAM. FACS analysis of sorted human EpCAM⁺ SSEA1⁻ SSEA4⁻ cells showed enrichment of human EpCAM⁺ cells comparable to that obtained with mouse cultures (Fig S5c). Selected EpCAM⁺ cells could grow in culture into homogeneously epithelial sheets (Fig S5d), and were enriched for definitive endoderm-restricted gene transcripts while being relatively depleted of neural and mesodermal transcripts (Fig S5e).

Conclusion

We have outlined a simple procedure for eliminating the tumorigenic potential from differentiated ESC populations of both mouse and human origin. MACS-sorted SSEA1⁻/SSEA3⁻/EpCAM⁺ cells never formed teratomas. Significantly, the enriched cells have limited proliferative capacity in vivo and are lineage-restricted; the gene expression pattern of sorted cells indicates posterior foregut regional endoderm specification. Freshly sorted cells can be amplified and differentiated in vitro to express key pancreatic progenitor proteins including PROX1 and SOX17. The sorted definitive endoderm population includes or gives rise in vivo to PDX1+ putative FGSC cells, which are capable of extensive proliferation and express a repertoire of genes indicative of endoderm-committed progenitor cells. Non-genetic methods to remove teratoma-forming potential that also enable isolation of desirable cell types may reduce therapeutic risks of using ESC-derived populations.

Materials And Methods

Cell Maintenance and Differentiation

Mouse ESC lines used were D3, purchased from ATCC, and R1 and YC5 ESC lines, generously provided by A. Nagy. Undifferentiated ESCs were maintained as previously described [30]. Cells sorted by MACS were subsequently cultured at 4 × 10⁵ cells per 15 mm well in MES medium without LIF plus 10% Serum Replacement (SR) and FGF10 (50 ng/ml). 10% FCS was added for 24 hrs to facilitate cell attachment and thereafter replaced with 5 μg/ml fibronectin; medium was changed daily. Confluent monolayers formed after 2–4 days were differentiated by coating with 10μl/well Matrigel (Collaborative Research-BD, cat. #354234) and feeding with a serum-free differentiation medium (SFDM) composed of DMEM/F12 (8mM glucose) containing 0.1 mM non-essential amino acids (NEAA), 2 mM glutamine, 0.11 mM β-mercaptoethanol, ITS (5 μg/ml insulin, 5 μg/ml apo-transferrin, 5 ng/ml selenite), 0.2% BSA, 10 mM nicotinamide, 10 nM Exendin-4 and 50 ng/ml FGF10 (R&D Systems). 8–10 days after Matrigel addition, cells were grown for another 7 days with SFDM containing 1% B27 (Invitrogen) instead of ITS (SFDM/B27), 10ng/ml HGF (R&D Systems) and 1 μM Betacellulin (BTC) (R&D Systems).

Subcutaneous nodules from sorted EpCAM⁺ cells were explanted by cleaning the nodule in PBS under a dissecting microscope, finely mincing the cleaned tissue with opposing scalpels, and placing it on an irradiated fibroblast feeder layer in medium consisting of MES minus LIF containing 20% SR, 1% B27, and 50 ng/ml FGF10. Within 1–2 weeks, areas resembling stem cells spread from the explant. The cell line described originated from differentiated D3 ESCs. Initially, cells were transferred by manual selection using a pulled Pasteur pipette. The addition of a ROCK inhibitor (Y-27632, Sigma-Aldrich) significantly aided survival during the transfer of early passages. Later passages were accomplished using limited dispase digestion, as single cells did not survive transfer.

MACS Separation

Cell separations were performed using MACS LD and LS columns according to the protocol of the manufacturer (Miltenyi Biotec), except that azide was not included and the incubation/wash solution was ice-cold DMEM/HEPES with 10% FCS.

Transplantation Studies

Teratoma formation was evaluated by injecting 1–2 × 10⁶ cells subcutaneously into 6–8 week old male NOD/SCID mice (Jackson Laboratory, Bar Harbor, ME). The maximal diameter of the masses was measured every 2–3 days for up to 23 weeks. In some cases, small nodules or large tumors were harvested and processed for immunofluorescence staining, hematoxylin and eosin staining, or RNA isolation.

Supplementary Material

01. Fig S1.

MACS-sorted cells do not express markers of undifferentiated ESCs. MACS-sorted cells were stained with antibodies to Nanog (top), Sox2 (middle) or Oct4 (bottom), as well as DAPI, to mark nuclei. Undifferentiated ESCs were stained in parallel as a control. Most sorted cells do not express any of these three proteins, although rare sorted cells do express Sox2. Scale bars, 100 μm.

02. Fig S2.

Protein expression in sorted EpCAM⁺ cells, 2–4 days post sort in vitro, before the addition of SFDM. Sorted cells in monolayers do not express pancreatic progenitor proteins including Pdx1, Hnf4α, Hnf6, Prox1 and Sox9. Note that cells incubated with Prox1 antibody do not have nuclear expression. Comparison with Fig 5 suggests that this is background staining. Scale bar, 50 μm.

03. Fig S3.

qRT-PCR analysis of freshly sorted EpCAM⁺ cells differentiated in vitro for a total of 15 days after Matrigel addition according to the differentiation scheme shown in Fig. 5a. Values are shown as fold change compared with Post-sort day 0 cells. There is continued expression of Sox 9 and Prox1; increased expression of SftpC and Pdx1, and decreased expression of Tat, TSHR, Nkx2.1 and Hoxb13, YY and MafB. *, p <0.05.

04. Fig S4.

Candidate foregut stem cells (FGSCs) were grown in culture for 10 passages (~7 weeks) and transcript analysis was performed by QPCR. (a) Results are shown as fold change relative to the FGSCs at time 0. A majority of genes did not change over the course of growth in vitro. *, p <0.05. #, not detected. (b) Table of ΔCt values of cells at the beginning and end of growth. Levels of transcript abundance can be inferred from ΔCt values (low ΔCt values correlate with high transcript abundance). N=9.

05. Fig S5.

MACS sorting of HuESC derivatives. (a) Scheme for MACS sorting of HuESCs. The first sort removes ESCs using anti-SSEA4, followed by removal of extraembryonic cells with anti-SSEA1. Finally, EpCAM⁺ definitive endoderm cells are positively selected. (b) Presort differentiated cells show co-expression of PDX1 and EpCAM at day EB_{14+ 21}. (c) FACS analysis of pre- and post-MACS sorted cells demonstrates significant enrichment of EpCAM⁺ cells in the sorted population. Prior to sorting, cells were maintained for 14 days as EBs in HuES medium containing DMEM/F12 plus 2 mM Glutamine, 0.1 mM NEAA, 0.11 mM mercaptoethanol and 20% SR. After EB plating, cells were maintained for 14 days in HuES medium, then switched for 28 days to SFDM/ITS to diminish growth of fibroblasts and enrich for endoderm cells. (d) Cells 4 days after sorting, grown in HuES medium. (e) Transcript analysis of genes expressed by sorted cells relative to unsorted cells shows enhanced levels of endodermal gene expression and diminished expression of neural and mesenchymal transcripts in sorted cells. This experiment was repeated twice with similar results. Scale bars, 50 μm (b), 100 μm (d).

06. Table S1.

List of primary antibodies used in this study.