Author contributions: J.P.: Conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing, final approval of manuscript; W.L.: Administrative support, collection and/or assembly of data, data analysis and interpretation; A.B., K.A.: Collection and/or assembly of data; O.I.: Conception and design, grant support, data analysis and interpretation, manuscript writing, final approval of manuscript.
First published online in STEM CELLS EXPRESS August 20, 2009.
Disclosure of potential conflicts of interest is found at the end of this article.
Identification and use of cell surface cluster of differentiation (CD) biomarkers have enabled much scientific and clinical progress. We identify a CD surface antigen code for the neural lineage based on combinatorial flow cytometric analysis of three distinct populations derived from human embryonic stem cells: (1) CD15+/CD29HI/CD24LO surface antigen expression defined neural stem cells; (2) CD15−/CD29HI/CD24LO revealed neural crest-like and mesenchymal phenotypes; and (3) CD15−/CD29LO/CD24HI selected neuroblasts and neurons. Fluorescence-activated cell sorting (FACS) for the CD15−/CD29LO/CD24HI profile reduced proliferative cell types in human embryonic stem cell differentiation. This eliminated tumor formation in vivo, resulting in pure neuronal grafts. In conclusion, combinatorial CD15/CD24/CD29 marker profiles define neural lineage development of neural stem cell, neural crest, and neuronal populations from human stem cells. We believe this set of biomarkers enables analysis and selection of neural cell types for developmental studies and pharmacological and therapeutic applications. STEM CELLS 2009;27:2928–2940
Differentiating human pluripotent stem cells mirror all stages of cell development and lineages  and could represent an unlimited source of cells for therapeutic paradigms in regenerative medicine [2, 3]. Mouse and human embryonic stem (ES) cells, induced pluripotent stem (iPS) cells, or nuclear transfer cells have been used experimentally to develop analytical and cell replacement approaches for neurological disorders [4–8]. Functional recovery has been achieved in animal models of disease [7, 9], and clinical data on cell therapy in the nervous system using human fetal material, although controversial , has shown proof-of-principle success [11–14]. Pluripotent stem cells, however, carry a risk of uncontrolled growth , and after transplantation, tumors of neural and non-neural (teratoma) tissue origin have been observed [7, 9, 16]. The occurrence of tumors in applications of pluripotent stem cell-derivatives mandates precise cell selection steps [17, 18]. Fluorescence-activated cell sorting (FACS) is an approach to eliminate non-neural cells from mixed cell preparations derived from ES cells for in vitro and in vivo studies [15, 19].
Such cell-sorting strategies require novel biomarkers to enable the selection of specific cell populations of potential therapeutic and scientific value for the fields of regenerative medicine and stem cell biology [15, 17, 18]. The combinatorial detection of surface markers by multicolor flow cytometry has been widely applied in the fields of hematology and immunology [20, 21], but has up to now only been marginally exploited in neurobiology [22–25]. A recent surface marker screen yielded an initial documentation of cluster of differentiation (CD) antigens expressed during human ES (hES) neural differentiation , but on the whole, the neural lineage remains yet to be defined according to a combinatorial surface antigen code .
Previous studies utilized genetic fluorescent markers such as Tau-GFP , Synapsin-GFP , Pitx3-GFP , or dye labeling  to isolate mature, differentiated neurons. However, clinical applicability of such cell isolation methods analogous to hematological diagnosis and therapy depends on the usage of surface markers. We and others previously applied FACS methodology optimized for neural cell selection  to successfully eliminate tumor-generating proliferative cells from ES cell- [15, 19, 27, 29] as well as iPS-cell-derived  neural cell populations. Importantly, FACS is routinely applied in hematological cell transplantation to generate clinical-grade cell preparations of high purity . In addition, there is considerable analytical value in the identification and isolation of multiple neural subsets exclusively by surface antigens: this eliminates the need for a genetic reporter and enables close monitoring of stem cell differentiation using a rapid quantitative readout. In summary, although the technology is feasible and available, it has not been extensively applied due to a lack of a comprehensive combinatorial marker analysis.
Here, our aim was to discover a surface antigen profile of neural lineage differentiation by identifying and characterizing specific neural cell subsets derived from pluripotent stem cells.
MATERIALS AND METHODS
Pluripotent Stem Cell Culture and Differentiation
Work with hES cells was approved by the Partners Embryonic Stem Cell Research Oversight Committee. Undifferentiated human ES cell lines H7 (WA-07, XX, passages 40-65) and H9 (WA-09, XX, passages 35-45) were cultured under growth conditions and passaging techniques previously described . In vitro analysis was done with both hES cell lines, the transplantation assay was done with H9. Undifferentiated hES cell cultures were maintained on mitotically inactivated human fibroblasts (D551; ATCC, Manassas, VA, http://www.atcc.org). Human ES cell neuronal induction and differentiation were achieved, essentially as previously described [7, 15]: a neural induction phase on stromal feeder cells with 300 ng/ml of Noggin was followed by careful manual selection of neural rosettes using microdissection (at 21 div) and subsequent culture on laminin/poly-ornithine substrate with N2-based medium (42 div and beyond) . Mouse ES and induced pluripotent stem (iPS) cell neuronal differentiation was done in a five-stage embryoid body-based protocol as previously described [6, 27, 31]. Cell lines used were as follows: mouse ES cell lines J1 and R1 (both ATCC), Pitx3-GFP (M. Li; see Hedlund et al. ); Sox1-GFP (A. Smith; see Chung et al. ); the Oct4-selected mouse iPS cell line O9 (R. Jaenisch; see Wernig et al., 2007, 2008 [6, 32]).
Neural Surface Antigen Staining
Cells were harvested and gently dissociated using TrypLE Express (Invitrogen/Gibco, Carlsbad, CA, http://www.invitrogen. com). Cells were filtered through cell strainer caps (35-μm mesh; BD Biosciences, San Diego, http://www.bdbiosciences.com) to obtain a single cell suspension (approximately 106 cells per ml for analysis, 0.5–2 × 107 cells per ml for sorting). Cell viability was routinely determined to be above 90% by trypan blue dye exclusion before use for analysis and sorting experiments. Species-specific antibodies were used whenever available; the CD15 and FORSE-1 antibodies were found to label neuroepithelial cell types on both mouse and human cell sources. For colocalization assays on neural precursors (Fig. 1B), H7 cells were infected at day in vitro (div) 21 with lentivirus containing a Nestin-GFP reporter (Isacson et al., unpublished). FORSE-1, SSEA-1, and SSEA-3 antibodies were obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the NICHD and maintained by the University of Iowa, Department of Biological Sciences, Iowa City, IA 52242. Surface antigens were labeled by incubating with the primary antibodies (see supporting information Table 1) for 20 minutes at room temperature, followed by incubation for 20 minutes with the appropriate Alexafluor-488 or Alexafluor-647 fluorescent secondary antibodies. Alternatively, coniugated antibodies were used (BD Pharmingen, San Diego, http://www.bdbiosciences.com). All washing steps were performed in cold phenol-free, Ca2+-free, Mg2+-free Hank's buffered saline solution (Invitrogen/Gibco) containing penicillin-streptomycin (Invitrogen/Gibco), 20 mM D-glucose (Sigma, St. Louis, http://www.sigmaaldrich.com), and 2% fetal bovine serum (HyClone, Logan, UT, http://www.hyclone.com) .
Flow Cytometry and Cell Sorting
The stained cells were analyzed and sorted on a fluorescence-activated cell sorter FACSAria (BD Biosciences) using FACSDiva software (BD Biosciences); data were additionally analyzed using FlowJo software (Tree Star, Ashland, OR, http://www.treestar. com). The fluorochromes were excited with the instrument's standard 488-nm and 633-nm lasers, and green fluorescence was detected using 490 LP and 510/20 filters and far-red fluorescence using 660/20 filters. Fluorescence was determined by analysis and gating against appropriate controls (unstained, secondary antibody-only, GFP-negative cells) at an identical stage of maturity and prepared in parallel. For triple staining, analysis was done in the green, red, and far-red channels. Cell populations of interest were gated to count a minimum of 10,000 live-gate events, excluding doublets, based on forward and side scatter of control samples. All flow cytometric analyses and sorts were repeated at least three times. Gates determining positivity were set so that less than 0.5% of positive events were present when acquiring the corresponding negative control. Gates for “high” (HI) versus “low” (LO) populations were set in comparison to the appropriate non-fluorescent control and adjusted to include visually discernable clusters, if revealed by combinatorial staining. To assure specificity, appropriate separation of sorted fractions was verified visually and by FACS reanalysis of the subsets defined in that manner (>80% enrichment post-sort) as described before . Prior to aseptic sorting, the nozzle, sheath, and sample lines were sterilized with 70% ethanol for 15 minutes, followed by washes with sterile water to remove remaining decontaminant. A 100-μm ceramic nozzle (BD Biosciences), sheath pressure of 20-25 PSI, and an acquisition rate of 1,000-3,000 events per second were used as conditions optimized for neuronal cell sorting (“gentle FACS”) as previously described .
Flow Cytometric Analysis of Intracellular Neural Antigens
Staining for intracellular antigens was done using commercially available fixation and buffer solutions IntraCyte fix solution, wash solution, and block reagent (Neuromics, www.neuromics. com). After careful validation and titration, the following antibodies were used at dilutions optimized for flow cytometric analysis: βIII tubulin (TuJ1), doublecortin, TH, MAP2, Pax3 (see supporting information Table 2). It should be noted that different primary antibodies against the intracellular antigens stain with varying levels of background fluorescence, enabling relative comparison between groups, but not an absolute quantitation. For combined surface and intracellular antigen detection, live cells were stained for surface antigens as described above, followed by fixation and permeabilization to stain the intracellular epitopes.
Embryonic Central Nervous System Tissue Dissection and Dissociation
Timed pregnant C57BL/6 mice (Jackson ImmunoResearch, West Grove, PA, http://www.jacksonimmuno.com) were anesthetized with intraperitoneal sodium pentobarbital (300 mg/kg) and decapitated. Primary fore- and midbrain tissue was obtained from E13 mouse embryos (Charles River, Wilmington, MA, http://www.criver.com) using a dissecting microscope. The overlying scalp tissue and the outer meninges were removed entirely to expose forebrain and midbrain tissue, which were dissected as described in detail elsewhere . For isolation of spinal cord tissue, a longitudinal paramedian cut along the dorsal midline was performed, and the spinal cord was isolated using forceps and microscissors. Dissected primary neural tissue was incubated with TrypLE express (Invitrogen/Gibco, Carlsbad, CA), and gently dissociated with fire-polished Pasteur pipettes. For FACS analysis and cell sorting, cells were resuspended in Hank's Buffered Salt Solution (HBSS)-buffered glucose solution, filtered through a 35 μm mesh and brought to a concentration of 0.5–2.0 × 106 cells/ml.
Immunofluorescence, Imaging, and In Vitro Analysis
Indirect immunofluorescence was performed on 4% paraformaldehyde-fixed cell cultures and brain tissue. Fixed cells were incubated in a blocking solution consisting of 10% normal donkey or goat serum (Jackson ImmunoResearch Laboratories Inc., West Grove, PA, USA) and 0.1% Triton X-100 (Sigma, St Louis) in 0.1 M phosphate-buffered saline (PBS) for 1 h at room temperature. Primary antibodies were diluted in blocking solution and added to the cells. Dilutions of the primary antibodies used were as listed in supporting information Table 2. Antibody concentrations for detection of surface antigens were determined by titration assays using flow cytometric analysis and immunocytochemistry, including staining of viable, attached ES cell and primary neural cultures as previously described . The appropriate fluorescence-labeled secondary antibodies (Alexafluor goat or donkey anti-rabbit, -mouse, or -sheep 488, 594, 647; Molecular Probes, Eugene, OR, http://probes.invitrogen.com; 1:500) were applied for visualization, and nuclei were counterstained with Hoechst 33,342 (Molecular Probes, Invitrogen, Carlsbad, CA; 5 μg/ml). On selected samples, the primary antibody was omitted to verify specificity of staining.
For visual microscopic analysis, quantification was performed by counters blinded to experimental groups on the stained coverslips using an integrated Axioskop-2 microscope (Carl Zeiss, Thornwood, NY, http://www.zeiss.com) and Stereoinvestigator image-capture equipment and software (MicroBrightField, Williston, VT, www.mbfbioscience.com). A contour was drawn around each coverslip to identify the area of interest. A physical dissector probe was utilized, and counting frames were placed in a systematically random manner at ∼100 sites per coverslip. For each experiment, an average of 2,000 cells was scored.
Quantitative RT-PCR Gene Expression Analysis
Total RNA was extracted by using the RNeasy kit and DNase I treatment (Qiagen, Valencia, CA). RNA samples from NS, NC, and ND populations were reverse-transcribed into cDNA using Sensiscript reverse transcriptase (Qiagen, Valencia, CA, http://www1.qiagen.com) and oligo dT as the primer. Polymerase chain reactions (PCRs) were set up in 25 μl reaction volume using SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA, http://www.appliedbiosystems.com). Primers for each candidate gene were designed using MacVector 7.0 and used with a final concentration of 250 nM. See supporting information Table 3 for a list of primers used. For each primer pair, duplicates of three independently collected samples were compared to quantify relative gene expression differences between these cells. Beta-actin was used as an internal control gene.
Cells and tissue samples were collected from and suspended in lysis buffer containing the following: 50 mM Tris-HCl, 0.15 M NaCl, 0.32 M sucrose, 1.0 mM EDTA, and 1% NP-40. In addition, phosphatase inhibitors I and II (1:100) and protease inhibitors (1:100) were added fresh before cell lysis (P2850, P5276, and P8340, respectively; Sigma). After cell lysis, the homogenate was centrifuged, a portion of the supernatant was reserved for protein determination (BCA Assay; Pierce, Rockford, IL, http://www.piercenet.com), and the remainder was stored at –20°C. Solubilized protein (40 μg) was mixed 1:1 with sample buffer and boiled for 5 minutes. The samples and molecular weight standards were separated using the Criterion precast 4-15% SDS–polyacrylamide gel system (Bio-Rad, Hercules, CA). After electrophoresis, the proteins were electrically transferred to polyvinylidene difluoride membranes at 100 V for 30 minutes. After transfer, blots were incubated in Tris-buffered saline with 0.1% Tween 20 containing 5% BSA. Subsequently, blots were incubated with the TuJ1 primary antibody (1:2000; Covance, Princeton, NJ, http://www.covance.com) at 4°C overnight. Horseradish peroxidase (HRP)-conjugated secondary antibodies were then applied, and immunoreactive bands were visualized with chemiluminescence (GE Healthcare, Little Chalfont, U.K., http://www.gehealthcare. com) and exposed onto film. Immunoblots were then stripped and reprobed for the expression of ß-actin (1:10,000; Abcam, Cambridge, MA, http://www.abcam.com) serving as a loading control. Optical density analysis (NIH Image, http://rsb.info.nih.gov/nih-image/) was used to determine the relative abundance of protein in each sample.
Transplantation of Purified Neural Cell Populations and In Vivo Analysis
Animal studies were approved by the Institutional Animal Care and Use Committee at McLean Hospital and Harvard Medical School. Naïve and unilaterally 6-hydroxydopamine-lesioned adult female Sprague-Dawley rats (200-250 g) were purchased from Taconic (Germantown, NY, http://www.taconic.com). After FACS purification (see above), differentiated hES cells were counted and resuspended at ∼25,000 viable cells per μl in the final differentiation medium. For each surgery, animals were deeply anesthetized with ketamine and xylazine (60 and 3 mg/kg, respectively, i.m.). Injections were performed as previously described . Two to four μl were slowly injected into striatum of the rats (anterior-posterior = 0; lateral = –2.8 from bregma and from –5.5 to –4.5 mm ventral from dura, with the tooth bar set at –3.3). For analysis at five weeks, naïve animals received NS, NC, and ND cell suspensions in bilateral grafts (eight per group). In addition, eight lesioned rats unilaterally received ND cell suspension to be analyzed at 10 weeks post-transplantation. Rats were immunosuppressed with cyclosporin A (15 mg/kg per day; Sandimmune; Sandoz, East Hannover, NJ, http://www.sandoz.com) starting 1 day prior to surgery. Five and/or ten weeks after transplantation, animals were terminally anesthetized by an intraperitoneal overdose of pentobarbital (150 mg/kg) and perfused intracardially with 70 ml of heparinized saline (0.1% heparin in 0.9% saline) followed by 100 ml of paraformaldehyde (4% in phosphate-buffered saline). Brains were removed, postfixed for 4 hours in 4% paraformaldehyde, equilibrated in sucrose (20% in PBS), and sectioned on a freezing microtome in 40-μm slices that were serially collected and stored in cryoprotectant. Two full series of sections were processed for visualization of the respective antigens using immunofluorescence staining, as described above. To identify human cells in the rodent brain, we used the human-specific antibody against human nuclear antigen (1:50; Chemicon/Millipore, Billerica, MA, http://www.millipore.com) and immunohistochemical characterization for human/primate NCAM (Eric-1; Santa Cruz Biotechnology Inc., Santa Cruz, CA, http://www.scbt.com) and human/primate Nestin (Neuromics, Edina, MN, http://www.neuromics.com). The sections were permeabilized with 0.1% Triton X-100 and incubated with primary antibodies in 2% normal donkey serum overnight at 4°C. After rinsing, sections were incubated with appropriate fluorescence-labeled secondary antibodies (Alexafluor; Molecular Probes/Invitrogen, Carlsbad, CA; 1:500) for 1 hour at room temperature, rinsed, and incubated with Hoechst 33,342 (5 μg/ml). Sections were mounted onto Superfrost Plus glass slides (Fisher Scientific, Pittsburgh, PA, http://www.fisherscientific.com), and confocal analysis was performed using a Zeiss LSM510/Meta station. For identification of signal colocalization within a cell, optical sections were kept to a minimal thickness, and orthogonal reconstructions were analyzed. Design-based stereology was performed on the specimens using an integrated bright-field microscope (Axioskop 2; Carl Zeiss, Thornwood, NY), confocal microscope (LSM510), and StereoInvestigator image-capture equipment and software (MicroBrightField, Williston, VT). Graft volumes and ki-67-positive cell counts were calculated using the Cavalieri estimator and optical fractionator probes, from one-sixth of the total sections. The coefficient of error was used to assess probe accuracy, and p < .05 was considered acceptable. The 3D reconstruction was created from one-sixth of the total sections using Neurolucida solid modeling software (MicroBrightField). The graft constituent cell ratios were calculated by blinded investigators counting cells in several [20–30] randomly selected high power fields.
The data were analyzed using a one-way analysis of variance and Tukey-Kramer multiple comparisons test or a two-tailed Student t-test, where appropriate (InStat; GraphPad, San Diego, CA, http://www.graphpad.com; JMP, Cary, NC, http://www.jmp.com). Calculated comparisons of p < .05 were considered significantly different. All reported values represent the mean ± SEM.
Identification and prospective isolation of specific cell subsets using novel markers would enable more detailed studies of neural lineage specification (Fig. 1A). We found that the nestin-positive population contained additional subpopulations that were only detected by analysis for differential expression of surface antigens (Fig. 1B). Distinct levels of CD15, FORSE-1, CD24, CD29, and CD56 antigens underlined the heterogeneity of neuronal differentiation of hES cells (Fig. 1C). Presence of these surface markers was also documented on a number of different neural cell types, including human ES cell-derived neurons and neuroblastoma cells, and varied depending on the time in vitro and/or stage of differentiation (supporting information Fig. 1A, 1B). We therefore investigated whether combinatorial analysis for these markers could yield specific expression patterns (codes) in the developing neural lineage.
Given the rising expression levels of the sialoglycoprotein CD24 (heat-stable antigen; nectadrin; small-cell lung carcinoma cluster four antigen; BA-1) during in vitro development from ES toward the neuronal differentiation (ND) stage (see supporting information Fig. 1A, 1B), this antigen was singled out as a potential marker for the prospective enrichment of differentiated neurons. Already at early stages of neural differentiation (div 14; NP), CD24 expression was observed as evident by utilizing a Sox1-GFP reporter mouse ES cell line  (Fig. 2 A, left panel; supporting information Fig. 2A). Presence of CD24 expression on mature neuronal cells types was determined by using transgenic mouse knock-in ES cell-derived Pitx3-GFP+ dopamine neurons  as an indicator of late differentiation in ES cell cultures (Fig. 2A, right panel). On human neural cultures derived from hES cells, bivariate FACS analysis identified CD24HI cells to be negative for early neural markers such as the FORSE1, CD15, and CD146 antigens (Fig. 2B). Furthermore, CD24HI cells were also found to be negative for CD133 (data not shown), a marker previously shown to represent a subset of CD15+ cells in hES neural differentiation , and did not costain with Pax 3, a neural crest marker (supporting information Fig. 2D). In vitro maturation of hES neural differentiation culture was paralleled by increased CD24HI surface antigen expression (see supporting information Fig. 2B–2D). Cell sorting of human ES cell-derived neural cell suspensions enabled the isolation of CD24HI and CD24LO subpopulations. Post-FACS cell cultures of the CD24HI population were found to be enriched for neurons (Fig. 2C, upper row) and contained less proliferative ki67+ cells than the CD24LO fraction (Fig. 2C, lower row). We also found that, using primary mouse brain cell preparations (E13), the neuronal fraction was enriched by selecting for CD24HI expression (Fig. 2C, supporting information Fig. 2E). CD24HI cultures displayed extension of neuronal processes, forming a dense network immunoreactive for neuronal markers such as Tuj1, MAP2, and synapsin. Viability of CD24HI-purified embryonic fore- and midbrain neuronal cultures was virtually unaffected by the sorting procedure (supporting information Fig. 2F). The CD24LO fraction enriched from hES neural differentiation expressed more Pax6 (14.1-fold enriched; 12.7 ± 0.5 versus 0.9 ± 0.0% in CD24HI), Vimentin (2.6-fold enriched; 6.8 ± 0.4 versus 2.6 ± 0.4% in CD24HI) and Nestin (2.7-fold enriched; 36.1 ± 1.1 versus 13.3 ± 2.0% in CD24HI), all markers characteristic of immature neural phenotype (Fig. 2F, left panel). In contrast, the CD24HI cells were strongly positive for central nervous system (CNS) neuroblast (doublecortin; 13.6-fold enriched; 43.4 ± 1.0 versus 3.2 ± 0.1% in CD24LO) and neuronal markers (TuJ1; 12.8-fold enriched; 46.0 ± 0.5 versus 3.6 ± 0.1%, Tau; 6.7-fold enriched; 32.8 ± 3.0 versus 5.0 ± 0.8% in CD24LO) (Fig. 2F, right panel). Our preliminary conclusion was that CD24HI surface antigen was a marker well correlated with the differentiation of neuronal cells.
In contrast, CD15 (stage-specific embryonic antigen-1, Lewis-X antigen), a more immature marker staining a carbohydrate epitope (fucosyl-N-acetyl-lactosamine), was absent on differentiated neurons, including dopamine neurons identified by Pitx3-GFP expression (Fig. 3A). Known as an established mouse ES cell and a myeloid differentiation marker [34–37], it is important to note its clear and distinct labeling of mouse neuroepithelium, as shown on midbrain sections of E13 mouse embryos (Fig. 3B). In human ES cell differentiation, CD15 was found to be strongly expressed on Sox1 and Sox2-positive neuroepithelial rosette structures (Fig. 3C). CD15 expression was thus determined to be a marker of immature, neuroepithelial cells, enabling their enrichment, or alternatively their depletion, for applications in developmental as well as cell therapeutic paradigms.
Similarly, the CD29 antigen (beta1-integrin) was found to predominantly label the neuroepithelial rosette structures typical of hES neural induction (Fig. 3D). CD29 expression decreased from 81.3 ± 6.8% at rosette stage (div 21) to 53.6 ± 4.7% at differentiation (div 42; three independent experiments; not shown). At those later stages of in vitro neural differentiation, remaining clusters of proliferative neural stem cells displayed enhanced expression of CD29 (supporting information Fig. 3A). Doublecortin-positive neuroblasts emerged from such more immature clusters and were negative for CD15 and also CD29 expression (Fig. 3D). Interestingly, a second population of mesenchyme-like proliferative cells, typically forming a halo around dense neuroepithelial cell groups, was identified to also strongly express CD29 (Fig. 3D; supporting information Fig. 3B). Upon FACS analysis and cell sorting experiments, the CD29HI fraction was enriched for proliferative spherical clusters as well as proliferative adherent cell types (Fig. 3E). In contrast, the CD29LO enriched fraction contained predominantly process-bearing post-mitotic neurons (Fig. 3E). CD29HI contained more nestin (41.5%) and less TuJ1 (25.3 ± 0.5%), CD29LO showed reduced nestin (9.6%) and increased TuJ1 (65.6 ± 1.1%); TuJ1 in this subset is determined by flow cytometric analysis of intracellular neural antigens (single experiment) (supporting information Fig. 3C, 3D). Consistent with those observations, the CD29LO fraction was also enriched for doublecortin-positive cells (50.2 ± 14 compared to 9.5 ± 3.5 in the CD29HI population; three independent experiments; supporting information Fig. 3E).
Our findings thus far suggested that the fraction of neurons was augmented by sorting for CD24HI-expressing cells, as well as for CD15− or CD29LO-expressing cells. In contrast, more proliferative cell populations were enriched by isolating the CD24LO, the CD15+, and/or the CD29HI cell subsets (see Figs. 1–3). In order to fully characterize specific neural subpopulations according to surface antigens, combinatorial FACS analysis for the above and a variety of complimentary surface markers  was performed (Fig. 4A): (i) CD29LO cells were found to also exhibit CD24HI and CD56HI, a surface antigen pattern consistent with differentiation towards neuroblasts and neurons [15, 38]. (ii) Precursor or stem cell markers, such as CD146 , FORSE1  or CD15  (Fig. 3), CD133 , respectively, were present on the CD29MID/HI subset. (iii) Finally, CD29HI cells were found to express surface antigens consistent with mesenchymal and/or neural crest phenotypes, such as CD271, CD57, and CD73  (Fig. 3). Key populations of neural lineage development were hereby defined according to a CD15/CD24/CD29 surface antigen code, as depicted in Figure 4B, 4C. A CD15+/CD24LO/CD29HI surface antigen profile defined presumptive neural stem/progenitor cells (NS). In contrast, CD15−/CD24LO/CD29HI expression was characteristic of a presumptive neural crest/mesenchymal cell population (NC). Finally, a CD15−/CD24HI/CD29LO surface antigen signature defined a presumptive neuronal/neuroblast population (ND). The temporal dynamics of CD15/CD24/CD29 surface expressions in vitro were consistent with NS, NC, and ND phenotypes (supporting information Fig. 4B, 4C): at later stages of in vitro maturation, a shift toward ND population was observed. However, also at prolonged differentiation in culture (up to 70 div analyzed), all three populations remained present (data not shown). Ki67+ cells appeared enriched in the NS population, whereas the central nervous system precursor marker doublecortin and the neuronal microtubule-associated protein (MAP)-2 were highly expressed in the ND population (supporting information Fig. 4A). We then used cell sorting experiments to isolate these subsets from typical heterogeneous cell populations differentiated from hES cells (H7 and H9; div 35-42) and found them to be morphologically clearly distinct (Fig. 4C). The CD15+/CD24LO/CD29HI subset (NS) was enriched for proliferative clusters, giving rise to neural precursor/neurosphere cultures (Fig. 4D). The CD15−/CD24LO/CD29HI subset (NC) presented with massive proliferation of flat, adherent cells after cell sorting. In contrast, the CD15−/CD24HI/CD29LO subset (ND) represented a purified culture of cells with neuronal morphology (Fig. 4D).
Subsequent detailed in vitro analysis of the neural cell populations identified according to CD15/CD24/CD29 expression codes confirmed the prospective isolation of distinct subsets consistent with phenotypes of neural stem/progenitor (NS), neural crest/mesenchymal (NC), and neuronal/neuroblast (ND) cells, respectively (Fig. 5; supporting information Figs. 2D; 3B, 3C; 4A–4C). Quantitative analysis of gene expression revealed characteristic expression patterns for the isolated subsets, with ND cells predominantly expressing βIII-tubulin (2.8-fold increased expression over NS, 6.3-fold increase over NC population; qRT-PCR; Fig. 5A). Investigations on the protein level mirrored this expression pattern, characterizing ND cells as the fraction with the highest presence of βIII-tubulin (Fig. 5B). Sorting for CD15−/CD24HI/CD29LO (ND) enabled the isolation of viable postmitotic near-pure neuronal cultures. Under high-purity sorting conditions, those cultures were devoid of ki67+ cells (Fig. 5C). Sphere formation assays underlined this reduced proliferative capacity of the ND (no spheres) versus the other populations (0.6 ± 0.2 and 3.2 ± 1.5 spheres per well in the NS and NC populations, respectively; supporting information Fig. 4E). We then tested the CD15/CD24/CD29 neural surface antigen code combination for its applicability as a marker set enabling the prospective isolation of specific neural subpopulations in transplantation experiments, for both therapeutic and basic developmental studies (Fig. 5D–5H). The three described populations (NS, NC, ND) were isolated from differentiated human ES cell cultures (H9; div 42) by FACS and subsequently grafted into rodent brain for histological analysis after 5 and 10 weeks post-transplantation. Consistent with the emerging hypothesis and designation of the combinatorial code (see above; Fig. 4B, 4C), grafts from the CD15+/CD24LO/CD29HI (NS) group showed neuroepithelial tumors, which displayed characteristic neural rosettes exhibiting immunoreactivity for Sox2, Nestin, Vimentin, and radial glial markers 3CB2, RC2 (Fig. 5D, 5E; and data not shown) (n = 8 grafts). The CD15−/CD24LO/CD29HI group (NC) showed similarly debilitating tumor formation of more diffuse composition with a strong component of migratory, mesenchyme-like cells. Confirming its characterization as neural crest/mesenchymal phenotypes, cells in these grafts were strongly CD271 positive (Fig. 5D; n = 8 grafts). Myosin-positive muscle cells in the host brain, a confounding component of neural grafts in previous studies [7, 9, 16], were exclusively found in this group (Fig. 5E). In contrast, CD15−/CD24HI/CD29LO (ND) grafts were composed of human-NCAM-positive cells, extending neuronal processes into the host brain tissue (Fig. 5F). Ki67-positivity in the graft area was decreased in the ND group (NS: 55.7 ± 21.4; NC: 54.3 ± 20.4; ND: 2.6 ± 1.0; 1.3 ± 1.0; 1.0 ± 0.2 × 103; supporting information Fig. 4D, 4E). No tumor formation was observed at either 5 (two independent groups, n = 5 each) or 10 weeks (n = 8 grafts). Graft volume in the ND groups ranged between 4.8 and 20.9% compared to that observed in the NS and NC groups; Fig. 5G, 5H). Graft analysis clearly demonstrated that the prospective isolation of specific neural subsets according to CD15/CD24/CD29 surface antigen expression resulted in distinct graft composition and characteristics. Being important from a therapeutic perspective, the CD15−/CD24HI/CD29LO (ND) group exclusively comprised neuronal grafts without tumor formation.
These experiments demonstrate that different combinations of the CD15, CD24, and CD29 surface antigens define distinct cellular subsets of neural lineage differentiation, including bona fide neural stem, neural crest/mesenchymal cells, and neuronal cells from human pluripotent stem cell cultures. We observed these biomarkers consistently on a variety of neural cell types and propose a model for selecting developing neural cells based on these distinct profiles (Fig. 5I; Table 1). Cell sorting using these neural surface antigen codes enabled the selection of specific cellular subsets from heterogeneous neural cell populations. After transplantation, the CD15−/CD24HI/CD29LO cell population yielded near-pure neuronal grafts, while the other subsets formed large neuroepithelial (CD15+/CD24LO/CD29HI) or neural crest/mesenchymal (CD15−/CD24LO/CD29HI) tissue tumors.
Table 1. Schematic surface antigen profile of the NS, NC, and ND populations identified and defined by CD15, CD24, and CD29 expression code
CD15 (Lewis-X antigen, SSEA-1), and also CD133 (Prominin-1), have previously been identified as putative markers of stem cell identity not only in the neural system [15, 22, 42, 43] but also in hematopoietic [44, 45] and other [36, 46–48] cell types, suggesting functional relevance for “stemness” properties in general, for example for adherence to stem cell niches . While CD15 and also the structurally related FORSE-1 antigen are strongly expressed on neuroepithelial cells, they are both downregulated upon further differentiation, which suggests that delamination from the neuroepithelial tube/rosette-like structures parallels surface expression changes of these glycolipid markers . This is consistent with the observation that CD15 is present on neural stem cells in vivo, residing in stem cell niches of the adult brain , including the recent identification of CD15 on nervous system tumor cells, such as medulloblastoma [52, 53]. Similarly, integrins play an important role in neural development , and CD29 specifically was observed on human neural stem cells obtained from fetal tissue . In addition, integrin signaling has been shown to be of functional relevance for both neural crest  and mesenchymal development [57, 58]. The distinct pattern of CD29 expression suggests that β1-integrin signaling, and regulatory events affecting cell adhesion, are of functional relevance in neural development not only in vivo but also in artificial in vitro systems. Again, analogous to physiological development in vivo, delamination of neural cells from neuroepithelium/neural tube-like structures can be observed , with differentiation toward neural crest/mesenchymal versus CNS neuronal phenotypes (see Fig. 3 D). In contrast, CD24 was identified to be upregulated with neural differentiation and neuronal maturation. We demonstrate that its surface expression is well correlated with known markers of differentiating neurons, such as Pitx3, TuJ1, and MAP2, and that high levels of CD24 surface expression are a valid marker for maturing neuronal cell types in in vitro neural differentiation (Table 1). Consistent with our findings, Uchida et al. have described only low levels of CD24 on human neural stem cells isolated from human brain , and Rougon et al. showed that CD24 labels neuroblasts in neurogenic niches of mouse brain and during development . Here, FACS for CD24HI was introduced to isolate purified neuronal cultures from primary rodent embryonic brain tissue, which provides a valuable tool to analyze pure neuronal subpopulations under defined conditions. In hES cell cultures, we observed that a separation into a CD24LO versus a CD24HI population occurred with the transition from a mixed neural precursor cell population to the neuronal differentiation stage (see supporting information Fig. 2). Yet, given the population dynamics in the dish  and the promiscuity of surface markers , a flow cytometric gating strategy according to a single marker would be insufficient to isolate a pure, neuronally differentiating population from heterogeneous stem cell cultures. Characterizing the CD15/CD24/CD29-defined populations for coexpression of additional neural surface markers may enable future studies on the relevance of these dynamic changes during in vitro neuropoiesis (see Fig. 4A). The presence of CD15/CD24/CD29 antigens on various neural cell types including mouse embryonic primary mid- and forebrain, spinal cord and stromal tissue, as well as ES and iPS cells, suggests more general applicability of this code (see supporting information Fig. 1B) not only for neurodevelopmental studies and cell therapeutic paradigms but potentially also in the diagnosis and study of neural tumors . It remains to be seen to which extent the dynamics of CD15, CD24, CD29 expression on these other cell types mirror the code described here for human neural cells derived from ES cells: the NS population, characterized as CD15+/CD24LO/CD29HI, comprises a CD133 population and positivity for FORSE-1, nestin, Pax6, Sox1. It generates proliferative neural clusters and spheres in vitro, and leads to the formation of rosette-containing neuroepithelial tumors reminiscent of medulloblastoma. We do not formally study the generation of NC from NS cells, but sorting for CD15 appears to generate neurosphere-like cultures that, in turn, are able to give rise to heterogeneous cultures of NS, NC, and ND populations (Fig. 5I). The precise level of multipotency of this overall population versus its subsets demands to be further elucidated , but all of the above features are consistent with neural stem cell character (see summary, Table 1). It will be clinically important to control and limit the presence of this proliferative population from a grafted cell suspension. In addition, the combinatorial code used here may serve to expand the biomarker library used to isolate neural stem cells from cell cultures as well as from primary tissue.
The NC population (CD15−/CD24LO/CD29HI) entails subsets expressing Pax3, CD57 (HNK1), and CD271 (p75), all of which have been described on neural crest cells [64, 41, 59]. Furthermore, it contains markers also present on mesenchymal derivatives, including CD271, CD73 (ecto-5-nucleotidase), and myosin [65, 58] (Table 1). Further analysis will be required to determine in which way neural crest derivatives such as peripheral neuron cells and melanocytes differentiate from this newly defined hES subset, and whether the mesenchymal cells are in fact generated via cranial neural crest-like cells in culture, or via mesodermal specification from the original pluripotent source . From our current focus on cell transplantation to models of neurodegenerative disease in the CNS, it is an important finding that excluding this population is a prerequisite for tumor elimination and enhances the purity of the graft, as evidenced by absence of myosin-positive cells.
Finally, the ND (CD15−/CD29LO/CD24HI) population was found to provide tumor-free neuronal grafts. In vitro, virtually pure neuronal cultures were established by sorting for this surface antigen combination. Markers co-expressed in this population include neuroblast and neuronal markers doublecortin, TuJ1, and MAP2, while more immature markers such as Pax6 and FORSE-1, the formation of proliferative clusters, are absent. While this study addresses one major obstacle to neural stem cell therapy, i.e. the formation of tumors, the derivation and stability of functional therapeutic phenotypes are still a challenge that needs to be solved. A better understanding of the cell populations present in culture and their interactions in the dish may contribute to achieving that.
SUMMARY AND CONCLUSION
In summary, these data provide a novel surface antigen code for neural lineage differentiation. Dynamic changes in surface expression of the identified antigens CD15, CD24, and CD29 in combination (Table 1) can be exploited for the experimental separation of key neural cell populations derived from developing human stem cells.
We thank Shreeya Karki and Casper Reske-Nielsen for excellent technical assistance. This work was supported by a National Institutes of Health grant (P50NS39793), the Michael Stern Foundation, the Orchard Foundation, the Consolidated Anti-Aging Foundation, and the Harold and Ronna Cooper Family.
DISCLOSURE OF POTENTIAL CONFLICTS OF INTEREST
The authors indicate no potential conflicts of interest.