|Home | About | Journals | Submit | Contact Us | Français|
Sox2 is expressed by neural stem and progenitor cells, and a sox2 enhancer identifies these cells in the forebrains of both fetal and adult transgenic mouse reporters. We found that an adenovirus encoding EGFP placed under the regulatory control of a 0.4 kb sox2 core enhancer selectively identified multipotential and self-renewing neural progenitor cells in dissociates of human fetal forebrain. Upon EGFP-based fluorescence activated cell sorting (FACS), the E/sox2:EGFP+ isolates were propagable for up to a year in vitro, and remained multilineage competent throughout. E/sox2:EGFP+ cells expressed more telomerase enzymatic activity than matched sox2-depleted populations, and maintained their telomeric lengths with successive passage. Gene expression analysis of E/sox2:EGFP-sorted neural progenitor cells, normalized to the unsorted forebrain dissociates from which they derived, revealed marked over-expression of genes within the notch and wnt pathways, and identified multiple elements of each pathway that appear selective to human neural progenitors. Sox2 enhancer-based FACS thus permits the prospective identification and direct isolation of a telomerase-active population of neural stem cells from the human fetal forebrain, and the elucidation of both the transcriptome and dominant signaling pathways of these critically important cells.
The live-cell specific identification of human neural stem cells has been problematic, in large part due to the lack of selective reporters for these cells. Transgenic mice with reportable neural progenitors have been established using regulatory sequences for the neuroepithelial filament nestin (Kawaguchi et al., 2001; Mignone et al., 2004) and the soxB1 transcription factor sox2 (Zappone et al., 2000; Ferri et al., 2004; Suh et al., 2007). However, transgenic reporting strategies do not lend themselves to use with human tissue-derived cells. Alternatively, virally-delivered nestin and musashi regulatory sequences have been used to isolate neural progenitor cells from fetal human brain tissue (Keyoung et al., 2001), but nestin and musashi are also expressed by lineage-restricted neuronal progenitors and glial progenitors, respectively, limiting their utility to restricted developmental periods and settings. Several surface antigens, such as LeX/CD15 (Capela and Temple, 2002) and CD133 (Uchida et al., 2000), have been reported to recognize neural stem and progenitor cells, but these markers appear neither specific nor selective in human brain tissue, limiting their utility.
In this study, we asked whether an enhancer regulating forebrain neuroepithelial sox2 expression might be used to selectively identify human neural stem and progenitor cells. Sox2 is expressed early in neuroepithelial development, and its expression persists in both mitotically competent neural stem cells and their transit-amplifying daughters (Zappone et al., 2000; Graham et al., 2003; Pevny and Placzek, 2005; Pevny and Nicolis, 2010). Nicolis and colleagues had previously defined a 5' regulatory region within the sox2 gene, which can direct sox2 expression in both neural stem and progenitor cells in the forebrain (Zappone et al., 2000). A small 0.4 kb core enhancer within this region includes a POU factor binding site, which appears sufficient to specify gene expression to the forebrain neuroepithelium (Catena et al., 2004; Miyagi et al., 2006). On this basis, we asked if viruses expressing fluorescent transgenes under the control of this sox2 neuroepithelial enhancer might be used to identify and isolate neural stem and progenitor cells from the human fetal forebrain.
Using adenoviral E/sox2:EGFP to transduce dissociates of the second trimester human VZ/SVZ, followed by EGFP-directed fluorescence activated cell sorting (FACS), we found that the resultant E/sox2:EGFP+ isolates differed from the larger cell populations from which they derived, in terms of their expansion competence, expressed transcriptome, and telomerase activity. The sox2+ isolates were multipotential and capable of sustained mitotic expansion, to a substantially larger degree than were matched sox2− isolates. Accordingly, they expressed higher levels of telomerase enzymatic activity than did E/sox2:EGFP-depleted fractions, and maintained their telomeric lengths with successive passage, consistent with the E/sox2-based enrichment of bona fide neural stem cells. Expression profiling of the E/sox2:EGFP+ cells then revealed that the sox2+ cells manifested a distinct expression signature, characterized by the transcriptional over-representation of both notch and canonical wnt pathways. Together, these observations indicate that adenoviral E/sox2-based fluorescence-activated cell sorting permits the direct isolation of telomerase expressing neural stem cells from the fetal human brain, and suggests a coherent set of transcriptional targets for both their dynamic modulation and phenotypic instruction.
Human fetal brain was taken from second-trimester aborted fetuses of 13–21 weeks gestational age (n=25 total); samples used for genomic analysis (n=4) were limited to 16–19 weeks g.a. Tissues were obtained from aborted fetuses, with informed consent and tissue donation approval from the mothers, under protocols approved by the Research Subjects Review Board of the University of Rochester Medical Center, as well as the Institutional Review Boards of both Weill Medical College of Cornell University, and the Albert Einstein College of Medicine (AECOM); the latter included the Human Fetal Tissue Resource of AECOM and Jacobi Hospital. No patient identifiers were made available to or known by any investigators besides N.S.; no known karyotypic abnormalities were included. The samples (n = 21 for culture, 4 for genomics) were collected into Hank's balanced salt solution (HBSS), and the telencephalic ventricular zone and subventricular layers (VZ/SVZ) were collectively dissected from the remaining forebrain, along the plane of the intermediate zone and subcortical anlagen. The VZ/SVZ was then dissociated as previously described (Keyoung et al., 2001). The cells were resuspended at 2–4 × 106 cells/ml in DMEM/F12/N2 containing 20 ng/ml bFGF (Sigma, St. Louis, MO), and plated in suspension culture dishes (Corning, Corning, NY).
An adenovirus was constructed to express EGFP under the control of the sox2 N2 enhancer, which specifies gene expression to the embryonic VZ. The 0.4 kb sox2 N2 enhancer (Catena et al., 2004) was PCR amplified and then subcloned into pENTR5'. This plasmid was recombined with a plasmid containing a ß-globin minimal promoter upstream of EGFP and a polyA tail, yielding pENTR/D E/sox2:P/ß-globin:EGFP:pA. The 3'-poly A tail, an SV40 early mRNA polyadenylation signal, was adopted from Clontech's pEGFP-N1 construct. The sequence is 50 bp long, as AAATA AAGCA ATAGC ATCAC AAATT TCACA AATAA AGCAT TTTTT TCACT. The resultant pENTR/D E/sox2:P/ß-globin:EGFP:pA plasmid was subsequently cloned into the pAd/PL-DEST vector (Invitrogen, Carlsbad, CA). The pAd/PL-DEST vector was then linearized with PacI, transfected into 293A cells, and viral plaques were allowed to develop for two weeks. The virus was purified using double CsCl2 centrifugation, yielding a titer of 1011 plaque forming units (pfu)/ml.
The pENTR/D E/sox2:P/ß-globin:EGFP:pA plasmid was introduced into the human fetal forebrain dissociates using FuGENE 6 Transfection Reagent (Roche), 12 hours after plating. Briefly, 1.5 μl FuGENE 6 was diluted with 100 μl Opti-MEM I (Invitrogen), mixed and incubated in room temperature for 5 min. 0.5 μg of either pENTR/D E/sox2:P/ß-globin:EGFP or a control vector, pENTR/D P/ß-globin:EGFP, was added to the FuGENE 6/Opti-MEM combination at a 3 (DNA):1 (FuGENE) ratio, then mixed and incubated at room temperature for 45 min. The mixture of FuGENE 6 and plasmid DNA was then added to fetal forebrain cells cultured in 0.5 ml fresh Opti-MEM, incubated at 37°C in 5%CO2. After 14 to 16 hrs, the transfection reagent was replaced with fresh culture media. E/sox2:P/ß-globin:EGFP expression was typically observed by day 3 after transfection.
The day after tissue dissociation, the human fetal forebrain cells were collected, counted and resuspended at a density of 2–4 × 106 cells/ml in DMEM/F12/N2 containing 20 ng/ml bFGF, then replated in suspension culture dishes. The cells were allowed to recover for a few hours before infection with AdE/sox2:EGFP, at 5 MOI for 8–12 hrs in DMEM/F12/N2 with 20 ng/ml bFGF. The cell suspension was then collected, transferred to a centrifuge tube and pelleted at 200g for 10 min to exclude remaining adenovirus. The pelleted cells were resuspended in new DMEM/F12/N2 with 20 ng/ml bFGF, and returned to the original suspension culture dishes. The infected cells were monitored daily until EGFP expression was detected, typically 3–5 days later, and were then dissociated with papain prior to FACS, as previously described (Keyoung et al., 2001).
Flow cytometry and sorting of E/sox2:EGFP+ and E/sox2:EGFP− cells was performed on a FACSAria (Becton Dickinson, San Jose, CA). Cells were washed twice with Ca++, Mg++-free HBSS, then incubated with 50 units of papain (Worthington) and 100 units of DNase I (Sigma, St. Louis, MO) in PIPES at 37°C for 10 minutes with gentle shaking. The samples were spun and then resuspended in 2 ml of DMEM/F12/N2 and dissociated by sequentially triturating with three serially narrowed glass Pasteur pipettes. The papain was inactivated with DMEM/F12/N2 plus 20% FBS and the cells were pelleted. The pelleted cells were resuspended in HBSS (minus MgCl2, CaCl2, MgSO4 and Phenol Red) and then passed over a 40 μm cell strainer (BD). The cells were resuspended to a final concentration of 3–6 × 106 cells/ml with 1.0 mg of 7-amino-actinomycin D. The cells were analyzed by forward and side scatter, for EGFP fluorescence through a 530 ± 30 nm band-pass filter and for 7-AAD fluorescence through a 695 ± 40 nm band-pass filter. The EGFP+ cells were sorted at 2,000–5,000 events/s using a purification mode algorithm. Either Ad/CMV/V5-GW/lacZ-infected cells or uninfected cells were used as a control to set the background fluorescence; a false positive rate of 0.1% was accepted so as to ensure an adequate yield. Sorted E/sox2:EGFP+ and E/sox2:EGFP− cells were plated onto poly-L-ornithine/laminin coated plates in DMEM/F12/N2 supplemented with 20 ng/ml bFGF for 2–5 days and then fixed for immunostaining.
After FACS, AdE/sox2:EGFP+ and AdE/sox2:EGFP− cells were distributed into 6-well plates at 50,000–100,000 cells/ml in serum-free medium supplemented with 20 ng/ml bFGF, 20 ng/ml EGF and 2 ng/ml LIF. After 8 days the number of primary neurospheres in each culture was determined. The primary neurospheres were then dissociated in papain and passaged. The passaged cells were then cultured for another two weeks and secondary spheres were observed. The number of secondary spheres was determined at 30 days. The secondary spheres were plated onto poly-L-ornithine/laminin coated plates in DMEM/F12/N2 plus 1% FBS, without mitogens, for 7–10 days to allow for differentiation.
Human fetal tissues were fixed with 4% PFA for 30 minutes at 4°C. The tissue was then washed three times with 0.1 M PBS. The samples were then mounted in O.C.T and cryosectioned at 15 μm. The sections were washed with 1× PBS, permeabilized with PBS containing 0.1% saponin and 1% normal donkey serum for 30 minutes at RT, and blocked with PBS containing 0.05% saponin and 5% normal donkey serum for 30 minutes. Sections were then labeled with goat anti-human Sox2 (1:1000, R&D Systems) and co-immunostained with one of the following antibodies: mouse anti-human GFAP (1:400, Sigma); mouse anti-human nestin (1:1000, Chemicon); mouse anti-human MAP2 (1:1000; Chemicon); or rabbit anti-human musashi (1:1000, courtesy of Dr. Hideyuki Okano). Alexa Fluor secondary antibodies, including both Alexa-488 and −594 conjugated donkey anti-goat and anti-mouse antisera, respectively, were used at 1:400 (Invitrogen, Carlsbad, CA).
After 2 days in vitro, unsorted cultures or FACS-sorted cultures were fixed for immunocytochemistry. Alternatively, spheres were plated onto poly-L-ornithine/laminin coated plates in DMEM/F12/N2 plus 1% FBS for 10 days. After 10 days, the spheres were fixed for immunocytochemistry. In either case, the cells were rinsed with HBSS, then fixed with 4% paraformaldehyde for 5 minutes at room temperature. The cells were incubated with primary antibodies overnight at 4°C and with secondary antibodies for 1 hour at 25°C. Primary antibodies included those against: sox2 (1:1000, R&D Systems); GFAP (1:400, Sigma); nestin (1:500, Chemicon); musashi (1:500, Chemicon); ßIII-tubulin (1:1000, Covance); and oligodendrocytic sulfatide, as recognized by MAb O4 (1:100, Chemicon).
Total RNA was extracted from E/sox2:EGFP+ and E/sox2:EGFP− cells, as well as that of their parental VZ/SVZ tissues, using RNeasy (Qiagen, Chatsworth, CA) and amplified using 3'-biased ribo-SPIA (NuGen Ovation). Amplified product was labeled and hybridized onto Affymetrix U133+2 arrays according to manufacturer's instructions (Affymetrix). Microarray analysis was performed in Genespring GX 7.3.1 (Agilent). Raw CEL data was processed using the MAS5 method. Poor quality data was removed by filtering those probe sets whose raw expression control levels did not exceed the average ratio of base:proportional coefficients as defined by the cross-gene error model in at least half of the arrays. Further filtration was performed using MAS5 generated present/absent calls to remove those probe sets called `absent' in all samples. Significantly variable genes were then defined by 1-way ANOVA followed by post-hoc t-test. P-values were corrected using the false discovery rate (Benjamini and Hochberg, 1995). Following identification of significant probe sets, annotation was performed using a custom SQL database and gene ontology analysis performed. Venn diagrams were formulated using Bioinformatics Resource Manager Software (BRM v2.0) obtained from the Pacific Northwest National Laboratory (http://omics.pnl.gov).
Parametric gene set enrichment analysis was performed using KEGG pathways in Bioconductor and R (Gentleman et al., 2004). Following RMA based normalization and filtration to retain only the most variably annotated probe set for each gene, parametric gene set enrichment analysis was performed comparing sorted E/sox2:EGFP+, depleted (E/sox2:EGFP−) and unsorted fetal VZ/SVZ tissue dissociates, using a linear modeling approach (n=4 each). P-values were corrected using FDR and significance assigned at 5% FDR (q=0.05).
Extracted total RNA was amplified using ribo-SPIA based whole transcriptome based-amplification (NuGen). The expression of 95 cell type marker and pathway-specific genes was assessed using a 96 gene Taqman low-density array (Applied Biosystems). The complete list of primer/probes is given in Supplemental table 1. The relative abundance of transcript expression was calculated by DDC analysis, and the expression data normalized to GAPDH. Genes whose expression was not detected in more than half of the RNA samples were excluded. Statistical analysis was performed on log2-transformed data and p-values calculated following a paired t-test. Further analysis of differentially expressed genes was then performed using Bioconductor and R (Gentleman et al., 2004) using a moderated t-test statistic with 5% false discovery rate cut-off (Smyth, 2004).
The telomeric repeat amplification protocol (TRAP) assay was conducted as described (Roy et al., 2004; Roy et al., 2007), using the TRAPeze telomerase detection kit (Chemicon), with the following modifications. The cells were collected by centrifugation, washed once with 1× PBS and extracted with 1× cold CHAPS lysis buffer (supplemented with 100U/ml RNaseOUT) Sample extracts corresponding to 10,000 or 20,000 cells were examined for telomerase activity using the following parameters: incubate at 30°C for 30 minutes then amplify by PCR in a thermocycler: 94°C for 30 sec denaturation, 59°C for 30 sec annealing, and 72°C for 1 min extension, for 33 cycles. 5 μl of loading dye was added to each sample and 25 μl of each sample was loaded onto a 12.5% non-denaturing PAGE gel. The gel was stained with SybrGreen at a dilution of 1:10,000 in 0.5× TBE for 40 minutes and then visualized with a 254 or 302 nm UV transIlluminator. The gel image was acquired with Kodak Molecular Imaging software, and the total products generated from each sample (TPG units) were quantified as described by the TRAPeze manufacturer.
Genomic DNA was isolated from cells using a genomic DNA extraction kit (High Pure PCR Template Preparation Kit, Roche) as per the manufacturer's instructions. Telomere length was determined as described (Roy et al., 2004; Roy et al., 2007), by non-radioactive Southern blot analysis of terminal restriction fragments obtained by the digestion of genomic DNA, using the TeloTAGGG Telomere Length Assay kit (Roche). 500 ng of DNA was digested using the restriction enzymes Hinf I and Rsa I. The digest DNA was then separated by electrophoresis on a 0.8% agarose gel in 1× Tris-acetate-EDTA (TAE) buffer. The digested DNA from the gel was transferred to a nylon membrane via Vacuum Blotter (Model 785, Bio-Rad) at room temp using 20 × SSC as a transfer buffer for 2 hrs. The membrane was hybridized with the Telomere specific probe, washed, incubated with anti-DIG-Alkaline phosphatase and then detected using CDP-Star chemiluminescence substrate. Determination of Mean TRF length is done as follows: ∑ (ODi)/∑ (ODi/Li) where ODi is the densitometer output and Li is the length of the DNA position at i. The amount of telomeric DNA was calculated by integrating the volume of each band, using Kodak Molecular Imaging software.
To assess the normal distribution of sox2 expressing cells in the mid-second trimester human forebrain, samples were fixed and immunostained for sox2, using an antibody whose lack of cross-reactivity with other soxB family members we verified by Western blot (Supplemental Figure 1). In 16 week g.a. fetal brain tissue, sox2+ cells were abundant in the VZ of the lateral ventricle; sox2 immunoreactivity (IR) was predominant in the VZ, in which virtually all cells were sox2+, but was less so in the SVZ. Abundant sox2-IR was also observed in the intermediate zone (Figs. 1A); these latter sox2+ cells of the IZ were excluded from our in vitro and genomics assessments by selective dissection along the SVZ-IZ border, which could be identified by a distinct change in tissue consistency in fresh tissue. In the fetal VZ/SVZ, sox2-IR was co-expressed with the neural progenitor filament nestin (Figs.1B–C) and the neural progenitor RNA binding protein musashi1 (Msi1) (Figs. 1D–E). At 16 wks gestational age, few cells expressed glial fibrillary acidic protein (GFAP) in the VZ, although GFAP+ cells were evident in the SVZ, many of which manifested radial glial morphology and co-expressed sox2 (Fig. 1F–G). Sox2+ cells were never noted to co-express the mature neuronal marker MAP2, either in the VZ/SVZ, IZ, or cortical plate (not shown).
Previous studies have reported that a 400 base pair sequence located 2 kb upstream of the translation start codon of the sox2 gene, designated the N2 region, acts as a core enhancer which specifies sox2 expression to both mouse and chick forebrain neural progenitor cells (Zappone et al., 2000; Uchikawa et al., 2003; Ferri et al., 2004) (Fig. 2A). To assess if the 0.4 kb N2 enhancer might be sufficient to direct sox2 expression in human forebrain neural stem and progenitor cells, we subcloned it upstream of a ß-globin basal promoter, 5' to an EGFP reporter. The resulting cassette, E/sox2(0.4):P/ßglobin: EGFP:polyA, was subcloned into the pENTR/D plasmid vector (Fig. 2B). The expression and phenotypic specificity of the resultant E/sox2(0.4):P/ßglobin:EGFP:polyA (hereafter referred to as E/sox2:EGFP) was assessed in dissociated human fetal forebrain cells, by plasmid transfection. followed by immunolabeling for sox2 and nestin. E/sox2:EGFP was observed in immature neural populations (Fig. 2C–D), that co-expressed both sox2 and nestin immunoreactivity. On that basis, we next sought to increase transduction efficiency, by generating an adenoviral vector containing the E/sox2-P:EGFP expression cassette. Adenoviral E/sox2-P/ßglob-EGFP:polyA (AdE/sox2:EGFP) was thereby generated in 293A cells, then titered in primary human fetal neural cells. Dissociates of the human fetal forebrain were then infected with AdE/sox2:EGFP, over a range of 1–50 multiplicities of infection (moi). A standard dose of 5 moi was chosen on the basis of achieving efficient transduction efficiency with no evident toxicity. After EGFP expression 3–5 days later, the cells were then fixed with 4% paraformaldehyde. Immunolabeling confirmed the expression of sox2 protein (Fig. 2E), as well as both nestin and musashi-1 (Fig. 2F–G), by 95–100% of E/sox2:EGFP+ cells. In addition, an average of 31.9 ± 3.0% of E/sox2:EGFP+ cells co-expressed GFAP (Fig. 2H); these latter cells typically exhibited a radial glial morphology, suggesting the radial cell phenotype of a fraction of the sox2-defined pool. Together, these data indicate that the 0.4 kb upstream enhancer of the sox2 gene identifies a sox2+/nestin+/musashi+/GFAP± neural population from the developing human brain.
To isolate sox2+ cells from the larger forebrain population, we next infected dissociated VZ/SVZ with AdE/sox2:EGFP, and then used FACS to extract the E/sox2:EGFP+ cell fraction. To this end, we exposed fresh dissociates of 13–21 wk. g.a. human fetal forebrain VZ/SVZ to 5 moi of adenovirus at a titer of 2 × 1011/ml. Positive and negative sort controls were also infected, with AdCMV:EGFP or AdCMV:lacZ, respectively. At 16 weeks and beyond, the VZ/SVZ were dissected free of the overlying IZ and cortical mantle, the border of the SVZ and intermediate zone presenting a definable border on gross examination. For the 13–15 wk g.a. samples, this border was not distinct, resulting in less discrete dissections.
Within 2 days of infection, a fraction of AdE/sox2:EGFP-infected cells were noted to express EGFP, indicating their transcriptional activation of the sox2 enhancer. At 5 days, these cultures were subjected to fluorescence-activated cell sorting (FACS), using matched cultures of AdCMV:lacZ infected or uninfected cells to establish gating thresholds (Fig. 3A–B). Cytometry revealed that the incidence of E/sox2:EGFP+ cells in these samples averaged 4.3 ± 0.79%, when pooled across all samples (13–21 wk g.a., n=21; mean ± SD) (Figs. 3C–D). To correct for the transduction efficiency of adenoviral E/sox2:EGFP infection, the transduction efficiency of their AdCMV:EGFP-infected positive controls was assessed by flow cytometry at 5 days, at matched viral exposures of 5 moi. These controls revealed that 40.3 ± 7.1% of AdCMV:GFP-exposed cells were GFP+; no toxicity whatsoever was observed at this dose. The absolute incidence of E/sox2:EGFP+ cells in each preparation was then estimated by multiplying the E/sox2:EGFP+ incidence by the resultant correction factor of 1/0.403=2.48, yielding an average net incidence of E/sox2:EGFP+ cells, across gestational ages, of 10.7%. Interestingly, no significant difference was noted in the incidence of E/sox2:EGFP+ cells as a function of gestational age, over the age range examined.
Microscopy confirmed that 2 hrs after FACS, virtually all sox2-sorted cells expressed EGFP fluorescence (Figs. 3E–F); in contrast, only rare EGFP+ cells were visualized in the E/sox2:GFP− fraction, and none were ever seen in uninfected controls (Fig. 3G–H). Immunolabeling confirmed that immediately after FACS, 85.3 ± 1.8% of the E/sox2:EGFP+ cells expressed detectable sox2-immunoreactivity, while only 20.5 ± 4.7% and 35.7 ± 5.2% of sox2:EGFP-depleted and unsorted cells, respectively, did so (Figs. 4A, D–E; n=3 experiments; >3500 cells scored). Similarly, whereas 65.9 ± 2.1% of E/sox2:EGFP+ cells expressed the neuroepithelial filament nestin, only 16.9 ± 2.9% and 31.7 ± 1.8% of the sox2:EGFP-depleted and unsorted cells, respectively, co-expressed nestin (Figs. 4B, F–G). By 3 days after FACS, E/sox2:EGFP+ cells almost uniformly expressed sox2 immunoreactivity (95.0 ± 1.05%), while all co-expressed both nestin (100 ± 0%) and musashi1 (100 ± 0%); in addition, a third immunolabeled as GFAP+ (31.9 ± 3.02%) (n=3; Figs. 4C, H–I). These results indicate that AdE/sox2:EGFP-based FACS isolates a neural progenitor population that uniformly co-express the neural stem and progenitor cell markers sox2, nestin and musashi1.
To assess the clonogenicity of sox2+ cells, E/sox2:EGFP+-sorted cells and their EGFP− remainders were next raised in low-density, serum-free, EGF, LIF and FGF2-supplemented suspension culture, and their production of neurospheres quantified. By way of comparison, matched cultures of unsorted and EGFP− cells were also included. Sorted cells were seeded at 50–100,000 cells/ml (Figs. 5A–B), and were cultured in suspension to yield primary neurospheres, which were then re-dissociated and passaged to generate secondary spheres. The E/sox2:EGFP+ cells generated small neurospheres as early as 2–3 days post-sort (Fig. 5C–D), while E/sox2:EGFP− cells largely remained as single cells or died in the depleted media. Within the 2 weeks thereafter, E/sox2:EGFP+ cells formed larger neurospheres (Figs. 5E–H), in contrast to EGFP− cells, which formed smaller spheres slowly if at all, while exhibiting ongoing death. At 2 weeks, the mean number of neurospheres generated from the E/sox2:EGFP+ and E/sox2:EGFP− fractions were 71.9 ± 22.4 and 12.9 ± 0.3 per 105 plated cells, respectively; 6-fold more neurospheres were generated in the E/sox2:EGFP+ pool than in its E/sox2-depleted remainder (p=0.058, n=3) (Fig. 5I). The E/sox2:EGFP+ neurospheres continued to express EGFP for at least 4 weeks, though the intensity of EGFP gradually weakened over time (not shown). Neurospheres derived from E/sox2:EGFP+ cells were readily maintained >6 months, with monthly passages. The number of neurospheres arising from E/sox2:EGFP+ cells rose from 90 ± 13/105 cells at first passage, to 476 ± 185/105 at passage 3 (Fig. 5J); serial expansion slowed thereafter, yet 151 ± 30 neurospheres were still generated per 105 cells at 5 months in vitro.
To define the lineage potential of E/sox2:EGFP+ cells, neurospheres sampled from serial monthly passages, taken over a 5 month period in vitro, were switched from serum-free conditions to conditions encouraging phenotypic differentiation. The latter included passage onto a laminin substrate in media supplemented with 1% FBS, in which the cultures were maintained for 10 days, then fixed and immunostained for neuronal and glial markers. Both ßIII-tubulin+ neurons and GFAP+ astrocytes were generated from single spheres, with smaller numbers of O4+ oligodendrocytes (Figs. 6A–D). With serial passage, robust multilineage cell genesis persisted, but a progressively lower ratio of neurons to astrocytes was observed. Thus, whereas at the third monthly passage (P3), 56.1 ± 9.8% (mean ± SD) of cells expressed neuronal ßIII-tubulin and 19.5 ± 3.7% expressed astrocytic GFAP, by P6, 44.5 ± 14.4% of cells expressed ßIII-tubulin and 23.8 ± 6.9% were GFAP+. Though the fall in the ratio of neurons to glia was slowly progressive over time, it was nonetheless significant (p=0.002). Although oligodendrocyte generation in this 1% FBS-supplemented media was less common, such that <2% of the E/sox2:EGFP-generated progeny typically expressed oligodendrocytic O4, at least some O4+ cells were nonetheless identified throughout serial passage (Fig. 6D). Thus, when allowed to differentiate in serum-supplemented media, E/sox2:EGFP-derived neurospheres uniformly generated both neurons and glia, reflecting the multilineage potential of their E/sox2:EGFP+ founders. Together, these results indicate that E/sox2:EGFP+-based selection significantly enriches the population of self-renewing neural progenitor cells derivable from the fetal human brain, and that the E/sox2:EGFP+-cells thereby isolated are capable of sustained expansion with preserved multilineage competence.
In both embryonic and somatic stem cell populations, the telomerase enzymatic complex functions to sustain telomeric length and hence mitotic competence, by counteracting the division-associated erosion of telomeres (Harley et al., 1990; Allsopp et al., 1992). As stem cells differentiate into restricted lineages, the transcription of hTERT, the rate-limiting component of the telomerase complex, falls. Their telomerase enzymatic activity quickly follows suit, as a result of which their telomeres predictably shorten with further division (Harley et al., 1990). Thus, stem cells are characterized by active hTERT transcription, and hence hTERT promoter activation, which is associated with robust telomerase enzymatic activity, as typically assessed by the TRAP assay. In contrast, most stably differentiated somatic cells lack telomerase activity, as a result of which their telomeres are shorter than those found in stem cells, having receded in inverse proportion to the number of divisions experienced by the cells (Wright and Shay, 2005).
Since E/sox2:EGFP+ cells are both self-renewing and multipotential upon their initial isolation from human fetal brain tissue, we asked whether E/sox2:EGFP-based FACS might enrich telomerase-expressing neural stem cells, and hence telomerase activity, from the larger population of brain cells. To test this hypothesis, telomerase enzymatic activity was examined after FACS in E/sox2:EGFP+ and E/sox2:EGFP− populations, using the telomeric repeat amplification protocol (TRAP) assay (Wright et al., 1995; Herbert et al., 2006). After E/sox2:EGFP+ and EGFP− populations were isolated from the VZ tissue of human fetal forebrain, their telomerase activities were assessed and compared to those of their parental VZ dissociates, after protein extraction. The TRAP-assessed telomerase activity was markedly higher in E/sox2:EGFP+ cells (23.9 units) than in their E/sox2:EGFP− (9.6 units) counterparts (Figs. 7A–B). Unlike the substantial differences in telomerase enzymatic activity, telomeric length did not differ between the sox2+ and sox2− pools. Instead, the telomere length assay (TLA) (Roy et al., 2007) revealed that the maximal length of telomeres of the E/sox2:EGFP+ sorted cells and their depleted remainder were roughly equivalent, at 15–17 kb (Fig. 7C). Yet given the early gestational age sampled here, and the slow decline in telomeric length after telomerase activity ceases, it is not surprising that we failed to note significant telomeric erosion in the sox2− pool, which is still mitotically active at this point in ontogeny.
To establish the genotype of human fetal sox2+ cells, we used Affymetrix microarrays to compare the gene expression patterns of AdE/sox2:EGFP+ cells to their corresponding EGFP− remainder, as well as to unsorted VZ, using 4 forebrain samples derived from 16–19 weeks g.a. fetuses. The total RNAs of E/sox2:EGFP+ and E/sox2:EGFP− cells, as well as that of their parental VZ/SVZ tissues, were extracted, amplified and analyzed with Affymetrix U133-Plus 2 arrays. A total of 4 matched sets of each of these 3 groups (sox2+, sox2−, and VZ/SVZ) were thus examined, for a total of 12 microarrays. Among 31,444 represented probe sets detected as present on at least one of the 12 chips, 10,012 were differentially expressed by 1-way ANOVA (5% FDR). Using a 4-fold difference cutoff (defined as significant by t-test, at 5% FDR), we found that among these transcripts, 2,187 were differentially expressed by E/sox2:EGFP+ cells relative to E/sox2:EGFP− cells, and 2,285 by E/sox2:EGFP+ cells relative to the VZ/SVZ tissue from which they derived. Subsequent Venn analysis revealed 1342 probe sets that were differentially expressed by E/sox2:EGFP+ cells relative to both E/sox2:EGFP− cells and VZ tissue (Fig. 8A). Of these, 706 were up-regulated, which were annotated to 492 unique genes; 636 were down-regulated, representing 411 unique annotations. Gene Ontology analysis of these transcripts revealed functionally related sets of differentially expressed genes, as summarized in table 1, with the complete list provided in Supplemental table 2, as well as at http://www.urmc.rochester.edu/ctn/goldman-lab/supplemental-data.cfm.
The gene expression pattern of the E/sox2:EGFP+ cells differed significantly from that of E/sox2:EGFP− cells, as well as from that of the parental VZ/SVZ from which each derived. In contrast, the expression patterns of the sox2− pool and VZ tissue largely overlapped (Fig. 8A). E/Sox2:EGFP+ cells expressed a number of known NSC genes, such as nestin (8-fold enriched relative to VZ, and 7-fold higher than in E/sox2:EGFP− cells), and CD133 (4.4-fold higher in E/sox2:EGFP+ cells than in parental VZ). Sox2 mRNA was 3.4 fold and 1.8 fold higher in the E/sox2:EGFP+ pool than in the E/sox2:EGFP− cells and VZ, respectively. Moreover, the other members of the sox B1 subgroup, sox1 and sox3, were expressed 2.0 and 8.2 fold higher in E/sox2:EGFP+ than in E/sox2:EGFP− cells, just as pax6 was 4.1 and 3.5-fold higher in E/sox2:EGFP+ than in E/sox2:EGFP− cells or VZ tissue, respectively (Table 2 and Fig. 8B). In addition, the radial glial transcripts GLAST and BLBP were both differentially over-expressed by E/sox2:EGFP+ cells, consistent with the progenitor phenotype of radial cells, and the observed co-expression of these transcripts in neurogenic radial cells (Ghashghaei et al., 2007; Suh et al., 2007). Similarly, GFAP mRNA was 5-fold higher in E/sox2:EGFP+ than E/sox2:EGFP− cells, consistent with the expression of GFAP by murine neural stem and progenitor cells (Doetsch et al., 1999; Garcia et al., 2004). Together, these data suggest that sox2-directed isolation selects a discrete population of cells that express transcripts typifying neural progenitor phenotype.
In contrast, E/sox2:EGFP+ cells were relatively depleted in genes typical of either terminally differentiated neurons and glia, or of their phenotypically committed progenitors. For instance, the DLX genes, mash1, and the mature neuronal markers, HuD and MAP2 were all depleted in E/sox2:EGFP+ cells, and were instead over-expressed within the corresponding E/sox2:EGFP− fractions (Table 1). Similarly, no markers typically expressed by mature oligodendrocytes were differentially expressed by E/sox2:EGFP+ NSCs (Table 1 and Fig. 8B). Together, these data indicate that E/sox2:EGFP+ cells exhibit a gene expression signature typical of neural stem cells.
E/sox2:EGFP+ cells exhibited differential expression of a number of receptors known to be involved in neural stem and progenitor cell expansion (Table 3). FGFR1, EGFR and LIFR were markedly over-expressed by human E/sox2:EGFP+ cells, in accord with the described actions of the fibroblast growth factors, EGF and TGFα, and LIF on these cells (Reynolds and Weiss, 1992; Vescovi et al., 1993; Morshead et al., 1994; Craig et al., 1996; Qian et al., 1997; Fallon et al., 2000; Doetsch et al., 2002; Bauer and Patterson, 2006; Bauer et al., 2007). Parametric gene set enrichment analysis of KEGG pathways (http://www.genome.jp/kegg/pathway.html) (Fig. 9A) revealed that the MAP kinase (Fig. 9B) and JAK-STAT (Fig. 10A) pathways were significantly regulated in E/sox2:EGFP+ cells (FDR corrected q-values = 0.023 and q = 0.033, respectively). In addition, E/sox2:EGFP+ cells expressed high levels of the MET receptor (13-fold higher in E/sox2:EGFP+ then E/sox2:EGFP− cells by microarray; 24-fold higher by qPCR, p=0.0002), suggesting a possible role for the c-met ligand HGF in regulating stem and progenitor cell turnover in the human VZ/SVZ.
In this regard, it is notable that the E/sox2:EGFP+ cells prominently over-expressed not only EGFR, but also its cognate ligand EGF (13.6-fold higher in E/sox2:EGFP+ than E/sox2− cells), and did so concurrently with over-expressing the BMP inhibitors noggin (16.6-fold higher by microarray; 11.8-fold by qPCR; p=0.04) and BAMBI (8.1-fold higher by microarray; 10.7-fold by qPCR; p=0.0005) (Table 3 and Supplemental Table 2). Given the pro-gliogenic roles of the BMPs, these data suggest that E/sox2-defined neural progenitors may exhibit a tonic suppression of BMP signals via a combination of BAMBI expression and noggin release, similar to the strategy of active BMP inhibition employed by adult glial progenitor cells. By inhibiting BMP signals concurrently with autocrine EGF signaled expansion, E/sox2:EGFP+-defined stem and progenitor cells would appear to have strong autocrine mechanisms by which to preserve their competence for undifferentiated expansion.
E/sox2:EGFP+ cells also expressed several categories of transcripts associated with undifferentiated self-renewal. Foremost among these were a number of wnt pathway members, both ligands and receptors. KEGG gene set enrichment analysis showed that wnt pathway transcripts were significantly over-represented in E/sox2:EGFP+ cells (q = 0.026) (Fig. 9C). Interestingly, hyper-geometric tests on KEGG pathways further revealed that gene sets associated with colorectal cancer, small cell lung cancer, prostate cancer and p53 regulated pathways were all differentially expressed by E/sox2:EGFP+ cells, relative to E/sox2:EGFP− cells (p<0.05 on genes over expressed in E/sox2:EGFP+ cells), highlighting the commonalities between self-renewal pathways in normal neuroectodermal development and cancer. In regards to individual up-regulated components of the wnt pathway, our microarray data indicated that E/sox2:EGFP+ cells over-expressed wnt16, a little-studied canonical wnt ligand closely related to wnt7 (Mazieres et al., 2005), as well as the wnt family receptors FZD-6, FZD-8, and FZD-10. Indeed, FZD-10 was over-expressed by 33- and 55-fold, respectively, relative to E/sox2:EGFP− cells and VZ tissue (Table 3). Quantitative real-time PCR confirmed the relative over-expression of these wnt receptor transcripts by E/sox2:EGFP+ cells, validating the microarray data; FZD-6, −8 and −10 mRNAs were expressed at levels respectively 50-fold, 18-fold and 180-fold higher in E/sox2+ than E/sox2− cells (Table 3). Interestingly, E/sox2:EGFP+ cells also differentially over-expressed the wnt antagonists DKK1 (14.4-fold higher in E/sox2:EGFP+ than in E/sox2− cells) and DKK3 (7.3-fold higher), suggesting that wnt-regulated neural progenitors might be inhibited by their neighbors.
Further manual pathway analysis revealed that besides wnt pathway components, E/sox2:EGFP+ cells also selectively over-expressed gene sets indicative of active notch signaling (Fig. 10B), which has been previously implicated in the regulation of neural stem and progenitor cell self-renewal. In regards to notch pathway components, E/sox2:EGFP+ cells were found to over-express notch2, notch3, hes1 and hes5, all positive activators and/or effectors of notch signaling (Table 3). Conversely, genes typically suppressed by notch receptor activation, such as numb and the bHLH transcription factors mash1, ngn1/2, and neuroD1/2, were all relatively under-expressed in E/sox2:EGFP+ cells relative to their sox2-depleted controls, again suggesting the maintenance of active notch signaling in E/sox2+ cells.
In this report, we used a sox2 enhancer-based selection strategy to isolate and profile uncommitted neural precursor cells from the second trimester human fetal ventricular and subventricular zones. These cells expressed sox2 protein, were multilineage competent, and proved capable of sustained self-renewal in vitro, indicating their inclusion of competent neural stem cells. In addition, these E/sox2:EGFP-defined isolates sustained both telomerase transcription and enzymatic activity, in contrast to their matched E/sox2:GFP-depleted populations, which failed to do so. Genomic analysis of the E/sox2:EGFP+ cell population, and comparison to the matched unsorted and sox2-depleted populations, revealed that E/sox2:EGFP+ cells differentially expressed a discrete set of transcripts suggestive of active wnt, EGF and notch pathway signaling.
Neural stem cells are defined as both multilineage competent for neurons and glia, as well as self-renewing. In humans, self-renewal competence is attended by sustained telomerase activity, which prevents telomeric erosion, thereby maintaining mitotic competence (Wright et al., 1996; Wright and Shay, 2005). Stem cell-derived daughter cells - which may be defined as transit amplifying progenitors, based upon their mitotic expansion concurrent with departure from the ventricular zone - include restricted neuronal and glial progenitors, as well as still-multipotential progenitors, all of which may share a loss of self-renewal competence (Goldman, 2005b; Goldman, 2005a). At least in development, this loss in self-renewal is associated with a fall in the telomerase activity of neural progenitors (Ostenfeld et al., 2000). As a result, whereas a sox2− pool devoid of self-renewing stem and progenitor cells would be expected to manifest little telomerase enzymatic activity, one might anticipate that a sox2+ pool, comprised of both neural stem cells and their derived transit-amplifying progenitor cells, would exhibit sustained telomerase activity. We observed this to be the case, as telomerase enzymatic activity was robustly higher in the E/sox2:EGFP+ cells, relative to their depleted remainder of E/sox2− cells.
This maintenance of telomerase activity by E/sox2+ cells was associated with their corresponding differential over-expression of telomerase-dependent transcripts, among others. Indeed, E/sox2:EGFP+ cells expressed several categories of transcripts associated with both mitogenic expansion and undifferentiated self-renewal. Likely mitogenic pathways were represented largely by genes associated with the MAP kinase pathways, activators of which were significantly up-regulated in E/sox2:EGFP+ cells. In particular, E/sox2:EGFP+ cells differentially expressed a number of receptor tyrosine kinases, that included the class 1, 4 and 6 receptor tyrosine kinases EGFR, FGF1/2R, and MET, respectively, each of which has been previously implicated in the expansion of neural stem and progenitor cells. Yet perhaps most striking in their degree of over-expression by E/sox2:EGFP+ cells were components of the wnt signaling pathway. Wnt ligands have been noted to comprise self-renewal signals for murine neural stem cells (Kalani et al., 2008), and may serve a similar function in human neural precursors as well. Nonetheless, we were surprised to note that in human fetal VZ/SVZ-derived E/sox2:EGFP+ cells, the most specifically and differentially over-expressed wnt ligand proved to be wnt16, a functionally obscure molecule that may signal through the canonical pathway (Mazieres et al., 2005). In addition, the wnt receptors FZD-6, FZD-8, and FZD-10 were all highly over-expressed, as were a number of wnt target genes. These over-expressed wnt targets included dickkopf3 (DKK3), follistatin (FST), Krupple-like factor 5 (KLF5), TWIST1, jagged1 (JAG1), NrCAM (NRCAM), RUNX2, autotaxin (ENPP2), and cyclin D (CCND1), among others. Several of these are of especial interest. The simultaneous over-expression of the wnt 16 and fzd-6, -8 and -10 together with DKK3, an antagonist of LRP5/6 and hence of wnt signaling, suggests the possibility of autocrine wnt signaling with concurrent lateral inhibition of neighboring wnt-regulated neural progenitors.
E/sox2:EGFP-defined cells were noted to differentially express a number of other wnt target genes, a number of which may serve to modulate wnt pathway activity. For instance, the E/sox2:EGFP-overexpressed transcript NrCAM has been shown to bind and regulate the receptor tyrosine phosphatase PTPRZ, which is involved in both radial cell maintenance and ß-catenin-dependent progenitor cell expansion, introducing yet another mechanism by which the undifferentiated self-renewal of E/sox2:EGFP+ cells might be regulated in trans. The differential up-regulation of these wnt-regulated genes in E/sox2:EGFP-defined progenitors is consistent with their distinct roles in both maintaining the undifferentiated self-renewal of early neuroepithelial stem cells, and in potentiating the phenotypic diversification of both neighbors and daughter cells into differentiated derivatives.
In regards to the latter, E/sox2:EGFP+ cells were found to over-express a number of serially activated components of the notch pathway, suggesting active notch signaling and concurrent repression of genes associated with the assumption of a differentiated fate. Notch2, notch3, hes1 and hes5, all positive activators and/or effectors of notch signaling, were all differentially over-expressed by E/sox2:EGFP+ cells (Fig. 10B). Conversely, genes typically suppressed by notch receptor activation, such as numb and the bHLH transcription factors mash1, ngn1/2, and neuroD1/2, were all relatively under-expressed in E/sox2:EGFP+ cells, again suggesting active notch signaling. In this regard, it is interesting to note that E/sox2:EGFP+ cells differentially expressed high levels of jagged1 (JAG), a trans-activating positive stimulus for notch signaling. The high level expression of JAG by E/sox2:EGFP+ cells suggests again that E/sox2:EGFP+ progenitors may exert a strong influence upon the turnover and fate of their neighbors.
E/sox2:EGFP+ cells also manifested high differential over-expression of noggin (NOG), a soluble antagonist of the bone morphogenetic proteins, and hence of BMP signaling. Given the pro-gliogenic actions of the BMPs (Mabie et al., 1997; Lim et al., 2000), the over-expression of noggin by sox2+ NSCs might serve to prevent their premature glial differentiation, and hence preserve their undifferentiated expansion competence (Kondo and Raff, 2004). Together, these data suggest the concurrence of wnt and notch signaling in E/sox2:EGFP-defined neural stem and progenitor cells, occurring in the context of a noggin-mediated minimization of concurrent BMP receptor-dependent signaling. Acting in concert, these signals act to ensure the cell-autonomous, active repression of terminal differentiation by these phenotypically plastic neural progenitor cells.
Thus, sox2 enhancer-based FACS permits the prospective and selective enrichment of a population of multipotential, self-renewing, and telomerase-expressing neural precursor cells from the fetal human brain. These E/sox2:EGFP+ cells included an hTERT+ fraction with active telomerase enzymatic activity, that likely defined the self-renewing fraction of neural stem cells within the larger pool of E/sox2:EGFP-defined neural progenitors. Most importantly, the E/sox2-based isolation of human neural stem and progenitor cells has permitted the definition of both the transcriptome and dominant signaling pathways of these cells. Assessed in the context of regional gene expression within the second trimester human forebrain (Johnson et al., 2009), these progenitor-selective pathways should permit the prediction of ligand-receptor interactions with both neural and non-neural cells within the host germinal matrix. Indeed, by defining their selective engagement of distinct receptor tyrosine kinase, wnt, notch and BMP signaling pathways, and by identifying the specific receptors employed by human neural progenitors, this analysis provides substantial molecular insight into how the expansion and fate of human neural stem and progenitor cells may be modulated for therapeutic benefit.
We thank Brad Poulis of the Human Fetal Tissue Resource of the Albert Einstein College of Medicine for assistance in tissue acquisition. Supported by the Adelson Medical Research Foundation, the Mathers Charitable Foundation, the James S. McDonnell Foundation, the New York State Stem Cell Science Board, and NINDS grants R01NS33106 and R01NS39559. S.N. is supported by Telethon and Fondazione Cariplo MIUR.