SEARCH

SEARCH BY CITATION

Keywords:

  • JMJD5;
  • Human embryonic stem cells;
  • G1 phase;
  • Pluripotency;
  • CDKN1A (p21)

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

In mammalian embryos, embryonic stem cells (ESCs) and induced pluripotent cells, a shortened G1 phase is correlated with the pluripotent state. To molecularly define this phase, we compared transcripts from the shortened G1 of human ESCs (hESCs) with those from the longer G1 of derived endoderm. We identified JMJD5, a JmjC (Jumonji C) domain containing protein that, when depleted in hESCs, causes the accumulation of cells in G1 phase, loss of pluripotency, and subsequent differentiation into multiple lineages, most prominently ectoderm and trophectoderm. Furthermore, we demonstrate that the JMJD5 phenotype is caused by the upregulation of CDKN1A (p21), as depleting both JMJD5 and CDKN1A (p21) in hESCs restores the rapid G1 phase and rescues the pluripotent state. Overall, we provide genetic and biochemical evidence that the JMJD5/CDKN1A (p21) axis is essential to maintaining the short G1 phase which is critical for pluripotency in hESCs. Stem Cells 2014;32:2098–2110


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

Cell cycle regulation is intertwined in the mechanisms underlying pluripotency maintenance in embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs) [1-7]. In human ESCs (hESCs) and iPSCs, the G1 phase is rapid, lasting only 2–3 hours, compared to 6–12 hours in other cell types [8-11]. Furthermore, in hESCs at any given time, 19% of cells are in G1 phase, compared to 40%–50% in other cell types [8]. Upon differentiation, hESCs acquire a lengthened G1 phase, while upon reprogramming, iPSCs shorten their G1 phase [4, 11-14]. This strong correlation between cells with an abbreviated G1 phase and the pluripotent state suggests that G1 timing may be an essential mechanism of pluripotency control.

The mechanisms underlying the connection between the abbreviated G1 and pluripotency are only beginning to be unraveled. Generally, pluripotent cells sustain high levels of CDK2, a major driver of the G1 progression into S phase, and low levels of CDKN1A (p21) and CDKN1B (p27), two cyclin-CDK2 inhibitors known to inhibit cell cycle progression [4, 15-21]. Expression of CDKN1A (p21) and CDKN1B (p27) increases upon differentiation and is associated with a lengthened G1 phase [13, 16, 22, 23]. In support of the involvement of CDK2 and CDKN1A (p21), inhibition of CDK2 leads to cellular arrest in G1, an increase in CDKN1A (p21) and CDKN1B (p27) expression, loss of pluripotency, and subsequent differentiation [16, 24-26]. Additionally, overexpression of CDKN1A (p21) displays a similar phenotype—the accumulation of cells in G1 phase and differentiation [3]. Interestingly, the same molecules appear to control the shortening of the G1 in reprogrammed cells as overexpression of CDKN1A (p21) inhibits reprogramming [3]. This manipulation of CDKN1A (p21) by various factors has been a major focus recently as a potential means to improve reprogramming frequencies [27-29].

While evidence mounts for a strong connection between an accelerated G1 and pluripotency being governed by CDK2 expression in the absence of CDKN1A (p21), the mechanisms underlying this connection are just beginning to be elucidated. MicroRNAs, particularly the miR-290 and miR-302 clusters, are important for the post-transcriptional regulation of CDKN1A (p21) in ESCs [30, 31]. These micro RNAs, which are expressed highly in ESCs, were shown to directly bind the CDKN1A 3 prime UTR and inhibit protein processing. More recently, it was shown that members of the JmjC family, including JMJD2A, JMJD2B, JAR1D1B, and JMJD5, can affect the expression of CDKN1A in various cell types, suggesting that the CDKN1A locus is sensitive to levels of these proteins [32-35]. Intriguingly, there may be a direct connection between the transcription factors which drive the pluripotent state and CDKN1A (p21) regulation as OCT4 can repress CDKN1A (p21) [36]. Overall, while much focus has led to an understanding of the signal transduction pathways involved in pluripotency and differentiation, the link between cell cycle and pluripotency is an unexplored, yet rich, area of investigation.

JMJD5 belongs to the JmjC (Jumonji C) domain containing protein family—many of which have been shown to epigenetically control gene transcription by catalyzing histone demethylation [37-45]. In addition, other JmjC family proteins including JMJD5 have been indicated as hydroxylases showing the diversity of function possible within this family [46-52]. Studies of several JmjC members suggest the involvement of JmjC proteins in embryonic developmental processes. JMJD1A and JMJD2C regulate ESC self-renewal by activating pluripotency-associated genes [53], while JMJD3 counteracts Polycomb activity at key loci, including the Nodal gene [54]. Furthermore, JMJD5 is necessary for proliferation of both MCF7 breast cancer cells and mouse embryonic fibroblasts (MEFs) [33, 55]. Although JMJD5 has not been implicated in pluripotency maintenance, it is necessary for embryonic development as deletion of JMJD5 in mice causes severe growth retardation and embryonic lethality at midgestation, likely due to the gross upregulation of CDKN1A (p21) [33, 56].

In this study, we elucidate the role of JMJD5 in hESCs. We find that JMJD5 is necessary for regulation of the G1 phase as well as for pluripotency maintenance. Furthermore, we find that loss of JMJD5 leads to accumulation of CDKN1A (p21) transcript and protein. We provide evidence that the role of JMJD5 is to repress CDKN1A (p21) by transcriptional and post-transcriptional mechanisms. Furthermore, by generating double JMJD5 and CDKN1A (p21)-depleted hESCs, we find that most, if not all, of JMJD5 function is mediated by CDKN1A (p21). Lastly, we show that the loss of the accelerated G1 precedes the loss of pluripotency.

Materials and Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

Culture and Differentiation hESCs

Undifferentiated hESCs were cultured on irradiated MEF feeders, see details in Supporting Information Experimental Procedures. Definitive endoderm differentiation was induced from hESCs using recombinant human Activin A as previously described [57-60]. For endoderm differentiation and embryoid body differentiation details, see Supporting Information Experimental Procedures.

Synchronization of hESCs and hESC-Derived Endoderm

hESCs and hESC-derived endodermal cells were synchronized at the G1 stage of the cell cycle by aphidicholin (Sigma-Aldrich, Ltd., Dorset, U.K., http://www.sigmaaldrich.com/). For additional details, see Supporting Information Experimental Procedures.

siRNA Design, Lentivirus Production, hESCs Transduction, and Stable Line Generation

JMJD5 siRNA were designed and cloned in the lentiviral vector LentiLox 3.7 [61]. For knockdown of CDKN1A (p21), the inducible lentiviral vectors TRIPZ containing shRNA targeting human CDKN1A and nonsilencing negative control were purchased form Open Biosystems (Open Biosystems, Inc., Huntsville, AL, http://www.openbiosystems.com). For the constitutive expression of JMJD5, full-length human JMJD5 cDNA was cloned from an expression plasmid pLTRE-Flag-KDM8 (Gift kindly provided by Dr. Yoshihiro Izumiya, University of California Davis School of Medicine) into the lentiviral vector pCDH (System Bioscience, Mountain View, CA, http://www.systembio.com, Cat. No. CD520A-1). siRNA design, lentivirus production, hESCs transduction, and stable line generation are all detailed in Supporting Information Experimental Procedures.

Cell Cycle Analysis and Flow Cytometry

The distribution of cells at specific cell cycle stages was evaluated by flow cytometry based on DNA content staining by propidium iodide. Pluripotency marker SSEA4 was analyzed by flow cytometry. Cells were stained with APC (allophycocyanin)-conjugated mouse anti-human SSEA4 antibody (R&D Systems, Inc., Minneapolis, MN, http://www.rndsystems.com). For additional details, see Supporting Information Experimental Procedures.

Analysis of Protein, mRNA, and miRNA

For Western blot, JMJD5 antibody (ab36104) was purchased from Abcam (Cambridge, U.K., http://www.abcam.com). Oct4 antibody (sc-9081), CDKN1A (p21) antibody (sc-397), and GAPDH antibody (sc-32233) were purchased from Santa Cruz (Santa Cruz Biotechnology, Inc., Dallas, TX, http://www.scbt.com). Quantitative RT-PCR (qRT-PCR) was performed following standard methods. Primer sequences are described in Supporting Information Table S1. To analyze miRNA, the total RNA was extracted by miRNeasy Mini Kit (Qiagen, Duesseldorf, Germany, http://www.qiagen.com). For additional details, see Supporting Information Experimental Procedures.

Chromatin Immunoprecipitation Analysis

Chromatin immunoprecipitation (ChIP) experiments were performed using Magna ChIP, a chromatin immunoprecipitation kit (MILLIPORE, Billerica, MA, http://www.millipore.com, Cat. 17–610) according to the manufacture's protocol. Complexes were immunoprecipitated with JMJD5 antibody (ab36104). ChIP-qPCR primers for regions in miR302 promoter are listed in Supporting Information Table S1.

3′SEQ Sequencing and Comparing Gene Expression Profiles Across G1-S Transition

Sequencing libraries were generated by first isolating total RNA using RNeasy Plus Mini Kit (Qiagen, Duesseldorf, Germany, http://www.qiagen.com) and then purifying mRNA using Dynabeads Oligo (dT)25 (Invitrogen, Carlsbad, CA, http://www.lifetechnologies.com) according to the manufacturer's instructions [62]. For 3′SEQ data, high-quality 36 bp sequence tags were aligned against the reference human genome (UCSC—hg18) using Eland (Illumina, INc., San Diego, CA, http://www.illumina.com) software. Read counts were centered by subtracting the experimental mean and dividing by the standard deviation of all reads within each experimental cell cycle phase sample (standard Z score). For additional details, see Supporting Information Experimental Procedures.

Statistical Analysis

Quantitative data are presented as the mean ± SEM. Significant was assessed by the two-tailed Student's t test.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

hESC-Derived Endoderm Accumulates Cells in G1

In order to compare transcripts expressed during the abbreviated G1 phase in hESCs with those expressed in a closely related cell type that has a longer or more standard G1 phase, we first examined whether endodermal cells derived from hESCs had a different cell cycle distribution compared to hESCs. To this end, we differentiated hESCs into definitive endoderm using Activin A in low serum conditions according to the previous studies [57-60]. We found that at 6 days postdifferentiation cells have morphology changes (Supporting Information Fig. S1A) and endoderm markers, such as SOX17 and GATA4, become highly expressed, while pluripotency markers, including OCT4, NANOG, SOX2, and SSEA4, are decreased (Supporting Information Fig. S1B–S1D). Importantly, when we examined the cell cycle profile for both hESCs and derived endoderm using flow cytometry based on DNA content staining by propidium iodide, we found that the cell cycle profile in hESC-derived endodermal cells is reminiscent of other human cells, in that it has a significant G1 phase accumulation (Supporting Information Fig. S1E). This finding is consistent with another recent report showing that hESC-derived endoderm has an extended G1 phase [13]. Overall, contrasting the cell cycle profile of the derived endoderm with that of the hESCs shows a significant increase of G1 phase cell population (52.7% ± 2.3% compared to 29.0% ± 0.4%) and a dramatic decrease of cells in S phase (29.5% ± 1.34% compared to 54.5% ± 1.45%) (Supporting Information Fig. S1E), suggesting that in early differentiated hESCs, the cell cycle machinery has dramatically altered within a short window of development.

Comparison of Synchronized hESC and Derived Endodermal Cell Populations

As hESC-derived endoderm and hESCs have significant differences in the timing of G1 phase, we systematically compared the shortened G1 phase in hESCs with the longer G1 phase in hESC-derived endoderm. To this end, we found that both cell types can be successfully synchronized in G1 phase by treatment with 20 µg/ml aphidicholin for 24 hours. Flow cytometry analysis showed that, prior to release (0 hour), 60.7% of hESCs and 66.1% of hESC-derived endodermal cells are blocked at G1 phase (Supporting Information Fig. S2A, S2B-a, S2B-d). Upon release from inhibition, we collected cells every hour and then determined the cell cycle distribution by flow cytometry. Within 3 hours of release, 75.5% of the hESCs were in S phase (Supporting Information Fig. S2A, S2B-c). By contrast, it took 6 hours for the derived endodermal cells to reach the highest S phase percentage at 51.3% (Supporting Information Fig. S2A, S2B-f). While we cannot exclude the possibility that synchrony is not maintained between the two populations after release, overall, these results suggest that hESCs transit from G1 to S in an accelerated fashion when compared with derived endoderm.

Elucidation of G1/S Transcripts Specific for hESCs

To elucidate molecular differences between the abbreviated G1 (hESCs) and the standard G1 (derived endoderm), we next examined the transcriptome during each cell cycle phase in both cell types. We carried out synchronization experiments in four replicates under identical conditions and prepared samples at G1, G1/S, and S phase for both hESCs and derived endodermal cells (Supporting Information Fig. S2B). We pooled the cell pellets of four replicates together, extracted mRNA and generated 3′ RNA-Seq (3′SEQ) libraries [62] for each cell cycle phase. For the sequencing data, 36 bp sequence tags were aligned against the reference human genome. We then used UNIPEAK [63] to identify expressed transcripts (see Methods for parameters). The 3′SEQ data are available in GEO (accession number GSE47117). We found 10,347 and 10,299 genes expressed in the G1 phase and G1/S phase of hESCs, respectively; 10,333 and 10,227 genes expressed at the G1 and G1/S phase of derived endoderm.

In order to identify genes involved in the rapid G1 phase of hESCs, we first selected genes expressed highly in hESC G1 phase, but not in the derived endoderm G1 phase, and, second, selected genes that fluctuated in expression in derived endoderm between G1 to G1/S phase, but not in hESCs. To this end, we calculated the transcriptional fold change between G1 and G1/S phase in hESCs and derived endoderm for each gene. We next identified 1,408 genes with a fold change greater than 1.5 in the hESC G1 phase as compared to the derived endoderm G1 phase, suggesting these factors might promote a rapid G1 phase specifically in hESC. Subsequently, as it has been shown that many cyclins remain constant in ESCs, but not other cell types [1, 2, 5, 15, 64], we identified genes within the 1,408 subset whose expression fluctuated in derived endoderm but not in hESCs between G1 to G1/S phase. Overall, we identified 51 genes that remain constant in hESCs but fluctuate in derived endoderm between G1 and G1/S phase.

Upon examination of annotated functions of these 51 genes, we identified one particularly interesting candidate, JMJD5, a JmjC (Jumonji C) domain containing protein, which has been suggested as a cell cycle and cell proliferation regulator [55]. We find that JMJD5 has high expression in both hESC G1 and G1/S phases and is expressed at a low level in G1 in derived endoderm, but increases dramatically at the G1/S phase (Supporting Information Fig. S3A). The differential expression pattern of JMJD5 between the two cell cycle types, and the lack of fluctuation during G1 to S phase in hESCs, suggested a distinct role in the rapid G1. Therefore, to examine JMJD5 more closely, we performed qRT-PCR and Western blot for JMJD5 in hESCs and derived endoderm. We found that both transcripts and proteins were consistently high from G1 to S phase in hESCs (Supporting Information Fig. S3B and Fig. 1A). Furthermore, we examined the cellular localization of JMJD5 by immunofluorescence and found that it was mainly nuclear localized in hESCs (Supporting Information Fig. S3C). Importantly, we also found that differentiated cells, including derived endoderm and embryoid bodies (at days 3, 7, and 10), have much lower JMJD5 expression than hESCs (Fig. 1B). Overall, the high expression of JMJD5 during G1 phase in hESCs—but not in the more differentiated endoderm—suggests a role specific for JMJD5 during the hESC G1 phase.

image

Figure 1. JMJD5 is expressed in the hESC G1 phase. (A): Western blot analysis of JMJD5 in the G1 to S phase in hESCs and hESC-derived endodermal cells. (B): qRT-PCR and Western blot analysis of JMJD5 expression in hESCs, endoderm (left), and in embryoid bodies over 10 days (right). Transcript levels were normalized to GAPDH. Data are presented as the mean ± SEM. *, p < .01; **, p < 5 × 10−6 (two-tailed t test relative to hESC). GAPDH was used as protein loading control. Abbreviation: hESC, human embryonic stem cell.

Download figure to PowerPoint

JMJD5 Is Necessary to Maintain Pluripotency in hESCs

To test whether JMJD5 is required for hESC propagation and pluripotency, we depleted its expression using siRNA technology. We found that two oligonucleotides, shJMJD5-A and shJMJD5-B, when introduced using lentiviral technology, were effective in knocking down the expression of JMJD5 at both the mRNA and protein levels in hESCs (Fig. 2A). In order to examine the phenotypic effect of hESCs depleted for JMJD5, we infected cells with lentivirus carrying either shJMJD5-A, B or a control vector (shControl). After 2–3 days, infected cells were transferred to new matrigel plate (passage 1) and then examined morphologically and molecularly after an additional 4–5 days (Supporting Information Fig. S4), which effectively avoided overcrowding. In addition to expressing short hairpin RNA (shJMJD5-A or shJMJD5-B), the lentiviral vector also coexpresses green fluorescent protein (GFP) as a reporter gene. Therefore, in this context, expression of GFP is an indicator of cells that have received either the JMJD5 shRNA or control vector. We found that while control hESC colonies formed well-defined boundaries (Fig. 2B-a, 2B-b), hESCs depleted for JMJD5 became enlarged and flattened (Fig. 2B-c–2B-f) and did not form compact colonies. To examine whether pluripotency was affected in the JMJD5-depleted hESCs, we examined OCT4, NANOG, and SOX2 mRNA levels by qRT-PCR and OCT4 protein level by Western blot in JMJD5-depleted cells and found a decrease over control (Fig. 2C). To further substantiate this finding, we examined the cell surface pluripotency marker SSEA3 by immunofluorescence and found a dramatic decrease in expression within the JMJD5-depleted cells (Fig. 2D). Lastly, we performed flow cytometry with a combination of SSEA4 and GFP and find a decreased population of double GFP and SSEA4-positive cells in JMJD5-A-depleted cells (Fig. 2F; data for shJMJD5-B were similar, data not shown). Upon quantification, we found that hESCs depleted for JMJD5-A have a decreased level of SSEA4-positive cells compared to control cells—64% ±1.5% and 83% ± 2.1%, respectively, indicating the loss of the pluripotency in JMJD5-depleted cells.

image

Figure 2. JMJD5 is essential for human embryonic stem cell (hESC) identity. (A): qRT-PCR and Western blot analysis of JMJD5 expression in control and JMJD5-depleted cells. (B): Depletion of JMJD5 led to loss of distinct colonies and enlarged flattened cell morphology. Infection of hESCs with GFP-marked lentivirus carrying control vector (a and b), shJMJD5-A, and shJMJD5-B (c-f). (a, c, and e) DIC. (b, c, and d) Live cell GFP images. (C): qRT-PCR analysis for OCT4, NANOG, and SOX2 in control and JMJD5-depleted hESCs; Western blot analysis of OCT4 expression in control and JMJD5-depleted cells. (D): Immunofluorescence against SSEA3 (right column), GFP (middle column), and DAPI (left column) in control and JMJD5-depleted hESCs. (E): qRT-PCR analysis with lineage-specific markers in control and JMJD5-depleted hESCs. (F): Flow cytometry analysis of SSEA4 in control and JMJD5-A-depleted hESCs. Cells were stained with primary SSEA4 antibody conjugated with allophycocyanin (x-axis). The y-axis indicates GFP expression and thus the lentivirus infected cell populations. The percentages indicate the cell populations in each of the four quadrants. Similar results were found in JMJD5-B-depleted hESCs (data not shown). For (A, C, and E), the levels of the transcripts were normalized to GAPDH. Data are presented as the mean ± SEM. *, p < .05; **, p < .01; ***, p < .005 (two-tailed t test relative to siControl). GAPDH was used as protein loading control. Scale bars = 50 µm. Abbreviations: DIC, differential interference contrast; GFP, green fluorescent protein.

Download figure to PowerPoint

Concomitant with the decrease in pluripotency markers, we show that markers for multiple lineages, including endoderm (SOX17), mesoderm (BRACHYURY), ectoderm (NESTIN and PAX6), and trophectoderm (HCG) are upregulated in JMJD5-depleted hESCs (Fig. 2E). Most significantly, the ectoderm markers NESTIN and PAX6 increased 5-fold and 13-fold, respectively; and the trophectoderm markers chorionic gonadotropin-a and -b increased 7-fold and 2-fold, respectively. Hence, depletion of JMJD5 results in multiple differentiated cells types, although cells are enriched for ectoderm and trophectoderm lineages. Collectively, the differentiated cell morphology, the reduced expression of pluripotency markers and induced lineage-specific markers strongly indicate that JMJD5 is necessary for the maintenance of pluripotency in hESCs.

JMJD5 Regulates hESC Proliferation

While we demonstrate that JMJD5 is necessary for pluripotency, we were interested to determine how this role interfaced with regulation of proliferation—specifically the G1 phase of the cell cycle. Initially, we observed that the JMJD5-depleted hESC colony numbers were reduced five- to sixfold when compared with controls (Fig. 3A). To quantify this effect, we counted cell numbers every day after passage 1 and find that, by day 4, JMJD5-depleted cells exhibited a three- to fourfold growth decrease compared to control cells (Fig. 3B). Additionally, we examined JMJD5-depleted and control hESCs for BrdU incorporation using flow cytometry. We found approximately 20% fewer double positive GFP+/BrdU+ cells in the JMJD5 depletion compared to controls (Fig. 3C). Taken together, our results indicate the proliferation of hESCs was inhibited by depletion of JMJD5.

image

Figure 3. JMJD5 regulates human embryonic stem cell (hESC) proliferation and cell cycle progression. (A): Colony formation was reduced in JMJD5-depleted hESCs. (B): Cell proliferation was compromised in JMJD5-depleted hESCs. Cells were counted at indicated time points in triplicate at passage 1 after lentivirus infection. (C): BrdU incorporation analyzed by flow cytometry shows decreased BrdU and green fluorescent protein double positive populations in JMJD5-depleted hESCs. Cells were stained with primary BrdU antibody conjugated with allophycocyanin. (D): Flow cytometry analysis of cell cycle profile assessed by propidium iodide staining in control and JMJD5-depleted cells. The quantified analysis of flow cytometry is shown in the far right panel. (E, F): Depletion of JMJD5 retains cells in G1 phase. Control and JMJD5-A-depleted hESCs were synchronized by aphidicholin at G1 phase (0 hour), released, and collected every hour. DNA was examined using flow cytometry after propidium iodide staining. Data plotted by percentages of cells in G1 and S phase over time. Data are presented as the mean ± SEM. *, p < .05; **, p < .01 (two-tailed t test relative to shControl).

Download figure to PowerPoint

JMJD5 Regulates the G1 Timing in hESCs

We next evaluated the effect of JMJD5 depletion on the hESC cell cycle. To this end, we used the same strategy as described above to infect, passage, and age the JMJD5-depleted cells (Supporting Information Fig. S4). Using flow cytometry, we found that hESCs depleted of JMJD5 have a marked increase in the fraction of cells in G1 phase compared to the controls (50.5% ± 2.4% in JMJD5-A-depleted cells and 56.3% ± 2.3% in JMJD5-B-depleted cells compared to 32.1% ± 3.4% in control cells). Furthermore, 25.4% ± 3.1% of JMJD5-A-depleted cells and 33.4% ± 4.1% of JMJD5-B-depleted cells transition into S phase, compared with 47.0% ± 4.2% in controls (Fig. 3D), strongly suggesting that JMJD5 plays a role in the hESC cell cycle.

To examine more precisely the G1 phase in JMJD5-depleted cells, we synchronized JMJD5-depleted and control hESCs to determine the kinetics of the cell cycle. Control and JMJD5-depleted cells were synchronized at G1 with 20 µg/ml aphidicholin for 24 hours. Upon release, the control and JMJD5-depleted cells were analyzed by flow cytometry every hour for 10 hours (Fig. 3E, 3F). At the time of release (0 hour), both control and JMJD5-depleted cells were efficiently synchronized at G1 phase (60%–70% cells). In the control, cells progressed to S phase far more rapidly than the JMJD5-depleted cells: At 3 hours, 60% of the control cells were in S phase, while at 5 hours, 73% were in S phase (Fig. 3E). In contrast, JMJD5-depleted cells showed significant retention in G1 phase: At 10 hours, only 42.5% cells were in S phase and the majority of the cells—46.8%—remained in G1 phase (Fig. 3F). This indicates that JMJD5 plays a role in the timing of the G1 phase in hESCs.

JMJD5 Negatively Regulates CDKN1A (p21) Transcription

To investigate the mechanism by which JMJD5 regulates the hESC cell cycle, we took a candidate approach and examined the expression of known G1-S transition components, including CDK2, CDK4, CYCLIN A1, CYCLIN A2, CYCLIN D1, CYCLIN E1, CDKN1A, and CDKN1B in JMJD5-depleted or control hESCs (Fig. 4A). We found that one of these, CDKN1A, was significantly upregulated compared to controls (11–14-fold; Fig. 4A). Other candidates, including CYCLINA1, were not affected, even though JMJD5 has been shown to directly regulate CYCLINA1 in MCF7 breast cancer cells [55]. Upregulation of CDKN1A is in agreement with a recent report showing an increase of Cdkn1a transcripts (4.17-fold) in jmjd5Δ/Δ MEFs cells and a robust increase of Cdkn1a transcripts in JMJD5 knockout E8.5 embryos [33, 56], suggesting that this interaction is conserved between species and cell types. These observations are consistent with a mechanism by which JMJD5 acts as a repressor of CDKN1A transcription during the G1-S transition in hESCs.

image

Figure 4. JMJD5 represses CDKN1A (p21). (A): qRT-PCR analysis of known G1/S regulators in control and JMJD5-depleted hESCs. (B): qRT-PCR analysis of JMJD5 and CDKN1A in control and JMJD5 overexpressing hESCs. (C): Western blot analysis of CDKN1A (p21) protein expression in control and JMJD5-depleted hESCs. GAPDH was used as protein loading control. (D): qRT-PCR analysis of miR-302 family expression with primers specific to the miR-302 transcript a, b, c, and d in control and JMJD5-depleted cells. The levels of the transcripts were normalized to U6 snoRNA. (E): Upper panel: schematic of the miR-302 promoter cloned upstream of the firefly luciferase gene. Lower panel: 293T cells and JMJD5 overexpressing 293T cells (293T-JMJD5) were cotransfected with pGL3-Basic, pGL3-mirR-302–525 bp, or pGL3-mirR-302–974 bp and a Renilla reporter plasmid pRL-SV40. Luciferase activity was measured 24 hours after transfection. Luciferase activity was normalized against Renilla activity and expressed as promoter activity relative to the empty vector (pGL3-Basic) in 293T cells. (F): Upper panel: schematic of where primers were designed over the miR-302 promoter. Lower panel: ChIP-qPCR using seven sets of primers in mirR-302 promoter with anti-JMJD5 antibody in control and JMJD5-A-depleted hESCs. Fold enrichment is normalized to IgG control. For (A and B), the levels of the transcripts were normalized to GAPDH. Data are presented as the mean ± SEM. *, p < .05; **, p < .005; ***, p < .001; ****, p < 1 × 10−4 (two-tailed t test). Scale bars = 50 µm. Abbreviation: hESC, human embryonic stem cell.

Download figure to PowerPoint

If JMJD5 is acting as a negative regulator of CDKN1A at the transcriptional level, then overexpressing JMJD5 in hESCs should be sufficient to decrease CDKN1A transcription. To test this hypothesis, we generated a hESC stable line that constitutively expresses human JMJD5 approximately 200-fold above control levels (Fig. 4B; see Methods for construction and propagation). Using this line, we examined the expression of CDKN1A by qRT-PCR. We found that CDKN1A expression decreased (more than 50%) in JMJD5 overexpressing hESCs when compared with the stable control hESC line (Fig. 4B). Significantly CDKN1A was the only G1-S transition component that changed expression—even CDKN1B did not change (data not shown). Overall, this supports a mechanism by which JMJD5 negatively regulates CDKN1A transcription in hESCs.

JMJD5 Depletion Leads to Accumulation of CDKN1A (p21) Protein

While CDKN1A transcripts are normally present in hESCs, the protein has not been detected, suggesting that it is not transcription per se that is critical for CDKN1A (p21) activity, but also post-transcriptional control [16, 17, 31]. Thus, in order for CDKN1A (p21) to play a key role in mediating pluripotency and G1 timing, the accumulated transcripts would clearly need to be translated. Therefore, we examined JMJD5-depleted and control cells for induction of CDKN1A (p21) protein using a Western blot approach and found a significant induction of the CDKN1A (p21) protein in the JMJD5-depleted hESCs (Fig. 4C).

JMJD5 Depletion Leads to Decreased miR-302 Expression

We next explored how JMJD5 might control the post-transcriptional regulation of CDKN1A (p21). As it was shown recently that CDKN1A (p21) is inhibited post-transcriptionally by the microRNA family miR-302 [31], we examined whether JMJD5 may regulate CDKN1A (p21) in conjunction with miR-302. To this end, we first extracted total RNA, including the small RNA fraction, in both the JMJD5-depleted and control hESCs. We then analyzed the expression of the miR-302 cluster by qRT-PCR with primers specific to miR-302-a, -b, -c, and -d. Interestingly, we found the expression of all four miRNAs of miR-302 cluster decreased significantly in JMJD5-depleted cells compared to control knockdown cells (Fig. 4D).

To determine whether JMJD5 directly activates the miR-302 promoter, we performed both luciferase assays and ChIP-qPCR. To this end, we used constructs that contained either a 0.5 kb or 1 kb fragment of the miR-302 promoter upstream of the luciferase gene (pGL3-mirR-302–525 bp and pGL3-mirR-302–974 bp) [65]. These constructs were introduced into either 293T cells or 293T stable line cells which constitutively overexpress human JMJD5 approximately 900-fold above control levels (data not shown). We observed a 1.7-fold increases in luciferase activity in the presence of JMJD5 overexpression (Fig. 4E), showing that the mir-302 promoter is sensitive to JMJD5 levels. Next, we performed ChIP-PCR using an anti-JMJD5 antibody to test whether JMJD5 could bind the miR-302 locus. We designed primer sets spanning the 1 kb promoter region (Fig. 4F; upper panel) and demonstrate that JMJD5 broadly associates with this region. This association is lost in JMJD5-depleted hESCs, supporting the idea that JMJD5 is recruited directly to the miR-302 promoter (Fig. 4F).

Generation of hESCs Depleted for Both JMJD5 and CDKN1A (p21)

While mechanistically JMJD5 serves to regulate CDKN1A (p21) at both the transcriptional and post-transcriptional levels, we next tested whether this interaction had any functional consequences. Therefore, to test functional interactions between JMJD5 and CDKN1A (p21), we performed epistasis analysis to determine whether depleting CDKN1A (p21) would rescue the JMJD5 depletion phenotype. To this end, we generated a double knockdown of JMJD5 and CDKN1A (p21) in hESCs. We first tested a TET-on lentiviral system containing shRNA for CDKN1A or control sequence in a CDKN1A (p21) overexpressing 293T stable line cells and found that two separate shRNA sequences could efficiently knockdown CDKN1A to 41.4% ± 4.6% and 38.5% ± 5.3% control levels, respectively (Supporting Information Fig. S5A). Once verifying the TET-on constructs, we generated two stable hESCs with doxycycline inducible expression—one with CDKN1A shRNA and the other with control shRNA. These hESC lines were effectively propagated and expressed shRNA and red fluorescent protein (RFP) in the presence of doxycycline. We verified that induced CDKN1A shRNA could effectively knockdown endogenous CDKN1A transcripts to 10% ± 3.1% in the presence of doxycycline (Supporting Information Fig. S5A).

To generate double depleted CDKN1A/JMJD5 hESCs, we treated the CDKN1A shRNA and control shRNA stable cells with doxycycline for 2 consecutive days. We then infected each of the two cell lines with the lentivirus carrying either the shJMJD5-A or control vector. Using the same protocol discussed above (Supporting Information Fig. S4), we transferred cells 2–3 days postinfection to matrigel (passage 1) and examined after 4–5 days. Doxycycline was present throughout in order to maintain CDKN1A shRNA or control shRNA expression. In the resulting double depleted, single depleted, or double control cells, RFP is a reporter for the control shRNA and the CDKN1A shRNA. GFP is a reporter for the shControl and the shJMJD5-A. Using qRT-PCR, we examined whether CDKN1A or JMJD5 was depleted when compared with controls. Interestingly, we demonstrate that CDKN1A is effectively inhibited in the CDKN1A shRNA/shJMJD5-A hESCs, but not in the Control shRNA/shJMJD5-A hESCs (Supporting Information Fig. S5B), validating the double depletion system.

JMJD5 Mediates the Pluripotent State by Repressing CDKN1A (p21)

We tested whether depletion of CDKN1A (p21) could rescue the loss of pluripotency found in the JMJD5-depleted hESCs. Thus, we examined the pluripotent state of hESCs depleted for both JMJD5 and CDKN1A (p21). Significantly, we observed that the morphology of colonies expressing both RFP and GFP within the CDKN1A shRNA/shJMJD5-A hESC resembled control hESCs, forming compact hESC colonies. This is in contrast to the Control shRNA/shJMJD5-A hESCs where expressing cells were flattened with no clear colony boundaries, similar to the phenotype when depleting JMJD5 alone (Fig. 5A). Similarly, there is no morphological defect in the colonies expressing both RFP and GFP within the CDKN1A shRNA/shControl hESCs, suggesting CDKN1A (p21) depletion alone does not affect hESC morphology (Supporting Information Fig. S5C). To more rigorously characterize the extent of the rescue, we used three channel flow cytometry detecting GFP, RFP, and SSEA4-APC (allophycocyanin, conjugated SSEA4 antibody). We found that the CDKN1A shRNA/shJMJD5-A hESCs (GFP+/RFP+ population) have a more extensive SSEA4-positive population compared to the Control shRNA/shJMJD5-A hESCs (GFP+/RFP+ population; 77.8% compared to 68.5%, green line compared to blue line in right panel, Fig. 5B). When we analyzed the SSEA4 contribution to cells within the same cultures, we found that—in the CDKN1A shRNA/shJMJD5-A hESCs—the percentage of SSEA4-positive cells is higher in GFP+/RFP+ cells (double depleted CDKN1A/JMJD5 cells) compared to GFP+/RFP cells (single JMJD5-depleted cells) (77.8% compared to 60.7%, green line of right panel compared green line of left panel, Fig. 5B). In contrast, we found that the control hESCs (Control shRNA/shControl) have far higher contributions of SSEA4-positive cells (96.7% and 98.7%, red line in right and left panel, Fig. 5B). We also found the CDKN1A shRNA/shControl hESCs have similar percentages of SSEA4-positive cells in GFP+/RFP+ cells (single CDKN1A-depleted cells) compared to GFP+/RFP cells (non CDKN1A-depleted cells) (95.7% compared to 97.6%, blue line compared to red line, Supporting Information Fig. S5D), suggesting CDKN1A depletion alone does not affect pluripotency of hESCs. Overall, these data indicate that depletion of CDKN1A (p21), at least partially, rescues the pluripotency defect in the JMJD5-depleted hESCs and demonstrates that an epistatic interaction of these two molecules is important in maintaining the pluripotent state.

image

Figure 5. CDKN1A (p21) and JMJD5 double depletion rescue the pluripotent state and rapid G1 phase. (A): CDKN1A (p21) and JMJD5 double depletion rescue the JMJD5 morphological phenotype in human embryonic stem cells (hESCs). Top panel: hESCs expressing both control shRNA (RFP) and JMJD5 shRNA (GFP). Bottom panel: hESCs expressing both CDKN1A shRNA (RFP) and JMJD5 shRNA (GFP). DIC image is at far left. (B): Three channel flow cytometry detecting GFP, RFP, and APC-SSEA4 (allophycocyanin, conjugated SSEA4 antibody) to examine the pluripotent state in JMJD5 and CDKN1A (p21) double depleted hESCs. Cells were sorted into GFP+/RFP+ and GFP+/RFP− cell populations (middle panel). GFP+/RFP+ cell populations were analyzed for SSEA4 expression (right panel). 96.7% of Control shRNA/shControl hESCs (red line); 68.5% of Control shRNA/shJMJD5-A hESCs (blue line); and 77.8% of CDKN1A shRNA/shJMJD5-A hESCs (green line) express SSEA4. GFP+/RFP− cell populations were also analyzed for SSEA4 expression (left panel). 98.7% of Control shRNA/shControl (red line); 65.4% of Control shRNA/shJMJD5-A hESCs (blue line); and 60.7% of CDKN1A shRNA/shJMJD5-A hESCs (green line) express SSEA4. IgG controls for each condition are marked in purple, cyan, and orange, but cluster in peak at far left and thus colors are indistinguishable. (C): Flow cytometry analysis of cell cycle phases assessed by propidium iodide staining in Control shRNA/shControl, Control shRNA/shJMJD5-A, and CDKN1A shRNA/shJMJD5-A hESCs in the presence of doxycycline. (D): Cell population numbers and colony numbers examined in Control shRNA/shControl, Control shRNA/shJMJD5-A, and CDKN1A shRNA/shJMJD5-A hESCs in the presence of doxycycline. Data are presented as the mean ± SEM. *, p < .05 (two-tailed t test). Scale bars = 50 µm. Abbreviations: DIC, differential interference contrast; GFP, green fluorescent protein; RFP, red fluorescent protein.

Download figure to PowerPoint

JMJD5 Mediates the Cell Cycle in hESCs by Repressing CDKN1A (p21)

We next examined whether repression of CDKN1A (p21) within hESCs depleted for JMJD5 would rescue the cell cycle defect. To this end, we used flow cytometry to detect cells at different cell cycle phases within the CDKN1A shRNA/shJMJD5-A hESCs or controls. In this analysis, we found that the inhibition of CDKN1A (p21) within the JMJD5-depleted cells led to a partial rescue of all cell cycle phenotypes, including the accumulation of cells in G1 phase, the decrease of cells in S phase, decreased proliferation, and decreased colony formation (yellow bar vs. red bar, Fig. 5C, 5D). These phenotypes were indeed observed in the Control shRNA/shJMJD5-A hESCs, consistent with earlier observations (red bar vs. blue bar, Fig. 5C, 5D). Furthermore, we show that CDKN1A depletion alone does not change G1 and S phase populations in the cell cycle (Supporting Information Fig. S5E). Overall, this demonstrates that JMJD5 is a key repressor of CDKN1A (p21) in hESCs and that the interaction between these two molecules is necessary for both maintenance of pluripotency and the cell cycle.

Cell Cycle Changes Precede Loss of Pluripotency in Depleted JMJD5 hESCs

As depletion of JMJD5 causes both the accumulation of cells in the G1 phase and the loss of pluripotency in hESCs, we next examined which phenotype occurred first. In other words, we sought to dissect whether this lengthened G1 phase simply occurs as the consequence of the loss of pluripotency. To this end, we performed a detailed time course of cellular progression during the depletion of JMJD5 in hESCs, during which we analyzed both cell cycle profiles and loss of pluripotency. We infected hESCs with the lentivirus carrying either the shJMJD5-A or control vector. We then took time points postinfection at days 1, 2, 3, 4, 6, and 7. At each time point, we used flow cytometry and qRT-PCR to examine the pluripotent state with the markers SSEA4, OCT4, NANOG, and SOX2 (Fig. 6 and Supporting Information Fig. S6). At day 3 after lentivirus infection, the percentage of JMJD5-depleted cells in the G1 phase significantly increased to 40.8% ± 1.3% compared to 26.9% ± 1.3% in the controls (Fig. 6A). In these same cells, the percentage of JMJD5-depleted cells in the S phase decreased to 36.9% ± 2.3% compared to 45.7% ± 1.39% in controls (Fig. 6B). We found no changes in SSEA4, OCT4, NANOG, or SOX2 levels within the JMJD5-depleted cells at this time point (Fig. 6C and Supporting Information Fig. S6). Significantly, at day 6 after infection, while cells in G1 continued to accumulate in the JMJD5-depleted cells, we observed the first evidence of decreased SSEA4, OCT4, NANOG, and SOX2 expression and by day 7 all markers were significantly decreased (Fig. 6 and Supporting Information Fig. S6). Importantly, we also found that CDKN1A was significantly induced at day 3 postinfection in the JMJD5-depleted cells, but not before, tracking well with the beginning of G1 accumulation. The correlation between CDKN1A (p21) expression and G1 accumulation also strongly suggests that the JMJD5/CDKN1A (p21) interaction is specific for the cell cycle and not a result of the loss of pluripotency and differentiation into a distinct cell type (Fig. 6D). While there is heterogeneity within these populations, there is a clear overall trend toward initial cell cycle loss followed by a decrease in pluripotency markers. Therefore, we suggest that JMJD5 regulates rapid G1 phase in hESCs by selectively activating CDKN1A (p21) and that this accumulation of G1 population precedes the loss of pluripotency.

image

Figure 6. Cell cycle defect precedes loss of pluripotency in JMJD5-depleted human embryonic stem cells (hESCs). (A, B): Flow cytometry analysis of cell cycle profile assessed by propidium iodide staining in control and JMJD5-A-depleted hESCs depicted as the percentage of cells in G1 (A) or S (B) phase at days 1–7 postinfection. (C): Flow cytometry analysis SSEA4 showing relative percentages of SSEA4+/GFP-positive cells and qRT-PCR for OCT4 in control and JMJD5-A-depleted hESCs. (D): qRT-PCR for JMJD5 and CDKN1A in control and JMJD5-A-depleted hESCs in days after lentiviral infection. The levels of the transcripts were normalized to GAPDH. Data are presented as the mean ± SEM. *, p < .05; **, p < .01; ***, p < .005; ****, p < 5 × 10−4 (two-tailed t test).

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

The maintenance of low levels of CDKN1A (p21) is a central link to the connection between the abbreviated ESC cycle state and the maintenance of pluripotency. CDKN1A (p21) has been shown to be necessary and sufficient for regulating the length of the cell cycle in both ESCs and during the reprogramming of fibroblasts [3, 25, 30]. Furthermore, these studies have demonstrated a fundamental role of CDKN1A (p21) in maintaining pluripotency in ESCs and driving reprogramming from fibroblasts—strongly suggesting that the regulation of protein levels of CDKN1A (p21) is central to both processes. In this study, we demonstrate that depletion of JMJD5 leads to the specific upregulation of CDKN1A (p21) transcripts and proteins in hESCs. We show that the inappropriate upregulation of CDKN1A (p21) inhibits the cell cycle and pluripotency in hESCs. Furthermore, we show that these phenotypes can both be rescued by knocking down CDKN1A (p21) within the JMJD5-depleted hESC background, strongly suggesting that these two proteins are functionally working together to drive these important processes.

While our data suggest that JMJD5 may regulate CDKN1A transcription, there is mounting evidence that post-transcriptional control of CDKN1A (p21) is a major mode of regulation [30, 31, 66]. Mechanistically, the lack of CDKN1A (p21) protein in hESCs was shown to be due to the direct binding of miR-302 miRNAs to CDKN1A (p21) mRNA and subsequent inhibition of translation [31]. Importantly, previous studies demonstrated that the miR-302 cluster is both essential for ESC cell cycle and promotes somatic cell reprogramming [30, 65, 67-71]. In this study, our data show that JMJD5 can regulate the miR-302 cluster expression and, in turn, modulates levels of the CDKN1A (p21) protein. Overall, our data implicate JMJD5 in the post-transcriptional regulation of CDKN1A (p21) by regulating the miR-302 cluster. This suggests that it may serve a dual function both to repress CDKN1A (p21) transcription and activate miR-302, the latter of which then regulates translation of the CDKN1A (p21) protein.

Although much correlative evidence implicates the accelerated G1 phase with maintenance of pluripotency in hESCs, including the observation that fibroblasts regain the accelerated G1 upon reprogramming [11, 12, 72], it remains a question to what extent the accelerated G1 plays a causative role in reprogramming and maintenance of pluripotency [6, 14]. Recently, it has been shown that G1 begins to lengthen soon after differentiation (12–16 hours) and that endodermal markers accumulate shortly thereafter, suggesting the two events are linked [13]. It has been also suggested that a longer G1 window might allow the accumulation of differentiation signals [5, 6, 14]. Importantly, recent studies have shown that, in human and mouse stem cells, the G1 phase cells have increased propensity to differentiation [73-75]. In contrast, a recent study has shown that in mESCs, elongation of the G1 phase fails to induce differentiation [76], suggesting that G1 timing is not critical for pluripotency. Whether this new paradigm is true for hESCs has yet to be determined. While this is still an open question, our experiments suggest that the loss of the accelerated G1 precedes the loss of pluripotency, suggesting a causative role in hESCs. Using a time course approach, we depleted JMJD5 and find that the first phenotypic effect is found at 3 days post knockdown when both CDKN1A expression is increased and the G1 phase is significantly accumulated. The second phenotypic effect—that of the loss of pluripotency—is not observed until 3 days later (6 days postdepletion) when we finally observe a decrease in the pluripotency marker SSEA4, OCT4, NANOG, and SOX2. In the context of JMJD5, this suggests that the initial defect occurs to the cell cycle and that the change in the cycling time subsequently may drive the loss of pluripotency. While this may be unique to the JMJD5 scenario, it lends support to the idea that the accelerated G1 phase is a fundamental regulator of the pluripotent state in hESCs and also supports the idea that efficient reprogramming may benefit from methods that reliably and rapidly accelerate the G1 phase in fibroblasts.

In this report, we examine transcripts on a genome wide level between the synchronized cell cycles of hESCs and endoderm and found many that were differentially regulated between the two cell types. Here, we only report on one of these genes, that of JMJD5. However, we provide both the extensive datasets of the G1, G/S, and S phase in both hESCs and endoderm which should serve as a valuable resource for examining more thoroughly gene expression in hESCs and derived endoderm. Similar to Medina et al. who recently elucidated the epigenetic control of cell cycle-dependent histone gene expression by microarray analysis at G1 to S phase between hESCs and normal diploid cells [77], this resource will serve as a useful tool for future studies.

Conclusions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

In this article, we propose that JMJD5 is playing a central role in defining the accelerated cell cycle and pluripotency in hESCs. We further suggest that JMJD5 mechanistically regulates the rapid hESC G1 phase by repressing CDKN1A transcription and by activating the miRNA-302 family which subsequently leads to inhibition of CDKN1A (p21) translation. Overall—by its role in repressing CDKN1A (p21)—we suggest that JMJD5 is a key factor in the cellular and molecular mechanisms underpinning the rapid cell cycle and pluripotency in hESCs.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

This work was supported by the California Institute of Regenerative Medicine, Grant RL1–00100. We thank Ziming Weng, Phil Lacroute, and Arend Sidow for performing high throughput sequencing, Dr. Edward Chuong for processing the 3′SEQ data and mapping to the human genome, Dr. Hua Tang and Dr. Hong Gao for valuable suggestions for data analysis, Dr. Christine Reid for critical reading of the manuscript, Dr. Yoshihiro Izumiya for kindly providing construct containing full-length human JMJD5 cDNA, and Dr. Alicia Barroso-delJesus for kindly providing constructs containing miR-302 promoter.

Author Contributions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

H.Z.: conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing, and final approval of manuscript; S.J.H.: collection and/or assembly of data and final approval of manuscript; J.B.: conception and design, financial support, administrative support, data analysis and interpretation, manuscript writing, and final approval of manuscript.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and Methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgments
  9. Author Contributions
  10. Disclosure of Potential Conflicts of Interest
  11. References
  12. Supporting Information

Additional Supporting Information may be found in the online version of this article.

FilenameFormatSizeDescription
stem1724-sup-0001-Suppinfo.docx42KSupporting Information
stem1724-sup-0002-Suppfig1.eps3009KSupporting Information Figure 1
stem1724-sup-0003-Suppfig2.eps1489KSupporting Information Figure 2
stem1724-sup-0004-Suppfig3.eps29933KSupporting Information Figure 3
stem1724-sup-0005-Suppfig4.eps1121KSupporting Information Figure 4
stem1724-sup-0006-Suppfig5.eps7684KSupporting Information Figure 5
stem1724-sup-0007-Suppfig6.eps815KSupporting Information Figure 6
stem1724-sup-0008-Supptbl1.docx23KSupporting Information Table 1

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.