Predicting stem cell fate changes by differential cell cycle progression patterns

Cell cycle progression and lineage commitment of stem cells appear to be tightly linked processes (Lange and Calegari, 2010; Orford and Scadden, 2008; Singh and Dalton, 2009). In particular, the early G1 phase of the cell cycle plays a crucial role in the differentiation of stem cells, as it represents a window of responsiveness to extrinsic cues (Blomen and Boonstra, 2007). Progression from G1 to S phase in most somatic cells requires the phosphorylation of the retinoblastoma protein (Rb) by the mitogen-dependent activity of cyclins, namely the CDK4-cyclin D and the CDK2-cyclin E complex (Harbour et al., 1999; Blomen and Boonstra, 2007). In marked contrast, Rb is hyperphosphorylated in mouse embryonic stem cells (ESCs) due to constitutive CDK2-cyclin E activity (Savatier et al., 1994; Stead et al., 2002). As a consequence, the fast cell cycle of self-renewing ESC bypasses commitment signals such as prolonged MAPK activation (Burdon et al., 1999; Burdon et al., 2002). ESCs are characterised by a very short G1 phase that makes up only about 15-20% of a cell cycle that is rather short per se (10 to 16 hours). During differentiation, G1 is markedly increased, as was shown, for example, in primate (Fluckiger et al., 2006) and human ESCs (Filipczyk et al., 2007), as well as in mouse and human induced pluripotent stem (iPS) cells (Ohtsuka and Dalton, 2008).

A similar lengthening of the cell cycle during differentiation has been observed for some somatic stem cells, including NSCs. During brain development, NSCs in the ventricular zone switch from self-renewing divisions, which expand the stem cell pool, to differentiating divisions, which give rise to neurons and committed basal progenitors (Götz and Huttner, 2005; Kriegstein and Alvarez-Buylla, 2009). These fate changes are associated with distinct kinetics in cell cycle progression: the length of G1 increases from 8 hours at the onset of neurogenesis (at E11) to 18 hours at the end (E16) (Takahashi et al., 1995). Furthermore, G1 lengthens specifically during acquisition of a neurogenic fate (Calegari et al., 2005) and during transition from apical progenitors to basal progenitors that are already committed to neuron production (Arai et al., 2011). In addition, numerous cell cycle regulators have been shown to affect stem cell fate, including cyclins and CDK inhibitors (Doetsch et al., 2002b; Kowalczyk et al., 2004; Kippin et al., 2005; Beukelaers et al., 2011). Prolongation of G1 by chemical inhibition of CDK2-cyclin E (Calegari and Huttner, 2003) or by RNAi-mediated silencing of CDK4-cyclin D (Lange et al., 2009) resulted in increased neurogenesis. By contrast, overexpression of CDK4-cyclinD led to an expansion of the progenitor pool, both during brain development (Lange et al., 2009) and in the adult hippocampus (Artegiani et al., 2011).

These and other data have led to the intriguing hypothesis that G1 lengthening is not only a consequence but can be a cause of differentiation (Götz and Huttner, 2005; Salomoni and Calegari, 2010). However, because cell cycle dynamics during stem cell fate changes have not yet been directly measured in live single stem cells, it has thus far not been possible to elucidate this bi-directional relationship in greater detail. We reasoned that a single-cell and high-throughput analysis of dividing stem cells that undergo well-defined cell fate changes could shed light on the link between cell cycle progression and fate changes. To analyse dynamic changes in cell cycle phases, we used stem cells that express FUCCI reporters (Sakaue-Sawano et al., 2008) whose expression we automatically and precisely detected over multiple rounds of cell division by our newly developed cell tracking software termed FUCCIJ. In order to record continuously the proliferation of individual cells, we employed a previously developed microwell array technology that allows trapping of single stem cells in confined areas for long-term cell culture and exposing them to niche cues of interest (Gobaa et al., 2011; Roccio et al., 2012).

Using this combined approach, we measured a total increase in adult NSC cycle length upon induction of differentiation that was mostly caused by a doubling of G1 length. Conversely, we found that treatment of NSCs with a cell-permeable CDK4 inhibitor induced an increase in the percentage of cells in G1 and promoted differentiation. The lengthening of G1 of committed cells resulted in progressively increasing red fluorescence intensity of the G1 FUCCI reporter. We took advantage of this intensity change to separate stem cells, based on their lower red fluorescence, from more committed cells in a heterogeneous cell population using FACS. In marked contrast, owing to their very short G1, ESCs were found to be mainly marked by the green S/G2-M FUCCI reporter. We again exploited this specific change in cell cycle pattern to substantially augment the proportion of reprogramming cells by sorting green cells early on during reprogramming from a NSC to an induced pluripotent stem cell-like state. Taken together, this approach may help to overcome the heterogeneity of stem/progenitor populations that hampers many population-level in vitro studies.

C57BL/6N and DBA/2N mixed heterozygous mice expressing either Cdt1-KO2 (B6.B6D2-Tg(FUCCI)596Bsi) or Gem-AG (B6;B6D2-Tg(FUCCI)504Bsi) were obtained by the Riken Institute (Japan). Animals were crossed to obtain double positive FUCCI mice.

FUCCI mice were sacrificed by decapitation after a short isoflurane anaesthesia. Subventricular zone and whole hippocampi were dissected on ice. Four 8-week-old mice were pooled for one extraction. The two neurogenic regions were cultured separately. Cells were cultured as neurospheres as described previously (Giachino et al., 2009). Hes5::GFP cells were isolated from early postnatal brains as described previously (Giachino et al., 2009). All cells were used between P2 and P5.

Time-lapse microscopy of single FUCCI cells captured in microwells was conducted as described previously (Gobaa et al., 2011; Roccio et al., 2012). Brightfield, red and green fluorescence were recorded. S/G2-M phase was measured as the time between first and last frame with green fluorescence. G1 was measured as the time between the first frame without fluorescence (after cell division) and the last frame with red fluorescence. This includes a short interval of S phase in which the cells were both green and red.

NSC differentiation on plastic dishes was performed on Laminin/Poly-L Ornithine (PLO)-coated plates in conditioned medium (from 3-day-old cell cultures). Two days later, medium was exchanged with DMEM-F12 supplemented with B27 and 2 ng/ml EGF and 2 ng/ml FGF. Cells were left in culture for 4, 6 or 10 days. Alternatively, cells were stimulated for 48 hours with 2% FBS and 1 μM retinoic acid and analysed 4 days after plating. Mouse ESC differentiation was induced by removal of LIF and addition of retinoic acid (2.5 μM).

The cell permeable cyclin-dependent kinase 4 (CDK4) inhibitor 2-bromo-12,13-dihydro-5H-indolo[2,3-a]pyrrolocarbazole-5,7(6H)-dione (Zhu et al., 2003) was added for 24 hours at a concentration of 1 μM to 3-day-old neurospheres that were dissociated and analysed by flow cytometry or re-plated on laminin-PLO-coated plates for differentiation assays. Alternatively, dissociated cells plated on laminin-PLO-coated plastic were treated with 1 μM CDK4i for 3 days.

Mouse E14 ESC were cultured in standard condition on 0.2% gelatin-coated plates in DMEM supplemented with 15% ES-screened serum (Fisher) and 1000 U/ml LIF (Millipore). Cells were infected with FUCCI lentiviral vectors (Riken Institute, Japan) and 3 days post-infection sorted by FACS to single cells using 96-well plates. Colonies expressing green fluorescence originating from cloned red cells (and vice versa) were selected as double-transduced clones. Nine clones were characterised by flow cytometry and clone #9 was further characterised by immunostaining and time-lapse imaging.

Hes5::GFP NSC (p2) were dissociated to single cells plated in a 12-well plate (1×10⁵ cells) in suspension. Cells were transduced with the Cdt1-KO2 vector, neurosphere-cultured for 4 days and transduced KO2⁺ cells sorted, expanded and characterised by flow cytometry and immunostaining.

NSC reprogramming was adapted from a published protocol (Kim et al., 2009). Lentiviruses were generated using the following plasmids: TetO-FUW-OSKM (Addgene plasmid 20321 containing Oct3/4, Sox2, Klf4 and c-Myc; for simplicity termed ‘4F’) and with FUW-M2rtTA (Addgene plasmid 20342 containing a transactivator). Transduction efficiencies were estimated by Oct4 staining 48 hours after doxycycline (dox: 2 μg/ml) induction. To induce reprogramming, 1×10⁵ FUCCI-NSC (4F line) were seeded in NSC medium in 12-well plates and grown as neurospheres. Dox was added to the medium at day 0. Three days after dox induction, neurospheres were dissociated and plated in PLL-coated plates in ESC medium in presence of LIF and dox. Colony formation was estimated at day 7 and 10. Alternatively, 3×10⁵ cells were seeded on PLL-coated 10 cm dishes and induced with dox for 3 days. Three days after induction, medium was changed to ESC medium containing LIF and dox. Cells were sorted 4 days after dox induction. 1000 cells were seeded per well in 96-well plates to establish colony-forming efficiency. 10,000 cells were seeded on microwell arrays for time-lapse imaging.

Paired t-tests were used to evaluate the statistical significance between two groups. One way-ANOVA with a Bonferroni post-test for multiple comparisons was used when more then two groups were compared. Differences in segregation in synchronous and asynchronous cell divisions were evaluated using a Pearson’s Chi square test.

Dynamic analysis of cell cycle progression by live-cell imaging has become feasible with the generation of the FUCCI reporter system (Sakaue-Sawano et al., 2008). To automatically and precisely detect cells based on the fluorescence of the two FUCCI reporters and to reconstruct genealogical trees of dividing cells, we developed a novel cell-tracking software as an ImageJ plug-in package (termed FUCCIJ) that allows extracting cell cycle phase lengths, fluorescence intensities, cell positions and finally entire lineage trees of single cells (Fig. 1; supplementary material Fig. S1; Movies 1, 2). FUCCIJ is delivered as two independent ImageJ plugins termed ‘Cell Lineage’ and ‘Tree Lineage’. After a preprocessing and filtering step automatic cell detection is performed. User-based editing allows correction of false-positive or false-negative detections. Lineage tree generation is based on the Hungarian algorithm as well as on custom-made combinatorial and probabilistic models in which theoretical expectations are combined with the results of detection.

To analyse stem cell cycle dynamics at single-cell level and with high temporal resolution and throughput, we trapped cells in hydrogel microwell arrays for time-lapse microscopy (Gobaa et al., 2011; Roccio et al., 2012) (supplementary material Fig. S1A). Individual microwells can be functionalized with desired biomolecules in order to expose cells to components of their native micro-environment (e.g. laminin1).

As a first model system to try to unveil the relationship between cell cycle progression and fate choice-making, we isolated adult neural stem and progenitor cells (for simplicity herein termed ‘NSCs’) from the subventricular zone (SVZ) and dentate gyrus (DG) of 8-week-old FUCCI mice. Under normal physiological conditions in vivo, we detected cycling (i.e. green; S/G2-M) cells in the SVZ and DG by confocal analysis of brain sections (supplementary material Fig. S2A). Lower numbers of green cells were found in DGs, indicating fewer cycling cells (Abrous et al., 2005). We isolated NSCs based on their neurosphere-forming capacity in medium containing EGF and FGFb from both neurogenic areas, but focused on SVZ-derived neurosphere-forming cells for subsequent in vitro characterisation.

Immunostaining revealed that 80-100% of the obtained cells were positive for the stem and progenitor cell markers Nestin (98.5±0.7%) and Sox2 (84±8%) (Fig. 2A). Flow cytometry analyses showed that the population contained 48.7±1.5% Cdt1-KO2-positive (KO2⁺) cells (red), 19.8±4.1% Gem-AG-positive (AG⁺) (green), 2.5±1.2% double-positive and 28.2±5.3% double-negative cells (Fig. 2B,C). These proportions were comparable for two independent SVZ isolations. Furthermore, no difference in cell cycle phase distribution was observed between cells growing as neurospheres compared with adherent culture conditions. DG-derived cells showed similar proportions of the four subpopulations (supplementary material Fig. S2B); however, in this case we detected a higher percentage of G1 when cells were cultured as spheres. To quantify the percentage of cells in mitosis (M), we immunostained cells for the mitotic marker phospho-S10-H3 and found that ∼5% of the cells stained positive (supplementary material Fig. S2C).

DNA staining using Hoechst confirmed that Cdt1-KO2⁺ cells represent a population of cells in G1 (2n cells), whereas Gem-AG⁺ cells contain twice the amount of DNA (4n cells) and intermediate amounts and are therefore in S/G2-M (Fig. 2B). Double-positive cells with intermediate DNA amounts mark cells at the onset of S phase. The double-negative population represented primarily 2n cells with a smaller percentage of 4n-cells (∼10%). FACS of the double-negative population and flow cytometry analysis after 4 days in culture, showed that 90% of the double-negative population consisted of cells that regained fluorescence in culture (supplementary material Fig. S3). This was also confirmed by time-lapse microscopy where ∼90% of the non-fluorescent cells reacquired red fluorescence within 2 to 3 hours of cytokinesis and regained green fluorescence upon cell cycle progression, whereas the remaining 10% of initially double-negative cells expressed only the red fluorescent protein. Therefore, the majority of double-negative cells are early G1 cells.

Imaging of single cells over 3 days allowed recording of approximately three consecutive cell divisions. Clone formation was very efficient, as 71±10% of all single cells proliferated extensively (Fig. 2D), indicating no phototoxicity despite prolonged fluorescence image acquisition. A small percentage of cells (2.9±2.3%) never divided during the 3-day experiment, and the remaining cells (23±7%) died at different time points during acquisition. Using FUCCIJ, we quantified the cell cycle characteristics of these cells at single cell level. We observed a remarkably broad distribution of the times taken by mother NSC to enter the first division (supplementary material Fig. S4A). This was primarily dependent on a similarly broadly distributed G1, whereas S/G2-M was relatively homogeneous (Fig. 2E-G). The average duration of G1 and S/G2-M phase was 12±7 and 6.6±1.6 hours, respectively. Notably, we observed a shortening of the cell cycle and G1 in consecutive divisions (supplementary material Fig. S4B). Overall, this resulted in G1 representing 66% of the total cell cycle, whereas S/G2-M made up 34% (Fig. 2H), matching the ratios determined by flow cytometry above.

To elucidate whether cell fate changes are linked to distinct changes in cell cycle progression, we compared G1 and S/G2-M in FUCCI-NSC cultured for 4 days under self-renewal conditions or differentiation-initiating conditions induced by growth factor withdrawal (Fig. 3). An end-point analysis showed a 1.3-fold increase in Cdt1-KO2⁺ cells (red), and a concomitant ∼60% decrease in the percentage of Gem-AG⁺ cells (green) in adherent cultures (Fig. 3A) and neurospheres (data not shown) upon growth factor withdrawal. Immunostaining of the mitotic marker phospho-S10-H3 also confirmed the significantly lower numbers of mitotic cells under differentiation-initiating conditions (1.1±0.5%) compared with self-renewal (5.7±2%) (supplementary material Fig. S5). Furthermore, Nestin expression was markedly reduced and we detected a significant increase in the expression of astrocytic (marked by GFAP) and neuronal fates (βIII-tubulin). After 10 days of differentiation, neurons made up 11.8±4.6%, oligodendrocytes 0.7±0.7% and astrocytes 38.4±14.7% of the population (supplementary material Fig. S6). In all cases, terminally differentiated cells displayed red fluorescent nuclei as expected.

To analyse the dynamics of single cell cycle progression during differentiation initiation, we continuously filmed cells for 3 days. Growth factor withdrawal caused a marked reduction in cell proliferation, as well as an increase in cell death (Fig. 3B). We also detected an increase in the percentage of non-dividing cells (4-13%), suggesting that these cells terminally differentiated. For 60% of the dividing cells, only one division occurred within 3 days and it was not possible to evaluate the duration of G1, as red fluorescence was maintained till the end of the experiment. Similar results were obtained in a 4-day time-lapse experiment (data not shown). It is therefore likely that these cells have exited the cycle and are on their way to terminal differentiation. In agreement with this, 85% of all cells cultured in the presence of growth factors were positive for Ki67, whereas after 4 days in low growth factors conditions, the percentage of Ki67-positive cells was reduced to 45% (data not shown). For 40% of cells that divided a second time within 3 days, we observed a significant increase in G1 from 12±7 to 21±14 hours (Fig. 3C,D). Concomitantly, we also observed a minor but significant increase in the duration of S/G2-M from 6.6±1.7 to 7.7±2.2 hours. Therefore, differentiation is accompanied by cell cycle lengthening, which may precede exit from the cell cycle.

To extract information on the synchrony of cell divisions, we analysed the difference in the length of G1 of the first generation of sister cells (Δt=G1a–G1b) (Fig. 3E). We defined divisions as asynchronous when the length of G1 of the paired daughters differed by more than 30%. Interestingly, the percentage of asynchronous divisions more than doubled when cells were moved from self-renewal to differentiation-inducing conditions (36.6 versus 76.9%) (Fig. 3F). The overall ratio of synchronous versus asynchronous divisions did not vary with increasing cell cycle length, whereas the percentage of synchronous divisions (Δt=0) dropped with increased cell cycle duration. Twenty three percent of cells with an average G1 shorter than 19 hours (average + 1 s.d.) showed this behaviour, whereas only 9.6% divided synchronously in cells with extended G1 (G1>19 hours) (Fig. 3G).

The marked increase in G1 length observed upon initiation of differentiation led us to design an inversed experiment to understand whether forced lengthening of G1 by chemical inhibition of CDK4 could influence NSC differentiation. In neurospheres or adherent cells treated with different concentrations of a cell-permeable CDK4 inhibitor (CDK4i), we detected an increase in the proportion of KO2⁺ cells (red) at the concentration of 1 μM (supplementary material Fig. S7A). Using a standard 10-day differentiation protocol, inhibition of CDK4 resulted in an approximate twofold increase in the number of neurons (supplementary material Fig. S8B, Fig. S7E), and a ∼30% decrease in the percentage of GFAP-positive cells (supplementary material Fig. S8B-D). However, we also measured ∼50% reduction (P<0.001) in the total number of cells in samples treated with CDK4i (supplementary material Fig. S8A), likely caused by cytotoxic and cytostatic effects exerted by this inhibitor (Sakaue-Sawano et al., 2011; Zhu et al., 2003). Therefore, in order to exclude effects related to differences in cell numbers (e.g. paracrine signals, differential cell-cell contacts or GF consumption), we assessed cultures after just 4 days of CDK4i-treatment where cultures were not confluent and cell numbers were comparable. Indeed, we detected a significant increase in KO2⁺ cells from 60±1.1 to 67±0.3%, as well as an increased fluorescence intensity compared with untreated cells (Fig. 4A). Furthermore, immunostaining showed an increase in the expression of βIII-tubulin in CDK4i-treated cells (Fig. 4B). As the large majority of these cells still expressed Nestin at this early time point, we believe that these cells are neuroblasts rather than neurons.

Next, we used the FUCCI system in combination with FACS to synchronize the population by sorting G1 cells and induce differentiation, either by growth factor withdrawal or by the addition of serum and retinoic acid (RA), specifically during G1 in KO2⁺ cells (Fig. 4C). As expected, we observed a more pronounced differentiation induction by addition of serum/RA compared with low growth factor conditions. CDK4i pretreatment induced a twofold increase in βIII-tubulin (Fig. 4D) and GFAP (Fig. 4E) expression in low growth factor conditions, and a 1.7-fold induction in presence of serum/RA compared with untreated cells, as shown by flow cytometric analysis. Interestingly, the percentage of KO2⁺ cells correlated with the percentage of differentiated cells in an exponential fashion, both in the case of GFAP expression and βIII-tubulin (Fig. 4F,G). These data show that the prolongation of G1 by chemical inhibition of CDK4 results in increased differentiation.

Owing to the fact that fluorescent protein expression in the FUCCI system is driven by a constitutively active CAG promoter and solely regulated by expression of the respective E3 ligases (APC^Cdh1 and SCF^Skp2 complexes), we reasoned that prolonged cell cycle phases would result in continuously increasing fluorescence intensity and that differences in fluorescence intensities might be used to distinguish different stem cell fates. Using FUCCIJ, we were able to precisely extract fluorescence intensity values of each cell at any time point and thus could directly test this. Indeed, cells that never divided constantly accumulated the red fluorescent protein (Fig. 5A). By marked contrast, dividing cells reached a lower maximum fluorescence intensity that linearly correlated with G1 (Fig. 5B), suggesting that differentiating cells have a distinctively higher expression of the red fluorescent protein. In addition, flow cytometry data showed that not only the percentage of Cdt1-KO2⁺ increases upon differentiation, but also the intensity of red fluorescence (Fig. 5C).

To corroborate these observations, we analysed by flow cytometry a heterogeneous population of partially differentiated cells. We found that cells expressing the neuronal marker βIII-tubulin belonged to the cell population expressing higher levels of the KO2 reporter (‘high-red’ population) (Fig. 5D,F). Conversely, Nestin- and EGFR-positive cells were mainly found in the population comprising lower red fluorescence (‘low-red’) (Fig. 5E,F). Similar results were obtained for differentiation induced by CDK4i treatment, both in the absence of growth factors or the presence of serum/RA (supplementary material Fig. S9). In all conditions examined, we observed on average a twofold enrichment in differentiated cells in the high-red compared with the low-red cells (supplementary material Fig. S9A,B), and an approximately linear correlation between the percentage of differentiated cells and the intensity of the red fluorescent protein marker (supplementary material Fig. S9C). Importantly, the same was also true in normally proliferating cells, where by qRT-PCR we identified in the high-red population a lower expression of stem cell markers and higher expression levels of differentiation markers (supplementary material Fig. S10). High-red cells also displayed an increased cell size compared with low-red cells. This could be explained either by an increase in cell size of morphologically differentiated cells (neurons, astrocytes and oligodendrocytes) or by an increase in cell size during the length of the cell cycle, as S/G2-M cells are larger than G1 cells (supplementary material Fig. S11A,B).

Heterogeneous stem cell populations despite identical growth conditions are a problem in many stem cell areas and distinct markers are often lacking that separate stem cells from already committed cells. To test whether differences in red fluorescence could be further exploited to enrich stem cells in heterogeneous cell populations, we first sorted cells by FACS based on the 20% highest and 20% lowest red fluorescence intensities, and re-plated them under differentiation conditions in low growth factor medium for 3 days. Directly after sorting, Nestin-positive cells were found to be ∼1.6-fold more abundant in the low-red population that also comprised 30% fewer βIII-tubulin-expressing cells (not shown). Furthermore, the number of cells recovered from the high-red population after 3 days was more than three times lower (P<0.001) (supplementary material Fig. S11C), showing that cells with higher fluorescence have a reduced proliferation rate in accordance with their committed state. Nevertheless, upon differentiation for 5 days, both cell populations were able to give rise to neurons and astrocytes to a comparable extent (supplementary material Fig. S11D).

To verify if fast-cycling (i.e. low red) cells indeed represent a population of neural stem cells, we took advantage of NSC isolated from Hes5::GFP transgenic mice (Basak and Taylor, 2007; Lugert et al., 2010). In this model, GFP expression reports Notch activity and was used to identify stem cells both during embryonic development and in the adult mouse brain. In addition, we have previously shown that neurosphere generation on laminin 1-functionalized hydrogel substrates is higher in cells expressing high levels of GFP (Roccio et al., 2012). Here, we examined the cell cycle progression of Hes5::GFP cells in greater detail.

Low GFP expressing cells showed a significant reduction in proliferation, as 91±8% of these cells were in G1/G0, in contrast to 75±2% of cells expressing high GFP levels (Fig. 5G). Hes5::GFP cells expressing the red FUCCI reporter construct Cdt1-KO2 showed similar cell cycle profiles to the parental cells with ∼75% of the cells being Cdt1-KO2 positive (Fig. 5H) and are tripotent, as confirmed by immunostaining (Fig. 5I). The remaining 10% of G1 cells did not express the reporter, suggesting they are early G1 cells. Under self-renewal conditions, where the majority of cells retained GFP expression (80%), we detected GFP-negative cells exclusively in cells expressing the highest levels of the red reporter (Fig. 5J,K). By contrast, high GFP cells were enriched in the S/G2-M population (Fig. 5L). These data show that it is indeed possible to further subdivide NSCs grown under identical conditions based on their cell cycle progression, with the fast-cycling cells in heterogeneous NSC cultures comprising stem or progenitor cells, and cells with increased G1 length being already committed.

An even more heterogeneous and rare stem cell fate change occurs during the process of reprogramming of somatic cells into induced pluripotent stem (iPS) cells, a process that is well known to be accompanied by dramatic changes in cell cycle progression (Smith et al., 2010). We wondered whether it would be possible to augment reprogramming by selecting cells with a specific cell cycle pattern.

We first analysed cell cycle progression patterns of FUCCI mouse embryonic stem cells (FUCCI-ESC) that we generated by lentiviral transduction and clonal expansion of wild-type E14 cells (Fig. 6; supplementary material Fig. S12). Under self-renewal conditions in the presence of LIF, FUCCI-ESC maintained expression of alkaline phosphatase, Nanog (supplementary material Fig. S12A,B) and Oct4 (not shown). FUCCI-ESC colonies displayed very high levels of green fluorescence (Fig. 6A-C) and when examined by flow cytometry, only 6±2.9% of the cells expressed the red G1 reporter. 54.9±1.7% of the population were Gem-AG positive, whereas 35.7±3% were double negative. Based on DNA profile, BrdU and phospho-S10-H3 staining (supplementary material Fig. S12C-E), we confirmed that red cells are predominantly in G1 (88%) and green cells in S/G2-M (84%). The blank population represented 65% of cells in G1 and 35% in S/G2-M. We believe that this is caused by rapid cycling, not allowing accumulation of the FUCCI constructs.

We next monitored by time-lapse microscopy changes in cell cycle progression during ESC commitment induced by LIF removal (Fig. 6D,E; supplementary material Fig. S12F-H). Under self-renewal conditions, FUCCI-ESC cycle very fast through G1 (2.2±1.5 hours), whereas upon differentiation induction, the length of G1 was found to double (4.5±2.6 hours, P<0.0001) within only the first two cell cycles (Fig. 6E). This striking change in ESC cycle progression patterns was indeed promising for using the FUCCI reporter system to increase the efficiency of reprogramming.

We first characterised the reprogramming of FUCCI-NSC generated by transduction of doxycyclin (dox)-inducible c-Myc, Klf4, Sox2 and Oct4 (for simplicity termed ‘4F’) using a polycistronic lentiviral vector (Carey et al., 2009; Kim et al., 2009; Kim et al., 2008) (supplementary material Fig. S13A,B). Transduction efficiency, calculated by flow cytometry from staining of Oct4 expression 48 hours after dox induction, was 27.5±8% (supplementary material Fig. S13C). Concomitantly, we detected an upregulation of c-Myc, Klf4, Sox2 and Oct4 at the mRNA level (supplementary material Fig. S13D).

Seven days after dox induction, colonies appeared that showed ESC-like morphology and displayed expression of the green FUCCI reporter that is indicative of highly proliferative clones (Fig. 6F). Efficiency was estimated as 0.9±0.7% with 4.7±5% of green cells at day 7 and 17.4±1.6% at day 10. Flow cytometric analysis revealed a dramatic decrease in Nestin expression from 71±2.5% to 10.5±0.2%, and an increase of Oct4 expression from 0.05±0.07% to 51.8±4.8% (Fig. 6G). Red cells were positive for the astrocytic marker GFAP (supplementary material Fig. S13E), whereas green and double-negative cells were positive for Oct4 (Fig. 6H; supplementary material Fig. S13E). In agreement with the above observation that green cells mark fast proliferating clones, these cells showed a reduced cell volume compared with red cells in G1. In the presence of mouse ESC medium containing FBS, only the reprogrammed clones continue to proliferate, whereas non-transduced and not reprogrammed cells differentiate and accumulate the red fluorescent reporter.

We took advantage of the underlying changes in cell cycle profiles and FUCCI reporter expression to sort cells by FACS early on during reprogramming. To this end, red and green cells were sorted 4 days after induction with dox (Fig. 6I, see scheme in supplementary material Fig. S13F). Strikingly, colony formation was detected only in the green-sorted population (Fig. 6J) with a colony-forming efficiency of ∼2.4±0.7%. Early selection of fast proliferating cells allowed us to enrich the number of reprogramming cells by at least 2.5-fold, an enrichment that enabled the continuous imaging of the reprogramming process at single cell level in microwell arrays (Fig. 6K). The majority of the proliferating green-sorted cells underwent nevertheless only one cell division (supplementary material Fig. S13F, upper panel). By contrast, red-sorted cells rarely divided during the time-lapse experiment (supplementary material Fig. S13F, lower panel). Ten days after sorting, Oct4-positive colonies were identified (Fig. 6L) and a cell cycle phase distribution of the FUCCI reporters similar to the one of ESC (Fig. 6C), with 17±3.8% red, 40±10% green and 54.6±11% blank cells, was measured (Fig. 6M). Finally, we examined the expression levels of the four reprogramming factors by qRT-PCR after 6 days from red- or green-sorted cells and observed a relative increase in expression by sixfold for c-Myc, fivefold for Sox2, 20-fold for Sox2 and 10-fold in the case of Oct4 (Fig. 6N). Colonies were further clonally expanded in the absence of dox for 15 days. The cell cycle profile, quantified by flow cytometry via Hoechst labelling, and the distribution of the FUCCI reporters was undistinguishable from the one of FUCCI-ESC (supplementary material Fig. S13G). These data show that the expression of the green FUCCI reporter, which appears in a rare subpopulation of cells early after induction of reprogramming, allows the enrichment of a cell population with highly proliferative capacity and expression of pluripotency factors.

Therefore, by mapping G1 and S/G2-M phases of single NSC and ESC undergoing self-renewal and differentiation, distinct cell cycle progression patterns can be identified that are indicative of different developmental stages of these cells (Fig. 7). The knowledge on the link between cell cycle progression and cell fate choice-making can be used, as shown here for example, to enrich for stem cells in heterogeneous populations or to isolate rare cells that undergo reprogramming. These findings may also help to elucidate the relationship between cell cycle progression, differentiation and reprogramming in other cellular systems.

By taking advantage of a powerful combination of high-throughput single cell culture on an established microwell array system (Cordey et al., 2008; Gobaa et al., 2011; Roccio et al., 2012), the FUCCI reporter system (Sakaue-Sawano et al., 2008) and a newly developed single cell tracking tool (FUCCIJ), we have systematically addressed how cell cycle parameters change during self-renewal, commitment and reprogramming.

Continuous imaging of NSC isolated from FUCCI reporter mice revealed a marked heterogeneity in cell cycle duration, mostly owing to large variations in the length of G1. We also detected a significant increase in total cell cycle length and the length of G1 upon differentiation. Previous studies had already highlighted a link between G1 lengthening and differentiation during embryonic brain development. To study the link between cell cycle progression and cell fate in adult NSC, we treated FUCCI-NSC with a CDK4 inhibitor and detected an increased fraction of cells expressing βIII-tubulin and GFAP 4 days after treatment, as well as an enhancement in FBS/RA-induced differentiation. Our data are thus in agreement with the view that prolonging G1 length increases the likelihood of commitment.

The standard in vitro culture conditions used here are different from the in vivo growth conditions of NSCs in the SVZ, where stem cells are mostly quiescent (Doetsch, 2003), whereas they become highly proliferative in vitro in the presence of abundant growth factors. Nevertheless, the capacity of cells to extensively self-renew and differentiate into three lineages are widely used criteria to define a NSC state (Conti and Cattaneo, 2010; Pastrana et al., 2011). These culture conditions are known to favour the expansion of both activated stem cells as well as transit-amplifying cells (TAPs) expressing the EGF receptor (Doetsch et al., 2002a; Pastrana et al., 2009). Our in vitro data show that the cell cycle kinetics determined by FUCCI (Fig. 7) match quite well with what was previously observed in vivo, where a cycle of about 18-24 hours was measured for activated stem cells and TAPs (Cameron and McKay, 2001; Doetsch et al., 1999; Hayes and Nowakowski, 2002). Under self-renewing conditions, the fast cycling cells, which correspond to the undifferentiated cells in the population, have a total average cell cycle of ∼20 hours and a G1 below 15 hours.

Recent studies showed that the in vitro culture of freshly isolated SVZ-derived NSC in the absence of growth factors quite faithfully mimics the lineage differentiation pattern found in vivo (Costa et al., 2011). Slowly dividing astrocytes, representing the in vivo quiescent stem cell, would then give rise to faster proliferating astro/radial glia, then to TAPs and finally differentiate to neuroblasts. The cell cycle parameter of the fast dividing astro/radial glia cells with a cell cycle length of ∼15 hours and their highly synchronous behaviour is in line with our findings. Therefore, we believe that whereas in the absence of growth factors this expansion phase is only temporally limited [five rounds of divisions were identified by Costa et al. (Costa et al., 2011)] and followed by terminal differentiation, in presence of growth factors this may be expanded through self-renewing divisions.

Single cell analysis and tracking of FUCCI-NSC allowed us to identify cell cycle parameters that correlate with the extent of differentiation. We used the accumulation of Cdt1-KO2 to distinguish between slow and fast cycling cells based on red fluorescence intensities. By analysing expression levels of several marker genes, we showed that the slow cycling cells express higher levels of differentiation markers, whereas fast cycling cells represent a population of more primitive cells. This was further confirmed by generating a Hes5::GFP-Cdt1-KO2 reporter line. In this system, the loss of GFP expression, which marks loss of stemness, was specifically associated with increased levels of red fluorescence, whereas high GFP-expressing cells showed higher percentage of S/G2-M cells compared with low GFP-expressing cells. This is in agreement with previous findings demonstrating higher clone-forming efficiency in the high GFP-expressing cells (Basak and Taylor, 2007; Roccio et al., 2012).

Fast cell cycle progression has previously been proposed as one hallmark of self-renewing mouse ESC that transit very quick through G1, presumably to avoid commitment signals and thus maintain a pluripotent state (Orford and Scadden, 2008; Singh and Dalton, 2009). Consistent with this notion, in mouse ESC transduced with the FUCCI reporter system, we measured average G1 lengths of only 2 hours with a cell cycle duration of ∼14 hours. Conversely, G1 extended very rapidly upon differentiation.

Based on the cell cycle analysis of FUCCI-NSC and FUCCI-ESC, we hypothesised that early changes in cell cycle profile during the reprogramming of NSC to iPS cells could be identified using our platform. Previous work had shown that primary colonies originating from fibroblast reprogramming were derived from cells that were smaller in size and cycled faster than non-reprogramming neighbouring cells (Smith et al., 2010). Indeed, upon overexpression of the four reprogramming factors c-Myc, Klf4, Sox2 and Oct4, we detected the appearance of proliferating cells based on the expression of the green Gem-AG FUCCI reporter. These cells represent a minor fraction (4% at day 4) of the total cell population that would be difficult to identify by other means such as by conventional marker expression at this very early time point. By sorting green cells, we were able to further enrich the population of reprogramming cells. Although this was crucial to enable imaging at a single-cell level and to track proliferating colonies reminiscent of ESC colonies, further analyses are needed to prove that these clones will fully reprogram and to dissect how cell cycle parameters change during this transition.

NSC reprogramming has been previously shown to be more efficient than for mouse fibroblasts, based on the fact that these cells may represent an intermediate stage between fibroblasts and iPS cells (Kim et al., 2008). However, the selection of fast proliferating cells may also improve the efficiency for other cell types. Acceleration of cell cycle progression through inhibition of the p53/p21 pathway or ectopic expression of Lin28 increased the speed of conversion into iPS cells (e.g. Hanna et al., 2009). Moreover, besides the well-established role of c-Myc in promoting cell cycle progression (Singh and Dalton, 2009), a direct relationship between regulators of pluripotency and cell cycle has been described (Card et al., 2008; Lee et al., 2010; Zhang et al., 2009). Our findings therefore confirm previous observations suggesting that the cell cycle is a key parameter in iPS cell generation. Accelerated cell division could increase the probability of iPS cell generation and/or cell cycle-specific events such as DNA replication might be the prerequisite for permitting the epigenetic changes to occur. The analysis of reprogramming events using the FUCCI cell cycle reporter may help to elucidate this issue.

In conclusion, the systematic analysis of changes in cell cycle parameters during proliferative and differentiating cell divisions allowed us to identify traits that are indicative of the stem cell and more committed cell state, and that can be exploited to improve differentiation and reprogramming conditions. In combination with genetic tools and other reporter gene systems, our approaches should provide new avenues to address pertinent questions in stem cell biology.

We thank Mukul Girotra for help with animal breeding and genotyping; Jalil Zerdani and Raphaelle Luisier for initial work on the FUCCIJ plug-in; Celine Ruegsegger for help with the characterisation of FUCCI-NSC; Francois Gorostidi for providing the lentiviral vectors; members of the Flow Cytometry Core Facility of EPFL for support with sorting experiments; Sebastian Jessberger for experimental tools; Verdon Taylor for providing Hes5::GFP cells; and Rudolf Jaenisch for providing the polycistronic vectors for iPS cell generation.