Time-Lapse Imaging Enables Pancreas Progenitor Tracking in Three Dimensions

To study how individual pancreatic progenitors contribute to pancreas expansion and to monitor their differentiation into endocrine progenitors, we conducted live imaging of explants of dorsal pancreatic buds from E12.5 Pdx1^tTA/+;tetO-H2B-GFP embryos (Fig. 1A). The buds were dissected and laid on a fibronectin-coated coverslip bottom plate, where they grew (Fig. 1B) [25,26]. After 24 h of settling time, we initiated high-magnification time-lapse live imaging in 3-D with 6-min intervals for up to 24 h. Nuclear H2B-GFP fusion protein enabled us to observe cell divisions and to track individual cell nuclei. At the end of the experiment, the explants were fixed and immunostained for markers of pancreatic progenitors (SOX9) and endocrine progenitors (NEUROG3) (S1F–I Fig.), while preserving the native green fluorescent protein (GFP) signal (S1G Fig.), which enabled us to match to the cells from the last frame of the time-lapse movies. The SOX9⁺ cells constituted the majority of GFP⁺ epithelial cells (S1I Fig.), and NEUROG3⁺ cells were mainly observed in the middle trunk region of explant (S1H Fig.) [8]. In spite of the constant exposure to laser, explants grew, showed active cell movements, apico-basal polarization, branching, acini morphogenesis, and differentiation similarly to explants that were not subjected to imaging (S1A–E,J,N,O Fig. and S1 and S2 Movies). After 18–24 h of time-lapse imaging (42–48 h post-dissection), NEUROG3⁺ cells were detected by immunostaining, showing that the differentiation process occurred ex vivo, albeit less efficiently than in vivo (S1 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Live imaging reveals both asymmetric and symmetric emergence of NEUROG3 cells.

(A) Scheme summarizing the genetic strategy to visualize PDX1⁺ pancreatic progenitors for live imaging. (B) Scheme of imaging and analysis. Pancreatic explants from E12.5 Pdx1^tTA/+;tetO-H2B-GFP embryos are cultured, and 3-D time-lapse imaging is done for 18–24 h. Then, the explants are immunostained for markers, and endocrine progenitor (NEUROG3) cells are back- and forward-tracked. (C) Model of pancreatic progenitor divisions. A PDX1⁺ progenitor can produce two PDX1⁺/SOX9⁺ progenitor daughters, two NEUROG3⁺ endocrine progenitor daughters, or one PDX1⁺/SOX9⁺ daughter and one NEUROG3⁺ daughter. (D) Still images of live imaging in 3-D maximum intensity projection from S3 Movie, illustrating a symmetric (P/P) division producing two progenitor daughters (blue spots). White nuclei correspond to H2B-GFP signal in the cells originating from the pancreas epithelium. (E) Still images from S3 Movie, illustrating an asymmetric (P/N) division producing two daughters with different fates (red spots). (F-I) Images of fixed explant with native GFP and nuclear staining DRAQ5 overlay (F) and immunostained for NEUROG3 (G) and SOX9/aPKC (H). Blue spots correspond to cells in (D) and red spots to cells in (E). Inset in (H) shows high magnification image of SOX9 staining. Note both blue spotted cells are SOX9⁺, but only one red spotted cell is SOX9^low (H), and the other red spotted cell is NEUROG3⁺ (G). (J) Still images from S4 Movie, demonstrating a symmetric (N/N) division producing two daughters with the same fate (pink spots). (K–N) Images of fixed explant with native GFP (K) and immunostained for NEUROG3 (L) and SOX9/aPKC (M). Pink spots correspond to cells in (J), and both are NEUROG3⁺/SOX9⁻. Inset in (L) shows NEUROG3 staining (four NEUROG3⁺ cells in a row) in high magnification. (O) Analysis of progenitor division patterns from live imaging. Total cell divisions are counted from four cropped positions from four live imaging movies, and fraction of NEUROG3-producing cell divisions is calculated from the corresponding positions (yellow bar over grey bar). NEUROG3-producing divisions (pink, blue, and purple bars) are counted from entire position of four movies. (P) Analysis of NEUROG3 emergence from four live imaging movies. 18.4% ± 5.0% cells emerge through P/N divisions, and 29.8% ± 14.2% through N/N divisions. 29.3% ± 5.9% do not exhibit prior division, and 22.4% ± 10.6% were either lost (17.5% ± 7.7%) or dead (8.9% ± 3.4%). Cells lost or gone out of frame were categorized as indeterminable (purple bar). Numbering denotes elapsed time in h:min, and in the cell division diagrams P indicates progenitor and N, NEUROG3 (D,E,J). Scale bars, 20 μm. Histograms and error bars represent the mean and standard deviation (n = 4). See S2 Table for further data.

https://doi.org/10.1371/journal.pbio.1002111.g001

Pancreatic Progenitors Divide in Three Different Modes That Differentially Contribute to Pancreas Expansion and Endocrine Differentiation

To determine how single progenitor cells contribute to balancing global pancreas expansion with endocrine progenitor generation (Fig. 1C), we systematically back-tracked NEUROG3⁺ endocrine progenitors in 3-D, as well as a random subset of SOX9⁺ pancreatic progenitors that were identified from immunostaining images and mapped onto the last frame of time-lapse movies. Once a cell division was observed through back-tracking, the sister cell was then forward-tracked, and its fate was determined by referring to the immunostaining. The tracking revealed that pancreatic progenitors divided in one of three different modes. The first type of division was symmetric, giving rise to two SOX9⁺ progenitor cells (P/P division; S3 Movie and Fig. 1D,F–I). The second type was asymmetric, giving rise to a SOX9⁺/NEUROG3⁻ pancreatic progenitor and a NEUROG3⁺ endocrine progenitor (P/N division; S3 Movie and Fig. 1E,F–I). The last type was symmetric endocrinogenic, producing two NEUROG3⁺ cells (N/N division; S4 Movie and Fig. 1J–N). In order to quantitatively account for each division mode, we analysed 1,628 divisions comprising all observed division events of Pdx1^tTA/+;tetO-H2B-GFP cells from randomly selected positions from four time-lapse movies. Thus, non-NEUROG3-producing divisions include both bi-potent progenitors and acinar cells, since Pdx1⁺ cells are multipotent at E13.5. This quantification revealed 6.6% ± 1.6% of divisions producing endocrine progenitors, and 93.4% ± 1.6% generating either progenitors or exocrine cells (Fig. 1O and S2 Table). Of all the divisions producing NEUROG3 cells that could be tracked, 56.3% ± 13.8% produced a SOX9⁺ cell and a NEUROG3⁺ cell (P/N division), and 43.7% ± 13.8% produced two NEUROG3⁺ cells (N/N division; Fig. 1O). We could determine the origin of approximately half of NEUROG3⁺ cells through P/N or N/N divisions in the past 24 h, while some NEUROG3⁺ cells either did not exhibit prior division or were either lost or dead during tracking (Fig. 1P). Cell death might be a consequence of the explant culture since apoptosis is very rare in the pancreas epithelium in vivo [27]. Taken together, these results show that at E13.5–14.5 most progenitors undergo symmetric renewing divisions, accounting for pancreas size increase, while the remaining progenitors are approximately evenly split into those undergoing symmetric endocrinogenic division and asymmetric division.

Clonal Analysis Confirms the In Vivo Existence of Three Progenitor Division Types Based on Their Progeny

While ex vivo imaging enables constant monitoring of cell behaviours, it is performed in a somewhat artificial context. In order to determine whether pancreas progenitors undergo the same pattern of symmetric and asymmetric divisions in an in vivo context, we devised a clonal lineage tracing strategy (Fig. 2A) using Hnf1bCreER mice. Previously, this line was used to demonstrate that the E13.5 HNF1B⁺ progenitor cells give rise to ductal and endocrine cells [3]. This could be accounted for either by individual cells giving rise to endocrine and ductal cells or by heterogeneity among HNF1B⁺ cells, some giving rise to endocrine cells and others to ductal cells. To investigate this question, we subjected pregnant mice carrying E13.5 Hnf1bCreER;mT/mG embryos to a single low-dose intraperitoneal injection of 4-hydroxytamoxifen (Fig. 2B) to label pancreatic progenitors at a clonal density. We optimized conditions for clonal tracking leading to two-cell clones at E14.5 (Fig. 2B–G, S3 Table, and S5 Movie). Since we know from the time-lapse experiments that the majority of daughters from the same cell do not move more than 30 μm apart, we called labelled cells within 30 μm of each other a clone (S8 Fig.). Reiterations with a 60 μm radius led to similar outcomes. Whole-mount immunostaining of 22 pancreata and detection of 244 two-cell clones revealed that the majority of progenitors in which the fate could be determined divided symmetrically (P/P; Fig. 2E) into two SOX9⁺ progenitors (59.8%; Fig. 2H). This proportion is lower than the 93.4% found in the explants, in part because the cells traced by HNF1B are a subgroup of PDX1⁺ and SOX9⁺ cells traced in the explants and some of the latter will give rise to acinar cells [2,4]. In vitro lineage tracing with Hnf1bCreER;mT/mG explants showed that 6.3% of clones became endocrine (S2 Fig. and S4 Table). This shows that the in vivo differentiation process is more efficient than in vitro differentiation. After the 24-h tracing period, we could not yet observe any INSULIN⁺ clones in vivo, suggesting NEUROG3⁻ or SOX9⁻ clones might be in transition to endocrine differentiation. Of the NEUROG3-producing two-cell clones in which the fate of both daughters could be determined, 61.8% were asymmetric, generating one NEUROG3⁺ daughter and a SOX9⁺ progenitor (P/N; Fig. 2F,H), and the remaining were symmetric with two NEUROG3⁺ daughters (N/N; Fig. 2G,H). As a consequence, more NEUROG3⁺ cells originated from symmetric divisions (Fig. 2I). These results thus provide in vivo evidence of asymmetric and symmetric endocrinogenic progenitor divisions, as well as of symmetric renewal of progenitors, confirming the modes of divisions detected by the explant tracking data.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Asymmetric and symmetric divisions revealed by in vivo lineage tracing of progenitors at clonal density.

(A) Scheme summarizing the genetic strategy to label HNF1B⁺ pancreatic progenitors with membrane-localized GFP reporter (mG) for lineage tracing. Upon CRE recombination, membrane-localized Tomato (mT) is excised, allowing mG expression. (B) Schematic overview of the lineage tracing strategy used to trace the fate of progeny from single progenitor cells labelled at clonal density. E13.5 pregnant mice carrying Hnf1bCreER;mT/mG embryos received a single intraperitoneal injection of 0.175 mg 4-OH tamoxifen. After 24 h, pancreata were subjected to whole-mount immunostaining, imaging, and 3-D reconstruction to detect recombined two-cell clones. (C) Model of two-cell clone lineage tracing. A HNF1B⁺ progenitor can produce two SOX9⁺ progenitor clones, two NEUROG3⁺ endocrine progenitor clones, or one PDX1⁺/SOX9⁺ and one NEUROG3⁺ clones. (D) Maximum intensity projection of 3-D reconstructed E14.5 dorsal pancreas after immunostaining for E-CADHERIN, SOX9, and GFP. Arrowheads indicate clones displaying recombination of the mT/mG reporter, detected by anti-GFP immunostaining, while membrane Tomato signal was diminished during staining process. (E) Optical sections from a whole-mount imaged dorsal pancreas demonstrating symmetric generation of SOX9⁺ progeny (P/P) from a single dividing progenitor cell. (F) Optical sections demonstrating clonal progeny with asymmetric fates, generating one NEUROG3⁺ and one SOX9⁺ daughter (P/N). (G) Optical sections demonstrating clonal progeny with symmetric NEUROG3⁺ fates (N/N). (H) Quantification of two-cell clone fate patterns after in vivo lineage tracing. 244 two-cell clones derived from 22 dorsal pancreata were scored according to SOX9 and NEUROG3 status. Indeterminable refers to clones that could not be categorized because of one or both daughters being both SOX9- and NEUROG3-negative after immunostaining. (I) Quantification of the number of NEUROG3⁺ cells generated by the different clone patterns. 84 NEUROG3⁺ cells were detected in 63 NEUROG3⁺ two-cell clones. Indeterminable refers to clones in which the second daughter was neither NEUROG3- nor SOX9-positive. Scale bars, 100 μm (D) and 3 μm (E–G). Histograms represent the mean (n = 22). See S3 Table for further data.

https://doi.org/10.1371/journal.pbio.1002111.g002

From the above data with regard to fate-determinable two-cell clones, we estimated expected ratios of P/P, P/N, and N/N divisions to be 69.9%, 18.6%, and 11.5%, respectively, after excluding indeterminable clones. Progenitors undergoing symmetric differentiating divisions will contribute all of their progeny to the differentiated pool, whereas asymmetrically dividing progenitors will contribute only one half of their progeny to this pool. We can therefore directly estimate the probability of differentiation of new-born cells to be 20.8% ([0.5 × 18.6] + 11.5)%, which is consistent with a net expansion of developing pancreas (S1.5 Text).

If sibling cells adopted their fate independently of each other, the expected fractions for each division type would be 62.7% for symmetric proliferative, 4.3% for symmetric differentiative, and 33% for asymmetric. Notably, these last two fractions deviate from the experimentally reported ones (Fig. 2H), contradicting the hypothesis of independent sibling fate allocation. This is further supported by statistical tests indicating a significant divergence from the independence ratios (S1.5 Text). Similar calculations can be made on the in vitro data leading to the same conclusion that a single conversion event leads to symmetric endocrine cell production (Fig. 1O and S1.5 Text).

A New Neurog3-RFP Reporter Reveals Synchrony in Differentiation and Differentiation Delays Suggestive of Cell Cycle–Dependent Priming

To investigate the dynamics of differentiation, we generated transgenic Neurog3-RFP reporter lines that can be used for live imaging together with H2B-GFP (Figs. 3A,B and S3A,B). Immunostaining for NEUROG3 and comparison with red fluorescent protein (RFP) from E14.5 Neurog3-RFP pancreas revealed that 40.1% ± 4.5% of NEUROG3⁺ cells were co-expressing RFP, while the remaining NEUROG3⁺ cells were RFP⁻ (S3C,D Fig.). Some discrepancies may be expected because of the transient nature of Neurog3 expression and the different onset and decay kinetics of the RFP protein compared to the NEUROG3 protein (S1.3 Text). Moreover, 77.1% ± 2.8% of RFP⁺ cells were NEUROG3⁻ due to probable delay and perdurance of RFP as compared to that of NEUROG3 [28], as also seen for other reporters [29–31]. This maintenance was attested by the detection of hormones in 18.5% ± 2.3% of RFP⁺ cells. To further address the reliability of the reporter and assess its incidence in our analysis, we compared this line to the enhanced yellow fluorescent protein (EYFP) knock-add-on allele, which has been reported to show a greater overlap with NEUROG3 protein [29] and which is, in principle, less susceptible to exogenous chromatin environments, being in the endogenous locus. Our imaging of explants expressing one allele of EYFP and one of RFP (S4A,B Fig.) showed that all RFP⁺ cells were also EYFP⁺ (S4C Fig. and S5 Table), indicating no false positives due to genomic environment. Single cell tracing showed that RFP was turned on 4.7 ± 1.1 h after EYFP was detected (S4D Fig.); 11.6% ± 3.7% of EYFP⁺ cells never became RFP⁺, indicating the system was largely faithful.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Extended live imaging with Neurog3-RFP reporter reveals the dynamics of progenitor cell cycle and differentiation.

(A) Scheme summarizing the genetic strategy to visualize PDX1⁺ pancreatic progenitors and NEUROG3⁺ endocrine progenitors for live imaging. (B) Model of pancreatic progenitor divisions with a Neurog3-RFP reporter. After second division of self-renewing progenitors, cell cycle length can be obtained. Using the Neurog3-RFP reporter, endocrine differentiation timing and synchrony can be obtained. (C) Still images from S6 Movie, demonstrating an asymmetric (P/R) division producing one Neurog3-RFP⁺ daughter and two other granddaughters (white spots) from a Pdx1^tTA/+;tetO-H2B-GFP;Neurog3-RFP explant. After the first division, one daughter turns on RFP (before elapsed time 30:00), and later the other daughter divides, producing two granddaughters (at elapsed time 42:12). (D-G) Images of fixed explant with native GFP (D) and immunostained for SOX9/aPKC (E) and Neurog3-RFP (F, staining for MYC-tag). White spots correspond to cells in (C), and one is RFP⁺ and two granddaughters are SOX9⁺ (E,F). Inset in (E) shows high magnification image of SOX9 staining. (H) Still images from S7 Movie, demonstrating a symmetric (R/R) division producing two Neurog3-RFP⁺ daughters (grey spots). After the division, both daughters turn on RFP. (I–M) Images of fixed explant with native GFP (I) and immunostained for SOX9/aPKC (J), NEUROG3 (K), and Neurog3-RFP (L, staining for MYC-tag). Both daughters are NEUROG3⁺/RFP⁺. (N) Fraction of RFP-producing cell divisions. Each category (pink, blue, and purple bars) was counted from three movies. (O) Analysis of RFP emergence from three live imaging movies. In three cases, RFP⁺ cells divided producing two RFP⁺ cells each (cyan bar), and the majority of RFP⁺ cells were either lost or moved out of frame during back-tracking (indeterminable, purple bar). (P) Fraction of asymmetric versus symmetric cell divisions from three different measurements: NEUROG3 tracking, Neurog3-RFP tracking, and in vivo clonal analysis. All three measurements exhibit equivalent rates of divisions. Numbering denotes elapsed time in h:min, and in the cell division diagrams P indicates progenitor and R, Neurog3-RFP (C, H). Scale bars, 20 μm. Histograms and error bars represent the mean and standard deviation (n = 3). See S6 Table for further data.

https://doi.org/10.1371/journal.pbio.1002111.g003

From time-lapse imaging extended to 48 h, we could observe a dynamic change of RFP signal in single cells from the onset of fluorescence: gradual increase and a subsequent decrease, which reflects the transient expression of NEUROG3 [32]. Our analysis predicts a short half-life of 5–6 h for RFP in a cell, most probably due to continuous laser exposure. We estimate a “perdurance” of detectable fluorescence of more than 20 h (see S5 Fig. and S1.3 Text) and a minimum delay between cell priming and RFP onset of approximately 5 h. Monitoring all events of RFP onset (n = 323; Fig. 4A) initially suggested waves of cellular differentiation at the tissue level. However, statistical analysis of the timing of onset events showed that these are also compatible with a stochastic process of cell differentiation with homogeneous differentiation rate (i.e., a Poisson process) throughout the imaging period (S6 Fig. and S1.4 Text). While we cannot rule out a periodic process underpinning Neurog3 expression, confirmation of this would require more data points.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Analysis of differentiation and cell cycle dynamics from live imaging.

(A) Analysis of Neurog3-RFP onset in four locations from three different time-lapse movies (n = 56, 89, 54, and 125, respectively). Each vertical bar symbol indicates an onset event, and the yellow area displays the probability, obtained by kernel density estimation, of an event occurring over time. These suggest that cell differentiation might not be a homogeneous process; however, further statistical analysis does not rule out this possibility (S1.4 Text). (B) Lag time of Neurog3-RFP onset between daughters derived from symmetric (R/R) divisions (n = 19). Symmetrically fated daughters exhibited synchronized expression of Neurog3-RFP, as pointed out by the highly correlated lag time between division and RFP onset (inset). (C) Lag time between division and Neurog3-RFP onset in asymmetric (P/R, n = 27) versus symmetric (R/R, n = 38) divisions. Note the data are pooled from three live imaging movies. RFP cells from P/R divisions took a significantly longer time to turn on RFP than cells from R/R divisions. Black-outlined red circles from P/R division (n = 14) indicate P/R divisions producing grand-daughters through progenitor daughter division. (D) Doubling time of progenitors originating from either symmetric (P/P, black circle) or asymmetric (P/N and P/R, pink circle) divisions. Doubling time of asymmetrically generated progenitors took longer than symmetrically generated progenitors. Statistical analyses were done using two-tailed Mann-Whitney test. *** p < 0.0001 and * p = 0.04 (C,D).

https://doi.org/10.1371/journal.pbio.1002111.g004

Similar to earlier tracking, RFP⁺ cells were back-tracked from the last time point in time-lapse movies, their prior division was monitored, and sister cells were forward-tracked. Quantifications (S6 Table) revealed that among the RFP-producing divisions where the fate of the two sisters was determinable, as follows: 60.2% ± 11.9% were asymmetric divisions producing a progenitor and a RFP⁺ daughter (P/R; S6 Movie, and Fig. 3C–G,N), and 39.8% ± 11.9% were symmetric divisions producing two RFP⁺ daughters (R/R; S7 Movie, and Fig. 3H–N). In these long time-lapse movies, many RFP⁺ cells moved out of frame or were lost due to weak GFP signal before acquiring RFP expression (Fig. 3O and S6 Table). Excluding those indeterminable RFP cells, 18.8% ± 6.6% were generated through P/R division, 25.0% ± 10.0% through R/R division, and 3.2% ± 2.8% through RFP division, while no division was seen during the movie duration for 53.0% ± 10.3% of RFP cells (Fig. 3O). These RFP tracking results thus confirmed the calculated proportions of asymmetric versus symmetric divisions established from the previous live imaging and in vivo clonal analysis (Fig. 3P).

The dynamic reporter revealed highly synchronized differentiation after divisions producing two NEUROG3 cells, the RFP signal being detected in both daughters within 0.8 ± 0.4 h of each other (Fig. 4B, correlation coefficient between lag times 0.98 ± 0.002). This outstanding synchrony confirms that it is unlikely that the two daughters are induced by independent events and suggests that mother cells have been primed to differentiate into NEUROG3 cells prior to their division. This observation further suggests a defined time between priming and NEUROG3 expression (or its RFP proxy). In addition, asymmetrically generated NEUROG3 cells exhibited a significantly longer lag time between the division and RFP onset, as shown in Fig. 4C, further supporting an interplay between cell cycle, the differentiation priming event, and the division mode.

These results on the contrasting dynamics of differentiation between cells stemming from symmetric versus asymmetric division events are obtained with the RFP reporter, for which we have established a false negative rate of 11.6% (S4C Fig.). This implies that, far from amplifying the differences between the dynamics of differentiation between the two groups, we might be underestimating them. Specifically, our reporter may miss a subset of Neurog3-expressing cells, thus leading to mis-allocation of around 11.6% of symmetric events to asymmetric and therefore homogenizing the two categories and reducing the differences between them (See below).

A Model of Cell Cycle–Dependent Stochastic Priming of Progenitors Provides Quantitative Insight into the Dynamics of Cell Differentiation

To try to understand the mechanism underlying the emergence of endocrine progenitors, we devised a simple mathematical model [33] of cell proliferation dynamics based on the lineage and differentiation dynamics data. The model is based on the premise that proliferating progenitor cells primed for Neurog3-dependent differentiation might either exit the cell cycle and become terminally differentiated or commit to complete the cell cycle and produce two terminally differentiated cells via mitosis (Fig. 5A–C). Thus in this model there are three, rather than two, cell types: (i) NEUROG3-primed (N) cells, which are post-mitotic; (ii) cells primed for differentiation but committed to cell cycle completion (L cells); and (iii) progenitor (P) cells, which will not differentiate (Fig. 5D). We assigned a probability q (the differentiation probability) for the differentiation of progenitors and a probability, θ, for primed cells to become N (1- θ to become L). Thus, the model describes the differentiation process in terms of two probabilities, which can be directly inferred from the lineage data (S1.6 Text), i.e. it has no free parameters. From the in vivo clonal analysis data, we estimate θ = 56.5% and, as we have already seen, q = 20.8%. This means that approximately one-half of the progenitors primed for differentiation become post-mitotic (P→N; 56.5%), while the other half (P→L[→N/N]; 43.5%) will undergo one last division before differentiating. Because this latter group of cells (L) is transient and contributes terminally differentiated cells (N), its expected abundance in the tissue is residuary. The model predicts that, at any given time, only 9.8% of cells in the developing tissue are primed progenitors committed to division-cycle completion (L), yet this small fraction accounts for 93.8% of the symmetric differentiative divisions (L→N/N) and might therefore explain our observation that the fates of sibling cells are linked (S1.6 Text). The remainder of symmetric differentiative divisions is interpreted to result from random, independent priming in two sister P cells.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Proposed model.

The observations from 3-D live imaging suggest that a distinct temporal induction of endocrine progenitor fate during the cell cycle may result in different fates of progenitors. As the majority of NEUROG3 cells are known to be largely post-mitotic [11], we propose models for three modes of cell division, resulting in NEUROG3 daughter differentiation according to a priming time point. (A) Asymmetrically fated daughter differentiation after progenitor cell division (P→P/N). Only one daughter may be induced (arrow) after the mother division, resulting in exit of cell cycle and its differentiation to NEUROG3, while the other daughter is fated as a progenitor, resulting in self-renewal. (B) Symmetrically fated daughter differentiation after progenitor cell division (P→N/N). Before mitosis, the progenitor may be induced to differentiate into an endocrine progenitor, complete the cell cycle, and divide, resulting in both daughters differentiating into NEUROG3. (C) Symmetrically fated daughters through symmetric division of endocrine progenitors (N→N/N). Considering endocrine progenitors post-mitotic, the progenitor may be induced to differentiate into endocrine progenitor, but has not yet finished the cell cycle before the cell actually differentiates into an endocrine progenitor. Therefore, to complete the cell cycle, a recently differentiated NEUROG3 cell may divide and give rise to two NEUROG3 daughters. (D–F) We have developed a mathematical model of cell cycle–dependent stochastic priming of progenitors to endocrine fate. (D) Schematic of the model in which pancreatic progenitors (P, green circles) stochastically are primed for differentiation with probability q. Primed cells can either exit the cell cycle and differentiate into NEUROG3 (N) with probability θ or conclude the cycle (L) and give rise to two NEUROG3 cells. (E) The proposed model accounts for the observed frequencies of each division mode and predicts differential RFP onset dynamics in asymmetric and symmetric divisions. “Experiment*” bar in Asymmetric category denotes asymmetric divisions accounting for a RFP daughter and a self-renewing RFP⁻ daughter (n = 14), whereas “Experiment” bar includes all the asymmetric divisions (refer to Fig. 4C). (F) The model with experiment-matching number of clones (out of 10,000 simulated clones) predicts a larger lag time between division and RFP onset in cells stemming from asymmetric divisions versus symmetric divisions. (G) Correlation of RFP lag times between sibling cells predicted by the model also matches that which was experimentally measured.

https://doi.org/10.1371/journal.pbio.1002111.g005

The model also allows multiple interpretations for the probability of becoming L versus N (e.g., exposure to differentiation signals, gene expression noise, etc.). One such interpretation is the timing of the priming event after division (i.e., θ can be construed as accounting for a cell cycle restriction point). For instance, if a cell is primed early after division it might differentiate and halt the cell cycle, whereas if the priming event occurs late in G1, the cell might have already committed to cell cycle completion. Such specific reading of the model, which we adopt hereafter, leads to a few qualitative predictions on the dynamics of differentiation. First and foremost, the vast majority of sibling cells (93.8%) from symmetric divisions will have a perfectly synchronized differentiation program. According to the model, differentiated cells stemming from symmetric divisions shall turn on the differentiation program, on average, much earlier than those from asymmetric divisions (S1.6 Text). To quantitatively account for these predictions and compare them to the experimental data, we performed computational simulations of the model (n = 10,000 clones, S11 Fig.) including the observed variability in the cell cycle length as well as the dynamics of the fluorescent reporter (Figs. 5E–G, S9, S11 and S1.3, S1.4, and S1.6 Text; data deposited in the Dryad repository: http://dx.doi.org/10.5061/dryad.4b58d [34]). The simulations reproduced the differences in the onset of the reporter in cells stemming from symmetric versus asymmetric divisions (Figs. 5E,F, S6E,F, and S9) and also for the high degree of synchronization between sibling cells (Figs. 5G, S6G, and S9). Furthermore, when we included the 11.6% false negative rate in the reporter (S4C Fig.) of the model, the results were not significantly affected (S10 Fig. and S1.6 Text).

The results of the model led us to experimentally characterize the cell cycle. We used FACS-sorting of Pdx1^tTA/+;tetO-H2B-GFP⁺;Ngn3-RFP⁻ cells marking pancreatic progenitors at E14.5 to establish their cell cycle partition and observed that 69% were in G0/G1, 27% in S and 5% in G2/M, whereas 98% of Neurog3-RFP⁺ cells were in G0/G1 (S7 Fig.). The progenitors thus spend the majority of their time in G1. Our hypothesis is that priming in early G1 would lead to differentiation and exit of the cell cycle, and its mother would thus have an apparent asymmetric cell division. In contrast, priming in late G1 after the cell has committed to complete the cell cycle through DNA replication and mitosis would lead to an apparent symmetric differentiative cell division.

Finally, simulations also predicted the existence of a residual fraction of primed cells that would turn on the reporter immediately before dividing. Noticeably, although most RFP⁺ cells did not divide, we observed 3 cases of RFP⁺ cell division producing six cells (Fig. 6, S8 Movie, and S6 Table) and thus accounting for 3.2% ± 2.8% of all tractable RFP cells. The average time between RFP signal acquisition and division was 1.7 ± 0.8 h. This result is in agreement with the previous estimations of the progression of NEUROG3 cells through S-phase (BrdU incorporation) [11] and further shows that the NEUROG3 cells can exceptionally progress through mitosis at an early stage of their life [29].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. A small number of Neurog3-RFP cells divide into two NEUROG3⁺ cells.

(A) Still images from S8 Movie, demonstrating division of Neurog3-RFP cell (white spots) in RFP channel from a Pdx1^tTA/+;tetO-H2B-GFP;Neurog3-RFP explant. (B–F) Images of fixed explant with native GFP (B) and immunostained for Neurog3-RFP (C, staining for Myc-tag), NEUROG3 (D) and Sox9/aPKC (E). Both daughters are NEUROG3⁺/RFP⁺. Numbering denotes elapsed time in h:min (A). Scale bars, 20 μm.

https://doi.org/10.1371/journal.pbio.1002111.g006

Analysis of the Expansion Potential of Progenitors

The longer movies enabled the observation of multiple rounds of division and the quantification of cell cycle parameters (Fig. 3B). They first confirmed that the daughter cells qualified as progenitors after asymmetric cell division based on SOX9 expression (Fig. 1H) were functionally behaving as progenitors. Indeed, among P/R divisions (n = 27), we observed 14 events where the RFP⁻ daughter divided, producing second-generation progeny (S6 Movie and Fig. 3C). Of those 14 cases, we observed two events where one or both granddaughters divided again, producing third-generation progeny. For all of those, immunostaining at the end point revealed that the RFP⁻ progeny were SOX9⁺ progenitors (Fig. 3E). In such cases, we could calculate the doubling time of daughter progenitor divisions, and we compared it between P/P and P/N or P/R divisions (Fig. 4D). The doubling time of self-renewing progenitors from P/P divisions was shorter than that from P/N or P/R divisions (p = 0.04, 21.0 ± 2.4 h and 26.5 ± 7.3 h, respectively). Moreover, the distribution of data points was greater in the asymmetric cell divisions. Finally, the time-lapse movies revealed that pancreatic epithelial cells were highly dynamic and that two daughters migrated to distances up to 64 μm apart from each other in the 24 h following division regardless of division mode (S8 Fig.).

Mice

Genetically engineered mice used for this study were as follows: Pdx1^tTA/+ [42], tetO-HIST1H2BJ/GFP (tetO-H2B-GFP) [43], Hnf1bCreER [3], Gt(ROSA)26Sortm4(ACTB-tdTomato,-EGFP)Luo/J (mT/mG) [44], Neurog3-EYFP [29], and Neurog3-RFP (S3 Fig.). For embryonic stage, noon of the day when vaginal plug appeared was referred as E0.5.

The Neurog3-RFP transgenic construct (S3A Fig.) was generated by fusing 7.6 kb of the Neurog3-promoter [2] with a reporter construct composed of a chimeric intron; turbo RFP (Evrogen); a nuclear localization signal (NLS); a Myc-tagC; a bovine growth hormone polyadenylation signal (bGH-PolyA). Transgenic mouse lines were obtained by pronuclear injection of the construct (Transgenic Core Facility, EPFL, Switzerland). Two different lines were obtained initially, exhibiting similar levels of RFP signal detectable by a wide-field fluorescent microscope, and one of the lines was used for this study. All animals were handled humanely according to the authorized protocols of Switzerland and Denmark.

3-D Live Imaging and Cell Tracking

Dorsal pancreata from E12.5 Pdx1^tTA/+;tetO-H2B-GFP or Pdx1^tTA/+;tetO-H2B-GFP;Neurog3-RFP were cultured on a fibronectin (Sigma)-coated coverslip, adapted from the previously published protocol [25]. GFP and RFP were readily detectable under wide-field fluorescent microscopes. We used a culture medium composed of Medium 199, 10% fetal calf serum, 1% Fungizone, and 1% penicillin/streptomycin (all from Gibco). After 24 h of culture that enabled stabilization of explant flattening to approximately 80 μm thick, pancreatic explants were imaged at a single-cell resolution using Leica SP5 or SP8 confocal microscopes with a 63X glycerol immersion objective in a humidified, heated, CO₂-controlled chamber. Tiled positions (9 [3x3] or 12 [3x4] tiles) were scanned in 256x256–280x280 format with around 40 μm Z-stack (voxel size, 0.506x0.506x1.3 μm³–0.880x0.880x1.25 μm³) every 6 min for 18–48 h. The GaAsP hybrid detection system (Leica HyD™) enabled a substantial reduction of laser power by 62.5% and increase in signal-to-noise ratio resulting in reduced scanning time, compared to conventional photomultiplier detectors. At each time point, it usually took approximately 5 min and 30 s to scan 9–12 tiled positions in 3-D. At the end point of image acquisition, the explants were fixed and prepared for whole-mount immunostaining.

Tiled images were stitched using either Leica AF6000 software or a custom-built Massive Stitcher plugin (Bioimaging and Optics Platform, EPFL, Switzerland) in Fiji. Imaris (Bitplane, Switzerland) software was used to track cells and their divisions in 3-D maximum intensity projection. Once immunostaining was done, NEUROG3⁺ endocrine progenitor cells from staining images were manually identified on the last frame of time-lapse movies with Pdx1^tTA/+;tetO-H2B-GFP explants by GFP superimposition. The identified endocrine progenitors were first back-tracked to monitor their prior divisions. Once a division was observed, the other sister was forward-tracked to the final frame, and its fate was determined from the immunostaining images. For time-lapse movies from explants with Ngn3-RFP in addition to Pdx1^tTA/+;tetO-H2B-GFP, RFP⁺ cells were back-tracked, and the fate of each sister was determined by immunostaining. For the quantification of total cell divisions, due to the technical difficulties in tracking all Sox9⁺ cells from the immunostaining, we did not trace all the individual cells from those 1,628 divisions, but rather subtracted the tracked divisions that produced NEUROG3 cells from the total number of divisions.

In Vivo Clonal Analysis

Pregnant mice carrying Hnf1bCreER;mT/mG embryos were injected intraperitoneally with 0.175 mg 4-hydroxy (4-OH) tamoxifen (H6278, Sigma Aldrich) at E13.5. 4-OH tamoxifen was prepared as a 10 mg/mL stock in 90% corn oil and 10% ethanol and diluted to obtain the desired dose. Embryos were harvested at E14.5, and the dorsal pancreas was isolated and subjected to whole-mount immunostaining for GFP, SOX9, and NEUROG3, as indicated below. The fixation procedure eliminates native GFP and Tomato signals. After whole-mount immunostaining, dorsal pancreata were dehydrated through an ascending methanol series and subjected to clearing in a 1:2 solution of benzyl alcohol to benzyl benzoate (BABB). Cleared samples submerged in BABB were mounted on glass depression slides and imaged whole-mount using a Leica SP8 confocal microscope with a 20X oil objective at a 1024x1024 format. 3-D reconstruction of whole-mount imaged pancreata was performed using Imaris (Bitplane), enabling detection of recombined clones while preserving the spatial organization of the pancreas, thereby ensuring detection of clonal progeny by allowing interclone distance measurements. Two-cell clones were identified in 3-D space, and categorized according to SOX9 and NEUROG3 status. Recombined cells were only considered to be of clonal origin if the distance between recombined cells was less than 30 μm after the tracing period, based on live imaging data (S8 Fig.). The results were not sensitive to this parameter as using 60 μm as a maximal distance to be considered as a clone led to the same proportion of the three types of division (S2 Data).

Hnf1bCreER;mT/mG embryos were also used for in vitro clonal analysis by explanting pancreata at E13.5 and growing these at the air–liquid interface on 0.4 μm filters (Millipore). Explants were subjected to a 6 h pulse of 25 nM 4-OH tamoxifen in 100% ethanol to induce labelling at clonal density. Following tracing for 48 h, explants were fixed and subjected to whole-mount staining and imaging as indicated below.

Immunohistochemistry

Whole-mount immunostaining was performed after live imaging or for pancreata harvested from the lineage tracing. After fixation with 4% paraformaldehyde (PFA) for 5 min on ice, samples were washed in phosphate buffered saline (PBS) for 5 min three times. Then, they were dehydrated through 50% and 100% methanol, and could be stored at −20°C until later use. When ready, samples were rehydrated through 50% methanol and washing solution, PBS+0.5% Triton X-100 (Tx100). Throughout the procedure, all the solutions contained 0.5% Tx100, and all the incubation was undergone in 4°C. After blocking overnight in blocking solution (1% Bovine serum albumin+0.5% Tx100), samples were incubated with primary antibodies (S7 Table) in blocking solution for 24–48 h. After washing, secondary antibodies were applied overnight, followed by washing. Alexa fluorophore conjugated secondary antibodies (Invitrogen) were used. Stained explants were kept in PBS and imaged using a confocal microscope. For quantification from explants, NEUROG3⁺ cells were counted manually, and H2B-GFP⁺ cells were counted using a custom-made macro in Fiji.

Immunostaining of frozen sections from E14.5 Neurog3-RFP pancreata was performed as previously described [6], and images were taken with a Leica DM5500 microscope. Quantification was obtained by manually counting immunopositive cells on every sixth section.

Statistical Analysis and Probabilities

Statistical analyses were done by two-tailed Mann-Whitney U-test using GraphPad Prism software. Values were presented as the mean ± standard deviation.