Research Article

Live imaging of neurogenesis in the adult mouse hippocampus

See allHide authors and affiliations

Science  09 Feb 2018:
Vol. 359, Issue 6376, pp. 658-662
DOI: 10.1126/science.aao5056

A window on hippocampal neurogenesis

Addition of new neurons to the adult brain is key to the hippocampal functions of learning and memory. Pilz et al. labeled individual progenitor cells in the mouse hippocampus and watched them in situ for the next 2 months (see the Perspective by Götz). The results revealed the developmental progression as progenitor cells gave rise to mature cells of the dentate gyrus.

Science, this issue p. 658; see also p. 639


Neural stem and progenitor cells (NSPCs) generate neurons throughout life in the mammalian hippocampus. We used chronic in vivo imaging and followed genetically labeled individual NSPCs and their progeny in the mouse hippocampus for up to 2 months. We show that NSPCs targeted by the endogenous Achaete-scute homolog 1 (Ascl1) promoter undergo limited rounds of symmetric and asymmetric divisions, eliciting a burst of neurogenic activity, after which they are lost. Further, our data reveal unexpected asymmetric divisions of nonradial glia-like NSPCs. Cell fates of Ascl1-labeled lineages suggest a developmental-like program involving a sequential transition from a proliferative to a neurogenic phase. By providing a comprehensive description of lineage relationships, from dividing NSPCs to newborn neurons integrating into the hippocampal circuitry, our data offer insight into how NSPCs support life-long hippocampal neurogenesis.

The hippocampus requires new neurons in the dentate gyrus (DG) throughout life for learning and memory (1). Failing or altered hippocampal neurogenesis has been implicated in a variety of diseases such as major depression and age-related cognitive decline (2, 3). On the basis of thymidine analog labeling, in vivo lineage tracing, and cell ablation studies, it has been proposed that radial glia-like (R) progenitor cells are the bona fide stem cells of the adult DG (49). According to the prevailing model of adult hippocampal neurogenesis, R cells self-renew—here defined as generating a daughter cell with equivalent molecular characteristics and potency—and give rise to proliferative nonradial glia-like (NR) cells that divide symmetrically to generate granule cells (3). However, the self-renewal capacity and lineage relationships of R cells remain controversial owing to the lack of longitudinal observations of individual R cells and their progeny within their niche (7, 8). Similarly to previous imaging approaches that have probed the dynamics of somatic stem cell behavior in the nonmammalian nervous system and other stem cell niches (1017), we used chronic in vivo imaging to track the fate of individual R cells over time within the adult DG.

Chronic imaging of individual brain stem cells

To label hippocampal R cells, we used mice expressing a tamoxifen (Tam)–regulable Cre recombinase under the control of the endogenous Achaete-scute homolog 1 (Ascl1) promoter crossed with a tdTomato reporter mouse line (Ascl1-tdTomato mice) (18). Ascl1-expressing cells constitute an essential population of neural stem and progenitor cells (NSPCs) in the adult DG (1820). Adult Ascl1-tdTomato mice were implanted with a cortical window that left the hippocampal formation intact and allowed for two-photon imaging (Fig. 1A and fig. S1A) (21). A single Tam injection induced sparse labeling of Ascl1-expressing cells that were classified as R or NR cells on the basis of morphological features and marker expression (Fig. 1, B and C; fig. S2; and movie S1). Only R cells were analyzed as a starting population. Individual clones were imaged every ~12 to 24 hours (unless otherwise indicated) and followed for up to 2 months (Fig. 1, D and E, and fig. S3). Imaged clones (n = 63) were characterized on the basis of behavioral and morphological criteria (methods, fig. S2, movie S2, and table S1), allowing for the construction of individual lineage trees (Fig. 1E, fig. S3, and movie S3). After imaging, the final fate of progeny was confirmed using immunohistochemistry (Fig. 1, E and F, and fig. S4).

Fig. 1 Chronic in vivo imaging of neurogenesis in the adult DG.

(A) Scheme illustrating the experimental approach allowing for chronic in vivo imaging of NSPCs in the adult DG of Ascl1-tdTomato mice. (B) Representative in vivo images of R and NR cells at 2 days post-induction (dpi). (C) Immunostained images showing Sox2-positive (green), Ascl1-tdTomato–labeled (red) R cells with GFAP (glial fibrillary acidic protein)–positive (white) radial processes and NR cells (Sox2-positive and GFAP-negative) in Ascl1-tdTomato mice at 2 dpi. (D) Selected imaging time points for two R cells (respectively indicated with open and filled arrowheads) over the course of 2 months, showing the emergence of two neuronal clones. Time points after Tam injection are indicated in each panel. Shown are collapsed z-stacks. The clonal expansion of individual R cell progeny and subsequent neuronal maturation can be seen. (E) Lineage tree deduced from tracking one R cell [open arrowhead in (D)] and its progeny. Identified cell types are color-coded, and lineage transitions are depicted depending on their certainty (methods). Each circle in the lineage tree represents an imaging time point. The y axis shows the duration of the imaging. (F) Post hoc immunohistochemical analyses of the clone shown in (D) (boxed area, day 59) confirm neuronal progeny with newborn cells positive for Prox1 (green) and negative for Sox2 (white). The horizontal view of the DG corresponds to the view obtained during in vivo imaging. Scale bars, 20 μm [(B) and (C)] and 50 μm [(D) and (F)]. d, days; GCL, granule cell layer.

In agreement with previous static clonal lineage tracing experiments, we found that in 8- to 9-week-old Ascl1-tdTomato mice, 67% (42/63) of R cells entered the cell cycle and became active during the time course of imaging. Of these active R cells, 88% (37/42) divided within the first 20 days and, as a population, gave rise to both neuronal and glial daughter cells (Fig. 1, D to F, and figs. S3 and S4, A to D) (19).

Once activated, cycling R cells divided 2.3 ± 0.1 times and persisted for 9.6 ± 1.4 days on average (Fig. 2A and fig. S3). We did not find Ascl1-targeted R cells that generated neuronal progeny and returned to long-term (>4 weeks) quiescence within the 2-month observational period in any analyzed clones (Fig. 2A and figs. S3 and S4E). This suggests that, once activated, Ascl1-targeted R cells do not reenter long-term quiescence but generate a burst of neurogenic activity before committing to terminal neuronal differentiation and loss (Fig. 2A and figs. S3 and S4, E to H). The average clone size derived from active R cells was 4.8 ± 0.5 cells (Fig. 2B and fig. S3). We found no evidence for terminal differentiation of R cells into astrocytes, which had been suggested for nestin-expressing NSPCs after several rounds of cell division (8) (figs. S3 and S4H). Thus, the self-renewal capacity of Ascl1-targeted R cells is temporally limited; this finding is similar to previous results obtained using population-based static analysis of nestin-labeled NSPCs (8).

Fig. 2 The mode of NSPC division is associated with individual cell division history.

(A) Self-renewal duration (time between first and last division in each lineage) of R cells (9.6 ± 1.3 days; n = 39). (B) Distribution of the final number of cells per active clone (n = 42 lineages). Open circles represent individual clones. (C) Chronic in vivo imaging before and after cell division illustrates asymmetric cell division of R cells. A large overlap is evident in the cellular morphology before (red arrowhead and red outline) and after (green arrowhead and green outline) R cell division. The black arrowhead points at the asymmetrically generated daughter cell. (D) A morphometric index (including circularity and process length; details are given in the methods) shows little deviation in cell morphology before and after cell division (9.6 ± 1.8%; n = 9). (E) Heat map representing the frequencies of modes of division of R cells (all divisions and division rounds 1 to 3; n = 68 divisions total). The division mode changes from predominantly asymmetric (division 1) to a more symmetric differentiating division (divisions 2 and 3). N, neuron; A, astrocyte. (F) Example of an asymmetric division of an R cell (lineage 40; fig. S3). (G) Example of a symmetric division of an R cell (lineage 13). Mother cells are indicated with arrowheads and daughter cells with arrows. (H) Heat map representing the frequencies of cell division modes of NR cells (all divisions and division rounds 1, 3, and 5; n = 153 divisions). (I) Example of a symmetric NR cell division (lineage 1). (J) Example of an asymmetric NR cell division (lineage 3). The NR daughter continues to divide. Mother and daughter cells are indicated as in (F) and (G). (K) Cell division time (TD) of R and NR cells for different divisions. (L) The times until next cell division of sister cells originating from a single R mother cell are correlated (Pearson’s ρ = 0.77; *P < 0.00001; n = 22 pairs). (M) The TD of sister cells originating from a single NR mother are not correlated (Pearson’s ρ = 0.44; P = 0.08; n = 16 pairs). Each plus sign represents a pair of sister cells. Red lines, means; error bars, SEM. Scale bars, 20 μm [(C), (F), (G), (I), and (J)].

Cell division history is associated with cell fate

We then considered the fate behavior of activated R cells and their progeny. Previously, it has been proposed that the predominant mode of R cell division is asymmetric (68, 22). However, without access to continuous in vivo cell tracking, evidence for asymmetric fate has been indirect. Morphological analyses of cell body and radial glia-like processes of R cells before and after cell division revealed that the morphology of R cells remained stable (Fig. 2, C and D, and fig. S5A), providing direct evidence for asymmetric cell divisions (68, 22). Although the majority of observed first cell divisions were asymmetric, generating a R cell and a NR cell (79.3%; Fig. 2, E and F), we found that 13.8% of first R cell divisions expanded the R cell pool through symmetric divisions (Fig. 2, E and G), mirroring the behavior found in static clonal studies of R cells targeted by a nestin promoter (7). With the further identification of direct neurogenic cell divisions of R cells (Fig. 2E and figs. S3 and S5, B and C), all three modes of division were observed, reminiscent of the fate behavior described for neural progenitors in the developing neocortex (23, 24). The majority (70.6%) of all R cell divisions led to the generation of NR cells that were identified on the basis of their lack of a radial process, their ability to enter the cell cycle, and their capacity to generate neuronal progeny at later stages during the imaging period (Fig. 2, E and F). In contrast to previously suggested models (3), we found not only symmetric, neurogenic cell divisions of NR cells but also a substantial fraction of asymmetric cell divisions (24.2% of all NR divisions), yielding one renewed NR cell and one neuronal daughter cell (Fig. 2, H to J) (9). NR cells underwent as many as six rounds of cell division (with an average of 2.9 ± 0.2 divisions); thus, NR cells are a major source of clonal expansion (Fig. 2, H to J, and fig. S3).

Chronic imaging also allowed us to analyze whether division times (TD) are correlated with previous cell divisions or within clonally related lineages. We analyzed the TD of R and NR divisions and found that, once the cells were activated, TD remained constant (Fig. 2K). TD times of sister cells were correlated among R cell daughters, whereas no correlation was detected for daughters of NR cells (Fig. 2, L and M), suggesting segregation of cellular features selectively in R cells that determine the daughter cells’ reentry into cell division.

Asymmetric segregation of cell death within clonal lineages

Not all newborn cells survive and become stably integrated into the DG circuitry. Two critical periods of cell death have been described previously: an early phase of cell death within the first days after cell birth and a later phase of neuronal selection that is activity-dependent and occurs about 1 to 3 weeks after new neurons are born (2527). Consistent with previous reports, we found an average frequency of cell death among the progeny per lineage of 59.6% and identified two waves of cell death occurring around 1 to 4 and 13 to 18 days after birth, respectively (Fig. 3, A and B) (2527). We identified interclonal variability, with some clones showing no cell death, whereas other clones completely disappeared over time (Fig. 3A). Furthermore, we found differences regarding the susceptibility to cell death among individual sublineages when analyzing levels of early cell death (until 7 days after cell birth; Fig. 3, C to F), suggesting that cell death is not evenly distributed among progeny. The underlying cause for the observed subtree-associated variability of cell death remains unknown, but it may involve intrinsic mechanisms—for example, retrotransposon-associated genetic alterations or unequal segregation of aging factors (28, 29)—rather than environmental, niche-dependent factors, such as growth factor availability (3). Supporting this interpretation, we found that surviving cells can lie interspersed among death-prone cells (Fig. 3G).

Fig. 3 Chronic in vivo imaging reveals variable susceptibility to cell death.

(A) The frequency of cell death in all chronically imaged lineages ranges from 0 to 100% (mean = 59.6%, n = 42; error bars, SEM). (B) Time point of cell death after last cell division. Two peaks of cell death occur before and after 7 days, with almost no cell death occurring >20 days after birth (n = 242). (C) Pictogram depicting the comparison of cell death frequencies in subtrees 1 and 2, derived after the initial R cell division, plus subsequent progeny within that subtree (Div2plus). (D) Difference in cell death frequencies in subtrees 1 and 2 in comparison with cell death frequency in the whole lineage [lineages with >25% difference in cell death rate are colored red; only early cell death (until day 7 after birth) was included]. (E) Pictogram depicting the comparison of cell death frequencies in the Div2plus subtrees and the subtrees derived from the second division after initial R cell division plus subsequent progeny within those subtrees (Div3plus). (F) Cell death frequencies are more asymmetrically distributed among Div3plus than among Div2plus sublineages, as demonstrated by the higher weighted standard deviation to the Div2plus subtree death frequency [Div2plus difference, 13.6 ± 1.6 (n = 34); Div3plus difference, 27 ± 2.2 (n = 33); *P < 0.0001, Wilcoxon rank sum test]. Weighted standard deviation was used to account for differences in subtree sizes within each clone. Red lines, means; error bars, SEM. (G) Locations of surviving newborn neurons (blue circles) relative to cells that underwent cell death (red circles). The observed times until death (calculated from the birth of the individual cell) for dying cells are 2.5 days (cells 1 and 2 in example 1) and 13, 3, and 4 days (cell 1, 2, and 3 in example 2). Spatial localizations of surviving and dying cells overlap. Scale bar, 100 μm.

Developmental-like program describes cell fate behavior

Inspection of reconstructed R cell lineages revealed a wide variability in neurogenic potential and fate outcomes (fig. S3). After induction, some R cells differentiated early, giving rise to one or two short-lived neurons, whereas others gave rise to more than 10 surviving neurons (Fig. 2B and figs. S3 and S4F). Despite this variability, some features of R cell fate behavior were conserved among all lineage trees. In particular, we found no instance of an asymmetrical R cell division being followed by a symmetric R cell duplication (Fig. 2E and fig. S3), suggesting either that, once activated, R cells move sequentially from a symmetrically to an asymmetrically dividing phase, or that the proliferative potential of R cells becomes progressively exhausted so that symmetrical duplicative divisions become increasingly scarce. In contrast, variability in the output of dividing NR cells was indicative of a conserved pattern of stochastic fate, with NR cells choosing between symmetric duplication, asymmetric division, and symmetric differentiation with probabilities independent of the cell generation (Fig. 2H and fig. S3).

Thus, on the basis of these observations, we considered the quantitative fate behavior of R cells. Given the sequential pattern of symmetric and asymmetric divisions, we questioned whether R cells might be following a developmental-like program, switching irreversibly from a phase of proliferative (symmetrical) divisions to a phase of neurogenic (asymmetrical) divisions, as observed during cortical development (Fig. 4A, fig. S6A, and table S2) (30). Using a statistical modeling approach, we quantitatively assessed the viability of this hypothesis by fitting the lengths of putative proliferative and neurogenic phases against average clonal properties (methods). This simple developmental-like paradigm yielded predictions of the proliferative output, the cell fate distributions, and the average clonal composition over time that were in agreement with the observed data within the theoretically predicted variability (Fig. 4, B to D, and methods). As a consistency check, we also assessed whether the observed sequential fate pattern could represent the chance outcome of stochastic fate behavior, with R cells becoming progressively biased away from self-renewal toward differentiation over time (figs. S6 and S7). However, given the observed cell fate frequencies and the number of observed lineage trees, we estimated the chance for such an outcome to be only 2.4% (methods). Thus, we conclude that a model in which the sporadic entry of R cells into the cell cycle activates a developmental-like program of fate, leading to a burst of neurogenic activity, provides the most plausible explanation of the lineage data.

Fig. 4 Modeling-based analysis suggests a developmental-like program for R cell fate behavior.

(A) In the model paradigm, R cells follow a defined program comprising a proliferative phase (duplications) that switches irreversibly into a neurogenic phase (asymmetric divisions and terminal differentiation). (B) Total number of newborn cells in successive time intervals of 5 days after induction from experiments (dark data set) and from simulations (light data set) for R and NR cells and neurons (N) (methods). Cell counts were pooled over 55 lineage trees with nonquiescent R cells and averaged over 500 realizations for each lineage tree, resulting in a total of 27,500 simulations (model parameters are given in table S2). Error bars indicate the range within which 95% of the simulation results fall. (C) Relative frequencies of different cell fates in successive time intervals of 5 days after induction from experiments and simulations, as in (B). Later time intervals (20 to 50 days) with small numbers of events have been pooled together. SD, symmetric differentiating divisions; A, asymmetric divisions; SR, symmetric self-renewing divisions; D, cell death. (D) Average clone content as a function of time from experiments (dots) and from simulations (lines) for different indicated cell types. Shaded areas indicate the regions within which 95% of the simulation results fall.

In this study, we used chronic imaging of individual R cells and their progeny to characterize the cellular dynamics underlying adult hippocampal neurogenesis. Our data show that, after activation, Ascl1-targeted R cells enter a developmental-like program, eliciting a burst of neurogenic activity. However, self-renewal is temporally limited: We did not observe repeated shuttling between quiescence and proliferation, leading to a loss of activated R cells. These findings do not rule out the previously reported presence of stem cells in the mammalian DG that shuttle back and forth between quiescence and activity, dividing for extended periods (7, 31, 32). Stem cell heterogeneity has been postulated, and the Ascl1-targeted population analyzed here may not include all subtypes that are capable of generating neuronal progeny in the adult DG (3335). Previous data suggested that about 10 to 15% of all granule cells are adult-generated in the mouse hippocampus. This indicates that adult neural stem cells generate 30,000 to 45,000 granule cells during the entire life span (5, 36, 37), which, as a fraction of the total neuronal population, appears to be lower in the rodent than in the human DG (38). We found that, once activated, individual Ascl1-targeted R cells generated 4.8 neurons on average. On the basis of previous estimates that the DG contains ~10,000 R cells in 2-month-old mice (8), the total number of cells that can be generated by Ascl1-targeted cells with the principles of clonal expansion described here (~45,000) appears to be sufficient to explain a substantial part of hippocampal neurogenesis. Our in vivo imaging results elucidate the cellular dynamics of physiological adult neurogenesis and form the basis to understand the molecular mechanisms governing the steps from dividing stem cells to newborn neurons.

Supplementary Materials

Materials and Methods

Figs. S1 to S7

Tables S1 and S2

References (3954)

Movies S1 to S3

References and Notes

Acknowledgments: We thank S. April for help with analyzing R cell morphology and D. L. Moore and D. C. Lie for comments on the manuscript. Funding: This work was supported by the European Research Council (to S.J. and F.H.), the Swiss National Science Foundation (BSCGI0_157859 to S.J.), the Zurich Neuroscience Center, and the Wellcome Trust (098357/Z/12/Z to B.D.S.). G.A.P. was supported by a European Molecular Biology Organization Long-Term Fellowship. M.B. was supported by a SystemsX transition postdoctoral fellowship. Author contributions: G.A.P. developed the imaging approach, performed imaging, analyzed data, and cowrote the manuscript. S.B. performed imaging, analyzed data, and revised the manuscript. S.C. codeveloped the imaging approach and performed imaging. M.B. analyzed data and revised the manuscript. D.J.J. and B.D.S. contributed to the concept, performed theoretical modeling, and cowrote the manuscript. F.H. contributed to the concept and revised the manuscript. S.J. developed the concept and wrote the manuscript. Competing interests: None declared. Data and materials availability: The data reported are presented in the main paper and the supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article