Live imaging of the Drosophila spermatogonial stem cell niche reveals novel mechanisms regulating germline stem cell output

Tissue maintenance relies on the output of adult stem cells that divide to self-renew and give rise to differentiating progeny. Residing in localized cellular micro-environments, or niches, stem cells receive molecular signals that maintain their fate and regulate their function (Morrison and Spradling, 2008). Although stem cells are considered to be long-lived relative to their differentiating progeny, they also undergo turnover (Doupe et al., 2010; Klein et al., 2010; Lopez-Garcia et al., 2010; Snippert et al., 2010; Wallenfang et al., 2006). Lost stem cells must be replaced to sustain the tissue, as depletion of the stem cell pool could result in tissue degeneration (Rossi et al., 2008). Contrary to previous models, where stem cells are considered to have low turnover rates, recent lineage tracing in the mouse testis, epidermis and intestine indicate that the stem cell pool is fairly dynamic in actively cycling tissues, and the fate of an individual stem cell cannot be reliably predicted (Doupe et al., 2010; Klein et al., 2010; Lopez-Garcia et al., 2010; Snippert et al., 2010). Formally, a stem cell can be replenished either by the division of an existing stem cell to produce two stem cells (symmetric division) or by the reversion of a differentiating cell into a stem cell (de-differentiation) (Fuller and Spradling, 2007; Morrison and Kimble, 2006). Both are thought to occur in Drosophila and mammalian stem cell systems (Barroca et al., 2009; Jiang et al., 2009; Nakagawa et al., 2007; Nakagawa et al., 2010). However, direct visualization of these events in real time in both normal and regenerating tissues has remained elusive.

Live imaging of stem cells in intact tissues has yielded significant advances towards understanding stem cell function. Tracking of marked undifferentiated spermatogonia in the intact mouse testis revealed the putative niche at blood vessel branching points (Yoshida et al., 2007), and live imaging of hematopoietic stem and progenitor cells within the bone marrow identified anatomical locations involved in proper stem cell function during engraftment (Lo Celso et al., 2009; Xie et al., 2009). Therefore, live imaging has tremendous potential to enhance our understanding of stem cells, particularly when applied to tissues, such as the Drosophila gonads, where stem cells can be unambiguously distinguished from their differentiating daughters (Fuller, 1998; Fuller and Spradling, 2007).

Drosophila spermatogenesis is initiated at the testis apex where ∼10 germline stem cells (GSCs) are attached to a group of stromal cells called the hub (Fig. 1A). GSCs always divide with their mitotic spindles positioned perpendicularly with respect to the hub; tethering of the older (mother) centrosome at the GSC-hub interface ensures this. Thus, upon division, one GSC daughter remains adherent to the hub, and the other is displaced away from the hub, becoming a gonialblast (Hardy et al., 1979; Yamashita et al., 2003; Yamashita et al., 2007; Yamashita et al., 2010). Gonialblasts undergo four rounds of mitotic amplification with incomplete cytokinesis to produce clusters of interconnected spermatogonia that eventually form sperm (Fuller, 1998). This niche also maintains ∼20 somatic stem cells known as cyst stem cells (CySCs) that flank GSCs and are also anchored to the hub (Gonczy and DiNardo, 1996; Hardy et al., 1979). CySCs divide asymmetrically to self-renew and produce cyst cells, two of which envelop each gonialblast. As the gonialblast divides, its accompanying cyst cells enlarge and sustain encystment. Both GSCs and CySCs are maintained by Janus Kinase-Signal Transducer and Activator of Transcription (Jak-STAT) signaling initiated from the hub, which secretes the ligand Unpaired (Upd) (Issigonis et al., 2009; Kiger et al., 2001; Leatherman and DiNardo, 2008; Tulina and Matunis, 2001). CySCs also contribute to the niche, as Bone Morphogenetic Protein (BMP) signaling from CySCs is required for GSC self-renewal. BMPs are insufficient to generate ectopic GSCs, suggesting that additional signals function in this niche (Kawase et al., 2004; Leatherman and DiNardo, 2010; Schulz et al., 2004; Shivdasani and Ingham, 2003). Thus, the Drosophila testis provides a genetically tractable system in which to study mechanisms regulating stem cells in vivo.

Clonal analysis has shown that Drosophila GSCs turn over with an approximate half-life of 2 weeks, but are continually replaced such that GSC number is relatively constant during adulthood (Wallenfang et al., 2006). Replacement GSCs can be derived from four- or eight-cell spermatogonial clusters that undergo de-differentiation (Boyle et al., 2007; Cheng et al., 2008; Mair et al., 2010; McLeod et al., 2010; Wallenfang, 2007; Wallenfang et al., 2006). Whereas GSCs, gonialblasts or two-cell spermatogonial clusters may also regenerate GSCs, these events have not been detected thus far using fixed imaging. We developed long-term live imaging conditions to directly observe GSC behavior during both steady-state conditions and upon recovery from genetically induced stem cell loss (referred to here as `regeneration') in intact Drosophila testes. Our data reveal that GSCs in the Drosophila testis niche display various modes of stem cell renewal, which are regulated in vivo by altered signals during regeneration.

P{nos::eGFP-Moe::nos 3′UTR} flies were a gift from R. Lehmann (Skirball Institute, NY, USA) (Sano et al., 2005). Hs-bam flies contain the P[w⁺; Hsp70-bam⁺]^18d transgene inserted on the X chromosome (Ohlstein and McKearin, 1997). nanos-Gal4-VP16 flies (Van Doren et al., 1998) (from E. Selva, University of Delaware, DE, USA) were crossed to UAS-GFP-Moesin Actin-binding domain flies (Bloor and Kiehart, 2001; Dutta et al., 2002) (from D. Kiehart, Duke University, Durham, NC, USA) to drive expression of GMA in germ cells. All other stocks were acquired from the Bloomington Stock Center (Indiana). GSCs were depleted from flies containing the Hs-bam transgene as described previously (Sheng et al., 2009).

Immunostaining was performed as described (Matunis et al., 1997). Primary antibodies: rabbit anti-Vasa at 1:200 (Santa Cruz); rabbit anti-GFP at 1:10,000 (Torrey Pines Biolabs); mouse anti-1B1 at 1:25; and mouse anti-Armadillo at 1:50 (all from Developmental Studies Hybridoma Bank, University of Iowa). Alexa fluor-conjugated secondary IgG (H+L) antibodies were diluted 1:200 for 405, and 1:400 for 488 conjugates. Secondary antisera were: goat anti-rabbit 488 and goat anti-mouse 405 (Molecular Probes/Invitrogen). Nuclei were counterstained using 1 μg/ml 4′-6-diamidino-2-phenylindole (DAPI) (Roche Molecular Biochemical).

Dissected testes were incubated in Schneider's media containing 10 μM EdU (Invitrogen kit C10338) for 20 minutes. Immunostaining and thymidine analog detection were performed as described previously (Leatherman and DiNardo, 2008).

Testis culture conditions were adapted from other publications (Prasad et al., 2007; Prasad and Montell, 2007). Testes were dissected into and rinsed twice with Schneider's insect medium (Gibco/Invitrogen 11720034, pH adjusted to 7.0) supplemented with 15% FBS (Sigma F3018) and 0.5× penicillin/streptomycin (Gibco/Invitrogen cat. 15140-122). Testes were then rinsed twice with the same supplemented Schneider's media with the addition of MitoTracker Red CMXRos (M7512, Molecular Probes) at a final concentration of 1 μM and 0.2 mg/ml insulin (Sigma I5500). Rinsed testes were transferred with ∼20 μl media onto a 60 mm coverslip, covered with ∼1 cm² Teflon sheet (YSI Life Sciences #5793), and sealed with Halocarbon oil 27 (Sigma H8773). Live images were acquired using either an inverted Perkin Elmer UltraVIEW spinning-disk confocal, inverted Leica SD4 spinning-disk confocal or upright Zeiss 710 NLO multiphoton (JHU SOM Microscope Facility). Imaging parameters used are described in Table 1. Up to eight different positions per slide were acquired using a programmable xy stage. As the muscle sheath was intact, occasional movies were not analyzed due to contractions. Furthermore, a range of movie times were scored, as sample drift in the x, y or z planes occasionally occurred and resulted in early termination of the movie as the sample shifted away from the focal plane. The length of the movie and the type of microsocpe used did not affect tissue viability.

Fixed testes were mounted in Vectashield and viewed with Zeiss LSM 5 Pascal. z-series of live testes were analyzed using the Pascal, Ultraview, Slidebook, Zen, Bitplane's Imaris or NIH ImageJ software.

As each testis contains on average 12.3 GSCs, we calculated the total amount of time GSCs were imaged (12.3 GSC/testis × 197 hours) and divided by the total number of GSC divisions observed during that time (104) to derive a cell cycle time of 23.3 hours.

Live imaging of GSC division shows that mitosis takes between 40 minutes and 1 hour to complete. The % dividing GSCs is estimated by dividing the length of mitosis by the length of the cell cycle (24 hours), which is between 3% and 4%.

The numbers of mitotic germ cells of a given stage (GSCs, gonialblasts, two-, four- or eight-cell spermatogonia) were counted in both fixed and live testes of genotypes P{nos::EGFP-moe::nos 3′UTR} and Hs-bam; nanos-Gal4; UAS-GMA/+, then divided by the total number of mitotic GSCs, gonialblasts, two-, four- and eight-cell spermatogonia to obtain the percent of total mitoses. In fixed tissues, cells were scored as being in mitosis if they had high levels of GMA enrichment at the cell cortex and condensed chromosomes. In live testes, cells were scored as having undergone mitosis if they produced daughter cells (Fig. 1C-F, Fig. 2A-D; see Fig. S1A-D in the supplementary material).

Flies were raised on standard molasses-based food, transferred to food comprising 2.8% agar and 10% sucrose for 14 days, then returned to standard food.

Averages were compared using two-tailed Student's t-test assuming unequal variances or ANOVA. All graphs were constructed and statistics were confirmed using Graphpad's Prism or R software. Comparison of live imaged events between steady-state and regeneration was conducted assuming that the probability of a given number of events occurred during steady-state with a Poisson distribution (mean equal to the total number of events observed during steady-state). Comparison of percentages was conducted using Fisher Exact or Chi-Squared tests.

In order to observe dynamic cellular processes in real time within a stem cell niche, we developed extended 4D (x, y, z and t) imaging with the ability to resolve individual germline cells. We imaged live testes expressing the Actin-binding domain of Moesin fused to GFP (referred to as GMA) in the early germline (Dutta et al., 2002), which reveals germ cell morphology in both live embryos (Sano et al., 2005) and fixed testes (Sheng et al., 2009). In live testes, this marker is enriched at germ cell cortices, at the GSC-hub interface, and at cell-cell contacts within syncytial spermatogonia (Fig. 1C-F) in a manner indistinguishable from that seen in fixed testes (Sheng et al., 2009). Germ cells in mitosis also displayed similar GMA localization to those found in fixed tissues (Fig. 1C-F; see Fig. S1A-D in the supplementary material).

As previous live imaging of the Drosophila testis apex was on shorter timescales (∼2 hours) (Cheng et al., 2008), we measured germ cell proliferation rates in our 27 movies of young wild-type testes to determine whether our extended live imaging conditions compromised cell viability. The length of our movies ranged between 2.6 and 12 hours (average 7 hours), with z stacks collected at 20-minute intervals. In our combined 197 hours of imaging data, 104 GSC and 166 spermatogonial divisions were observed (Fig. 1B-F; see Fig. S1A-D in the supplementary material), indicating cells were viable and could proliferate given our imaging conditions. We calculate the cell cycle length of a single GSC to be ∼24 hours during live imaging; this is comparable with previous estimates of GSC cell cycle length (Wallenfang et al., 2006) and mitotic index (Yamashita et al., 2003). Consistent with this calculation, we never observed the same GSC divide twice in our movies. The proportion of dividing GSCs, gonialblasts and two- to eight-cell spermatogonia was similar between fixed and live imaged testes (Fig. 1B), and we did not observe significant changes in mitochondrial morphology during imaging (see Fig. S6A,B in the supplementary material), further indicating that imaging conditions did not compromise cell viability or proliferation.

GSCs in young flies are reported to always divide with spindles oriented perpendicular to the hub to produce one GSC still attached to the hub and one differentiating daughter cell displaced away from the hub, thus accomplishing asymmetric division (Cheng et al., 2008; Yamashita et al., 2003). We tracked all 104 GSC divisions until the end of each movie to determine the outcome of the division. With the exception of one case (see Fig. S2 in the supplementary material), every GSC that divided during live imaging (103/104) initially displaced a daughter cell away from the hub, confirming that this stereotypical GSC behavior occurs reliably in our live imaging conditions (Fig. 2A-D). Furthermore, daughter cells remained displaced from the hub in 95 out of 104 divisions for the remainder of the movie (see Movie 1, Fig. S7 and Table S2 in the supplementary material). Thus, asymmetric division is the predominant outcome following GSC divisions in vivo (Cheng et al., 2008; Yamashita et al., 2003) and during live imaging.

While tracking wild-type GSCs, we were surprised to see seven GSC-daughter pairs swivel such that the daughter cell initially distal to the hub changed its position and gained direct contact with the hub (Fig. 2E-H; see Movie 1, Fig. S7 and Table S2 in the supplementary material). This event has never been reported to occur in this system, and would not have been identified in fixed images. In all seven cases, both daughter cells were maintained next to the hub for the remainder of the movie (2-5 hours), suggesting that GSCs in the testis can produce two stem cells from one division. We refer to this phenomenon here as `symmetric renewal'. To determine whether both daughters acquire (or retain) GSC character, we analyzed the expression of the well-characterized GSC marker STAT92E (Boyle et al., 2007; Chen et al., 2002; Issigonis et al., 2009; Johansen et al., 2003; Leatherman and DiNardo, 2008; Read et al., 2004; Tulina and Matunis, 2001). STAT92E was detected in all GSCs (n=556) in fixed young wild-type testes (not shown). Thus, although GSC divisions are stereotypically oriented, daughter cells displaced away from the niche occasionally enter the niche and become stem cells. As we do not detect single gonialblasts or any type of spermatogonial cluster reverting into GSCs in 197 hours of imaging, our data strongly support the hypothesis that symmetric renewal of existing GSCs is the primary mode for GSC regeneration in young testes.

Previous clonal analysis in fixed testes demonstrated that GSC loss occurs in wild-type testes upon aging (Wallenfang et al., 2006). Using live imaging, we detected 16 cases of GSC loss in young wild-type testes, each characterized by a GSC-daughter pair progressively losing hub contact over ∼20-40 minutes time (Fig. 2I-L; see Movie 1, Fig. S7 and Table S2 in the supplementary material). GSC loss occurred at random times between 20 and 250 minutes (average 120 minutes) after the beginning of imaging, suggesting that the loss is not due to disruption of tissue integrity caused by imaging. Interestingly, we never observed GSCs without interconnected daughters losing hub contact, suggesting that GSC loss does not occur in the late G2 or M phases of the cell cycle (see Fig. S3 and Table S1 in the supplementary material). GSC-daughter pairs that detached from the hub stayed interconnected and appeared morphologically healthy for the remainder of the imaging time (from 2 to 9 hours), suggesting that differentiation rather than death occurred following loss. Consistent with this hypothesis, STAT92E was not detected in pairs of germ cells detached from the hub (n=44 testes), and apoptotic GSCs have never been detected in fixed wild-type testes (Boyle et al., 2007; Brawley and Matunis, 2004; Issigonis et al., 2009; Sheng et al., 2009). Together, these data suggest that stem cell loss during steady-state is via symmetric differentiation characterized by loss of adhesion of a mother/daughter pair followed by differentiation of both cells.

We next sought to confirm that symmetric renewal of GSCs occurs in vivo, and is not an artifact of live imaging conditions. As symmetric renewal is distinguished by the movement of a daughter cell into the niche while attached to a GSC, we looked for such pairs of cells in fixed tissues by incubating testes briefly (20 minutes) in media containing the thymidine analog EdU prior to fixation. Similar to wild-type female Drosophila GSCs (Carpenter, 1981; de Cuevas and Spradling, 1998), male GSCs and their daughter cells undergo S-phase while interconnected (Fig. 3A,B), and every EdU-positive GSC (n=361) was connected to a daughter. We then assessed the position of EdU-positive GSC-daughter pairs with respect to the hub. Ninety-five percent of EdU-positive GSCs were attached to daughters displaced away from the hub (Fig. 3A). However, 5% of EdU-positive GSC-daughter pairs were positioned such that both cells contacted the hub (Fig. 3B). These cells were morphologically similar to symmetrically renewing GSCs observed during live imaging (Fig. 2G,H), as both cells were enriched for GMA at the hub-germ cell interface. It is unlikely that these EdU-positive cells are two adjacent GSCs that coincidentally both labeled and both completed abscission of their daughter cells prematurely, as we did not detect single EdU-positive gonialblasts nearby (n=374 gonialblasts). Furthermore, pulse-chase BrdU/EdU labeling experiments (see Fig. S4 in the supplementary material) and fusome and Anillin morphology (see Fig. S3 in the supplementary material) revealed that GSC-daughter abscission occurs in G2, many hours after S phase (see Table S1 in the supplementary material). Finally, a spherical fusome was detected between the two cells, and EdU label intensity was similar within each pair, as expected if both cells were interconnected and entered S phase simultaneously (Fig. 3B, inset). As we did not detect movement of two-cell spermatogonial clusters into the niche during our live imaging, we do not believe these pairs of cells represent de-differentiating spermatogonia. Instead, they represent symmetrically renewing GSCs, which are in vivo products within this stem cell niche.

We next looked for signs of symmetric differentiation in fixed EdU-labeled testes. GSC-daughter pairs that have recently detached from the hub are indistinguishable from two-cell spermatogonia in fixed tissues. However, we infrequently saw examples (three cases in over 300 GSCs) of EdU-positive GSC-daughter pairs with fine protrusions projecting between adjacent GSCs to contact the hub (Fig. 3C). As these are fixed images, we cannot determine whether the pairs of cells are moving towards or away from the hub. However, in our live imaging, we saw only the latter. Thus, these examples most probably represent GSC-daughter pairs that are detaching from the hub in unperturbed wild-type tissues, resulting in symmetric differentiation.

Having devised an assay for identifying symmetric renewals in fixed tissue, we wanted to confirm that it behaved as expected in a genetic background known to increase the frequency of GSC renewal. Thus, we analyzed centrosomin (cnn) homozygous and trans-heterozygous mutants, where randomization of the division plane is thought to yield increased numbers of GSCs (Inaba et al., 2010; Yamashita et al., 2003). As expected, whereas the overall EdU labeling frequency remained similar to controls (∼30%), the percentage of GSC-daughter pairs where both cells touched the hub increased significantly from 5% in heterozygous controls to over 15% in the mutants (Fig. 3D,F), consistent with the finding that GSCs lacking Cnn can place daughter cells into the niche due to disruption of spindle orientation (Yamashita et al., 2003). In contrast to previously published findings, we did not see a significant increase in the number of GSCs per testis compared with sibling controls (Fig. 3E), suggesting that factors other than division orientation regulate overall GSC number.

We next asked whether changes in GSC division outcome could be detected in physiologically relevant circumstances such as poor nutrition. Protein starvation increases the rate of GSC loss in the Drosophila ovary and testis (Hsu and Drummond-Barbosa, 2009; Mair et al., 2010; McLeod et al., 2010). We confirmed that protein starvation reduces the number of GSCs compared with age-matched controls (8.6±1.8 versus 10.5±2.3, P<0.05, Fig. 3G,H). This effect is fully reversed by returning starved flies to standard food (Fig. 3G,H) (McLeod et al., 2010). As an increase in de-differentiation or misoriented centrosomes was not previously detected during GSC recovery (McLeod et al., 2010), we hypothesized that symmetric renewals were responsible for the modest increase in GSC number. Therefore, we determined the percentage of EdU-positive GSC-daughter pairs with both cells contacting the hub; this was significantly higher both after starvation (14%) and after 1 day of feeding (21%) compared with fully fed age-matched controls (3%), suggesting that nutrition levels affect the frequency of symmetric renewals (Fig. 3I). This increase in symmetric renewal was transient and returned to baseline after 5 days of feeding (6.5%). Together, our data show that the frequency of symmetric renewals responds to environmental and genetic perturbations.

In order to better understand regeneration in vivo, we made 26 movies (254 hours) of testes undergoing de-differentiation. Although de-differentiation occurs in wild-type testes, the probability of observing it in young flies is very low (Cheng et al., 2008). To capture de-differentiation events during live imaging, we used flies that had first been genetically depleted of GSCs and were in the process of recovering their stem cells. During recovery, high levels of spermatogonial dedifferentiation replenish lost GSCs (Brawley and Matunis, 2004; Sheng et al., 2009). In our assay, heat shock-induced ectopic expression of the differentiation factor Bag-of-Marbles (Bam) reduces the number of GSCs to less than one per testis, but testes regain wild-type numbers of GSCs after 4-5 days of recovery at the permissive temperature (Sheng et al., 2009). As our previous work suggested that de-differentiating spermatogonia were undergoing cellular rearrangements between 12 and 36 hours post heat-shocks (Sheng et al., 2009), we focused on these timepoints.

Fixed images suggested that spermatogonia, normally thought to be immobile, can acquire motility and move to make hub contact during dedifferentiation (Sheng et al., 2009). Our movies validated this hypothesis, as we saw 25 cases where spermatogonia initially not contacting the hub move in and established hub contact (Fig. 4A-C; see Fig. S7, Table S2 and Movie 2 in the supplementary material). Following contact, which occurred despite the presence of cyst stem cells occupying the positions directly next to the hub (see Fig. S6 in the supplementary material) (Sheng et al., 2009), GMA became enriched at the hub-GSC interface (Fig. 4B). Spermatogonia of most developmental stages were able to revert; we saw two gonialblasts, 14 two-cell, four four-cell and one eight-cell cluster establish hub contact during our movies. Before initiation of movement, the proximal edges of the reverting spermatogonia were between 2 and 11 μm (average 4 μm) away from the edge of the hub. Cells that moved to contact the hub were not always those closest to the hub (Fig. 4A-C, cells 3 and 4), indicating that proximity to the hub does not necessarily correlate with reversion. Our previous work showed that some spermatogonia had protrusions on their cell surface during de-differentiation (Sheng et al., 2009), and we observed similar protrusions on live spermatogonia (see Movie 2 in the supplementary material). However, protrusions only occasionally correlated with cell movement towards the hub, indicating they may reflect other cellular events such as changes in encystment by somatic cells. Live imaging has now directly demonstrated the ability of spermatogonia to move into the niche, gain hub contact and then resume divisions (Fig. 4D-F and Movie 3 in the supplementary material), thereby reverting to a less differentiated cell.

Live imaging also allowed us to visualize fragmentation of spermatogonial clusters, a process inferred to occur specifically during dedifferentiation (Brawley and Matunis, 2004; Cheng et al., 2008; Kai and Spradling, 2004; Sheng et al., 2009). In live wild-type testes, cytoplasmic stretching occurs only between the GSC and daughter gonialblast during abscission (see Fig. S1G-I in the supplementary material), and germ cells within a cluster of interconnected spermatogonia always divide synchronously (see Fig. S1A-D in the supplementary material). In testes undergoing dedifferentiation, we observed stretching of the cytoplasm between interconnected spermatogonia followed by mitoses of resulting smaller clusters at different timepoints. For example, two pairs of cells arise from a stretched four-cell spermatogonial cluster; one of the resulting pairs divided over 2 hours before the other, further indicating they were no longer interconnected (see Fig. S5 in the supplementary material). In 254 hours of imaging, we saw two four-cell, one eight-cell and one 16-cell cluster fragment into two-cell clusters, and one eight-cell fragment into two four-cell clusters. As we observed only five clear examples, fragmentation is either rare or our methods are not optimized for detecting it. Nonetheless, live imaging clearly shows spermatogonial clusters near the hub fragmenting into smaller clusters while entering the niche – hallmarks of de-differentiation that had never been directly visualized.

GSCs in the ovary are able to divide parallel to niche-generating cells after stem cell turnover (Xie and Spradling, 2000), and it has been proposed that male GSCs could also divide with a mitotic spindle parallel to the hub during GSC regeneration (Fuller and Spradling, 2007; Spradling, 2008). Therefore, we wanted to determine whether testes undergoing regeneration produce stem cells that repopulate the niche via symmetric renewals derived from divisions with a misoriented spindle. Surprisingly, even in testes undergoing regeneration, 91% (58/64) of germ cell mitoses at the hub produced a daughter cell that was initially displaced from the hub, indicating that misoriented divisions are not the primary method to generate additional stem cells during regeneration (see Fig. S7 in the supplementary material). However, the proportion of misoriented hub-contacting germ cell divisions in regenerating testes is slightly higher than wild type (6/64 versus 1/104, P≤0.05; see Movie 3 in the supplementary material), consistent with previous findings that GSCs derived from de-differentiation have higher incidence of misoriented centrosomes and division planes (Cheng et al., 2008).

The stem cell output of tissues in steady-state versus tissues undergoing regeneration is probably different in order to fulfill the cellular requirements of each situation. We compared the relative frequencies of events resulting in asymmetric division, symmetric renewal, symmetric differentiation and reversion during steady-state (total events=119) and regeneration (total events=102). During regeneration, we saw 47 cases of asymmetric outcome, 22 cases of symmetric renewals, 25 cases of spermatogonial reversion and 8 cases of loss (see Fig. S7, Table S2 in the supplementary material). Although the relative frequency of germ cell loss was not significantly changed during regeneration, the relative frequency of asymmetric division decreased by about one-half (Fig. 5; see Table S2 in the supplementary material). Furthermore, both symmetric renewal and reversion increased over threefold (Fig. 5; see Table S2 in the supplementary material). Thus, symmetric renewal and reversion are regulated processes, and the overall status of the tissue can direct stem cell behavior via changes in cell signaling.

Live imaging of the Drosophila germline stem cell niche has directly demonstrated many aspects of GSC behavior that were impossible to observe in fixed tissues. Asymmetrically oriented divisions do not necessarily determine asymmetric cell fate, but can occasionally result in the production of two GSCs. This is the primary mechanism by which GSCs are replenished in healthy tissues to compensate for GSC loss. As GSC-daughter pairs are adjacent to the hub and are enriched in the maintenance factor STAT92E (Boyle et al., 2007; Issigonis et al., 2009; Leatherman and DiNardo, 2008), the process of symmetric renewal is probably distinct from de-differentiation of spermatogonia (which are non hub-adherent and express the differentiation factor Bam). The frequency of symmetric renewal increases during GSC recovery after protein starvation, and during GSC regeneration after genetically induced stem cell depletion. In the latter case, where the rate of GSC regeneration is higher, GSCs are concurrently derived from de-differentiating spermatogonia, a process characterized by movement, fragmentation and adhesion to the hub by spermatogonial cells. Together, these data demonstrate that lost GSCs can be regenerated by multiple mechanisms, some or all of which may be similar to events occurring in other stem cell systems.

As changes in stem cell output are observed during regeneration, signaling from support cells or from systemic factors may underlie these effects. Niche-generating cells, transit amplifying daughter cells or even differentiated daughter cells may potentially signal to stem cells and modulate their division output. In the Drosophila testis, GSC maintenance depends on Jak-STAT signaling initiated from the hub, but it is not known whether this same pathway regulates division outcome. As STAT-null GSCs are rapidly lost from the niche (Brawley and Matunis, 2004; Kiger et al., 2001; Leatherman and DiNardo, 2010; Tulina and Matunis, 2001), low levels of Jak-STAT signaling due to fluctuations in gene expression may be sufficient to cause GSC loss. In support of this hypothesis, three out of 556 GSCs examined for STAT92E expression had low levels of this protein. However, the mRNA expression pattern of the Jak-STAT pathway ligand Upd is unchanged during de-differentiation (Sheng et al., 2009), suggesting that genes other than Upd may affect symmetric renewals. BMP signaling, which is required for GSC maintenance (Kawase et al., 2004; Leatherman and DiNardo, 2010; Schulz et al., 2004; Shivdasani and Ingham, 2003), is a good candidate. Combining live imaging with genetic tools for monitoring levels of signaling pathway activation in the Drosophila testis will provide a powerful platform for understanding how cell signaling affects the outcome of stem cell divisions in real time.

Our observation that both symmetric renewal and GSC loss occur when the GSC is attached to a daughter cell suggests that there may be a cell cycle-specific gene expression profile that primes the cells for these events to occur during S or early G2 in the cell cycle. We speculate that the abscission accompanying symmetric renewal is similar to that occurring in GSC-GB pairs, another G2 event. Cell cycle regulation, which is characterized by a short G1 phase and relatively long S phase, maintains pluripotency in many types of cultured stem cells (Singh and Dalton, 2009). As GSCs in the Drosophila testis have short G1 phases (see Table S1 in the supplementary material), and Drosophila GSCs require distinct cell cycle regulators (Wang and Lin, 2005), investigation of cell cycle regulation of Drosophila GSC division outcome may be informative.

We showed that GSCs in both centrosomin mutants and starved wild-type flies have increased frequencies of symmetric renewal, but surprisingly, there is no corresponding rise in GSC numbers. These results suggest that increased symmetric renewal is counterbalanced by increased GSC loss. Cnn mutant GSC are reported to have abnormal cell morphology and often appear to be detaching from the hub, suggesting an overall maintenance defect (Yamashita et al., 2007). During starvation, lowered insulin signaling results in GSCs loss, and this effect can be rescued by overactivation of insulin signaling (McLeod et al., 2010; Ueishi et al., 2009). Our results indicate that symmetric renewals of GSCs undergoing oriented divisions are the source of new GSCs. Starved flies initially have low insulin signaling, but when returned to normal food for a day have higher insulin signaling (McLeod et al., 2010). However, we find that both timepoints exhibited increased symmetric renewals, leading us to believe that activation of insulin signaling does not directly modulate division outcome. Perhaps during starvation, lowered insulin signaling causes GSC loss, which in turn triggers a compensatory increase in symmetric renewal. However, symmetric renewals are not able to fully compensate for the loss, yielding an overall decrease in GSC number. When flies are re-fed and insulin signaling returns to normal, GSCs are no longer rapidly lost, and the same rate of symmetric renewal is now able to increase overall GSC number. Together, our results suggest that the behavior of stem cells within the niche is much more dynamic than previously expected, and indicate that GSC number is controlled by the relative rates of symmetric renewal versus loss, not by the orientation of the division plane.

Why do the majority of Drosophila GSCs undergo asymmetric division if symmetric renewal plus symmetric differentiation produces the same output? As GSCs and CySCs function together within the niche during spermatogenesis, robust division orientation of both populations may enable differentiating germline cells to be generated at a rate that matches cyst cell production. Asymmetric divisions may also prevent clonal expansion of stem cells harboring harmful mutations within the niche, which can compete for niche occupancy (Issigonis et al., 2009; Johnston, 2009). However, clonal expansion may not always be harmful; mammalian niches regularly progress towards mono-clonality with stem cells exhibiting neutral drift dynamics (Klein et al., 2010; Lopez-Garcia et al., 2010; Snippert et al., 2010). Perhaps symmetrically renewing divisions are not detrimental to mammalian systems because mammalian niches are not as constrained spatially, and mammalian stem cells are often motile (Morrison and Spradling, 2008; Nakagawa et al., 2010; Yoshida et al., 2007). So far, asymmetric division in Drosophila testes correlates with optimal GSC function, as it becomes less robust with aging (Cheng et al., 2008). Whether symmetric divisions increase during aging has not been examined, but it might occur because GSCs are thought to be lost more frequently due to decreased maintenance cues (Boyle et al., 2007; Wallenfang, 2007). Interestingly, depleting STAT92E from GSCs displaces them from the hub, yet they are not lost from the tissue. Instead, they associate with BMP-producing CySCs, which probably promote GSC renewal. However, GSC division orientation is now randomized (Leatherman and DiNardo, 2010); suggesting that their output is composed of symmetric renewals and symmetric differentiation. Furthermore, APC2 mutants that affect centrosome position and E-cadherin mutants that have misoriented divisions still have wild-type GSC numbers (Inaba et al., 2010). Together, these observations suggest that the Drosophila testis stem cell niche does not require invariant asymmetric GSC division outcomes.

As mammalian stem cells are thought to undergo symmetric renewal in combination with stochastic differentiation, rather than strict asymmetric divisions, GSCs in Drosophila may share more aspects of stem cell behavior with mammalian systems than has been previously assumed. We observe wild-type GSCs losing niche attachment and directly differentiating, which is consistent with reports that subsets of undifferentiated spermatogonia in the mouse testes can directly differentiate (Nakagawa et al., 2010; Yoshida et al., 2006). Although Fig. 2E-L shows a lost GSC being replaced by a neighboring GSC undergoing symmetric renewal, this was our only example where these events are coupled together. Thus, stem cell loss and symmetric renewal may occur stochastically in Drosophila GSCs, as in the mouse testis (Klein et al., 2010). We also show that differentiating spermatogonia revert into GSCs, which is consistent with findings that differentiating spermatogonia can contribute to the stem cell pool during reconstitution of spermatogenesis in the mouse testes (Barroca et al., 2009; Nakagawa et al., 2007; Nakagawa et al., 2010). Therefore, our system provides an ideal platform for determining regulators of stem cell loss and replacement in vivo that may also be conserved in mammalian tissues.

We gratefully thank our colleagues who have supplied us with suggestions, stocks and technical assistance. We thank Dr Margaret de Cuevas for editing, Dr Geraldine Seydoux for providing comments on the manuscript, Drs Mohit Prasad and Denise Montell for imaging advice, Dr Melanie Issigonis for STAT29E immunostaining, and the Johns Hopkins University SOM Microscope Facility for technical assistance. This work was supported by NIH grants HD052937 and HD040307 (E.L.M.). Deposited in PMC for release after 12 months.