Piwi reduction in the aged niche eliminates germline stem cells via Toll-GSK3 signaling

Transposons are known to participate in tissue aging, but their effects on aged stem cells remain unclear. Here, we report that in the Drosophila ovarian germline stem cell (GSC) niche, aging-related reductions in expression of Piwi (a transposon silencer) derepress retrotransposons and cause GSC loss. Suppression of Piwi expression in the young niche mimics the aged niche, causing retrotransposon depression and coincident activation of Toll-mediated signaling, which promotes Glycogen synthase kinase 3 activity to degrade β-catenin. Disruption of β-catenin-E-cadherin-mediated GSC anchorage then results in GSC loss. Knocking down gypsy (a highly active retrotransposon) or toll, or inhibiting reverse transcription in the piwi-deficient niche, suppresses GSK3 activity and β-catenin degradation, restoring GSC-niche attachment. This retrotransposon-mediated impairment of aged stem cell maintenance may have relevance in many tissues, and could represent a viable therapeutic target for aging-related tissue degeneration. Retrotransposons are known to be involved in aging. Here the authors show that the retrotransposon silencer Piwi contributes to germline stem cell loss during aging in Drosophila.


Results
Aging reduces Piwi expression in the GSC niche. Our group and others have used the Drosophila ovary to study the decline of GSC number with age 12,13 . In this model system, GSCs are located at the anterior tip of the germarium and form direct contacts with cap cells (CpCs, the major GSC niche component), which are adjacent to the terminal filament (TF) and anterior escort cells (ECs) (Fig. 1a); together, these cells form the GSC niche 14 . A single GSC gives rise to a cystoblast, which later generates a functional oocyte 15 . Interestingly, we found that Piwi was highly expressed in every cell of the young germarium, except the TF 16 , but its expression was nearly absent in CpCs of aged flies ( Fig. 1b and Supplementary Fig. 1a-g). To examine the role of Piwi in adult CpCs, we individually used two independent UAS-RNAi lines (piwi RNAi-1 and piwi RNAi-2 ) with the nichespecific driver, bab1-GAL4 17 , under the control of temperature sensitive GAL80 ts18 . Flies expressing piwi RNAi were cultured at 18°C to suppress GAL4 activity during developmental stages, and were switched to 29°C after eclosion to degrade GAL80 ts and activate GAL4. Piwi protein expression remained in CpCs of 1week-old bab1 > piwi RNAi-1 and bab1 > piwi RNAi-2 germaria (data not shown), but was nearly absent in piwi-knockdown (KD) CpCs at 2 weeks as compared to controls (bab1 > gfp RNAi ) (Fig. 1c). We then examined the number of GSCs in newly eclosed (Day 1), 2-and 5-week-old bab1 > piwi RNAi flies by observing their anteriorly anchored fusome (a membranous cytoskeletal structure) adjacent to CpCs 19 . As previously reported 12,13 , the number of GSCs in control flies decreased with age ( Fig. 1d and Supplementary Table 1). This decrease was accelerated in piwi-KD lines ( Fig. 1d and Supplementary Table 1). Compared to newly eclosed germaria, 83% of GSCs remained in 2-week-old control germaria (n = 108); however, only 66 and 68% of GSCs were found in 2-week-old bab1 > piwi RNAi-1 (n = 94; P < 0.001) and bab1 > piwi RNAi-2 germaria (n = 101; P < 0.001), respectively. Although newly eclosed bab1 > piwi RNAi-2 flies carried higher baseline numbers of GSCs compared to bab1 > piwi RNAi-1 flies and controls (probably due to their different genetic backgrounds), the rate of GSC loss in bab1 > piwi RNAi-2 was comparable to bab1 > piwi RNAi - Fig. 1 Aging reduces Piwi expression in the niche leading to GSC loss. a The anterior region of Drosophila germarium. TF terminal filament, CpC cap cell, E escort cell, GSC germline stem cell, CB cystoblast. b Piwi expression is decreased in CpCs during aging (green, Piwi; blue, Vasa for germ cells; red, Tj for nuclei of CpCs and ECs). Scale bar, 5 μm. Scale bar in inset image, 2.5 μm. c Piwi expression is decreased in 2-week-old CpCs of bab1 > piwi RNAi germaria (green, Piwi; blue, Vasa; red, Tj). Arrowheads, ECs. d Left, GSC number is decreased in 2-week-old bab1 > piwi RNAi-1 germaria (red, LamC for CpC nuclear envelopes; red, Hts for fusomes; blue, DAPI for DNA). Dashed white circles, CpCs; yellow circles, GSCs. Right, GSC number in bab1 > gfp RNAi-1 , bab1 > piwi RNAi-1 , and bab1 > piwi RNAi-2 germaria at 1 day (D1), 2 weeks (2W), and 5 weeks (5W) after eclosion. The numbers of analyzed germaria are shown above each bar. Percentage (%) of remaining GSCs (left y-axis) represents average GSC number at each time-point normalized to D1 within the same genotype; % of germaria (right y-axis) indicates the proportion of germaria carrying 0, 1, 2, 3, or 4 GSCs for the indicated genotype and age. Error bars of left y-axis, S.E.M., ***P < 0.001 (0.00011 for 2W piwi RNAi-1 and 0.0008 for 5W piwi RNAi-1 ); ****P < 0.0001, yellow and green outlines on symbols indicate that the % of remaining GSCs is significantly lower than D0 of the same genotype. Green outline on symbols indicates that the % of remaining GSCs is significantly lower than symbols with yellow outline within same genotype. Student's t-test was used for comparisons of % remaining GSCs. All images (b, c, and d) are the same magnification and share same scale bar with b. All inset images (b, c) are the same magnification and share same scale bar with insert image of b. At least two biological replicates were performed for each experiment. than control flies. This 32-34% loss of GSCs from D1 to 2 weeks in bab1 > piwi RNAi flies was similar to the loss observed in control flies from D1 to 5 weeks (Fig. 1d), indicating that depletion of Piwi in the niche accelerated the GSC aging phenotype by 3 weeks. From these data, we conclude that Piwi expression in the young niche maintains GSC number.
Niche Piwi silences transposons to maintain GSCs. We next tested if Piwi silences transposons in the niche. We first monitored transposon activity in the germaria with gypsy-lacZ, a transcriptional reporter of the gypsy retrotransposon 20 , which is among the most active endogenous insect retroviruses [21][22][23] and is known to be silenced by a Piwi-piRNA complex 20 . The gypsy-lacZ reporter was not expressed in the 1-week-old germarium but was strongly expressed in CpCs of 5-week-old flies, indicating derepression of transposons in the aged niche ( Fig. 2a and Supplementary Fig. 1h, i). As expected, gypsy-lacZ expression was present in CpCs of bab1 > piwi RNAi-1 but not in control flies at 2 weeks after eclosion (Fig. 2b). Consistently, gypsy transcripts, examined by in situ hybridization, were also highly increased in 2week-old piwi-KD CpCs ( Supplementary Fig. 2). Interestingly, gypsy transcripts were also highly abundant in control and piwi-KD TFs ( Supplementary Fig. 2), where Piwi was not expressed. We noticed that in 2-week-old bab1 > piwi RNAi-1 germaria, anterior ECs also showed reduced Piwi expression (see Fig. 1c, arrowhead) but did not express gypsy-lacZ (Fig. 2b). In addition, knocking down Piwi specifically in adult ECs did not cause GSC reduction (Supplementary Fig. 3 and Supplementary Table 1), supporting our previous conclusion that Piwi in CpCs maintains GSCs.
To examine which retrotransposons are suppressed by Piwi, we isolated CpCs (GFP-positive) from bab1 > piwi RNAi-1 gfp and bab1 > gfp flies for RNA sequencing ( Supplementary Fig. 4). Notably, TF cells were isolated with the CpCs, as they also expressed GFP. However, TF cells do not express Piwi, or only express it at very low levels 24 (see Supplementary Fig. 1e). Interestingly, a number of retrotransposons were already expressed in control niche cells (Fig. 2c). For instance, copia accounted for 27% of the total reads (164014 reads per kilobase per million mapped reads, RPKM; Supplementary Table 2). In piwi-KD niche cells, total retrotransposon transcripts were increased (214625 RPKM), with copia transcripts remaining very high (Fig. 2c and Supplementary Table 2). Increases in the expression levels of ZAM (47-fold), mdg1 (10-fold), blood (10fold) and gypsy (1.26-fold) were observed in piwi-KD compared to control niche cells (Fig. 2c). This moderate increase of gypsy transcripts in isolated piwi-KD niche cells could be due to the presence of TF cells, which expressed high levels of gypsy transcripts (See Supplementary Fig. 2). Therefore, it appears that Piwi preferentially silences a subset of retrotransposons in the niche.
To determine whether the increase of retrotransposons in piwi-KD CpCs caused GSC loss, we suppressed reverse transcription of replicating retrotransposons by feeding bab1 > gfp RNAi-1 and bab1 > piwi RNAi flies with (−)-L-2′,3′-dideoxy-3′-thiacytidine (3TC), a cytidine analog that is clinically used to inhibit reverse transcription of human immunodeficiency virus and Hepatitis B virus for suppressing viral replication mediated by complementary DNA (cDNA) generation 25,26 . Consistent with our previous experiment, GSC maintenance in 2-week-old bab1 > piwi RNAi flies was significantly decreased as compared to controls (Fig. 2d,  e and Supplementary Table 1), although piwi RNAi-2 expression had a milder effect on GSC loss (26%) in this experiment ( Fig. 2e and see Fig. 1d). Two-week-old control flies with or without 3TC treatment after eclosion showed comparable GSC numbers, while GSC loss in 2-week-old bab1 > piwi RNAi flies was restored to the control level upon feeding with 3TC for 2 weeks after eclosion (Fig. 2d, e and Supplementary Table 1).
To test the role of elevated retrotransposons in piwi-KD niche cells for GSC maintenance, we used the only two publicly available RNAi lines to suppress gypsy and copia in the piwi-KD niche. Strikingly, knocking down the retrotransposon gypsy, but not copia, in the piwi-KD niche restored GSC number to the control level (Fig. 2e, and Supplementary Table 1), suggesting the retrotransposon type-specific effect on GSC maintenance. Thus, Piwi-mediated silencing of retrotransposons in the niche, and inhibition of gypsy transposons in particular, serves to maintain GSCs. Interestingly, rescue by knockdown of gypsy retrotransposon was also found with regard to the neurodegeneration phenotype in a fly TDP-43 model 21,27 . The relative importance of this particular retrovirus may be due to the facts that the gypsy retrotransposon is highly active, encodes an endogenous retrovirus, and its replication can generate de novo insertions in aged fly brain 9 . Overall, we show that gypsy upregulation is at least largely responsible for the GSC-loss phenotype, but our results do not completely rule out roles for other retrotransposons on the GSC-loss phenotype.
Niche Piwi supports GSC anchorage. Previous studies reported that transposon jumping leads to cell death via DNA damageinduced Chk2/p53 signaling pathway 21,[28][29][30][31] . Although approximately 30% of GSCs were lost in 2-week-old piwi-KD flies, CpC number was not affected ( Supplementary Fig. 5a). In addition, DNA damage, as revealed by γ-H2Av 32 , was observed in the CpCs of 30% of 5-week-old germaria (n = 120) accompanied by a reduction of CpC number ( Supplementary Fig. 5a-c). However, removing one copy of p53 or co-knockdown of loki (encoding fly Chk2) in the piwi-KD niche did not restore GSC numbers (Supplementary Fig. 5d and Supplementary Table 1). These results suggest that niche function, but not niche cell survival, is disrupted when retrotransposons are activated.
To maintain GSCs, CpCs provide both Dpp stemness signals 33 and physical contact via cell-cell adhesion, which is mediated by trans-interaction between extracellular domains of E-cadherin 34 and further strengthened by the cytoplasmic interaction of Cadherin-Catenin-Actin 35 . We did not observe decreased dpp mRNA production in the niche, nor did we find decreased Dpp signaling in the GSCs of bab1 > piwi RNAi ovaries ( Supplementary  Fig. 6). Therefore, we next asked whether the piwi-KD niche exhibits defective GSC anchoring by examining E-cadherin, Armadillo (Arm, the β-catenin orthologue in fly) and filamentous (F)-actin levels. In bab1 > piwi RNAi germaria, total expression of E-cadherin was the same as control in CpCs (Fig. 3a, f), however, its expression was enriched in CpC-CpC junctions but dramatically decreased in CpC-GSC junctions (Fig. 3a, g). In contrast, Arm expression in bab1 > piwi RNAi germaria was reduced in CpCs (including CpC-GSC junctions) but remained relatively high in the junction between CpCs (Fig. 3b, h, i). This localization of Arm may explain the enrichment of E-cadherin in CpC-CpC junction of the piwi-KD niche. Arm reduction was also found in flies with CpCs carrying homozygous mutations of piwi 1 or piwi 2 (null and hypomorphic allele, respectively) 36 , induced by mitotic recombination 37 (Supplementary Fig. 7). Consequently, F-actin expression was also decreased in the junction between piwi-KD CpCs and GSCs ( Supplementary Fig. 8). We further showed that Arm expression in the niche was necessary for E-cadherin accumulation in the CpC-GSC junction for GSC anchorage ( Supplementary Fig. 9a-c); during aging, Arm and F-actin expression levels were decreased in the CpC-GSC junction (Supplementary Fig. 9e-h), along with the decreased Ecadherin 13 . These results suggest that Arm affects E-cadherin expression and cellular localization. Thus, we conclude that Piwi in the niche promotes Arm expression to support E-cadherinmediated niche-GSC adhesion.
Retrotransposons impair niche-GSC anchorage via GSK3. Notably, arm mRNA expression was not decreased in piwi-KD CpCs (see Supplementary Fig. 3e), raising the possibility that Piwi may control β-catenin stability via Glycogen synthase kinase 3 (GSK3), which phosphorylates β-catenin for degradation 38 . Indeed, suppressing the expression of Shaggy (Sgg, the GSK3 orthologue in fly) or overexpressing a constitutively active form of Arm (Arm S10 , which cannot be phosphorylated by GSK3 39,40 ) in piwi-KD CpCs rescued Arm expression (Fig. 3d, h) Fig. 1d. Note that GSC number of piwi RNAi-2 alone at D1 in 7th and 10th columns is the same data shown in Fig. 1d. % of germaria (Right y-axis) indicates the proportion of 2-weekold germaria carrying 0, 1, 2, 3, or 4 GSCs. Error bars of left-y-axis in e, S.E.M. n.s., non-significant. Yellow outline on symbols indicates that the % of remaining GSCs is significantly lower than D0 of the same genotype. Student's t-test was used for comparisons of % remaining GSCs. All images (a, b, and d) are the same magnification and share scale bar with a. All inset images (a) are the same magnification. At least two biological replicates were performed for each experiment.  Retrotransposons increase GSK3 activity to impair niche-GSC anchorage. a, c E-cadherin expression was evaluated at the niche-GSC junction (arrows) of 2-week (W)-old bab1 > piwi RNAi-1 , bab1 > piwi RNAi-1 sgg RNAi , and bab1 > piwi RNAi-1 arm S10 germaria (green, E-cadherin; red , LamC for CpC nuclear envelopes; red, Hts for fusomes). Scale bars in left and right images are 5 and 2.5 μm, respectively. b, d, e Arm expression in the niche-GSC junction of 2week-old bab1 > piwi RNAi-1 , bab1 > piwi RNAi-2 , bab1 > piwi RNAi-1 ;sgg RNAi , bab1 > piwi RNAi-2 ;gypsy RNAi , and germaria of bab1 > piwi RNAi-1 with 10 mM 3TC treatment (green, Arm; red, LamC and Hts). Same magnification of images (a-e, left), and of enlarged images (a-e, right) share the left and right the scale bar in a, respectively. f, g Relative expression of E-cadherin in the niche (f) or niche-GSC junction (g) of germaria with indicated genotypes. h, i Relative expression of Arm in the niche-GSC junction (h) or niche (i) in the germarium of indicated genotypes. j Relative percentage (%) of GSC number in 2-weekold germaria of indicated genotypes (normalized to GSC number at D1). Note that GSC number of gfp RNAi alone are the same data as shown in Fig. 1d. % of germaria (Right y-axis) indicates the proportion of 2-week-old germaria carrying 0, 1, 2, and 3 or 4 GSCs. k pSer9-GSK3β expression in the niche of 2week-old bab1 > gfp RNAi and bab1 > piwi RNAi-1 germaria with or without 10 mM 3TC treatment (green, pSer9-GSK3β; red, F-actin for CpC plasma membranes). Scale bars, 5 μm; all inset images in k are same magnification (scale bar, 2.5 μm). Dashed line, the anterior edge of the germarium (a-e); asterisks, CpCs; dashed circles outline niche CpCs (k). Box-Whiskers plots (f, g, h, i, and k) show the central lines for median values, edges of box represent upper (75th) and lower (25th) quartiles and whiskers show minimum and maximum values. Numbers of analyzed niches/niche-GSC junctions and GSCs are shown above or within each bar (f-k). Error bars, S.E.M. in j. Statistical analysis, Student's t-test for f and j; One-way ANOVA for g, h, i, and k. ****P < 0.0001. At least two biological replicates were performed for each experiment. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-16858-6 ARTICLE NATURE COMMUNICATIONS | (2020) 11:3147 | https://doi.org/10.1038/s41467-020-16858-6 | www.nature.com/naturecommunications localization in the CpC-GSC junction (Fig. 3c, g), and GSC number (Fig. 3j). Further, 3TC treatment or co-knockdown of gypsy retrotransposon in bab1 > piwi RNAi flies also restored Arm expression in piwi-KD CpCs (Fig. 3e, h). These results suggest that retrotransposons cause Arm degradation via GSK3, which is highly regulated by its phosphorylation status 41 .
Phosphorylation of GSK3 at Tyr216 (pTyr216) is an activating event, while Ser9 phosphorylation (pSer9) is inhibitory. In piwi-KD CpCs, expression of pSer9-GSK3 was reduced (Fig. 3k), but not pTyr216-GSK3 or total GSK3 ( Supplementary Fig. 10), and this reduction was prevented by 3TC treatment (Fig. 3k). We also found that pSer9-GSK3 levels in CpCs were decreased in 7-weekold control flies ( Supplementary Fig. 11a, b), and overexpressing a constitutively active form of GSK3 (Sgg S9A ) in the niche reduced Arm and E-cadherin expression, as well as GSC number (Supplementary Fig. 11c-g). Therefore, Piwi in CpCs maintains expression of Arm by suppressing GSK3 activity via inhibitory phosphorylation.
Piwi supplementation in the aged niche slows GSC loss. Since constitutively overexpressing piwi in CpCs of adult flies delayed age-dependent GSC loss ( Supplementary Fig. 12), we proceeded to ask if GSC loss mediated by niche Piwi deficiency can be reversed or slowed down by intervention within a specific timewindow during aging and whether the Piwi-Arm regulatory axis is involved in age-dependent GSC loss. To address these questions, we supplied Piwi or Arm expression to the aged niche by utilizing a niche GeneSwitch GAL4, 2261 42 , which can be activated by the steroid RU486 (mifepristone) 43,44 . Of note, this driver shows some basal leaky activity in the absence of RU486 42,45 . Overexpressing either Piwi or Arm S10 in 3-or 6week-old niches by feeding flies with RU486 for 1-week delayed GSC loss ( Supplementary Fig. 13a-f and Supplementary Table 1). Consistent with these results, overexpressing Piwi in the 6-weekold niche for 1 week also partially rescued Arm expression and Ecadherin distribution in the CpC-GSC junction, although total Ecadherin expression levels remained low ( Supplementary  Fig. 13g-l). These results suggested that Piwi-independent regulation of E-cadherin expression levels and GSC loss occurs during aging, and together, the findings reflect the critical involvement of transposon activation in the aged niche.
Toll-GSK3 signaling in Piwi-deficient niche causes GSC loss. Replicated retrotransposons may integrate into new loci within the genome and affect expression of genes that mediate inhibitory phosphorylation of GSK3. To test whether this occurs, we isolated GFP-positive piwi-KD niche cells (CpCs, anterior ECs and TFs) and other ovarian non-niche somatic cells (GFP-negative) from 2-week-old bab1 > gfp & piwi RNAi-1 germaria for whole genome sequencing ( Supplementary Fig. 14a). We found 70 new transposon insertion sites in the genome of the piwi-KD niche compared to the control genome; 53 insertion events were associated with 22 different retrotransposons (Supplementary Table 3). However, none of these new insertions were located in genes known to regulate inhibitory phosphorylation of GSK3. Surprisingly, we did not find any new insertions of the gypsy retrotransposon. We also did not identify any de novo gypsy transposition events in the aged CpCs using the gypsy-TRAP reporter system 9 , which contains Ovo binding hotspots for gypsy insertion ( Supplementary Fig. 14b-d). These results suggest that GSK3 activation in the piwi-KD niche is likely mediated by a retrotransposition (mediated by retrotransposon cDNA)-independent mechanism.
Like the Drosophila gypsy retrotransposon, the human retrotransposon Human endogenous retrovirus K (HERV-K) can generate endogenous virus 46 , which is present in some patients with amyotrophic lateral sclerosis 47 . HERV is known to induce immune response via Toll-like receptor (TLR) signaling 48,49 , which activates GSK3 by reducing inhibitory phosphorylation, leading to β-catenin degradation in enterocytes of mice 50 . It is therefore possible that retrotransposons may induce Tollmediated signaling in the piwi-KD niche to activate GSK3 and subsequently cause GSC loss. Therefore, we next asked if Toll signaling is involved in the transposon-GSK3 axis that impairs GSC maintenance. In the fly, Toll and Toll-7 signaling may be activated by virus infection 51,52 . Moreover, Toll-5 is known to interact with Toll and is involved in Toll signaling activation 53 . We found that depletion of Toll or Toll-5, but not Toll-7, in piwi-KD CpCs rescued GSC loss (Fig. 4a, b and Supplementary Table 1) and knocking down Toll or Toll-5 alone in the CpCs did not cause increased maintenance of GSCs (Supplementary Table 1). Decreased levels of pSer9-GSK3 and Arm, and mislocalization of E-cadherin in the piwi-KD CpCs were also rescued by depletion of Toll or gypsy retrotransposon (Fig. 4b, c and Supplementary Fig. 15).
Niche Piwi inhibits Viral-like particles and Toll signaling. Upon activation of Toll-mediated immune signaling, Cactus (orthologue of mammalian IκB 54 ) is phosphorylated and degraded in the cytoplasm to release Dorsal (orthologue of mammalian NF-κB 54 ). Dorsal then enters the nucleus, where it activates transcription of genes encoding antimicrobial peptides (AMPs) 55 . To test whether this immune signaling pathway is activated in piwi-KD CpCs, we examined expression of Cactus and Dorsal. We found that Cactus and Dorsal were highly expressed in the cytoplasm of bab1 > gfp RNAi CpCs (Fig. 4d), while both Cactus and Dorsal expressions were dramatically reduced in the cytoplasm of aged CpCs, piwi-KD CpCs, and CpCs that overexpressed a constitutively active form of Toll, [Toll 10b ] 56 (Fig. 4d). Notably, we did not observe nuclear location of Dorsal, nor did we find increases of known AMPs in isolated piwi-KD CpCs by transcriptome analysis. This lack of AMP signals may have been due to amplification bias in the cDNA from isolated CpCs, our examination of unregulated AMPs, or quick removal of Dorsal from the nucleus. Furthermore, knockdown of gypsy or toll in the piwi-KD niche or 3TC treatment of bab1 > piwi RNAi-1 flies did not reverse Cactus degradation (Fig. 4f), suggesting that other Toll receptors may be involved in the activation of immune signaling, and this process may be independent of the effects of gypsy and retrotransposition. More interestingly, we observed the presence of viral-like particles (VLPs) by transmission electron microscopy in the cytoplasm of many piwi-KD (12/37 = 32% of germaria from ten ovaries) and aged CpCs (8/36 = 22% of germaria from ten ovaries), but VLPs were not observed in control CpCs (0/22 germaria from 10 ovaries) (Fig. 5a). Meanwhile, 3TC treatment reduced the number of VLPs in piwi-KD CpCs (Fig. 5b, c), suggesting that retrotransposon replication can increase transposon transcripts for VLP assembly. However, VLPs are not reduced in the piwi/gypsy-double KD CpCs (Fig. 5a); this result is consistent with our finding that gypsy did not undergo retrotransposition and suggests that VLPs may be derived from nongypsy retrotransposons.
According to the results of our genetic and 3TC treatment experiments (Fig. 6a), we predicted that activation of Toll signaling in the aged or piwi-KD niches would both promote GSK3 activity and degrade Cactus. However, we found that knockdown of gypsy, or inhibition of reverse transcription for virus cDNA generation in the piwi-KD niche suppressed GSK3 activity but did not restore Cactus levels nor did it completely clear VLPs. We therefore speculate that retrotransposons other than gypsy may contribute to VLP generation and Cactus degradation. Furthermore, co-knockdown of piwi and toll in the niche cannot prevent Cactus degradation, suggesting that other Toll receptors may be involved in Toll-Cactus signaling. However, we cannot rule out the possibility of Piwi-dependent, Toll-independent Cactus expression. Nevertheless, our results suggest that in the aged niche, decreased Piwi expression results in generation of retrotransposon-derived viral materials (such as viral transcript, cDNA, or protein) and coincident activation of Toll signaling, which acts through GSK3 to degrade Arm (Fig. 6b). This reduction in Arm disrupts Cadherin-Catenin-Actin adhesion, compromising GSC maintenance.

Discussion
It is possible that this pathway for removal of GSCs from the aged or piwi-KD niche would protect the integrity (avoid retrotransposon integration) of genomic material that is transmitted to    Fig. 16). Hence, Piwi may play a similar role in the GSC niche of both sexes, but the individual retrotransposons that are suppressed by Piwi may differ between males and females. Intriguingly, flies carrying a mutation of piwi lacking a nuclear localization signal (Piwi NT ) were reported to display elevated gypsy expression and normal GSC number 57 , which appears to contradict our findings. However, gypsy expression in piwi NT mutant ovaries is only half as abundant as it is in piwi 2 mutant ovaries 57 . In addition, piwi null mutant ovaries are extremely small, while piwi NT mutant ovaries display normal morphology and carry egg chambers, each with a monolayer of follicle cells, where gypsy is expressed when Piwi is deleted 58 . Thus, the reported elevation of gypsy is likely to partly rely on expression by follicle cells, which are far away from GSCs. In addition, Piwi protein contains a Piwi domain, which has been shown to cleave RNA in vitro 59 . Based on this activity, cytoplasmic Piwi might possibly participate in degrading retrotransposons, in addition to its well-characterized role in suppressing retrotransposon transcription.
It is not clear how retrotransposons, retrovirus or virus materials (including transcripts, cDNA made by reverse transcription and/or proteins) in the cytoplasm are detected by Toll receptors in CpCs. Using the Drosophila RNAi Screening Center Integrative Ortholog Prediction Tool (DIOPT) 60 , we found that Toll and Toll-5 are the respective orthologues of human TLR7/9 and TLR4 (Supplementary Table 4), which sense viral materials 61 . TRL7/9 are present and active in endosomes within the cytoplasm to sense viral materials 61 ; similarly, endosomal location of Toll has been reported, and its activation requires the endocytic pathway 62 , suggesting that Toll may also participate in virus sensing. It is not clear if Toll-5 senses viral materials. Nevertheless, it is known to dimerize with Toll, and it is required for Toll-mediated immune activation 53 . Another open question is why aging only reduces Piwi expression in the GSC niche. Given that many types of mammalian stem cell niches lose their function during aging 63,64 , it would be interesting to know if transposon activation may broadly account for this decline.
Inflammation is associated with aging-related disease 65 . A very recent study showed that aging-induced elevation of LINE1 retrotransposons triggers inflammation in mice, and 3TC treatment reduces this inflammation 66 . Together with this previous study, our results may shed light on the long-known association among retrotransposons, inflammation/innate immunity, and Alzheimer's disease (AD) 67,68 . The pathogenesis of AD is known to involve increased GSK3 activity as a result of Aβ42 accumulation 69 , although the underlying mechanisms are still not clear. Our study raises the possibility that retrotransposon-immune signaling-GSK3 regulation observed in the fly GSC-niche unit may also occur in mammalian neural cells. Indeed, GSK3 activity in an AD model in human embryonic kidney 293 cells was suppressed by 3TC treatment (Supplementary Fig. 17). Moreover, in an AD fly model with human Aβ42 overexpression in the eye, GSK3 activity was suppressed and neural activity was partially rescued by 3TC treatment (Supplementary Fig. 18). In addition to AD, several other aging-related diseases were reported to exhibit both derepressed retrotransposons and elevated GSK3 activity, including multiple sclerosis 70,71 , amyotrophic lateral sclerosis 72,73 , and cancer 74,75 . Given that GSK3 plays a crucial role in inflammation 76 , our results not only show that retrotransposon derepression impairs stem cell maintenance via   Toll-GSK3 signaling but also highlight the possibility that retrotransposon reactivation may contribute to aging-related tissue dysfunction and diseases via immune signaling-mediated GSK3 activation.

Methods
Fly stocks and culture. Drosophila stocks were maintained at 22-25°C on standard medium (normal diet). yw, w 1118 , Canton S, and Oregon were used as wildtype controls. Null p53 -ns (a gift from Dr. John M. Abrams, University of Texas Southwestern Medical Center, U.S.A.) alleles have been described previously 57,77,78 . myc-piwi, a genomic construct with a Myc-tag inserted between Piwi amino acid residues 3 and 4, was used to monitor Piwi expression 16 84 . The bab1-GAL4, ptc-GAL4 and c587-GAL4 lines have been previously described 90,91 . 2261-GAL4 GeneSwitch (GS) is a driver that activates expression of UAS transgenes in TF and CpCs upon the binding of RU486 (mifepristone) to the progesterone receptor-GAL4 chimera 42,44 . Other genetic elements are described in Flybase (http://flybase.org). Flies expressing RNAi or transgenes driven by bab1-GAL4 or ptc-GAL4 also carried a temperature sensitive (ts) mutant GAL80 under the control of a tubulin promoter (tub-GAL80 ts ) and were cultured at 18°C to suppress GAL4 activity during developmental stages; the flies were switched to 29°C after eclosion to degrade GAL80 ts and activate GAL4. For experiments, 10-15 females plus three wild-type male flies were maintained in a plastic vial and fed with normal diet plus dry yeast powder. For RNAi knockdown experiments, food (normal food with dry yeast powder) was changed every other day.
For the experiments involving co-knockdown of piwi and gypsy, gypsy RNAi and piwi RNAi-2 were used together, as these transgenes are respectively located on the second and third chromosomes, allowing for efficient of the presence of both RNAi lines.
Genetic mosaic analysis. Genetic mosaic clones were produced by Flipase (FLP)/ FLP recognition target (FRT)-mediated mitotic recombination 37   Pharmacological treatments. Flies were cultured at 25°C before RU486 treatment and maintained as described above; at the desired age, flies were switched to vials containing normal diet plus a wet yeast paste mixed with RU486 (M8046, Sigma). A 10 mg/mL stock solution of RU486 (mifepristone; Sigma) was made in pure ethanol (Sigma). For RU486-containing wet yeast preparation, 0.28 mL of RU486 stock solution, 0.1% blue food color additive (to confirm food intake; #861146, Sigma) and 1.52 mL ddH 2 O were mixed well and then added to 1 g active dry yeast (RED STAR) for a final concentration of 3.6 mM RU486. For 3TC (2′, 3′-Dideoxy-3′-thiacytidine, #L1295, Sigma) treatment, newly eclosed flies were cultured in plastic vials containing normal diet plus a 3TCcontaining wet yeast paste. For 3TC-containing wet yeast preparation, 11.5 mg 3TC, 0.1% blue food color and 3.3 mL ddH 2 O was mixed well and then added to 1.7 g active dry yeast for a final concentration of 10 mM 3TC. Control flies were fed with the same diet but without RU486 or 3TC. Food was changed daily until flies were dissected. Image analysis and quantification. GSCs were identified by the presence of a fusome (labeled with anti-Hts antibody), which is an intracellular structure adjacent to CpCs. CpCs contain ovoid nuclear envelopes that were stained by LamC. Rhabodomere numbers in each Ommatidium were analyzed by Fiji. Regions of interest for analyzing protein expression were outlined manually. To analyze the levels of Piwi, gypsy-lacZ, pSer9-GSK3, pTyr216-GSK3, total GSK3 and pMad in germarial cells, Image J (National Institutes of Health [NIH], U.S.A.) was used to measure the mean fluorescent intensity (arbitrary units) of the confocal z-section with the highest fluorescent signal from each cell type. The protein levels were normalized to the indicated proteins (used as a staining control) in the same cell type, or in other germarial cells (Supplementary Fig. 19). Tj in the same cell type was used for normalization of Piwi and gypsy-lacZ in CpCs, and Piwi in ECs. Expression of Piwi in GSCs was normalized to Vasa expression in GSCs. For pSer9-GSK3, pTyr216-GSK3 and total GSK3 levels in CpCs, the mean fluorescent intensity was normalized to the level in TFs. In RU486-treatment experiments, Piwi expression in CpCs was normalized to Piwi expression in follicle cells. To analyze the total amounts of E-cadherin and Arm expressed in the niche, confocal z-stacks of CpCs were merged. The mean fluorescence intensity of the merged image was measured with ImageJ (version 1.50i) and normalized to the mean fluorescence intensity for E-cadherin or Arm in a germ cell of a 16-cell cyst (which is derived from a CB) in the same germarium. To analyze E-cadherin, Arm or F-actin expression in the niche-GSC junction, the confocal z-section with the highest signal of E-cadherin, Arm or F-actin in the niche-GSC junction was analyzed and normalized to the signal for E-cadherin or Arm in a germ cell in a 16-cell cyst, or the signal for F-actin at the junction between CpCs of the same germarium. To increase sample size, 20 flies of each genotype were collected for dissection. Only a few germaria from each ovary were examined, such that the total number of examined germaria reached about 100 for GSC number measurements and 10 for expression intensity measurements. To assess the distribution of E-cadherin at the interface between GSCs and CpCs, three-dimensional reconstructions were made from confocal z-sections using Imaris (version x64, 9.1.2) (Bitplane).
Statistics and reproducibility. Quantitative data were recorded and graphed in Excel (Microsoft) or GraphPad Prism 8 software (version 8.3.0). P-values were calculated from two-tailed unpaired Student's t-test or chi-squared test in Excel for comparison between two groups, and One-way ANOVA in Prism (GraphPad) for multiple comparisons. P-values are showed as exact value or asterisks in the graphs and legends of figures. At least two biological replicates were performed for each experiment. Representative images were taken from at least two technical replicate experiments with similar results or from independent biological samples with similar phenotypes.
RNA in situ hybridization. RNA in situ hybridization was performed as previously described 92 . In brief, ovaries were dissected and fixed overnight with 4% paraformaldehyde in PBS (1% DMSO and 0.1% DEPC) at 4°C. Ovaries were washed with PBST (PBS with 0.1% Tween-20), dehydrated by a series of ethanol solutions with increasing concentration (25%, 50%, 75%, and 100%) and kept at −20°C for at least 16 h. Ovaries were then rehydrated by a series of ethanol solutions with decreasing concentration and treated with proteinase K (50 µg/mL in PBST; Sigma) for 5 min at RT, followed by post-fixing in 4% paraformaldehyde for 30 min. The tissues were prehybridized in hybridization solution (50% formamide, 5x SSC, 0.1% Tween-20, 50 µg/µL heparin, 100 μg/mL yeast t-RNA and 10 µg/mL salmon sperm DNA) at 60°C for 1 h, then hybridized with hybridization solution containing Digoxigenin (DIG)-labeled probes at 60°C overnight. Hybridized ovaries were washed with 50 and 25% hybridization solution diluted in 2x SSC, PBST, and then incubated with 3% H 2 O 2 in PBST to inactivate endogenous perioxidase (POD). Samples were then blocked with 2x blocking buffer in maleic acid (Roche) at RT for 1 h, then incubated with anti-DIG-POD (1: 500, Roche # 11207733910) in blocking buffer at 4°C overnight. After washing, signals were developed using a TSA plus fluorescence kit (Perkin Elmer).
CpC sorting for transcriptome analysis. The transparent portions of 60 of 2week-old bab1 > gfp and bab1 > gfp & piwi RNAi-1 ovaries (switched to 29°C after eclosion) were dissected in GIM with 10% FBS. Ovaries were further dissociated in 2.5% Trypsin solution (Sigma) containing 600 units (U) Collagenase type I (Thermo Fisher Scientific) for 30 min at 29°C with vigorous shaking; throughout the incubation, samples were vortexed every 5 min. Cell suspensions were filtered twice through a 70-μm nylon mesh. Cells were collected by centrifugation at 2000 × g for 5 min at 4°C, then re-suspended in GIM containing 10% FBS. Dead cells were labeled with propidium iodide (PI) (2 μg/mL). Three hundred GFPpositive cells (CpCs, TFs and few anterior ECs) from bab1 > gfp and bab1 > gfp & piwi RNAi-1 were collected into wells of 96-well plate containing lysis buffer from SMART-Seq HT kit (Takara Bio USA, Inc.) by fluorescence-activated cell sorting (FACS) on a FACSAriaII (BD Biosciences) (Supplementary Fig. 20a). Two biological replicates were used for the analysis. Total RNA was subjected to reverse transcription for cDNA generation in a 96-well plate, according to the manufacturer's instructions for the SMART-Seq HT kit. cDNA was further amplified by PCR (11 cycles), also according to manufacturer's protocol for the SMART-Seq HT kit. Amplified cDNA was purified by Agencourt AMPure XP (Beckman Coulter, Inc.). All RNA-seq protocols followed the manufacturer's instructions (Illumina). cDNA was qualified with a Bioanalyzer 2100 on a RNA 6000 labchip (Agilent Technology), and library construction was carried out with an Agilent SureSelect Strand Specific RNA library Preparation Kit for 75SE (Paired-End) sequencing on a Solexa platform. The sequences were determined by sequencing-by-synthesis technology with a TruSeq SBS kit. Sequencing data (FASTQ reads) were generated using the Welgene Biotech pipeline based on the Illumina base-calling program, bcl2fastq v2.20, at 14.5 M to 18.9 M (million reads) per sample. The sequences were then filtered to obtain qualified reads. Sequence quality trimming was performed with Trimmomatic v0.36, using a sliding-window approach. The transposon and gene expression levels were calculated as FPKM (Fragment Per Kilobase of transcript per Million mapped reads). For transposon analysis, the "Transposable elements (canonical set)" database was downloaded from FlyBase and used as a reference for transposon annotations. For coding-gene expression analysis, the reference gene annotations were retrieved from FlyBase.
CpC sorting and genomic sequencing and analysis. The transparent portions of hundreds of 2-week-old bab1 > gfp & piwi RNAi-1 ovaries (switched to 29°C after eclosion) were dissected in GIM with 10% FBS. Ovaries were further dissociated in 2.5% Trypsin solution (Sigma) containing 600 units (U) Collagenase type I (Thermo Fisher Scientific) for 30 min at 29°C with vigorous shaking; throughout the incubation, samples were vortexed every 5 min. Cell suspensions were filtered twice through a 70-μm nylon mesh. Cells were collected by centrifugation at 2000 × g for 5 min at 4°C, then re-suspended in GIM containing 10% FBS. Dead cells were labeled with propidium iodide (PI) (2 μg/mL). GFP-positive cells (piwi-KD CpCs, TFs-which do not express Piwi-and a few anterior ECs) and GFPnegative ovarian somatic cells (germ cells were excluded by their large size) were collected by fluorescence-activated cell sorting (FACS) on a FACSAriaII (BD Biosciences) (Supplementary Fig. 20b).
Genomic DNA was extracted from GFP-positive niche cells (CpCs and TF) and GFP-negative ovarian non-niche somatic cells (like ECs and FCs, which served as the piwi-KD control), as well as 50 ovaries from newly eclosed bab1 > gfp & piwi RNAi-1 of flies (cultured at 18°C to silence GAL4 activity for the genotype control). Genomic DNA of whole ovaries was extracted by phenol-chloroform extraction, according to the instructions from Vienna Drosophila Resource Center (VDRC). Genomic DNA extracted from sorted GFP-positive niche cells and GFP-negative ovarian non-niche somatic cells (~600 cells for each sample) was amplified using the REPLI-g Single Cell Kit (Qiagen) according to the manufacturer's instructions. In brief, isolated cells were lysed by adding Buffer D2 and incubating on ice for 10 min. After adding stop solution, amplification master mix containing Phi DNA polymerase was added to the cell lysate and incubated at 30°C for 2.5 h. Amplified genomic DNA was then purified using a DNeasy Blood and Tissue Kit (Qiagen) with EB elution buffer. Purified genomic DNA of isolated cells and of whole ovaries was subjected to genomic DNA sequencing by HiSeq Rapid v2 (paired-end 2 × 150 base pair). Transposon insertion sites were detected by Transposon Insertion and Deletion AnaLyzer (TIDAL) 94 based on the reference genome BDGP Release 6 (dm6). The software packages and annotation files were automated, and file format conversion, read trimming, mapping, and polymorphic transposon detection were performed seamlessly. Transposon insertion events were categorized by the location and types of the inserted sequences. New TE insertions found in GFP-positive niche cells but not in GFP-negative cells or ovary samples were called.
Transmission electronic microscopy (TEM). TEM was performed similar to a previous study 95 , with minor modifications. Fly ovaries were immersed in fixation solution (2.5% glutaraldehyde, 1% tannic acid and 2% paraformaldehyde in 0.1 M sodium cacodylate buffer) overnight at 4°C. Tissues were then washed in 0.1 M cacodylate buffer with 0.2 M sucrose and 0.1% CaCl 2 three times at 4°C (10 min for each wash). Next, ovaries were fixed with 1% osmium tetroxide, OsO 4 in 0.1 M cacodylate buffer for 1 h at 4°C and washed three times with ddH 2 O at 4°C (5 min for each wash). Ovaries were then stained with uranyl acetate at 4°C for 1 h and washed with ddH 2 O three times at 4°C (5 min for each wash). Stained samples were dehydrated with increasing concentrations of ethanol (30%, 50%, 75%, 90% and 3 × 100%, 5 min each) at 4°C and infiltrated by 100% ethanol: Spurr = 3:1, 1:1, 1:3 (each for 1 h) and pure Spurr four times (1 h each time). Spurr was allowed to polymerize for 48 h at 60°C. Ultrathin sections were sectioned with a diamond knife (DiATOME) at on a microtome and stained with 4% uranyl acetate for 20 min. After rinsing with H2O six times, ovaries were immersed in Reynolds lead citrate for 10 min. Then, slices were rinsed with H2O and mounted to on copper slot grids and observed under TEM. After mounting, germarial sections were examined by with a Tecnai G2 spirit TWIN transmission electron microscope (FEI Company) equipped with a Multiscan Gatan camera (Gatan) at an accelerating voltage of 120 kV.
Electroretinogram (ERG). Electroretinograms (ERGs) were performed by immobilizing flies after 2-or 10-day light exposure with PVA glue on glass slides. The recording and reference electrodes were filled with 2 M NaCl while the ERGs were carried out. An iWorx 404 multichannel recorder was used to amplify the electrode voltage (iWorx Systems Inc., NH, UAS), and signal was filtered through an intracellular microelectrode amplifier (Walter Instrument Co., CT, USA). Light stimuli (1 s) were administered by a computer-controlled white LED system, and the ERGs were recorded using iWorx LabScribe2 software.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Transposon expression profiles (Raw FPKM) are provided in Supplementary Table 3