Fragile X mental retardation protein coordinates neuron-to-glia communication for clearance of developmentally transient brain neurons

Significance

Brain circuits are remodeled by the removal of neurons via glial engulfment and phagocytosis. The mechanism requires neuron-to-glia signaling to identify the targeted neuron, recruit glia, and trigger phagocytosis. This normal optimization process goes awry in neurological disease states such as fragile X syndrome (FXS), a leading heritable cause of intellectual disability and autism spectrum disorders. Epigenetic silencing of the mRNA-binding translational regulator fragile X mental retardation protein (FMRP) elevates mothers against decapentaplegic (Mad) transcript levels to increase phospho-Mad (pMad) signaling and prevent brain circuit remodeling. In neurons, this FMRP-dependent regulatory network interacts with insulin receptor (InR) phospho–protein kinase B (pAkt) control of two neuron-to-glia signals driving phagocytosis. This mechanism is critical for normal and diseased brain circuit remodeling.

Abstract

In the developmental remodeling of brain circuits, neurons are removed by glial phagocytosis to optimize adult behavior. Fragile X mental retardation protein (FMRP) regulates neuron-to-glia signaling to drive glial phagocytosis for targeted neuron pruning. We find that FMRP acts in a mothers against decapentaplegic (Mad)-insulin receptor (InR)-protein kinase B (Akt) pathway to regulate pretaporter (Prtp) and amyloid precursor protein-like (APPL) signals directing this glial clearance. Neuronal RNAi of Drosophila fragile X mental retardation 1 (dfmr1) elevates mad transcript levels and increases pMad signaling. Neuronal dfmr1 and mad RNAi both elevate phospho–protein kinase B (pAkt) and delay neuron removal but cause opposite effects on InR expression. Genetically correcting pAkt levels in the mad RNAi background restores normal remodeling. Consistently, neuronal dfmr1 and mad RNAi both decrease Prtp levels, whereas neuronal InR and akt RNAi increase Prtp levels, indicating FMRP works with pMad and insulin signaling to tightly regulate Prtp signaling and thus control glial phagocytosis for correct circuit remodeling. Neuronal dfmr1 and mad and akt RNAi all decrease APPL levels, with the pathway signaling higher glial endolysosome activity for phagocytosis. These findings reveal a FMRP-dependent control pathway for neuron-to-glia communication in neuronal pruning, identifying potential molecular mechanisms for devising fragile X syndrome treatments.

Neuron-to-glia communication has critical roles in controlling brain circuit remodeling (1–3). Neuronal signaling induces glial phagocytosis from synapses to whole neurons, crucial in normal brains and in neurodevelopmental disorder conditions (2, 4–6). Disruption of neuron-to-glia communication causes aberrant neuronal pruning, resulting in defects ranging from defective synaptic transmission (7, 8) to impaired circuit wiring (9) to transient neuroinflammation (10, 11). A key case is fragile X syndrome (FXS), a leading intellectual disability and autism spectrum disorder (12, 13), typically caused by the epigenetic loss of fragile X mental retardation protein (FMRP) owing to expanded CGG repeats in the 5′-untranslated region of fragile X mental retardation 1 (fmr1) (13, 14). In neurons, FMRP binds to specific transcripts to regulate protein translation during brain circuit development and later plasticity, including synaptic connectivity remodeling (15) and intercellular signaling mechanisms (16, 17). In the Drosophila FXS disease model, FMRP loss blocks removal of the developmentally transient pigment-dispersing factor (PDF)-Tri peptidergic neurons from the juvenile brain (15). Cell-specific RNAi studies show that FMRP is required only in neurons, not glia, to transcellularly activate glial phagocytosis driving PDF-Tri neuron clearance (18). Thus, FMRP-dependent neuron-to-glia communication drives targeted neuron pruning, but the molecular mechanisms remain largely unknown.

FMRP regulates bone morphogenic protein (BMP) and insulin-like peptide (ILP) signaling (19). Activated Drosophila BMP receptors phosphorylate mothers against decapentaplegic (pMad) to control gene transcription, including the insulin receptor (InR) (20, 21). InRs phosphorylate protein kinase B (pAkt) (22, 23) to suppress dendrite pruning (22). Importantly, mouse FMRP binds smad messenger ribonucleic acid (mRNA) (Drosophila mad homologue) (23), and Drosophila FMRP regulates pMad signaling levels in neurons (24). Moreover, FMRP loss elevates InR-dependent signaling (25). In neuron-to-glia communication, pretaporter (Prtp) and amyloid precursor protein like (APPL) from neurons both activate glial phagocytosis (26, 27). Drosophila Prtp traffics to the neuron surface to bind the glial phagocytotic receptor draper (Drpr) (26, 28). Loss of neuronal FMRP decreases glial Drpr expression (18), consistent with FMRP-dependent Prtp signaling. Drosophila APPL has a cleavable N terminus, and APPL release from neurons activates glial phagocytosis (27). Glia take up secreted APPL to maintain Drpr expression and up-regulate Rab GTPases, activating the glial endolysosomal network for the neuron clearance mechanism (27). Taken together, these studies suggest that neuronal FMRP interacts with neuronal InR, pMad, and pAkt signaling cascades to tightly regulate Prtp and APPL neuron-to-glia communication controlling the glial phagocytosis of target neurons during circuit remodeling in the juvenile brain.

Here, we use Drosophila brain PDF-Tri neuron removal via glial phagocytosis to study the neuron-to-glia communication remodeling mechanism, assaying both the early pruning steps at 1 day post-eclosion (dpe) and end-stage clearance (5 dpe) (18). We discover that neuronal FMRP binds mad mRNA to restrict pMad signaling in neurons. Surprisingly, however, we find that both neuronal Drosophila fragile X mental retardation 1 (dfmr1) and mad RNAi similarly block PDF-Tri neuron removal, indicating a more complex regulatory mechanism. Consistently, we find pMad is a positive transcription factor for InRs driving downstream pAkt signaling but that pMad also indirectly inhibits pAkt, inducing the phenocopy between neuronal dfmr1 and mad RNAi conditions. We find both neuronal dfmr1 and mad RNAi similarly decrease Prtp neuron-to-glia signaling, resulting in reduced glial phagocytic activity and a block of PDF-Tri neuron clearance, whereas loss of neuronal InR and pAkt has that opposite phenotype of elevating Prtp to accelerate neuronal removal. We also discover that the FMRP-pMad-InR-pAkt pathway positively regulates neuronal APPL signaling to induce glial Rab7-driven endolysomal activation for PDF-Tri neuron clearance. Taken together, we conclude that neuronal FMRP-pMad and InR-pAkt cascades coordinate an integrated regulatory decision network governing neuron-to-glia communication via neuronal Prtp and APPL signaling ligands that drive glial phagocytosis for targeted neuron pruning from brain circuits.

Results

Neuronal FMRP Regulates pMad Signaling to Mediate PDF-Tri Neuron Clearance.

Previous genetic studies have shown that PDF-Tri neuron removal over the first 5 d in the early juvenile brain requires phagocytosis clearance by two cooperating glial subclasses (16, 19). To begin to dissect this mechanism further, we imaged interactions between glia and neurons early and late in this clearance process at 1 day posteclosion (dpe) and 5 dpe. To image neuron–glia interactions, we first used projected confocal microscopy to visualize glial localization relative to PDF-Tri neurons at 1 dpe during the early stages of the phagocytic removal (Fig. 1A). In 2D projections, the glia-specific reversed polarity (Repo) nuclear marker shows glial cells closely surrounding the PDF-Tri neurons, with multiple glia in direct contact along the entire length of the neuronal processes (Fig. 1 B, Top). In magnified three-dimensional (3D) reconstructed views, multiple glia (arrows) contact the PDF-Tri neuron, converging on the remodeling neurons from multiple directions (Fig. 1 B, Bottom and Movie S1). To label the glial membrane, we expressed UAS-mCD8::GFP with repo-Gal4 to show the PDF-Tri neuronal cell body covered by glial projections (Movie S2), further demonstrating close interaction between glia and PDF-Tri neurons. Previous studies have shown neuronal FMRP signals glial phagocytosis for PDF-Tri neuron removal (18). To test this mechanism over the time course of clearance, we assayed the PDF-Tri neuron area at 1 dpe (early in the removal process) and 5 dpe (at the end of the removal process) in neuronal driver controls (elav-Gal4/+) and with dfmr1 RNAi (Fig. 1C). At 1 dpe, controls (elav/+) already show dramatically reduced PDF-Tri neuron area compared to dfmr1 RNAi (elav>dfmr1 RNAi) animals (Fig. 1D). Two-way ANOVA indicates that neuronal FMRP loss causes retention of significantly more PDF-Tri neuron area at 1 dpe. By the end of the removal process (5 dpe), controls show near-complete PDF-Tri neuron loss, whereas dfmr1 RNAi results in neuron maintenance (Fig. 1C). The control PDF-Tri neuron area is enormously reduced compared to neuron retention with dfmr1 RNAi, a highly significant difference by two-way ANOVA (Fig. 1D). To specifically test FMRP in PDF-Tri neurons, we assayed PDF driver controls (PDF-Gal4/+) alone and with targeted dfmr1 RNAi (SI Appendix, Fig. S1A). FMRP loss in PDF neurons significantly lowers PDF-Tri neuron clearance at 1 dpe (SI Appendix, Fig. S1B). Taken together, these findings indicate that neuronal FMRP plays critical roles in driving normal PDF-Tri neuron clearance in the early juvenile brain.

Fig. 1.

Glial phagocytic removal of PDF-Tri neurons requires FMRP regulation of pMad. (A) PDF-Tri neuron (anti-PDF, green) and glial cells (anti-Repo, magenta) in the central brain at the early glial phagocytic removal (0-1 dpe). (Scale bar, 20 μm.) (B) Progressively higher magnified views of PDF-Tri neuron and glia interaction. [Scale bars, magnified 2D (Right Top), 5 μm; magnified 3D (Right Bottom), 5 μm on X, Y, and Z axes.] (C) Double labeling of the neuronal driver control (elav/+, Left) and with dfmr1 RNAi (Right) at day 1 (Top) and day 5 (Bottom) after eclosion. (Scale bar, 20 μm.) (D) Two-way ANOVA of the PDF-Tri neuron area quantification at both time points (data shown as mean ± SEM; 1 dpe: control 1,761.0 ± 149.80 vs. dfmr1 RNAi 2,615.0 ± 148.30, P = 0.0047; 5 dpe: control 455.9 ± 104.10 vs. dfmr1 RNAi 1,767.0 ± 219.90, P < 0.0001). There is no significant difference in glia density between elav/+ and elav>dfmr1 RNAi (P = 0.121). (E) Representative anti-pMad western blot of neuronal elav/+ control and with dfmr1 RNAi. GAPDH is the loading control. (F) Normalized pMad levels from the western blots (control 1.00 ± 0.034 vs. dfmr1 RNAi 2.14 ± 0.121; Student’s t test, P < 0.0001). (G) RNA immunoprecipitation and normalized quantification of FMRP-bound transcripts. GFP-trap beads pull down neuronal YFP::FMRP together with negative control GFP (GFP-stau 1.00 ± 0.009 vs. FMRP-stau 1.979 ± 0.163, Student’s t test, P = 0.0001; GFP-mad 1.00 ± 0.019 vs. FMRP-mad 1.653 ± 0.137, Student’s t test, P = 0.0008). (H) Normalized mad mRNA levels in elav/+ control and with dfmr1 RNAi (control 1.00 ± 0.045 vs. dfmr1 RNAi 1.182 ± 0.037, Student’s t test, P = 0.009). Dot plots show all data points. Statistical significance indicated as **P < 0.01, ***P < 0.001, and ****P < 0.0001.

We recently discovered that dfmr1 null neurons accumulate high levels of phosphorylated mothers against decapentaplegic (pMad), indicating FMRP limits neuronal pMad signaling (24). To study this mechanism of FMRP-pMad regulation in the context of PDF-Tri neuron removal from the developing brain, we first tested pMad levels with neuronal dfmr1 RNAi in staged brain western blot analyses (Fig. 1E). Comparisons with the neuronal driver controls (elav/+) show that dfmr1 RNAi results in a consistently elevated pMad level. Quantified analyses indicate that neuronal dfmr1 RNAi causes a significantly higher pMad signaling level, greater than twofold increase over the matched control (Fig. 1F). Since mouse RNA immunoprecipitation (RIP) sequencing has shown that FMRP binds smad mRNA (Drosophila mad homologue) (23), we next assayed for an RNA-binding interaction. To analyze FMRP-bound transcript levels, we used GFP-trap beads to pull down a FMRP::YFP fusion compared with a GFP-negative control from brain neurons (24). As FMRP has been shown to bind staufen (stau) mRNA (24), we use stau as the immunoprecipitation positive control (Fig. 1G). With quantitative real-time PCR (qPCR) measurements, we found a >1.5-fold immunoprecipitation enrichment for mad mRNA occurs with FMRP::YFP pull-down as normalized compared to the GFP-negative control (Fig. 1G). This is comparable to the approximately twofold immunoprecipitation for the stau positive control. These results show FMRP binds mad mRNA in brain neurons, so we next tested possible FMRP-dependent effects on mad transcription levels (Fig. 1H). Compared with the driver control (elav/+), reducing neuronal FMRP (elav>dfmr1 RNAi) causes a significant increase in mad mRNA levels (Fig. 1H). This suggests that neuronal FMRP destabilizes bound mad transcripts to reduce the amount of protein produced. Taken together, these findings indicate that FMRP binds mad mRNA and reduces mad transcript levels, thereby presumably restricting pMad intracellular signaling.

During intracellular signaling, the phosphorylation of Mad (pMad) generates an active transcription factor that translocates into the nucleus to directly regulate gene expression (20). We therefore next wanted to determine whether knockdown of neuronal mad mRNA impairs pMad signaling and whether neuronal pMad signaling regulates targeted PDF-Tri neuron removal. To test the first question, we compared driver controls (elav/+) to neuronal mad RNAi (elav>mad RNAi) to find a significant reduction in pMad protein levels. To test whether neuronal pMad regulates PDF-Tri neuron clearance, we therefore reduced pMad in neurons by neuronal mad RNAi (Fig. 2A). Surprisingly, this knockdown results in impaired PDF-Tri neuron clearance similar to the dfmr1 RNAi condition (compare to Fig. 1 C and D). Starting early in the glial clearance process (1 dpe), and particularly at the end of the removal mechanism when PDF-Tri neurons are normally absent (5 dpe), neuronal mad RNAi results in a striking retention of most of the neuronal architecture (Fig. 2A). Statistical analyses indicate that neuronal mad RNAi causes significantly reduced PDF-Tri neuron removal compared to driver controls (elav/+) at 1 dpe and an even stronger significant maintenance at 5 dpe (Fig. 2B). To specifically test FMRP in PDF-Tri neurons, we assayed PDF driver controls (PDF-Gal4/+) alone and with mad RNAi (PDF>mad RNAi; SI Appendix, Fig. S1A). Loss of pMad signaling in PDF neurons very significantly lowers PDF-Tri neuron clearance at 1 dpe (SI Appendix, Fig. S1B). Since this conclusion is opposite to expectations, a second FMRP-pMad interactive pathway was suspected to result in the dfmr1 RNAi phenocopy. As both FMRP and pMad interact with insulin receptor (InR) signaling (19, 20, 29), we therefore next focused on testing InR dysregulation in neuronal dfmr1 and mad RNAi conditions. With qPCR assays, InR expression is oppositely modulated by neuronal dfmr1 and mad RNAi (Fig. 2 C and D), which fits the inhibitory regulation from FMRP to pMad. However, this fails to explain the phenocopy result. We therefore next explored interactions downstream of InR signaling.

Fig. 2.

PDF-Tri neuron clearance through FMRP-pMad signaling up-regulates pAkt level. (A) PDF-Tri neuron (green) and glia (magenta) labeling in the neuronal driver control only (elav/+, Left) and with neuronally driven mad RNAi (Right) at day 1 (Top) and day 5 (Bottom) after eclosion. PDF-Tri neuron alone is shown to the right for each image. (Scale bar, 20 μm.) (B) PDF-Tri neuron area quantification for both genotypes at both time points (1 dpe: control 1,894.0 ± 125.10 vs. mad RNAi 2,771.0 ± 118.60, two-way ANOVA, P = 0.0020; 5 dpe: control 265.20 ± 86.97 vs. mad RNAi 1,852.0 ± 230.80, two-way ANOVA, P < 0.0001). There is no significant difference in glia density (anti-Repo) between elav/+ and elav>mad RNAi (P = 0.642). (C and D) Normalized InR mRNA levels with the neuronal elav/+ driver control compared to dfmr1 RNAi (C, elav/+ 1.00 ± 0.036 vs. dfmr1 RNAi 1.283 ± 0.0441, Student’s t test, P = 0.0006) and mad RNAi (D, elav/+ 1.00 ± 0.018 vs. mad RNAi 0.851 ± 0.020, Student’s t test, P = 0.0001). (E) Representative anti-pAkt western blots with the neuronal elav/+ driver control compared to both neuronal dfmr1 RNAi (Top) and neuronal mad RNAi (Bottom) conditions. GAPDH is the loading control. (F and G) Western blot quantification for both conditions compared to elav/+ driver control (F, elav/+ 1.00 ± 0.007 vs. dfmr1 RNAi 2.06 ± 0.203, Student’s t test, P = 0.0004; G, elav/+ 1.00 ± 0.041 vs. mad RNAi 2.14 ± 0.178, Student’s t test, P < 0.0001). Dot plots show all data points. Statistical significance indicated as **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Activated InR phosphorylates protein kinase B (pAkt) (30, 31), and FMRP loss elevates pAkt in neurons (29). To test whether pAkt is the dysregulated player in the FMRP-pMad signaling, we first used western blots to measure pAkt levels in driver controls (elav/+) compared with dfmr1 and mad RNAi (Fig. 2E). Quantification shows pAkt up-regulated by both neuronal dfmr1 and mad RNAi, with a significantly approximately twofold enrichment in both cases (Fig. 2 F and G). To test pAkt in PDF-Tri neurons, we colabeled with anti-PDF and anti-pAkt. Both dfmr1 and mad RNAi increase pAkt in PDF-Tri neurons at 1 dpe (SI Appendix, Fig. S1 C–E). These findings suggest pAkt upregulation generates the neuronal dfmr1 and mad RNAi phenocopy blocking PDF-Tri neuron removal. Consistently, neuronal pAkt RNAi dramatically increases the rate of PDF-Tri neuron clearance (Fig. 3A). Due to the complete neuron loss at 5 dpe in controls, we only measured the PDF-Tri neuron area at 1 dpe to show significantly accelerated neuronal removal early in the process (Fig. 3B). As decreasing neuronal pMad results in aberrant pAkt elevation, we hypothesized that correcting neuronal pAkt levels in the reduced pMad background should restore normal clearance. With mad and akt double RNAi, PDF-Tri neuron clearance is restored to control levels (Fig. 3 C and D), supporting this hypothesis. To test possible Gal4 dilution with double UAS constructs, we also expressed akt RNAi with control gfp RNAi. This double RNAi still significantly promotes neuron loss (SI Appendix, Fig. S2 A and B). We also overexpressed akt (elav>akt OE) to find slower PDF-Tri neuron loss at 1 dpe (SI Appendix, Fig. S2 C and D). PDF-Tri neuron clearance defects from InR knockdown (Fig. 3 E and F) and constitutive activation (InR CA; SI Appendix, Fig. S2 C and D) are also consistent with InR acting as a positive regulator to pAkt (31). Importantly, InR and akt RNAi cause no significant effect on PDF-Tri neurons in pre-eclosion pharate adults (SI Appendix, Fig. S3 A–D), supporting a specific role for InR-pAkt signaling in PDF-Tri neuron clearance. We finally tested InR-pAkt signaling within PDF-Tri neurons. Consistent with pan-neuronal RNAi, both PDF-Gal4-driven InR and akt RNAi promote PDF-Tri neuron clearance, while targeted akt and mad double RNAi correct the akt phenotype (SI Appendix, Fig. S4 A and B). These results show FMRP-pMad cross talk to InR-pAkt signaling regulates PDF-Tri neuron removal. However, this intracellular neuronal network must act to control intercellular signaling to drive glial phagocytosis, so we next turned to testing the neuron-to-glia communication mechanism.

Fig. 3.

Neuronal knockdown of both InR and Akt accelerates PDF-Tri neuron clearance. (A) PDF-Tri neuron (anti-PDF, green) and glia (anti-Repo, magenta) double labeling in the neuronal driver control (elav/+, Left) and akt RNAi (Right) at day 1 posteclosion during the early stages of the glial phagocytic removal process. (Scale bar, 20 μm.) (B) Quantification of the PDF-Tri neuron area in the two conditions (elav/+ 1,578.0 ± 132.0 vs. akt RNAi 927.10 ± 117.70, Students’ t test, P = 0.0017). There is no significant difference in glia density between elav/+ and elav>akt RNAi at 1 dpe (P = 0.778). (C) The same colabeling with neuronal mad and akt RNAi driven in double knockdown combination at day 1. (Scale bar, 20 μm.) (D) Quantification of the PDF-Tri neuron area in the two conditions (elav/+ 1424.0 ± 48.69 vs. mad; akt double RNAi 1,308.0 ± 100.10, Students’ t test, P = 0.357). (E) PDF-Tri neuron and glia double labeling in the neuronal driver control (elav/+, Left) and with InR RNAi (Right) at day 1. (F) Quantification of the PDF-Tri neuron area in the two conditions (elav/+ 1,886.0 ± 196.10 vs. InR RNAi 1,273.0 ± 94.27, Students’ t test, P = 0.008). There is no significant difference in glia density between elav/+ and elav>InR RNAi at 1 dpe (P = 0.464). Dot plots show all data points. Statistical significance indicated as **P < 0.01 and not significant (ns; P > 0.05). (Scale bar in all images, 20 μm.)

Neuronal FMRP-pMad-InR Network Regulates Pretaporter Signaling for Glial Phagocytosis.

Neuronal pretaporter (Prtp) binds the glial draper engulfment receptor to drive the glial phagocytosis of neurons (26). Neurons traffic Prtp from the endoplasmic reticulum (ER) to the plasma membrane cell surface to activate targeted glial phagocytosis (26). We therefore next tested neuronal Prtp knockdown (elav>prtp RNAi) compared to the driver control (elav/+) to test whether PDF-Tri neuron removal involves Prtp signaling (Fig. 4A). In assaying the full time course of the glial phagocytosis mechanism, we find strongly reduced PDF-Tri neuron clearance at both 1 dpe (early) and 5 dpe (late) with neuronal Prtp knockdown (Fig. 4A). As usual, the aberrant retention is more obvious at 5 dpe, although the PDF-Tri neuron maintenance here is reduced relative to both the dfmr1 and mad RNAi conditions (compare to Figs. 1 and 2), suggesting the combinatorial action of parallel neuron-to-glia signals (see below). Two-way ANOVAs show the PDF-Tri neuron area with neuronal prtp RNAi is significantly greater early in the glial removal process (1 dpe) and also at the end of clearance (5 dpe) when neurons are normally eliminated in the elav/+ driver controls (Fig. 4B). These results suggest neuronal Prtp is involved in PDF-Tri neuron clearance by inducing glial phagocytosis. We therefore next tested whether the above neuronal FMRP-pMad and InR-pAkt network modulates Prtp signaling. With qPCR assays, both neuronal dfmr1 and mad RNAi reduce prtp levels (Fig. 4 C and D), consistent with the inhibition of PDF-Tri neuron clearance. Moreover, both neuronal InR and akt RNAi elevate prtp levels (Fig. 4 E and F), consistent with accelerated PDF-Tri neuron clearance. These combined interactions provide a mechanism to explain the bidirectional change in PDF-Tri neuron removal in these two pathway conditions downstream of the linked neuronal regulation. Taken together, these findings show the neuronal FMRP-pMad and InR-pAkt network controls Prtp signaling to regulate the glial phagocytic clearance of PDF-Tri neurons. However, another neuron-to-glia intercellular signaling mechanism seemed to be acting in parallel.

Fig. 4.

Neuron-to-glia pretaporter signaling acts downstream of FMRP, pMad, and InR. (A) PDF-Tri neuron (green) and glia (magenta) colabeling with the neuronal driver control (elav/+, Left) and neuronal pretaporter RNAi (prtp, Right) at both day 1 (Top) and day 5 (Bottom). (Scale bar, 20 μm.) (B) Quantification of the PDF-Tri neuron area for both the genotypes at both early and late time points (1 dpe: elav/+ 1,151.0 ± 77.48 vs. prtp RNAi 1,620.0 ± 93.28, two-way ANOVA, P = 0.0148; 5 dpe: elav/+ 29.51 ± 14.21 vs. prpt RNAi 574.0 ± 164.10, two-way ANOVA, P = 0.0432). There is no significant difference in glia density between elav/+ and elav>prtp RNAi (P = 0.571). (C) Normalized quantification of prtp mRNA levels in the neuronal driver control (elav/+, Left) and with neuronal dfmr1 RNAi (Right) (elav/+ 1.00 ± 0.016 vs. dfmr1 RNAi 0.717 ± 0.027, Students’ t test, P = 0.0001). (D) Normalized quantification of prtp mRNA levels in the neuronal driver control (elav/+, Left) and with neuronally targeted mad RNAi (Right) (elav/+ 1.00 ± 0.018 vs. mad RNAi 0.702 ± 0.089, Students’ t test, P = 0.0051). (E) Normalized quantification of prtp mRNA levels in the neuronal driver control (elav/+, Left) and with neuronal InR RNAi (Right) (elav/+ 1.00 ± 0.021 vs. InR RNAi 1.329 ± 0.022, Students’ t test, P < 0.0001). (F) Normalized quantification of prtp mRNA levels in the neuronal driver control (elav/+, Left) and with neuronal akt RNAi (Right) (elav/+ 1.00 ± 0.07045 vs. InR RNAi 1.507 ± 0.127, Students’ t test, P = 0.0130). Dot plots show all data points. Statistical significance indicated as *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Neuronal FMRP-pMad and InR-pAkt Network Regulates Amyloid Precursor Protein Signaling.

In the mouse FXS model, FMRP restricts neuronal amyloid precursor protein (APP) levels (32, 33). In Drosophila, the amyloid precursor protein-like (APPL) homologue cleaved N terminus released from brain neurons is engulfed by glia to activate their phagocytic function during brain neuron removal (27). To test whether neuronal APPL is required for PDF-Tri neuron clearance, we next compared driver controls (elav/+) with neuronal APPL knockdown (elav>appl RNAi) at 1 dpe (early) and 5 dpe (late) in the glial phagocytosis clearance process (Fig. 5A). Brain imaging shows more PDF-Tri neuron retention with neuronal appl RNAi compared to driver controls at 1 dpe and 5 dpe, with the difference more obvious late in glial removal (Fig. 5A). Quantification shows the removal of PDF-Tri neurons is significantly reduced with neuronal appl RNAi at both the 1 dpe and 5 dpe time points (Fig. 5B). These results are consistent with recently reported APPL requirements in glia-mediated neuronal removal (27). We next tested if FMRP and APPL interact to regulate PDF-Tri neuron removal. Single dfmr1/+ and appl^d/+ heterozygotes have no significant impact on PDF-Tri neuron clearance compared to the genetic background control w¹¹¹⁸, whereas the dfmr1/+; appl^d/+ trans-heterozygote significantly increases PDF-Tri neuron retention (SI Appendix, Fig. S5 A and B). To investigate how neuronal FMRP-pMad and InR-pAkt signaling might regulate this APPL function, we next combined qPCR and western blots to measure APPL expression in targeted RNAi studies for neuronal dfmr1, mad, and InR RNAi (Fig. 5 C–E). Quantified analyses show a similar significant reduction in appl transcript levels with both dfmr1 and mad RNAi compared to the elav/+ driver controls (Fig. 5 C and D), consistent with the PDF-Tri neuron persistence in both these conditions. Likewise, neuronal InR knockdown results in an even more significant ~50% reduction in appl mRNA levels (Fig. 5E). Moreover, neuronal dfmr1, mad, InR, and akt RNAi also decrease APPL protein levels (Fig. 5F). The dfmr1 and mad results are again consistent with the delayed neuronal removal, but the decreased APPL with InR and akt RNAi opposes the phenotype of accelerated PDF-Tri neuron removal. Taken together, neuronal FMRP-pMad and InR-pAkt cascades similarly regulate neuronal APPL levels. However, we still wished to test the role of neuronal signaling in glial activation for PDF-Tri neuron removal.

Fig. 5.

Neuronal APPL acts in the same pathway to facilitate PDF-Tri neuron clearance (A) PDF-Tri neuron (anti-PDF, green) and glial cell (anti-Repo, magenta) double labeling in the neuronal driver control (elav/+, Left) and appl RNAi (Right) at both day 1 (Top) and day 5 (Bottom). The PDF-Tri neuron alone (green) is shown to the Right. (Scale bar, 20 μm.) (B) Quantification of the PDF-Tri neuron area for both genotypes at both early and late time points (1 dpe: elav/+ 1,620.0 ± 158.70 vs. appl RNAi 2,469.0 ± 165.40, two-way ANOVA, P = 0.0072; 5 dpe: elav/+ 175.0 ± 51.34 vs. appl RNAi 954.0 ± 272.50, two-way ANOVA, P = 0.0195). There is no significant difference in glia density between elav/+ and elav>appl RNAi (P = 0.746). (C–E) Normalized quantification of appl mRNA levels in neuronal driver controls (elav/+, Left) compared with neuronal dfmr1 RNAi (C, elav/+ 1.00 ± 0.012 vs. dfmr1 RNAi 0.79 ± 0.013, Students’ t test, P < 0.0001), mad RNAi (D, elav/+ 1.00 ± 0.032 vs. mad RNAi 0.70 ± 0.042, Students’ t test, P = 0.0009), and InR RNAi (E, elav/+ 1.00 ± 0.011 vs. InR RNAi 0.58 ± 0.030, Students’ t test, P < 0.0001). (F) Representative anti-APPL western blots with the neuronal elav/+ controls (Left in each pairing) and neuronal dfmr1 RNAi (Left), mad RNAi (second from Left), InR RNAi (second from Right), and akt RNAi (Right). (G–J) Western blot quantification normalized to neuronal elav/+ controls (Left in each pairing) for dfmr1 RNAi (G,  elav/+ 1.00 ± 0.039 vs. dfmr1 RNAi 0.78 ± 0.037, Students’ t test, P = 0.0018), mad RNAi (H, elav/+ 1.00 ± 0.024 vs. mad RNAi 0.69 ± 0.045, Students’ t test, P = 0.0001), InR RNAi (I, elav/+ 1.00 ± 0.031 vs. InR RNAi 0.68 ± 0.073, Students’ t test, P = 0.0011), and akt RNAi (J, elav/+ 1.00 ± 0.040 vs. akt RNAi 0.77 ± 0.045, Students’ t test, P = 0.0037). Dot plots show all data points. Statistical significance indicated as *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001.

Neuronal FMRP-pMad and InR-pAkt Network Up-Regulates Glial Endolysosomal Activity.

Neuronal InR signaling inhibits glial engulfment (34, 35), with reduced endolysosomal activation in glia (36). The APPL N terminus released from brain neurons up-regulates the glial Rab GTPases endolysosomal network for glial phagocytosis (27). Thus, the InR-pAkt-APPL regulation of glial activation has been established. To test whether neuronal FMRP-dependent pMad and Prtp signaling plays roles in transcellularly regulating glial endolysosomal activation, two binary transgenic systems were used in parallel: 1) neuron-targeted RNAi lines driven with Gal4/UAS, and 2) glia-targeted mCD8::GFP plasma membrane labeling with LexA/LexAOP (Fig. 6A). All assays were done on glia immediately adjacent to the PDF-Tri neuron cell bodies at 1 dpe to investigate early-stage activation in the pruning glial population. As endolysosomal sorting starts with the Rab5 GTPase in early endosomes (27, 37), we first tested anti-Rab5 labeling in glia. None of the neuronal RNAi manipulations cause any change in the glial Rab5 signal, indicating early endosomal activity is not affected. Late endolysosomal sorting employs the Rab7 GTPase on late endosomes and lysosomes (27), so we next tested anti-Rab7 labeling in glia. Neuronal dfmr1, mad, and prtp RNAi all dramatically depress glial Rab7 (Fig. 6A). The driver controls (elav/+) consistently have larger, more numerous Rab7 puncta in glia adjacent to the PDF-Tri neurons, whereas all three neuronal knockdowns (dfmr1, mad, and prtp RNAi) have obviously smaller Rab7 puncta that are relatively difficult to detect (Fig. 6A). Quantification of glial Rab7 volume shows the most significant reduction with neuronal dfmr1 RNAi normalized to the elav/+ control (Fig. 6B). Consistent with the above regulatory pathway, both neuronal mad and prtp RNAi also exhibit significantly reduced glial Rab7 volume during PDF-Tri neuron removal (Fig. 6 C and D). Together with established neuronal InR and APPL signaling of glial activation (29, 34–36), these findings show neuronal FMRP-pMad-Prtp signaling triggers glial endolysosomal phagocytosis activation for neuron clearance in the juvenile brain.

Fig. 6.

Neuronal FMRP-pMad-Prtp signaling activates the glial endolysosomal pathway. (A) Glial cells immediately adjacent to the PDF-Tri neurons in the central brain marked with membrane mCD8::GFP (green) and colabeled with endolysosomal anti-Rab7 (magenta) at day 1 during the glial phagocytic removal process. The neuronal driver control alone (elav/+, Left) is compared to driven dfmr1 RNAi (second from Left), mad RNAi (second from Right), and prtp RNAi (Right). The endolysosomal Rab7 is shown alone in the bottom row for all four genotypes. (Scale bar, 20 μm.) (B–D) Glial Rab7 volume normalized to the neuronal driver control (elav/+, Left in each pairing) compared to neuronal dfmr1 RNAi (B, elav/+ 1.00 ± 0.186 vs. dfmr1 RNAi 0.318 ± 0.129, Students’ t test, P = 0.0078), neuronal mad RNAi (C, elav/+ 1.00 ± 0.178 vs. mad RNAi 0.537 ± 0.119, Students’ t test, P = 0.041), and neuronal prtp RNAi (D, elav/+ 1.00 ± 0.123 vs. prtp RNAi 0.479 ± 0.147, Students’ t test, P = 0.014). Dot plots show all data points. Statistical significance indicated as *P < 0.05 and **P < 0.01.

Discussion

We have discovered an integrated mechanism of neuronal FMRP-dependent network signaling that regulates neuron-to-glia communication to drive the glial phagocytic removal of targeted neurons from an otherwise maintained brain circuit. Specifically, within the neurons, RNA-binding FMRP restricts the translation of bound mad transcripts to limit phosphorylated Mad (pMad) signaling, which, in turn, inhibits phosphorylated Akt (pAkt) to promote glial phagocytosis for neuron removal. In parallel, the neuronal insulin receptor (InR) regulates pAkt signaling in a second intersecting cascade controlling the neuronal clearance mechanism. This bone morphogenic protein (BMP) and insulin-like peptide (ILP) neural decision-making network (19) controls neuron-to-glia communication regulating glial phagocytosis function for targeted neuron removal. Neuronal pretaporter (Prtp) is a ligand for the draper (Drpr) engulfment receptor on glia (26). The FMRP-pMad pathway promotes neuron-to-glia Prtp signaling to induce glial phagocytosis for neuron clearance, whereas the InR-pAkt pathway suppresses Prtp signaling to repress glia-mediated neuron removal. Neuronal amyloid precursor protein like (APPL) is released via a cleavable N terminus to activate Rab7 GTPase endolysomes in glia (27). The FMRP-pMad regulatory pathway promotes this neuron-to-glia APPL signaling, consistent with inducing glial phagocytosis for neuron clearance, and the intersecting InR-pAkt pathway also up-regulates this signaling, suggesting additional roles in neuron removal. Overall, these findings indicate neuronal FMRP coordinates signal transduction cascades to provide cross talk downstream of two signaling inputs (BMP and ILP), which provides output in the form of two neuron-to-glia signaling ligands (Prtp and APPL) that regulate glial phagocytosis for the clearance of targeted neurons from the juvenile brain.

From mRNA Transcription to Phosphorylated Protein Signaling.

We suggest RNA-binding FMRP limits pMad signaling levels by reducing the number of mad transcripts available for translation. FMRP is established to maintain protein translational homeostasis by modulating RNA target stability (38, 39). Ribosome profiling and transcriptome sequencing demonstrate imbalanced FMRP-targeted mRNA levels are common in the mouse FXS model brain due to reduced stabilization (38). In mouse RNA immunoprecipitation (RIP) sequencing, FMRP binds smad (Drosophila mad homologue) transcripts widely in the 5′-UTR, within the coding region, and in the 3′-UTR (23). With both RIP and qPCR measurements, we find Drosophila FMRP likewise binds mad mRNA and that FMRP loss increases both mad transcripts and pMad protein levels within brain neurons. We therefore suggest a causal effect correlation, although other indirect regulatory mechanisms are also possible (40, 41). We find loss of neuronal pMad decreases InR levels but elevates InR-dependent pAkt signaling, suggesting cross talk inhibition from pMad to pAkt. The pMad transcription factor (20) positively regulates InR expression. Downstream of the InR, pAkt regulates numerous targets, including target of rapamycin complexes (42) and the transcription factor forkhead box O (43), to regulate cell fate decisions (44, 45). In the brain neuronal fate decision, we suggest pMad inhibits pAkt signaling as a major cross talk regulation within the neuron-to-glia signaling network. With western blot assays, we find reducing neuronal pMad levels causes no changes in Akt protein levels but a approximately twofold increase in pAkt signaling. Thus, FMRP-dependent pMad signaling modulates InR-pAkt signaling at two levels: as a positive transcription factor regulating InR expression and as an inhibitor of downstream pAkt signaling. We conclude this cross talk provides critical regulation for targeted neuron pruning from the juvenile brain.

Brain Circuit Pruning in Fragile X Syndrome (FXS).

Neuron removal from brain circuits is a normal mechanism in neurodevelopment, often mediated by glial phagocytosis (46–48). During the clearance process, glia engulf neurons and degrade internalized debris in endolysosomes (49, 50). In glia, the engulfed neuronal debris is first sorted in Rab5 GTPase early endosomes (27), then trafficked from early-to-late endosomes with accompanying increased intravacuolar acidification, and finally delivered to Rab7 GTPase lysosomes where acid hydrolases complete the degradation process (51). In both rodent and Drosophila FXS disease models, this glial phagocytosis mechanism is severely impaired, resulting in a failure to prune neurons in brain circuits (18, 52). In the Drosophila FXS model, loss of neuronal FMRP decreases the glial Drpr engulfment receptor, and loss of glial Drpr blocks glial phagocytosis to prevent neuron removal (18). This previous study generated the hypothesis that the impaired brain circuit neuron removal in the FXS disease model is caused by the loss of neuron-to-glia communication driving glial phagocytosis (15). With neuron-specific RNAi trials, we find the neuronal FMRP-pMad pathway triggers Drpr ligand Prtp signaling from neurons to mediate glial engulfment and phagocytosis, whereas the neuronal InR-pAkt pathway represses Prtp to oppose the glial removal mechanism. Furthermore, we find targeted knockdown of neuronal FMRP, pMad, and Prtp impedes glial endolysosomal activation with reduced Rab7 expression, showing the neuronal FMRP-dependent network is required for glial phagocytosis function. These findings are consistent with the impaired glia-mediated neuron pruning in FXS disease models (18, 52) and provide insight into the molecular mechanisms of this defect. Clearly, the decision to eliminate neurons by glial phagocytosis rests on multiple, distinct signaling inputs and is executed via parallel avenues of neuron-to-glia communication.

Neuronal Prtp vs. APPL Regulation of Glial Phagocytosis.

We find that loss of either neuronal Prtp or APPL hampers glial phagocytosis neuron removal from the juvenile brain circuit but that loss of either signal alone causes a less severe impairment than blocking the neuronal FMRP-pMad pathway. These results suggest that Prtp and APPL act combinatorially downstream of FMRP-pMad, consistent with both neuronal Prtp and APPL driving the glial phagocytic clearance of targeted neurons (26, 27). Prtp is a transmembrane protein trafficked from the endoplasmic reticulum (ER) to the cell surface of a neuron thus marked for removal, where it binds the Drpr engulfment receptor on glia (26). Through this mechanism, Prtp surface presentation downstream of the neuronal FMRP-pMad pathway is proposed to directly signal glial engulfment. In contrast, the cleaved extracellular domain of transmembrane APPL from neurons activates the glial endolysosomal network (27). This provides a separable function for APPL, suggesting why two signaling ligands mediate neuron removal by the glial phagocytosis mechanism. The Prtp signaling changes closely match the neuron clearance phenotypes of all neuronal dfmr1, mad, InR, and akt RNAi experiments. Likewise, the APPL signaling changes are consistent with FMRP-pMad regulation but not the InR-pAkt pathway. We therefore propose Prtp and APPL coregulate glial phagocytic activity with different strengths or independently participate in glial phagocytosis via other regulatory mechanisms. To test the possibility that Prtp interacts with APPL in this clearance mechanism, we could test the putative cross talk interaction at molecular and/or genetic levels. Conversely, neuronal Prtp and APPL may be independent signals in neuron-to-glial communication linked predominantly or solely through upstream FMRP regulation. Future studies will dissect these combinatorial signals downstream of inputs targeting the clearance of neurons from the juvenile brain.

Neuronal InR vs. Glial InR Roles in Glial Phagocytosis.

The pigment-dispersing factor (PDF) brain circuit mediates circadian clock functions, with the developmentally transient central PDF-Tri neurons presumably driving eclosion timing (15). Multiple signals appear to coordinate the targeted clearance of PDF-Tri neurons from the juvenile brain, including insulin-like peptide (ILP) signaling (18). The insulin receptor (InR) is present in both neurons and glia, acting to coordinate intercellular communication and neuronal remodeling (1, 22, 53–55). For glial signaling, InRs are proposed to be activated by secreted ILPs from the remodeling neurons to stimulate transduction cascades promoting glial phagocytosis (18, 54). Glial InR phosphorylation triggers pAkt production, which, in turn, elevates Drpr engulfment receptor levels to drive glial phagocytosis (54). Consistently, genetically increasing glial InR levels elevates glia-dependent neuron removal of the PDF-Tri neurons (18). In the Drosophila FXS disease model, loss of FMRP represses InR phosphorylation in glia, whereas glial InR activation restores PDF-Tri neuron pruning (18). For neuron signaling, we propose that InR-mediated pAkt signal transduction regulates neuron-to-glia communication in this same remodeling process. Similarly, InR signaling in Drosophila dendrite arborization neurons inhibits their developmental pruning (22). Our work shows neuronal InR loss elevates Prtp signaling to drive glial phagocytosis, thus accelerating PDF-Tri neuron clearance from the juvenile brain. Neuronal InRs may be responding to autocrine ILP signaling from the remodeling neurons (18, 54) or ILPs from another source which is coordinating neuron and glia functions to properly regulate the glial phagocytosis neuron removal process. Future work will investigate how the different intercellular signals act on both neurons and glia to ensure the exact targeting and timing of the coordinated PDF-Tri neuron clearance from the juvenile brain.

Materials and Methods

Drosophila Genetics.

All Drosophila stocks were reared on cornmeal/agar/molasses food at 25 °C. Loss-of-function mutants used include dfmr1^50M (24) and appl^d Bloomington Drosophila Stock Center (BDSC# 43632) alleles (27). Transgenic drivers used include pan-neuronal elav-Gal4 (24), PDF-Gal4 (BDSC# 6899) (18), and glial repo-LexA (BDSC# 67096) (56). The controls include the w¹¹¹⁸ genetic background (24) and the transgenic drivers alone. All genetic crosses and recombinations to make double transgenic insertions were performed with standard methods. Transgenic stocks were obtained from the Bloomington Drosophila Stock Center (BDSC) and are listed in SI Appendix, Table S1. UAS lines alone have been tested to show no significant effect on the PDF-Tri neuron phenotype (SI Appendix, Fig. S6 A–C).

Immunocytochemistry.

Brains from synchronized, staged animals were dissected in phosphate-buffered saline (PBS), fixed in 4% paraformaldehyde in PBS for 30 min at room temperature (RT), and then blocked for 1 h at RT in PBS with 0.2% Triton X-100 (PBST) + 1% bovine serum albumin. Primary antibodies were diluted in the blocking buffer for incubation overnight at 4 °C, followed by washing 3X with PBST. Secondary antibodies were incubated for 3 h at RT and then washed 3X with PBST. Primary antibodies used were mouse anti-PDF [Developmental Studies Hybridoma Bank (DSHB), PDF C7,1:50], rabbit anti-Repo (gift from Dr. Benjamin Altenhein, University of Cologne, Germany, 1:1,000), chicken anti-GFP (Abcam, ab13970, 1:1,000), rabbit anti-Rab5 (Abcam, ab31261, 1:500), rabbit anti-pAkt (phospho S473, Cell Signaling, 4060S, 1:100), and mouse anti-Rab7 (DSHB, Rab7, 1:100). Secondary antibodies used were goat anti-mouse 488 (Invitrogen, A11001, 1:250), donkey anti-rabbit 555 (Invitrogen, A31572, 1:250), goat anti-chicken 488 (Invitrogen, A11039, 1:250), and goat anti-mouse 633 (Invitrogen, A21050, 1:250).

Confocal Imaging.

Brains were imaged on a Zeiss LSM 510 META laser scanning confocal microscope using 40X, 63X, and 100X oil immersion objectives (24). All collected images were projected in Zen software. Identical microscope setting was used to collect and analyze all genotypes for all trials within any given experiment. For three-dimensional (3D) images, the cropped Z-stack was processed with the 3D viewer icon in FIJI software to visualize PDF-Tri neurons, Repo-positive glia, and pAkt puncta. PDF-Tri neurons with glia colabeling Z-stacks were maximum projected in FIJI software to quantify the outlined neuronal area. Glial Rab5/7-labeled Z-stacks were sum projected to analyze total volumes. All Rab volumes were quantified only within labeled glia (repo-Gal4>mCD8::GFP membrane marker). The Rab volumes were divided by the imaged glial volume for normalization. The pAkt puncta within labeled PDF-Tri neurons were quantified based on 3D views. The number of pAkt puncta were divided by the PDF-Tri neuron area to quantify the pAkt puncta density.

Western Blots.

Staged heads (3 to 5 h after eclosion at 25 °C) were homogenized in lysis buffer (20 mM HEPES, 10 mM EDTA, 100 mM KCl, 1% (v/v) Triton X-100, and 5% (v/v) glycerol, pH 7.4) with 2X protease and phosphatase inhibitor cocktail (Thermo Scientific, WL334851). Homogenized lysate samples (50 μL lysis buffer/head) were incubated on a rotor for 15 min at 4 °C and then centrifuged at 14,000 rpm for 15 min at 4 °C. The supernatant was denatured with 1X NuPAGE LDS sample buffer (Invitrogen, NP0007) for 10 min at 100 °C. Protein samples were run in 4 to 15% Mini-PROTEAN TGX gels (Bio-Rad, 4568086) and then transferred onto PVDF membranes using the Trans-Blot Turbo Transfer System (Bio-Rad). PVDF membranes were blocked in TBS Intercept Blocking Buffer (Li-COR# 220121) for 1 h at RT, followed by incubating with primary and secondary antibodies. Primary antibodies used were rabbit anti-Smad3 (phospho S423+S425, Abcam, ab52903, 1:1,000), rabbit anti-pAkt (phospho S473, Cell Signaling, 4060S, 1:1,000), goat anti-GAPDH (glyceraldehyde 3-phosphate dehydrogenase) (Abcam, ab157156, 1:2,000), and chicken anti-APPL (a gift from Dr. Doris Kretzschmar, Oregon Health and Science University, USA; 1:1,000). Secondary antibodies used were goat anti-chicken 800 (Invitrogen, SA5-10076, 1:10,000), donkey anti-goat 680 (Invitrogen, A21084, 1:10,000), goat anti-mouse 800 (Invitrogen, SA535521, 1:10,000), and goat anti-rabbit 800 (Invitrogen, SA535571, 1:10,000). Blots were scanned with an Odyssey CLx Imager (Li-COR). Fluorescent band intensities were quantified using Image Studio Lite software (Li-COR). To normalize protein levels, the fluorescent intensity of targeted proteins was divided by the loading control of glyceraldehyde 3-phosphate dehydrogenase (GAPDH).

Quantitative Real-Time PCR.

RNA was extracted for ~30 heads at 1 day posteclosion (1 dpe). The RNA extraction procedure followed the instructions of the RNeasy Plus Micro Kit (Qiagen, 74034). RNA (2 μg) was reversely transcribed in 20-μL reaction mixture of SuperScript VILO Master Mix (Invitrogen, 11755-050), with cDNA (1 μL) then mixed with 19 μL diluted Power SYBR Green PCR Master Mix (Applied Biosystems, 4367659) for real-time PCR amplification using the Bio-Rad CFX96 system. The quantified PCR cycle number was normalized to the internal control GAPDH. All primers used in this study are listed in SI Appendix, Table S2.

RNA Immunoprecipitation.

The procedure was done as in Song et al. (24). Briefly; lysates were prepared from ~30 heads at 1 dpe with 500 μL RNase-free lysis buffer (20 mM HEPES, 100 mM NaCl, 2.5 mM EDTA, 0.05% (v/v) Triton X-100, 5% (v/v) glycerol, and 1% β-mercaptoethanol) with 1X protease inhibitor cocktail (cOmplete Mini EDTA-free tablets, Sigma-Aldrich) and 400 U RNase inhibitor (Applied Biosystems, N8080119). To precipitate RNA complexes, precleared lysates were incubated with 15 μL magnetic GFP-trap beads (ChromoTek) for 3 h at 4 °C, followed by washing 3X with lysis buffer. To purify bound RNAs, the washed beads were incubated with a 500 μL TRIzol and chloroform mixture (Ambion, 15596026) for 10 min. Subsequently, 1 µL glycogen was applied to carry RNAs for the precipitation by mixing with 100 µL 2-propenol. The above qPCR methods were performed to analyze the reverse-transcribed RNAs.

Statistical Analyses.

All statistical analyses were performed with GraphPad Prism software 9.0. All datasets were subject to ROUT outlier tests with Q set to 1%. Unpaired two-tailed Student’s t tests with 95% confidence of comparisons were used for all two-way datasets (PDF-Tri neuron areas, Rab7 volumes, pAkt puncta density, western blot assays, and qPCR assays). All datasets with more than two comparisons (day 1 and day 5 PDF-Tri neuron areas with dfmr1 RNAi, mad RNAi, prtp RNAi, and appl RNAi) were analyzed by two-way ANOVA tests with a 5% alpha significance level. The comparison with more than two datasets was analyzed using one-way ANOVA tests (PDF-Tri neuron area). Data are presented as the mean ± SEM. Significance is indicated by P value measurements shown as P < 0.05 (*), P < 0.01 (**), P < 0.001 (***), and P < 0.0001 (****). Values of P > 0.05 are deemed not significant (ns).

Supplementary Material

Appendix 01 (PDF)

Movie S1.

PDF-Tri neuron and glia interaction with nuclear labeling to show cell bodies Glial cells (anti-Repo nuclear labeling, magenta) contact a PDF-Tri neuron (anti-PDF, green) during early neuronal removal (0-1 dpe). Scale bar shown in μm on X, Y and Z axes.

Movie S2.

PDF-Tri neuron and glia interaction with membrane labeling to show surface Glia (repo-Gal4>mCD8::GFP membrane labeling, green) contact a PDF-Tri neuron (anti-PDF, magenta) during early neuronal removal (0-1 dpe). Scale bar shown in μm on X, Y and Z axes.

Data, Materials, and Software Availability

All study data are included in this article and SI Appendix.