Phagocytosis and self-destruction execute dendrite degeneration of Drosophila sensory neurons at distinct levels of NAD+ reduction

After injury, severed dendrites and axons expose the “eat-me” signal phosphatidylserine (PS) on their surface and degenerate by disassembly. While axon degeneration is controlled by a conserved “axon-death” pathway that is thought to activate self-destruction, how PS exposure is regulated by this pathway and whether PS-induced phagocytosis contributes to neurite breakdown in vivo remain unknown. Here we show that in Drosophila sensory dendrites, PS exposure and self-destruction are triggered by two distinct levels of NAD+ reduction downstream of Sarm activation. Surprisingly, phagocytosis is the main driver of dendrite degeneration induced by both genetic NAD+ disruptions and injury. Furthermore, the axon-death factor Axed is only partially required for self-destruction of injured dendrites, acting in parallel with PS-induced phagocytosis. Lastly, injured dendrites exhibit a unique rhythmic calcium flashing that correlates with self-destruction. Therefore, a special genetic program coordinates PS exposure and self-destruction in injury-induced dendrite degeneration in vivo.


INTRODUCTION 29
Physical insults to the nervous system often disrupt neuronal connectivity and function by 30 damaging dendritic or axonal processes of neurons. Injured axons break down through a series of 31 stereotypical events collectively called Wallerian degeneration (WD) (Coleman and Freeman,32 2010; Waller, 1850). Dendrites undergo a similar program of degeneration after injury (Tao and 33 Rolls, 2011). Before neurons can regenerate their processes and restore connections, the debris 34 from damaged neurites has to be promptly cleared by phagocytes, which are cells that engulf 35 dead cells or cell debris (Sapar and Han, 2019). Inefficient clearance can lead to 36 neuroinflammation and further exacerbate the damage to the surrounding tissues (Davies et al.,37 explain why dendrite degeneration was not detected previously in Nmnat mutant C4da neurons, 138 considering that the membrane GFP marker used to label dendrites in the previous study is 139 rapidly degraded by epidermal cells once engulfed and thus cannot visualize phagosomes (Han et 140 al., 2014;Wen et al., 2011). 141 Nmnat protects neurons by both synthesizing NAD + and functioning as a chaperon 142 protein (Ali et al., 2013). To verify that the observed dendrite degeneration is due to the loss of 143 Nmnat enzymatic activity, we tried to rescue Nmnat KO neurons by overexpressing Wld S , which 144 contains full NMNAT activity (Mack et al., 2001), and Wld S-dead , a mutant version of Wld S that 145 (H) Quantification of debris coverage ratio, which is the percentage of debris area normalized by dendrite area ratio. Number of neurons: same as in (G). Kruskal-Wallis (One-way ANOVA on ranks) and Dunn's test, p-values adjusted with the Benjamini-Hochberg method.
For all quantifications, ***p≤0.001; n.s., not significant. The significance level above each genotype is for comparison with the control. Black bar, mean; red bar, SD. cannot synthesize NAD + but maintains the chaperon function (Avery et al., 2009). Wld S 146 overexpression (OE) rescued the degeneration of Nmnat KO neurons and restored dendrite 147 morphology to the wildtype level ( Figures 1C, 1G, and 1H), while Wld S-dead OE did not change 148 dendrite length or debris level of Nmnat KO neurons ( Figures 1D, 1G, and 1H). These results 149 suggest that NAD + reduction is responsible for the dendrite degeneration of Nmnat KO neurons. 150 We next asked whether Sarm plays a role in Nmnat KO-induced dendrite degeneration. 151 Indeed, Sarm LOF completely blocked dendrite degeneration of Nmnat KO neurons ( Figures 1F,  152 1G, 1H), as evident in the absence of dendrite debris in epidermal cells. However, these neurons 153 showed more variable dendrite length ( Figure 1G) compared to Sarm LOF alone ( Figures 1E,  154 1G, and 1H), perhaps due to the loss of Nmnat chaperon function (Wen et al., 2011). 155 Importantly, these results demonstrate that Sarm is required for the dendrite degeneration of 156 Nmnat KO neurons. All the results above together suggest that Nmnat LOF in neurons cause 157 spontaneous dendrite degeneration through Sarm-mediated NAD + loss. 158

PS exposure-mediated phagocytosis causes dendrite degeneration of Nmnat KO neurons 159
To ask whether Nmnat KO also causes neuronal PS exposure, we used an established method to 160 visualize PS exposure on C4da dendrites. In this method, the fluorescent PS sensor GFP-Lact is 161 expressed by the larval fat body and secreted into the hemolymph (Sapar et al., 2018). As C4da 162 dendrites are largely exposed to the hemolymph, GFP-Lact labeling allows visualization of 163 dynamic PS exposure on dendrites in intact live animals (Sapar et al., 2018). We found that GFP-164 Lact strongly labeled Nmnat KO neurons at distal branches that underwent degeneration ( Figure  165 2B), while wildtype C4da neurons showed no labeling ( Figure 2A). Interestingly, we also 166 observed weaker GFP-Lact labeling on dendrite segments that did not display obvious signs of 167 degeneration (outlined in Figure 2B), suggesting that PS exposure may precede dendrite 168 breakdown, instead of being merely a consequence of dendrite degeneration. This conclusion  As Sarm is required for dendrite degeneration of Nmnat KO neurons, we asked if Sarm 173 also regulates PS exposure in these neurons. Nmnat KO neurons showed no PS exposure in the 174 Sarm mutant background ( Figure 2D), suggesting that Sarm-mediated NAD + reduction is also 175 responsible for inducing the observed PS exposure. 176  (J) Quantification of dendrite length. n = number of neurons: control (n = 13, 7 animals); Nmnat KO (n = 13, 7 animals); drpr -/-(n = 14, 8 animals); Nmnat KO + drpr -/-(n = 22, 14 animals); ATP8A OE (n=17, 9 animals); Nmnat KO + ATP8A OE (n=23, 9 animals). One-way ANOVA and Tukey's test. (K) Quantification of debris coverage ratio. Number of neurons: same as in (J). Kruskal-Wallis (One-way ANOVA on ranks) and Dunn's test, p-values adjusted with the Benjamini-Hochberg method. (L) A time series of Nmnat KO + drpr -/dendrites. Yellow arrowheads indicate growth of dendrites compared to the previous frame; blue arrows indicate retractions of dendrites compared to the previous frame. Scale bar, 10 μm. See also Video 2. In all panels, neurons were labeled by ppk-CD4-tdTom (A-H) and ppk-MApHS (I). C4da-specific KO was induced by ppk-Cas9. For all quantifications, *p≤0.05, **p≤0.01, ***p≤0.001; n.s., not significant. The significance level above each genotype is for comparison with the control. Black bar, mean; red bar, SD. We previously found that overexpressing Wld S in C4da neurons blocked fragmentation and PS 196 exposure of ablated dendrites at 10 hrs after injury (AI) (Sapar et al., 2018). To further 197 investigate the role of the WD pathway in neuronal PS exposure, we examined the effects of 198 overexpressing Wld S and knocking out Sarm in C4da neurons at 24 hrs after laser-severing of 199 dendrites. Neuronal-specific KO was conducted using SOP-Cas9, which is active in precursor 200 cells of da neurons, to minimize potential gene perdurance (Poe et al., 2019). As expected, Wld S 201 OE blocked degeneration and clearance of injured dendrites also at 24 hrs AI ( Figure 3C, as 202 compared to the control in Figure 3B), even though the injured arbors were greatly simplified as 203 compared to uninjured dendrites ( Figures 3A and 3E Figures 3H and 3I). The above data in dendrite injury together suggest that 210 Sarm-mediated NAD + reduction causes both PS exposure and degeneration of injured dendrites. 211 (I) Quantification of GFP-Lact binding on injured dendrites. The GFP-Lact intensity is shown for both background epidermal regions and injured dendrites. The outliers in Sarm KO dendrite dataset correspond to infrequent degenerating terminal branches. Background: epidermal regions without dendrites. n = number of measurements: control background (n = 39) and control dendrite (n/a, no remaining dendrites), 7 animals; Wld S OE background (n = 18) and Wld S OE dendrite (n = 18), 4 animals; Sarm KO background (n = 48) and Sarm KO dendrite (n = 48), 9 animals. In all panels, neurons were labeled by ppk-MApHS (A-C; only tdTom channel is shown), ppk>CD4-tdTom (D), and ppk-CD4-tdTom (F-H). Neuronal-specific KO was induced by SOP-Cas9. For all quantifications, One-way ANOVA and Tukey test; ***p≤0.001; n.s., not significant; n/a., not applicable. The significance level above each genotype in (E) is for comparison with the control. Black bar, mean; red bar, SD. Scale bars, 25 μm.

Ectopic PS exposure causes injured dendrites of Wld S OE and
For all quantifications, *p≤0.05, ***p≤0.001; n.s., not significant. The significance level above each genotype is for comparison with the control. Black bar, mean; red bar, SD. data strongly suggest that injury-induced PS exposure is a major driver of dendrite breakdown by 230 activating phagocytic attack by epidermal cells. 231
For all quantifications, **p≤0.01; ***p≤0.001; n.s., not significant. The significance level above each genotype is for comparison with the control. Black bar, mean; red bar, SD. The remaining axons of peb KO neurons did not show defects in axon degeneration after injury 286 ( Figures S1E-S1G). These results suggest that Peb may be a neuronal type-specific modulator of 287 the WD pathway and is not required for dendrite degeneration of da neurons. 288

Injured dendrites undergo severe membrane rupture during dendrite fragmentation 289
To further understand how injury induces dendrite degeneration, we investigated the extent of 290 membrane rupture during dendrite breakdown using a split GFP-based assay. In this membrane 291 rupture assay ( Figure 7A), neurons express myristoylated tdTom-GFP(1-10) and the fat body 292 secretes GFP(11)x7 into the hemolymph. GFP(11)x7 will bind myr-tdTom-GFP(1-10) and 293 reconstitute fluorescent GFP only when the dendrite membrane is ruptured to allow diffusion of 294 GFP(11)x7 into the cytoplasm of neurons. While reconstituted GFP was not detected in 295 uninjured wildtype dendrites ( Figure S2A) or degenerating dendrites of Nmnat KO neurons 296 ( Figures 7B, open arrowheads, and 7D-7E), it was observed in severed wildtype dendrites that 297 were undergoing blebbing and fragmentation (Figures 7C, arrowheads, and 7E), as well as on 298 membrane pieces shed from injured dendrites ( Figure 7C, arrows, and 7D). In time-lapse movies, 299 GFP signals were visible at low levels in injured branches soon after laser ablation, likely due to 300 GFP(11)x7 entry through the injury site. The signals were kept at constantly low levels until 301 injured dendrites fragmented, when GFP signals rapidly increased in large dendrite particles 302 (Video 3). These results suggest that injured dendrites undergo severe membrane rupture during 303 fragmentation. In contrast, Nmnat KO neurons experience much milder disruptions of membrane 304  integrity, even though they are losing membranes due to the attack of epidermal cells, probably 305 because their dynamically growing branches are efficient in repairing membrane damages. 306

Injured dendrites exhibit calcium flashing that is suppressed by Wld S and axed loss 307
The observed membrane rupture of injured dendrites is consistent with phagocytic attacks of 308 epidermal cells on PS-exposing dendrites ( Figure 7C). To understand potential signaling events 309 that may lead to dendritic PS exposure and membrane rupture, we examined calcium dynamics  (Figures 8B and 8C, Video 7). Considering 334 that Annexin V-binding to PS and accumulation on dendrite surface are likely slower than 335 calcium activation of GCaMP6s, our data support the idea that PS-mediated phagocytosis causes 336 dendrite membrane rupture and the final calcium surge. 337 (L) Quantification of average flashing amplitude (GCaMP6s brightness difference between maximum peak and neighboring minimum peak) captured in 2 s-interval time-lapse images. n=number of dendrite arbors: wildtype 0.5-1.5 hrs AI (n= 19, 5 animals); Wld S -overexpressing 2 hrs AI (n=8, 4 animals); axed KO 0.5-1 hrs AI (n=7, 3 animals). One-way ANOVA and Tukey's test. For all quantifications, ***p≤0.001; n.s., not significant. The significance level above each genotype is for comparison with the control. Black bar, mean; red bar, SD.

Because the unique pattern of calcium flashing is absent in uninjured dendrites and 338
Nmnat KO neurons, we suspected that it may play an active role in promoting degeneration of 339 injured dendrites. If so, factors that can block dendrite degeneration may also alter the calcium 340 flashing. Indeed, Wld S OE dramatically reduced calcium fluctuations in injured dendrites and 341 eliminated the quiescent and surge phases for the entire duration of our time-lapse imaging (13 342 hrs) (Figures 8E, 8J, S3B, and Video 8). In addition, using time-lapse imaging at a higher 343 temporal resolution (2 s/frame), we found that wildtype injured dendrites displayed calcium 344 flashes at a frequency of 0.4-3/min (Figures 8G and 8K, and Video 9) while injured dendrites of 345 before the quiescence and surge phases (Figures 8F, 8J, S3C, and Video 12), which was further 349 confirmed by time-lapse imaging at the high temporal resolution (Figures 8I, 8K, 8L, S3C, and  350 Video 13). Overall, injured dendrites of Wld S OE and axed KO neurons showed much smaller 351 average amplitudes of calcium fluctuations compared to the wildtype control ( Figure 8L). PS to engage in phagocytosis-mediated non-autonomous degeneration. Therefore, in our revised 392 model, dendrites respond to at least three distinct, increasingly severe levels of NAD + reduction 393 by eliciting different molecular events (Figure 9). Between the NAD + level required for Sarm 394 activation (SA level) and the level that initiates self-destruction (SD level), Sarm activity lowers 395 NAD + to a level that causes neurons to expose PS on their surface (which we call the PSE level). 396 This PS exposure is sufficient to cause phagocytosis-mediated dendrite degeneration, which can 397 be completely prevented by blocking engulfment activity of phagocytes. However, below the SD 398 level, neurites spontaneously fragment even in the absence of phagocytosis. 399 Our results also suggest a direct correlation between the kinetics of NAD + reduction and 400 the severity of neurite degeneration. Nmnat KO is expected to cause slow NAD + reduction, due 401 to gene perdurance and the time required for natural NAD + turnover, and correspondingly causes 402 engulfment-dependent dendrite degeneration only in late 3 rd instar larvae. In contrast, Sarm GOF 403 OE should lead to a more rapid NAD + depletion and in fact causes engulfment-dependent 404 dendrite degeneration as early as the 1 st instar and dendrite self-destruction by the 3 rd instar. 405 Injury is known to cause even more rapid NAD + reduction in axons (Wang et al., 2005) and is 406 correlated with the fastest dendrite degeneration -initiation at around 4 hrs AI and completion 407 usually by 10 hrs AI. 408 How does NAD + reduction cause PS exposure? A direct consequence of NAD + loss is the 409 decline of neurite ATP levels due to the requirement of NAD + in glycolysis and oxidative Genetic analyses in Drosophila identified Axed as a key switch of WD, whose activity is 420 absolutely required for axon degeneration caused by injury and genetic depletion of NAD + 421 (Neukomm et al., 2017). How does Axed regulate neurite degeneration? Our data suggest that 422 Axed is not required for PS-mediated phagocytosis but contributes to self-destruction of injured 423 dendrites, placing its activation below the SD level of NAD + (Figure 9). Surprisingly, Axed 424 seems to play a minor role in dendrite degeneration, as its LOF only slowed down but did not 425 block self-destruction, indicating the existence of other factors that promote self-destruction of 426 injured dendrites. Importantly, these results suggest that dendrites and axons may rely on 427 different mechanisms of self-destruction.

Live imaging 508
Animals were reared at 25°C in density-controlled vials (60-100 embryos/vial) on standard 509 yeast-glucose medium (doi:10.1101/pdb.rec10907). Larvae at 125 hours AEL (wandering stage) 510 or stages specified were mounted in 100% glycerol under coverslips with vacuum grease spacer 511 and imaged using a Leica SP8 microscope equipped with a 40X NA1.30 oil objective. Larvae 512 were lightly anesthetized with isoflurane before mounting. For consistency, we imaged dorsal 513 ddaC neurons from A1-A3 segments (2-3 neurons per animal) on one side of the larvae. Unless 514 stated otherwise, confocal images shown in all figures are maximum intensity projections of z 515 stacks encompassing the epidermal layer and the sensory neurons beneath, which are typically 8-516 10 μm for 3 rd instar larvae. 517

Injury assay 518
Injury assay at the larval stage was done as described previously (Sapar et al., 2018). Briefly, 519 larvae at 90 hr AEL were lightly anesthetized with isoflurane, mounted in a small amount of 520 halocarbon oil under coverslips with grease spacers. The laser ablation was performed on a Zeiss 521 LSM880 Confocal/Multiphoton Upright Microscope, using a 790 nm two-photon laser at 522 primary dendrites of ddaC neurons in A1 and A3 segments. Animals were recovered on grape 523 juice agar plates following lesion for appropriate times before imaging. imaging after ablation, larvae were pre-mounted in the imaging chamber and subjected to laser 542 injury. The larvae were then imaged 0.5-1 hours after ablation. For calcium imaging before and 543 immediately after ablation, images were captured on a Zeiss LSM880 Confocal/Multiphoton 544 Upright Microscope on which the ablation was performed. 545

Image analysis and quantification 546
Image processing and analyses were done in Fiji/ImageJ. Methods for tracing and measuring 547 C4da neuron dendrite length have been previously described (Poe et al., 2017). Briefly, the 548 images were segmented by Auto Local Threshold and reduced to single pixel skeletons before 549 measurement of skeleton length by pixel distance. The dendrite debris measurement has been 550 described previously (Sapar et al., 2018). Briefly, a dendrite mask was first generated from 551 projected images by Auto Local Threshold in order to create a region of interest (ROI) by 552 dilation to map areas within one-epidermal-cell diameter (40 μm) from dendrites. Dendrite debris 553 within the ROI was converted to binary masks based on fixed thresholds. Different thresholds 554 were used for ppk-C4-tdTom and ppk-Gal4 UAS-CD-tdTom as they have different brightness. 555 The dendrite pixel area (ADen), debris pixel area (ADeb), and ROI area (AROI) were measured 556 and dendrite coverage ratio was calculated based on the following formula: 100ꞏAdebꞏAROI 557 /(AROI-ADen)ꞏADen. For measuring Lact-GFP, two regions at empty epidermal regions were 558 measured as background levels. TdTom signals on dendrites were used to generate dendrite 559 masks for measurement of GFP within the masks. For kymographs, we used a custom macro 560 based on the Straighten function to extract a strip of pixels centered at the selected dendrite 561 branch. The maximum intensity pixel in the strip at each distance was used to generate a single-562 pixel line for each time frame. The final kymographs were displayed using the Fire lookup table 563 (LUT). 564

Statistical Analysis 565
R was used to conduct statistical analyses and generate graphs. (*p < 0.05, **p < 0.01, and ***p 566 < 0.001). Statistical significance was set at p < 0.05. Data acquisition and quantification were 567 performed non-blinded. Acquisition was performed in ImageJ (batch processing for debris 568 coverage ratio and fragmentation ratio, manually by hand for GFP-Lact binding) and Microsoft 569 Excel. Statistical analyses were performed using R. We used the following R packages: car, stats, 570 multcomp for statistical analysis and ggplot2 for generating graphs. Some graphs were made in 571 Excel using its native plotting functions. For the statistical analysis we ran the following tests, 572 ANOVA (followed by Tukey's HSD) when dependent variable was normally distributed and 573 there was approximately equal variance across groups. When dependent variable was not 574 normally distributed and variance was not equal across groups, we used Kruskal-Wallis 575 (followed by Dunn's test, p-values adjusted with Benjamini-Hochberg method) to test the null 576 hypothesis that assumes that the samples (groups) are from identical populations. To check 577 whether the data fit a normal distribution, we generated qqPlots to analyze whether the residuals 578 of the linear regression model is normally distributed. We used the Levene's test to check for 579 equal variance within groups. The state of neuronal degeneration or fragmentation was compared 580 using the Freeman-Halton extension of Fisher's exact test. dendrites. Timestamp is relative to the first frame, with a 3-min interval between each frame. 784 Video 6. Calcium dynamics in injured wildtype dendrites, related to Figure 8. 785 Time-lapse movie of laser-injured C4da dendrites from 1 to 11 hrs AI showing calcium 786 dynamics in both severed dendrites and those attached to the soma. Timestamp is relative to the 787 first frame, with a 3-min interval between each frame. 788 Video 7. AV-mCard labeling and calcium dynamics of injured dendrites, related to Figure  789

790
Time-lapse movie of laser-injured C4da dendrites from 1 to 6 hrs AI showing labeling of injured 791 dendrites by the PS sensor AV-mCard and GCaMP6s signals. Timestamp is relative to the first 792 frame, with a 3-min interval between each frame. 793 Video 8. Calcium dynamics in injured Wld S OE dendrites, related to Figure 8. 794 Time-lapse movie of laser-injured Wld S OE C4da dendrites from 1 to 13 hours AI showing 795 GCaMP6s signals in severed dendrites. Timestamp is relative to the first frame, with a 3-min 796 interval between each frame. 797 Video 9. Calcium dynamics in injured wildtype dendrites with a higher temporal 798 resolution, related to Figure 8. 799 High temporal-resolution time-lapse movie of laser-injured wildtype C4da dendrites around 1 hr 800 AI showing calcium dynamics. Timestamp is relative to the first frame, with a 2-sec interval 801 between each frame. 802 Video 10. Calcium dynamics in injured Wld S OE dendrites at a higher temporal resolution, 803 related to Figure 8. 804