High-resolution in vivo imaging of regenerating dendrites of Drosophila sensory neurons during metamorphosis: local filopodial degeneration and heterotypic dendrite–dendrite contacts

Neuronal circuits that are formed in early development are reorganized at later developmental stages to support a wide range of adult behaviors. At Drosophila pupal stages, one example of this reorganization is dendritic remodeling of multidendritic neurons, which is accomplished by pruning and subsequent regeneration of branches in environments quite distinct from those in larval life. Here, we used long-term in vivo time-lapse recordings at high spatiotemporal resolution and analyzed the dynamics of two adjacent cell types that remodel dendritic arbors, which eventually innervate the lateral plate of the adult abdomen. These neurons initially exhibited dynamic extension, withdrawal and local degeneration of filopodia that sprouted from all along the length of regenerating branches. At a midpupal stage, branches extending from the two cell types started fasciculating with each other, which prompted us to test the hypothesis that this heterotypic contact may serve as a guiding scaffold for shaping dendritic arbors. Unexpectedly, our cell ablation study gave only marginal effects on the branch length and the arbor shape. This result suggests that the arbor morphology of the adult neurons in this study can be specified mostly in the absence of the dendrite–dendrite contact.


Introduction
The dendritic arbor of the neuron is the subcellular compartment that receives and processes synaptic or sensory inputs, thereby effecting appropriate responses of animals (Hausser & Mel 2003;London & Hausser 2005). Because its architecture influences the neuronal function, for example, by restriction of the number and type of inputs, it is important to investigate how dendrites acquire their characteristic size and morphology (Jan & Jan 2010). The spatial pattern of the dendritic arborization is not fixed by branch elongation in early development, and it is also sculpted at later stages by branch retraction and elimination, and the mature pattern is maintained throughout animal life (Wong & Ghosh 2002;Luo & O'Leary 2005).
Da neurons grow two-dimensional dendrites underneath the epidermis and on the musculature during late embryonic and larval stages Kim et al. 2012). During metamorphosis, larval dendritic arbors of surviving neurons are totally pruned and subsequently regenerated, and many of the rebuilt arbors persist throughout adult life (Williams & Truman 2005a;Shimono et al. 2009;Yasunaga et al. 2010). This large-scale structural conversion from larval to the adult type is considered to be one example of dendritic and axonal reorganizations without death of the parent neurons, which is crucial for supporting stage-specific behaviors of the animal (Consoulas et al. 2000;Luo & O'Leary 2005). It should be noted that after the pruning, regenerating neurons do not simply recapitulate dendrite formation in early development, because they are exposed to environments that are distinct in spatial and hormonal respects from those in the embryo and larva. For example, larval da neurons regulate their growth in coordination with the expanding body wall (Parrish et al. 2009); however, pupal da neurons somehow have to control their arbor shape and size in a body whose volume has been predetermined by the nutritional status during larval development (Hietakangas & Cohen 2009;Mirth & Shingleton 2012). Thus, investigating programs controlling the dendritic remodeling is expected to provide novel mechanistic insights into dendrite morphogenesis.
One prominent advantage of observing pupal da neurons is that they are accessible to long-term in vivo time-lapse recordings. In fact, regeneration and elaboration of a dorsolateral da neuron, ddaE, can be tracked for up to nearly 2 days, and this neuron actively migrates to a new position on the body wall (Williams & Truman 2004). We previously provided a systematic anatomical map of all persistent neurons (Shimono et al. 2009), whereby we can study intrinsic and extrinsic mechanisms of dendritic remodeling in vivo. In this study, we focused on three other da neurons: ldaA, ldaA-like and v'ada whose expansive arbor occupies the entire lateral plate of the abdomen (right in Fig. 1; also shown later in Fig. 3G). ldaA and ldaA-like are a pair of closely associated cells that develop arbors of similar morphology (Shimono et al. 2009), and we hereafter designate them as ldaA/Alike that belong to a single cell type. These two cell types, v'ada and ldaA/A-like, protruded filopodia from regenerating branches and underwent cycles of extension and retraction and local degeneration. Branches extending from the two cell types overlapped with each other, and we explored its functional relevance.  Shimono et al. (2009). (middle) Selected frames of the pruning and regeneration phases that focused on the ventral region of the hemi-segment (see also Movie S1 in Supporting Information). Pupal stages are indicated as hours after puparium formation (APF). In these images, we were able to track v'ada (magenta arrow) throughout this recording, whereas ldaA/ldaA-like was first identified approximately 40 h APF (magenta arrowhead). A posterior-directed branch remained at 28 h APF (white arrowhead) and was subsequently pruned.

Results and Discussion
Dynamic extension, withdrawal, degeneration and stabilization of filopodia of regenerating branches We tracked v'ada neurons from the large-scale pruning phase of the dendrites until their regeneration over a period of 2 days ( Fig. 1 and Movie S1 in Supporting Information). After larval branches were pruned and cleared, v'ada cell bodies started shifting ventrally (arrows in Fig. 1). Subsequently what appeared on the focal planes approximately 40 h after puparium formation (APF) was ldaA/A-like with ventrally extending branches (arrowheads in Fig. 1), which encountered dorsally growing branches of v'ada. Below, we describe our detailed observations of regenerating v'ada branches and then focus on the dendrite-dendrite contact between v'ada and ldaA/ A-like neurons.
To monitor dynamics of dendrite regeneration, we performed time-lapse recordings for several hours at high spatiotemporal resolution, at multiple pupal stages. As soon as larval dendrites were pruned, dynamic processes protruded from the v'ada cell body ( Fig. 2A), followed by the emergence of primary branches ( Fig. 2B-D). Tips of the primary branches were capped with growth cone-like structures (Movie S2 in Supporting Information). The processes longer than 20 lm were distributed along the entire length of the branches, and they underwent active extension and withdrawal. Throughout this study, we designate these dynamic processes as filopodia and distinguish them from branches that were thicker and more stable.
Until approximately 55 h APF, we occasionally observed degeneration of filopodia ( Fig. 2E-E" and Movie S3 in Supporting Information). Compared to the branch pruning by 20 h APF (Williams & Truman 2005a), this degeneration was more local and rapid. Degradation was restricted to the filopodia, whereas primary branches were not severed or retracted. Detached filopodia were entirely fragmented within 10 min, in contrast to an hour that was taken by the branch pruning (e.g., see fig. 3 of Williams & Truman 2005a). This time difference may reflect distinct cytoskeletal organizations between the degenerating filopodia and the pruned branches that are rich in microtubules (Williams & Truman 2005a;Lee et al. 2009). The degeneration could be reminiscent of axonal degradation in the zebrafish peripheral nervous system, where repulsive interactions between isoneuronal axons occasionally execute local degradation and control shapes and sizes of sen-sory arbors (Sagasti et al. 2005;Grueber & Sagasti 2010), and also in the nociceptive neuron PVD of Caenorhabditis elegans, where dendrites are locally broken off after dynamic outgrowth, and it is proposed that this series of events eliminates extra branches (Oren-Suissa et al. 2010;Smith et al. 2010). It remains to be elucidated whether the degradation we found contributes to final spatial patterns of regenerating dendritic branches or not. The axonal degeneration in the fish system is triggered also by contacts between different cells; likewise, we did find instances where branches of v'ada encountered those of ldaA/A-like and then filopodia were degraded (described later in Fig. 4A and Movie S6 in Supporting Information).
At 60 h APF onwards, the arbor underwent a persistent increase in complexity and size with a rapid expansion of primary branches and generation of higher-order branches, and still displayed high filopodial activity on both tips and stalks of the branches ( We previously imaged dendrites by using pan-neuron or pan-da neuron markers and found that the ldaA/ A-like arbor is intermeshed with that of v'ada in the adult (right in Fig. 1; Shimono et al. 2009). To address how the two neuronal types develop individual dendritic arbors and whether dendritic branches of the two interact with each other, we differentially labeled them (Fig. 3).
ldaA/A-like expanded its arbor ventrally in a unidirectional fashion, whereas v'ada developed a more radial arbor (Fig. 3A-D). Ventrally directing branches of ldaA/A-like encountered dorsal ones of v'ada approximately 50 h APF and they partially fasciculated with each other, which was seen at later stages as well (yellow arrows in Fig. 3C-G; high-power images are shown later in Fig. 4). At 60 h APF, ldaA/A-like arbors were more elaborated and larger than those of v'ada, and branches of these two neuronal types looked tightly associated with each other (Fig. 3E). The differences in arbor size and complexity of the two cell types were reversed after 72 h APF, making v'ada more expansive than ldaA/A-like ( Fig. 3F). These spatiotemporal profiles of the dendritic development raised the possibility that ldaA/ A-like, starting its arbor elaboration first, may serve as a guidepost to facilitate dorsal extension of later-growing v'ada branches, and that this dendritedendrite action eventually contributes to the establishment of adult dendritic arbors. This speculation prompted us to observe the dendrite-dendrite contact in more detail.
Time-lapse recordings showed that at least some of the filopodia sprouting from leading branches of v'ada and ldaA/A-like degraded upon initial contacts ( Fig. 4A and Movie S6 in Supporting Information). Nonetheless, branches of both neurons started contacting with each other, and the overlapping region extended over time (white arrowheads in Fig. 4A). Thus, it is unlikely that the degeneration in this context elicits mutual repulsion of the branches. High-power images of the regions at 50 and 60 h APF showed that the overlap occurred not only between the branches, but also between filopodia and branches, and between filopodia of the different neuronal types (Fig. 4B,C). In 1-day-old adults, the v'ada arbor exhibits prominent radial-to-lattice transformation (Shimono et al. 2009;Yasunaga et al. 2010); the fasciculation was seen both in dorsoventral (DV) segments of the branches, which are embedded between muscles, and in anterior-posterior (AP) segments as well (Fig. 4D).
Roles of the dendrite-dendrite interaction have been shown in vivo in the context of avoidance between homotypic cells and self-avoidance of isoneuronal branches (Grueber & Sagasti 2010;Jan & Jan 2010;Matsubara et al. 2011;Lefebvre et al. 2012). In contrast, the heterotypic contact we observed implied its role as a guiding scaffold, as seen in many instances of local and/or transient cell-cell interactions in organizing neural circuit formation (Chao et al. 2009;Grueber & Sagasti 2010;Jan & Jan 2010).

Attempts to selectively ablate ldaA/A-like by toxic gene expression
To address the above hypothesis, we used a system to eliminate either ldaA/A-like or v'ada selectively by expressing cell-death-inducing genes before the contact took place. We found that the expression of a dominant negative Rab5, DRab5[S43N] (Entchev et al. 2000), efficiently killed ldaA/A-like when driven by C161-Gal4 (Shepherd & Smith 1996), and arbor formation of v'ada looked impaired in those pupae (results not shown). However, this Gal4 system also induced gene expression in v'ada, and although the expression was weaker than that in ldaA/A-like (see details in the Fig. 3 legend), it complicated the interpretation of the results. In an attempt at shutting off the leaky expression in v'ada, we combined C161-Gal4 with ppk-Gal80 (Yang et al. 2009). Unfortunately, however, ppk-Gal80 repressed Gal4 activity in ldaA/ A-like as well as v'ada and could not be used for ldaA/ A-like selective elimination (results not shown).
Therefore, we searched for Gal4 stocks that could induce gene expression only in ldaA/A-like before or from the onset of the v'ada-ldaA/A-like contact. Of 156 lines tested, including a collection of taste-receptor gene drivers (Weiss et al. 2011), we isolated seven new Gal4 lines that were specific to ldaA/A-like ( Fig. S1 and Table S1 in Supporting Information). Unfortunately, however, none of them was early or strong enough for our purpose.

Laser ablation of ldaA/A-like produces only marginal effects on the v'ada arbor
Instead of genetically eliminating the neurons, we ablated ldaA/A-like using a laser and examined how the arbor size of v'ada was affected (Fig. 5). Because the origin of ldaA-like is not known (Shimono et al. 2009), the earliest stage when we were able to target both ldaA and ldaA-like was 40-42 h APF before the contact. We ablated ldaA/A-like in abdominal hemisegment 4 or 5 (A4 or A5) and then imaged v'ada arbors at 72 or 85 h APF in the irradiated hemi-segments and in control hemi-segments (Fig. 5A-D). For each arbor, we quantified total branch length and the growth index along the DV axis (Fig. 5E-H) and assessed the effect of the ablation of ldaA/A-like on the v'ada arbor growth.
We had expected poor growth of at least dorsal branches of v'ada in the absence of ldaA/A-like in the same hemi-segments. However, visual observation did not show such gross morphological defects (Fig. 5A,  B). Consistently, our quantitative analysis at the two time-points showed that the ablation of ldaA/A-like affected both parameters of v'ada arbors only marginally ( Fig. 5E-H), and it was difficult to conclude that the ablation caused any significant differences. We also performed a converse ablation experiment, where v'ada was ablated either in wandering larvae or at 40-42 h APF and imaged ldaA/A-like at 72 h and 96 h APF. Again, we did not find obvious morphological defects in shaping arbors of the remaining neurons ( Fig. S2 in Supporting Information; n = 5). Although we could not exclude the possibility that branches of the ablated neurons were invisible but somehow remained partially intact long after the ablation, our results under the conditions used showed that elongation and branching of dendrites did not appear to slow down in the absence of the heterotypic dendrite-dendrite contact and that the shape and size of the dendritic arbor in the regeneration phase could be specified almost normally. It could still be the case that the heterotypic contact might contribute in a longer term to restricting overshooting of branches or maintaining their arbors during adult life.
Other possible mechanisms that shape dendritic arbors other than the dendrite-dendrite contact It has been shown that the extracellular matrix (ECM) secreted by the epidermis makes significant contributions to shaping dendritic arbors of da neu-rons. In larvae, dendrite-substrate interactions ensure preventing crossings of isoneuronal branched or selfavoidance Kim et al. 2012), whereas local degradation of the ECM in newly eclosed adults plays a pivotal role in reshaping radial arbors of v'ada into the lattice shape (Yasunaga et al. 2010). It could be that branch elongation of v'ada and ldaA/A-like neurons may be particularly promoted in a spatially restricted fashion by an unknown ECM distribution between the cell bodies of the two cell types, which secondarily results in the dendrite-dendrite contact. How the arbor growth is restricted at the end of pupal development remains to be elucidated. The expansive arbor of v'ada occupies the entire pleura (the lateral plate of the adult abdomen), but its dorsal and ventral borders do not necessarily about those of ddaE neuron in the tergite and v'ada in the contra-lateral hemi-segment, respectively (see fig. 3 in Shimono et al. 2009). Given that the pupal dendrites cope with spatial and hormonal environments distinct from those in larvae, we should be able to attain novel mechanistic insights into shaping dendritic arbors with the v'ada arbor as an assay system and the in vivo time-lapse imaging developed in this study.

Image acquisition of whole-mount animals and quantitative analysis
To acquire images of da neurons at pupal stages, the collected pupae were transferred to a plate filled with water, washed and dried on 3MM Chr paper (Whatman). Each pupa was taken out of its puparium carefully by forceps, mounted in PBS on a slide between spacers made of vinyl tape, and covered with a 24 9 24 mm cover slip (No. 1, MATSUNAMI). For imaging, adult abdomens, heads, wings and legs were removed and mounted in 50% glycerol as described above. Images of ldaA/A-like and v'ada neurons at A4 and/or A5 were acquired by laser scanning confocal microscopy (Ni-konC1) with 488-or 543-nm lasers with a 1-lm Z-step and total 15-20 lm depth. Images were processed by using Photoshop, Illustrator (Adobe Systems), and ImageJ. Neurocyte software (Kurabo) was used for the quantification of total dendritic length. Statistical analysis was performed by R program (version 2.14.0; The R Foundation for Statistical Computing).

Time-lapse recording
Each pupa was taken out of its puparium as described above and mounted on a 35-mm glass-bottomed dish (3911-035, IWAKI). In mounting, we folded legs and put abdomens on the dish and tilted them to retain an appropriate angle to observe v'ada and/or ldaA/A-like neurons. Details of image acquisition are essentially described previously (Harumoto et al. 2010), except for Movie S1 (Supporting Information). Briefly, Movies S2-S6 (Supporting Information) were acquired by using a spinning-disk confocal scan head (CSU10; Yokogawa), an Olympus IX71 microscope and an EM-CCD camera (DU-888; Andor Technology). Fluorescent proteins were excited with a 488-nm line and a 561-nm line, and signals were detected with a 500-to 550-nm and 580-to 640nm band-pass filter, respectively. The exposure time was 500 ms for both 488 and 561 nm, and the EM gain of the camera was 10009. The output of the 488-and 561-nm laser power was 10% and 40%-50%, respectively. Typically, single confocal planes were taken with a 0.7-lm Z-step and total 15 -20 lm depth, at 30-s or 60-s intervals. The above hardware was driven by MetaMorph (Molecular Device), and acquired data were processed with MetaMorph and ImageJ. After image acquisition, each pupa was kept at 25°C and their survival was confirmed to at least the pharate stage or the adult stage. Movie S1 (Supporting Information) was acquired by using a Leica TCS SPE (see its legend).

Ablation experiment
In our ablation experiments, pupae were mounted as described in 'Image acquisition of whole-mount animals and quantitative analysis', except water was substituted for PBS. Both cell bodies and trunks of dendrites of ldaA/A-like were targeted at 40-42 h APF through a 1009 UPlanApo objective that was attached to a microscope (BX51; OLYMPUS), and ablated by using a 337-nm N2 laser at a frequency of 20 Hz for 5-10 s (Micropoint, Photonics Instruments). This condition was harsher than what was used for ablating da neurons in embryos and larvae (Sugimura et al. 2003). Fluorescence of cell bodies was no longer detected after 1 min. The pupae were kept at 25°C until 72 or 85 h APF. To collect six hemi-segments at 72-74 h APF where both ldaA and ldaA-like were ablated, we targeted 19 hemi-segments at 40-42 h APF (Fig. 5E,F). Similarly, we targeted 12 hemi-segments at 40-42 h APF to collect five hemi-segments at 85-87 h APF where the ablation was successful (Fig. 5G,H).  Fig. 3C,E,G, respectively. Dorsal dendrites of v'ada (left, green in the merged image at right) and ventral dendrites of ldaA/A-like (middle, magenta at right). Green, red and yellow arrows mark the overlap of filopodia and branches between v'ada and ldaA/A-like. In the middle panels of 'B' and 'C', note that the cell body (B) and proximal dendrites (B and C) of v'ada were labeled more weakly than ldaA/A-like. Orange arrow in 'D' points to a spiracle. Genotype: (A) 477-Gal4/UAS-mCherryCAAX, UAS-mmRFP; ppk-EGFP/ppk-EGFP (B and C) UAS-mCherryCAAX, UAS-mmRFP/+; C161-Gal4, ppk-EGFP/ppk-EGFP, and (D) ppk-CD4::tdTom/ppk-CD4::tdTom; C161-Gal4, UAS-mCD8::GFP/ +. Scale bars, 25 lm. Figure S1 ldaA/A-like-specific Gal4 drivers. Figure S2 ldaA/lda-A like developed arbors in the absence of v'ada branches.