Daily ultrastructural remodeling of clock neurons

SUMMARY

Animals have endogenous time-keeping mechanisms to measure time and adjust their physiology to cyclical environmental changes. These biological clocks rely on neural networks to orchestrate circadian rhythmicity. Some clock neurons undergo a daily remodeling of their morphology -known as circadian structural plasticity- that is expected to impact their connectivity and function. This process remains poorly defined at the subcellular level, preventing a full understanding of how it relates to other cyclical properties. In this study, we generated 3D electron microscopy reconstructions of adult Drosophila brains at three key time points to examine ultrastructural changes in the terminals of core clock neurons. We found that neuronal varicosities are the organizational units underlying these structural changes, as they contain the functional elements required for neuronal communication (active zones and dense core vesicles) and for their metabolic support (mitochondria). Varicosities appear to change in number during the day-night cycle, while exhibiting differences in the number of active zones, in the accumulation of dense-core vesicles, both full and fused, and in the number, shape and size of the mitochondria. These results suggest an interplay between structural and functional plasticity that was not appreciated to date. We propose that circadian plasticity of presynaptic varicosities modulates the influence of specific clock neurons onto the circadian network. Given the conservation across timekeeping mechanisms, ultrastructural changes might underlie circadian shifts in neuronal connectivity across species.

ETOC Blurb

A cluster of Drosophila clock neurons remodel their axonal arbors daily. Using volumetric electron microscopy at different times of day, Ispizua, Rodriguez-Caron and colleagues reveal ultrastructural cycles—affecting synapses, vesicles, and mitochondria—that accompany the remodeling, linking structural plasticity to changes in the neurons’ capacity to influence downstream targets.

Graphical Abstract

INTRODUCTION

The selective pressure imposed by the day-night cycle through evolutionary time has favoured the emergence of circadian clocks. These are endogenous mechanisms that exist in almost all forms of life, and align biochemical, physiological and behavioral functions to the external environmental cycles, thereby increasing fitness^1,2. In the fruit fly Drosophila melanogaster, the circadian clock comprises a network of almost 250 neurons³. Among them, the small Lateral ventral Neurons (s-LNvs, four per hemisphere) express and release Pigment-Dispersing Factor (PDF), a neuropeptide that is necessary to maintain rhythmic behavior under constant laboratory (aka free-running) conditions^4–8. These neurons also express short Neuropeptide F (sNPF), regulating feeding and sleep, and the inhibitory neurotransmitter glycine, which contributes to the synchronization of the circadian network^9–11. The s-LNvs project their axons to the dorsal protocerebrum, a key neuropile for brain integration. There, they exhibit rhythmic and stereotyped changes in their arborization pattern¹², being more elaborated in the morning and less so at night. Indirect markers of neuronal activity, such as anti-PDF immunoreactivity¹³ and the abundance of BRP^short-RFP (which labels presynaptic sites¹⁴), are also rhythmic and peak in the morning¹⁵. Indeed, those markers and the complexity of the axonal arbour increase and decrease upon forced depolarization and silencing, respectively, linking structural remodeling to neuronal activity^16–18. This form of structural plasticity is not exclusive to the s-LNvs: other neurons in the fly’s clock circuit and optic neuropile also show similar remodeling, and analogous pacemaker neurons in the mouse brain show circadian cycles of growth and retraction^19–22. These observations suggest that cyclical structural remodeling may represent a common property of clock neurons. Nevertheless, the need to rely on indirect markers to track multiple cellular components without affecting cellular physiology^23,24 undermines a comprehensive understanding of the phenomenon.

In this work, we explored how these manifestations of rhythmic neuronal activity interconnect at the subcellular level, using quantitative volumetric electron microscopy (EM). We analyzed three timepoints representing a key moment during the structural remodeling cycle: Zeitgeber Time 22 (ZT22, two hours before lights-ON), ZT2 (two hours after lights-ON), and ZT14 (two hours after lights-OFF). We labeled the s-LNvs by expressing an EM-detectable marker targeted to the mitochondria²⁵. We then generated 3D volumes of their terminals using Serial Block-face Scanning Electron Microscopy (SBEM) and developed a semi-automated convolutional neural network (CNN) algorithm to aid 3D reconstruction. We show that the peptide-filled dense-core vesicles (DCVs) accumulate and fuse to the plasma membrane mainly in the morning, in parallel with a peak in the number of presynaptic sites. Moreover, such changes are associated with a marked shift in mitochondrial shape and volume between early morning and night. Finally, the number of varicosities —regular membrane enlargements that function as en passant boutons—also cycles, being higher in the morning. We strengthened our findings using fluorescent reporters to study structural changes at the population level, and confirmed that daily oscillations are dependent on a functional circadian clock. We propose that the s-LNvs modulate their influence onto the circadian network by adjusting both the number and the content of their varicosities throughout the day. Under this scenario, daily variations in neuronal transmission are realized through the discrete and regulated change of the varicosities. Thus, the s-LNvs re-align the release of neurotransmitters and neuropeptides in space and time, while optimising mitochondrial architecture for better metabolic support.

In summary, we report ultrastructural evidence of a rhythmic change in the number and relative composition of varicosities in the s-LNvs, which points to a daily assembly/disassembly cycle. We suggest it may represent an efficient strategy to orchestrate daily fluctuations in connectivity, and ensure rhythmic but coherent transmission of information to the circadian network by the s-LNvs.

RESULTS

Mitochondrial labeling allows rapid segmentation of Drosophila neurons in 3D EM volumes

To investigate circadian plasticity at the ultrastructural level, we generated 80×80×100 μm 3D EM volumes containing the s-LNv terminals. To guide the identification of the neurons of interest we took advantage of APEX2, an engineered pea peroxidase²⁵. The enzyme catalyses the deposition of osmophilic diaminobenzidine (DAB), which is visualized as electron dense in EM images. To avoid obscuring cytoplasmic details of synaptic terminals we generated a novel reporter line targeting APEX2 to the mitochondrial matrix. In addition, to facilitate the identification of the region of interest through correlated light and electron microscopy (CLEM), APEX2 was fused to the fluorescent protein mKO2. The resulting construct, UAS-mito::mKO2::APEX2 (referred to as mito::APEX2) was expressed along Pdf-GAL4 specifically in the lateral ventral neurons (LNvs, small, s- and large). We first confirmed that its expression does not affect the circadian clock per se, by assessing locomotor activity patterns (Figure S1B and Table S1). As expected, mito::APEX2 expression in the LNvs resulted in a dotted confocal signal. To assess neuronal structure we co-expressed cytoplasmic RFP in the LNvs using Pdf-RFP.

Based on previous studies, we chose to process for EM the time points showing the most dramatic structural changes in the terminals of the s-LNvs: early morning (ZT2), early night (ZT14) and late night (ZT22). Briefly, ZT2 shows maximum spread of the neuronal terminals, ZT14 shows minimum spread with no change in total length, and ZT22 shows retracted terminals¹⁵. Given the labour-intensive nature of the technique, we analyzed one brain per timepoint, selected for stereotypy and overall sample quality (Figure 1A). Female brains were used to allow comparison to existing connectomes^26–28. DAB-labeled mitochondria (Figure 1B, arrowheads) guided the manual segmentation of the images and the generation of models (Figure 1C) with IMOD, a program used for 3D reconstruction from electron tomography or EM serial sections²⁹.

Figure 1. Mitochondrial labeling allows rapid segmentation of Drosophila neurons in 3D EM volumes.

(A) Schematic diagram of the experimental protocol. RFP expressed in PDF-positive neurons enabled the identification of the s-LNv terminals. mito::APEX2 and DAB stained the mitochondria for SBEM. (B) Labeled mitochondria (white arrowheads) identified the s-LNv terminals. (C) 3D-models of the s-LNv terminals after manual segmentation showing them together (left) or individually (right) at each time point. (D) Representative neurites from each volume display the segments defined as primary (magenta), secondary (green) or tertiary (purple) neurites and boutons (yellow). Primary neurites are defined as the longest projection extending from the fasciculated axon bundle, secondary neurites are those stemming from the primary ones, etc. Short protrusions including a terminal varicosity are labeled as boutons. (E) The graph shows the quantification of the neurite length according to their order (as defined in d). The total number of neurites per order and boutons is indicated at the top. (F) Volume of each terminal per time point (n = 4 neurons). In all graphs, error bars indicate the standard error of the mean (SEM). Asterisks indicate statistically significant differences: * p < 0.05, ** p < 0.01, *** p < 0.001. Non-significant differences are not shown. Details can be found in Table S4. See also Figures S1, S2, S5, and Table S1.

A direct comparison of the segmented neurons with the corresponding confocal images obtained exploiting mKO2 and RFP expression provided compelling evidence that the former correspond to the s-LNvs (Figure S2). To make the segmentation of subsequent 3D-EM volumes more accessible, we used the data obtained from ZT2 and ZT14 to train a semantic classifier to recognize and segment cells containing labeled mitochondria. Furthermore, we developed an algorithm utilizing convolutional neural networks for the segmentation of free and fused DCVs (Figure S3).

The unequivocal identification of individual neurons enabled the detailed assessment of the complexity of the neurites throughout the day. We defined as primary neurites the neuronal processes that originate from the cell body, usually the longest, while secondary neurites originate from the primary, and so forth (Figure 1D). Interestingly, while the total number of primary, secondary and tertiary neurites remained relatively constant (Figure 1E, numbers above symbols) their length changed according to the time-of-day. Generally, secondary and tertiary neurites were longest at ZT2. This, together with the tendency of having more varicosities and terminal boutons, would explain the complex morphology of the early morning¹⁵. Conversely, the overall volume was significantly reduced late at night (Figure 1F). We confirmed that the varicosities distributed along the s-LNv neurites function as en passant boutons concentrating all the elements required for communication (mitochondria, DCVs, presynaptic sites)³⁰. Throughout the manuscript the varicosities will be the focus of our analyses.

Daily changes in DCV recruitment and release

DCVs carry neuropeptides from the soma to the neuronal terminals. PDF and sNPF are the main neuropeptides produced by the s-LNvs³¹, contributing to the synchronization of the clock neurons and to the regulation of sleep, respectively^4,10,32. Using the manually segmented EM models, we identified both free DCVs and those that had entered the release path (fused DCVs, fDCVs). We observed differences in the number of free and fused DCVs across the day (Figures 2A–2F), but for the most part, DCVs were localized within varicosities at any time (Figures S5D and S5G). The total number of free DCVs was higher at ZT2 compared to ZT14 and ZT22 (Figures S5B and S5C). The number of DCVs in the medial and distal portions of the terminals showed significant differences at night (Figure 2C).

Figure 2. Daily changes in DCV recruitment and release.

(A and D) On the left panels, 3D-models of neuronal terminals showing free (blue dots, A) and fusing (red dots, D) DCVs. Scale bars = 10 μm. On the right panels, representative EM images show DCVs and fDCVs (white arrowheads). The plasma membrane is highlighted in blue and red in A and D, respectively. Scale bars = 1 μm. ZT2, ZT14, and ZT22 indicate the time points analyzed. (B and E). Density per neuron (total number/volume of the neuronal terminal) of free (B) and (E) fusing DCVs at each time point (n = 4 neurons). (C and F) Graphs show the quantity (number/varicosity) of free (C) and fusing DCVs (F) per varicosity at each time point. Quantification in proximal, medial and distal refers to the number of objects segmented in the initial, subsequent and terminal thirds of each neurite. Thirds are defined by the length of the longest neurite (See Figure S5A for details); n = 10–29 varicosities. (G) Schematic diagrams showing the experimental design (top) and the logic of the GFP reconstitution in the pre-synapse (GRIP) experiments (bottom). Pdf-GS > GRIP animals were raised in normal food and then induced with RU486 supplemented food for 96 h prior to dissection. DCVs decorated with Sytα::GFP₁₁ fuse to the plasma membrane containing CD4::GFP_1–10 to release their content. Thus, GFP is reconstituted at release sites. (H) s-LNv terminals expressing cytoplasmic RFP (magenta) and GRIP (green). The images are maximum intensity Z projections of confocal images acquired with an Airyscan detector. Individual varicosities are shown in insets 1 and 2. Scale bars = 10 and 1 μm. (I) Representative confocal images of GRIP (maximum intensity Z projections); white dashed lines indicate the terminal outline. Scale bars = 10 μm. (J) GRIP signal at each time point. Each symbol represents the averaged signal value for the terminals of one brain. Darker symbols correspond to the images shown in I. N (number of experiments) = 2. n₁ (sample size per time point)=13, 9, 9 brains. n₂=15, 12, 10 brains. In all graphs, the sample size per time point is indicated above the symbols, except in the analyses per neuron, in which n = 4. Error bars indicate the standard error of the mean (SEM). Asterisks indicate statistically significant differences: * p < 0.05, ** p < 0.01, *** p < 0.001. Non-significant differences are not shown. Details can be found in Tables S2–S4. See also Figures S1, S4, S5, and Table S1.

fDCVs (Figures S5E and S5F), albeit fewer than free DCVs, were also more numerous in the early morning. Thus, DCV release may be a function of the quantity of DCVs available. Consistent with previous observations, fDCVs were largely excluded from presynaptic densities³³. The number of DCVs and fDCVs per terminal volume (DCV and fDCV density, Figures 2B and 2E) were higher at ZT2. However, at ZT22, fDCV density increased dramatically, in parallel with the growing excitability of the s-LNvs towards dawn³⁴.

To confirm the daily regulation of neuropeptide release in a larger sample, we designed a new reporter of DCV fusion based on the recovery of fluorescence upon reconstitution of two split (non-fluorescent) GFP components. We called it GFP reconstitution in the presynapse (GRIP) because of its design. Briefly, one of the split components was fused to Synaptotagminα and specifically targeted to the lumen-facing side of the DCV membrane (UAS-Sytα::GFP₁₁). The complementary part was tethered to the extracellular portion of the transmembrane carrier protein CD4 (UAS-CD4::GFP_1–10³⁵). The concomitant expression of both fragments in the same neuron enables the reconstitution of fluorescent GFP at the sites of fusion. Thus, GFP fluorescence becomes a proxy of DCV release (Figure 2G, lower panel). The design of the GRIP reporter relies on the specific trafficking of Sytα to DCVs³⁶, but to further validate this tool we immunostained PDF and Sytα::GFP_1–10, which should both decorate DCVs, and quantified their co-occurrence in s-LNv dorsal terminals. Nearly all the PDF staining co-occurred with the Sytα::GFP_1–10 signal, underscoring its specificity (Figure S4A). In addition, we examined the locomotor activity profiles of the reporter lines and confirmed that they are rhythmic (Figure S1C, upper panel).

When using the GRIP reporter at ZT2, ZT14 and ZT22 (Figure 2G, upper panel), most of the fluorescence was observed within the varicosities (Figure 2H), as described for the EM dataset. Notably, DCVs fusion peaked early in the morning, reaching a minimum at the beginning of the night, and increased again towards the end of the night (Figures 2I and 2J), consistent with our observations made through SBEM in single volumes (Figure 2E). Moreover, these changes were abolished in animals kept under constant light (LL), a condition known to eliminate molecular and behavioral rhythmicity^37,38 (Figure S1C, lower panel), further supporting the circadian regulation of neuropeptide availability and release (Figures S4C–S4E).

Presynaptic sites are more abundant in the morning

Synaptic sites are trans-cellular structures that organize fast neurotransmission between neurons. At the presynaptic site (PS), many proteins organized in complex assemblies facilitate the docking of neurotransmitters-containing clear vesicles to the plasma membrane and their release after a depolarization event^39,40. PSs were detected in the SBEM images (Figure 3A, model and arrowheads). They were reminiscent of T-bars and were surrounded by dense accumulations of unresolved structures that are likely clear vesicles. The majority of the synapses, if not all, were polyadic – contacting more than one postsynaptic neuron – and situated almost exclusively in varicosities and terminal boutons (Figure 3A, right panel, asterisks, and Figure S5J). As reported previously³³, these structures serve as exclusive docking sites for small clear vesicles, with no evidence of dense core vesicle release at those locations.

Figure 3. Presynaptic sites are more abundant in the morning.

(A) Segmented PSs in 3D models of neuronal terminals (left, green dots; scale bars = 10 μm) and in representative EM images (right, white arrowheads; scale bars = 1 μm). White asterisks indicate putative post-synaptic structures. (B) Total number of PSs per time point including the objects located outside of the varicosities. n = 4 neurons. (C) Number of PSs per varicosity per region at each time point. n = 10–29 varicosities. (D) PS density (total number/neuronal volume) at each time point. n = 4 neurons. In all graphs, the sample size per time point is indicated above the symbols, except for the analyses per neuron (see Figure 2 legend). Error bars indicate the standard error of the mean (SEM). Asterisks indicate statistically significant differences: * p < 0.05, ** p < 0.01, *** p < 0.001. Non-significant differences are not shown. Details can be found in Tables S2 and S4. See also Figure S5.

The morning sample exhibited a two-fold increase in the total number of PSs in comparison to the volumes obtained at night (Figure 3B). Such a difference results from a slightly higher number of PSs per varicosity in the medial and distal segments (Figure 3C, Table S2) together with a higher number of varicosities at ZT2 (Figure 5B, right panel). When the number of presynaptic sites was normalized by the total volume of each neuron, the presynaptic density showed a decrease at the beginning of the night period, followed by an increase at ZT22, which might be explained by the reduced volume of the terminals at that time (Figure 3D). Altogether these results confirm that differences in terminal complexity are associated with changes in presynaptic connectivity.

Figure 5. Varicosities are plastic throughout the day.

(A) Representative model of the varicosities observed at different time points. Individual mitochondria are labeled with asterisks in different colors. Scale bars = 1 μm. (B) Total number of varicosities per neuron and region (n = 4 neurons). (C) Distance between consecutive varicosities at each time point (n = 49–53). (D) Varicosity girth at each time point (n = 50–64). In all graphs, the sample size per time point is indicated above the symbols, except for the analyses per neuron (n = 4). Error bars indicate the standard error of the mean (SEM). Asterisks indicate statistically significant differences: * p < 0.05, ** p < 0.01, *** p < 0.001. Non-significant differences are not shown. Details can be found in Tables S3 and S4.

Circadian structural plasticity correlates with changes in mitochondria shape and number

Mitochondria are dynamic organelles that exist in a fused-fragmented equilibrium. In extremely polarized cells, like neurons, they are usually found in close proximity to active zones⁴¹. They provide energy and serve as a Ca²⁺ reservoir⁴². In the s-LNv terminals, we identified marked time-of-day differences in the total number and morphology of mitochondria (Figure 4A, Figure S5K), which were mostly detected within varicosities (Figure S5M). No significant differences were observed in the total volume occupied by mitochondria (Figure 4B). Interestingly, at ZT2 we observed small and rounded mitochondria, usually more than one per varicosity (Figure 4A, right panel, arrowheads). However, at the end of the night we observed fused and elongated structures (Figure 4A), as indicated by a higher mitochondrial complexity index (MCI, Figure 4C). MCI approaches 1 in round mitochondria, whereas values >1 denote progressively complex and elongated shapes⁴³ (Figure 4C, left panel). Our data suggests that during the four-hour interval between ZT22 and ZT2, the elongated mitochondria characteristic of late-night undergo fission. To confirm these observations in a larger sample, we resorted to the fluorescent reporter mito::GFP⁴⁴ to assess, through confocal microscopy, mitochondria number and complexity at ZT2, ZT14, and ZT22. Consistent with the EM data, we detected mitochondria both in the center of the varicosities and as elongated structures within small processes (Figures 4D and 4E). Additionally, we confirmed both the reduction in the number of mitochondria and the increase in MCI towards the end of the night (Figures 4G and 4H). Upon transfer to LL, time-of-day differences were dampened and phase shifted compared to LD (Figure S1D, lower panel, and Figures S4F–S4H). This confirms circadian regulation but also uncovers a strong homeostatic control over mitochondria number and complexity.

Figure 4. Circadian structural plasticity correlates with changes in mitochondria shape and number.

(A) Segmented mitochondria (taken as individual objects, hence labeled in different colors) in modeled neurons (left, scale bars = 10 μm) and in representative EM images (right, scale bars = 1 μm) at each time point (white arrowheads). (B) Total mitochondrial volume per neuronal volume at the indicated time points (n = 4 neurons). (C) Mitochondrial complexity index (MCI) per mitochondrion per time point. Each symbol represents one mitochondrion (n = 82–144). Examples of different shapes of mitochondria along with their corresponding MCI are shown on the left. (D) Z projection of s-LNv terminals expressing mito::GFP. (E) Varicosities and processes from the terminals in (D) expressing RFP in the cytoplasm (magenta) and GFP in the mitochondria (green). Mitochondria are located both, in the center of the varicosities and within the thin processes as well. (F) The image shows the mito::GFP signal in (D) together with the segmentation of the structures as recognized by the Mitochondria Analyzer plugin. (G) Total number of mitochondria per brain per time point. Each symbol represents one brain (n = 16–18). (H) MCI per mitochondrion per time point (n = 1465–2014). The variable was transformed using the natural logarithm and is shown as log (MCI). In all graphs, the sample size for each time point is indicated above the symbols, except for the analyses performed per neuron where n = 4. Error bars indicate the standard error of the mean (SEM), except for (H), which shows the median and the quartiles. Asterisks indicate statistically significant differences: * p < 0.05, ** p < 0.01, *** p < 0.001. Non-significant differences are not shown. Details can be found in Tables S2–S4. See also Figures S1, S4, S5, and Table S1.

Varicosities are plastic throughout the day

Neuronal varicosities appear like beads-on-a-string structures in both insects and mammals, acting as discrete communication units dispersed along the axon⁴⁵. In the s-LNv terminals, varicosities have an approximate diameter of 1.8 μm at ZT2. At all timepoints, they contain mitochondria and presynaptic sites and function as fusion hubs for neuropeptide-loaded DCVs (Figure 5A and Figures S5D, S5G, S5J and S5M). The total number of varicosities per neuron was higher in the morning than at night, although the differences were not uniform, with the medial and distal parts of the volumes showing the largest variation (Figure 5B). By contrast, the distance distribution between varicosities remained stable at all time points (Figure 5C), suggesting that new varicosities are added at rather regular intervals. Their girth, however, was reduced at ZT22 (Figure 5D).

Given their structural and functional features, we propose that changes in varicosity number and content link structural remodeling to functional plasticity. In this framework, daily changes in varicosity number and content provide a modular strategy for adjusting connectivity and information transfer within the circadian network.

DISCUSSION

Semi-automated pipeline for 3D-EM segmentation

Volumetric electron microscopy is an effective high resolution imaging method, generating extensive datasets from intricate three-dimensional samples. In this study, we developed a combination of an established marker (mito::APEX2) with a new informatics interface to significantly reduce the processing load of 3D-EM data (Figure S3). The mito::APEX2 marker, when used in concert with either the GAL4 or LexA expression systems, enables cell-specific targeting while circumventing the considerable opacity associated with membrane-bound APEX2, which has the potential to obscure synaptic sites and other fine ultrastructural features. Furthermore, it is possible to identify a cell without having to produce a volume that encompasses it entirely. This enables the reduction of the area of interest to specific zones of interest. Two manually segmented volumes (ZT2 and ZT14) were employed to train a semantic classification network to recognize labeled mitochondria. This resulted in the development of models for both the mitochondria and the neurons that contained them. Furthermore, an algorithm utilizing convolutional neural networks was developed for the additional segmentation of DCVs. Together, these advances provide a streamlined and accessible pipeline for targeted 3D-EM volume generation, expanding the potential of EM and connectomics^26–28 to address specific biological questions.

Circadian control of vesicle transport and release

Neurons use neuropeptides to convey messages that extend beyond individual synapses, producing broader cellular changes compared to the membrane potential shifts typically induced by classical neurotransmitters⁴⁶. The s-LNvs produce at least two types of neuropeptides, PDF and sNPF. Both neuropeptides are transported within DCVs to the dorsal terminals of the s-LNvs, where they are captured until release⁴⁷.

Since the early 2000s, it has been known that PDF levels cycle in a circadian manner, peaking in the morning within the dorsal terminals of the s-LNvs¹³. However, the dynamics of neuropeptide accumulation and release have only been analyzed indirectly, through fluorescent protein fusions with exogenous peptides⁴⁸. Our work highlights the circadian nature of DCV accumulation within the dorsal terminals of the s-LNvs. We observe that free DCVs accumulate largely within the s-LNv varicosities, surrounding the mitochondria. Our data further shows that this accumulation decreases significantly at night, especially in the distal varicosities. A similar pattern is observed in DCVs undergoing fusion with the plasma membrane: fusion predominantly occurs within the varicosities and outside the presynaptic densities, a pattern also described using fluorescent markers and expansion microscopy³³. Furthermore, we demonstrate that, similar to free DCVs, there are significantly more fusion events during the morning compared to nighttime, with a more pronounced effect in distal varicosities. The detection of fused DCVs at night indicates that, although most of the peptides are expected to be released from the s-LNv terminals at dawn — associated to the higher frequency of bursts of action potentials^34,49— these neurons still release neuropeptides at night, despite the circadian variation in firing frequency^49–51. This was further confirmed in acutely fixed brains through GRIP. The reconstituted GFP signal was mainly restricted to varicosities, and the degree of fusion at the time points examined, judged by GFP intensity, closely resembles the EM data. In fact, a similar daily pattern was previously described employing a fluorescent DILP chimera at the s-LNv dorsal terminals, further confirming a temporal control in DCVs release⁵².

Interestingly, distal varicosities exhibited the greatest degree of plasticity in terms of the number of DCVs they accommodate. Both the number of free and fused vesicles varied with a similar phase. This indicates that circadian modulation is likely to affect vesicle transport towards the dorsal projections, and that the number of vesicles fusing to the membrane may be contingent upon their availability.

The morning structure is more closely integrated with the network

Chemical synapses are functional structures that organize communication between neurons. The presynaptic neuron fuses vesicles filled with classical neurotransmitters, the postsynaptic cell receives and interprets these neurotransmitters via membrane receptors, which induce downstream changes in membrane potential. In electron microscopy images of the Drosophila nervous system, presynaptic sites appear as clusters of clear vesicles organized around a T-bar structure formed by scaffold proteins, whereas postsynaptic sites only exhibit a slightly darker contrast in the membrane adjacent to the synaptic cleft^30,53. Upon segmenting these structures in our volumes, we noticed that most synapses are polyadic, meaning that a single presynaptic density is associated with multiple postsynaptic neurons, consistent with previous observations^26,27,54. Building on previous observations locating presynaptic sites to varicosities in brain slices, we confirmed that this organization extends across the entire s-LNv processes, with synapses almost exclusively confined to varicosities. While the number of presynaptic sites per varicosity was largely stable, we found a trend towards fewer presynaptic densities at night, particularly in distal regions. In any case, the morning organization displays approximately 50% more presynaptic sites than the two volumes taken at night (Figure 3B), consistent with previous reports based on samples examined using fluorescent protein labeling and 3D laser-scanning light microscopy¹⁵.

Changes in mitochondrial structure imply variation in the signaling valence of the terminals and/or in energy demand

In most varicosities, vesicles and presynaptic sites are organized around one or more mitochondria. Given that both active transport and vesicle fusion rely on ATP, an organization into functional units appears to facilitate a localized ATP supply, thereby supporting essential communication processes. A recent study on mitochondrial morphology across the hemibrain demonstrated a correlation between mitochondrial shape and the identity of neuronal processes. Specifically, dendrites were observed to contain elongated mitochondria, while axons exhibited smaller, rounder ones⁵⁵. In this work, we show that the s-LNv terminals have small and round mitochondria in the morning, and elongated and more complex mitochondria at night. Prior studies have shown that these terminals exhibit both pre- and post-synaptic markers. However, the abundance of somatodendritic markers increases at night⁵⁶, together with the enhanced responsiveness to the optogenetic activation of different groups of neurons that are their likely circadian partners^15,20. Taken together, these observations suggest that the increase in mitochondrial complexity at night may be indicative of a functional transition from “axonal” to “dendritic” terminals, which would revert on the next morning.

An alternative, but not mutually exclusive hypothesis is that changes in mitochondria morphology indicate a stress response. Just before dawn, the s-LNvs begin to fire at higher frequencies³⁴, and at the same time they increase their presynaptic sites and show augmented recruitment and fusion of DCVs. Such an increased demand for energy could result in mitochondrial stress, resulting in the fragmented, rounded forms characteristic of ZT2. Conversely, at night, as neuronal activity decreases, mitochondria tend to fuse into larger, elongated structures with reduced total surface area. This configuration may enable collective evaluation of mitochondrial health and membrane potential equalization⁵⁷, preparing the mitochondria for another round of metabolic stress at the start of the new day.

In mammals, some evidence suggests that the circadian clock controls the abundance and morphology of mitochondria by regulating biogenesis, fission/fusion and mitophagy⁵⁸. Along those lines, the reduction in overall mitochondrial volume at the night onset could be indicative of clearance of damaged mitochondria through mitophagy. While electron microscopy images are unable to capture movement, observations of elongated mitochondria outside varicosities at night suggest that dismantling these functional units may contribute to the reduced complexity observed during this period.

Varicosity changes link structure to function

Circadian structural plasticity was initially described through Z-axis projections derived from confocal microscopy images of brain samples collected at different time points. Thus, changes in complexity were characterized using Sholl analysis⁵⁹ and later on, defined by an image analysis algorithm⁵⁶. Hypotheses regarding the types of subcellular changes underlying oscillations in complexity included the growth/disappearance of new neurites, the extension/retraction of existing neurites, and changes in the degree of fasciculation (i.e., proximity between neurites)¹². However, distinguishing between these alternatives requires identifying individual neurites, an impossible task given the resolution of confocal microscopy. While labeling individual neurons randomly could potentially circumvent this limitation, these stochastic labeling strategies not only exhibit low efficiency within the s-LNvs, but also do not consistently label the same neuron⁶⁰. Our data suggests that the number of neurites is similar across the three time points taken for analysis: in general, each neuron has one or two secondary neurites and only a few show tertiary neurites. Hence, changes in structural complexity would depend primarily on the growth/retraction (length) of individual neurites together with the generation of new varicosities and terminal boutons. Given the subtle nature of these changes, they are likely obscured by inherent structural differences across time points and the limited number of observations achieved with this level of resolution.

Interestingly, the high-resolution technique employed herein enabled the identification of additional changes. One such change was related to volume, whereby terminals occupied a greater space in the morning compared to the night, reminiscent of what was described in the L2 lamina neurons¹⁹. Another change was observed in the number of varicosities. This novel finding reveals a new dimension of circadian remodeling as it demonstrates that structural changes involve not only alterations in size or complexity, but also the addition of new, larger varicosities containing new synapses and peptide release sites. While most varicosities exhibited mitochondria, free and fused DCVs and PSs, a small fraction (about 5–6%) did not include any mitochondrion, while still containing free and fused DCVs and PSs. No varicosities contained exclusively PSs and, interestingly, no “empty” varicosities were found at any given time point.

This work provides evidence for daily changes in the size and volume of the s-LNv terminals, in addition to changes in the total number of varicosities, PSs, free and fused DCVs and mitochondria. These dramatic structural changes are likely to reflect functional differences, a claim further supported by the clear shift in the number and complexity of mitochondrial morphology, i.e. bioenergetics, across the day. The data collected thus far do not provide unequivocal evidence regarding the mechanisms that underlie the assembly and disassembly of varicosities. In the mammalian brain, high-frequency stimulation has been shown to induce axonal remodeling, whereby a transient enlargement of boutons is followed by a sustained widening of the axons, which in turn leads to functional changes⁶¹. Clock neurons in flies and mice exhibit circadian changes in membrane excitability, causing differential firing rates throughout the day^62–64. While this phenomenon likely contributes to the structural plasticity described herein¹⁵, it is tempting to speculate that there may be circadian components regulating membrane and/or vesicle biogenesis and transport, to support the wide extent of structural remodeling we uncovered. These observations and hypotheses, which have important implications for clock disorders and other neurological diseases, will inform our future research.

RESOURCE AVAILABILITY

Lead contact

Requests for information and resources should be directed to M. Fernanda Ceriani (fceriani@leloir.org.ar).

Materials availability

All data generated or analyzed during this study are included in this published article and its supplementary information files. Any other data can be requested to the lead contact.

Data and code availability

The custom-developed algorithm, designed for segmentation of mito::Apex2 labeled EM volumes, is publicly accessible on the GitHub repository https://github.com/labCeriani/SegmentEM. Researchers can freely download, modify, and utilize the code to encourage widespread adoption and collaboration. A detailed README file within the repository provides comprehensive instructions for installation, usage, and a breakdown of the algorithmic approach. For inquiries, please open an issue on the repository or contact francisco.tassara0@gmail.com or fceriani@leloir.org.ar.

STAR METHODS

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Fly rearing

All fly strains used in this manuscript are detailed in the Key Resources Table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Chicken polyclonal anti-GFP	Aves Labs	Cat#GFP-1020; Lot#GFP3717982; RRID: AB_10000240
Rabbit anti-RFP	Rockland	Cat#600-401-379; Lot#48710; RRID: AB_2209751
Rat polyclonal anti-PDF	Depetris-Chauvin et al 16	N/A
Alexa Fluor 488-conjugated AffiniPure donkey anti-chicken	Jackson ImmunoResearch Labs	Cat#703-545-155; Lot#160696; RRID: AB_2340375
Cy3-conjugated AffiniPure donkey anti-rabbit	Jackson ImmunoResearch Labs	Cat#711-165-152; Lot#79424; RRID: AB_2307443
Cy5-conjugated AffiniPure donkey anti-rat	Jackson ImmunoResearch Labs	Cat#712-175-153; Lot#134917; RRID: AB_2340672
Chemicals, peptides, and recombinant proteins
RU486, Mifepristone (≥98%)	Sigma-Aldrich	Cat#M8046; CAS#84371-65-3
Glutaraldehyde 25% solution	Electron Microscopy Sciences	Cat#16220 Lot#2200218
Paraformaldehyde, prills	Electron Microscopy Sciences	Cat#19202 Lot#190402-07
Cacodylic acid, sodium salt, trihydrate	TED PELLA	Cat#18851 Lot#210928-10 CAS#6131-99-3
Glycine	Electron Microscopy Sciences	Cat#16200
Potassium ferrocyanide	Baker analyzed A.C.S.	Cat#3114
Thiocarbohydrazide	Electron Microscopy Sciences	Cat#21900
L-aspartic acid	Sigma Aldrich	Cat#A93100 CAS#56-84-8
Lead nitrate	Electron Microscopy Sciences	Cat#17900
Durcupan ACM resin, Component A,M	Sigma-Aldrich	Cat#44611
Durcupan ACM resin, Component B	Sigma-Aldrich	Cat#44612
Durcupan ACM resin, Component C	Sigma-Aldrich	Cat#44613
Durcupan ACM resin, Component D	Sigma-Aldrich	Cat#44614
Osmium tetroxide 4% Aqueous Solution	Electron Microscopy Sciences	Cat#19140
3,3′-Diaminobenzidine	Sigma-Aldrich	Cat#D8001-10G Lot#WXBD3217V CAS#91-95-2
Sodium Hydroxide solution, 1N	Electron Microscopy Sciences	Cat#21170-01 Lot#190606-11
DRAQ5	Invitrogen	Cat#65-0880-92 Lot#DR05201
Vectashield	Vector Laboratories	Cat#H-1000; Lot#ZF0221
Triton X-100	Sigma-Aldrich	Cat#T9284; CAS#9036-19-5; Pcode: 1003426861
Donkey serum	Sigma-Aldrich	Cat#D9663
Goat serum	Natocor	Cat#734
Experimental models: Organisms/strains
D. melanogaster: Pdf-GeneSwitch: w*;;P{PdfGS}3	Bloomington Drosophila Stock Center, Depetris-Chauvin et al 16	BDSC: #80956; FlyBase: FBst0080956
D. melanogaster: ;{UAS-mito::mKO2::V5::APEX2}VKO2;	This paper	N/A
D. melanogaster: Pdf-GAL4: y[1]w[*];P{w[+mC]=PdfGAL4.P2.4}2;	Bloomington Drosophila Stock Center	BDSC: previously #6900, now available as part of #25031; FlyBase: FBtp0011844
D. melanogaster: white: w[1118];;	Bloomington Drosophila Stock Center	BDSC: #5905; FlyBase: FBal0018186; RRID: BDSC_5905
D. melanogaster: UAS-mito::GFP: w[1118]; P{w[+mC]=UAS-mito-HA-GFP.AP}2/CyO;	Bloomington Drosophila Stock Center	BDSC: #8442; FlyBase: FBti0040803; RRID:BDSC_8442
D. melanogaster: Pdf-RFP: w;Pdf-RFP;	Ruben et al 65	N/A
D. melanogaster: ;;UAS-CD4::GFP1-10	Feinberg et al 35	N/A
D. melanogaster: UAS-Syb::spGFP1-10	Macpherson et al 68	N/A
D. melanogaster: w;UAS-Sytα::GFP11/SM6a;	This paper	N/A
D. melanogaster: w;;UAS-Sytα::GFP1-10/TM6c	This paper	N/A
Oligonucleotides
Primer F1: 5’-CAAATTGGCGTCTATGCCTCTG-3’	Sigma-Aldrich
Primer R1: 5’- TTTCTAGACTAGTTCTTTGAACTTTTGGCCG-3’	Sigma-Aldrich	N/A
Primer F2: 5’- ATCAATGACAACACATTCACTCTCAGCCGGAGAAAT GGACATTG-3’	Sigma-Aldrich	N/A
Primer R2: 5’- CGTTTTGAGCAGGTGGTATACAGCGCTG-3’	Sigma-Aldrich	N/A
Primer FXSGL: 5’- AACTCGAGCAAACATGCGCGATCACATGGTGCTGC ACGAGTACGTGAACGCCGCC-3’	Sigma-Aldrich	N/A
Primer RXSGL: 5’- GTGTTGTCATTGATCCTCCGCCTCCGCTTCCTCCGC CGCCGGTGATGCCGGCGGCGTTCACGTACTCGTG- 3’	Sigma-Aldrich	N/A
Primer F3: 5’-AACTCGAGCAAACATGCGCG-3’	Sigma-Aldrich	N/A
Primer R3: 5’-CGTTTTGAGCAGGTGGTATACAGC-3’	Sigma-Aldrich	N/A
Primer F4: 5’- AACTCGAGCAAACATGTCCAAAGGAGAAGAACTG-3’	Sigma-Aldrich	N/A
Primer R4: 5’- TCCGCCGCCTGTTCCTTTTTCATTTGGATCTTTGC-3’	Sigma-Aldrich	N/A
Primer F5: 5’- GAAAAAGGAACAGGCGGCGGAGGAAGCGGA-3’	Sigma-Aldrich	N/A
Primer R2: 5’- CGTTTTGAGCAGGTGGTATACAGCGCTG-3’	Sigma-Aldrich	N/A
Recombinant DNA
mito::mKO2::V5::APEX2	Lam et al 25	N/A
Plasmid: pJFRC81-10XUAS-IVS-Syn21-GFP-p10	Pfeiffer et al 67	RRID:Addgene_36432
Plasmid: pUAST	Brand & Perrimon 69	N/A
Software and algorithms
IMOD	University of Colorado Boulder	RRID:SCR_003297; https://bio3d.colorado.edu/imod
Zen software	Zeiss	RRID:SCR_013672; https://www.zeiss.com/microscopy/es/productos/software/zeiss-zen.html
R, version 4.1.0	R Core Team	RRID:SCR_001905; https://www.R-project.org/
GraphPad Prism, version 8.0.2	GraphPad Software	RRID:SCR_002798; www.graphpad.com
ClockLab, version 6.1.04	Actimetrics	RRID:SCR_014309; https://actimetrics.com/products/clocklab/
Fiji (Fiji Is Just ImageJ), version 2.16.0	Schindelin et al 71	RRID:SCR_002285; https://imagej.net/downloads
Data related to semi-automatic segmentation	This paper	https://github.com/labCeriani/SegmentEM

Flies were grown and maintained at 25°C in vials containing standard cornmeal yeast agar medium under 12:12 h light:dark (LD) cycles. Adult-specific GRIP expression was accomplished through GeneSwitch¹⁶. GeneSwitch expression was induced transferring 2- to 5-day-old males and females to vials containing food supplemented with RU486 (mifepristone, Sigma, USA) in 80% ethanol to a final concentration of 200 μg/ml for exactly 96 hours prior to dissection. The Pdf-GeneSwitch (pdf-GS) and the UAS-mito::mKO2::V5::APEX2 (described in this manuscript) lines were generated in our laboratory¹⁶. Pdf-GAL4 (#6900, now available as part of #25031), white¹¹¹⁸ (#5905) and UAS-mito::GFP (#8442) were obtained from the Bloomington Stock Center. Pdf-RFP was generously provided by J. Blau (New York University, USA). GRIP stocks were generated in the Rosato Lab (Leicester University, UK) and UAS-CD4::GFP_1–10 was provided by K. Scott (University of California, Berkeley, USA).

Generation of transgenic lines

The mito::mKO2::V5::APEX2 sequence was provided by the Ellisman Lab²⁵. A Kozak sequence was added upstream of the mito::mKO2::V5::APEX2 fragment. This sequence was synthesized and cloned into pJFRC81–10XUAS-IVS-Syn21-GFP-p10 (Addgene, USA; described in Pfeiffer et al⁶⁵) by GeneScript (GeneScript, USA), replacing the GFP coding region. Transgenic lines were produced by BestGene (BestGene, USA) by Phi31 site specific transformation into attP2 (third chromosome) or VKO2 (second chromosome) sites.

For the GRIP lines, the two different GFP fragments, GFP_1–10 and GFP₁₁, were fused through a flexible linker to the N-terminus end (pointing to the lumen of vesicles) of Synaptotagminα (Sytα). For GFP₁₁, Sytα was amplified via PCR between exons 2 and 6 using genomic DNA extracted from wild type flies and primers F₁: 5’-CAAATTGGCGTCTATGCCTCTG-3’ and R₁: 5’-TTTCTAGACTAGTTCTTTGAACTTTTGGCCG-3’, which incorporate restriction sites for Xba I and Pst I. After digestion with both enzymes, this fragment 1 (f1) was cloned into pBluescript II KS (−), generating pBs-f1. Another fragment (f2), which contains the coding sequence of exon 1 from Sytα fused to the beginning of exon 2, was amplified using the primers F₂: 5’-ATCAATGACAACACATTCACTCTCAGCCGGAGAAATGGACATTG-3’ and R₂: 5’-CGTTTTGAGCAGGTGGTATACAGCGCTG-3’. A third fragment (fXSGL) was produced annealing, elongating and amplifying in a PCR reaction the overlapping oligonucleotides: F_XSGL: 5’-AACTCGAGCAAACATGCGCGATCACATGGTGCTGCACGAGTACGTGAACGCCGCC-3’ and R_XSGL: 5’-GTGTTGTCATTGATCCTCCGCCTCCGCTTCCTCCGCCGCCGGTGATGCCGGCGGCGTTCACGTACTCGTG-3’. This contains a Xho I cutting site, the start methionine, the GFP₁₁ and the linker (GGGSGGGGS), and ends with exon 1 Sytα coding sequence. Finally, fragments f2 and fXSGL were mixed and amplified via PCR with the primers F₃: 5’-AACTCGAGCAAACATGCGCG-3’ and R₃: 5’-CGTTTTGAGCAGGTGGTATACAGC-3’. The resulting fragment was digested with Xho I and Pst I and cloned into the pBs-f1 previously generated, producing the pBs[Xho I-GFP₁₁-linker-Sytα-Xba I].

For GFP_1–10, the fragment was amplified by PCR using genomic DNA isolated from the fly line UAS-syb::spGFP_1–10⁶⁶ with primers F₄: 5’-AACTCGAGCAAACATGTCCAAAGGAGAAGAACTG-3’ and R₄: 5’-TCCGCCGCCTGTTCCTTTTTCATTTGGATCTTTGC-3’. The produced fragment (f4) contains a Xho I recognition site and part of the linker sequence. Another fragment (f5) was amplified using f3 as a template and primers F₅: 5’-GAAAAAGGAACAGGCGGCGGAGGAAGCGGA-3’ and R₂: 5’-CGTTTTGAGCAGGTGGTATACAGCGCTG-3’. f4 and f5 were mixed and amplified via PCR using primers F₄ and R₂. The resulting fragment (f6) was digested with Xho I and Pst I and cloned into pBs-f1 also restricted with the same enzymes, producing pBs[Xho I-GFP_1–10-linker-Sytα-Xba I].

The DNA sequences encoding the fusion proteins were cloned into pUAST⁶⁷ in the Xho I-Xba I sites and undesired mutations were ruled out by sequencing. Transgenic lines were produced by the University of Cambridge Fly Facility (UK) using p-element transformation. Lines expressing split GFP components in the plasma membrane (UAS-CD4::GFP_1–10) were already available and described³⁵.

METHODS DETAILS

Behavioral analysis

A detailed description of genotypes examined can be found in Table S1. Animals were raised and entrained to LD 12:12 cycles at 25°C since early development; 3–5 day old males were placed in glass tubes (5 mm diameter × 65 mm length, TriKinetics, USA) in Drosophila Activity Monitors (DAM, TriKinetics) for data acquisition. Activity was monitored with infrared detectors and the DAMSystem3 data collection software (TriKinetics) under LD 12:12 cycles at 25°C for 3 days followed by constant darkness (DD) or constant light (LL) for additional 10 days. A single experiment was performed (N = 1); the number of animals per genotype ranged from 19 to 32 (n = 19–32).

Brain preparation for CLEM

Briefly, at ZT2, ZT14 and ZT22, 2- to 4-day old female flies expressing Pdf-GAL4, Pdf-RFP > UAS-mito::APEX2 were anesthetized on ice and decapitated. Heads were fixed in 2.5% glutaraldehyde and 2% paraformaldehyde (PFA) in 0.15 M cacodylate buffer (CB, pH 7.4) on ice for 1 hour. After removing the fixative, brains were dissected in fresh 0.15 M CB and washed 3 times for 10 minutes in the same buffer solution. To facilitate the penetration of the staining, an optic lobe was removed using fine forceps, leaving the adjacent region of interest undamaged. Brains were transferred to a coverslip and imaged at 20X, 40X, and 60X in an Olympus FluoView 1000 microscope (Olympus, Japan). Out of 8–10 prepared brains, we selected one per time-point based on how well its s-LNv arbor matched the stereotypical morphology documented in prior studies^12,15 as well as the overall quality of the preparation. Immediately after, brains were treated for 15 minutes with 20 mM glycine in 0.15 M CB on ice to quench any unreacted glutaraldehyde. For the DAB reaction, a preincubation step in 2.5 mM DAB solution (25.24 mM stock in 0.1 M HCl) in 0.15 M CB for 15 minutes on ice was included. For staining, 0.03% H₂O₂ containing DAB solution was added to the brains on ice for 40 minutes. The staining buffer was washed off with ×5 washes using 0.15 M CB. Then, brains were individually placed inside scintillation vials into 1% OsO₄ in 0.15 CB for 40 min at 4°C. Next, the OsO₄ solution was removed and, without washing, a reduced solution of 1.5% potassium ferrocyanide in 0.15 M CB containing 2 mM CaCl₂ was added. Brains were incubated for 1.5 h at 0°C and 30 minutes RT. After thoroughly washing 3 × 10 minutes with ddH₂O, brains were transferred to 0.5% aq. thiocarbohydrazide for 15 minutes at RT. Subsequently, brains were washed 3 × 10 minutes on ddH₂0 and stained with 2% aq. OsO₄ for 30 minutes at RT. Brains were washed with ddH₂O at RT 3 × 10 minutes and then stained with 0.5% aq. uranyl acetate for 30 minutes at RT. For the last staining step, after 3 × 10 minutes washes with ddH20 at RT, brains were incubated with 0.05% lead aspartate solution for 30 minutes at 60°C and 1 h at RT. The brains were washed 3 × 10 minutes on ddH₂0 and dehydrated in crescent percentages of ethanol and finally to a staining solution at specific temperatures and precise intervals, by using a freeze substitution chamber Leica EM AFS2 (Leica, Germany). Brains were individually transferred to small baskets and incubated as follows:

Temperature	1°C to −5°C	−5°C to −10°C	−10°C to −15°C	−15°C to −20°C	−20°C to −25°C	−25°C	−25°C	−25°C to 22°C
Solution	10% EtOH	30% EtOH	50% EtOH	70% EtOH	80% EtOH	90% EtOH	97% acetone 0.2% UA 1% OsO4	97% acetone 0.2% UA 1% OsO4
Time	30 min	30 min	30 min	30 min	30 min	30 min	33 hs	6 hs

Brains were then washed twice with dry acetone and placed into 50:50 Durcupan ACM:acetone overnight. Brains were transferred to 100% Durcupan resin overnight. Brains were then embedded in tin containers and left in an oven at 60°C for 72 hours. To proceed with micro-computed tomography and SBEM, brains were carefully trimmed in approximately ~1 mm square pieces, leaving the minimum resin, and then mounted on Gatan SBEM specimen pins with conductive silver epoxy.

Micro-CT, ROI definition and SBEM

The Micro-Computed Tomography (Micro-CT) tilt series were collected using a Zeiss Xradia 510 Versa (Zeiss X-Ray Microscopy, Germany) operated at 80 kV (87 μA current) with a 20X magnification and 0.3836 μm voxel size for ZT2, 0.4687 μm voxel size for ZT14, and 0.4688 μm voxel size for ZT22. Micro-CT volumes were generated from a tilt series of projections using XMReconstructor (ZEISS X-Ray Microscopy, Germany). Since no visible signal from the DAB deposition was seen in the Micro-CT, neuroanatomical features such as the alpha lobe of the mushroom bodies and the protocerebral bridge were used to correlate the computed tomography data to the confocal images. A 80 × 80 × 100 μm region of interest (ROI) containing the s-LNv terminals was defined for each time point. SBEM was accomplished using GeminiSEM 300 (ZEISS, Germany) equipped with a Gatan 3View system (Gatan, AMETEK, US) and a focal nitrogen gas injection setup. This system allowed the application of nitrogen gas precisely over the blockface of the ROI during imaging with a high vacuum to maximize the SEM image resolution, as described in Deerink et al⁶⁸. Images were acquired in 2.5 kV accelerating voltage and 1 μs dwell time, Z step size was 50 nm, raster size was 16k × 14k for ZT2, 20k × 20k for ZT14 and 20k × 24k for ZT22 Z dimension was ~900 to 1500 image samples. Volumes were collected using 90% nitrogen gas injection to samples under high vacuum. Once volumes were collected, the histograms for the slices throughout the volume stack were normalized to correct for drift in image intensity during acquisition. Digital micrograph files (.dm4) were normalized and then converted to MRC format. The stacks were converted to eight bits and volumes were manually traced for reconstruction using IMOD (University of Colorado Boulder, US)²⁹. The s-LNvs were identified as neurons containing labeled mitochondria. They were segmented starting 2 microns below the first ramification of the first branching neuron of the bundle. DCVs were identified as dense, round structures of 90–200 nm. When connected to the plasma membrane, DCVs were marked as fusing DCVs. Synaptic sites were identified by integrating several ultrastructural features, especially in cases where material loss due to serial sectioning obscured traditional T-bars. We looked for 1) regions where the presynaptic membrane exhibited a distinct electron-dense thickening —indicative of active zone scaffold proteins—, 2) at least one nearby dense postsynaptic specialization and 3) the presence of clustered clear vesicles or remnants of a T-bar structure. These three elements were segmented as scattered objects. Mitochondria were recognized by shape, matrix pattern and high contrast due to DAB staining, and segmented as closed objects. Following the segmentation process, a single round of independent manual proofreading was conducted. The data pertaining to the total number of scattered objects, neuronal and mitochondrial volume, and surface area were extracted using the -imodinfo command. The varicosities were first defined as swellings greater than 0.8 μm in the axon. Then, the objects found within the model were counted. For the purpose of quantification, the primary neurite on each neuron was defined as the longest path from the point of origin of the structure. Subsequently, the relative position of each varicosity within a given neuron was normalized in relation to the length of the primary neurite of that neuron. Secondary neurites were defined as the longest protrusion emerging directly from a primary neurite. Tertiary neurites were defined as protrusions from secondary neurites. Boutons were defined as small sprouts containing one varicosity, similar to synaptic spines.

The MCI was calculated using the equation described in Vincent et al ⁴³:

$$MCI=\frac{mitochondrialsurfac{e}^3}{16\times mitochondrialvolum{e}^2\times {\pi }^2}$$

Semi-automated segmentation software

Three main neural networks support the tool’s segmentation tasks. First, a semantic segmentation model identifies the s-LNv terminals by detecting labeled mitochondria, which is then followed by a high-precision model that refines the boundaries of the mitochondria. A convolutional neural network (CNN) is then employed for the detection of vesicles within the boundaries of neurons. The segmentation models utilize Residual Attention U-Net on small patches (6.4 × 6.4 μm) to optimize efficiency, subsequently reassembling the patches to facilitate full volume segmentation. Neuron and mitochondrial annotations from previous IMOD data, refined in Python and Fiji (Fiji Is Just ImageJ)⁶⁹, provided ground truth for training on 4,150 images.

For vesicle detection, a “mobile classifier” (developed by PyTorch) uses a CNN⁷⁰ with four layers, max-pooling, and fully connected layers on 250 × 250 nm patches to estimate the probability of vesicle presence. A second “refinement classifier” is employed to reduce the number of false positives, utilizing larger 500 × 500 nm patches to provide greater context around vesicle centers. The classifiers were trained and validated on a dataset comprising thousands of vesicle and non-vesicle images.

Finally, using py-clesperanto⁷¹ for GPU-accelerated 3D segmentation, each structure is labeled to facilitate metric calculations, like structure count and distribution within the volume. Figure S3 provides an overview of the algorithm development.

GFP reconstitution in the presynapse (GRIP)

Pdf-RFP;Pdf-GS,UAS-CD4::GFP_1–10 > UAS-Sytα::GFP₁₁ (GRIP) flies kept under LD 12:12 at 25°C were induced in RU486 as described. At ZT2, ZT14 and ZT22, males and females were collected in empty vials, briefly anesthetized on ice and their brains dissected in cold PBS under red light. Dissected brains were fixed in PFA 4% (pH 7.5) for 1 hour at room temperature in darkness. Samples were washed in cold PBS for 15 minutes and then for 5 minutes. Finally, brains were mounted in Vectashield (Vector Laboratories, USA) with their posterior surface upwards to allow access to the dorsal protocerebrum. Z-stacks of the s-LNv dorsal projections were taken immediately after mounting using a confocal microscope Zeiss LSM 880 (Zeiss) with a 40X water objective (N.A 1.2) employing the Zen software (Zeiss). The same objective and an Airyscan detector were employed to acquire the image in Figure 2H. Two independent experiments with n = 9–15 brains per time point were analyzed (Figure 2J).

For observations in constant light (LL), the experiment was replicated as described above, but one group of animals was kept in LD 12:12 as a control, while another group was transferred to LL the day after RU486 induction. Flies in both light regimes were dissected at 6 am, 10 am and 10 pm. For the control group, these time points are ZT22, ZT2 and ZT14. For the treated group, these times represent 70, 74, and 86 hours under LL conditions. Details on the experimental design can be found in the diagrams included in Figure S4C. A single experiment with n = 5–12 brains per time point was performed (Figure S4E).

Immunofluorescence and imaging

For validation of the GRIP tool, males and females expressing Pdf-RFP;Pdf-GS> ;;UAS-Sytα::GFP_1–10 were induced for 96 h with RU486 as described and dissected at ZT2. Note that, unlike GRIP flies that enable GFP reconstitution, these animals only express the larger GFP fragment tagged to the DCVs enabling detection through the incubation with a polyclonal GFP antibody. One experiment with n = 9 brains was performed (Figure S4B).

For mitochondrial analysis, male and female flies expressing ;Pdf-GAL4,Pdf-RFP;> ;UAS-mito::GFP; were exposed to the same light regimes and dissected at the same time points as in the GRIP experiment (LD and LL). One experiment with n = 13–18 brains per time point was performed (Figures 4G, 4H, S4G, and S4H).

Flies were anesthetized with CO₂, decapitated, and heads fixed in PFA 4% (pH 7.5) for 50 minutes at 25°C. Brains were dissected in PBS containing 0.1% Triton X-100 (PT) and washed 4 × 5 minutes. Samples were incubated ~48 h at 4°C in PT 0.1% with 5% donkey serum or 7% goat serum; the primary antibodies (1:500) used were chicken anti-GFP (Aves Labs, USA), rabbit anti-RFP (Rockland, USA), and rat anti-PDF (generated in our laboratory¹⁶). After washing 4 × 10 minutes in PT 0.1%, brains were incubated for 2 h at room temperature with the corresponding secondary antibodies (1:500): Alexa Fluor 488 anti-chicken, Cy3 anti-rabbit, and Cy5 anti-rat (Jackson ImmunoResearch, USA). Brains were then washed 4 × 15 minutes in PT 0.1% and mounted posterior side up in Vectashield (Vector Laboratories) with a spacer made from double-sided tape (Rapifix, Argentina). Z-stacks of s-LNv dorsal projections were acquired on a Zeiss LSM 880 using an Airyscan detector and a 40X water-immersion objective (NA 1.2) for GRIP or a 63X oil-immersion objective (NA 1.4) for mitochondria imaging.

QUANTIFICATION AND STATISTICAL ANALYSIS

The statistical analyses were performed using the R software, version 4.1.0⁷² (R Core Team, Austria). All graphs were made in GraphPad Prism, version 8.0.2. In all graphs, error bars indicate the standard error of the mean (SEM), unless otherwise indicated in the figure legend. Asterisks indicate statistically significant differences: * p < 0.05, ** p < 0.01, *** p < 0.001. Non-significant differences are not shown. Exact values for analyses and p values for contrasts are included in Tables S2–S4. The sample size (n) is indicated above the symbols in graphs that have been statistically analyzed, and the type of sample is described in the corresponding figure legend. In the analyses performed per neuron, n = 4. The number of experiments performed is referred to as N. Outliers were defined with a first box plot exploration and confirmed with standardized residual analysis after modelling.

Locomotor behavior analysis

Dead animals were removed from the analysis. Rhythmicity (as power-significance) of the remaining flies was estimated using ClockLab software (Actimetrics, USA). Additionally, data were binned into 5 minutes and flies with a stereotyped actogram and a defined peak over the significance line in the periodogram (p < 0.05) were selected as rhythmic. Percentage of rhythmic and arrhythmic animals and the mean period for each genotype were calculated. Data can be found in Table S1.

Electron microscopy volumes analysis

All the objects analyzed throughout the manuscript are the result of manual segmentation of the EM images.

Parameters displayed per neuron per time point (volume, DCVs, fDCVs and PSs density, total number of objects and total number of objects per region, mitochondrial volume and total number of varicosities)

Datasets were analyzed using one-way ANOVA followed by t-test with Bonferroni corrections from the emmeans package (v. 1.10.5)⁷³. Assumptions of used models were checked with check_model function from performance package (v. 0.14.0)⁷⁴. Details can be found in Table S4.

DCVs, fDCVs, and PSs per varicosity per region, and number of mitochondria

A model was applied to each neuronal region and object. Generalized linear mixed models with a Poisson distribution were fitted using the lme4 package (v. 1.1–35.5⁷⁵), with neurites considered as a random effect. In cases where the assumptions were not met (over-dispersion), a Conway–Maxwell–Poisson distribution was used employing the glmmTMB package (v. 1.1.10⁷⁶). Then, an analysis of deviance (based on type II Wald chi-square tests) was performed and Bonferroni corrections were applied to post hoc comparisons. Details can be found in Table S2.

Mitochondrial complexity index (MCI)

The mitochondrial complexity index was examined as a function of time point. A linear mixed model was fitted, and the variable was transformed using the natural logarithm. One-way ANOVA and a post hoc analysis with the Bonferroni test was conducted to compare between time points. Details can be found in Table S3.

Varicosity girth

The data was fitted to a linear mixed model in which neurons were considered as a random variable. One-way ANOVA and a posteriori contrasts were performed with the Bonferroni test. Details can be found in Table S3.

Fluorescence images analysis

Co-occurrence analysis

To quantify co-occurrence of two signals, Z-stacks acquired from brains stained to detect GFP (DCVs), RFP (cytoplasm) and PDF were processed in Fiji. Channels were separated and the average background of three random regions was subtracted from the entire image. Each channel was then converted into a binary mask using Otsu thresholding. To restrict signals to the neuronal terminal, GFP and PDF masks were separately multiplied by the RFP mask (Process > Image Calculator > Multiply), and the resulting images were further multiplied to generate a final mask representing the GFP–PDF and PDF-GFP overlap. Particle analysis was performed on each mask, and the overlap area (GFP × PDF) was normalized to either GFP × RFP or PDF × RFP to obtain the relative co-occurrence values shown in Figure S4B.

GFP reconstitution in the presynapse

GFP signal in the dorsal projections of s-LNvs was quantified using Fiji. Maximum intensity Z projection of Pdf-RFP signal was used to generate a mask of the region of interest (ROI). The ROI was used over a maximum intensity Z projection of GFP signal to quantify the mean intensity and then moved to the background to measure it. Background was subtracted to the terminal signal to get the final values displayed in the graphs. For representative images in Figure 2I and Figure S4D, background was measured in each case and subtracted to the entire image. Additionally, the upper limit of the display range for the signal was set to 60 in all images to allow direct comparison and ensure signal visualization in the Z projections included in the figures. For statistical analysis, the fluorescence intensity in the terminals for each time point was modeled using a general linear mixed model, using the nlme package (v. 3.1–164⁷⁷), in which the replica number was treated as a random variable. Furthermore, the model incorporates a parameter that allows the adjustment of the variance. One-way ANOVA was performed and subsequent analysis was conducted using the Bonferroni test. Details can be found in Table S3.

Mitochondrial analysis

To quantify mitochondria number, volume, and surface area in fluorescence images, image stacks were processed in Fiji using the Mitochondria Analyzer plugin⁷⁸. All datasets were analyzed in 3D, preserving voxel calibration metadata to obtain measurements in physical units (μm, μm², μm³). For each acquired Z-stack, channels were separated, and the RFP channel was converted to 8-bit and segmented in 3D using Otsu’s automatic threshold. The resulting binary mask was dilated to ensure complete coverage of the cytoplasmic volume and used to restrict the GFP signal to the neuronal compartment through voxel-wise multiplication. The masked GFP stack was then processed with the Mitochondria Analyzer plugin, which combines local filtering, adaptive thresholding, and morphological operations to segment mitochondria in three dimensions. The following parameters were applied for segmentation in Mitochondria Analyzer: rolling subtraction = 1.20, sigma radius = 1.90, enhance max = 2.0, scale_0 = 2.6, threshold range from 0.50 to 0.80, gamma adjustment = 0.90, method = Weighted Mean, block size = 0.8, c-value = 10, with despeckling, outlier removal, and filling operations enabled. These settings were empirically optimized to maximize separation of individual mitochondrial structures while minimizing background noise. Quantification was performed with the analysis module of the plugin, which automatically extracted the number of segmented mitochondria, as well as their individual volumes (μm³) and surface areas (μm²). All images within a dataset were processed all together with identical parameters. Quality control was performed by visual inspection of representative datasets to confirm that the RFP mask accurately delimited the cytoplasmic volume, that no extracellular signal was retained after masking, and that segmented objects reflected mitochondrial structures. Statistical analysis and assumption verification for mitochondrial number modeling, as well as MCI, followed the same modeling criteria as those applied to EM data analysis, respectively.

Supplementary Material

SUPPLEMENTAL INFORMATION

Document S1. Figures S1–S5 and Tables S1–S4.

Highlights.

• Electron microscopy reveals daily changes to the ultrastructure of s-LNv projections

• Peptidergic and synaptic communication peaks in the early morning

• Mitochondrial function cycles in sync with axonal remodeling

• Varicosity number and content changes underlie daily fine-tuning of connectivity