The Neuronal Circuit of the Dorsal Circadian Clock Neurons in Drosophila melanogaster

Unless stated otherwise, fly strains used in this study were reared on standard cornmeal medium with yeast at 25 ± 0.2°C and 60 ± 5% relative humidity under light-dark cycles (LD) of 12:12 h. All fly lines used are described in(including references). Supplementary Table S1

To reveal the anatomy of specific clock neurons, GAL4 or split-GAL4 lines were crossed to a membrane-bound GFP reporter (10xUAS-myr::GFP) or to a cytoplasmic myc-tagged GFP reporter (20UAS-6xGFP, Supplementary Table S1). They were co-stained with antibodies against the clock protein Period (PER) and different antibodies against several neuropeptides to verify their neurochemistry (see immunohistochemical procedure below). To reveal the post- and presynaptic sites of the neurons of interest, we crossed the GAL4-drivers to UAS-DenMark::mCherry and UAS-nSyb::EGFP, respectively (Supplementary Table S1).

A driver stock for usage with the Flybow system (see below) was built by balancing the GAL4-lines listed in Table 1 and crossing them to y w;hs-mFlp5^MH12/CyO; TM2/TM6B or to y w; GlaBc/CyO;hs-mFlp^5MH3/TM6B depending on which chromosome the GAL4 insertion was located. Experimental flies were obtained by crossing the balanced GAL4/hs-mFlp5 lines to either hs-Flp¹;+;FB2.0B^49b or to hs-Flp1;FB2.0B^260b;+ (Supplementary Table S1).

For labeling individual clock neurons we used the revised multicolor Flybow system with the Flybow2.0B reporter construct, which carries an FRT-site flanked stop-codon upstream of the fluorescent protein sequences (Shimosako et al., 2014). The stop codon can be removed by Flp-recombination, induced by a heat shock-promotor controlled hs-Flp DNA-recombinase. Experimental flies were kept at 18°C (18 ± 0.2°C and 60 ± 5% relative humidity with a LD cycle of 12:12) to prevent uncontrolled Flippase recombination events. Three heat shocks (37°C) of 45–60 min were applied on three consecutive days to induce Flippase-recombinase activity at different larval and pupal stages until the clock network was fully developed, which appears to be the case in late pupal stages (Helfrich-Förster et al., 2007a). Adult flies (4–5 days after eclosion) were dissected and the brains were immunolabeled with anti-GFP, anti-mCherry, and anti-nc82 antibody-solution, as described in Schubert et al. (2018).

To reveal putative post-synaptic partners of selected clock neurons, we applied the trans-Tango technique (Talay et al., 2017) in combination with new clock driver lines (Sekiguchi et al., 2020). The crosses were reared at 25°C and the progeny transferred at 18°C after eclosion for several days, to obtain an accumulation of the post-synaptic signal. In general, we sampled flies after 17, 28, 30, 34, and 53 days at 18°C and found that the post-synaptic signal strength increases significantly with time. Presynaptic neurons were revealed by anti-GFP and postsynaptic neurons by anti-Hemagglutinin (HA) antibody staining (see below and Supplementary Table S2).

Fluorescent immunocytochemistry was performed according to the protocol described in Schubert et al. (2018). In brief, male flies were fixed as whole animals for 2–3 h in 4% paraformaldehyde after they have been entrained to LD12:12 cycles for 4–5 days after eclosion. For anti-PER staining, fixation was performed 1 h before lights-on (ZT23), since PER accumulation is maximal at this time (Zerr et al., 1990). For the other antibody staining, flies were fixed around ZT3. For staining against Cryptochrome (CRY), the flies were kept several days under constant darkness to allow CRY to accumulate. After dissecting the brains in phosphate-buffered saline (pH 7.4) they were incubated in 5% normal goat serum in PBT 0.5% overnight and then incubated in the primary antibody solution for ∼2 days. Fluorescent tagged secondary antibodies were applied for at least 3 h. All used antibodies are listed in Supplementary Table S2. The immunostained brains were aligned on a specimen slide, embedded in Vectashield 1000 mounting medium (Vector Laboratories, Burlingame, CA, United States), and stored at 4°C until scanning. If not otherwise stated at least 10 different brains were stained for each antibody or antibody combination.

Fluorescence protein expression and antibody staining were visualized and scanned with a Leica TCS SP8 or a Leica SPE confocal microscope (Leica Microsystems, Wetzlar, Germany), Leica SP8 was equipped with hybrid detectors, a photon multiplier tube, and a white light laser for excitation, using the laser and detector settings as described in Shimosako et al. (2014). Leica SPE was equipped with a photomultiplier tube and 488, 532, and 635 nm solid-state lasers for excitation.

We used a 20-fold glycerol immersion objective (HC PL APO, Leica Microsystems, Wetzlar Germany) for whole-mount scans and obtained confocal stacks with 2 µm z-step size and 1024 × 1024 pixels. For a more detailed view, we used a 63-fold glycerol objective (HC PL APO, Leica Microsystems, Wetzlar Germany) and scanned the brains with a resolution of 2048 × 2048 pixels and a z-resolution of 1 µm. All focal planes were scanned three to four times and the frames were averaged to reduce background noise. The hybrid detectors of SP8 were used with photon counting mode and each focal plane was scanned and accumulated four times. The obtained confocal stacks were maximum projected and analyzed with Fiji ImageJ (Schindelin et al., 2012). Besides contrast and brightness, no further manipulations were done to the confocal images. For displaying the depth information in maximum projections we used the plug-in “Z-stack Depth Color Code” (v.0.0.2) for Fiji. For the reconstruction of the DN₂s_, we used the simple neurite tracer plug-in (Arshadi et al., 2021). The reconstructions were visualized using the natverse libraries (v 0.2.4, Bates et al., 2020) for RStudio (v 1.3.1093). The anti-nc82 staining (anti-bruchpilot) was used to align our stacks to the Janelia Farm Research Campus standard brain (JFRC2) by using the computational morphometry tool kit graphical user interface plug-In for Fiji (CMTK GUI, Rohlfing and Maurer, 2003; Jefferis et al., 2007). We compared our anatomical results with the results gained in the hemibrain of a single female fly (Scheffer et al., 2020) and the reconstructions of the DN_1as by Marin et al. (2020) available for the full adult female brain (FAFB, Zheng et al., 2018).

For a comparison of our findings with the reconstructed neurons of the electron microscopic data set from the hemibrain and the FAFB, we used the natverse libraries (v 0.2.4, Bates et al., 2020) and hemibrainr package (v.0.5.0, Bates and Jefferis, 2022) for R (v. 4.0.5) via RStudio (v 1.3.1093). The neuron-specific information was obtained from the hemibrain dataset (Scheffer et al., 2020, v 1.2.1) of the neuprint server (neuprint.janelia.org↗). For the FAFB dataset (Zheng et al., 2018) the reconstructions were obtained from the Virtual Fly Brain CATMAID server (fafb.catmaid.virtualflybrain.org↗). The annotation of the neurons was used as a reference and the identity of the neurons was verified by their morphology and the orientation in the brain if possible. For morphological comparisons, the neurons and brain regions of the hemibrain were transformed into the JRC 2018F template space (Bogovic et al., 2020) using xform_brain (nat.templatebrains v 1.0). When different neuron types were shown in the same brain, the neurons were additionally mirrored to the other hemisphere. Connectivity data was visualized using the ggplot2 package (v3.3.5, Wickham, 2016).

The DN_1as express PER from the first larval instar onward (Kaneko et al., 1997). They are glutamatergic (Hamasaka et al., 2007; Collins et al., 2012) and in adult flies, they express additionally the neuropeptides IPNamide (Shafer et al., 2006) and CChamide1 (Fujiwara et al., 2018). Furthermore, they express the intracellular blue-light receptor Cryptochrome (CRY) (Yoshii et al., 2008). We addressed the two DN_1as with Flybow using the R16C05-GAL4 driver (Pfeiffer et al., 2008). Using this method, we were able to trace individual DN_1as in 30 brains, whereas in 11 brains only a single DN_1a was labeled and could be analyzed at the single-cell level. In addition, the two DN_1as were stained together in the DN_1a-specific split-GAL4 line R43D05-p65.AD; R93B11-DBD (Sekiguchi et al., 2020). In the following, we will describe the morphology of the DN_1as according to the results of both staining methods compared to the reconstructions at the electron microscopic level (Supplementary Video S1).

The cell bodies of both DN_1as are located in the anterior superior cell body rind dorsal to the superior lateral protocerebrum (SLP) (Shafer et al., 2006). Initially, the DN_1as project ventrally along the surface of the SLP into the posterior-most part of the SLP. The projections of both cells bifurcate at the level of the posterior boundary between the lateral horn (LH) and the SLP and invade the posterior-most part of the LH. Here, the neurites of both DN_1as branch extensively and innervate the ventromedial region of the posterior LH (Figure 2C), together with the s-LN_v terminals. Both neurons send projections medially that terminate dorsally to the Kenyon cells of the mushroom bodies in the lateral and dorsal accessory calyx (CA, Figures 2A,C, arrow). Further projections run ventrally to the accessory medulla (AME, Figure 2, open arrowhead). Concerning the shape and extent of arborizations in the posterior SLP and LH, the two neurons differed slightly. The varicose arborizations of one DN_1a appeared slightly more restricted around the CA whereas the other DN_1a showed less dense projections, extending slightly more medially in the SLP as well as more ventrally in the posterior LH. The biggest difference between the two DN_1as represented the single neurite of the neuron with the slightly broader arborization pattern, reaching from the dense ramifications around the CA in the dorsal SLP (Figure 2 double arrowhead). In roughly one-third of the DN_1as analyzed with Flybow the projections to the AME were not visible. However, the DN_1a-specific split-GAL4 line revealed that both DN_1as project ventrally (Figure 2C open arrowhead). This projection ran parallel to the neurites of the Pigment Dispersing Factor (PDF)-positive s-LN_vs until it reached and joined the posterior optic commissure (Figures 2A,C, open arrow). Along with this major fiber bundle, the ventral DN_1a projections invaded the ipsilateral AME and often ran onto the surface or along the anteromedial edge of the medulla (Figures 2A,C). Due to the vast overlap of their arborizations in most parts, distinguishing the two neurons in the polarity staining was difficult. We observed signals of the post- and presynaptic markers in the posterior ventromedial LH, probably originating from both DN_1as. The presynaptic vesicles in the projection to the AME were consistently and strongly labeled (nSyb::EGFP), whereas only a faint DenMark (TLN::mCherry) signal (postsynaptic marker) was found in these neurites (Figures 2D,E).

The reconstructions of the hemibrain largely support our observations and the two annotated DN_1as (#264083994 and #5813022274) have the same characteristic arborization pattern as described above (Figure 2F). The reconstructions by Marin et al. (2020) using the FAFB dataset (Figure 2F, neuron #4129363 and #3973434) support additionally our observation that the DN_1as differ in the neurite reaching in the dorsal SLP (Figure 2F double arrowhead). Output synapses (presynaptic terminals) were present throughout the neurites, but less densely in the one fiber of DN_1a #264083994 that projects dorsally into the SLP. This dorsally projecting fiber carried mainly input synapses (postsynaptic terminals) (Figure 2B double arrowhead).

Both DN_1as appear strongly linked to the s-LN_vs. The neurites of both neuron types have been shown to largely overlap with each other (Fujiwara et al., 2018), an observation that we confirm with our study. The DN_1as overlap with the terminals of the s-LN_vs in the SLP and with the dendrites of the s-LN_vs in the AME. Also functionally, the DN_1as and the s-LN_vs are connected as was shown by previous work: the DN_1as signal via glutamate and the neuropeptide CCHamide1 to the s-LN_vs (Hamasaka et al., 2007; Collins et al., 2012; Fujiwara et al., 2018; Song et al., 2021) while the s-LN_vs signal via glycine and PDF to the DN_1as (Shafer et al., 2008; Yoshii et al., 2009; Frenkel et al., 2017). To further investigate this mutual synaptic connection and to identify synaptic outputs of the DN_1as to other downstream neurons we performed trans-Tango and visualized the results by anti-HA in flies of different ages (17, 28, 30 days).

When driving trans-Tango in the DN_1as, several clock neurons (PER-positive) were co-labeled by the HA-antibody (Figures 3A,B): one to two DN₃s, the Ion transport peptide (ITP)-positive LN_d and fifth LN, one to four s-LN_vs (Figures 3A,B, white arrows) and sometimes weakly one to four l-LN_vs as already shown by Song et al. (2021). Also, the DN_1as themselves were labeled in some brains. In addition, one characteristic neuron of the dorsal fan-shaped body (dFB) was marked by anti-HA. Its cell body was located in the posterior brain (Figure 3A magenta arrowheads) with neurites running into the eighth layer of the FB and laterally toward a dense fiber hub close to the LPNs, in which several clock neurons arborize and the DN_1as contribute few fibers (Figure 3A, magenta arrows; see also later). Further non-clock neurons were labeled in the lateral, dorsolateral and posterior dorsal brain. These were located close to the LN_ds, DN₃s, and DN_1ps, respectively, but were not double-labeled by anti-PER (Figures 3A,B, open magenta arrowheads). While the DN₃s and most of the LNs were already stained in young flies, the neuron projecting to the dFB was only marked in older flies.

When driving trans-Tango in the s-LN_vs, we found that at least one DN_1a is postsynaptic to the s-LN_vs (Figure 3C). In addition, the three CRY-negative LN_ds were marked, indicating that they are also postsynaptic to the s-LN_vs.

The trans-Tango staining was rather consistent with the presynaptic (output) sites of the DN_1as revealed on the ultrastructural level in the hemibrain (Figure 4A). Most output synapses were found in the superior protocerebrum (SLP and SMP) to the ITP- and CRY-positive LN_d and the fifth LN, but no outputs were found to the s-LN_vs (Figure 4A) and very few to DN₃-like neurons (see below). The virtual lack of output to the DN₃s is understandable because these neurons are not unequivocally annotated in the hemibrain, but it is less clear why the s-LN_vs are not among the listed postsynaptic partners of the DN_1as. Most probably, they form only a small amount of synapses with these clock cells that have been overlooked. The same might be true for the input synapses from the s-LN_vs that we detected in our trans-Tango staining. The s-LN_vs are not listed as input partners of the DN_1as in the hemibrain (Figure 4B). Nevertheless, input from the s-LN_vs to the DN_1as was also revealed in a recent functional study, strongly indicating that our observation is valid (Song et al., 2021).

As expected from the slightly different morphology of the two DN_1as, DN_1a #264083994 with the predominantly dendritic fiber in the SLP received slightly more input from neurons located in this brain region. Otherwise, the two DN_1as show large similarities in their synaptic connections (Figure 4C) and seem to signal to each other (Figure 4B).

While the main DN_1a input and output synapses are formed in the SLP and posterior lateral protocerebrum (PLP), their neurites in the AME appear mainly presynaptic. The DN_1as get especially strong input (20% of their input synapses) in the dorsal and lateral accessory calyces that are located adjacent to the SLP (Figure 4C). Most of this input appears to come from temperature sensing neurons in the antennal lobes (VP3+_vPN/VP2+_adPN as well as VP2_adPN, Figure 4B) as was also revealed by Marin et al. (2020).

The DN_1ps are a demonstrably heterogeneous group in respect to their neurochemistry and morphology. About half of the approximately 15 cells per brain hemisphere express CRY and the PDF receptor (PDFR) (Shafer et al., 2008; Yoshii et al., 2008; Im and Taghert, 2010). Five to six of the CRY-positive DN_1ps are positive for glutamate (Hamasaka et al., 2007; Guo et al., 2016) and the Diuretic Hormone 31 (DH31, Kunst et al., 2014; Goda et al., 2016) and at least four of these CRY-/DH31-positive neurons express additionally Allatostatin C (AstC, Díaz et al., 2019; Zhang et al., 2021) and CNMamide (Jung et al., 2014; Abruzzi et al., 2017; Jin et al., 2021; Zhang et al., 2021). As recently shown, four AstC-positive DN_1ps additionally express the receptors for PDF and DH31 (Goda et al., 2018; Lin et al., 2022). The neurochemistry of the CRY-/PDFR-negative DN_1ps is so far unknown.

It is not surprising that the DN_1ps differ also in their morphology. The latter has been described in detail in three studies coming from different labs (Chatterjee et al., 2018; Guo et al., 2018; Lamaze et al., 2018). These studies used combinations of different GAL4, LexA, GAL80, and split-GAL4 lines to address different subgroups of the DN_1ps and stochastic flip outs to label single neurons. The most important line, which we used also in our study, is the Clk4.1M-GAL4 line. It targets about 10 out of the 15 DN_1ps comprising the majority (∼6 out of 7) of the CRY-positive neurons plus ∼4 CRY-negative cells (Figure 2; Zhang Y. et al., 2010, Zhang et al., 2010 L.; Chatterjee et al., 2018). The R18H11-GAL4 line from the FlyLight library collection (Pfeiffer et al., 2008) originates from the gene coding for PDFR and targets only the five to six CRY-/PDFR-positive neurons that are additionally glutamate- and DH31-positive (Kunst et al., 2014; Goda et al., 2016; Guo et al., 2016, 2018; Chatterjee et al., 2018). In contrast, the R18H11-LexA line additionally drives in few CRY-/PDFR-negative DN_1ps (Chatterjee et al., 2018; Guo et al., 2018), presumably because PDFR expression is not completely absent in the PDFR-negative characterized neurons, but only largely reduced, so that it is still revealed by strong drivers. Nevertheless, none of the used lines targets all DN_1ps, so additional types of DN_1ps may exist (especially CRY-negative ones) that have not yet been characterized.

Previous studies have principally identified two main types of DN_1ps: One type remaining ipsilaterally and projecting to the anterior optic tubercle (AOTU) (the so-called anterior projecting dorsal neurons), while the other type crosses the midline of the brain through the pars intercerebralis but lacks projections to the AOTU (Chatterjee et al., 2018; Lamaze et al., 2018). According to Chatterjee et al. (2018), the DN_1ps projecting to the AOTU express CRY, whereas it is unclear which DN_1ps of the other type are CRY-positive.

We analyzed the DN_1ps of 22 Clk4.1M-GAL4 brains in combination with the Flybow-reporter to reveal potential differences in the morphology of the DN_1ps. As it turned out, the modified Flp-recombinase had not been activated sufficiently, and only four brains showed individually labeled neurons. Nevertheless, we were able to identify three different types of DN_1p subgroups. All DN_1p cell bodies were located in the dorsal posterior cell body rind innervating the posterior superior medial protocerebrum (SMP) and SLP, as well as the posterior ventromedial LH and dorsal parts of the PLP, where they followed the s-LN_v projections in ventral direction, although no fibers reached the AME (Figure 5).

A single labeled DN_1p largely resembled the previously described CRY-positive neurons that invade the AOTU and form pronounced varicosities in this region (Figure 5A, called a-DN_1p in Lamaze et al., 2018). The fibers of this neuron formed two loops running from posterior to anterior and terminated in the AOTU (open arrowheads). One loop ran around the SLP and the second one ventrally on the surface of the superior clamp (SCL) and the inferior clamp until both converged shortly before the AOTU. In addition, fibers from this neuron remained posterior and followed the s-LN_v projections in the PLP (double arrowhead) for a short distance toward the AME. Most significantly, all neurites of this neuron remained ipsilaterally and did not cross the midline of the brain in the middle dorsal commissure (MDC). A similar neuron is described in Figures 1C,D of Lamaze et al. (2018) and called anteriorly projecting DN_1p (a-DN_1p).

The other labeled DN_1ps appeared to belong to the second type of DN_1ps described by Lamaze et al. (2018), Chatterjee et al. (2018), and Guo et al. (2018). These neurons were called vc-DN_1p_neurons by Lamaze et al. (2018) and sent fibers through the MDC to the contralateral SMP and on the ipsilateral side, their projections in the PLP appeared to reach slightly further toward the AME than that of the first DN_1p type (Figure 5B). We could identify two different subtypes with this projection pattern. One neuron of the first subtype was individually labeled (Figure 5B right and yellow neuron in Figure 5C), whereas the presence of the second subtype (Figure 5C green) was concluded from the samples with varying numbers of visible cells. Fibers of the second subtype started to form a loop around the SLP as was typical for the two CRY-positive DN_1ps described above, but they did not complete this loop and did not reach the AOTU (green neuron in Figure 5C). A very similar neuron (called vc-DN_1p) is described in Figure 1C of Lamaze et al. (2018). Furthermore, similar DN_1ps were found in Jin et al. (2021; Figure 5E in Jin et al.) and Zhang et al. (2021; Figure 3C in Zhang et al.) among the CNMamide-positive neurons. Since the latter are CRY-positive, we assume that this type of DN_1p is CRY-positive. In case of the yellow colored neuron in Figure 5C, we do not know about its CRY expression. Therefore, no statement about CRY is made. The same applies for all further statements regarding the DN_1ps.

Seven DN_1ps are annotated in the hemibrain, which were again divided into two types (called A and B in the hemibrain) (Figures 6–8 and Supplementary Video S2). The two type B neurons (cyan in Figure 6) resemble CRY-positive neurons described above, which remain ipsilateral and project into the AOTU. The five type A neurons all project contralaterally and do not enter the AOTU (green and magenta in Figure 6). Among these, the DN_1p depicted in green closely resembles the just described CRY-/CNMamide-positive DN_1ps that send fibers into the anterior SLP and follow the s-LN_v projections in the PLP several micrometers (green in Figure 6). The other four type A neurons do not show any fibers following the s-LN_v projections in the PLP, while the fibers extending toward the anterior SLP were generally present (magenta in Figure 6). The DN_1p neuron type that is not annotated in the hemibrain is depicted in Figure 5C (yellow). This neuron completely lacks the anterior running fibers but shows prominent neurites following the s-LN_v projections in the PLP. A similar neuron is described by Lamaze et al. (2018; vc-DN_1p neuron in Figure 1D) making us confident that this is a true subtype.

The DN₂s consist of only two neurons that have been described for the first time 25 years ago (Kaneko et al., 1997). Their importance in the clock network is highlighted by their early appearance during larval development, together with the s-LN_vs, the fifth LN, and the DN_1as, but their function and fine anatomy are yet unclear. The somata of the DN₂s are located near the dorsal terminals of the s-LN_vs and do not seem to express CRY (Benito et al., 2008; Yoshii et al., 2008). The arborizations of both DN₂s appear to remain in the dorsal protocerebrum and cross the midline of the brain through the pars intercerebralis, to terminate close to the cell bodies of their contralateral counterparts (Helfrich-Förster, 2005). However, due to the wide expression of the clock drivers available (i.e. Clk856-GAL4, Gummadova et al., 2009) and the vast overlap of the fibers of other clock neurons in the SLP, it was impossible to finely trace the neurites of the DN₂s so far.

We used the Clk206-GAL4 driver in combination with cry-GAL80 to restrict the expression of Flybow exclusively to the CRY-negative clock neurons, which comprise the DN₂s. Unfortunately, GAL80 was not sufficient to repress GAL4 expression in the CRY-positive s-LN_vs and DN_1as, except for the fifth LN. Consequently, none of the 17 brains analyzed contained individually labeled DN₂s. Nonetheless, since all additionally marked neurons have been described at single-cell resolution in our previous study (Schubert et al., 2018), we were able to specifically assign certain projections to known clock neuronal morphologies and therefore to characterize the unique pattern of the DN₂s. We found two main projections of the DN₂s - one toward the medial brain passing to the contralateral hemisphere, the other projection running laterally. The latter runs dorsolaterally around the SLP and invades the anterior-lateral part of the AOTU. The loop around the SLP reminds strongly of the projections of the CRY-positive DN_1ps described above. Since the staining with the Clk206-GAL4;cry-GAL80 line was only partly satisfying, we looked for split-Gal4 lines that label specifically the DN₂s.

Indeed, the just described DN₂ projection pattern was confirmed by using a split-GAL4 line that specifically labels one DN₂ and none of the other clock neurons (R43D03-AD; VT003234; Sekiguchi et al., 2020, Supplementary Video S3) and by the Clk9M-GAL4 line combined with Pdf-GAL80, which has been reported to label one or both DN₂ (Kaneko et al., 2012; Goda et al., 2016). The new split-GAL4 DN₂ driver was considerably weaker than the Clk206-GAL4 and the Clk9M-GAL4 driver, but we got satisfying labeling with a cytosolic GFP reporter that accumulates GFP stronger than the membrane-bound GFP does (Supplementary Table S1). In the case of Clk9M-GAL4; Pdf-GAL80, membrane-bound GFP was sufficient for labeling the DN₂s. As expected, the novel split-GAL4 line marked only one DN₂ and this cell was always CRY-negative (Figure 10A). With the Clk9M-GAL4; Pdf-GAL80, we could see two DN₂s in only one brain (out of 16, Figure 10B). In all other brains, again only one DN₂ was labeled, and in about half of the cases, the labeled DN₂ was weakly CRY-positive (Figure 10B). This suggests that the two DN₂s can be distinguished by their CRY expression level, and that Clk9M-GAL4 drives GFP either in the CRY-negative or in a weakly CRY-positive DN₂. Nevertheless, we did not see any difference in morphology between the different individually labeled cells, suggesting that the two DN₂s are very similar. In the following, we will describe the typical morphology of the DN₂s in more detail.

Shortly after leaving the soma, the DN₂ neurites bifurcate into two main branches. As found by Flybow, one runs to the contralateral brain hemisphere (arrow in Figure 10A), the other runs laterally (green arrowhead). The laterally running branch projects into the dorsal SLP and then runs further anteriorly forming the characteristic loop around the SLP toward the AOTU, but the DN₂ fibers do not reach it (Figures 10–12). Often this branch splits into two prominent fibers (Figure 11A). Before the split-off of the lateral branch(es), some less intensively stained fibers turn ventrally and follow the PDF-positive s-LN_v terminals for several micrometers (Figure 10, green double arrowheads). Generally, the ipsilateral remaining DN₂ fibers were in close vicinity to the s-LN_v terminals, suggesting an interaction between the two clock neuron types (see also below). In one case (out of 15 brains), fibers of the s-LN_v terminals even followed the dorsolaterally projecting DN₂ fibers (magenta arrowhead in Figure 11A). The contralaterally projecting fibers of the DN₂s showed pronounced varicosities and terminated in the superior protocerebrum slightly before they reached the contralateral DN₂ somata (white double arrowheads in Figures 10A, 11A). Sometimes we saw fine non-varicose fibers extending from the varicose fibers further laterally (open double arrowhead, Figure 10A).

The arborizations of the DN₂s did not only largely overlap with the PDF-positive fibers of the s-LN_vs but also with those of the ITP-positive fibers from the fifth LN and LN_d (Figure 11B). This was especially true in the SMP to SLP, in which the arborizations of the ipsilateral DN₂s were less varicose than those running to the contralateral hemisphere. As for ITP-positive clock neurons, the close vicinity to the DN₂s was maintained throughout the SMP where both clock neuron types cross the midline of the brain in the MDC (Figure 11B).

Comprising approximately 40 cells, the DN₃s are by far the largest clock neuron cluster in the brain of the adult fly. Several studies have already revealed the presence of larger and smaller DN₃ cell bodies, and that some of them are situated more anteriorly in the brain compared to the others (Helfrich-Förster, 2003; Shafer et al., 2006; Helfrich-Förster et al., 2007b). This is in line with our observations made on 43 brains, in which the Flybow-reporters were expressed under the control of the pan-clock-neuronal Clk856-GAL4 driver. The somata of approximately 15 DN₃s are located posterior to the LH, whereas the remaining cells reside more anteriorly in the dorsal and lateral cell body rind of the LH, respectively. The majority of DN₃s possess rather small cell bodies and only about four to five neurons have larger somata. Half of the larger cells are found among the small posterior DN₃s, while the other two are found among the more anteriorly located neurons of this group. Interestingly, the projections of the large DN₃s strongly resemble the arborization pattern of the ITP- and CRY-positive LNs (fifth LN and LN_d) (Schubert et al., 2018). Their projections initially run along the posterior boundary between the LH and the SLP (Figures 14A,B open arrowhead). The first branching occurred on the posterior surface of the brain at the level of the trijunction between the LH, SLP, and SCL (Figures 14A,B, arrows). Here, some of the fibers from the large DN₃ projected anteriorly to the AOTU via three paths. The fiber bundle running along the first path, projects around the surface of the SLP. This is the same path (or better loop) used by the CRY-positive DN_1ps and the DN₂s projecting towards the AOTU. The second path towards the AOTU coincides with the second loop of the CRY-positive DN_1ps that runs along the trijunction of SLP, SCL, and superior intermediate protocerebrum. The third path originates in the lateral PLP and runs along the surface of the anterior ventrolateral protocerebrum. These different fiber bundles running in the different paths are hard to identify in Figure 14 (anterior projecting fibers in yellow) and are best seen in Supplementary Video S4.

Other fibers of the DN₃s remained initially posterior (blue colors in Figure 14A) and passed ventrally through the PLP until they reached the ipsilateral AME (Figure 14A white double arrowhead). On their way to the AME, numerous side-branches separated from the main bundle and formed a kind of fiber hub (Figure 14A double open arrowhead) where they overlap with the ITP- and CRY-positive LNs (seeSupplementary Video S5).

Most fibers of the DN₃ projected medially forming a dense varicose fiber network in the SMP (Figures 14A,B, arrowheads). From there, a portion of the projections crossed the midline of the brain via the MDC and superior arch commissures (SAC) reaching into the contralateral hemisphere (Figures 14A,B).

We observed a different arborization pattern in brains in which only the small DN₃s were labeled compared to brains in which both, the large and small DN₃s were stained (Figure 14B). In these brains, the two hemispheres were connected exclusively via the MDC, with no additional projections through the SAC (Figure 14B). However, there are even more apparent differences between the small and large DN₃s: the small DN₃s lack the fiber bundle towards the AOTU and the projections ending in the AME. The fibers of the small DN₃s, that pass through the PLP toward the AME terminate in the fiber hub of the PLP halfway to the AME described above (Figure 14B double arrowheads). Thus, only the four to five large DN₃s send fibers into the AOTU and AME (Figure 14C).

No DN₃s could be annotated with certainty in the hemibrain dataset until now. However, there are possible candidates available of which three subtypes can be distinguished. While two subtypes comprise only five (named DN₃) and six (named DN₃B) neurons, respectively, the third subtype comprises 15 neurons (named DN₃A) (Supplementary Video S4). Comparing the morphology of these DN₃ candidates with our findings, we are fairly confident that the subtype with five neurons comprises the large DN₃s, as their projection pattern is consistent with our descriptions (Figure 14D). The only difference is that there are no fibers reported running through the SAC in the hemibrain that we observed in our Flybow staining, but this might be caused by a general lack of contralateral projections in the hemibrain, which we discussed previously (Reinhard et al., 2022).

The second group of DN₃s candidates from the hemibrain that matches our description is the DN₃A subtype comprising 15 neurons. These neurons largely resemble the small DN₃s. As we found in our staining they project halfway down to the AME and terminate in the PLP fiber hub.

The third subtype of DN₃ candidates in the hemibrain completely lacks the projections extending ventrally towards the AME in the PLP. Otherwise, it looks similar to the small DN₃s. We did not see this subtype in our staining, but most likely, the DN₃s comprise more subtypes than we and others could identify so far.

The small DN₃s were also labeled by Guo et al. (2018) as common neurons between the Clk856-GAL4 and the R77H08-LexA driver lines (Guo et al., 2018 Supplementary Figure S1A). Furthermore, Meiselman et al. (2022) recently found the large (and several small DN₃s) labeled by a specific split-GAL4 line. These authors report the same arborization pattern that we describe here. They also found a faint commissure in the superior brain, but unfortunately, we could not distinguish between the SAC and MDC according to their figures (Figure 2K in Meiselman et al., 2022).

Due to the lack of specific driver lines, we have not yet been able to perform a polarity analysis of the DN₃s. It will be most promising to do so in the future with the recently published split-GAL4 line. In addition, we assume that the hemibrain DN₃ candidates will be certainly annotated as DN₃s as soon as further experiments allow a more accurate comparison in respect of their polarity and synaptic partners.

The DN_1as express two neuropeptides (CCHamide1 and IPNamide) (Shafer et al., 2006; Fujiwara et al., 2018) and one neurotransmitter (glutamate) (Hamasaka et al., 2007). They are functionally strongly interconnected with the s-LN_vs via neuropeptide signaling. While the s-LN_vs signal to the DN_1as via PDF (Shafer et al., 2008; Im et al., 2011), the DN_1as signal to the s-LN_vs via CCHamide1 (Fujiwara et al., 2018). Here, we show by trans-Tango that the two neurons are additionally interconnected via regular synapses. This fits with previous trans-Tango studies (Song et al., 2021) as well as immunohistochemical results demonstrating that the s-LN_vs express metabotropic glutamate receptors that can be activated by glutamate from the DN_1as (Hamasaka et al., 2007). Vice versa, the DN_1as might express glycine receptors that can be activated by glycine released from the s-LN_vs, although this has not yet been proven (Frenkel et al., 2017). Manipulations of the reciprocal peptidergic connections between the DN_1as and the s-LN_vs affect the flies’ free-running period, their morning and evening activity as well as their siesta under light-dark conditions (Fujiwara et al., 2018). The synaptic connections between the two neuron groups have not been specifically manipulated, but the knockdown of glutamate receptors in the s-LN_vs increases the free-running period and the nocturnal activity under light-dark conditions (Hamasaka et al., 2007). Furthermore, downregulation of glutamate synthetase in all dorsal clock neurons increases the rhythmicity of flies under constant light conditions (de Azevedo et al., 2020), indicating that glutamate signaling from the DNs is involved in the general light sensitivity of the clock. Although it is unlikely that this effect is mediated exclusively by the DN_1as (some DN_1ps and DN₃s are also glutamatergic as shown by Hamasaka et al. (2007) and Collins et al. (2012)), the DN_1as appear to contribute strongly to the light detection. At night, their silencing reduces light-induced startle responses, whereas during the day, PDFR-dependent signaling to the DN_1as limits light-induced startle responses (Song et al., 2021). As we show here, the DN_1as get strong input from the dorsal and lateral accessory calyces. The dorsal accessory calyces transfer visual information from the lobula (Li et al., 2020). Thus, DN_1as may directly receive light information from the optic lobe.

On the other hand, the lateral accessory calyces convey thermosensory information from neurons in the antennal lobes (Marin et al., 2020). While about half of the clock neurons are intrinsically light-sensitive (via CRY), none of them is per se temperature-sensitive (Sehadova et al., 2009) but they rely on temperature input from peripheral sensory neurons (the aristae on the antennae, the chordotonal organs and the antennal AC neurons (Tang et al., 2017; reviewed in; George and Stanewsky, 2021)). Indeed, the DN_1as get strong synaptic input from several temperature sensing projection neurons from the antennal lobes. Alpert et al. (2020) showed that temperatures below the preferred temperature for flies (∼25°C) silence the DN_1as via GABAergic synapses from such thermosensory projection neurons. This silencing leads to reduced activity and increased sleep in the morning and evening, which can only be overwritten by light and PDF signaling. Indeed, we found that the DN_1as signal not only to the morning neurons (the s-LN_vs) but even stronger to the evening neurons, namely the CRY- and ITP-positive fifth LN and LN_d. In addition, the DN_1as have few output synapses to tangential neurons arborizing in the second and eighth layer of the dFB, which overlap with the fibers of several clock neurons in the PLP. The dFB is known to be involved in the control of sleep (reviewed in Bertolini and Helfrich-Förster, 2021).

Altogether these results demonstrate that the DN_1as play a central role in light- and temperature integration of the circadian clock and can prominently influence the morning and evening neurons as well as the sleep of flies.

We could show that the DN_1ps consist of more than the originally described two types of CRY-positive and CRY-negative neurons. In the hemibrain, seven of the ∼15 DN_1ps are annotated. The seven neurons are classified into type A and type B neurons, but the type A neurons appear heterogeneous. At least one type A neuron looks different from the others (green neuron in Figure 6). We found the same neuron type by Flybow and identified an additional one that was not yet annotated in the hemibrain. Lamaze et al. (2018) depict a further DN_1p type in their supplement that was neither identified in the hemibrain nor in the present study. These two previously undescribed DN_1ps project contralaterally like the type A neurons but differ in their projections toward the AOTU and PLP. The DN_1p identified here lacks the anterior projections to the AOTU but shows prominent fibers to the PLP (Figure 6D right panel), whereas the DN_1p identified by Lamaze projects anteriorly to the AOTU and lacks fibers in the PLP (Supplementary Figure S1A in Lamaze and Stanewsky, 2020). These findings demonstrate that the DN_1ps encompass many neurons with different morphology and most likely also different neurochemistry. Single-cell RNA sequencing data underline this conclusion: according to their gene expression, the DN_1ps comprise five to six distinct clusters (Ma et al., 2021). This heterogeneity could explain the different and partly contradicting findings regarding their function(s) (reviewed in Lamaze and Stanewsky, 2020). We will not repeat these details here but instead, focus on the main findings adding new insights that derive from the present study.

The DN_1ps appear to comprise morning and evening neurons (Chatterjee et al., 2018), which is strongly supported by our findings. Type A DN_1ps receive strong synaptic input from the evening neurons (ITP positive fifth LN and LN_d) and signal even stronger back to them. Thus, there exists a prominent mutual connection between the type A DN_1ps and the evening neurons, suggesting that the five annotated type A DN_1ps are by themselves evening neurons. Nevertheless, the seven neurons are heterogeneous and at least one of them (green neuron in Figures 5, 6) gets synaptic input in the PLP, a region where many clock neurons arborize. This DN_1p may additionally receive input from morning neurons (s-LN_vs) via PDF or from the LPNs via DH31 (Reinhard et al., 2022). So far this is the only annotated type A DN_1p neuron that is most likely CRY-positive, expresses the PDF receptor and additionally receives synaptic input in the PLP. It is tempting to speculate that all DN_1ps that overlap with the s-LN_v projections in the PLP are CRY-positive, but this has to be proven by future studies. The type B DN_1ps are all CRY-positive, largely overlap with projections from the s-LN_vs, and express the PDFR. Thus, they are connected via PDF with the morning neurons and can contribute to the morning activity of the flies.

At the same time, DN_1ps consist of sleep- and activity-promoting neurons (Guo et al., 2018; Lamaze et al., 2018). Through their fibers projecting to the pars intercerebralis, they affect wakefulness by signaling to neurosecretory cells expressing the Drosophila homolog of the mammalian corticotropin-releasing factor (Cavanaugh et al., 2014; Kunst et al., 2014). In contrast, the anterior projections of the DN_1ps to the AOTU, appear to play a prominent role in sleep control because in the AOTU, CRY-positive type B DN_1ps signal on sleep-promoting ring neurons of the ellipsoid body, influencing their neuronal activity (Lamaze et al., 2018). Here, we could confirm the intimate connection of the DN_1ps to the AOTU. Moreover, we could show that the projections of the DN_1ps toward the AOTU largely overlap with fibers from the DN₂s and large DN₃s. The DN_1ps strongly signal via these projections to the DN₂s (see below).

DN_1ps are also known to promote the siesta of the flies (Guo et al., 2016) and to inhibit activity at low temperatures (Yadlapalli et al., 2018). They respond to cool temperatures (∼16°C) with a strong increase in firing rate, while they are inactive at warm temperatures (>25°C) (Yadlapalli et al., 2018). In contrast to the DN_1as, the DN_1ps do not arborize in the lateral accessory lobes, but they seem to receive temperature information from thermoreceptors in the aristae of the antennae and chordotonal organs in the body via signals from anterior cells (AC) which are known to be involved in temperature sensing pathways (Hamada et al., 2008; Yadlapalli et al., 2018; Jin et al., 2021). Thus, the DN_1ps appear to be a major gateway for temperature sensing into the circadian neural network, which continuously integrates temperature changes to coordinate the timing of sleep and activity. Most likely, the DN_1ps pass this information to other clock neurons, which do not receive temperature input directly but are similarly involved in temperature-related behavior (e.g. DN₂, Kaneko et al., 2012; Goda et al., 2016).

It is worth noting that the so far annotated type A DN_1ps are quite polar and have their synaptic output sites in the contralateral SMP. The same is true for the DN₂s, which we will discuss in the following. This feature might be important for integrating information from both hemispheres.

The two DN₂s, similarly to the DN_1as, show a special association with the s-LN_vs. All three clock neuron groups express PER cyclically already during the larval stages, whereas PER in the larval DN₂s cycles in antiphase to the other clock neurons (Kaneko et al., 1997). This anti-phase cycling depends on PDF signaling from the s-LN_vs to the DN₂s (Picot et al., 2009) suggesting that the DN₂s are PDFR positive, although they are reported to be CRY-negative in adults (Benito et al., 2008; Yoshii et al., 2008). Here, we report weak CRY-immunostaining in one of the DN₂s, strongly suggesting that the PDFR is also present since they are always co-expressed (Im et al., 2011).

The DN₂s are important for temperature entrainment in larvae (Picot et al., 2009) and temperature preference in adults (Kaneko et al., 2012). In contrast to the DN_1as, the DN₂s do not arborize in the lateral accessory calyces and we could not detect any direct input from temperature sensing neurons of the antennal lobes. Nevertheless, the DN₂s could get information indirectly via other clock neurons. The most promising candidates are the type B CRY-positive DN_1ps that get input from thermosensory AC neurons and that strongly contact the DN₂s. Furthermore, Tang et al. (2017) suggested that AC neurons project also directly to the DN₂s. In addition to synaptic input, the DN₂s may receive neuropeptidergic input from the DN_1ps via DH31 (Goda et al., 2016).

The DN₂s themselves do barely signal to other clock neurons, suggesting that they are pure output neurons of the clock network. As mentioned, they are highly polar as they receive synaptic input in the ipsilateral SLP and have output synapses in the contralateral SMP. As we show here by trans-Tango, their main output sites are the mushroom bodies and neurosecretory cells in the pars intercerebralis. These two sites are highly involved in the regulation of sleep, arousal, and metabolism (see above and discussion in Reinhard et al. (2022)). Consequently, DN₂s can impact both processes and adapt them to different environmental temperatures.

The DN₃s are the largest and, until very recently, also the less well-characterized group of the dorsal neurons. They are heterogeneous, not only regarding their size and morphology but also in respect to their gene expression, which divides them into two clusters (Ma et al., 2021). Here, we confirm previous findings that the DN₃s consist of neurons with large and small somata with different morphologies, but they may be composed of further subgroups, as suggested by data from the hemibrain.

Some (∼4) of the DN₃s express a homolog of the vertebrate circadian gene nocturnin, coding for a deadenylase involved in mRNA decay (Nagoshi et al., 2010), and up to 17 DN₃s express the gene quasimodo, coding for a Zona pellucida domain protein (Chen et al., 2011; Buhl et al., 2016). Both proteins are involved in the light perception of the clock. Other observations suggest an additional role of the DN₃s in the sensory integration of temperature cues into the circadian system (Harper et al., 2016). Consistent with this observation, up to 10 DN₃s neurons express the DH31 receptor, and together with the DN₂s cells, they appear necessary for regulating the temperature preference of the flies at the beginning of the night (Goda et al., 2016). The large DN₃s are additionally responsive to PDF (Shafer et al., 2008). Furthermore, some DN₃s express glutamate, and the majority of them express the neuropeptide AstC (Guo et al., 2018; Díaz et al., 2019; Meiselman et al., 2022). Díaz et al. (2019) suggest that the AstC-positive DN₃s signal to at least one LN_d and in this way influence the evening activity of the flies. Guo et al. (2018) demonstrate a sleep-promoting role of the DN₃s, especially of the small DN₃s. Meiselman et al. (2022) show that the AstC-positive DN₃s regulate the cold-induced dormancy of the flies, supporting the role of the DN₃s in temperature-dependent behavior.

All these different roles are supported by morphological data. Especially the arborizations of the large DN₃s reach the entire clock network. They form arborizations in all four fiber hubs of the clock network (AME, SMP, ventral PLP and the loop in the SLP). In the AME they might not only receive light input (Li et al., 2018) but also communicate with all lateral neurons. In the SMP, they overlap with all clock neurons besides the LN_vs. In the ventral PLP hub (double open arrowhead in Figure 14A) their fibers largely overlap with the cell bodies of the LPNs and with fibers from the evening neurons (ITP-positive fifth LN and LN_d) and the small DN₃s. In the SLP loop, they are again in close contact with the evening neurons as well as with many DNs. Consequently, they may communicate with all these neurons. From our trans-Tango data, we already know that one to two larger DN₃s get input from the DN_1as. Connectivity data from the hemibrain suggest weak synaptic input from the type B CRY-positive DN_1ps to some of the small DN₃s. The large DN₃ even appear to have dense arborizations in the lateral accessory calyces, which would predispose them to receive information from the thermosensory neurons of the antennal lobes. Moreover, they cross the midline of the brain in two commissures, meaning that they can integrate information from both brain hemispheres. The small DN₃s form a prominent fiber network in the SLP and SMP and run ventrally in the PLP where most of them appear to terminate in the fiber hub formed by the arborizations of the large DN₃s, fifth LN, and LN_d. Thus, they contribute to the wide connectivity of the large DN₃s within the clock network.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.