Candidates for photic entrainment pathways to the circadian clock via optic lobe neuropils in the Madeira cockroach

The compound eye of cockroaches is obligatory for entrainment of the Madeira cockroach's circadian clock, but the cellular nature of its entrainment pathways is enigmatic. Employing multiple‐label immunocytochemistry, histochemistry, and backfills, we searched for photic entrainment pathways to the accessory medulla (AME), the circadian clock of the Madeira cockroach. We wanted to know whether photoreceptor terminals could directly contact pigment‐dispersing factor‐immunoreactive (PDF‐ir) circadian pacemaker neurons with somata in the lamina (PDFLAs) or somata next to the AME (PDFMEs). Short green‐sensitive photoreceptor neurons of the compound eye terminated in lamina layers LA1 and LA2, adjacent to PDFLAs and PDFMEs that branched in LA3. Long UV‐sensitive compound eye photoreceptor neurons terminated in medulla layer ME2 without direct contact to ipsilateral PDFMEs that arborized in ME4. Multiple neuropeptide‐ir interneurons branched in ME4, connecting the AME to ME2. Before, extraocular photoreceptors of the lamina organ were suggested to send terminals to accessory laminae. There, they overlapped with PDFLAs that mostly colocalized PDF, FMRFamide, and 5‐HT immunoreactivities, and with terminals of ipsi‐ and contralateral PDFMEs. We hypothesize that during the day cholinergic activation of the largest PDFME via lamina organ photoreceptors maintains PDF release orchestrating phases of sleep–wake cycles. As ipsilateral PDFMEs express excitatory and contralateral PDFMEs inhibitory PDF autoreceptors, diurnal PDF release keeps both PDF‐dependent clock circuits in antiphase. Future experiments will test whether ipsilateral PDFMEs are sleep‐promoting morning cells, while contralateral PDFMEs are activity‐promoting evening cells, maintaining stable antiphase via the largest PDFME entrained by extraocular photoreceptors of the lamina organ.

The clock is innervated by eight adjacent groups of neurons that are abundant of colocalized neuropeptides (Reischig & Stengl, 2003b).
Among them are four clusters of PDF-immunoreactive (PDF-ir) neurons: two in the lamina (dorsal and ventral PDFLAs) and two next to the AME (anterior and posterior PDFMEs). While the arborizations of the four groups of PDF clock neurons were reconstructed and embedded into a standard atlas of the cockroach brain (Wei, el Jundi, Homberg, & Stengl, 2010), individual branching patterns of single PDF neurons could not be resolved. Nevertheless, functional and neuroanatomical studies suggested that PDF neurons take part in different circadian clock circuits reviews: Stengl & Arendt, 2016;Stengl et al., 2015). It was hypothesized that they are involved in gating of photic entrainment pathways, in synchronization among clock cells, and in gating of clock outputs. Thus, they set the phase for circadian rest-activity (sleep-wake) rhythms.
Multiple-label studies revealed that PDFLAs that mostly colocalize 5-HT and FMRFamide immunoreactivities innervate the accessory laminae and the proximal lamina (Petri, Stengl, Würden, & Homberg, 1995). Furthermore, PDFLAs connected both lamina neuropils via the anterior fiber fan to the AME without sending processes to the midbrain . Their general branching pattern suggests that they obtain direct input from green-sensitive short compound eye photoreceptors and carry photic-phase information to AME clock neurons. Contralaterally projecting anterior PDFMEs connect both AMEs to ipsi-and contralateral lamina-and midbrain neuropils (Reischig & Stengl, 2002;Söhler, Stengl, & Reischig, 2011). All the posterior PDFMEs remain ipsilateral with unknown branching patterns. Additionally, undiscerned PDF neurons sent a side branch into the median-layer fiber system of the medulla Giese et al., 2018;Reischig & Stengl, 2002, 2003b. Based on previous findings Page, 1983;Page, Caldarola, & Pittendrigh, 1977) it was suggested that PDFLAs are clock inputs while PDFMEs serve as clock outputs. Furthermore, it was concluded that contralaterally projecting PDFMEs gate locomotor activity rhythms via terminals in ipsi-and contralateral premotor areas of the midbrain. Instead, ipsilaterally remaining PDFMEs gate sleeppromoting neuronal circuits, in synchrony with external light-dark cycles reviews: Stengl & Arendt, 2016;Stengl et al., 2015).
Endogenous circadian locomotor activity rhythms are relayed via unknown clock outputs to ipsi-and contralateral premotor areas (Page, 1978;Page et al., 1977). The light entrainment pathways that synchronize these endogenous rhythms of the cockroach circadian clock with external light dark cycles are not known. Undiscerned compound eye pathways from both the ipsi-and contralateral compound eyes synchronize the clock via phase advances at dawn and phase delays at dusk (Nishiitsutsuji-Uwo & Pittendrigh, 1968;Roberts, 1965Roberts, , 1974review: Homberg, Reischig, & Stengl, 2003). While histamine-ir short photoreceptor cells terminate in the lamina, long histamine-ir photoreceptor cells branch in a distal, so far undiscerned layer of the medulla (Lösel & Homberg, 1999). However, no direct innervation of the AME by the compound eye was found. Therefore, interneurons relay photic entrainment information from lamina and/or medulla to the AME. Intracellular recordings identified different types of lightsensitive optic lobe interneurons with somata next to the AME . At least two groups of light-sensitive medial neurons (MNes) connected the AME to unknown layers in the medulla and via the anterior fiber fan to lamina and accessory laminae . Furthermore, polarization-sensitive ventromedial neurons (VMNes) connected both AMEs via the posterior optic commissure to median layers in the ipsi-and contralateral medulla . Therefore, next to PDFLAs, MNes and VMNes relay ipsi-and contralateral photic input to the cockroach clock (Reischig & Stengl, 2003b). Furthermore, the accessory laminae were hypothesized to be innervated by extraocular photoreceptors also in the Madeira cockroach, as first described in beetles (Fleissner et al., 2001). There, accessory laminae are the first-order optic neuropil of the lamina organ, which is an extraocular photoreceptor organ, next to the lobula organ that innervates the AME as first-order optic neuropil (Fleissner et al., 2001).
Here, we searched further for candidates of light entrainment pathways to the cockroach circadian clock. With multiple-label immunocytochemistry, combined with histochemistry and backfills, we specified layers of lamina and medulla, identifying respective termination sites of compound eye photoreceptor neurons. Then, neuroactive substances that were suggested to be involved in photic entrainment were located to specified layers of lamina, accessory laminae, and medulla. Backfills from the contralateral optic stalk obtained detailed information where contralateral light information is processed in optic lobe neuropils. Based on these studies, we suggested a new hypothesis on how multiple parallel-light entrainment pathways entrain circadian rest-activity rhythms (graphical abstract). Furthermore, we showed for the first time that the accessory laminae have glomerular subcompartments and do not receive histaminergic innervation from the compound eye. Instead, they expressed strong acetylcholinesterase (AChE) activity, suggesting cholinergic input from the lamina organ (Fleissner et al., 2001).

| Animals
Cockroaches of R. maderae were reared in laboratory colonies with 12:12 hr light-dark cycles. Temperature was kept at 25 C and relative humidity at 50%. Animals were fed with dried dog food, potatoes, and carrots, and had access to water ad libitum. All experiments were performed with adult male cockroaches.

| Backfill experiments
For neurobiotin (Vector Laboratories, Burlingame, CA) backfill experiments, the animal was first anesthetized in ice water and then fixated on a wax petri dish. A constant CO 2 inflow kept the animal stunned during the experiment. To expose the optic lobe, a rectangle on one side of the head capsule was excised and tracheae as well as fatty tissue were carefully removed. Insect saline was used during the experiment to prevent drying. In the next step, either one optic nerve or the optic stalk was cut through with a precision shear. A glass capillary filled with a drop of neurobiotin was slipped over the severed optic nerve (0.07% neurobiotin) or the optic stalk (5%), respectively, and fixed with modeling clay. Finally, the animal was placed in a humidity box for almost 24 hr at 4 C to allow intracellular transport of the tracer. The next day, the brain was removed from the head capsule and processed for (immuno-) histochemistry. To label the neurobiotin, an optimum dilution of dye-coupled streptavidin with 1:100 for Cy2 and 1:300 for Alexa 405 (Dianova) was added in the incubation step of secondary antibodies.

| Imaging
The evaluation of AChE histochemistry was carried out with the transmission mode of the confocal laser scanning microscope (CLSM; TCS SP5, Leica). All preparations labeled with fluorescence dyes were scanned with the confocal fluorescence mode. If AChE histochemistry was performed on the same section, transmission and fluorescence channels were scanned together. Preparations were scanned with a Leica HCX PL apochromate 20×/0.7 multi-immersion or 63×/1.20 water objective.

| RESULTS
Photic entrainment pathways of the Madeira cockroach circadian clock, the accessory medulla (AME), are not well described. Compound eye photoreceptor neurons are required for photic entrainment of the cockroach clock, but they do not contact the AME directly (Lösel & Homberg, 1999). To determine which interneurons relay photic information from photoreceptor cells to the clock, multiple-label immunocytochemistry was combined with histochemistry and backfill studies. First, different layers of lamina and medulla were distinguished, as possible functional subdivisions. Then, multiple-label studies confined termination sites of short and long photoreceptor neurons to specific layers in lamina and medulla. Next, we searched for interneurons that interfaced photoreceptor neurons and circadian clock neurons. We employed antiserum against pigment-dispersing factor (PDF) as marker for cockroach circadian clock neurons and antisera against neuroactive substances predicted to be employed in photic entrainment (review: Stengl et al., 2015). Finally, backfills from the contralateral optic stalk allowed to identify arborizations of contralateral visual pathways in lamina and medulla layers.

| Identification of 10 main layers in the medulla and three layers in the lamina via AChE histochemistry
With different neuroanatomical techniques we characterized distinct layers in lamina and medulla neuropils. Enzyme-mediated AChE histochemistry was employed previously to identify cholinergic neurons (Eckenstein & Sofroniew, 1983;Homberg et al., 1995;Karnovsky & Roots, 1964;Satoh, Armstrong, & Fibiger, 1983;Tago et al., 1986). It proved to be suited best for the differentiation of three lamina layers (LA1-3) and 10 main medulla layers (ME1-10) distally to proximally in horizontal optic lobe sections ( Figure 1a; Table 2). AChE-labeled neuronal projections were detected in all lamina and medulla layers at varying intensities. Highest staining intensities were observed in LA1, LA3 in the accessory laminae, ME1, ME3, ME5, ME8, ME10, and the AME ( Figure 1a; Table 2). Antisynapsin (n = 10) and anti-horseradish peroxidase (HRP; n = 10) immunocytochemistry also allowed for the differentiation of 10 ME layers. However, layer boundaries were less defined compared to AChE staining (Figure 1b,c). In the lamina anti-synapsin immunocytochemistry labeled LA1 and LA2, but not LA3 (n = 10). With anti-HRP immunofluorescence no reproducible distinction of lamina layers was possible (n = 10). Consequently, in all following preparations immunocytochemistry was combined with AChE histochemistry to assign staining patterns to respective layers.
In summary, AChE histochemistry distinguished three layers in the lamina, strongly stained accessory laminae, and revealed 10 layers in the medulla.
The different optic neuropil layers might encode parallel retinotopic maps processing distinct visual cues together with extracellular space coordinates. Our further studies will determine in which of these optic lobe layers the spatial map is combined with temporal information provided by the circadian clock. Furthermore, future studies will determine whether AChE-stained neuropils are cholinergic.  3.3 | One CRZ-ir MNe per optic lobe connects two distinct dorsal compartments of ME4 to the AME, strongly overlapping with PDF-ir branches  Table 2). ME6 could be further subdivided into contralaterally innervated ME6.1 and only ipsilaterally innervated ME6.2 ( Figure 4f; Table 2). Strongest contralateral innervation received the accessory laminae, ME1.1, ME3 (not shown), ME4, and ME6.1.

| PDF clock cells arborize in accessory laminae
Neurobiotin-labeled fiber bundles in the lobula valley tract were counted (23.2 ± 4.6; Table 3). Single bundles usually contained several axons and innervated just one or two specific ME layers (Figure 4e;  Backfilled, contralateral pigment-dispersing factor-immunoreactive (PDF-ir) fibers innervated the accessory medulla (AME) and projected via the AFF to pLA and ALA (open arrowheads; g), but not to ME4. ALAs were innervated by ipsi-and contralateral PDF-ir fibers (a-d,g,h) and were also strongly acetylcholinesterase (AChE)stained (c,h). Confocal laser images of horizontal (a-f,h) or frontal (g) brain sections through the ipsilateral optic lobe, stained with neurobiotin backfills from the COS (cyan), consecutive triple labeling with anti-PDF immunocytochemistry (magenta), and AChE histochemistry (white). Boxed areas (a,c) were enlarged (e-h). Lines indicate layer boundaries. a/pPDFMEs, anterior/posterior PDF-ir ME neurons (dashed ovals); d/vPDFLAs, dorsal/ventral PDF-ir LA neurons (dashed ovals); d, distal; do, dorsal (for g); p, posterior (for a-f,h). Scale bars = 50 μm [Color figure can be viewed at wileyonlinelibrary.com] anterior fiber fan projecting in ME1.1, in ME4 due to innervation via the GABA-ir medial layer fiber tract, and in the AME that is innervated by all the GABA-ir tracts (Figure 5a,b; Table 2). With antiserum against 5-HT we stained PDFLAs next to the accessory laminae, as well as somata next to the AME (not shown; Giese et al., 2018;Petri et al., 1995). Strongest expression of 5-HT immunoreactivity was observed in LA1-3, the accessory laminae, ME1.1, ME2, and the AME (Figure 5c,d; Table 2). In summary, both GABA-and even more 5-HT-ir neurons could interact with short photoreceptor terminals in the lamina, as well as with long photoreceptor neurons in ME2. While GABA appears to be dominant at arborization sites of PDF-ir clock neurons, 5-HT dominates at termination sites of photoreceptor cells.
The strong innervation of the accessory laminae by 5-HT further supports their photoreceptive function.
3.6 | ME4 is connected tightly to the circadian clock and exhibits multiple neuropeptide immunoreactivities  Table 2). Additionally, these three antisera, but not PDF antiserum-labeled ME2 where long histamine-ir photoreceptor cells terminated and ME3 that receives T A B L E 3 Innervation of specific medulla (ME) layers by fiber bundles (mean ± SD) backfilled from the contralateral optic stalk ME layer Fiber bundles (n = 5)  strong contralateral optic lobe inputs (Figures 2a,b, 4e, and 6a-f; Table 2). Interestingly, both ME4 and AME were labeled by all neuroactive substances tested in this study, and most of them expressed strong staining intensity (Figures 1-6; Table 2). As described above ( 3.7 | PDF and AChE signals, but not histamine overlapped in the glomeruli of the accessory laminae As shown previously, GABA-and 5-HT-ir terminals innervate the accessory laminae ( Figure 5; Table 2). Employing multiple-label immunocytochemistry the dorsal (not shown) and ventral accessory lamina  Table 2). In contrast, CRZ (Figure 7i,j; Table 2) and AT immunoreactivity (Figure 7k,l; Table 2) were not found in F I G U R E 8 (a,b) Model of cockroach photic entrainment pathways from UV-sensitive photoreceptors to sleep-promoting and from greensensitive photoreceptors to activity-promoting clock circuits. Both comprise of different pigment-dispersing factor (PDF) processing circadian clock neurons in the Madeira cockroach. (a) The largest, ipsi-and contralaterally projecting anterior PDF-immunoreactive (-ir) neuron (largest cPDFME; cyan) is suggested to keep sleep-promoting morning (M) circuits and activity-promoting evening (E) circuits at antiphase via an endogenous rhythm of PDF release. This largest clock neuron is suggested to be activated at increasing, high light levels via cholinergic input in accessory laminae (singular: ALA, plural: ALAs) via UV-sensitive extraocular photoreceptors (purple circle) of the lamina organ (LAO). Also, we suggest an inhibitory input pathway from green-sensitive short photoreceptors (green) in lamina (LA) layer LA3 to the largest cPDFME at low light levels. (b) Thus, light-dependently maximal PDF release from the largest cPDFME and the PDF-ir LA cells (PDFLAs) occurs during the day (open bar) in the accessory medulla (AME). The PDF rises in the AME during the day activate ipsilateral remaining iPDFMEs via excitatory PDF autoreceptors. The iPDFMEs were suggested to be sleep promoting, belonging to the morning oscillator (M). Furthermore, UV-sensitive long compound eye photoreceptor neurons (purple) that terminate in medulla (ME) layer ME2 advance iPDFMEs via medial neurons (MNes; orange) connecting ME2, ME4, and AME. Also, other MNes are suggested to relay extraocular UV input from the ALAs to iPDFMEs via the AME. At the same time PDF release during the day inhibits contralaterally projecting cPDFMEs via inhibitory autoreceptors. The cPDFMEs were suggested to be activity promoting, being part of the evening oscillator (E). The cPDFMEs lock onto dusk via processes in the LA adjacent to terminals of green-sensitive short photoreceptors from the compound eye. In parallel, PDFLAs and MNes such as the AT-ir MNe also provide light inputs from LA and ALAs into the AME. Arrows: excitatory connections, no arrows but bars: inhibitory connections, black bar: night. MLFT, median layer fiber tract; ZT, Zeitgebertime; LO, lobula; d, distal; p, posterior [Color figure can be viewed at wileyonlinelibrary.com] accessory laminae. To summarize, we found no evidence for histaminergic photoreceptor terminals in the accessory laminae, but for cholinergic, GABAergic, serotonergic, and neuropeptidergic innervation.

| DISCUSSION
In search for light entrainment pathways to the cockroach circadian clock, the accessory medulla (AME

| The number of layers in lamina and medulla are conserved among hemi-and holometabolous insects
In hemi-as well as in holometabolous insects the main photoreceptor organs are the compound eyes. The lamina and medulla, as well as most nested neuropils of the lobula (Rosner, von Hadeln, Salden, & Homberg, 2017) maintain topographic organization with regular parallel columns that form a retinotopic map of the external environment, thus, encoding external space (Strausfeld, 1976). Tangential neurons interconnect columns, branching perpendicularly to the columns and, thereby forming layers (= strata; Campos-Ortega & Strausfeld, 1972).
These layers serve parallel visual processing and encode different parameters of photic information such as light intensity, contrast, form, color, and motion embedded into spatial information (review: Borst, Haag, & Reiff, 2010). As tangential neurons interconnect corresponding layers between optic neuropils, they are the neuronal elements of parallel visual information processing. Because a main role of the circadian clock is to guarantee synchronization between the organism's endogenous timing and the timing of the external light dark cycle, we searched for direct/indirect connections between the circadian clock and terminals of compound eye photoreceptor axons.
Thus, we concentrated our analysis on tangential neurons connecting layers of lamina and medulla to the AME. We did not focus on the lobula which is involved in small-and large-field motion detection, in looming responses, and stereopsis (Aptekar, Keleş, Lu, Zolotova, & Frye, 2015;Nordström & O'Carroll, 2006). A detailed analysis of the lobula of the Madeira cockroach when compared with lobulae of other species was published recently (Rosner et al., 2017).
Consistent with the structure of the medulla in the fruitfly Drosophila melanogaster (Fischbach & Dittrich, 1989), in butterflies (Hamanaka et al., 2012;Heinze & Reppert, 2012), in the locust Schistocerca gregaria (Homberg, Brandl, Clynen, Schoofs, & Veenstra, 2004), in the mantis Hierodula membranacea (Rosner et al., 2017), and in another study of R. maderae (Rosner et al., 2017), we distinguished 10 layers of the medulla, despite the fact that we employed different techniques compared to previously published work. However, only eight medulla layers were described in Pieris brassicae (Strausfeld & Blest, 1970) and different species of Calliphora (Strausfeld, 1970). It was suggested before that these differences do not result from different staining methods. Rather, differences in layer organization, even between related species, were assumed to result from functional adaptation to different lifestyles during evolution (Heinze & Reppert, 2012). In our study, we wanted to determine in which optic lobe neuropils/layers photoreceptor cells could contact PDF-processing circadian pacemaker neurons of the Madeira cockroach as possible circadian light entrainment pathways.

| Parallel light entrainment pathways to morning or evening oscillator circuits of the insect circadian clock
The fruitfly D. melanogaster expresses a bimodal activity pattern with a peak in the morning and a peak in the evening, anticipating lights on or off. Consistent with the two oscillator models proposed previously for vertebrates (Daan & Pittendrigh, 1976), two separate circadian clock networks were identified for the generation of the morning and evening activity peak of Drosophila (Grima, Chélot, Xia, & Rouyer, 2004;Rieger, Shafer, Tomioka, & Helfrich-Förster, 2006;Stoleru, Peng, Agosto, & Rosbash, 2004). While the morning (M) oscillator network is coupled to dawn, being advanced by light, the evening (E) oscillator couples to dusk, being delayed by light (Helfrich-Förster, 2009). In fruitflies, the PDF-ir small LNvs are M oscillator cells, which control the activity of Drosophila in the morning, while PDF-ir large LNvs together with neurons processing other neuropeptides are part of at least three E oscillator circuits controlling activity and sleep at dusk and during the night (Helfrich-Förster, 2017;Johard et al., 2009;Schlichting, Díaz, Xin, & Rosbash, 2019;Schubert, Hagedorn, Yoshii, Helfrich-Förster, & Rieger, 2018).
In contrast to the holometabolous crepuscular fruitfly, the Madeira projecting PDFMEs were predicted to be activity-promoting E oscillator cells ( Figure 8) that are delayed by light at dusk. Furthermore, M cells were predicted to be active only during the light period, while antagonistic E cells were active only during the dark period per day Page, 1978;review: Stengl & Arendt, 2016).
In the fruitfly, the AME is a hub for light inputs from different organs that guarantee photic entrainment of the circadian clock (Li et al., 2018). The Drosophila circadian clock receives monosynaptic input from the Hofbauer-Buchner (HB) eyelet, an extraocular photoreceptor organ in the brain's optic lobes (Helfrich-Förster et al., 2002;Hofbauer & Buchner, 1989). Furthermore, the fruitfly AME receives indirect inputs from the compound eye via interneurons. Interestingly, the HB eyelet directly activates the PDF-ir M oscillators via cholinergic inputs, while via histaminergic input it directly inhibits the PDF-ir large LNvs that are arousal neurons and are part of the E oscillator circuit (Schlichting et al., 2016). Furthermore, the HB eyelet antagonizes with its differential connections the indirect photic inputs of the compound eye to the respective M and E oscillators dependent on daytime and on light regimes Schlichting et al., 2016;Schlichting, Weidner, et al., 2019). Thus, in analogy to the Drosophila circadian system, we would expect that lamina and lobula organs play a similar role as the extraocular HB eyelet, providing opposite inputs to M and E oscillator circuits and antagonizing compound eye inputs clock circuit-and light level-dependently (Figure 8).
4.3 | Cholinergic lamina and lobula organs as extraretinal photoreceptor neuropils are hypothesized to relay ambient light levels indicative of day or night to the circadian clock In beetles, lamina and lobula organs appear to be extraocular photoreceptive organs that send photoreceptor terminals to their respective firstorder optic neuropils: the accessory laminae and the AME (Frisch, Fleissner, Fleissner, Brandes, & Hall, 1996). In the Madeira cockroach, the lamina and lobula organs were also suggested to house extraocular photoreceptors that transmit ambient light information to accessory laminae and to the AME, the circadian clock. Consistent with this hypothesis, they were stained with antisera against UV-opsin and CRYPTOCHROME (Fleissner et al., 2001;Hofer, 2004). In D. melanogaster, CRYPTOCHROME is a blue-light-sensitive chromophore that directly affects clock proteins in the molecular feedback loop of the circadian clockwork (Ivanchenko, Stanewsky, & Giebultowicz, 2001). Furthermore, in beetles, lamina and lobula organs were labeled with antiserum against the circadian clock protein PERIOD (Frisch et al., 1996), supporting a close connection to the circadian system. In R. maderae, the accessory laminae are glomerular, nonretinotopic neuropils in proximity to the lamina organ, and are innervated by light-responsive MNes (OL2; ). Here, we show that accessory laminae do not receive histaminergic input from compound eye photoreceptors. Instead, they are strongly labeled with AChE histochemistry, indicative of cholinergic innervation.
Thus, it is likely that the visual input of the accessory laminae was obtained from cholinergic photoreceptors of the lamina organ, reminiscent of the lamina organ in beetles . Consequently, we hypothesize that the lamina organ provides cholinergic input to the cockroach clock (Figure 8), reminiscent of the HB eyelet in Drosophila (Li et al., 2018). Because at least two separate glomeruli per accessory lamina could be discerned that were innervated to opposite extends by PDF and

| The largest PDFME keeps rest-promoting M and activity-promoting E circuits in stable antiphase
As the largest contralaterally projecting cPDFME that arborized in accessory laminae, in LA3, and AME expresses excitatory ACh and inhibitory GABA receptors  it could be activated via cholinergic input during the day and inhibited GABA-dependently during the night.
Thus, differential light inputs from extraocular lamina organ and compound eyes would orchestrate it daytime-dependently (Figure 8a). Light regime-dependent modulation of light inputs was also observed in Drosophila indicating the circadian network gates light inputs daytimedependently and light input-dependently (Lazopulo, Lazopulo, Baker, & Syed, 2019;Schlichting, Weidner, et al., 2019). Thus, we predict that the largest PDFME releases PDF during the day, at all sites where it overlaps with other PDFMEs, such as in the AME. There, PDF release would affect all clock neurons that express PDF receptors. As there was a strict correlation between the branching pattern and PDF sensitivity in AME clock cells ) the ipsilaterally remaining PDF-sensitive clock neurons (such as PDFMEs) were activated by PDF, while contralaterally branching PDF-sensitive AME neurons (such as PDFMEs) were inhibited by PDF ( Figure 8). As PDF neurons orchestrate sleep-wake cycles in the Madeira cockroach (Reischig & Stengl, 2003a;Stengl & Homberg, 1994), we suggest that the ipsilateral PDFMEs that are active during the rest phase of the cockroach are sleep promoting. Because contralateral circadian clock neurons were suggested previously to control locomotor activity rhythms (Page, 1978) and because contralateral PDFMEs are inhibited during the day, when cockroaches rest, we suggest that they are activity promoting. When light levels decline at dusk the largest PDFME is not activated any more light-dependently and its GABAergic inhibition overturns its cholinergic activation, releasing the contralateral PDFMEs from its PDF-dependent inhibition. Thus, the largest PDFME that itself does not express PDF autoreceptors could maintain sleep-promoting M and activity-promoting E circuits in stable antiphase. Future experiments will test whether indeed the largest PDFME is required for stable sleep-wake cycles and is activated during the day via cholinergic input from lamina and lobula organs.
4.5 | Different parallel light entrainment pathways are suggested to antagonistically connect M and E clock circuits in the Madeira cockroach As green light was activity promoting and UV light was inactivity promoting in cockroaches there are parallel antagonistic pathways to premotor areas connected to specific photoreceptors (Zhukovskaya, Novikova, Saari, & Frolov, 2017). Thus, we hypothesize that also antagonistic, parallel photic entrainment pathways from different wavelength photoreceptors connect to the circadian clock that mediates behavior daytime-dependently similar to Drosophila clock circuits (reviews: Helfrich-Förster, 2019; Stengl & Arendt, 2016). We predict that at night green-sensitive short compound eye photoreceptors provide excitatory input in LA3 to contralateral PDFMEs as activitypromoting E cells while inhibiting ipsilateral PDFMEs (Figure 8b). In contrast, during the day UV-sensitive long compound eye photoreceptors provide excitatory photic input to rest-promoting ipsilateral PDFMEs and inhibitory light input to activity-promoting contralateral PDFMEs in the AME via interneurons that connect ME2, ME4, and AME ( Figure 8). Current experiments are testing this hypothesis.
4.6 | The CRZ-ir MNe is part of the advancing light entrainment pathway to M circuits closely interconnected with ipsilateral PDFMEs while AT neurons as part of the delaying entrainment pathway may connect to contralateral PDFMEs In support of this hypothesis is the single CRZ processing MNe that arborizes in ME4, extends a process to ME2 where UV-sensitive long photoreceptors terminate and strongly arborizes in the AME (Arendt et al., 2017;Petri et al., 1995). As injections showed that CRZ only advances the clock at dawn this cell would be part of the M circuit that should connect to ipsilateral PDFMEs. Indeed, it overlaps with the PDF-ir median-layer fiber system in a dorsal subregion of ME4 that backfills proved to originate from ipsilateral PDFMEs only. It is not known which two to four ipsilateral PDFMEs send an ipsilateral side branch from the anterior fiber fan into ME4, forming the medianlayer fiber system (Reischig & Stengl, 2002). However, because these PDF-ir fibers of the median-layer fiber system do not colocalize 5-HT, nor orcokinin (ORC), nor FMRFamide, nor leucokinin, it is likely that they originate from the posterior PDFMEs Petri et al., 1995;Söhler et al., 2011). Future experiments will examine whether posterior, but not anterior PDFMEs express CRZ receptors.
Also in the AME the CRZ cell could contact ipsi-as well as contralateral PDFMEs. Physiological studies will examine whether ipsi-and contralateral AME neurons respond antagonistically to CRZ, as predicted by our hypothesis.
As injections of AT revealed only phase delays, AT-ir MNes are suggested to be part of the phase delay pathway to E cells. The AT immunoreactivity was strongest in the AME, in ME5, and ME9. Thus, AT-ir neurons could contact contralaterally projecting anterior PDFMEs as part of the E circuit in the AME. Whether in ME5 and ME9 E cells arborize remains to be examined. Future experiments will test whether PDFLAs with arborizations in the proximal lamina relay input from green-sensitive photoreceptor neurons of the compound eye E circuits of the AME. Thus, in the AME both M and E circuits could be antagonistically connected via redundant neuropeptide and neurotransmitter pathways, to keep both circuits in antiphase or to shift their respective weight, depending on strength, frequency, and duration of the light inputs. 4.7 | Multiple neuropeptidergic circuits in the medulla may relay advancing and delaying photic and nonphotic phase information via ME4 to and from the circadian clock Injections of different neuropeptides or neurotransmitters into the vicinity of the circadian clock determined Zeitgebertime (ZT)-dependent phase shifts of locomotor activity rhythms. Therefore, these substances were part of input pathways to the clock (Arendt et al., 2017;reviews: Stengl & Arendt, 2016;Stengl et al., 2015). All phase-advancing and/or phasedelaying neuroactive agents appear to be present in different neurons innervating ME4 and AME. Thus, ME4 is a major input area for circadian clock neurons and also receives outputs from the clock. So far, the functions of these many clock inputs and outputs are not known. However, we assume that these circadian clock neurons release their respective neuropeptides ZT-dependently to control the phase of various physiological and behavioral rhythms in the cockroach (reviews: Stengl & Arendt, 2016;Stengl et al., 2015).
The strong innervation of ME4 as transition layer between the distal and proximal ME is striking and resembles the serpentine layer in other insects (Fischbach & Dittrich, 1989;Hamanaka et al., 2012;Heinze & Reppert, 2012;Xi, Toyoda, & Shiga, 2017). ME4 is densely innervated by the GABA-ir medial-layer fiber tract and by many neuropeptidergic neurons such as the PDFMEs. All neuropeptides and neurotransmitters examined were found in ME4, apparently processing parallel lines of clock inputs and clock outputs. Accordingly, Schubert et al. (2018) in D. melanogaster also described clock with dorsal receptive fields. Apparently in the same region of the medulla ipsi-and contralaterally projecting VMNes of the AME arborize that process polarization vision .
Thus, we hypothesize that the CRZ cell receives polarization information from a so far not described dorsal area of the cockroach compound eye, reminiscent of the dorsal rim area of locusts (review: Homberg, 2015). As the polarization pattern has its highest contrast in the UV range, these cells may be connected to UV-light detecting polarization-sensitive photoreceptor neurons. Future experiments will test this hypothesis and will determine whether ME4 is the first interaction of the circadian clock with the polarization vision pathway to allow for navigation oriented in space and time, CRZdependently.
To summarize, we hypothesize that light inputs from cholinergic extraocular photoreceptors in the lamina and lobula organs entrain sleep-wake rhythms of the Madeira cockroach via modulation of endogenous PDF release by the largest PDFME that maintains restor activity-promoting clock circuits in stable antiphase. Furthermore, parallel, either UV-sensitive compound eye photoreceptor neuron terminals in ME2 or green-sensitive terminals in the LA antagonistically inhibit or activate M or E oscillator circuits via different MNes or via different PDF processing neurons. Thus, ZT-dependently via distinct neuropeptide release, phase relationships of internal rhythms are controlled in synchrony with external light-dark cycles (Figure 8). Future experiments will test our hypothesis.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.