A SUBSET OF DORSAL NEURONS MODULATES CIRCADIAN BEHAVIOR AND LIGHT RESPONSES IN DROSOPHILA

SUMMARY

A fundamental property of circadian rhythms is their ability to persist under constant conditions. In Drosophila, the ventral Lateral Neurons (LNvs) are the pacemaker cells driving circadian behavior under constant darkness. Wild-type flies are arrhythmic under constant illumination, but flies defective for the circadian photoreceptor CRY remain rhythmic. We found that flies overexpressing the pacemaker gene per or the morgue gene are also behaviorally rhythmic under constant light. Unexpectedly, the LNvs do not drive these rhythms: they are molecularly arrhythmic and PDF – the neuropeptide they secrete to synchronize behavioral rhythms under constant darkness - is dispensable for rhythmicity in constant light. Molecular circadian rhythms are only found in a group of Dorsal Neurons: the DN1s. Thus, a subset of Dorsal Neurons shares with the LNvs the ability to function as pacemakers for circadian behavior, and its importance is promoted by light.

INTRODUCTION

Circadian rhythms give the sense of time to cyanobacteria and most eukaryotes, so that these organisms can adapt their physiology and behavior to daily environmental variations. These rhythms are generated by an endogenous, self-sustained molecular pacemaker (Dunlap, 1999). In Drosophila, this pacemaker is a transcriptional feedback loop (Hardin, 2005). Two proteins, PER and TIM, repress their own gene transcription by blocking the activity of two transcription factors: CLK and CYC. A set of kinases (DBT, CKII, SGG) and a phosphatase (PP2A) regulate PER and TIM phosphorylation, and therefore their stability and activity, so that the cycle lasts 24 hours. A second feedback loop regulates CLK expression. PDP1 and VRI are positive and negative transcriptional regulators of the clk gene, respectively, while CLK regulates positively their circadian expression (Hardin, 2005). The first loop is absolutely essential for circadian rhythms, but the function of the second loop still needs to be established. It might be important for the robustness of circadian rhythms, or their stability (Emery and Reppert, 2004). A strikingly similar molecular architecture that involves two interlocked feedback loops is also found in mammals (Shearman et al., 2000).

Recent studies have begun to elucidate the neural circuitry underlying circadian rhythms in Drosophila. This crepuscular animal shows two peaks of activity: around dawn and before dusk. Two separate groups of cells control these two peaks of activity: the ventral Lateral Neurons (LNvs) - that express the neuropeptide PDF - control the morning peak of activity, while the dorsal Lateral Neurons (LNds) and possibly two Dorsal Neurons (DNs) control the evening peak (Grima et al., 2004; Stoleru et al., 2004). A recent study suggests that a specific LNv that does not express PDF might also contribute to the evening peak (Rieger et al., 2006). The PDF positive LNvs have another crucial function: they maintain circadian rhythms in constant environmental conditions (constant darkness and constant temperature to be precise; Renn et al., 1999). These cells are believed to synchronize the other groups of circadian neurons through the rhythmic secretion of PDF (Park et al., 2000; Stoleru et al., 2005). The absence of the LNvs or of PDF results in rapid loss of behavioral rhythmicity under constant darkness, severely reduced amplitude of tim mRNA oscillations, and desynchronization of PER cycling within different groups of circadian neurons (Lin et al., 2004; Peng et al., 2003; Renn et al., 1999).

Since the LNvs control the oscillations of other circadian cells, they must be properly synchronized with the environment. The light:dark (LD) cycle is a crucial environmental cue. The LNvs receive two kinds of photic input. First, these cells are directly blue-light sensitive because they express the photoreceptor CRY (Emery et al., 2000b). Second, photoreceptive organs that express rhodopsins (eyes, ocelli, Hofbauer-Buchner eyelets) all contribute to a certain degree to the synchronization of the LNvs (Helfrich-Forster et al., 2001; Rieger et al., 2003).

CRY is thought to be the primary circadian photoreceptor, because it functions within circadian neurons (Emery et al., 2000b). Flies defective for CRY show very severe circadian photoreceptive defects. They cannot respond to short light pulses, while pulses as short as 1 minute can change the phase of circadian rhythms by several hours in wild-type flies (Egan et al., 1999; Stanewsky et al., 1998). They also react abnormally to constant light. Under these conditions, wild-type flies are arrhythmic, but flies without a functional CRY input pathway have a 24-hr period rhythm, as if they were in constant darkness (Emery et al., 2000a). Rescuing CRY function only in the LNvs is sufficient to significantly restore circadian behavioral light responses (Emery et al., 2000b). This indicates an important autonomous role of the LNvs in CRY dependent light responses. However, since these responses are not completely restored to normal, there might be other cells that contribute to CRY photoreception.

Here, we show that a gain-of-function mutation in the circadian pacemaker can also protect flies from the disruptive effects of constant light. Indeed, flies overexpressing the key pacemaker gene per are robustly rhythmic under constant illumination. Interestingly, our results demonstrate that the cells maintaining these behavioral rhythms are not the LNvs, but a subset of Dorsal Neurons of the DN1 group. Thus, these poorly characterized neurons play a central role in the control of circadian rhythms and the modulation of circadian responses to constant light.

RESULTS

Flies overexpressing per are rhythmic under constant light

The circadian behavior of wild-type flies is dramatically affected by the presence of constant light. The flies become arrhythmic after a day or two, while under constant darkness they would remain rhythmic for weeks (Konopka et al., 1989). This circadian response to constant light is dependent on the circadian photoreceptor CRY. cry^b flies, that carry a severely hypomorphic cry mutation (most likely a null mutation), remain rhythmic under constant light, with a periodicity of 24 hours, as if they were under constant darkness (Figure 1A; Emery et al., 2000a).

Figure 1. Flies overexpressing per are rhythmic in LL.

(A) Average double-plotted actograms of 16 wild-type (control) and 16 cry^b flies under constant light. Both groups of flies were first synchronized to a light dark (LD) cycle (grey shadings indicate the dark phase), and then released under constant light (LL, indicated with an arrow head). Wild-type flies become rapidly arrhythmic under constant light, while cry^b flies are robustly rhythmic, with a 24-hr period. Each day (except the first) is plotted twice: on the right half of the actogram and then on the left half, on the next line. The first day is plotted only once on the left half of the first line. Arrhythmic flies are included in all average actograms, including those of figure 4 and 5.

(B) Average double-plotted actograms of flies overexpressing per in different groups of circadian cells under LL conditions (16 flies per genotype). per was overexpressed in different groups of circadian neurons using a combination of tissue-specific GAL4 and GAL80 drivers to drive UAS-per. per overexpression was driven by either the tim-GAL4 driver (tg4-per) or the pdf-GAL4 driver (pg4-per). tim-GAL4 was also combined with the pdf-GAL80 (tg4-pg80-per) and cry-GAL80 (tg4-cg80-per) repressive transgenes (see the Material and Methods section for details). Note that overexpression of per in the PDF positive LNvs is neither sufficient, nor required for rhythmicity under constant light. (C). Percentage of rhythmic flies for each genotype shown in (B), and for their controls (the driver and repressor transgenes without uas-per). The average period is indicated above the bars for the robustly rhythmic genotypes. See also Table I for more details.

Interestingly, we found that when we overexpressed per with the tim-GAL4 driver (Emery et al., 1998; Kaneko and Hall, 2000; Kaneko et al., 2000), which is active in every cell with circadian rhythms (genotype: y w;tim-GAL4/+; uas-PER/+), almost all flies showed a robust ca. 26.8-hr period phenotype under 200 lux constant light (LL; Figure 1B and C, and Table I). The vast majority of control flies were arrhythmic (Table I). Only a few flies showed residual rhythmicity of weak amplitude; their period was similar to that observed in constant darkness (DD). Under DD, per overexpressing flies had a longer period than their control (25.7 hr vs 24.8 hr; Table S-I), but that period length was shorter than under LL (25.7 hr vs 26.8 hr). Thus, manipulating the level of PER expression, a central element of the molecular circadian pacemaker, protects flies from the disruptive effects of constant light. However, while severe mutations in the CRY input pathway result in flies that are blind to constant light (Emery et al., 2000a; Koh et al., 2006), flies overexpressing per are still partially responsive to LL.

Table I. Behavior of flies overexpressing per and morgue under LL conditions (200 lux).

Genotypes with robust rhythms are highlighted in bold. Rhythmic flies have a power greater than 10 and a width greater than 2. See Ewer et al. (Ewer et al., 1992) for power and width definition

gene	GAL4 driver	GAL80 driver	Number of flies	Number of rhythmic flies	Period average (±st.dev.)	Power average (±st.dev)
per	tim-GAL4	-	31	28	26.8±1.5	51.1±31.1
per	pdf-GAL4	-	32	1	25.5	41.9
per	tim-GAL4	pdf-GAL80	26	25	26.7±1.7	56.1±32.2
per	tim-GAL4	cry-GAL80	30	7	25.2±0.9	33.9±12.0
per	cry-GAL4	-	15	0	-	-
per	gmr-GAL4	-	28	0	-	-
morgue	tim-GAL4	-	30	29	26.2±0.6	52.7±16.6
morgue	pdf-GAL4	-	30	1	16.4	17.0
morgue	tim-GAL4	pdf-GAL80	32	32	26.4±0.3	65.8±18.3
morgue	tim-GAL4	cry-GAL80	32	5	24.7±1.7	25.4±12.7
-	tim-GAL4	-	28	6	24.6±2.6	29.1±16.3
-	pdf-GAL4	-	28	0	-	-
-	tim-GAL4	pdf-GAL80	28	7	25.1±1.8	28.3±13.0
-	tim-GAL4	cry-GAL80	27	4	23.2±1.7	13.8±2.2
-	cry-GAL4	-	16	2	21.3±0.2	13.8±0.4
-	gmr-GAL4	-	14	0	-	-

Non-PDF circadian neurons maintain circadian behavioral rhythms under constant light

The LNvs are the cells maintaining circadian rhythms under constant darkness (Lin et al., 2004; Peng et al., 2003; Renn et al., 1999). In their absence, flies become rapidly arrhythmic, within 2–3 days. Moreover, CRY expression in the LNvs has been reported to significantly restore responses to constant light in cry^b flies (Emery et al., 2000b). Thus, the simplest explanation for why flies overexpressing per remain rhythmic in LL is that somehow the LNvs have lost most of their light sensitivity. Therefore, we tested whether restricting per overexpression to these cells would result in LL rhythmicity. We drove per overexpression with pdf-GAL4, a driver that is specifically expressed in the LNvs in the adult fly brain (Renn et al., 1999). Unexpectedly, this restricted per expression did not result in LL rhythmicity (Figure 1B and C, and Table I). This suggests that the LNvs are not the critical cells for circadian rhythms in LL. To verify that this result was not due to a lower level of per expression in flies with the pdf-GAL4 driver compared to those with the tim-GAL4 driver, we drove per overexpression in flies with the tim-GAL4 driver, but excluded this overexpression from the LNvs with the pdf-GAL80 repressive transgene (Stoleru et al., 2004). These flies no longer overexpress per in the LNvs (Figure S-1), but still do so in most (if not all) other clock neurons (data not shown). They also have a normal period length in DD, which indicates that the period lengthening was due to overexpression of per in the LNvs (Table S-1). Nevertheless, the tim-GAL4 / pdf-GAL80/ UAS-per flies were as rhythmic as the tim-GAL4 / UAS-per flies in LL, and the period length of their behavior was identical (Figure 1B and C, Table I).

Moreover, when tim-GAL4 was used in combination with cry-GAL80, rhythmicity was greatly reduced, and the period of the few remaining rhythmic flies was shortened to 25.2 hours (Figure 1B and C, and Table I). cry-GAL80 blocks tim-GAL4 expression in the LNds and the PDF negative LNv, in addition to the PDF positive LNvs (Stoleru et al., 2004). Most likely, it also represses tim-GAL4 driven expression in all the other DNs since cry is expressed in these cells, but this repression is not as complete. cry is also expressed in the eyes, and cry-GAL80 could thus potentially block tim-GAL4 in this tissue as well. However flies with overexpression of per driven by the eye-specific gmr-GAL4 driver remained completely arrhythmic in constant light (Table I).

Taken together, these results indicate that dorsally located circadian neurons (or possibly the unique PDF-negative LNv) modulate the responses to constant light and share with the PDF-positive LNvs the ability to maintain circadian rhythms over a long period of time under constant conditions. To confirm that the LNvs were not rhythmic in flies overexpressing per with tim-GAL4 in LL, we measured PDP1 levels in these cells by immunohistochemistry. PDP1 shows robust circadian oscillations with a very narrow concentration peak between ZT18 and ZT21 under LD conditions (Cyran et al., 2003). Thus, PDP1 is an excellent phase marker. We monitored tim-GAL4/UAS-per flies behaviorally in LL and dissected the brains of the flies that were rhythmic to determine whether PDP1 oscillates in their LNvs. As shown on Figure 2, PDP1 did not oscillate in the LNvs of flies collected at CT2 and 17, which are the predicted peak and trough time points for PDP1 staining in flies with 26.8-hour period rhythms on the 3^rd day of LL. As expected, we did not observe oscillations at CT10 and 21 either, which would have occurred had the LNvs continued to oscillate with a period close to that of wild-type flies (data not shown). This proves that the LNvs are not circadianly functional under LL, even when per is overexpressed. This strengthens the notion that circadian neurons other than the PDF-positive LNvs can maintain circadian rhythms in LL on their own.

Figure 2. The LNv and LNd neurons do not show molecular circadian oscillations in LL when per is overexpressed.

The LNv and LNd neurons were immunostained with anti-PDP1 (red) in the adult brain of flies overexpressing per (per O/E, genotype: yw;tim-GAL4/+; UAS-per/+) and control (yw) flies. Anti-PDF (green) was used to localize the PDF (+) small and large LNvs.

(A). Expression of PDP1 under a Light:Dark cycle (LD). PDP1 protein is highly concentrated in the nuclei of the small LNvs (s), the large LNvs (L) and the LNds at ZT21 in adult brains of control flies. As expected, PDP1 (+) cells are not detected at ZT10. A similar pattern of staining was observed in adult brains of flies that overexpress per (per O/E), suggesting normal circadian oscillations under this light regime. There was a notable exception (indicated with asterisks) in the large LNvs (L), which did not express detectable PDP1 levels at ZT21. The LNvs might be particularly sensitive to PER dosage, which is a repressor of pdp1 transcription.

(B). Expression of PDP1 in constant light (LL). Adult brains were dissected during the third day in LL at the predicted peak (CT2) and trough (CT17) of PDP1 abundance, taking into consideration the long circadian period of flies overexpressing PER. PDP1 protein is highly concentrated in the nuclei of the small LNvs (s), the large LNvs (L) and the LNds at both CTs in adult brains of the control as well as the per O/E flies, even though these flies are behaviorally rhythmic. Subjective scoring did not reveal any significant variations of staining intensity in any of the LN subtypes (data not shown)

Circadian oscillations persist in DN1 neurons in LL when per is overexpressed

To determine which cells might generate behavioral rhythms under constant light, we studied PDP1 staining in non-PDF circadian neurons of tim-GAL4/UAS-per flies. We first focused on the LNds, since these cells are believed to be the E-cells critical for the control of the evening activity (Grima et al., 2004; Stoleru et al., 2004). No oscillations of PDP1 staining could be detected in this group of neurons in per overexpressing flies under LL. PDP1 was constantly high, as 5–6 cells with PDP1 nuclear staining were detected at all 4 time points tested (Figure 2 and data not shown). A PDF-negative LNv has recently been implicated in the control of the evening peak of activity as well, and could underlie the long period behavioral rhythms observed in cry^b flies under LL when these flies split their behavior into a short and long period component (Rieger et al., 2006). In several brains dissected either at CT2 and CT17, we observed high PDP1 levels in a cell closely associated with the LNvs that was PDF negative, but we cannot be certain that this cell was the PDF-negative LNv. Indeed, additional PDP1 positive, PDF negative cells were seen in the vicinity of the LNvs. These cells did not appear to show circadian oscillations of PDP1 either.

However, when we looked at the DN1 group, we clearly saw a much larger number of positive cells for PDP1 staining at CT2 compared to CT17, which are the predicted PDP1 peak and trough for per overexpressing flies, taking into account their long period phenotype in LL (Figure 3). On average, we saw ca. 8 positive neurons at the predicted trough for PDP1, while there were ca. 13 positive neurons at the predicted peak. Thus, a subset of DN1s oscillates in LL when per is overexpressed. Importantly, the number of PDP1 positive cells was low at both CT10 and 21, even though during the LD cycle there were low at ZT10 and high at ZT21 (Figure 3). Therefore, the period of the molecular oscillations in the oscillating subset of DN1s is not 24 hours, but is longer by several hours. This fits well with the period of the circadian behavior of per overexpressing flies in LL. This result strongly supports the idea that it is a subset of dorsal neurons that controls circadian behavior under constant light.

Figure 3. A subset of DN1 neurons show molecular circadian oscillations in LL when per is overexpressed.

DN neurons were identified by anti-PDP1 (red) immunostaining in the adult brain of per overexpressing (per O/E, genotype yw; tim-GAL4/+; UAS-per/+) and control (yw) flies. Projections coming from the LNvs are stained with anti-PDF (green),

(A). Expression of PDP1 in the DNs under a Light:Dark cycle (LD). PDP1 protein is highly concentrated in the nuclei of the DN1, the 2 DN2 and the DN3 neurons in both control and per O/E brains at ZT21. As expected, PDP1 (+) cells are not detected at ZT10.

(B). Expression of PDP1 in the DNs in constant light (LL). PDP1 protein is highly concentrated in the nuclei of the 2 DN2 and the DN3 neurons at all CTs tested for both the control and the per O/E brains. As shown in the quantification to the right of the confocal sections, about 15 DN1s are PDP1 (+) at the two time points tested in the control brains (Number of brain hemispheres quantified: 4 for CT2 and 4 for CT17). On the other hand, per O/E brains show robust PDP1 oscillation in a subgroup of DN1 neurons with a peak at CT2, which is the predicted peak of PDP1 expression for per overexpressing flies (Number of brain hemispheres quantified: 8 for CT21, 11 for CT2, 8 for CT10 and 14 for CT17). Student’s t tests show that the differences between the number of PDP1 positive cells at CT2 and the other time points is statistically significant (P < 10^−4,, indicated with an asterisk), while there are no statistically significant differences between the other time points. This strongly suggests that this group of oscillating DN1s is responsible for the rhythmic behavior observed in LL. Subjective scoring did not reveal any significant variations of staining intensity in PDP1 positive DNs (data not shown).

We also examined the DN2 and DN3 groups in LL with per overepxression (figure 3). The DN2s did not oscillate; both DN2 neurons were PDP1 positive in most brains at all 4 time points. In the DN3 group, PDP1 did not appear to oscillate either, even though we cannot exclude that a small subset of these ca. 40 neurons were rhythmic. In conclusion, robust molecular oscillations are restricted to a subset of DN1s under constant illumination in per overexpressing flies. This result, combined with our genetic data, indicates that these are the neurons maintaining circadian behavior in LL. Therefore, they play an important role in the neural circuits regulating circadian rhythms.

The DN1s also drive circadian rhythms in flies overexpressing morgue under constant light

To obtain an independent confirmation of the important role of the DN1s in the circadian neural circuits, we turned to flies overexpressing morgue (Wing et al., 2002). In a genetic screen that will be described in details elsewhere, we isolated several genes that can protect flies from the disruptive effects of constant light when overexpressed with the tim-GAL4 driver (A.M, M. E.-L., Michael Rosbash and P.E., unpublished results). The strongest phenotype was observed with morgue, and was very similar to that observed with per overexpression: a long period phenotype of 26.2 hr (Figure 4A and C, and Table I). In DD, circadian behavior was normal (Table S-1).

Figure 4. LL rhythmicity is dependent on the DN1s in flies overexpressing morgue, but does not require PDF.

(A) morgue was overexpressed with the same combinations of transgenes described in figure 1. As with per overexpression, the PDF cells are not the cells critical for LL rhythmicity. The average period is given for the genotypes that were robustly rhythmic. See also Table I for more details.

(B). The LNv, LNd and DN neurons were immunostained with anti-PER (red) in the adult brain of flies overexpressing morgue (genotype: yw;tim-GAL4/UAS-morgue) in constant light (LL). Anti-PDF (green) was used to localize the PDF (+) small and large LNvs. Adult brains were dissected during the third day in LL at the predicted peak (CT6) and trough (CT17) of PER nuclear abundance, taking into consideration the long circadian period of these flies. PER protein is highly concentrated in the nuclei of the small LNvs (sLNvs), the large LNvs (LLNvs), the 5^th PDF(-)LNv (*), the LNds, the 2 DN2s and the DN3s at both CTs. Staining intensity does not show significant oscillations in these cell groups (data not shown). However, as shown in the quantification to the right of the confocal sections, morgue overexpressing brains show robust PER oscillation in a subgroup of DN1 neurons with a peak at CT6 (Number of brain hemispheres quantified: 8 at CT6, 5 at CT17). The difference in the number of PER positive DN1s between the two time points is statistically significant (asterisk, P < 0.005).

(C) Average double-plotted actograms of flies overexpressing morgue with the tim-gal4 driver (genotype yw; tim-GAL4/UAS-morgue; +, n=16), flies overexpressing morgue with the tim-gal4 driver in a pdf⁰¹ background (genotype yw; tim-GAL4/UAS-morgue; pdf⁰¹, n=28) and control flies (genotype yw; UAS-morgue/+; pdf⁰¹, n=8) under LL conditions. The arrowhead indicates when the flies were released in LL.

As shown on Figure 4 and Table I, flies overexpressing morgue were very robustly rhythmic in LL conditions when tim-GAL4 was used, but not when pdf-GAL4 was used. The addition of pdf-GAL80 to tim-GAL4 flies had no effect, further demonstrating that non-PDF cells are important for constant light rhythmicity. Finally, as with per overexpression, blocking morgue overexpression with cry-GAL80 led to arrhythmicity in constant light.

We then determined which circadian cells are oscillating at a molecular level in the brains of morgue overexpressing flies. We used a PER antibody for these experiments. The staining was done on the 3^rd day of LL at CT6 and CT17, which are the predicted peak of nuclear PER accumulation and its concentration trough, respectively, based on the period length of the behavior (Shafer. et al., 2002). The results were strikingly similar to those observed with PDP1 staining in per overexpressing flies (Figure 4B). The LNvs did not oscillate, including the PDF-negative LNv that was this time unambiguously identified. The LNds did not cycle either. However, there were very clear molecular oscillations in a subset of DN1s. Both DN2s were strongly stained at CT6. Staining was more variable at CT17, but some brains still had both DN2s that were PER positive. The DN3s did not appear to cycle. Thus, as observed with per overexpression, robust molecular oscillations are limited to a subset of DN1s when morgue is overexpressed. These results indicate that the DN1s are maintaining circadian behavioral rhythms in LL when morgue is overexpressed, and strongly support the notion that these cells play a central role in the control of circadian rhythms.

PDF is not required for circadian behavioral rhythms under constant light

In DD, the LNvs synchronize behavior and the other brain circadian neurons such as the DN1s through the rhythmic secretion of PDF from their dorsal projection (Park et al., 2000; Peng et al., 2003, Lin et al., 2004; Stoleru et al., 2005). Since the LNvs are molecularly arrhythmic in morgue or per overexpressing flies, PDF secretion should be arrhythmic too. However, the DN1s might be able to induce rhythmic PDF secretion even when there is no functional circadian clock in the LNvs. Indeed, the DN1s send projection toward the LNvs (Kaneko and Hall, 2000), and per⁰ flies in which PER expression (and thus circadian rhythms) has been rescued in every neuron except the LNvs show morning anticipation, even though this anticipatory behavior is normally controlled by the LNvs (Stoleru et al., 2004). Rhythmic PDF secretion driven by the DN1s could even feedback and help the DN1s to remain rhythmic in LL. To determine whether PDF secretion is required for rhythmic behavior in LL, we overexpressed morgue in pdf⁰¹ mutant flies (Renn et al., 1999). Under a light:dark cycle, these flies showed the typical advance in the phase of the evening activity found in pdf⁰¹ flies. As expected, most of them became arrhythmic in DD (Table S-II), although we observed more rhythmicity than in pdf⁰¹ control flies (the degree of residual rhythmicity varies in PDF deficient flies of different genetic background; see Renn et al., 1999). The period of the rhythmic morgue overexpressing pdf⁰¹ flies was short, as previously observed with the rhythmic pdf⁰¹ flies (Renn et al., 1999). In LL however, ca. 60% of morgue overexpressing pdf⁰¹ flies remained rhythmic (Table II and figure 4C). This demonstrates that output from the LNvs is dispensable for rhythmicity in constant light, and reinforces the notion that the DN1s can function independently of the LNvs when light is present. However, it should be noted that the behavioral rhythms observed in LL without PDF are not as robust as those observed in the presence of PDF (higher degree of arrhythmicity, lower amplitude) and their period is about one hour shorter than control, as observed in DD. Thus, although PDF is not needed for LL rhythms, it influences their property.

Table II.

Behavior of pdf⁰¹ flies overexpressing (or not) morgue under LL conditions (200 lux). Genotypes with robust rhythms are highlighted in bold.

Genotype	Number of flies	Number of rhythmic flies	Period average (±st.dev.)	Power average (±st.dev)
pdf01	20	0
y w; tim-GAL4/+; pdf01	25	5	25.2±1.7	27.0±21.6
y w; +/UAS-morgue; pdf01	25	0
y w;tim-GAL4/UAS-morgue; pdf01	65	37	25.0±1.0	29.2±12.3
y w; tim-GAL4/ UAS-morgue; +	16	15	26.2±0.7	40.5±14.2

Inhibition of the CRY input pathway allows the DN1s to remain rhythmic in LL

CRY is responsible for the arrhythmic behavior observed under constant light, presumably because under these conditions it constantly degrades the pacemaker molecule TIM (Emery et al., 2000a; Stanewsky et al., 1998). Thus, the mechanism that allows the DN1s of flies overexpressing per or morgue to escape the disruptive effects of constant light might be a repression of the CRY input pathway. If this hypothesis were correct, we would expect that other behavioral circadian responses to light would be affected in these flies. Wild-type flies delay their clock after a short early night light pulse, while they advance their behavior with a late night light pulse. In flies with cry mutations, these responses are severely reduced or absent (Busza et al., 2004; Stanewsky et al., 1998). We therefore tested the ability of flies overexpressing morgue to respond to short light pulses. These flies responded to short light pulses like flies with a hypomorphic mutation in cry (cry^m; Busza et al., 2004): phase shifts could be detected, but they were very severely reduced compared to control (figure 5A). This result strongly suggests that the CRY input pathway is inhibited in morgue overexpressing flies.

Figure 5. Inhibition of the CRY input pathway allows the DN1s to remain rhythmic in LL.

(A) Phase response curve for wild-type flies (Canton-S strain, cs, solid black line) and flies overexpressing morgue with the tim-gal4 driver (genotype yw; tim-GAL4/UAS-morgue; +, solid red line). For comparative purposes, the response of cry^m flies (adapted from Busza et al. 2004, dashed line) has been included. Flies were entrained under a 12-hour light:12-hour dark regime. The light intensity during the day was 1000 lux. The flies were then pulsed during the last night of the light:dark regime at 3000 lux for 5 min, and then left in constant darkness. Their phase was compared to those of flies that had not been pulsed. Phase change is plotted on the y axis; phase delays and advances are shown as negative and positive values, respectively. The x axis represents the Zeitgeber Time (ZT) of the light pulse. Data are averages of three independent experiments; Standard Deviations between experiments are shown.

(B) Average double-plotted actograms of cry^b flies in which cry expression was rescued either with the pdf-gal4 driver (genotype yw; pdf-GAL4/UAS-cry; cry^b, n=13) or the tim-gal4 driver (genotype yw; tim-GAL4/UAS-cry; cry^b, n=15) under LL conditions. The arrowhead indicates when the flies were released in LL.

(C) Expression of PDP1 in the DNs of LNv-rescued cry^b in constant light (LL) at CT10 and CT21. DN neurons were identified by anti-PDP1 (red) immunostaining in the adult brain of cry^b flies in which cry expression was rescued with the pdf-gal4 driver (genotype yw; pdf-GAL4/UAS-cry; cry^b). Projections coming from the LNvs are stained with anti-PDF (green). As shown in the quantification on the right of the confocal sections we observed robust coherent PDP1 oscillations in ca. 6–7 DN1s (Number of brain hemispheres quantified: 7 for both time points). Student’s t tests show that the difference between the number of PDP1 positive cells between the two time points is statistically highly significant (P<10^−4,, indicated with an asterisk). Only the DN1s showed coherent circadian oscillations (see Table S-III). This indicates that this group of circadian neuron is responsible for the rhythmic behavior observed in LL.

In an earlier study, it was shown that cry^b mutant flies expressing wild-type CRY in the LNvs only (genotype: y w; pdf-GAL4/UAS-cry; cry^b) are partially rhythmic under constant light (Emery et al., 2000b). About half of the LNv-rescued cry^b flies were rhythmic. We wondered whether the DN1s might be the pacemaker neurons in these flies. We first monitored LNv-rescued cry^b under our current experimental conditions and found that about 50% of them were rhythmic in LL for at least 6 days (Figure 5B and data not shown). As expected, CRY expression with tim-GAL4 fully rescued the cry^b phenotype under LL (i.e. all the flies were arrhythmic). To determine whether the DN1s are the cells generating LL rhythms, we measured PDP1 levels in the brains of LNv-rescued cry^b flies. Since these flies exhibit ca. 24-hr period rhythms, we dissected the brains at CT21 (predicted peak) and CT10 (predicted trough). As expected, no oscillations could be detected in the LNvs of LNv-rescued cry^b flies since they express CRY (Table S-III). PDP1 levels were lower than those observed in flies overexpressing per or in wild-type flies, suggesting that the clock in the LNvs is frozen at a different time point in LNv-rescued cry^b flies. This could be due to a more extensive degradation of TIM, since CRY should be overexpressed in these cells. We could not identify the LNds in these brains, presumably because PDP1 levels were very low. PDP1 levels were also constantly low in a subset of DN1s (Figure 5C and Table S-III). The number of DN2 positive cells was higher at CT21 than at CT10 (Table S-III), but this oscillation was not statistically significant. Finally, staining in the DN3 was low in all brains at CT10, but the number of PDP1 positive cells varied considerably at CT21 (Table S-III). This suggests that the DN3 might be oscillating, but that after three days in LL their oscillations are not synchronized properly any more, probably because they do not get synchronization signals from the LNvs. However, we observed robust, coherent PDP1 oscillations in ca. 6–7 DN1s (figure 5C).

These results are very important. First, they confirm that a subset of DN1s play the role of pacemaker cells for circadian behavior in LL. Second, since this last set of results is obtained in flies with a cry loss-of-function mutation, rather than flies overexpressing a specific gene, the conclusion is that the DN1s are intrinsically able to control and generate self-sustained circadian behavioral rhythms when light is present. Their ability to do so when overexpressing morgue or per is thus not due to a gain-of-function that would have given them a property that they do not usually have. The DN1s thus play an important role in the control of circadian behavior and its responses to light.

DISCUSSION

Recent studies have shown that two groups of cells control circadian behavior. The PDF positive LNvs are called morning cells (M-cells), and the LNds evening cells (E-cells), because they control the anticipatory behavior observed before dawn and dusk respectively (Grima et al., 2004; Stoleru et al., 2004). In addition, the LNvs are the cells maintaining circadian behavior in constant darkness and controlling the phase of most circadian neurons of the brain (Lin et al., 2004; Peng et al., 2003; Renn et al., 1999; Stoleru et al., 2005). In their absence, circadian behavior rhythms are lost after a few days in DD. Surprisingly, our results show that a functional circadian clock in the LNvs is actually not necessary for long-term behavioral rhythms. In flies overexpressing PER, the LNvs are no longer circadianly functional under constant illumination. No oscillation of the circadian protein PDP1 can be detected and yet these flies remain rhythmic for at least 7 days. Moreover, limiting per overexpression to circadian neurons that do not express PDF is sufficient to obtain circadian behavioral rhythms under constant environmental conditions.

We believe that the neurons maintaining circadian behavior independently of the LNvs are not the E-cells. Indeed, when per is overexpressed, we did not see any sign of circadian oscillation in the neurons that are thought to control the evening activity: the LNds (Grima et al., 2004; Stoleru et al., 2004). In addition, the PDF negative LNv that might also contribute to the evening activity (Rieger et al., 2006) did not cycle in LL when morgue was overexpressed. Moreover, flies with per overexpression driven by cry-GAL4 were completely arrhythmic under constant light (Table I). cry-GAL4 is one of the critical GAL4 driver used to define the E-cells (Stoleru et al., 2004). Importantly, we actually detected molecular circadian oscillations in only one group of cells when per was overexpressed: the DN1s. Due to the high number of DN3s, we cannot rule out that a few cells in the DN3 groups also oscillate. Interestingly, Veleri et al. (2003) have previously shown that a subset of DN3 neurons can maintain their own circadian oscillations in DD, in the absence of circadianly functional LNvs. However, these DN3 cells were not able to generate rhythmic behavior in DD. While it is possible that light is a necessary co-factor for these self-sustained DN3s to participate in the control circadian behavior, we favor the hypothesis that it is the DN1s that maintain circadian rhythmicity in LL. This idea is strongly supported by several additional findings. First, the phase of PDP1 molecular oscillations in the DN1s on the 3^rd day of LL fits well with the long period of the circadian behavior observed under these conditions in per overexpressing flies. Second, the behavioral observations made with morgue overexpression also suggest that the critical cells for rhythmicity are not the LNvs, and PER staining in morgue overexpressing flies gave us an independent confirmation that robust circadian molecular oscillations are restricted to the DN1s in LL. Finally, in LNv-rescued cry^b flies, only the DN1s show robust, coherent circadian rhythms in phase with the behavioral rhythms. Remarkably, the DN1s can maintain circadian behavior in LL even when PDF is absent. This indicates that they can work autonomously of LNv output. Interestingly, not all DN1s do oscillate in LL, only about 6–7 cells most likely. This shows that the DN1 group is heterogeneous. This is not surprising, since the different groups of circadian neurons were named based on their location in the brain, not on their function or developmental lineage. There is ample evidence for heterogeneity of morphology, gene expression and behavior within these different groups of cells, including the DN1s (see for example Rieger et al., 2006; Shafer et al., 2006).

Thus, a subset of DN1s can control and generate circadian behavioral rhythms. They must therefore play an important role in the circadian neuronal circuits. Since ablation of the M cells and E cells results in flies with no morning and evening activity, and no self-sustained rhythms in DD (Stoleru et al., 2004), this could mean that the DN1s are usually functioning downstream of the M and E cells. This is further supported by the fact that in the absence of the neuropeptide PDF - believed to be the critical synchronizing signal secreted by the M cells – the DN1s cannot maintain their circadian rhythms in the long run in DD (Lin et al., 2004; Peng et al., 2003). The DN1s can thus probably function as a relay connecting the LNvs with the neurosecretory cells of the pars intercerebralis (PI), believed to play an important role in the control of locomotor behavior (Helfrich-Foerster et al., 1998: Kaneko and Hall, 2000). A LNvs-DN1-PI pathway has also been suggested based on the anatomical studies of the projections of the small LNvs and the DN1s (Kaneko and Hall, 2000). The expression of the receptor for PDF in at least a subset of DN1s also supports the existence of a functional connection between them and the LNvs (Hyun et al., 2005; Lear et al., 2005; Mertens et al., 2005). The implication of this connection is that in wild-type flies under LL, the LNvs should constantly send a disruptive signal to the DN1s, presumably the non-oscillating secretion of PDF.

This leaves us with the following question: if the LNvs and rhythmic PDF secretion are normally required for the DN1s to be rhythmic, why are the DN1s able to free themselves from the disruptive effects of constant light, while at the same time becoming independent of the LNvs? Our results show that an important mechanism is the inhibition of the CRY-dependent light input pathway. Indeed, morgue overexpressing flies are defective in the CRY-dependent behavioral responses to short light pulses, and cry loss-of-function mutations also result in rhythms driven by the DN1s. In the case of per overexpression, we presume that TIM role is reduced, since one of its major functions is to protect PER from proteasomal degradation (Grima et al., 2002; Ko et al., 2002; Price et al., 1995). TIM is the target of CRY, thus its reduced importance would result in DN1s that are less sensitive to the CRY input pathway. In addition, overexpression of SHAGGY, which inhibits CRY signaling, also results in LL rhythms driven by dorsal neurons (Stoleru et al., 2007). However, under natural environmental conditions, inhibition of the CRY input pathway is probably not required for the DN1s to participate in the control circadian rhythms. Indeed, even in the polar regions of the globe that experience constant light conditions during the summer, the elevation of the sun varies during the day, and this should result in variations of temperature sufficient to synchronize the DN1 circadian clock (Yoshii et al., 2005).

The mechanism by which the DN1s avoid to become arrhythmic in LL as a result of the molecular arrhythmicity of the LNvs, which should result in constant PDF secretion, is not clear yet. It is possible that the presence of light inhibits PDF signaling and thus promotes the role of the DN1s. Light input could come from the eyes, ocelli, or from the DN1s themselves (Rieger et al., 2003). Alternatively, as mentioned in the result section, the DN1s could induce rhythmic PDF secretion. The fact that PDF is not required for LL behavioral rhythms does not exclude this possibility, particularly since the robustness of the rhythms is improved by the presence of PDF.

Interestingly, per and morgue overexpression results in a very similar long period phenotype under LL, which could suggest that these two molecules coincidentally affect the period length of the circadian molecular pacemaker in the same way. In DD however, per overexpression does affect behavioral period length, while morgue does not. The long period phenotype observed in LL actually probably reflects the fact that the CRY input pathway is not completely blocked in the DN1s of per or morgue overexpressing flies. Indeed, under very low light intensity, wild-type flies exhibit a long period phenotype as well (Konopka et al., 1989). In addition, morgue overexpression does not completely block the CRY-dependent responses to short light pulses (figure 5A). Finally and most importantly, LNv-rescued cry^b flies - in which the CRY input pathway is completely non-functional in the DN1s – have 24-hr period rhythms. The LNv-rescued cry^b flies show nevertheless a higher degree of arrhythmicity than normal cry^b flies, or than flies overexpressing morgue or per. This might be due to the desynchronization observed within the DN3 group of circadian neurons. Indeed, the DN3s do not appear to be desynchronized in per or morgue overexpressing flies.

A previous report had already shown that LNv-rescued cry^b flies are partially rhythmic (Emery et al., 2000b), and this was interpreted as evidence for a functional role of CRY directly in the LNvs. Our new results show that expression of CRY in the LNvs is probably not very important for the response to constant light. The DN1s are the important cells for this response. Does this mean that CRY is not a photoreceptor in the LNvs? We believe it actually does function as a photoreceptor in the LNvs as well. CRY is expressed in these cells (Emery et al., 2000b; Klarsfeld et al, 2004), and LNv-rescued cry^b flies show very significantly rescued responses to short light pulses. Preliminary experiments with morgue overexpression limited to the LNvs confirm a predominant role of these cells for light pulse responses (A. M. and P. E., unpublished data). Thus, the CRY input pathway might mediate response to short light pulses by its action in the LNvs, and constant light responses by its action in the DN1s.

In summary, our work underscores the importance of the DN1s in the control of circadian behavior and responses to light. Earlier genetic studies have indicated that the DN1s modulate the sensitivity of the circadian network to light:dark cycles of very low light intensity (Klarsfeld et al., 2004). Our results significantly extend this observation by showing the profound impact the DN1s have on the response to constant light and by demonstrating that these cells not only modulate circadian light responses, but can also become the driving force controlling circadian locomotor behavior, and this in the absence of environmental cues and functional LNvs. This confers upon them a unique status among non-PDF circadian neurons. One of our striking results is that genetically identical flies rely either on the LNvs or the DN1s for the control of their circadian rhythms, depending on the presence or absence of light. Indeed, the LNvs determine period length in our experiments with per overexpression in DD, but in LL the DN1s set the pace. That the presence or the absence of light can so remarkably shift the dominance from one cell group to the other strongly suggests that the relative contribution of the LNvs and DN1s to the control of circadian rhythms change during the course of the year, particularly at high latitude. The DN1s, which interestingly generate evening activity (figure 1,4 and 5), would play a more prominent role in the control of circadian behavior during the long days of the summer, while the LNvs would be more important when photoperiods are shorter (see also Stoleru et al., 2007).

MATERIAL AND METHODS

Drosophila Stocks and Transgenics

The following Drosophila strains were used in this study: y w; UAS-per 24 (Kaneko et al., 2000), cry^b ss (Stanewsky et al., 1998). y w; tim-GAL4 (Emery et al., 1998), y w; pdf-GAL4 (Renn et al., 1999). cry^b flies with rescued wild-type CRY expression have been described previously (Emery et al., 2000b). To overexpress morgue, we used a P-element containing UAS binding sites inserted in front of the morgue gene: EP(2)2367 (Rorth et al., 1998; Wing et al., 2002). EP(2)2367 and tim-GAL4 were separately combined with the pdf⁰¹ mutation (Renn et al, 1999) in y w background, so that pdf⁰¹ mutant flies overexpressing morgue could be generated. A y w; tim-GAL4, pdf-GAL80/CyO; pdf-GAL80/ TM6B strain was generated in two steps to be able to drive per overexpression in every clock cells expect the LNvs. First the 2^nd chromosome tim-GAL4 insertion was meiotically recombined with a chromosome containing two 2^nd chromosome insertions of the pdf-GAL80 transgene (Stoleru et al., 2004). Recombinants with darker eye colors than the parental strains were selected, which should contain both tim-GAL4 and at least one copy of pdf-GAL80. The presence of both pdf-GAL80 and tim-GAL4 in the recombinants was confirmed by PCR. Then, we added two copies of the pdf-GAL80 transgene carried on the 3^rd chromosome. The final stock thus contains at least three copies of pdf-GAL80, which ensure a very strong repression of GAL4 in the LNvs (see figure S1). To further restrict per overexpression, a y w; tim-GAL4/CyO; cry-GAL80/TM6B strain was generated by combining a 2^nd chromosome insertion of tim-GAL4 with two cry-GAL80 insertions on the 3^rd chromosome (Stoleru et al., 2004).

Behavioral Analysis

Locomotor activity of male flies (1–5 days old) were measured with Trikinetics Activity Monitors (Waltham, MA, USA) for 3 full days under 12 hr light: 12 dark conditions (LD) followed by 6 full days of either constant light or constant darkness at 25°C. For almost all experiments, a light intensity of ca. 200 lux was used. Data analysis was performed with the FAAS software (Grima et al., 2002). Rhythmic flies were defined by χ² periodogram analysis with the following criteria: power 10, width 2 (Ewer et al., 1992). The group activity actograms were generated using a signal-processing toolbox (Levine et al., 2002) implemented in MATLAB (MathWorks). Phase Responses Curves were generated essentially as described in Busza et al. (2004).

Whole-Mount Immunohistochemistry and Quantitation

Adult Drosophila (1–5 days old) were entrained to 3 days of LD and then released in LL. Adult brains were dissected during the second and third day of LL. Immunotainings for PER were performed as previously described (Lear et al., 2005). The same protocol was used for PDP-1 staining. The anti-PDP1 antibody was a generous gift from Justin Blau and was used at a concentration of 1:5,000. For the PDF staining, anti-PDF (generous gift from Michael Rosbash) was used at a concentration of 1:400 and then visualized using a FITC-anti-rat secondary antibody (Jackson Immunoresearch, PA, USA). All samples were mounted in BioRad antifade reagent, and viewed on a Zeiss LSM5 Pascal confocal microscope. At least two experiments for each time points were performed and produced comparable results. The images presented in the figures are overlays of several confocal stacks (1μm) obtained using ImageJ software (freely available at http://rsb.info.nih.gov/ij/). After discriminating between the different neurons on the overlays, the brains were scored for the number of PDP1 or PER positive cells and staining intensity by an observer who was blind for genotypes and time points. Staining intensity was subjectively scored from 0 (no staining) to 5 (high staining) for all PDP1 positive cells of two to three representative brain hemispheres, except for the DN3s. For this group, ca. 10 representative neurons were scored.