Light triggers a network switch between circadian morning and evening oscillators controlling behaviour during daily temperature cycles

Under standard LD conditions Drosophila melanogaster exhibits crepuscular behavior with periods of activity centered around the light transitions in the morning and evening. Interestingly, the evening activity peak observed in LD, is also observed in LLTC [29], albeit with a phase advance compared to LD. We hypothesized that the timing of this activity peak is controlled by the same clock neurons, known as the evening oscillator (EO). Restricting clock function to these cells is sufficient to drive the evening peak in LD [11]. To test whether this oscillator is required for controlling the behavioral evening activity phase, we interfered with clock function in restricted groups of clock neurons using the dominant negative form of cycle (UAS-cyc^DN) [31] (Figs 1 and S1, Table 1). Both Mai179-Gal4 and cry-Gal4[19]- are expressed in the evening oscillator but also in the s-LNv and the l-LNv [11, 30]. However, Mai179-Gal4 is only weakly expressed in morning oscillator neurons [11]. Furthermore, while cry-Gal4[19] expression is restricted to clock neurons, Mai179-Gal4 is broadly expressed in non-clock neurons. Flies were exposed to three days of LD followed by two to three days in LL, and subsequent LLTC, which was delayed by 5 h compared to LD (Fig 1A). Interestingly, morning anticipation in LD was not affected by the expression of cyc^DN in the s-LNv (Fig 1A–1C), presumably because clock function in the DN1p is sufficient [12, 18]. As expected, expression of cyc^DN in the morning and evening oscillator with both drivers strongly reduced the amplitude of the evening peak, which anticipates the environmental transition in both LD and LLTC (Fig 1A–1C). While the startle response to lights-off and the following rapid activity decline in LD was not affected, both Mai179-Gal4 and cry-Gal4[19] > cyc^DN flies showed increased activity in the cryophase during LLTC (Fig 1A–1C). In the following we focus on the amplitude of the anticipatory behavioral evening peak, because it is under tight clock-neuronal control and also influenced by the environment. A decrease in amplitude can be due to reduced synchronization within the population, or to reduced speed of activity increase resulting in a delayed activity peak. To first quantify the amplitude reduction, we calculated the slope for each fly (the speed of activity increase, Δactivity/Δt). The two time points (start of activity increase and activity peak) were determined based on median activity levels (S1A Fig and Materials and Methods). Using this method, we observed a strong decrease of the slope in both Mai179>cyc^DN and cry[19]>cyc^DN flies in LD and LLTC (Figs 1D, 1E and S1B). To test whether this decrease was due to a decrease of synchronization, or a (synchronized) delay of anticipatory activity, we compared the median of the slope value (Slo_exp) with the slope of the median (Slo_the). Therefore, if fly activity is highly synchronized between individuals, the ratio of Slo_exp/Slo_the is close to 1 (i.e., most individuals behave similar to the median). While the Slo_exp/Slo_the ratio for the controls was close to 1, it was reduced to ~0.5 for both Mai179>cyc^DN and cry[19]>cyc^DN flies (S1B Fig). In addition, to easily visualize if a population is synchronized, we compared the percentage of flies with a Slo_exp > ½ x Slo_the with flies that are below this value. Indeed, > 80% of the control flies increase their locomotion with Slp_exp > ½ Slp_the, compared to only 50–60% of the Mai179>cyc^DN and cry[19]>cyc^DN flies (S1C Fig, blue bars percentage of flies with Slo_exp > ½ of Slo_the ‘anticipating’, orange bars percentage of flies with Slo_exp < ½ of Slo_the ‘non-anticipating’). Taken together, these results indicate that expression cyc^DN in both evening and morning neurons reduces behavioral synchronization of evening activity in both LD and LLTC.

How temperature cycles entrain the molecular clock in the brain is not known. Previous observations suggest that temperature entrainment uses multiple molecular and probably neuronal circuits involving peripheral thermosensors [32–36]. We therefore tested, whether the absence of a functional molecular clock in the Mai179 cells could disturb the other clock cells in the brain. For this, we dissected brains and quantified PER levels on the 6^th day of LLTC at four different time points. First, we confirmed the circadian clock disruption after cyc^DN expression in Mai179 cells, because we observed only 3 out of 6 LNd at ZT0 (Fig 2A), while PER levels were constitutively low in the other Mai179 cells (Fig 2C). In the remaining CRY^– LNd, we observed normal PER oscillations, albeit with a slight amplitude decrease (Fig 2B). Normal PER cycling was also maintained in the other, largely CRY^– neurons, not expressing Mai179 (Fig 2D), indicating that they receive independent temperature input. Flies with ablated morning and evening oscillator neurons also exhibit synchronized PER oscillations in the DN1-3 during temperature cycles in DD, further supporting the existence of multiple temperature inputs into the brain clock [37].

In order to restrict cyc^DN expression exclusively to the EO we combined Mai179-Gal4 with Pdf-Gal80 (Materials and Methods). To test the efficiency of Pdf-Gal80 repression, we dissected Mai179-Gal4 Pdf-Gal80 flies expressing both GFP and cyc^DN in LD at ZT2 and analyzed PER and GFP signals. The absence of GFP expression in the PDF⁺ cells confirmed the efficiency of Pdf-Gal80. Also, as expected, PER was absent from 2–3 CRY⁺ LNd, but in the 5^th sLNv PER expression was similar as in the l-LNv (Fig 3A), suggesting that in this genotype the effect of cyc^DN expression in the 5^th s-LNv was not 100% penetrant. Behaviorally, in LLTC the amplitude of the evening peak was significantly reduced (Fig 3B–3D), demonstrating that a functional clock in the evening cells is necessary for synchronizing the evening activity peak in LLTC. Interestingly, although the slope is reduced in LD, Mai179-Gal4 Pdf-Gal80 > cyc^DN flies are still synchronized (S2A and S2C Fig), whereas in LLTC they are desynchronized (S2B and S2C Fig). The reduced, but otherwise synchronized slope in LD suggests a delayed activity peak, which is indeed visible on the first day in LL (Fig 3B). Next, we applied an even more restricted driver to disturb clock function in a subset of the EO neurons. spE-Gal4 is expressed in the 5^th s-LNv and the CRY⁺ LNd, but not in the DN1a [38]. Interestingly, expression of cyc^DN does not alter the shape of behavioral evening activity in both LD and LLTC (S3 Fig). Although this may indicate a prominent role for the DN1a in regulating evening activity, we think the weak expression of the spE-Gal4 driver allows for sufficient clock function in the 5^th s-LNv and the CRY⁺ LNd evening cells of spE > cyc^DN flies (see S5 Fig).

To determine if clock function within the MO or other clock neurons is also required for the evening activity peak in LLTC, we screened additional clock neuronal Gal4 drivers (S4B Fig and Table 1). Interestingly, only Clk856-Gal4, which drives expression of cyc^DN in all brain clock neurons, showed a significant decrease of the synchronized evening behavior, while none of the other drivers showed significant differences compared to the respective controls in LLTC. In particular, neither Pdf-Gal4 nor DvPdf-Gal4 which are both expressed in all of the PDF⁺ MO neurons, had an effect on the evening peak (S4 Fig). In agreement with previous findings [22], this suggests that a functional clock in the MO is not necessary for synchronizing behavior in LLTC. Moreover the DvPdf-Gal4 result shows that expression of cyc^DN in only two out of the six EO neurons (1 CRY⁺ LNd and the 5^th s-LNv, Table 1) is not sufficient to interfere with synchronizing evening activity in LLTC (S4 Fig).

To underpin the role of the EO in controlling evening activity, we ablated these neurons using the Gal4/UAS system to drive expression of the pro apoptotic genes head involution defective (hid) and reaper (rpr). Because of the broad expression pattern of Mai179-Gal4 outside the brain clock, activation of UAS-hid,rpr using this driver led to lethality. In contrast, cry[19]>hid,rpr flies are viable and show the almost complete disappearance of the evening peak in LD and in LLTC (Figs 4 and S5A–S5C). Because cry[19]-Gal4 is also expressed in the morning oscillator neurons, we next used spE-Gal4 in order to ablate the CRY⁺ LNd and the 5^th s-LNv [38, 39]. Similar to the spE>cyc^DN experiments, there was no obvious difference between spE>hid,rpr flies and their parental controls in LD and the slope was not significantly different (Figs 4A, S5A and S5B). However, compared to controls, spE>hid,rpr flies showed a reduced Slo_exp/Slo_the ratio (0.67 for spE>hid,rpr compared to 0.86 and 0.89 for both parental controls, respectively), which is correlated with a 15% increase of flies showing non-anticipatory behavior (S5C Fig). During LLTC, spE>hid,rpr flies show a strongly reduced amplitude of the evening activity, accompanied by an increase of the percentage of desynchronized flies (Fig 4A–4D), confirming the role for the evening neurons during temperature synchronization in constant light. To determine ablation efficiency we dissected spE>hid,rpr flies in LD at ZT2-ZT3 and immunostained the brains with PER as a marker. Although not 100% penetrant, in the majority of brains only 3 to 4 LNd could be detected, presumably representing the CRY^– LNd subset. In addition, the 5^th s-LNv was detectable in 50% of the hemispheres, although it was never detectable in both hemispheres of the same brain (S5D and S5E Fig). Hence, although the incomplete penetrance of this driver is most likely responsible for the weak behavioral phenotype in LD, the stronger phenotype observed in LLTC indicates a more prominent role for the CRY⁺ PDF^– neurons in LLTC compared to LD.

To confirm that the MO neurons are not required for the synchronization to LLTC, we induced their ablation by expressing hid and rpr using Pdf-Gal4 and DvPdf-Gal4. As a positive control, we also ablated all clock neurons using Clk856-Gal4. Indeed, Clk856 > hid,rpr were the only flies that were completely desynchronized both in LD and LLTC (S6A and S6C Fig). As expected, both Pdf > and DvPdf > hid,rpr flies revealed altered behavior in LD, revealing the typical lack of morning anticipation and advance of the evening peak (S6A and S6B Fig). Interestingly, we also observed increased desynchronization in LD, particularly with the Pdf-gal4 driver (S6C Fig). During LLTC, both Pdf > and DvPdf > hid,rpr flies showed robust synchronization (S6 Fig). The slightly decreased synchronization after DvPdf-Gal4 ablation can be explained by the expression of this driver in a small subset of EO neurons (5^th s-LNv and 1 CRY⁺ LNd). In summary, the ablation experiments strongly support an essential role for the EO neurons for synchronizing fly behavior to temperature cycles in LL.

Flies behave differently in LLTC compared to DDTC, even though the temperature regime remains the same [29]. For example in DDTC (25°C:16°C), flies are synchronized to the beginning, while in LLTC they are synchronized to the end of the thermophase ([29], Fig 1). If, depending on the light condition, the neuronal network balance switches from one state to another, the EO should not be operating during DDTC. To test this, we interfered with clock function by expressing cyc^DN in both the MO and EO neurons (cry[19]-Gal4 > cyc^DN) or only in the EO (cry[19]-Gal4, Pdf-Gal80 > cyc^DN). Strikingly, while cry[19]Gal4 > cyc^DN flies exhibit a very broad activity peak, covering almost the entire warm phase, flies with cyc^DN expression restricted to the EO have their activity peak in the 1^st half of the warm phase, indistinguishable from the controls (Fig 5A and 5B), and suggesting that the MO controls behavior under these conditions [29, 37]. In addition, while in DD and constant temperature cry[19]-Gal4 > cyc^DN flies become arrhythmic, due to the expression of cyc^DN in the s-LNv pacemaker neurons [31], cry[19]-Gal4, Pdf-Gal80 > cyc^DN exhibit synchronized rhythmic activity (Fig 5A and 5C). This demonstrates both, the effectivity of Pdf-Gal80-mediated repression of cyc^DN expression in the s-LNv morning cells, and the previous synchronization to temperature cycles (based on the maintenance of the activity phase obtained during DDTC). Because Mai179-Gal4 expression is weaker in the MO compared to the EO neurons [11], we also analyzed Mai179>cyc^DN in DDTC: Strikingly, while synchronization was strongly reduced in LLTC (Fig 1), Mai179>cyc^DN flies exhibited robustly synchronized activity rhythms in DDTC (S7 Fig), however with a phase delay compared to controls (S7A and S7B Fig). Nonetheless, after switching to DD, flies free run with the same phase as during the previous DDTC regime, indicating stable synchronization of the circadian clock and confirming weak s-LNv expression of this driver (S7A and S7C Fig). Restricting Mai179 > cyc^DN expression to the EO, by introducing Pdf-Gal80, restored the phase to the beginning of the thermophase as in controls (S7A and S7B Fig), showing the weak cyc^DN expression in the s-LNv is enough to change the phase in DDTC. In summary, these results show that while the EO controls behavioral synchronization during LLTC, the MO takes over this role in the absence of light.

An important question in chronobiology that so far remains unanswered is the mechanism of entrainment by light and temperature when these two environmental parameters are actually not so reliable considering their substantial daily variation [40]. Notably, substantial weather-inflicted variations of temperature and light intensity can occur during the day and these must not lead to circadian clock resetting. Nonetheless, animals behave quite differently depending on the current environmental status regardless of their clock entrainment status. For example, fruit flies lose their two anticipatory activities present in LD 25°C at different constant temperatures. At warm temperature (≥30°C), flies only anticipate the light-on transition while at colder temperatures (<20°C) they only anticipate the lights-off [41]. On the other hand, when temperature cycles are used as Zeitgeber in constant darkness, the absolute temperatures determine the phase of the activity peak, with TCs of 25°C:16°C and 29°C:20°C inducing activity in the first half or the second half of the thermophase, respectively [29, 37]. In constant light however, both temperature cycles result in an activity peak during the second half of the thermophase [29].

Here, we demonstrate that the EO that drives the evening peak in LD25°C is necessary to drive the evening peak in LLTC25°C-16°C, while the MO that drives the morning peak in LD25°C controls the phase in DDTC25°C-16°C. According to our model the ambient environmental condition modulates the clock network balance (Fig 6). Hence, the environmental condition must be taken into account in order to understand the role of each oscillator. Previously, trying to understand how temperature cycles synchronize flies, a similar approach has been performed, although in the absence of light and different absolute temperatures, but with the same amplitude of temperature oscillations (29°C-20°C). Consistent with our findings, ablating both the MO and EO using a cry-gal4 line lead to largely desynchronized behavior [37] (Fig 5). As mentioned above, in DDTC 29°C:20°C conditions flies exhibit an afternoon activity peak, and ablating the PDF neurons does not affect this synchronization [37]. In contrast, our data show that a functional clock in the PDF neurons is important for synchronized morning activity in DDTC 25°C:16°C (Figs 5 and S7). Following our model, we postulate therefore that the DN1p clock neurons also contribute to the behavioral evening activity during warm temperature cycles in constant darkness. The DN1p can drive evening activity during DDTC 29°C:20°C and during low intensity LD cycles when the temperature is constant, while at high light intensity LD cycles they support morning activity [18]. Furthermore, the DN1p receive warm input via the TrpA1 expressing AC neurons [42]. Interestingly however, TrpA1 is not required for temperature entrainment as such, but shapes behavior under warm conditions [42–44]. Notably, while the loss of TrpA1 function has no effect on the behavioral phase in DDTC 25°C:16°C, TrpA1 mutant flies show an advanced increase of activity and a reduced siesta in DDTC 29°C:20°C compared to controls [43]. Hence, taken together these data re-enforce our model where, depending on the environmental condition, the clock network can change its balance to set behavioral activity phase, bypassing a change of clock entrainment (Fig 6).

Furthermore, our data suggest that the length of the light period (here 12 h vs 24 h) influences the involvement of the EO. Ablating the PDF neurons not only advances the evening peak in LD but it also drastically reduces the synchronization of the flies (S6B and S6C Fig), while they remain strongly synchronized in LLTC. Furthermore, while ablating the EO, even with incomplete penetrance, drastically affected the presence of the synchronized evening peak in LLTC, it only slightly reduced synchronization in LD. Hence, although the clock of the EO can be entrained independently to light [11], our data indicate that the PDF neurons have a strong influence on the robustness of EO-output in LD25°C but not in LLTC. This suggests that the extent of dark periods increases the influence of the PDF neurons during the day. The evening peak occurs under the same environmental condition (lights-on and 25°C), however with a slight advance in LLTC compared to LD (Fig 1). Therefore, we propose that the past experience during the night/cryophase determines the influence of one group toward the other. In summary, these data confirm that the network status varies with the environmental condition, and therefore it is crucial to consider the specific environmental conditions when deciphering the different contributions of circadian network components in regulating behavioural activity. Our simple environmental protocol provides a template to further test and extend this model. For example, using temperature cycles as Zeitgeber, we can now modify light quality and intensity to test how the clock network responds to these light variations while the clock is stably entrained.

Synchronization of clock protein oscillations in LPN neurons is preferentially sensitive to temperature [21–23]. However, we have previously observed that these neurons are not necessary for rhythmic behavior under TC conditions in both LL and constant darkness [29]. Here, we confirm the non-requirement of clock function within the LPN for rhythmic locomotor behavior in LLTC (S4B Fig), suggesting that LPN sensitivity for molecular synchronization to TC serves another function unrelated to entrainment. Indeed, a recent study shows that the LPN are activated above 27° and they are important for increased siesta sleep at 30°C [45]. CRY is expressed in about 1/3^rd of the clock network in the brain and plays an important role in light entrainment [7, 46], consistent with the idea that CRY⁺ neurons are more sensitive to light and CRY^– cells are more sensitive to temperature [23]. Notably, isolated brains can still be entrained to LD cycles [47], and in the absence of visual input all clock neurons are entrained to LD [46], indicating that the CRY⁺ neurons synchronize the remaining CRY^– neurons non-cell autonomously [48]. However, here we demonstrate an essential role of the CRY⁺ neurons in controlling rhythmic behavior in LLTC. Nevertheless, even in the absence of all CRY⁺ neurons, flies exhibit weak synchronized behavior with an evening phase during temperature cycles (in LL and DD), suggesting that CRY^– clock neurons (subsets of DN1_p and LNd, DN2, and DN3) also play a role in temperature synchronization [29, 37]. Nevertheless, because they are not able to instruct the CRY⁺ neurons during TC (Fig 2), their role in temperature entrainment is not as prominent as that of CRY⁺ neurons in LD (see above). Moreover, ablation of PDF⁺ or all three groups of LN (LNv, LNd, LPN) still allows for weak behavioral evening synchronization to 30°C: 25°C TC in LL [22]. Furthermore, the fact that in the absence of a clock in the Mai179⁺ cells (Fig 2), after ablation of all CRY⁺ cells [37], or in the absence of all LN [22], molecular oscillations in other clock neurons can be synchronized, supports the idea that multiple and independent pathways contribute to temperature entrainment [32, 34–36].

Flies were raised in a 12 h:12 h light dark (LD) cycle on a medium containing0.7% agar, 1.0% soy flour, 8.0% polenta/maize, 1.8% yeast, 8.0% malt extract, 4.0% molasses, 0.8% propionic acid, and 2.3% nipagin at 25°C and 60% relative humidity. The following fly lines were used in this study: Mai179-GAL4/CyO [11], cry[19]-GAL4 [30], spE-Gal4 (JRC-MB122B) [38], DvPdf-GAL4 [49], R16C05-Gal4 (BL69492) [39], Clk4.1M-GAL4 [17, 18], Clk9M-Gal4 [50], Pdf-GAL4 [51], Clk856-GAL4 [52];UAS-cyc^DN [31], Pdf-GAL80 (BL80940), GFP-cry [53]. spLPN-GAL4 (R11B03-p65.AD; R65D05-GAL4.DBD) [54] was provided by Taishi Yoshii. Mai179-GAL4, UAS-cyc^DN, Clk9M-GAL4, and Clk4.1-Gal4 have been outcrossed for 5 generations to iso31 to standardize the genomic background [55]. We recently showed that only flies carrying at least one copy of the natural ls-tim allele, encoding both the light sensitive S-TIM and less light sensitive L-TIM protein [27] are able to synchronize to temperature cycles in constant light, while flies homozygous for the s-tim allele (encoding only S-TIM) cannot [27]. Therefore, all strains analyzed in this study carry at least one copy of ls-tim.

Analysis of locomotor activity of male flies was performed using the Drosophila Activity Monitoring System (DAM2, Trikinetics Inc., Waltham, MA, USA) with individual flies in recording tubes containing food (2% agar, 4% sucrose). Briefly, DAM2 activity monitors containing LD-entrained flies were placed inside a light- and temperature-controlled incubator (Percival Scientific Inc., Perry, IA, USA). Fly activity was monitored for at least 11 days with the first 2 days in LD25°C (700–1000 lux generated by 17W F17T8/TL841 cool white Hg compact fluorescent lamps, Philips) followed by three days in LL 25°C and then LLTC 25°C-16°C with a shift delayed by 5h relative to the previous LD. The LD behavior analysis was performed on the last day of LD, while LLTC behavior was analyzed on the 6^th day of LLTC.

Plotting of actograms was performed using a signal-processing toolbox [56] implemented in MATLAB (MathWorks, Natick, MA, USA). The 24h locomotor activity plots were generated using a custom excel macro [42]. Activity was averaged into 30 minute bins and normalized to the maximum individual activity level. The median of this normalized activity was plotted, because it is a more representative parameter of behavioral synchrony compared to the mean. To measure the slope, we manually determined the latest time point of the minimum median (ZT_min) and the maximum median (ZT_max) before Zeitgeber transition. The only exception is shown in S7C Fig, where the downhill slope was analyzed in the subjective day of DD2, because controls show a better synchronization while decreasing their activity, compared to the period when activity increases. We calculated for each fly the derivative of the line between ZT_max and ZT_min: Slope = (Act_max- Act_ZTmin)/(t_ZTmax-t_ZTmin). Time was counted in minutes, but rounded to full hours in figures and legends for convenience. The box plots were made using Excel. The statistical tests were performed using the freely available Estimation Statistics [57].

Adult male Drosophila brains were immuno-stained as described previously [55]. Briefly, brains were fixed in 4% paraformaldehyde for 20 min at RT, and blocked in 5% goat serum for 1 h at RT. Primary antibodies used were as follows: rabbit anti-DsRed (Clontech)– 1:2000; mouse anti-PDF (Developmental Studies Hybridoma Bank, DSHB)– 1:2000; mouse anti-Bruchpilot (nc82, DSHB)– 1:200; chicken anti-GFP (Invitrogen)– 1:1000 –rabbit anti-PER [58] 1:15000. Alexa-fluor secondary antibodies (goat anti-rabbit 555, goat anti-chicken 488, goat anti-mouse 488; Invitrogen) were used at 1:2000 except for labeling anti-BRP where goat anti-mouse 647 where a dilution of 1:500 was used. Confocal images were taken using an inverted Leica LSP8. PER signal intensity was quantified using an 40x oil-objective and imageJ. Pixel intensities were normalized to the background after background subtraction [(signal-background)/background] using Excel.