Responses of activity rhythms to temperature cues evolve in Drosophila populations selected for divergent timing of eclosion

Many insects show rhythmic emergence of adults from their pupal cases. These rhythms persist in the wild and in the laboratory under both light:dark (LD) cycles and constant conditions (Saunders, 2002). Among them, a large proportion of species time this remarkable phenomenon to occur just after dawn. These include, for instance, the yellow dung fly (Scopeuma stercoraria), the Queensland fruit fly (Dacus tryoni), moths (Pectinophora gossypiella and Heliothis zea) and many Drosophila species (Saunders, 2002). These observations have raised two interesting questions. What is the mechanism by which organisms restrict emergence to certain times of the day? Why is emergence predominantly restricted to dawn in so many insect species? These questions have been of interest for many decades, and we are now aware of the presence of biological time-keeping mechanisms (also referred to as circadian clocks) that generate and drive rhythmicity in many aspects of behaviour, and across almost all living beings (Dunlap et al., 2004). To the question of why emergence is restricted to dawn, Colin S. Pittendrigh in the mid-1950s hypothesised that organisms must have evolved to time emergence to the time of the day when humidity is high and temperature is low. This was thought to allow efficient wing expansion in pharate adults and therefore enable survival (Pittendrigh, 1954).

Pittendrigh's hypothesis implied that timing of emergence and modalities that allow sensation of and responses to temperature and/or humidity are intimately linked. Adaptations to capitalise on a temporal niche are thought to be multi-tiered such that multiple aspects of behaviour, physiology and morphology evolve together (Daan, 1981). For instance, in addition to circadian clock properties, waxy cuticles to prevent water loss and enhanced vision are thought to have evolved in diurnal insects, while improved sound and olfaction are thought to have evolved in nocturnal birds and mammals (Daan, 1981). However, clear demonstration of the genetic association of various aspects of physiology and circadian clock properties via the evolution of behavioural timing has been lacking. With the goal of understanding such relationships, our laboratory generated and currently maintains four large and outbreeding D. melanogaster populations that are artificially selected for morning and evening adult emergence (first described in Kumar et al., 2007).

In relation to Pittendrigh's 1954 hypothesis, we have recently demonstrated that laboratory selection for evening timing of emergence (as opposed to morning) is strongly associated with the co-evolution of enhanced temperature sensitivity of the circadian clock circuit regulating adult emergence rhythms (Abhilash et al., 2019). This clearly demonstrates a genetic correlation between behavioural phase and temperature responses, and is in agreement with the above hypothesis. Although Pittendrigh's argument was made based on cycles of both temperature and humidity, due to technical limitations, there are hardly any studies on the role of humidity in emergence rhythms. Effects of temperature, however, are fairly well studied in both adult emergence and locomotor activity rhythms (Pittendrigh, 1954; Konopka, 1972; Das and Sheeba, 2017; Selcho et al., 2017).

It is important to note here that: (i) while the emergence rhythm is a population level phenomenon (each individual can emerge from its pupal case only once), locomotor activity is an individual level rhythm, and furthermore (ii) adult emergence and adult locomotor activity are two very different physiological processes occurring at two entirely different life stages of the fly life cycle. Despite this, the first study that isolated and described the effects of period (per) mutation on behavioural rhythms in Drosophila found that both emergence and activity/rest rhythms are affected in a similar manner for all alleles of the per locus (Konopka and Benzer, 1971). Subsequently, it was demonstrated that the small ventral lateral neurons (s-LNvs) are necessary to drive behavioural rhythms in eclosion and locomotor activity under constant conditions (Myers et al., 2003; Grima et al., 2004; Stoleru et al., 2004), thereby illustrating the close overlap in timing machinery regulating both rhythms. Additional evidence also comes from the fact that the free-running period (FRP) estimated using eclosion and activity/rest rhythms are strongly positively correlated with each other in flies from our morning and evening selected populations (Kumar et al., 2007; Nikhil et al., 2016b), again highlighting a common machinery regulating both behaviours. In light of these similarities in organisation of the circadian clock across rhythms spanning two very different behaviours, we asked if temperature sensitivity of the clock regulating the adult locomotor activity rhythm also evolved in the evening emerging flies. In reference to the way the circadian network is organised, what does this imply?

To address this question, we first subjected our flies to simulated jetlag of 6 h phase advance (equivalent to eastward travel, e.g. London to Bangkok) and 6 h phase delay (equivalent to westward travel, e.g. London to Chicago) using temperature cycles alone. We found that flies from the evening populations resynchronise to phase-shifted temperature cycles much faster than the morning and control populations. This result indicated differences in the temperature-sensitive components of the circadian circuit in the morning and evening flies. To further understand the nature of differences in sensitivity, we explored their behaviour under temperature cycles with different durations of warm phase under otherwise constant darkness. We analysed various aspects of the activity/rest rhythm – phase, accuracy (day-to-day variation in phases), power of the rhythm (see Materials and Methods) and consolidation of rhythm under entrainment. We also examined period of the rhythm and its amplitude under constant darkness post-entrainment to the aforementioned temperature cycles. Subsequently, we also analysed the behaviour of these flies under LD cycles at two different constant ambient temperatures to understand the degree of waveform plasticity under different constant temperatures. Our results suggest that selection for divergent timing of emergence behaviour is also associated with increased temperature sensitivity of the clock regulating activity/rest rhythms. Interestingly, we found that this increased sensitivity is brought about predominantly by the evening bout of activity.

All experiments reported in this study made use of four sets of Drosophila melanogaster populations early_(1–4), control_(1–4) and late_(1–4) each derived in our laboratory through artificial selection for morning eclosion (early) and evening eclosion (late). Each of the four sets have independent genetic architecture and are referred to as ‘blocks’. Selection was induced in each of these sets separately and simultaneously. Every generation ∼300 eggs were collected and dispensed into vials that were then maintained in a light-, temperature- and humidity-controlled cubicle under a 12 h:12 h light:dark (LD12:12) cycle at 25±0.5°C and 65±5% relative humidity. On days 9–13 after egg collection, only flies emerging in a 4 h window starting 3 h before lights-on and ending an hour after lights-on (or ZT21–ZT01, where ZT00 is zeitgeber time 00 and refers to the time at which lights come on) were collected to form the breeding pool for the next generation of the early flies. Similarly, all the flies emerging on the same days but between ZT09 and ZT13 were collected to form the breeding pool of the next generation of late flies. For control populations, flies emerging throughout the day were collected to form the next generation. All adult flies were maintained in Plexiglas cages with culture medium in Petri dishes at a roughly 1:1 sex ratio and adult density of ∼1500 flies per cage. On day 18 after the previous egg collection, flies in the cages were provided with food supplemented with live yeast paste (as a protein supplement) and on day 21, eggs were collected in the same manner described above for the new generation. Therefore, our flies were maintained on a 21-day discrete generation cycle and all flies used in the experiments reported here had undergone over 280 (∼16 years) generations of selection. All experiments were performed on progeny of flies that experienced one generation of common rearing, to avoid maternal effects on traits being measured which may confound our interpretations (Bonduriansky and Day, 2009).

Approximately 300 eggs from each of the 12 populations were collected (as during maintenance) and dispensed into 5–10 vials and maintained under our standard stock maintenance regime. Two sets of 32 three- to five-day-old virgin males were collected, and under minimal CO₂ anaesthesia were transferred to 5 mm locomotor tubes. These sets were then recorded using the Drosophila activity monitor (DAM) system under 12 h:12 h thermophase:cryophase cycle (TC12:12; thermophase: 28°C; cryophase: 19°C) for four to five cycles before simulating the jetlag. One of these sets was given a 6 h advance phase shift, while the other set was given a 6 h delay phase shift for 10 cycles before all the flies were transferred to constant darkness at 19°C for a few cycles to judge phase control (an essential property of circadian clocks wherein phase of activity on first day in constant darkness and temperature follows from the phase on last day of temperature cycles).

In the second batch of experiments, three sets of 32 three- to five-day-old virgin male flies were collected in the same manner as described in the previous paragraph. One set each was then subjected to TC06:18, TC12:12 and TC18:06, under constant darkness for 7 days (thermophase: 28°C; cryophase: 19°C). On day 8, flies transitioned from TC to constant cryophase of 19°C for 6–8 days, to enable analyses of FRP.

In the third batch of experiments, two sets of 32 three- to five-day-old virgin males were collected as described above. One set was recorded under LD12:12 (∼70 lux during the photophase) at 19°C and the other at 28°C. Both these sets were recorded under their respective conditions for 7–8 days, before being transferred to constant darkness (DD) under their respective constant temperatures. Flies were maintained under free-running conditions for 6 days, to allow estimation of the FRP. We used FRP data of all our stocks to facilitate comparisons with the experiments reported here from a previous experiment performed by Lakshman Abhilash on entrained behaviour, which is published elsewhere (Nikhil et al., 2016b).

To estimate rates of re-entrainment to 6 h advance and delay, we marked phases of offset for each fly on each day for all 12 populations using RhythmicAlly (Abhilash and Sheeba, 2019). We then calculated the daily phase relationship as phase of offset of activity – phase of offset of thermophase. For each pre-jetlag cycle, we calculated the average phase relationship across flies. Then we computed the average inter-individual variation in among-fly phase relationships. Subsequently, we multiplied this measure by 1.96 to get a 95% confidence band around the mean inter-cycle phase relationship. We did this for all populations and then examined the dynamics of phase relationship change across days for each fly. A fly was considered re-entrained when its phase relationship re-entered the confidence band and stayed inside the band for at least two subsequent cycles. The number of cycles taken for each fly to re-entrain was used as a measure of number of transients taken for re-entrainment. These values were averaged across flies for obtaining block means. Two separate two-factor mixed model randomised block design ANOVAs were used to analyse the effect of selection on number of transients taken for re-entrainment, each for the 6 h advance regime and 6 h delay regime. Selection was used as a fixed factor and block as a random factor.

We analysed the activity profiles for all populations under TC06:18, TC12:12 and TC18:06. Raw DAM data were scanned and monitor files were saved in 20 min bins. These data files were analysed using RhythmicAlly (Abhilash and Sheeba, 2019). Individual profiles were downloaded and were re-organised to 1 h bins. Activity counts were then averaged across flies within each block to obtain population-wise profiles. As on multiple previous occasions (e.g. Nikhil et al., 2014; Srivastava et al., 2019), centre of mass (CoM) was used as a non-subjective phase marker (ψ_CoM) so that changes in the waveform under different regimes are captured reliably. Owing to the bimodality of activity profiles under TC12:12 and TC18:06, an angle doubling transformation was performed before computing ψ_CoM (Batschelet, 1981). For TC06:18, ψ_CoM was computed without the angle doubling transformation. While we understand that the activity/rest profiles of D. melanogaster are typically bimodal with the morning and evening bouts of activity being regulated by different cells in the adult brain (Helfrich-Förster, 2017), independently marking their phases is not a reliable reflection of phase. We argue that the ψ_CoM is a better phase marker because under TC cycles (i) the morning activity has a sharp masking component, and (ii) the evening activity is blunt as opposed to the sharp peaks under LD cycles (see Fig. S1). These make it difficult to identify the true peaks of activity, and hence these measures of peak phase (see Discussion) are less reliable than the ψ_CoM. FRP of flies experiencing constant conditions post-aforementioned entrainment regimes was quantified using the Chi-squared periodogram implemented in RhythmicAlly (Abhilash and Sheeba, 2019).

We used circular r as a proxy measure of normalised amplitude (due to its significance in describing the consolidation of a peak), as has been used before (Abhilash et al., 2019). Similar to the computation of phase, angle doubling was performed on activity profiles under TC12:12 and TC18:06 only. Intrinsic amplitude of each of these stocks were estimated using ActogramJ (Schmid et al., 2011). First, average actograms for each block were generated using data post-entrainment to their respective temperature cycles. Then a Chi-squared periodogram analysis was done on each population to estimate the average FRP. Subsequently, activity profiles were generated using modulo-FRP for each block. Amplitude was measured as the maximum activity count – minimum activity count in each of these profiles. To estimate accuracy (cycle-to-cycle variation in entrained phase), we calculated ψ_CoM for each fly and each cycle during entrainment. Accuracy was defined as inverse of standard deviation in ψ_CoM across cycles for each fly. These values were then averaged across flies to obtain block means.

Phases, FRP, intrinsic amplitude, amplitude and accuracy of entrainment were analysed using two-factor mixed model randomised block design ANOVAs, wherein selection was treated as a fixed factor and block as a random factor.

To examine behaviour under LD12:12 at different constant ambient temperatures, average profiles were obtained as described above. Using the population-wise profile data we calculated a ratio of total activity during the day to the total activity during the night for each population (day/night ratio). These were quantified for profiles under both temperatures. Furthermore, we were interested in asking if anything about the waveform in different temperatures changed differently across populations. For this, we used the 1 h binned activity profiles and computed difference in activity level at each time point between two temperatures. This difference was squared and the sum of these squared differences (SSD) across the entire cycle was calculated as a measure of deviance of rhythm waveform in the two temperatures. From the 1 h binned profiles, we also computed total activity in a morning window (ZT01–06) and an evening window (ZT06–11) to assess the individual contributions of morning and evening bouts of activity to potential differences in temperature sensitivity. FRP for all these flies under constant darkness at 19 and 28°C was assessed in RhythmicAlly (Abhilash and Sheeba, 2019) using the Chi-squared periodogram. Similar analyses were performed to estimate FRP of flies under DD at 25°C.

The day/night ratio, total morning, total evening activity counts and FRP post-entrainment to LD cycles under different temperatures were analysed using three-factor mixed model randomised block design ANOVAs using block means. In these ANOVAs selection and temperature were treated as fixed factors and block was treated as a random factor. The SSD values were analysed using a two-factor mixed model randomised block design ANOVA wherein selection was used as a fixed factor and block was used as a random factor.

All statistical analyses were followed by a Tukey's honestly significant difference (HSD) post hoc test, to generate error bars that facilitate easy visual hypothesis testing. All results were considered significant at α=0.05.

As a first step to understand if our early and late flies differ in sensitivity of their circadian clocks to temperature, we analysed their behaviours to simulated jetlag of 6 h phase advance and 6 h phase delay under TC12:12. From the representative actograms, one can see that when flies were subjected to a 6 h phase advance regime, all three stocks resynchronised to the phase-shifted TC cycles fairly quickly and took approximately the same time (Fig. 1A, top). However, it appears that all stocks took much longer to re-entrain to a 6 h phase delay; but the late stocks re-entrained sooner than the early and control stocks (Fig. 1A, bottom). These patterns were clearly visible when we analysed the dynamics of phase relationships for each stock across days for both phase-shifted regimes (Fig. 1B). The traces in Fig. 1B indicate average phase relationships pre- and post-simulated jetlag for all three stocks; these traces are bound on either side by a mean 95% confidence interval estimated from all replicate blocks. The error bars on the lower and upper limits of the 95% interval are standard error of the mean across four replicate blocks. This trace is provided for visual inspection of how phase relationships vary with days and were used to estimate entrainment (see Materials and Methods); the number of transients were used to statistically analyse differences in the rates of re-entrainment (Fig. 1C).

We found that all three stocks took fewer than two cycles to re-entrain to 6 h phase-advanced TC cycles and there were no among-stock differences in number of transient cycles (F_2,6=2.37, P>0.05; Fig. 1B, top and C, top). However, under the 6 h phase-delayed TC cycles, while the early stocks took about 5.3 cycles and the control stocks took ∼4.6 cycles, the late stocks only took ∼3.2 cycles to re-entrain (Fig. 1B, bottom and C, bottom). These among-stock differences were statistically significant, as was revealed by the significant main effect of selection on number of transients (F_2,6=17.53, P<0.05). Results from this experiment suggest the presence of a circadian clock circuit with enhanced temperature sensitivity in the late stocks. We conclude this, rather than attributing longer τ enabling faster re-entrainment to phase-delay regimes (Nikhil et al., 2016b), based on results from a similar jetlag experiment using light as a time cue (Nikhil et al., 2016a). Under a 9 h delay paradigm, the late stocks took significantly longer to re-entrain compared with the early stocks, thereby implying that our results indeed indicate strong temperature-sensitive components in the circadian network of the late stocks (Nikhil et al., 2016a).

Additionally, while it appears as though the early and control stocks show anticipation of the onset of thermophase in our representative actograms (Fig. 1A, bottom), this is not observed across all individuals as is clear from the averaged actograms (Fig. S2).

To further understand whether underlying differences in clock sensitivity to thermal cues contribute to plasticity in phases under different durations of warmth in warm:cold cycles (TC cycles), we examined their behaviour under three different thermoperiods (TC06:18, TC12:12 and TC18:06) under otherwise constant darkness. Across thermoperiods, most activity for both early and late stocks were restricted to the thermophase, as can be clearly seen in the actograms (Fig. 2A,B). Moreover, it appears as though under TC12:12 and TC18:06 that the late chronotypes had delayed evening activity compared with the early chronotypes [Fig. 2A (middle and right) and B (middle and right)]. Therefore, we quantified entrained phases and found that under TC06:18 the ψ_CoM was not different among stocks (F_2,6=4.20, P>0.05; Fig. 2C, top left), whereas, as expected from the actograms, the ψ_CoM of late chronotypes was significantly delayed compared with that of early chronotypes under both TC12:12 (F_2,6=7.10, P<0.05) and TC18:06 (F_2,6=5.91, P<0.05; Fig. 2C, top middle and top right). While under TC06:18 the early stocks appear to start their activity earlier in the cryophase, we argue that this is not statistically significant, based on CoM, which incorporates changes in the entire waveform to describe mean phase (Zar, 1999).

Furthermore, earlier experiments from our laboratory have reported the presence of FRP differences among the stocks under DD at 25°C (Kumar et al., 2007; Nikhil et al., 2016b). Our results, therefore, implied that entrainment to temperature cycles in these stocks can be explained within the framework of the non-parametric model of entrainment, a key prediction of which is that longer FRP is associated with delayed phase (Pittendrigh and Daan, 1976a). To test this, we first analysed FRP values of the flies that experienced the presented TC and were transferred to DD at 19°C. We found that there was no statistically significant among-stock difference in FRP when flies were transferred to constant conditions after TC06:18 (F_2,6=1.70, P>0.05; Fig. 2C, bottom left; Table 1). Interestingly, after exposure to TC12:12 the late chronotypes showed significantly longer FRP in constant conditions compared with that of the early stocks (F_2,6=5.90, P<0.05; Fig. 2C, bottom middle; Table 1). However, although there were among-stock differences in phase under TC18:06, there was no among-stock difference in FRP (F_2,6=2.10, P>0.05; Fig. 2C, bottom right; Table 1). Typically, the difference in FRP between early and late stocks after entrainment to LD and different constant ambient temperatures varies from ∼0.7 to ∼0.9 h (Fig. 5) (Kumar et al., 2007; Nikhil et al., 2016b). However, here there is no difference under short and long thermoperiods and the difference persists under TC12:12, but is greatly reduced (∼0.38 h). This, we argue, reflects differential response of FRP of the stocks to temperature as a zeitgeber (see Discussion).

Importantly, we have shown previously that similar experiments under three different photoperiods, i.e. LD06:18, LD12:12 and LD18:06, yield very different results compared with the results described above. We have found that while the late stocks show significantly delayed phase under short photoperiod relative to the early stocks, they are not different from the early stocks under LD12:12, and are significantly advanced under long photoperiod (Nikhil et al., 2016b; Abhilash and Sharma, 2020). Furthermore, post-entrainment to all three photoperiods there were significant differences in the FRP among these stocks, with the late stocks exhibiting longer τ than the early stocks (Abhilash, 2020). We have discussed the implications of these results being that the non-parametric model of entrainment is unable to account for such patterns of period and phases in our populations (Abhilash and Sharma, 2020). However, the relationships between period and phase observed under different thermoperiods are largely in agreement with predictions from the non-parametric model of entrainment.

Results from both previous sections were suggestive of among-stock differences in phase-dependent sensitivity of the circadian clock to temperature, typically characterised by a phase response curve (PRC – phase-dependent sensitivity of the clock measured in terms of phase-shifts; see Moore-Ede et al., 1982). However, there are inherent complexities of temperature pulse phase-resetting experiments and typically, only very small phase-shift values are obtained in response to long duration of pulses, as is discussed elsewhere (Chandrashekaran, 2005). Therefore, we tested predictions under the assumption of divergent temperature PRCs of the early and late stocks, without constructing the phase response curves of these stocks to temperature pulses.

Previous studies have linked divergent PRCs to differences in intrinsic amplitude of the circadian oscillator, its amplitude under entrainment, power of periodogram and accuracy of entrainment (Beersma et al., 1999; Vitaterna et al., 2006; Brown et al., 2008; Nikhil et al., 2016a). We examined these properties in flies that were exposed to TC and subsequently placed in DD, under constant temperatures. Firstly, we found that the late stocks had significantly higher amplitude of entrainment, estimated using circular r (a measure of how sharp the peak is; see also Fig. S3A), under both the asymmetric thermoperiods TC06:18 (F_2,6=20.77, P<0.05) and TC18:06 (F_2,6=9.17, P<0.05; Fig. 3A, left and right). However, there was no significant among-stock difference in the amplitude under entrainment to TC12:12 (F_2,6=4.38, P>0.05; Fig. 3A, middle). Furthermore, there was no significant among-stock difference in intrinsic amplitude under DD post-TC06:18 (F_2,6=0.67, P>0.05) or post-TC18:06 (F_2,6=1.26, P>0.05; Fig. 3B, left and right). However, the late stocks had significantly higher intrinsic amplitude, post-entrainment to TC12:12 (F_2,6=9.86, P<0.05; Fig. 3B, middle). These results imply higher amplitude expansion of the late stocks under specific TC regimes, thereby suggesting stronger temperature sensitivity in these stocks. We reasoned that periodogram power during entrainment could provide additional evidence for enhanced temperature sensitivity in the late stocks. We found that power was significantly higher for late stocks only under entrainment to TC12:12 (F_2,6=5.83, P<0.05; Fig. 3C; see also Fig. S3B). Subsequently, we analysed the accuracy of entrainment under all three regimes. We found that under short thermoperiods there was no among-stock differences in accuracy of entrainment (F_2,6=1.24, P>0.05; Fig. 3D, left; see also Fig. S3C). Under both TC12:12 and TC18:06 there was a significant main effect of selection such that the late chronotypes showed significantly increased accuracy of entrainment (TC12:12: F_2,6=9.20, P<0.05; TC18:06: F_2,6=6.46, P<0.05; Fig. 3D, middle and right; see also Fig. S3C).

In summary, we found evidence for the evolution of temperature sensitivity in the activity/rest rhythms in populations selected for divergent timing of adult emergence rhythm. Additionally, these results also suggest that most features of entrainment can be well explained within the framework of the non-parametric model described above, which makes use of predictions using the FRP and the PRC of the clock.

As selection for evening emergence contributed to enhanced phase plasticity of emergence rhythms even under LD and different constant ambient temperatures (Abhilash et al., 2019), we next analysed the activity/rest behaviour of our stocks under these regimes. Visual inspection of the activity/rest profiles of early, control and late stocks under 19°C indicated higher evening activity in the late chronotypes (Fig. 4A, left). However, profiles under 28°C looked largely similar (Fig. 4A, right), except the slightly increased morning activity in the late flies.

To quantify this, we calculated the ratio of total day-time activity to total night-time activity (day/night ratio, henceforth), and found that although there was a significant main effect of temperature regime such that there was higher day-time activity under 19°C as has been shown earlier (F_1,3=98.10, P<0.05; Majercak et al., 1999), there was no significant effect of selection×temperature regime interaction (F_2,6=0.54, P>0.05; Fig. 4B, left), thereby implying that all three stocks responded similarly to cool and warm ambient temperatures.

We then examined divergence in waveform across stocks and regimes by calculating SSD (see Materials and Methods) as a measure of extent of plasticity that early, control and late stocks show in response to different constant ambient temperatures. Although the SSD in late stocks is almost twice as much as that in early and control stocks (Fig. 4B, right), the ANOVA did not detect a main effect of selection, thereby implying that the stocks respond similarly to change in ambient temperature (F_2,3=4.69, P=0.059).

However, as the SSD difference among stocks was marginally non-significant and there were trends of among-stock differences, we quantified total activity in a morning window (ZT01–06) and an evening window (ZT06–11) for all three stocks under 19 and 28°C. We found that there was no significant effect of selection×temperature regime interaction on total morning (F_2,6=1.33, P>0.05; Fig. 4C, left) or total evening activity (F_2,6=2.56, P>0.05), but there is a clear trend of the late stocks suppressing evening activity more strongly under 28°C relative to the early and control stocks (Fig. 4C, right). These results are suggestive, although not strongly, of increased plasticity of the activity/rest waveform in the late chronotypes in response to different constant ambient temperatures under LD cycles.

While analysing features of the activity/rest rhythm under different regimes, one result that piqued our curiosity was the absolute scale on which FRP varied post-entrainment to different thermoperiods (Fig. 2C, bottom; Table 1), compared with the FRP values post-entrainment to LD cycles under different temperatures (Fig. 5; Table 1). While period values ranged from 23.1 h (early) to 23.5 h (late) and 23.3 h (early) to 23.5 h (late) post-entrainment to TC06:18 and TC18:06 (Table 1), respectively, overall period values were much lower post-entrainment to TC12:12 (Table 1). The period values ranged from 22.8 h in the early stocks to 23.2 h in the late stocks (see Fig. 2C, bottom; Table 1). Moreover, the absence of statistically significant differences in FRP between early and late stocks post-entrainment to TC06:18 and TC18:06, and the difference under TC12:12, implies stock-specific responses of FRP to cycling temperatures (Table 1).

We then examined the FRP of these stocks under DD at 19, 25 and 28°C post-entrainment to LD12:12 at these respective temperatures. We found that there was a statistically significant main effect of selection such that the late stocks had significantly longer FRP than the early stocks (F_2,6=189.70, P<0.05; Fig. 5; Table 1). Also, there was a significant main effect of temperature such that FRP lengthened with increase in temperature (F_2,6=50.00, P<0.05; Fig. 5; Table 1). Such over-compensation of FRP to changing temperatures in insects is an already established phenomenon (see Saunders, 2002). However, there was no statistically significant selection×temperature regime interaction (F_2,6=1.10, P>0.05; Fig. 5). These results indicate that temperature entrainment protocols may contribute to after-effects, despite the network being temperature compensated. This adds an additional dimension to the mechanisms of entrainment to temperature cues.

In this study we were interested in examining features of entrained activity/rest behaviour of early and late stocks under a variety of temperature cues in order to (a) test responsiveness of the circadian clocks regulating activity/rest rhythms of these stocks to temperature cues, (b) understand the mechanisms of entrainment that can account for such entrained behaviours, and (c) as a consequence, understand the similarities, if any, in the organisational principles of the circuit regulating emergence rhythms and activity/rest rhythms.

In the case of our early and late flies, as all stocks re-synchronise more quickly to 6 h phase advances (Fig. 1) compared with delays, we infer that the temperature pulse PRC of our stocks must have overall larger advance zones than delay zones. Furthermore, our results demonstrated that late stocks re-synchronised significantly faster to 6 h phase delay than early and control stocks (Fig. 1), thereby implying that the late stocks have a larger delay zone than the other stocks. This would suggest the co-evolution of high-amplitude temperature pulse PRCs of the circadian clock governing activity/rest rhythms in the late stocks in response to selection for evening adult emergence. High-amplitude PRCs also imply increased phase variation (Pittendrigh and Daan, 1976a), higher amplitude and power of rhythm (Vitaterna et al., 2006; Brown et al., 2008; Nikhil et al., 2016a) and increased accuracy of entrainment (Beersma et al., 1999).

In relation to the aforementioned predictions, we obtained curious results when we analysed phases of entrainment in our stocks under TC cycles with three different durations of the thermophase. We found that while there was no among-stock difference in phases of entrainment under short thermoperiods (Fig. 2C, top left), the late stocks showed significantly delayed phase compared with the early stocks under both TC12:12 and TC18:06 (Fig. 2C, top middle and top right). In both these cases, because the early and late stocks did not individually differ from the control stocks, it is not possible to comment upon the individual stock's contribution to phase lability under different temperature regimes. However, it is possible to still conclude that the small among-stock differences in lability, which may be present although not statistically detectable, could be due to among-stock differences in the PRCs.

Phase difference among the stocks, however, can be explained using the non-parametric model of entrainment (Pittendrigh and Daan, 1976a; Daan and Aschoff, 2001). The model posits that the difference between FRP and the period of the external environment is adjusted, during entrainment, via phase shifts due to the time cue. This response is characterised using a PRC (discussed above). Therefore, individuals with longer FRP are expected to show delayed phase under entrainment and vice versa, and this has found ample experimental evidence (Pittendrigh and Daan, 1976a; Sharma et al., 1998; Daan and Aschoff, 2001; Roenneberg et al., 2005; Wright et al., 2005; Rémi et al., 2010; Srivastava et al., 2019). This, however, is thought to occur under the assumption that the PRC is a fixed entity in all these individuals. Therefore, if individuals have divergent PRCs they will not necessarily show such a relationship between FRP and entrained phase. We found that phases under entrainment to TC06:18 were not different among stocks, and neither were the FRP of these stocks under constant conditions post-entrainment to TC06:18 (Fig. 2C). In case of entrainment to TC12:12, phases of the late stocks were delayed and their FRP was also longer under constant darkness post-entrainment (Fig. 2C). These two results are in agreement with the general rule outlined above. However, the absence of such a relationship between FRP and entrained phase under TC18:06 reveals that although entrainment is in agreement with the non-parametric model, there is compelling support in favour of the co-evolution of divergent temperature pulse PRCs in our stocks.

To further garner support for divergent PRCs in our early and late stocks, we examined other features of entrainment to TC cycles. We found evidence for increased amplitude expansion, higher power of periodogram and higher accuracy in the late stocks, all of which are indicative of high-amplitude PRCs. It is also interesting to note, at this point, that all the phase variations in the late stocks under different TC cycles is driven by the change in the evening bout of activity (see Fig. 2B). This can be clearly seen when we examined phases of morning and evening peaks of activity under all three TC cycles (Fig. 6). We found that the phase of morning peak of activity was similar across all stocks under TC06:18, and identical in all stocks under TC12:12 and 18:06 (Fig. 6, top). We could not perform statistical analyses to compare the phase of morning peak of activity across stocks owing to the lack of variance in all stocks in all regimes, except the late stocks under TC06:18. Furthermore, we found that the phase of evening activity peak in the late stocks was significantly delayed under TC12:12 and TC18:06 compared with the early stocks (Fig. 6, bottom). Importantly, while the phases of early and control stocks do not differ from each other, the late stocks are phase delayed significantly from both early and control stocks under TC18:06. This implies that plasticity in waveform in response to temperature regimes is predominantly brought about by the response of late stocks, consistent with the idea of high-amplitude temperature PRCs in these populations. Our results of changes only to the evening peak of activity suggest that the effect of temperature acts on period of the molecular clockwork regulating only the evening activity. While an interesting possibility, this must await further experimentation.

Our results indicate that the late chronotypes have stronger temperature responsiveness compared with the early chronotypes, and this is predominantly brought about by the behaviour of the evening bout of activity; Hall (2003) has previously reported similar enhanced response of evening activity bout to temperature. In an earlier study, we have shown that temperature responsiveness of the eclosion rhythm of our late chronotypes is enhanced relative to the early chronotypes (Abhilash et al., 2019). This enhanced temperature sensitivity for the eclosion rhythm, we discuss, is perhaps due to the temperature-sensitive oscillator (PG clock) regulating the eclosion rhythm. However, we know that the evening bout of activity is regulated by different cells (E-cells or evening oscillators) in the adult brain, i.e. the LNds (lateral dorsal neurons) and DN1s (dorsal neurons). Interestingly, we also know that the DN1s are temperature sensitive (Stoleru et al., 2004; Murad et al., 2007; Stoleru et al., 2007; Zhang et al., 2010; Gentile et al., 2013; Yadlapalli et al., 2018). Therefore, our results indicate that while enhanced temperature responsiveness of both eclosion and activity/rest rhythms evolve in the late chronotypes, they perhaps do so using distinct physiological mechanisms. Typically, the eclosion rhythm is thought to be regulated by a hierarchically arranged set of two oscillators – one light-sensitive master clock that drives a temperature-sensitive slave oscillator (Oda and Friesen, 2011; Pittendrigh, 1974). The activity/rest rhythm, however, is described to be governed by a mutually coupled oscillator system, in which historically, neither component is described to be sensitive to temperature (Pittendrigh and Daan, 1976b; Helfrich-Förster, 2009). Interestingly, the evolution of enhanced temperature-sensitive components of the circadian clock network in the late stocks for both these rhythms suggests remarkable conservation in organisation principles regulating the two rhythms with respect to the hierarchical system of organisation. Furthermore, our results imply (i) a previously unrecognised role of the hierarchical model of organisation in regulating activity/rest rhythms, and (ii) potential overlap between temperature-sensitive oscillators and the evening oscillators.

Plasticity of FRP under different temperature regimes pose an extremely interesting conundrum. In this regard, two key results are important to note: (i) FRP post-entrainment to different TC cycles are shorter by >1 h relative to the FRP post-entrainment to LD cycles under different constant temperatures (Figs 2C, bottom and 5; Table 1), and (ii) FRP post-entrainment to short and long thermoperiods behave similarly but different from FRP post-entrainment to TC12:12 (Fig. 2C, bottom; Table 1). Temperature compensation of the FRP is a crucial prerequisite for calling an oscillatory physiological process a circadian clock (Moore-Ede et al., 1982; Saunders, 2002; Dunlap et al., 2004). Such compensatory mechanisms implied that in order for entrainment to occur in response to temperature time cues, only phase shifts must occur (as predicted by the non-parametric model) and not period changes (as predicted by the parametric model, which posits that the zeitgeber's effect is integrated over the cycle to constantly modulate the angular velocity of the clock, and therefore allow entrainment). However, we find that despite this being the case, FRP post-entrainment to different thermoperiods varies (Fig. 2C, bottom). While the period value averaged over all stocks post-entrainment to TC12:12 is ∼22.98 h, FRP post-entrainment to TC06:18 and TC18:06 lengthens and is ∼23.32 and ∼23.43 h, respectively (Table 1). If these responses were due to compensatory mechanisms, one would expect opposite effects on FRP post-entrainment to short and long thermoperiods. Therefore, we think that these reflect some form of after-effect due to entrainment to TC cycles. Additional support for temperature-dependent after-effects also comes from the result that FRP values shorten after being under the influence of different TC cycles relative to values after being under the influence of LD cycles and different constant ambient temperatures. Importantly, we find that late stocks show significantly longer FRP post-entrainment to TC12:12, while there is no among-stock difference under the two other thermoperiods (Fig. 2C, bottom; Table 1). This suggests, although weakly, that the FRP of late stocks are less likely to change in response to temperature cycles relative to early stocks. While there have been many reports of after-effects of light regime on FRP (Dunlap et al., 2004), there are very few on after-effects of temperature cycles (Balzer and Hardeland, 1988). Our results, to the best of our knowledge, provide first hints of temperature after-effects on FRP in Drosophila activity/rest rhythms, implying that temperature may also contribute to entrainment via parametric means, warranting further, more detailed documentation of effects of temperature cycles on FRP.

In conclusion, we find enhanced temperature sensitivity of the activity/rest rhythm in late chronotypes. Interestingly, altered rhythm phase under different temperature regimes of the late stocks is driven by changes in the evening bout of activity. Furthermore, analyses of properties of entrainment and FRP implied that results can be explained under the assumption of the evolution of divergent temperature pulse PRCs, a matter worthy of further study.

We are very grateful to the late Professor Vijay Kumar Sharma (V.K.S.) for providing us with the tools that enabled us to carry out this research, and for a wonderful work environment that enabled us to undertake this study. We thank Srishti Priya and Arijit Ghosh for help with experiments, and the two reviewers for reading a previous version of our manuscript and providing some very valuable suggestions.