Introduction

Intuitively, an organism could optimize its behaviour and physiology by responding to daily and seasonal changes in the environment. Yet virtually all organisms from Cyanobacteria to humans have an internal circadian clock that allows them to anticipate daily environmental changes and to alter their behaviour and physiology accordingly.

The roles of these internal clocks in our lives can perhaps most clearly be understood by seeing what happens when our clocks become desynchronized from the environment. In Major League Baseball, the effect of jet lag on West Coast teams that travel to the East Coast (but not vice versa) increases the chance of East Coast teams winning home games. This effect decreases as the visiting team acclimatizes during the course of a three- or four-game series. This small, yet statistically significant effect, as recorded from 1991–1993, may even have accounted for the Atlanta Braves winning their division by one game from their West Coast rivals in 1991 and 1993 [2].

In addition to controlling the timing of sleep/wake cycles and thus influencing alertness, circadian clocks in mammals have been shown to control rates of drug detoxification, bone growth, liver regeneration and cell division [3–5]. Circadian rhythm disruptions can lead to depression, obesity and higher incidences of cancer [6–10] and have even been implicated in the sensitivity of an organism to drugs of abuse [11–13]. In other words, circadian rhythms are very important for the normal well-being of an animal because they enable an organism to anticipate and respond to environmental changes before they happen. In contrast, reindeer that live in the Arctic abandon daily activity rhythms during summers of constant light and winters of constant darkness since an endogenous circadian clock is apparently unnecessary without daily environmental changes to anticipate [14].

The beginnings of circadian research

The existence of an endogenous clock was first reported by French geophysicist Jean-Jacques d’Ortous de Mairan in 1729 [15]. Having observed a daily cycle of leaf opening and closing in heliotrope plants, he asked what would happen to this rhythm in the absence of environmental cues by moving the plants to his dark wine cellar. He found that leaves continued to show daily cycles of opening and closing even in constant darkness (DD), indicating the existence of an internal clock. Since these rhythms run with an approximately 24 h repeating period under constant conditions, they are termed circadian—‘about a day’.

Many of the pioneering experiments on circadian rhythms were performed in Drosophila. In one classic paper [16], Colin Pittendrigh demonstrated that the clock that drives rhythms in eclosion (hatching of adults from their pupal case) free-runs with a circadian rhythm in constant darkness, can be reset by light delivered during darkness and runs with an ∼24 h period over a 10°C temperature range—a phenomenon known as temperature compensation. This last feature of molecular clocks is important not just for cold-blooded animals like Drosophila, but also for mammals given the daily fluctuations in body temperature, and especially for hibernating animals. It was the identification of the first clock gene mutants by Konopka and Benzer [17] in Drosophila that opened a door which ultimately led to a detailed molecular understanding of how intracellular clocks tick and how they are reset by light. However, the mechanisms underlying temperature compensation remain mysterious. More recently, this molecular understanding has been used to move towards an understanding of the neural circuits driving circadian behaviour. So what have we learnt from Drosophila?

Circadian rhythms in Drosophila

When Konopka and Benzer [17] initially screened mutant flies for disrupted circadian rhythms, others were sceptical that mutating a single gene would have much effect on a complex behaviour. But Konopka and Benzer succeeded in identifying the first three clock mutants, all in the period (per) gene: a null allele (per 01) that led to a complete loss of rhythmic behaviour, and two alleles that left rhythms intact but gave flies either short-period (19 h) or long-period (27 h) rhythms in DD.

The ensuing 35 years of circadian biology have made circadian rhythms the best understood behaviour at the molecular level. A number of factors have probably contributed to this, including: (1) Circadian rhythms are not essential in the lab, although they presumably do contribute to survival in the wild [18, 19]. (2) The first four clock genes identified are non-essential genes with few roles outside the clock. Indeed, they seem dedicated to rhythmic processes, as their other reported effects include altered courtship song rhythms and recovery from sleep deprivation [20, 21]. This can be contrasted with the first two learning and memory genes identified in Drosophila, dunce and rutabaga, which are essential genes with general roles in cyclic AMP signal transduction [reviewed in 22]. (3) Circadian rhythms are robust and easily quantifiable at the single fly level. This allowed screens of single F1 progeny [23] rather than having to generate lines of flies—and the same approach was successful in identifying clock mutations in mice [24]. (4) Functional clocks in Drosophila are found not only in the pacemaker cells in the brain that drive behavioural rhythms but also in peripheral oscillators in sensory neurons (e.g. photoreceptor cells in the eye and olfactory neurons in antennae) that probably drive daily rhythms in sensory sensitivity [25]. The number of photoreceptor cells per fly make the eye clock ideal for quantitative analysis of RNA and protein level oscillations and circadian regulation of post-translational modifications such as phosphorylation. One screen for clock mutants even used a per-luciferase reporter gene with the signal coming largely from the eye to identify mutants that no longer rhythmically produced luciferase [26]. (5) Finally, the absence of many of the core clock components in Drosophila cell lines has allowed the reconstitution of certain aspects of the clock in vitro (see later).

The techniques available to a field of research influence its direction, and the relative ease of biochemically studying the clock means that we now have a good molecular understanding of clock genes. As genetic tools have become available that make it possible to study and manipulate individual neurons, circadian researchers are now able to explore how individual clock cells communicate, and how the clock translates environmental inputs to modulate behavioural outputs. In this review, we highlight some recent findings, focussing first on the molecular clock and then moving towards a more neurobiological understanding of circadian behaviour.

The Drosophila molecular clock

All circadian clocks characterized to date are based on negative feedback loops. In the current model of the Drosophila molecular clock [reviewed more extensively in 27], the basic helix-loop-helix transcription factors CLOCK (CLK) and CYCLE (CYC) activate expression of per and timeless (tim, see Fig. 1). PER requires TIM for stabilization, but TIM is light sensitive, so PER and TIM levels do not rise until dark. PER/TIM dimers then accumulate in the cytoplasm, and enter the nucleus ∼6 h after lights off. PER and TIM dissociate inside the nucleus, and PER represses transcription by preventing the CLK/CYC dimer from binding the per and tim promoters [28, 29]. The length of the delay before nuclear entry of PER/TIM is in part dependent on the activity of the Glycogen Synthase Kinase ortholog SHAGGY (SGG) [30]. SGG overexpression leads to hyperphosphorylated TIM and the PER/TIM heterodimer translocates into the nucleus ∼4 h early, shortening the period as CLK/CYC activity is inhibited prematurely [30]. Many other factors contribute to this loop, including DOUBLE-TIME (DBT), a Casein Kinase Iɛ homologue [23], Casein Kinase II [31, 32], Protein Phosphatase 2A (PP2A) [33], the F-box protein SLIMB [34, 35], and the blue light photoreceptor CRYPTOCHROME (CRY), that also doubles as a transcriptional repressor in some clock cells [26, 36, 37]. Attention has largely focussed on how these proteins regulate the stability of PER and TIM, although DBT and PP2A also regulate CLK activity [38], with hypophosphorylated CLK associated with maximal per transcription [29, 38].

Fig. 1
figure 1

The two loops of the Drosophila molecular clock. In the first loop, CLK and CYC directly activate transcription of per and tim. CRY mediates light-dependent degradation of TIM and is partly responsible for preventing the accumulation of PER/TIM heterodimers during the day. At night, PER and TIM accumulate in the cytoplasm and enter the nucleus, where PER inhibits CLK/CYC activity. The timing of nuclear entry and the stability of PER are controlled by SGG phosphorylation of TIM and DBT phosphorylation of PER. In the second loop, PDP1ɛ acts as an activator and VRI a repressor of Clk transcription. As Pdp1ɛ and vri are direct targets of CLK/CYC, this creates a second feedback loop

Full size image

A second, interlocked loop regulates rhythmic expression of Clk and probably of cry. Here, CLK/CYC activates transcription of vrille (vri) and PAR-domain protein 1ɛ(Pdp1ɛ), which encode related basic leucine zipper transcription factors [39–41]. VRI and PDP1ɛ regulate Clk transcription, with VRI acting as a repressor and PDP1ɛ an activator of transcription, causing Clk RNA levels to cycle in opposite phase to per, tim, vri and Pdp1ɛ [39, 40]. vri RNA and protein levels peak ∼3–6 h before those of Pdp1ɛ, presumably underpinning the daily rhythm in Clk expression [40].

A two-loop clock is found in flies, mammals and Arabidopsis, suggesting that it is an optimal system for accurate, temperature-compensated 24 h timekeeping. Although rhythms can be achieved with much less (see below), this complexity may be required so that the clock can be fine-tuned to the environment, and to facilitate the regulation of outputs at different times of day by altering the transcriptional properties of the cell through daily oscillations in transcription factor activity (e.g. CLK/CYC and VRI/PDP1ɛ).

Reconstructing the clock

A recent experiment in Cyanobacteria found that temperature-compensated 24 h rhythms in KaiC phosphorylation were observable for at least three days when purified KaiA, B and C proteins were mixed together in a test tube. Thus, just three Cyanobacteria clock proteins are sufficient for generating accurate rhythms, and rhythmic transcriptional regulation is unnecessary [42].

Similarly, PER and TIM transfected into Drosophila S2 cells replicate the long delay between their cytoplasmic accumulation and nuclear entry as observed in vivo [43]. Fluorescence resonance energy transfer (FRET) levels were measured between CFP-labelled PER and YFP-labelled TIM proteins to reveal the dynamics of their interaction and accumulation. Maximum levels of FRET were observed from ∼30 min after transfection, demonstrating that PER and TIM interact almost immediately. PER and TIM accumulated in speckled foci in the cytoplasm, where they remained together for ∼5 h. Importantly, using Konopka and Benzer’s original long-period PER mutant extended the delay in nuclear entry of PER and TIM, as seen in vivo [44]. Thus, PER and TIM dynamics generate an accurate ‘interval timer’ even in naive S2 cells, a delay which presumably contributes to the 24 h molecular clock.

Interestingly, high levels of FRET disappeared just before nuclear entry, and the onset of PER nuclear accumulation often preceded TIM. Thus, PER and TIM may not enter the nucleus together, as also suggested by in vivo studies where PER was detected in the nucleus before TIM [45]. Foci of transfected CRY protein were previously observed in the cytoplasm of S2 cells, and these aggregations were dependent on PER, TIM and light [46]. Care has to be taken with interpreting data from S2 cells, as equivalent foci have not been identified in vivo and also since there are contradictory conclusions in the literature between the findings from S2 cells and arrhythmic pacemaker neurons [47, 48]. Nevertheless, S2 cells seem to be useful for studying specific portions of the molecular clock, and the ability to mimic the dynamics of PER/TIM nuclear entry in S2 cells should help understand the biochemistry of this interval timer.

Sensory inputs to the clock: light

The primary input to the circadian clock is light, with flies active during the day and asleep at night. There are two ways by which Drosophila clock cells receive light information—through the endogenous blue light photoreceptor, CRY, and, for cells with appropriate neural connections, from the eye: Projections from the eye photoreceptor cells contact the l-LNvs, whereas the Hofbauer–Buchner eyelet contacts the s- and l-LNvs [reviewed in 49]. The clocks of mutant flies lacking photoreceptors and CRY are not entrainable by light [50].

The best characterized effect of light on the Drosophila clock is on TIM degradation, and this rapid response enables the molecular clock to respond to daily and seasonal changes in light. At the behavioural level, a light pulse during darkness delays or advances the timing of onset of activity on the next day, depending on the timing of the light pulse. At the molecular level, a light pulse in the early evening degrades cytoplasmic TIM, which delays PER accumulation and thus, progression of the molecular clock. Consequently, the timing of activity on the next day is also delayed. Conversely, a light pulse late at night degrades nuclear TIM, freeing PER to repress CLK/CYC activity earlier than normal, advancing timing of the onset of activity on the next day.

The importance of light as a clock input is underlined by the effect of prolonged light. Whereas flies show molecular and behavioural rhythms in constant darkness, they become arrhythmic in constant light (LL). Strangely, Drosophila cry b mutants show both molecular and behavioural rhythms in LL. This is surprising given that cry b flies can still detect light via the visual system and can entrain to LD cycles [51–53], but is probably partly due to increased TIM stability in cry b mutants in some clock neurons.

CRY-dependent TIM degradation involves the recently identified F-box protein JETLAG (JET) [54]. Like cry b mutants, jet mutants are rhythmic in LL, have normal DD behaviour and show reduced responses to light pulses, suggesting a defect in the light input pathway [54]. This effect is dependent on the genetic background, as rhythmic behaviour is only observed in jet c mutants when they have one of two naturally occurring tim alleles that differ by 23 AA at the N-terminus [55, 56]. The functional difference between these two TIM isoforms remains to be determined.

Sensory inputs to the clock: temperature

Quite sensibly, Drosophila have a mid-day siesta and avoid activity during the hottest part of the day. At colder temperatures, this siesta is reduced as the evening activity peak moves earlier in the day. This response is controlled by the splicing of an intron within the 3′ UTR of per [57]. Regulation of per splicing integrates seasonal information, as it responds to both temperature and light [57–59]. At low temperatures, or under the short photoperiods associated with colder days, per splicing levels are increased at least in photoreceptor cells where this phenomenon has largely been studied. This leads to earlier processing of the per transcript and earlier accumulation of PER protein, resulting in an earlier phase of evening activity, and so reducing the mid-day siesta. This allows the behaviour of a fly to be fine-tuned to any given day across the seasons [57–59]. Rescue of per 01 mutants with a per transgene where the intron cannot be spliced fails to rescue this adaptive response [57]. This suggests that per splicing is regulated in the same way in pacemaker neurons as in the eye, and that regulated per splicing is responsible for the shift in mid-day siesta in response to temperature changes.

Information regulating per splicing is received through a signalling pathway involving the NorpA Phospholipase-C [58, 59]. Furthermore, NorpA has subsequently been shown to be a general factor involved in temperature entrainment of behaviour, as NorpA mutants fail to entrain behavioural rhythms to temperature cycles in LL. NorpA likely acts as part of a signalling cascade relaying temperature information to the clock, possibly though PER [56]. A second, as yet un-cloned mutation, nocte, also specifically abolishes temperature entrainment, whilst leaving light entrainment intact [60].

Temperature entrainment of the clock and temperature compensation appear to be independent of one another. Neither the regulation of per splicing nor the temperature entrainment roles of NorpA and nocte seem to contribute to temperature compensation [58, 60]. Instead, naturally occurring polymorphisms in a Threonine–Glycine (Thr-Gly) repeat region in the per gene may provide a clue. 99% percent of wild-type Drosophila strains have a per gene encoding 14, 17, 20 or 23 Thr-Gly repeats [61]. These alleles affect temperature compensation: (Thr-Gly)20 display a slightly short ∼23.7 h period that remains constant over a wide range of temperatures [62]. (Thr-Gly)17 has a period closer to 24 h than (Thr-Gly)20 at higher temperatures, but overall, is less well temperature compensated. This difference in temperature compensation explains the highly significant cline in (Thr-Gly)17 and (Thr-Gly)20 allele distribution in wild populations through Europe [61]. (Thr-Gly)20 is more common in the North where temperatures are more variable, whereas (Thr-Gly)17 is more common in Southern Europe where temperatures are warmer and therefore an allele giving a 24 h period at higher temperatures is ideal [62]. Several long- and short-period mutations of per and tim also affect temperature compensation [63–66], although this may simply reflect that these alleles generate temperature-sensitive proteins rather than affecting parts of the temperature compensation mechanism per se.

Social inputs to the Drosophila clock

Although Drosophila is not generally considered a social animal, its circadian clock is influenced by social signals, as are the clocks of bees, rodents, fish and humans. Flies pre-housed together show greater synchrony in behaviour when allowed to free run individually than those pre-housed apart [67]. Adding arrhythmic per 01 mutants to a group of rhythmic flies reduces (but does not abolish) synchronization. Furthermore, adding a small number of ‘visitor’ flies with an earlier phase of activity than their hosts advances the phase of the hosts’ activity, demonstrating that flies communicate circadian information. These social signals are airborne, as pumping air from a vial containing flies in LD to flies in DD helped synchronize flies individually housed in DD. Furthermore, mutants lacking normal olfactory responses are unaffected by per 01 visitors.

One important result was that per 01 visitor flies do not disrupt synchronization of flies in which per expression is limited to a subset of central brain neurons, the Lateral Neurons (LNs, see below). One interpretation of this result is that social entrainment is mediated via the peripheral clock in the antenna. However, another possibility is that social entrainment is mediated via non-LN central brain pacemaker clocks such as the Dorsal Neurons (DNs, see below). The identification of the relevant neural and molecular substrates for social entrainment of the clock will be very exciting.

Functional groups of clock neurons

Clock protein oscillations can be detected in six major groups of neurons in the Drosophila brain which have been anatomically divided into Lateral and Dorsal Neurons (LNs and DNs; see Fig. 2). The LNs are subdivided into large and small ventral lateral neurons (l- and s-LNvs), which produce the neuropeptide Pigment Dispersing Factor (PDF), and dorsal lateral neurons (LNds). DNs are subdivided into three groups (DN1–3), although it was recently found that DN1s are a heterogeneous group consisting of 2 DN1a anterior neurons that produce the neuropeptide IPNamide, and the more posterior DN1s whose transmitter(s) remain unknown [68]. The clocks in all of these neurons oscillate in wild-type flies in DD; even the l-LNvs, while initially arrhythmic, regain rhythms after several days [69]. Overall, clock gene oscillations in ∼100 neurons control rhythmic behaviour of the whole organism [70]—but what are the roles of these individual groups of clock neurons?

Fig. 2
figure 2

Clock neurons in the Drosophila brain. One lobe of an adult per 01 mutant brain stained for PDF (blue), PDP1ɛ (green) and TIM (red). Levels of PDP1ɛ and TIM are high in all six groups of clock cells as a result of the per 01 mutation. Inset on right shows a closer view of LNs stained for PDP1ɛ (green) and TIM (red)

Full size image

Three initial studies suggested the importance of the s-LNvs in driving behavioural rhythms: (1) the presence of just one s-LNv in a disconnected mutant is sufficient for a fly to display some rhythms, while flies with no s-LNvs remaining are arrhythmic [71]; (2) cry b mutants are rhythmic in DD, with the s-LNvs the only cells to show rhythms in TIM protein levels, at least in LD cycles [26]; and (3) PDF is only produced in LNvs, and Pdf null mutants become arrhythmic after 1–2 days in DD [72]—thus LNvs are important for rhythms. However, the persistence of rhythms for the first 2 days in DD in Pdf mutants and in flies in which PDF cells have been ablated indicates that other cells are also required to drive normal behavioural rhythms.

Morning and evening peaks are controlled by separate clock neuron groups

Although flies can sustain behavioural rhythms for weeks in DD, they normally live in a constantly changing environment. A long-standing prediction of Pittendrigh and Dann [73] was that separate morning and evening oscillators would help an organism adapt to seasonal changes. In an LD cycle, Drosophila activity peaks in the morning and evening, anticipating lights on and off. Pdf 01 mutants or flies with PDF cells ablated are rhythmic in LD, but they only anticipate dusk (the “evening peak”) not dawn (the “morning peak”) [72]. Thus, the LNvs are a good candidate for the Morning oscillator (M) cells.

To identify the Evening (E) cells, locomotor activity rhythms were tested in flies which had subsets of clock neurons ablated [74] or had per restored in an otherwise per 01 mutant background [74, 75]. The presence or absence of a morning and/or evening peak was then correlated with the presence or absence of a functional clock within each neuronal group. This is summarized in Fig. 3. These approaches confirmed that the PDF-expressing LNvs control the morning peak, as any genotype lacking clock function in these cells shows no anticipatory morning peak (with one exception, see below). Furthermore, clock function in the s-LNvs alone is necessary and sufficient for the morning peak, making the s-LNvs the M cells [74, 75].

Fig. 3
figure 3

Morning/Main and Evening cells. Presence (green) or absence (red) of a functional clock in each group of clock cells in different genetic backgrounds. Cells in orange retain some clock function. “?” represents unknown clock function

Full size image

Conversely, the LNds, a PDF-negative LNv, and possibly a small number of DN1s contribute to the evening peak (E cells). The neurotransmitter(s) controlling the evening peak is unknown, but signalling by the neuropeptide IPNamide, recently shown to be expressed in a subclass of DN1 neurons that project to the s-LNvs [68], may be involved in this process.

Under constant conditions, the function of the M and E cells is slightly different. Pittendrigh and Dann’s [73] model predicted that in constant light, the M oscillator will free run with a short period and the E oscillator a long period. cry b flies show a weak morning and a strong evening activity peak in LL [53]. After a few days, the evening peak splits into a long (25.2 h) and a short (22.5 h) period component; the short component has a similar period to the weak morning activity peak. A single LNd and the 5th PDF-negative s-LNv are the only neurons with a long period molecular clock under these conditions, making them strong candidates for the E oscillator neurons. Similarly, the s-LNvs display a short period, fitting their role as the M cells. The splitting of the evening peak in LL into both a short and long component suggests that the M cells also contribute to the evening peak of activity, an addition to the original model, suggesting that these cells should be named ‘Main’ rather than ‘Morning’ neurons [53].

The relationship between M and E cells was probed further by over-expressing sgg in the LNvs, shortening the period by ∼3 h in DD [74]. This advanced tim RNA accumulation in both the s-LNvs and E cells and the timing of both morning and evening peaks. Over-expression of sgg in the E cells advanced the timing of tim accumulation only in the DN2s and l-LNvs, and not in the LNds or 5th LNv. Presumably, a signal from the s-LNvs overrides the endogenous period of the E cells. This confirms that the s-LNvs are the ‘Main’ oscillator, necessary and sufficient for rhythmic activity in DD and for determining the period of DD rhythms [76]. But if the s-LNvs contribute to both morning and evening peaks, what is the function of the E cells?

The LNds and possibly the 5th s-LNv also show robust oscillations in DD, but are unable to drive rhythmic locomotor activity independent of the s-LNvs. To determine the function of the E cells in DD, the period of the E cells was accelerated by over-expression of sgg. In contrast to expression in the M cells, this did not affect the free-running period. However, it did reduce the time between morning and evening peaks by ∼2 h; expression in M cells alone had no effect on this interval [76]. Thus, the M cells set the overall period of the system and presumably send a daily resetting signal to the E cells, which then drive an evening peak in response to this signal. The time between the signal from the M cells and the appearance of an evening peak is dependent on the period of the E cells’ clock: If it runs fast, then the evening peak is earlier [76]. In this way, changes in temperature or day length that tend to affect the evening and not the morning peak [77] can be accommodated by the clock without disrupting the underlying 24 h period. As the LNds’ clock is advanced in Pdf 01 mutants, this suggests that PDF signals normally maintain phasing and amplitude of LNd rhythms perhaps by delaying the timing of PER nuclear entry [78]. Thus, PDF may be the daily resetting signal from M to E cells.

These experiments, together with the lack of anticipatory peaks in clock-gene mutants such as per 01, suggest that anticipatory activity in LD is clock-dependent. However, this is confounded by the observation of anticipatory evening activity in per 01 ; cry b double mutant flies that should lack a functional clock in all clock cells [79]. Anticipation is TIM-dependent, as both per 01 ; tim 01 ; cry b triple mutants and tim 01 ; cry b double mutants showed no anticipation. per 01 ; cry b mutants are arrhythmic in DD and LL, so rhythmicity is dependent on an LD cycle. One possibility is that the cry b mutation prevents the immediate degradation of TIM in response to light. The resultant light-driven oscillation in TIM then restores enough clock function (to the E cells?) to generate rhythmic evening activity in LD. However, TIM does not enter the nucleus of either the M or E cells in per 01 ; cry b mutants, making it unclear exactly how much clock function is sufficient to drive LD rhythmicity [79].

Clock neural networks

In order for different groups of clock neurons to control a single behaviour, they need to form a single network. Although the PDF-expressing LNs can be considered the ‘Main’ oscillator cells, they need the support of the rest of the clock network to control locomotor behaviour.

There are several pieces of evidence that support a model of networked clock neurons. Firstly, although Pdf null mutants become arrhythmic in DD, rhythms of PER oscillation persist in pacemaker LNs [72, 78]. However, the timing of PER nuclear entry between individual Pdf mutant s-LNvs becomes desynchronized in DD [78]. The importance of PDF for synchronization is likely reduced when individual neurons can receive light via CRY, hence the relatively normal LD behaviour of Pdf 01 mutants. PDF could normally keep the s-LNvs synchronized either by direct signalling or via the whole clock circuit. A circuit explanation seems more likely because the s-LNvs do not produce the PDF Receptor (PDFR) [80–82]. Thus, a PDF signal from the s-LNvs is relayed back to the other s-LNvs via at least one other clock neuronal group.

Secondly, although the morning and evening oscillators function independently, the LNds, DN1s and the PDF-negative LNv can drive normal LD rhythms (even including the morning peak) in per 01 flies where per expression is restored everywhere except the PDF cells [75]. The restoration of the morning peak suggests that rhythmic activity of E cells is sufficient to drive the appropriately timed release of PDF from M cells. However, when the PDF expressing cells are ablated or do not produce PDF, no anticipatory morning peak is observed [72].

Finally, the ability of a functional clock in PDF neurons to drive rhythmic behaviour depends on the clock state of the rest of the network. PDF-cell specific expression of per in a per 01 mutant background rescued both molecular and behavioural rhythms [75]. Similarly, molecular rhythms in only LNvs are sufficient for behavioural rhythms in cry b mutants [26]. Conversely, PDF-cell specific expression of cyc in a cyc mutant background rescued molecular oscillations of tim RNA in the s- and l-LNvs, but failed to restore behavioural rhythms of flies [69]. How can these results be reconciled? One idea is that the status of the clock in the remainder of the clock neuron circuit (see below) may or may not render them permissive to respond to rhythmic outputs of the PDF cells. Certainly the molecular functions of PER and CRY as transcriptional repressors are different enough from CYC (a transcriptional activator) to leave the basal states of mutant cells very different. What this means at the molecular and electrophysiological level remains to be explored.

In summary, the ability of a fly to remain highly rhythmic for weeks in DD is probably the result of a network of clock neurons keeping each other synchronized rather than the highly accurate cell autonomous timekeeping of individual clocks in neurons.

Outputs: neurotransmitters and membrane excitability

In contrast to the relatively detailed understanding of the molecular clock and the beginnings of an understanding about the roles of the different clock neurons, very little is known about the step in between: how a clock is coupled to neuronal outputs. There are two specific questions here: What are the neurotransmitters released by individual clock neuron groups? And how does an intracellular clock control neuronal activity to convey a circadian message?

Of the six main groups of clock neurons in the central brain, we know embarrassingly little about their neurotransmitters: We know only that PDF is produced in the LNs, and IPNamide in a subset of DN1s. Mutations that eliminate PDF production phenocopy ablation of LNvs, suggesting that PDF is the major signal from LNvs to control locomotor behaviour. However, neuropeptides are often co-produced in neurons with a more classical small molecule neurotransmitter, and these remain to be identified in clock neurons. Certainly, additional signalling molecules must be present to explain how LNvs control light avoidance in larvae and cocaine sensitivity in adults independently of PDF [11, 83].

How a molecular clock controls rhythmic neuronal activity (as would be predicted by analogy with the mammalian suprachiasmatic nucleus [84]) is presumably tied to its rhythmic transcription. Indeed the transcriptional state of the s-LNvs seems to determine their output levels, at least in one simple system: light avoidance by larvae.

Drosophila larvae are intrinsically photophobic: When wild-type larvae are aligned down the middle of a half-covered Petri dish, ∼70% of larvae will be on the dark side after 15 min [83]. Ablation of the larval visual system, Bolwig’s Organ (BO), causes larvae to distribute randomly between light and dark (‘blind’). BO projects to the larval LNvs, and ablation of LNvs also causes larvae to distribute randomly, indicating that the LNvs are necessary for light avoidance, and suggesting that they transmit a signal they receive from BO. Light avoidance is under circadian control: Wild-type larvae are most sensitive to light at dawn and least sensitive at dusk, and this is clock gene-dependent, since per01 and tim01 mutants (high CLK/CYC activity) are constitutively blind, whereas ClkJrk and cyc0 mutants (low CLK/CYC activity) are constantly highly photophobic. Clock gene modulation is dependent on the LNvs, as the larval photoreceptor cells that make up BO have no clock [83], in contrast to adult photoreceptors.

The blindness of per and tim mutants can be rescued by increasing the light intensity—presumably increasing the amount of transmitter released by BO—whereas LNv-ablated larvae remain blind even under bright light conditions. Conversely, Clk and cyc mutant larvae are still photophobic at low light levels where wild-type larvae are blind. Thus, clock modulation of larval light avoidance is dependent on the transcriptional state of the LNvs, consistent with the excitability of LNvs being under clock control. The time of day at which larval LNvs are most sensitive to light (dawn) corresponds well with their presumed activity as adult morning cells. Pupal LNvs drive the daily rhythm in eclosion, and this also peaks at dawn, suggesting that morning behaviour is under the control of the same intracellular mechanisms in the s-LNvs throughout a fly’s lifetime. This could be achieved by circadianly regulated transcription of genes that alter the neuronal activity of the cell and/or the production of the appropriate output signal. Such genes could include ion channels and their modulators and/or enzymes involved in transmitter production.

The M and E neurons control behavioural outputs at opposite times of day. This could mean that they fire in antiphase—and this would occur despite the oscillations of clock proteins appearing broadly similar between these cells. To explain this, we can borrow an analogy from mammals, which have clocks in many tissues outside the nervous system including the liver and the heart. Although the molecular clocks in these tissues are similarly organized, they are coupled to rhythmic expression of genes with tissue-specific functions (e.g. Alcohol dehydrogenase in the liver and Fibrillin-1 in the heart [85]). Perhaps the output pathways downstream of the molecular clocks in different clock neurons in flies are also very different, enabling them to be active at different times of day.

Finally, there is increasing evidence that clocks are not just housed within neurons, but that the neuron itself is a part of the clock [86, 87]. Electrically silencing the LNs stops their free-running rhythms—thus membrane electrical activity is coupled to the molecular clock. Again, this remains to be understood at the molecular level. A circuit explanation could be invoked here too, as the molecular clock in electrically silenced LNvs runs down in DD rather than stopping immediately.

Conclusion

Researchers studying circadian rhythms in Drosophila are leaving their molecular biology roots and are moving towards a more holistic understanding of how flies keep accurate time. There are clearly many unanswered questions that will occupy the field for many years, some of which have been mentioned here. Presentations at recent meetings highlight this trend, with researchers reporting the effects of more natural lighting conditions on the clock (including moonlight) and trying to tie together temperature and light changes over 24 h. Ultimately, we should have a good understanding of how flies keep in internal harmony with their continuously changing external environment.