Cognitive ecology is ultimately interested in understanding the constraints applied by perceptual bias and their consequences in the natural world. At the heart of perception lies an animal's ability to recognize change and to make predictions based on the way change has played itself out in the past. Critical to this behavior is the ability to measure time. Without suggesting that it is the basis of predictive science, the perception of time is the substance of causal understandings.
Animal event timing refers to the process that an animal undergoes in order to `recognize' an interval of time. In a very anthropomorphic way, one can ask the question this way: How does an animal know the difference between 2 minutes and 2 hours? Or does it even know at all? Absolute time may be an artificial consequence of man-made clocks, but animals do behave on temporally defined schedules and many of them are observed to answer questions in the wild that require a specific estimate of time. Central place foragers must be able to find their way home, and many of them communicate information about distance to forage sites after they do so. Prey species adjust vigilance schedules to match predator density. Almost all animals will give up a foraging site if resource intake drops too low. In all of these cases, animals must judge the temporal duration of behaviors.
A unified theory of animal event timing requires that we know two things: how animals do it and why they do it. To our advantage, psychophysical studies of animal timing have already established that some animals do measure time, and this will provide us with a basis for conceptualizing event timing in a very definitive way. The purpose of this review will be to formalize what we know about animal event timing by addressing contributions from psychophysics, molecular genetics, neuroscience, and evolutionary ecology. The goal of this review will be to consolidate and build upon that information in a way that furthers our understanding of animal event timing.
The course of this review will follow a route that begins with what is known about animal event timing from psychophysics. This will be followed by a discussion of the molecular mechanisms of circadian clocks. Circadian mechanisms will help us to understand what the clock is not and also how it might operate at the cellular level. I will then suggest a possible mechanism for event timers that agrees with the psychophysical and physiological evidence for where these clocks are and how they operate. Finally, I will take a tour of ecologically relevant behaviors where ethologists may hope to find event timers at work in the world.
The Russian physiologist Ivan Pavlov (1849-1936) was the first to publicize event timing in animals. He recognized the trait when his salivating dogs began to `wait' to salivate after lengthy durations of the conditioned stimuli (story retold by Roberts, 1998). Since that time, psychophysics has developed a number of techniques for assaying animal event timing. Before exposing the established contributions of psychophysics to animal event timing, I would like to present the three most prevalent assays, as I will refer to them frequently.
One of the first and most easily used assays of animal event timing is the fixed-interval procedure (FI). FI schedules present a subject with a lever and reinforce lever presses after a fixed interval of time. There is no deterrent for lever presses prior to the reinforcement, lending subjects to press the lever at will until food is finally procured. Animals that are able to temporally regulate behavior based on experience typically show scalloped response curves in cumulative response records over time (Figure 1). The scallop is created by the increase in response rate as the reward time approaches. The alternative to this behavior is the `break-and-run' response, which is characterized by a short pause in procuring the reward followed by a steady response rate until the next reward (Figure 2). The inference from the break-and-run response is that the animal is indifferent to the temporal interval.
A more informative variation of FI is the peak procedure. In the peak procedure, animals learn that a lever will deliver food after a certain interval has passed since the initiation of a light cue. The difference between peak procedure and FI is that in the peak procedure, approximately 20 percent of the time, there is no food reward. Instead, the light cue stays on for a few minutes regardless of how the animal responds. It is during these no reward trials that responses are recorded. Inevitably the response rate increases up until the time of the reward and then decreases until the light is turned off (Figure 2). The peak rate matches the peak time.
The bisection procedure requires the subject to discriminate between two signals, long and short in duration (e.g., by pressing the left or right levers, respectively). The point of subjective equality (PSE) is the signal duration at which the subject responds with equal probability long and short. This data is usually presented as in Figure 3. This procedure has direct relevance to models of foraging theory, because it verifies the animal's ability to discriminate temporal intervals either in transit time between patches, in time between captures at a patch, or in handling time.
While the principles that I report here are not established for all animals, they do seem to be well established for vertebrates (i.e., birds, mammals, and fish). I will point out exceptional species in `Species Comparisons' (Section 2.6).
The ability to discriminate two temporal cues is reduced in a predictable way as the duration of those cues is increased. This property, known as Weber's Law, describes a linear relationship between the duration of a stimulus, I, and the measure of uncertainty associated with that stimulus, \triangle I. The relationship
|
The results from the peak procedure resemble a Gaussian distribution. Longer intervals between the beginning of the light cue and the reward lead to wider distributions (Roberts, 1998). Regardless of the length of the duration, the shape of the curve is conserved (Figure 1b).
In the bisection procedure, increasing the time between signals increases the accuracy, and longer duration signals require greater disparity in length in order to achieve the same levels of discrimination (Bizo and White, 1997; Roberts, 1998)
This is another critical observation for establishing the ability of animals to forage with the approving support of optimal foraging theorists. That is, if animals are unable to distinguish between the reward reliability of different temporal cues, then they certainly aren't going to be choosing patches based on the auspicious nature of environmental cues.
That rats can associate different cues with different frequencies of reward was again shown with the peak procedure. Light was associated with an 80 percent probability of food, and a tone was associated with a 20 percent probability of food. Animals responded to the tone at about one quarter of the peak response rate for the light (Roberts, 1981). In fact, studies of risk-sensitive foraging have established that animals can do far better than this in the wild. I will discuss this in more detail in Section 5.5.
It is difficult to imagine why animals would need to temporarily pause the clock. Nonetheless, pigeons and rats are clearly capable of this (Roberts, 1981; Roberts, 1998). An animal trained in the peak procedure will respond to a ten second signal blackout during the middle of a trial by moving its peak response rate back by ten seconds. This is true for a number of different blackout times, with the interesting caveat that longer blackout times lead to a slow resetting of the clock (de Vaca et al., 1994).
The first evidence that a temporal sense was not temperature compensated was provided long before the establishment of a temporal sense in non-human animals. Hoagland (1935) made this discovery after he subjected his sick wife to various temporal acuity tests while measuring her temperature. He noticed that at hotter internal temperatures her counting rate was faster than at lower temperatures. To exclude illness, he did the same for volunteers after short periods in a freezer. His data set exhibits an exquisite linear relationship between the inverse of temperature, 1/T, and the log speed of counting.
In perhaps the only ecological study of direct interval timing, the parasitoid wasp Trichogramma dendrolimi was demonstrated to lay eggs in its insect host based on the duration of its walk across the host's long axis (Schmidt and Pak, 1991). The hosts are eggs of larger insects, which vary greatly in size. This lends itself to an adaptive measure of host size. Too many or too few eggs laid on the host can lead to either starvation or under-utilization of resources. The host crossing can take between 0.5 and 20 seconds depending on the host size, but for identically sized hosts the speed is increased at higher ambient temperatures. At higher temperatures the wasp also lays its eggs faster. However, the wasp is able compensate for this temperature adjustment, and lays the same number of eggs in identically sized hosts regardless of the ambient temperature. Presumably, the wasps have a reduced estimate of elapsed time at lower ambient temperatures to compensate for their reduced speed (Schmidt and Pak, 1991).
In circadian studies of the courtship song of Drosophila melanogaster, the timing pattern of interpulse-intervals is temperature compensated and directly correlated with the duration of the free-running circadian clock. Under the assumption that the wasp's internal estimate of time is affected, in the same way that Hoagland's wife was, then this is our first evidence that circadian clocks are unrelated to event timing. I will pursue this argument further in Section 2.
The linearity of temperature effects are limited. Severe heat stress tends to reduce response rates, and very severe heat stress tends to change the motivational state of the animal altogether (e.g., they try to escape) (Richelle and Lejeune, 1980). Another problem with these results is that the temperature changes could be affecting other physiological properties of the organism with little effect on the clock. An experiment by Rozin (1965) looked at temperature effect in the goldfish Carassius auratus. In FI trials, goldfish at different temperatures showed similar scalloped response curves that were different in absolute but not relative response rates over the trial interval. This suggests that the metabolic rate changes do not actually change the relative features of the behavior but do change the absolute rate of response. It is still necessary to establish that in peak procedure trials the animal doesn't exhibit a lag in peak response. Even if the animal doesn't, it is still possible that temperature increased pacemaker speeds lead to more accurate clocks in judging small intervals. This remains to be tested.
Increased rates of reinforcement increase the relative temporal sense so that, in the bisection procedure, the PSE is moved towards the `long' response. This is supported by several studies that show altered PSE for different between trial durations (Fetterman and Killeen, 1991; Morgan et al., 1993; Bizo and White, 1997).
The consequences of reinforcement rate on the subjective perception of time are quite significant. I was unable to uncover any reference to this effect in ecological studies of foraging. To the extent that this phenomenon is real, it precludes the possibility of optimal foraging in the sense of Charnov's marginal value theorem (Charnov, 1976). In this theory, the ability of the animal to record accurate measures of time is directly responsible for its preference for certain food types and times of patch departure. The prediction of this reinforcer rate relativity is that animals enjoying a particularly fruitful patch will actually leave earlier than expected because their event timers are running fast. Thusly, their perception of the rate of food intake is reduced. For as long as the increased rate lasts, their subjective temporal perception will insist that intervals between food are actually longer than they really are, inclining them to give up perfectly acceptable patches under the marginal value theorem.
Animals clearly show different capacities for learning and different sensory biases (Bitterman, 1975; Dukas, 1998). The capacity for learning temporal intervals has clearly evolved in vertebrates and is summarily supported in at least one parasitoid wasp.
Essentially all vertebrates show scalloped responses (Richelle and Lejeune, 1980). The consensus among fish studies is that some do and some don't (Eskin and Bitterman, 1960; Richelle and Lejeune, 1980). The African mouthbreeder (Tilapia macrocephala) apparently fails to show a scalloped response, whereas goldfish (Carassius auratus) appear to have no problem with it (Rozin, 1965). Rather surprising, honeybees (Apis mellifera) show break-and-run behavior in pseudo-naturalistic environments with delays of 20 or 90 seconds (Grossman, 1973).
Are insects in general capable of timing short intervals? It is certainly possible that we have not yet asked the question in the appropriate way. It is equally possible that the genesis of certain brain structures coincident with vertebrate evolution mark a significant dichotomy in the evolutionary history of event timing. The evolution of the vertebrate forebrain facilitates the possibility of a homologous structure with the hippocampus in all vertebrates. The role of the hippocampus in spatial and temporal learning and memory is widely demonstrated (Thompson et al., 1982; Ono et al., 1995; Kesner, unpublished). Given their capacities for spatial learning in navigation, the possibility of an analogous structure in insects should not be ruled out (Dyer, 1998). I will address this relationship in more detail in Section 4.
Leads on the molecular and cellular mechanisms of event timing are sparsely distributed. Molecular geneticists have yet to isolate an event timer in animals. Neuroscientists have the power to isolate function of gross anatomical regions by lesion and transplant, but cell number limits functional understanding almost completely to neural simulations. Nor are neuroscientists equipped to investigate genetic and molecular function. The point of discussing circadian and ultradian clocks is twofold. First, it will allow us to make the distinction between mechanisms of endogenously controlled time and those of perceived time. Second, it will expose a molecular clock as a kind of null model against which we can address future questions.
Animal timing refers to a broad class of behaviors. These include general life history events such as when to stop making sperm and the age of first reproduction. They also include rhythmic behaviors with periods on the order of a year (circannual), longer than a day (infradian), a day (circadian), or shorter than a day (ultradian). The suggestion that animals have evolved to organize their lives in time could refer to any of these levels of behavioral timekeeping. On a very basic level, animal event timing refers to none of these.
The fundamental mechanisms of animal event timing are sensory based. Those of rhythmic behaviors and the scheduling of large scale life events are fundamentally based on endogenously controlled, often genetically predetermined, timetables. Circadian rhythms are 24 hours long not because the animal learns a 24 hour period in its lifetime, but because it is genetically predisposed to timing an interval approximating the rotation of the earth. On another planet, or in a lab where light-dark cycles (LD) are manipulated, circadian clocks are far from functional in terms of recording the length of the LD cycle. Daily activity cycles appear to fall into the same category. They are driven in well defined ways by circadian rhythmicity. It is a rare occurrence for an animal to learn when to be most active in the LD cycle (but see Section 5.4).
This section is also designed to point out that the clock used to measure event duration is not necessarily the same clock that measures time of day. The observation that the same clock measures differently timed intervals can be confounded by this time-of-day clock. For example, intervals of 24 hours do not need to be measured, because they can be posted to the time-of-day physiology. I will provide several lines of evidence in this section that the time-of-day clock is useless for measuring intervals different from 24 hours and show that it is quite distinct from animal event timers.
Circadian regulation of gene expression is well established in plants (i.e., Arabidopsis; Millar et al., 1995), fungi (i.e., Neurospora; Dunlap, 1996), insects (i.e, Drosophila; Iwasaki and Thomas, 1997), unicellular microorganisms (Lloyd, 1998), and vertebrates (Takahashi, 1995). I will describe in detail only the Neurospora and Drosophila system, because they share an idiosyncratic relationship that is telling with respect to circadian dynamics.
The filamentous bread mold, Neurospora crassa, exhibits a circadian rhythm in its asexual spore formation (conidiation) (Edmunds, 1988). The frq gene encodes a central element in this circadian rhythm. Both the frq gene mRNA and its protein product, FRQ, cycle in amount with a period of approximately 24 hours (Dunlap, 1996). In a 24 hour LD cycle, frq mRNA and FRQ levels are at their lowest point near the middle of the dark phase. Slowly rising, they peak approximately 10 to 12 hours later, with FRQ levels always lagging behind mRNA levels by about three hours.
The vinegar fruit fly, Drosophila melanogaster, shows oscillations in two of its circadian clock gene transcripts (per and tim) that are exactly out of phase with frq (Hardin et al., 1992). Their periods are the same, but when frq is beginning to rise, per and tim mRNAs are beginning to fall. Like FRQ, PER lags behind its mRNA transcript by several hours (Iwasaki and Thomas, 1997). Interestingly, a recently identified protein in mammals, mPer1, sharing sequence similarity with the Drosophila PER protein, oscillates in phase with frq (Shigeyoshi et al., 1997).
In the absence of light (DD), all of these gene products show free-running periods of about 24 hours. The phase of gene expression can also be directly entrained by light, but this response is limited to certain intervals of the cycle. For example, in Neurospora, when FRQ levels are low, a pulse of light leads to rapid transcription of the frq mRNA, moving the phase of oscillation forward in time (Crosthwaite et al., 1995). Given the swiftness of the response, it is believed that light acts directly on the frq promoter (Dunlap, 1996). mPer1 shows a similar response to light (Shigeyoshi et al., 1997). In Drosophila, light has little effect on per mRNA but reduces levels of TIM. Constant light (LL) breaks down the entire rhythm (Power et al., 1995). Given this evidence, it is not surprising that manipulations by light are unable to change the period of the circadian rhythm. An early rising sun simply moves the phase of the rhythm forward, but if the sun rises to early or too late, the clock is unaffected.
Light is not the only source of entrainment for the circadian clock. Temperature is also a cue (Iwasaki and Thomas, 1997). More interestingly, there is recent evidence that the circadian clock can be affected by conditioned stimuli (Amir and Stewart, 1996). There is some disagreement about the extent of this phenomenon. There is evidence that social contacts in humans can synchronize circadian pacemakers (Hastings, 1997), but recent research on blind subjects and manipulated light schedules in sighted subjects supports the necessity of light as an entrainment cue (Czeisler, 1995). Pioneering work by Winfree (1980) led him to predict strong resetting of the circadian clock by light across a wide array of species, and this appears to hold true for most cases.
After translation, the PER and FRQ proteins are localized to the cytoplasm. In short order, FRQ enters into a binding complex with (at least) itself and PER enters into a 1:1 heterodimeric complex with TIM, as suggested by studies in the yeast two-hybrid system (Dunlap, 1996). These relationships appear to stabilize FRQ and PER in the cytosol. They also appear to be necessary for translocation to the nucleus, where they either directly or indirectly suppress their own expression (Aronson et al., 1994; Dunlap, 1996). In the case of PER and TIM, other proteins are indubitably involved in this final step as neither are able to bind to DNA directly.
Although there are physiological temperature limits for circadian rhythmicity, there is no effect of temperature within these limits (Iwasaki and Thomas, 1997). It is perhaps better put that there are no effects of stable temperatures, as temperature changes can elicit rises or falls in gene expression (Edery et al., 1994; Rensing et al., 1995). The mechanism by which this temperature compensation operates has been worked out in detail for Neurospora.
Within the first 100 codons of the frq gene there are three methionine codons (AUG). These are represented by codon 1, 11, and 100 (Liu et al., 1997). Codon number 2 is not used to initiate the sequence under normal conditions, but codons 1 and 100 are. In other words, in wild type Neurospora there are two FRQ proteins, FRQ100-989 and FRQ1-989, and either of them alone is sufficient for circadian rhythmicity. At low temperatures, FRQ100-989 is preferentially transcribed. At high temperatures, FRQ1-989 is preferentially transcribed. The ratio between the absolute levels of the two gene products is thusly controlled by temperature. Removal of either form disturbs the ability of the clock to compensate for physiological temperature extremes (Liu et al., 1997).
It is possible that other mechanisms operate to compensate for temperature. Both per and frq encode an internally repetitive array of Thr-Gly codons. In D. melanogaster this region is polymorphic in length in both natural and laboratory populations (Kyriacou et al., 1992). Work done by Rosato et al. (1996) on D. melanogaster shows a significant latitudinal cline in this repeat sequence ranging from Europe to North Africa. These sequences show a direct relationship with the ability of the flies to maintain a compensated circadian rhythm at different temperatures (Sawyer et al., 1997). Furthermore, deleting the Thr-Gly region produces flies that have temperature sensitive circadian periods (Kyriacou et al., 1992).
Ultradian behaviors appear to oscillate in circadian time. This is best understood in D. melanogaster. In a screen looking for mutants of the pupal-adult eclosion phenotype, Konopka and Benzer (1971) identified three mutants of the per gene: pers showed 19 hour cycle, perL1 showed a 29 hour cycle, and per01 appeared to be arrhythmic. pers and perL1 mutations are due to single amino acid substitutions. per01 encodes a stop codon at the 460th residue (Yu et al., 1987). The amazing thing about these mutants is that their ultradian behaviors, like activity and courtship song cycles, are exactly proportional to their circadian behaviors (Kyriacou et al., 1992). Where wild type male flies have 60 second courtship song cycles, pers flies have 40 second cycles, perL1 flies have 80 second cycles, and per01 flies show little evidence of cycling at all.
Drosophila melanogaster females show a preference for 55 second songs and Drosophila simulans females prefer 35 second songs. It is a reasonable assumption then that the various per mutations would show changes in female preference for male song duration. In fact, just the opposite is true (Kyriacou et al., 1992). D. melanogaster females, regardless of their per genotype, prefer 55 second songs. This provides a fundamental difference between ultradian transmitter and receiver mechanisms. It also suggests that the `timer' in Drosophila is not part of an ultradian oscillation. This may be one of the more telling observations about circadian and event timing, given the peculiar absence of bees to discriminate short intervals despite their mastery of circadian time.
The famed ``Zeitgedachtnis" (time sense) of bees is widely reported (Wilson, 1971; Saunders, 1982; Seeley, 1995). There is hardly a shortage of evidence that bees can relocate almost anything provided it is presented at 24 hour intervals (Saunders, 1982; Moore et al., 1989). As well, `marathon' dancers who return to the hive dance floor to dance for hours following a particularly fruitful foraging trip compensate for the suns motion as the day passes (Wilson, 1971). However, in FI trials, honeybees show no evidence that they can learn anything about a 2 minute interval (Richelle and Lejeune, 1980).
Despite this evidence for a distinction between perceived time and circadian time, the relationship between ultradian and circadian behaviors is a profound one. The Syrian hamster tau mutant has a circadian rhythm that is shortened by four hours from the wild type. Wild type females have approximately 30 minute periods of cortisol and luteinizing hormone fluctuations that are slightly shortened in the tau mutant (Loudon et al., 1994). The recently cloned circadian Clock gene in the mouse was isolated in a massive screen for mutants that exhibited altered wheel running activity rhythms in constant darkness (King et al., 1997a,b). Close inspection of King et al.'s (1997) wheel running data shows that the wheel running activity cycle is very stereotypical in its ultradian oscillations.
The pineal gland's circadian rhythmicity and direct control of numerous hormonally controlled behaviors is further evidence for this relationship. At the receiving end of the suprachiasmatic nucleus, the pineal gland is located at the posterior dorsal aspect of the diencephalon. The rhythmic release of melatonin is its most obvious circadian feature, as melatonin is directly responsible for transmitting the circadian LD signal to the rest of the organism (Menaker, 1997). Melatonin is known to affect human body temperature and influence thermoregulation in lizards. Given the definitive relationship between body temperature and sleep (Wever, 1992), the pineal gland has a clear role in the sleep-wake cycle. In an experiment reported by Lavie (1992), eight young male adults experienced 20 minute ``days" for 48 hours in the 7/13 minute sleep-wake paradigm. In this experiment subjects were instructed to attempt to fall asleep or to resist sleep for 7 and 13 minutes every twenty minutes. The results show a well defined sleep-wake cycle at 24 hours.
It is not surprising that studies of interval timing exhibit circadian influence on attention and memory. One particular study by Meck (1991) tested rats for their ability to discriminate 2 and 8 second durations (bisection procedure) over the course of the day. There was no effect of circadian rhythmicity on the clock rate-the PSE did not change. However, the overall sensitivity to time (measured by the variability of the response) was highest during the dark phase and lowest during the light phase. This relationship has also been demonstrated for honeybee arrival times, with more accuracy in the morning than later in the day (Moore et al., 1989). If there is a clear relationship between circadian clocks and event timing, this is likely to be it. Experiments on human isolation in Antarctica, submarines, and underground are typically difficult to interpret because subjects often lose their abilities to concentrate; it is as if the mechanisms that control attention are lost (Harrison et al., 1989).
Unfortunately, the vast majority of psychophysical literature fails to mention the time of day at which experiments were performed or any salient features of the LD cycle. Ecological studies on animal behavior are seldom better: the unspoken assumption being that the behavior, once initiated, is stereotypical. Variance is regarded as `noise' in the form of genetic variance or sensitivity of the animal to subtle environmental factors. However, given Meck's results (1991), it may be variance in the animal's sensitivity or general attention that is responsible. Accounting for this relationship is paramount in establishing criteria for animal motivation and attention.
A second issue with respect to the importance of recording LD cycles is that animals that aren't reared in 24 hour LD oscillations don't exhibit circadian rhythms (Richelle and Lejeune, 1980). For example, if the LD cycle at a particular geographical location or in a particular lab is highly unpredictable, then the activity rhythms observed in animals reared there may be quite arrhythmic or may be based on other cues such as temperature or resource availability.
Lastly, animals that exhibit circadian rhythms may be biased towards remembering 24 hour event times. Honeybees are exceptionally good at following the daily peaks in pollen and nectar production by visiting only certain species of flowers at certain times of day (Saunders, 1982; Seeley, 1995). However, bees are unable to learn to return at non-24 hour periods. Kestrels (Falco tinnunculus) and starlings (Sturnus vulgaris) have also been observed to follow food abundance in time and both can be trained to return to a feeder at specific times of the day (Bell, 1991). Food-anticipatory activity rhythms have also been observed for rats fed at 24 hour intervals (Rosenwasser, 1984). Further work needs to be done to understand this 24 hour memory bias. Given the nature of temporal memory, which I discuss further in `How Event Timers Might Work' (section 4), it is probably also the case that there is an annual memory bias as animals may be able to assess their seasonal context simultaneously with important events.
More work needs to be done to verify the distinction between circadian rhythms and event timers. A first step would be to quantify the temperature effects on Drosophila song delivery and reception. Here we can make a clear distinction between metabolic rates and perceptual and behavioral time. Another possibility would be to investigate giving up densities in flies during multi-patch foraging. Patch departure times should reveal under or overestimation of patch resources for pers and perL mutants, respectively, if circadian genes are controlling perceived time.
Molecular examinations of mechanisms involved in event timing require a phenotype for the isolation and cloning of relevant genes. A designer behavior would be particularly useful in this respect. Tim Tully (DeZazzo and Tully, 1995) has made superb use of electrical stimulus to dissect memory formation in Drosophila. To determine if flies are capable of learning temporal durations, flies could be trained in a periodic shocking regime. For example, every 60 seconds the experimenter runs a current through the cage. If flies have event timers, they may learn to associate flight at specific temporal intervals with the absence of shock. Reward paradigms might also be useful, in which flies are rewarded with food at consistent intervals. Once the phenotype is established, various per mutants could be assayed for defects in event timing. These mutants could then be assayed for more ecologically relevant behaviors associated with fitness. Caenorhabditis elegans might also be a useful subject for molecular study of event timers, as there is some preliminary evidence that they can learn 15 second intervals of time (Galloway and Rankin, pers. obs.).
In some respects, the observation of circadian rhythms in unicellular organisms offers up the possibility that all cells contain circadian rhythms. While it is difficult to discount this possibility, expression patterns of circadian genes and brain lesion and transplant studies support specific localization of circadian clocks and event timers.
It is fairly well established in vertebrates that there is one circadian pacemaker: the suprachiasmatic nucleus (SCN) (Ralph and Hurd, 1995). The SCN is located over the optic chiasm rostral to the supraoptic nucleus. Cross-genotype transplants of SCN for Syrian hamster tau mutants have unambiguously defined the SCN as the home of the mammalian circadian period (Ralph et al., 1990). The SCN also shows strong expression of the mPer1 and Clock genes, which are rapidly induced by exposure to light (Shigeyoshi et al., 1997; Hastings, 1997).
SCN neurons in hypothalamic slices are observed to fire rhythmically at around 8 to 10 Hz during the day and 2 to 4 Hz at night (Wagner et al., 1997; Hastings, 1997). Isolated SCN neurons show a spontaneous rate of firing at near the same rates as in slices (Hastings, 1997). They also show a higher order rhythm in frequency over the circadian day. Despite the rhythmic firing, the SCN reaches most of its targets via thalamus and hypothalamus (e.g., the pineal gland) mediated hormonal control (Hastings, 1997).
SCN function in natural environments is still poorly understood. In the laboratory, rodents rendered arrhythmic by SCN lesions live normal lifespans (DeCoursey and Krulas, 1998). Besides circadian arrhythmia, the hibernation cycle is also affected by SCN lesions. SCN lesioned female squirrels (Spermophilus lateralis) were observed to hibernate for almost 2 years in a laboratory setting (Ruby et al., 1996). Psychophysical experiments on event timing in SCN lesioned animals remain to be done. A more ambitious study of SCN-lesioned chipmunks returned these animals to the wild after surgery. Unfortunately, after 3000 hours of fieldwork over slightly more than two years, there was essentially no observed effect of the SCN removal (DeCoursey and Krulas, 1998). The SCN-lesioned chipmunks did show evidence of brief arrhythmia and nighttime `restlessness' in the wild, but the activity cycles were largely the same for all chipmunks studied. There was also no significant effect on survivability, reproduction, or winter torpor duration.
There is no obvious relationship between the SCN and event timers. On the other hand, event timing does have several direct and indirect relationships with the hippocampus. The hippocampus has played a starring role in mammalian learning and memory since the lesioning of H.M.'s medial temporal lobes in the 1950s. After the surgery, H.M. was completely unable to form new declarative memories (Churchland and Sejnowski, 1994). Declarative memories are akin to semantic memory in the sense that the memory is based on the learning of semantic statement. Procedural memory, for which H.M. showed only minor deficit, is based on a kind of implicit function learning. For example, H.M. was perfectly capable of learning a motor skill, but he would be unlikely to remember that he had learned it.
Since that time, a great deal of attention has been paid to role of the hippocampus in the formation of spatial memory. In food storing birds, damage to the hippocampus disrupts memory for storage sites (Krebs et al., 1989). As well, bird species that store food have significantly larger hippocampal formations than those that don't. The operation of the spatial function of the hippocampus appears to work via location coding neural cells (place fields) (Mizumori et al., 1996). When the animal returns to a specific spatial location, similar cells in the hippocampus fire. This information appears to be primarily visually coded.
The function of the hippocampus has also been established in the formation of episodic memory (Fletcher et al., 1997). Episodic memory is typically associated with the ability to recollect past events, however the precise meaning of the term has become increasingly ambiguous since its introduction by Endel Tulving (1972). It was coined to represent any type of memory that was not strictly lexical (semantic memory), and thus it would represent essentially all temporal forms of learning. Recent evidence using positron emission tomography (PET) finds the retrieval of episodic memory events localized in the right prefrontal cortex (Fletcher et al., 1997) with a limited functional role played by the hippocampus.
Hippocampal lesions are shown to operate in the formation of temporal memory in rats (Meck et al., 1984). Once the memory is formed, the hippocampus becomes less important. Effects of hippocampal lesions include a lack of avoidance for previously visited maze sites, the inability to withhold or inhibit previously learned response patterns, and the foreshortening of temporal memories (Meck et al., 1984; Meck, 1988)
Given the hippocampal evidence, it seems likely that the hippocampus operates as a kind of ticking backdrop upon which episodic events can be hung until they are stored in a longer-term reference memory. Certainly hippocampal patterns are utilized in numerous ways, as similar place fields do fire in new locations over the lifetime of a rat (Mizumori et al., 1996).
This kind of hippocampal tagging has been modeled in detail for sequential spatial memory using a large scale simulation of hippocampal function (Wallenstein and Hasselmo, 1997; Wallenstein et al., 1998). Multicompartmental pyramidal cells are shown to have synchronizing behavior over multi-trial learning and this is suggested as a mechanism for sequential learning. The pyramidal cells that fire apparently chaotically before the learning trial, use this pre-randomization to settle into nonpredictable but highly sensory specific patterns of firing. This is reminiscent of the kind of unsupervised self-organization exhibited by Kohonen networks (Haykin, 1994). It is different in that neighboring events in space share similar contextual patterns of hippocampal cell firing. In this way, when a sensory stimulus arises that is familiar, the cells that fire in response to that stimulus partially stimulate the `memory' of neighboring events.
Is it possible that temporal interval discrimination works by a similar mechanism? Coexisting with place fields, we would expect to find an analogous kind of `time field'. By necessity, the time field could not operate exactly like the place field. In space, sensory input is constant, and rats could run spatial information through the hippocampal filter persistently during the trial. This would perpetuate the synchronous cell firing in a way that time fields might be unable to do. For example, in order for a rat to learn the temporal duration between signal and reward, it must sense the interval. In space, the interval is exogenously applied-the space field moves and the sensory cells respond. In time, the interval must be measured by an internal clock.
What could the clock be if it is not related in some way to the circadian clock? The observation of high-frequency intracortical oscillations by electroencephalograph (EEG) provides a basis for a possible clock hand. It has been suggested before that composite cortical waveforms measured by EEG operate as a pacemaker in duration timing (Treisman et al., 1994; Artieda and Pastor, 1996). Unfortunately, EEGs measure waveforms produced by populations of presumably loosely coupled cells, and it is very difficult to isolate particular areas of the brain for specific analyses. Nonetheless, evidence for the EEG-pacemaker hypothesis has been provided by work showing interactions between auditory click rates, certain EEG components, and the simultaneous assessment of duration (Treisman et al., 1994).
A more specific neural pacemaker central to the hippocampus would offer more support for time fields as a measure of temporal acuity. This may be provided by observations of theta and gamma oscillations from in vivo recordings of the hippocampus (Wallenstein and Hasselmo, 1997). In the sequential place field model referred to above, theta and gamma oscillation are produced by GABAergic receptor inhibition of recurrent collaterals among pyramidal cells and between pyramidal cells and non-pyramidal neurons. Verbally, the effective nature of this system is to `iterate and check' at each time step, such that internal and external signals are integrated in a meaningful way. Certain cells fire for longer intervals and become associated with the duration, such that a sequence of population patterns is gradually learned in progressive trials.
Iterate and check, however, might be only half of the story. Animals involved in temporal training tasks often behave in a peculiar but stereotypical way that might be further related to the spatiotemporal integration of the hippocampus. This behavior is characterized by seemingly unrelated activities between the stimulus and the reward. For example, a rat might chew its tail; a monkey might jump around its cage in a repetitive way; a human might tap her finger or shake her head. It is also observed that animals engaging in this `collateral behavior' are more efficient in their response time than animals which don't perform these behaviors (Richelle and Lejeune, 1980). These behaviors could act as a kind of context counting. Assuming the animal is unable to count (or asked not to, in the case of humans), they may, in the process of learning the interval, learn sequential behaviors associated with the particular sequence of population patterns in the hippocampus. This makes perfect sense in terms of the spatiotemporal aspects of hippocampal learning. It also provides a physiological basis for the behavioral theory of timing and multiple-time-scale theory. Not only does the animal iterate and check, but it reinforces the dynamic pattern of cellular events by engaging in context specific behavior.
Evidence for an embedded, context specific memory is supported by research on temporal memory in humans (Friedman, 1993). An appropriate, but less physiological, model for how events are recollected is the theory of reconstructive memory. Reconstruction of remembered events is based on recognition of an event with respect to extant cues during the event interval. Reconstructive memory explains otherwise anomalous characteristics of memory, like primacy (enhanced memory for initial events), scale effects (e.g., more accuracy for time of day than month or year), and facilitative effect of background temporal structure (Friedman, 1993). Subjects may actually be self-generating temporal structure through collateral behaviors. Reconstructive memory also supports bias towards memory of events with more endogenous and external cues, as in 24 hour and season memories.
Since most animals do not have a hippocampus, I would now like to discuss a smaller timer, one that can easily be carried by individual cells. It shares its basic molecular features with that of the circadian clock, except for one thing. This clock is linear. It operates on a simple mechanism of stimulus induced production followed by decay. The clock, which I will call the decay timer, is probably at work in countless ultradian behaviors, and potentially in primitive event timers. For example, the nematode Caenorhabditis elegans exhibits a distinctive foraging pattern following periods of starvation (Hills, pers. obs.) The behavior changes in a characteristic way over time, such that the nematode prevents foraging in areas where it has already searched. The most feasible explanation for this behavioral change is some kind of decay timer, with dynamic protein levels responsible for the tonic level of firing in specific neurons. The decay timer also has the potential for measuring different intervals of time by adjusting the amplitude of the initial expression or the response level at specified thresholds. In this way, animals with different life experiences can learn to adapt their timer to specific environments. The fundamental metabolic properties of the decay timer also make it agreeable with temperature effects on clock speed.
Experiments designed to distinguish between decay timers and time fields are probably bound for failure. The reason is that the time fields model is perfectly capable of acting like a decay timer, and may in fact consist of numerous decay timers that set the context for contiguous spatial and temporal phenomenon. However, there is a rather deep distinction between these different event timers in the form of the credit-assignment problem as it is established in the psychological literature (Machado, 1997; Staddon and Higa, unpublished). The credit-assignment problem is based on the animal's attention to the reward relevant cue. How, for example, does a rat learn that the onset of the red light signals food in 40 seconds and that changes in air temperature are unrelated? I believe that one of the premises of the decay timer must be that the credit is assigned in the evolutionary history of the animal. Bacteria, bees, and other invertebrates do not learn to assign a particular stimulus to a decay timer. That is given to them for free. On the other hand, the decay timers in the hippocampus are actually used to solve the credit-assignment problem. They do this by maintaining the firing rates of certain pyramidal cells even after the specific response stimulus is gone. This allows contiguous events in time and space to be contextually associated with neighboring events (Wallenstein et al., 1998). The relevant cues to which any given event timer are sensitive are probably intimately related to cognitive bias for certain environmental cues. Thus, negative results on event timing experiments may be limited to telling us about a very specific environmental stimulus (see `Navigation', Section 5.2).
The remainder of this review will be devoted to understanding why animals need event timers. One might argue that there is no need, all ecologically relevant behaviors could just as easily be performed without this faculty. I recognize this argument not because it is helpful to understanding the behavior, but because it elucidates a very real constraint on our understanding of animal event timing. In the natural world, it is very difficult to tell by what cues an animal makes its decisions.
With respect to this argument, we are at a slight advantage. We know the behavior exists. At this point we are merely in the business of knowing why. But how can we know why if we aren't sure when the behavior really exists? Pigeons in cages can learn the difference between 2 and 8 seconds. Does this mean that an osprey uses time to capture prey? It does not. Does this mean that squirrels make assumptions about how much time they would have to escape given the appearance of a predator in a particular location? It does not mean this either.
The first step toward understanding the adaptive contribution of event timers is to recognize the contribution they would make if they were being used. This will provide us with some sense of the situations that might have facilitated the evolution of event timers. The following tour through event timing in the natural world will focus primarily on the domains of animal behavior that would show a positive fitness relationship in the presence of a cost-free event timer. I will discuss the difficulties with assessing event timer cost in `Conclusions' (Section 6).
Communication is a distinctly temporal behavior. It requires transmitting signals in sequence and duration such that a receiving organism `understands' the message. There are cases when the timing of symbolic components of the message are unimportant, for example, when one is merely trying to get another's attention. Here we are strictly interested in the relationship between syntactical elements in time and the way in which that conveys information. More specifically, we are interested in animals that learn how to communicate.
In the case of songbirds, songs are acquired by listening to other birds (Beecher et al., 1998). The primary reason for this appears to be ecological. For song sparrows, the song repertoire is usually learned after the bird leaves its birthplace and during the first season of territory establishment (Beecher, 1994). In this way, the song sparrow learns the social communication strategies of its neighbors. This is further evidenced by song sparrows showing a preference for the learning of already shared songs (Beecher et al., 1998).
The fact that birds can learn frequency and durational components of a song immediately implies a usefulness for event timing. If one bird intends to mimic the call of another, then it must be able to record that call in memory. Functionally, the mechanism that records the call is an event timer. Whether or not this ability can be generalized to record the times of nonsyntactic events remains to be established.
Linguistic studies in humans recognize a temporal component, but a clear understanding of exactly how that component is manifested is far from understood (Port et al., 1995). Some models of language acquisition are, however, distinctly similar to contemporary models of hippocampal function in their use of recurrent networks (Port et al., 1995; Wallenstein et al., 1998). There is also evidence that perceived time is shorter for familiar auditory signals than it is for unfamiliar signals, suggesting that perceived time is not absolute (Kowal, 1984). The main difficulty with results from linguistic studies, as it is for essentially all studies on communication, is that it is extremely difficult to separate the meaning from the message.
Insect communication is broadly understood to carry information in its gross rhythmicity. There is clear evidence that insects distinguish likely mates by the gaps between signals, the interpulse interval (IPI) (Michelsen et al., 1985; Kyriacou et al., 1992). In the case of Drosophila this IPI transmitter appears to be under genetic control with a well defined relationship with circadian rhythms.
What about the receiver? Moths are useful for studying auditory transduction because they have a well defined and relatively simple ear, with one or two receptor cells attached to an accessible tympanum. The evidence for moths suggests that specific cells can operate as frequency filters (Michelsen et al., 1985). In this way, insects could completely bypass any need for event timers.
Animal navigation involves a gamut of sensory acuities to various environmental signals. As a consequence, there are countless ways to avoid non-circadian clocks: Celestial and magnetic compasses are used by birds, insects, and fish; honeybees use polarized light; wasps have memory for landmarks; salmon can find their breeding grounds by smell; amphipod crustaceans use the slope of the ground (Daan, 1981; Dyer, 1998; Gould, 1998). In fact, the only evidence for an event timer in navigation is that some animals appear to know how far they have gone.
This is best exemplified by the waggle dance of honeybees. Remarkably, honeybees returning to the nest can inform other workers of the exact whereabouts of a forage site. The waggle dance transmits direction by establishing an angular relationship between the sun and the forage site in the form of a linear waggle movement at the same angle from the vertical axis of the hive. The distance to the site is transmitted by the distance of the linear waggle (Figure 5). The worker increases her waggle distance by about 75 milliseconds per 100 meters of foraging distance (Seeley, 1995). The use of a decay timer in transmitting this signal is not altogether problematic. Time based on metabolic costs are probably far more prevalent than times based strictly on the observation of events. The energy expenditure in flight could be used as the assay of flight duration. Bees may leave the hive or return to it with tuned energy stores so as to accomplish this. That another bee watches or follows the dance and then knows the distance to the forage site implies a more complicated mechanism.
The behavior of honeybees in the wild makes the FI data for honeybees particularly cumbersome (Grossman, 1973). It suggests that despite Grossman's effort to simulate a naturalistic environment in order to measure honeybee event timing, he may have been asking the question in the wrong way. There is no evidence that honeybees are confronted with anything that replenishes itself at one minute intervals in the wild (Seeley, 1995). A honeybee may return to the hive to communicate the exact location of a new nectar source and still be unable to learn that a flower takes 20 seconds to fill up. This entire behavior pattern appears to follow the logic of the specificity of decay timers. In general, if one expects to find event timers in insects, then one needs to accept the possibility that these event timers are very phylogenetically derived and highly specified. Whatever the cost of the hippocampus, it is probably far out of reach of the typical invertebrate.
The role of chemotaxis or navigation by thermal gradients is typically overlooked in the standard navigation literature. Gradient navigation requires either spatially or temporally separated samples of the environment. Which of these to choose depends critically on the size of the organism. For Escherichia coli, and similarly sized animals, the primary difficulty lies in the signal-to-noise ratio introduced by Brownian motion (Berg, 1993). Temporally spaced samples are of limited utility if past positions are equally likely to be in any direction from the present position. It is for this reason that as size decreases, spatial mechanisms become more informative (Dusenberry, 1998). While there is support of spatially distributed processing of information in E. coli (Grebe and Stock, 1998; Dusenberry, 1998), there is also evidence that they weight temporal experiences over time (Segall et al., 1983).
The use of event timers might facilitate gradient taxis, but it isn't necessary. In the case of E. coli, the mechanism may operate by receptor activation of a kinase, resulting in an increase in phosphorylation of a flagellar motor response regulator (Grebe and Stock, 1998). This is essentially a detailed description of a decay timer. In good times, the cytosol fills up with these response regulator proteins, maintaining the forward motion of the bacterium. If the bacteria leaves the source of the favorable receptor ligand, the regulator protein decays until the flagellar motor reverses direction, tumbling the bacteria onto a new course. The tumbling event probably resets a threshold of regulator proteins so that the bacterium can move forward before turning again.
Reproduction involves a highly defined series of behaviors. Beginning with locating a mate and potentially securing a territory, animals undergo a very disciplined and stereotyped sequence of behaviors typically ending with the offspring's departure (Hills, pers. obs.).
To locate a mate, an animal must know where and when to look. I was unable to find any evidence that animals learn the temporal aspects of this behavior. In most cases, the timing of particular mating behaviors appears to be a possible form of sympatric speciation. For example, the temporal staggering of male mating flights in East African army ants and the nocturnal activity patterns of moths maintain a temporal separation in reproductive timing (Daan, 1981).
Evidence for a potential relationship between event timing and circadian rhythmicity was found in the time sharing behavior of parent doves (Streptopelia risioria) (Silver and Bittman, 1984). The female spends up to 18 hours on the nest each day and is temporarily relieved by the male during one six hour interval. If the male is prevented from starting his sit bout at the appropriate time, the female will return at her circadian time regardless of the duration of the male's effort. The male, however, will dispute the rightful sitter with her until his six hours are finished. Given that it is unknown whether the male learns the appropriate sitting interval, it is impossible to conclude that the dove has an event timer. The male dove might simply have a genetically timed six hour alarm clock in its brain that is started by sitting on the nest.
How might this alarm clock work? Presumably, in a very similar way to the operation of light on the circadian clock. In the case of the male dove, sitting on the nest causes a pulse of gene expression in the `sit' gene. The SIT proteins are degraded over time until they reach a minimum threshold level, at which point the dove gets up. This is, of course, the decay timer.
Animals attempting to avoid predators can do so in a number of ways. They can wait for them to arrive and then try to escape. They can time their activity out of phase with predator foraging. Or they can attempt to satiate predators with fellow members of the species. All of these behaviors could gain from event timers.
There are several methods of escape in the sense that I use it above. An animal might run for cover, or it might trick the predator into thinking it is not a prey item. In either case, learning the principles of predator vigilance can increase foraging efficiency. In some cases, the presence of predators may actually enhance foraging efficiency (Holtcamp et al., 1997).
For an animal foraging in the open, attention to predators is necessary for survival. But how much time should an animal devote to vigilance? Animals that economize predator vigilance strike an optimal relationship between eating and being eaten (Dukas, 1998). Potential environmental factors are the nearest possible escape and the proximity of possible predators. In the latter case, there is evidence that adult ground squirrels with obstructed views of their surroundings are more vigilant than those with clear views (Arenze and Leger, 1997). Juveniles were undeterred, suggesting the behavior learned.
A useful trick against predators is feigning death. Vigilance is helpful here, but of equal importance is how long one should stay down for. Anxious resurrection will indubitably lead to real mortality. But staying dead until a predator unbeguiled by death arrives is an equally poor outcome. Domestic chicks perform the death feigning behavior instinctively. Predictably then, the time spent inanimate appears to be affected by the circadian period (Richelle and Lejeune, 1980), suggesting a connection with the courtship song of flies.
Predator avoidance can also take the form of knowing when predators are active and choosing to be active at other times. This is exemplified by the behavior of baby alligators (Alligator spp.), which when heavily preyed upon by African fish eagles (Haliatus vocifer) move from diurnal to nocturnal activity rhythms (Curio, 1976). This suggests a phenotypic plasticity in the way behaviors are linked to circadian clocks, reminiscent of honeybees learning the daily cycles of nectar production.
By far one of the most popular forms of predator avoidance is feeding them your neighbors. Plants do this in the form of masting (Silvertown, 1980). Periodical cicada (Magicicada spp.) are one of the more artful exemplars of this phenomenon (Lloyd and Dybas, 1966). Having the longest known life cycles of any insect (barring some queen ants), they emerge from the ground to mate, lay eggs, and die within weeks of one another every seventeen or thirteen years, depending on the species. The behavior is hypothesized to be a predator satiating mechanism above ground and a predator avoidance mechanism below ground (Lloyd and Dybas, 1966).
The predator satiating mechanism works on the premise that predators are limited in their maximal intake rate of prey. This can be due to simple satiation or to prey handling times. For example, when guillemot fledglings (Uria lomvia) jump from their breeding cliffs the probability of death is significantly enhanced if the bird jumps alone (Daan, 1981). The usual strategy is to jump with everyone else, so that the fledglings are shielded from the predatory glaucous gulls (Larus hyperboreus) by other members of their cohort. If the glaucous gulls did not have a maximal intake rate they would presumably eat everyone as soon as they were exposed.
Animals acquire resources in countless ways. Temporal perception is useful in many of them. Speciation mechanisms are undoubtedly related to competitive exclusion in competition for resources. A possible force of sympatric speciation is via resource partitioning in time. In this case, animals forage at different times of day but still eat the same foods, as is observed in several species of tern, lizards, crustaceans, and gastropods (Schoener, 1970; Schoener, 1974). This presumably reduces competition while simultaneously economizing daily time use. Whether or not these behaviors are learned is unestablished.
A basic assumption of optimal foraging theory is that animals recognize something about resource distribution. This recognition can be more or less behaviorally plastic depending on the cognitive faculties of the animal. If resource distribution is relatively stable over time, a species may evolve a patch departure schedule that is based on generations of trial and error without regard for the present environmental conditions. At the other extreme, an animal with an event timer could measure the rate of food intake at different patches or with different foods and compare them so as to optimize its schedule in the future. It could also measure the time between patches and incorporate this into its overall strategy. The marginal value theorem assumes that animals know resource distribution, transit and handling times even before they begin foraging (Charnov, 1976; Valone and Brown, 1989). This provides a useful null model against which to compare animal behaviors.
Yet, we know most animals require patch assessment before they can make optimal foraging decisions (Valone and Brown, 1989). Constraints on forager memory and resource changes over time force the reinvestigation of patches (Belisle and Cresswell, 1997). These costs to the perceptive forager are not well quantified.
Animals do use temporal aspects of resource distribution to make decisions about patch departure. Among central place foragers, there is a positive correlation between distance traveled to the foraging site and the patch residence time (Kacelnik, 1984).
Studies of risk sensitive foraging have exposed a sensitivity to the variance of resource acquisition even when the mean is unchanged (Real and Caraco, 1986). For example, honeybees prefer stable rewards to unstable rewards, regardless of the mean. This requires an event timer. The adaptive explanation for stable versus unstable preferences is based on minimizing the risk of starvation. I present a graphical argument for this in Figure 6. When animals are experiencing a positive energy budget they should prefer stable resources, as this minimizes the probability of starvation. When experiencing a negative energy budget animals should be risk prone because unstable resource distributions represent the highest probability of recovering (Stephens, 1980).
The data on animal preferences does not entirely corroborate this argument (Ha et al., 1990; Bateson and Kacelnik, 1998). Explanations for this phenomenon are based on cognitive constraints related to time perception and memory. Animal's may discount time in different ways depending on past experience or genetic predisposition, or they may average rate intake over different intervals. Animals also have certain constraints on their abilities to discriminate event times, as typified by Weber's Law (Bateson and Kacelnik, 1998). I suggested earlier that animals may also suffer from distorted perceptions of time based on intake rate. Mathematical models incorporating intake rate effects will help to explain how this perceptual bias affects observed foraging behaviors.
A kind of first impression among animals, called side bias, sometimes confounds psychophysical results (Ha et al., 1990). Side bias seems to refer to some unknown force controlling the animal's behavior. Experimenters typically make an effort to remove these animals from the analyses. Nonetheless, every animal may experience this kind of bias with variable time reinforcement schedules. Large initial rewards could lead to particularly strong cognitive bias. A series of large rewards might also instill a memory of a rare event that keeps the animal coming back. Exactly how the temporal sequence of events establishes memory biases is still an open question.
Another temporal factor in foraging is the effect of time horizons (Krebs and Kacelnik, 1984). Time horizons undoubtedly affect the behaviors of animals that are able to anticipate the ends of foraging bouts. Late in the day an animal may choose to continue foraging in a poor patch because it doesn't have enough time to get to a better one. A mechanism to avoid this problem involves organizing a series of patches in time and visiting them so as to maximize resource gain over the duration. Traplining fits this criteria.
Traplining is a behavior seen in bats and number of birds and frugivorous primates. It involves following a prespecified path during the daily foraging bout (Bell, 1991). Time horizons undoubtedly affect traplining schedules, but, once scheduled, traplining provides a short term answer to the time horizon problem. A similar behavior pattern is cropping. Cropping involves visiting locations at intervals that allow for resource replenishment. Cody (1971) observed various species of finches cropping seeds in the Mohave desert at the base of a mountain range. These birds moved their foraging sites to different distances from the mountain each day, scheduling visitation rates to match replenishment rates. Insect eating shore birds also appear to crop along the shore. The ant Veromessor pergandei makes radial changes in its foraging pattern outside the nest of between 14 and 17 degrees each day (Bell, 1991). In some manifestations of cropping, an event timer could help an animal know when to return to a foraging site.
In order for an American osprey (Pandion haliaetus) to intercept a catfish (Chaca chaca) in shallow water, it must perfectly time its descent and penetration of the water to match the location of its prey. Individual osprey have been observed catching many different kinds of fish, and this immediately suggests that osprey learn to anticipate the position of their prey by observing something about individual fish (Bent, 1961). Numerous predators intercept moving prey (Curio, 1976), whether it be wolves (Canis lupus) taking down moose (Alces alces) in the Yukon or golden eagles (Aquila chrysaetos) catching rabbits (Lepus spp.) in plain country. The behavior seems to be a general one. But does it require an event timer?
For insects, the answer is probably no. At least in tiger beetles, the method of pursuit and capture is to constantly move where the prey was, with rapid halts to reupdate the prey's current location (Gilbert, 1997). The other alternative is that the insect measures the velocity of its prey and moves to where the prey will be. Evidence of the latter does not exist.
Do osprey pursue like tiger beetles? It is too early to tell. Computer imaging of predator and prey paths, like those done for the tiger beetle (Gilbert, 1997), are not yet being used for larger animals. In the case of fish pursuit by birds, even simple video analysis is constrained by simultaneous water and air analyses. Animals moving afield of the equipment is bound to be problematic. Still, this is likely to be the most informative method for determining the nature of larger predator pursuit and the mechanisms involved.
There is another case where predator pursuit may involve timing and that is in group foraging efforts. Members of a concerted predatory effort must `understand' their particular duties in relation to other members. For example, observations of predatory groups breaking up to surround prey on scales at which they are not visible to one another requires an estimation of other group members in space and time (Curio, 1976). Similarly, knowing when to takeover in the pursuit of prey in serial efforts necessitates an understanding of when to act. Temporal event timers may actually be quite remedial in behaviors of this type. That is to say, the behavior necessitates a level of cognitive ability far more costly than a mere stop watch. Knowing the traits of other individuals in the group, recognizing fatigue or opportunity, being at the right place at the right time, all of these things require clocks plus ample cognitive space for allocating memories and learned predatory skills. Yet, surprisingly, honeybees appear to perform essentially the same feats of temporal economy within the hive (Moore et al., 1998).
The role of the hippocampus is well established in associative learning assays of the conditioned-stimulus-unconditioned-stimulus (CSUS) type (Thompson et al., 1982; Ono et al., 1995). Typically in CSUS trials an animal is trained to associate an otherwise noninformative CS (e.g., an odor) with a previously meaningful US (e.g., pain or reward). The archetypical example is that of Pavlov and his bell-stimulated salivating dog. The paradigm takes advantage of the fact that the animal has some unconditioned response (UR) to the US, so that the experimenter can verify association of the CS by omitting the US.
As one might expect, there is a clear relationship between the timing of various components of the CSUS and the ability of the animal to learn the association. For example, there is an optimal time of CS length that maximizes learning rate (Cooper, 1991). This is reminiscent of the credit-assignment problem and suggests that duration of environmental signals may affect attention for those signals when they are subsequently paired with relevant stimuli.
At the level of the synapse, long term potentiation (LTP) in the hippocampus is a form of learning based on the modulation of synaptic gain between the presynaptic and postsynaptic cell (Churchland and Sejnowski, 1994). NMDA glutamate receptors are required for LTP, and their voltage-gated properties necessitate that the postsynaptic cell be firing for a certain temporal duration while the presynaptic cell is firing in order for the NMDA receptors to be activated. NMDA activation does appear to operate on gene transcription (Bading et al., 1993) and this may be a mechanism for the modulation of long term synaptic gain. Several phases exist for hippocampal LTP and these have been compared to the various phases exhibited in vertebrate memory (DeZazzo and Tully, 1995). A recent finding suggests that this mechanism may also operate in a retrograde fashion by weakening synapses when the presynaptic cell fires after the postsynaptic cell action potential (Markram et al., 1997). This mechanism is not an event timer in the usual sense; it is more akin to a coincidence detector. But it does exhibit the flavor of event timers and is certainly relevant to the contiguity of the CSUS sequence.
More specifically, the hippocampus appears to be required in order to get beyond coincidence limited association. In trace conditioning assays, in which a puff of air follows a brief tone, the hippocampus is required for association of nonoverlapping stimuli (Thompson et al., 1982). Hippocampal lesions only allow learning when the CS and US are overlapping in time.
Hippocampal-type event timers offer substantial associative powers in the development of tool use and causal recollection. A Japanese macaque (Macaca fuscata) washing a potato must somehow learn to associate the improved potato with the action that preceded it. A more common example of this kind of temporally gapped causal association is the learned food avoidance response exhibited in a large number of vertebrates. If an animal becomes sick from eating a toxic food, it may learn to avoid that food in the future. There is clearly a temporal relationship here because delay of adverse consequences inevitably limits the association of the food with the effect (Stephens and Krebs, 1986).
Time is easily one of the more slippery subjects in four dimensions. Our linear and subjective experience of it make it rather difficult to define. Understanding how other organisms experience it is, thusly, quite a leap. Though, when one stops to take it all in, a considerable amount is actually known about how animals perceive time. Molecular geneticists are in fact on the verge of discovering the mechanisms behind event timing. Ecologists as well are refining their ideas to incorporate evidence from psychophysical studies (Bateson and Kacelnik, 1998).
One of the major points of progress is the realization of distinctions between circadian time and event timing. Circadian timers are temperature compensated, while event timers are not. Circadian timers are not useful for recording event intervals that deviate from the 24 hour LD cycle, whereas event timer accuracy appears to be a function of the linear increment in duration (Weber's law). Vertebrate circadian and event timers also appear to be isolated to different areas of the nervous system, but this remains to be corroborated in invertebrates. Circadian time also affects the variance in event timer responses, but not the duration. These distinctions provide us with a basis for understanding the relationship between the two mechanisms and for understanding why, in an adaptive sense, an animal is biased towards certain environmental stimuli and patterns of behavior that follow geophysical rhythms.
I have also presented possible mechanisms for event timers that are in agreement with the available evidence from psychophysics and neurophysiology. Time fields are an analogous structure to space fields, and may in fact operate by precisely the same mechanism. Decay timers are presented as an explanation for essentially all timed events (as the hippocampus probably contains decay timers, potentially in the form of ion channel conformation changes). The difference between decay timers and time fields may simply be in the number of cells used and the allocation of receptors to specific sensory cues. This difference may also explain failures to show evidence for event timers in invertebrates, as a consequence of sensory bias.
The problem with event timers is one that is shared by much of the literature on the adaptive value of cognitive mechanisms. We don't yet understand the costs of nervous tissue (Aiello, 1997). Costs associated with changes or new development of nervous tissue are still far from quantified. Genetic perturbations may be bringing us closer to this, but the synergistic nature of the nervous tissue will presumably confound our efforts for some time. In the SCN example, a portion of the brain was removed with no observable effect over a two year span. This suggests that a trait of nervous systems in general may be their plasticity. The costs and adaptive value of phenotypic plasticity, even though animal event timing is one of them, are still very much in the dark. Why not perceive everything, record perfect memories of it all, and be able to tell me to the minute (without looking at a clock) how long you've been reading this?