Evolution of amniotes

According to the current view of evolutionary relationships among jawed vertebrates, several lines of early aminiotes derived from a common ancestor during the Carboniferous in the Paleozoic era, c.a. 320 million years ago (Carrol, 1988). Amniotes therefore constitute a monophyletic group composed of synapsids (leading to mammlas), diapsids (leading to dinosaurs and birds), and anapsids (leading to extinct reptiles; the linkage to the present turtles is questioned), the classification based on the patterns of temporal openings in the skull as the critical cue. Ancestors of mammals are supposed to date back to the amniotic origin, and showed a massive diversification during the whole period of the Permian. Most groups of the primitive mammals perished at the Permian mass extinction, however, some survived, giving rise to the Triassic cynodonts. Accordingly, all of the late Mesozoic and Cenozoic mammals are supposed to have stemmed from this group. During the era of great reptiles, or dinosaurs, the cynodonts stayed relatively small in their diversity. With their small size and high metabolic activities, shrews-like ancestors survived without major changes until the dawn of the Cenozoic era.

Origin of birds

Origin of modern birds dates back to the Jurassic in the Mesozoic era, about 200 million years ago. The idea that the birds are rooted in the therapod dinosaurs has gained more and more supports from recent fossil records of common features shared by birds and therapods such as wishbone, breastbone, and feathers (Norell et al., 1997; Qiang et al., 1998). Although the intensive and earnest research activities have suffered from a fossil forgery (Zhou et al., 2002), steady lines of evidence have been accumulated for the therapod origin of the modern bird. However, consensus has not yet been reached, and an alternative hypothesis of older origin of the modern birds is still holding.

Bird brain

In accordance with the evolutionary relationships, brain of the amniotic vertebrates share many features in common. Neural organizations of subtelencephalic structures such as spinal cord, medulla oblongata, cerebellum, pons, mesencephalic, and diencephalic structures (optic tectum, tegmentum, thalamic and hypothalamic nuclei) are basically comparable wide among different classes of amniotes (Butler and Hodos, 1996). On the other hand, correspondence of telencephalic structures is much more vague, and has long been debated, seeking for the genuine homological relationships.

Traditional nomenclature

Traditional nomenclature has been used since it was summarized by Ariens-Kappers and his colleagues (1936), but now the presently used terminology has proved to be terribly misleading. For the brain atlases available to date, see Kuenzel and Masson (1988) for the domestic chick, and also see Karten and Hodos (1967) for the pigeon. According to the traditional view, most of the avian telencephalon was equated to sub-regions of the basal ganglia (or striatum) in the mammalian brain, and the nuclei were given names with “striatum” as post-fix; e.g., paleostriatum, archistriatum, neostriatum, and hyperstriatum. Actually, Golgi study (Tömböl et al., 1988a, 1988b; Tömböl, 1995) shows that cytoarchitecture of these avian telencephalic nuclei are somewhat similar to the mammalian striatum.

Genuine homologies

However, data obtained by analyses of embryonic gene expression patterns (Fernandez et al., 1998; Puelles et al., 2000), and detailed neuro-chemical examinations of transmitter and receptor types together with hodological data for neuronal connectivities (Reiner et al., 1998; Durstewitz et al., 1999), revealed that a considerable portion of these “striatal” structures have nothing to do with the mammalian striatum. Instead, structures in the dorsal telencephalon could be homologous to the mammalian cortex (Shimizu, 2001; Medina and Reiner, 2000), even though they lack laminated (layered) architecture and pyramidal neurons characteristic of the mammalian cortex. The ventro-medial telencephalic (sub-pallial) structures, on the other hand, proved to be highly conservative in their neural characters (Reiner et al., 1987; Reiner et al., 1998), thus some of them could deserve the post-fix of “striatum.” It remains still controversial as to whether the major evolutinary changes occurred at the transition from amphibians to the amniotes (Reiner et al., 1998), or at the transition from finned anamniotes to tetrapods (i.e., at the origin of amphibians) (Marín et al., 1998).

Nomenclature reform

The traditional terminology is now under a reform in the contemporary view of the evolution of telencephalon. Organized by E.D. Jarvis (Duke University, USA) and H. Karten (University of California San Diego, USA), comparative neuroanatomists formed a platform called “Avian Brain Nomenclature Exchange” (refer to the website  http://jarvis.neuro.duke.edu/nomen/). We will soon find the final form of the nomenclature report to be published, and most of the avian researchers will follow the proposal. The basic idea underlying the reform is that the inappropriate post-fix “striatum” should be removed, leaving many of the abbreviations unchanged. In this review, we will follow the traditional (and therefore incorrect) terminology, but state the homological relationships to the mammalian counterparts so far as reasonable consensus has been reached.

ARE model

Attempts have been therefore accumulated toward finding common features of imprinting shared by other learning paradigms such as sexual imprinting, operant conditioning and Pavlovian conditioning (see Hollis et al. 1991 for review). Theoretical study using an abstract neural-net model (Analysis-Recognition-Execution model, or ARE model; Bateson and Horn, 1994) has been actually successful in unifying the learning paradigms in terms of common representations shared by distinct learning processes. Different learning paradigms could be understood in terms of distinct combination and distinct changes in connectivities among the presumed sub-processes of A, R, and E. However, biological implementation of the sub-processes (such as A, R, and E) into relevant brain structures remained totally untouched.

Brain mechanisms

Due to the high tractability of ducklings, goslings or domestic chicks as experimental subjects and reproducibility of the learning, the underlying brain mechanisms have been intensively studied in terms of relevant brain regions involved, underlying neurochemical cascades, and accompanying morphometric changes in neural structures (see reviews by Horn, 1985, 1998; Bolhuis and Honey, 1998; Bolhuis, 1999). Research activities have been concentrated on a telencephalic region abbreviated as IMHV (or, intermediate medial hyperstriatum ventrale). Note that the IMHV has nothing to do with the mammalian striatum; readers are rather requested to regard the term “IMHV” as a label for a distinct brain region, instead of “a portion of ventral striatum” that is just incorrect. The IMHV was initially identified as a region where the training procedure of imprinting selectively enhanced the uptake of radio-active uracil (Horn et al., 1979) and also of radioactive 2-Deoxyglucose (Kohsaka et al., 1979). Hodological (tract-tracing) study (Bradley et al., 1985) revealed that the IMHV have reciprocal connections wide with telencephalic structures that include hyperstriatum accessorium (area analogous to the primary visual cortex in mammals; not a “striatal” region) and archistriatum (a complex of structure analogous to the limbic and somato-motor cortices in mammals; also see below), suggesting that the IMHV could function as a site for association of signals issued from multimodal sensory inputs (see also Durstewitz et al., 1999).

Acquisition and retention

Localized lesions placed in the bilateral IMHV (i.e., IMHV regions in both right and left hemispheres) actually proved to prevent the chicks from successful learning in the imprinting paradigm (McCabe et al., 1981); therefore, IMHV is necessary for acquisition. When the IMHV was lesioned soon (within 3 hr) after the imprinting training, on the other hand, the ablated chicks also showed significantly less selective approaches at test accomplished at 24 hr post-training (McCabe et al., 1981); the IMHV is also necessary for retention at least for several hr after the end of training. When the bilateral lesions were made much later (6 hr or afterwards), however, the ablated chicks showed selective approaches at test; the IMHV is no longer required for recall (McCabe et al., 1982).

Permanent and transient storages

In a further series of sequential unilateral IMHV lesions (i.e., the right IMHV was ablated, and subsequently the left IMHV was lesioned, or vise versa), functional laterality has been shown in the involvements of the IMHV in memory formation (for detailed review, see McCabe, 1991). Briefly, the left IMHV is supposed to be a long-term storage site for the imprinting memory, whereas the right IMHV acts as a buffer storage (Cipolla-Neto et al., 1982; also see Bolhuis and Honey 1998). The right IMHV is required for another memory trace to be formed outside of the bilateral IMHV with a considerable delay (6 hr or longer). In other words, the memory traces are supposed to be represented in multiple brain regions, and copies are subsequently delivered from the right IMHV to other regions. The memory trace stored outside of the IMHV is referred to as S′ [S-dash], although its location has not been identified so far. Since the IMHV was assumed to be the major storage site of permanent memory, studies on the neural basis of imprinting have been concentrated on IMHV.

Recently, Nicol, Horn and their colleagues have been successful in analyzing single neuron activities in freely-behaving chicks both during and after the imprinting training (Nicol et al., 1995, 1998; Horn, 1998; Horn et al., 2001). Population of neural correlates of the imprinting object (such as coding of the color and the shape) in IMHV increased as the training proceeded, thus yielding direct evidence for the IMHV as a constituent of the memory system. For the system-level understanding of imprinting, however, we must specify what aspects of behavioral execution the IMHV is responsible for.

Recognition of occluded image and biological motion

Imprinting has been useful also in revealing the cognitive capability of chicks. Selective approaches toward partly occluded imprinting object have suggested that the chicks can utilize the partial visual features for recognition (Regolin and Vallortigara, 1995). Further analysis of orienting behaviors toward a hidden imprinting object has successfully shown that the chicks can maintain the location of invisible (hidden) object for up to 3 min (Vallortigara et al., 1998), similarly to the delayed matching-to-sample task. Object permanence and working memory have, however, not been proved unequivocally in the chick. Chicks can also recognize the imprinting object by its biological motion, or point-light animation sequences depicting a walking hen (Johanson's biological motion; Regolin et al., 2000). All these facts suggest a high degree of similarity in the capability in visual cognition between newly-hatched chicks and humans.

Taste aversion

Passive avoidance learning has some features common with the taste aversion learning (Mazur, 2002); in both cases, the memory is formed only after one-trial experience of association, and the chick learns to avoid the object. However, these two paradigms can be clearly distinguished. In the taste aversion, the subject animal was given a food, and subsequently an intra-peritoneal injection of a LiCl solution that makes the animal feel ill several hr afterwards. The taste-aversion is assumed to represent a case of classical conditioning, with the food acting as a conditioned stimulus, and the illness as an unconditioned stimulus. However, the taste-aversion does not require a strict contingency of events to be associated; the induced illness causes the subject to recall the characteristics of food that was ingested several hours previously. In the passive avoidance, on the other hand, a strict temporal contingency is required between pecking the bead and tasting the bitter liquid; with a delayed delivery of the bitter liquid by only 5 min, chicks failed to form the avoidance memory (M. Aoki and T. Matsushima, unpublished data).

Fitness gains of the chick's high performance in this task could be that the capable chicks have higher chance of survival because they can avoid bitter-tasting, therefore possibly poisonous objects. This argument is however questionable because of the following reasons. First, the bitter taste does not necessarily mean a poisonous food; the taste-aversion paradigm would be much more adaptive in this context. Second, the avoidance memory quickly generalizes within 24 hr after the training, so that the learned chicks would have a risk of avoiding even edible food items with slightly different colors (Aoki et al., 2000).

Common brain mechanisms

The underlying brain mechanisms for the formation of passive avoidance memory have been intensively studied in terms of neurochemical and morphological correlates (see reviews by Rose, 1991, 1995; Rose and Stewart, 1999). Rose's research strategy has been to identify specific changes at the molecular level, which have direct correspondence with the memory formation (Rose, 1993). Most importantly, time course of the changes must be compatible with development of the learning. The passive avoidance task is appropriate in this context, because the memory is established in single and short training trial. Experimenters can thus identify the exact instant when the memory is formed; in the imprinting paradigm, on the other hand, chicks are exposed to the imprinting object for a couple of 1-hr-long training sessions.

Neurochemical approaches

Again, the IMHV proved to be involved in the passive avoidance (Rose and Csillag, 1985; Davies et al., 1988). Learning-specific permanent changes were identified also in another brain region referred to as LPO (lobus parolfactorius) (Stewart et al., 1987; Csillag, 1999). The LPO constitutes the medial part of striatum, that is homologous to a complex of caudate-putamen / nucleus accumbens in mammals. In the IMHV, enhanced metabolic activities immediately after the training leads, through enhanced expression of immediate early genes (c-fos, but also see Yanagihara et al. (2000) for ZENK or zif/268) and expression of late response genes such as those coding cell adhesion molecules (NCAM and L1), to morphological changes in both pre- and post-synaptic structures. The permanent changes in LPO include; increase in the length of thickening of the post-synaptic density (indicative of the active zone) (Stewart and Rusakov, 1995), and enhanced neurogenesis in the post-hatch and post-training period (Dermon et al., 2002). Although not all of these events have been fully understood in their functional roles, the cellular / molecular studies proved to be extraordinarily fruitful when applied to such a simple association learning as passive avoidance.

Underlying cognition

For system-level understanding, however, the passive avoidance task fails to give us few clues for elucidating the neural representations. First, chicks are trained once, and tested once for recall; any neuronal activities recorded in single trials cannot be a basis for reliable functional analyses. Second, memory contents of chicks are too much simplified; chick is either recalling (successfully avoiding the bead) or amnestic (pecking at the bead), without telling how the chick recognized the aversive bead.

Right or left

Contemporary researches on the spatial memory in chicks emerged from a psychological study on the right-left asymmetry in position learning, indicative of a functional lateralization of telencephalic hemispheres (Vallortigara and Zanforlin, 1986; Vallortigara et al., 1988). Basically, the subject chicks (1-2 weeks post-hatch) are tested in a rectangular arena, the front wall of which is equipped with a pair of food container boxes. Chicks are introduced from the entrance on the other side of the arena, approach to the boxes, and are requested to peck either one of these two boxes; pecks at the correct box is immediately rewarded by an opening of the box for chicks to gain the food inside (Vallortigara et al., 1996).

Taking an advantage of biased preferences for food items, it has been shown that the chicks memorize both of the content (“what” information) and the position (“where” information) of the boxes (Cozzutti and Vallortigara, 2001), reminiscent of the “episodic-like” memory in jays (Clayton & Dickinson, 1998). Briefly, under a control condition, chicks approaches to the box of their preferred food. When the chicks were fed sufficiently with the preferred food, the satiated chicks would re-orient to the other box, presumably due to the reduced attractiveness of the over-fed food; this process is referred to as “devaluation.”

Center of a place

Further elegant experiments developed by the same group of Italian psychologists revealed that chicks adopt two distinct strategies in spatial localization (Tommasi et al., 1997). In this paradigm, chicks were trained to find a food item hidden at the center of a training arena. The food was initially placed on the surface, and subsequently hidden in the sawdust on the floor. By simply observing the locations where the subject chick scratched the floor in a test arena, experimenters could study how the chick localized the center. The trick is that the test arena differed from the training arena in either the size (with the shape being identical) or the shape (with the size being identical). In order to localize the center, the chicks could utilize either the absolute distance from one wall (local absolute cue), or depend on the equal distance from both of the opposing walls (global relational cue) (Tommasi and Vallortigara, 2000). Surprisingly, the right and the left telencephalic hemispheres differed in localizing strategies; chicks with the operational right hemisphere (with its right eye covered by eye-patch) adopted the global cue, whereas chicks with the left hemisphere (with its left eye covered) searched for food based on the local cue (Tommasi and Vallortigara, 2001). Further unilateral lesion of hippocampus suggested that the global and local cues are separately stored in the right and left hippocampi, respectively (Tommasi et al., 2003).

Position as supplementary cue for association

Position could serve an important cue for the chicks, which depend on seeds and grains scattered unevenly in their foraging ground. In a reinforced concurrent choice task, quail chicks proved to recognize beads primarily by color, and secondarily by position (N. Aoki and T. Matsushima, unpublished); the positional cue appeared operational only when the color cue was no longer available. It will be extremely interesting to see if the IMHV-lesioned chicks (therefore, possibly color-blind subjects; see below) could discriminate objects by the second supplementary positional cues. So far, on the other hand, color-cue dependent object discrimination proved to remain intact in the domestic chicks with bilateral hippocampal lesions (S. Nakajima and T. Matsushima, unpublished data), suggesting a possible double dissociation of neural representations of color and position.

Color map

With their tetra-chromatic nature of the retinal cone photoreceptor cells (ultra-violet, blue, green, and red; Bowmaker et al., 1997), domestic chicks are supposed to be endowed with acute sense of colors. Visual discrimination task with food reinforcement actually demonstrated that domestic chicks have accurate color memory for foraging (Osorio et al., 1999); the pattern proved much less significant. Basically identical conclusion was drawn in the quail chicks, in which visual memory was examined by selective habituation and passive avoidance task (Aoki et al., 2000). It is to be emphasized that chicks have a context-independent representation of colors. Subjective distance of a green measured from memorized image of a yellow was identical to the distance of the yellow from the green image (Aoki et al., 2000); quail chick could have an internal color map as reference.

Genetic basis of color preference

Experimental manipulations of color perception must be carefully accomplished, because the chicks have innate preference to specific colors and the preference is genetically determined (Kovach, 1980). With traditional genetic selection, Kovach established several lines of quails with innate blue- and red-preference (blue- and red-line). Furthermore, quail chicks can be imprinted to the color opposite to their original preference by simply exposing the subject chicks to the color (Kovach, 1990). It is to be noted, however, that the genetically determined color preference reflects a selective choice for shorter (blue-line) or longer (red-line) wavelength, respectively. When confronted with a concurrent choice between yellow and green, chicks of the blue-line chose green over yellow; in the red-line, on the other hand, the same test revealed yellow preference over green. Innate color preference thus could represent a process, which is distinct from that involved in the color map based discrimination (Aoki et al., 2000).

Neural basis of color discrimination

In parallel with the two distinct processes of color discrimination, two relevant brain regions have been pointed out; a telencephalic region (IMHV) and a subtelencephalic region (dorsomedial thalamus). In a series of lesion experiments in passive avoidance task in domestic chicks, it has been shown that a post-training lesion placed to bilateral IMHV failed to cause amnesia (Gilbert et al., 1991), in contrast to the pre-training lesion experiment (Davies et al., 1988); it was thus concluded that the IMHV is required for acquisition, but not for recall, reminiscent to the functional involvement of the right IMHV in the imprinting (see above; Cipolla-Neto et al., 1982). Further examination of the post-lesion effects revealed, however, that the lesioned chicks avoided the bitter-tasting bead by some (yet unidentified) non-color cues (Patterson and Rose, 1992); memory-based color discrimination was selectively impaired. In contrast, lesion experiments on the innate color preference revealed that even total telencephaloectomy (the whole telencephalon aspirated on the hatching day) does not impair posture, sensorimotor coordination for pecking, locomotion, and selective approach to the genetically preferred color (Kovach and Kabai, 1993). Much smaller lesion localized in the dorsomedial thalamic complex proved to attenuate the genetically determined color preferences (Csillag et al., 1995); lesions to an ascending visual pathway (nucleus rotundus) failed to have effects. Most probably, color is multiply represented in the chick brain, with distinct controls over the behavioral executions.

Shapes

Objects might also be recognized by the shape cue. Actually, the domestic chicks with bilateral IMHV lesions were successful in avoiding the bitter bead by non-color cue(s) as has been described above (Patterson and Rose, 1992); the shape cue was supposed the most plausible candidate for discrimination. However, to date, even intensive examinations failed to reveal the chicks' ability to discriminate objects by shapes in quail chicks (Sakai et al., 2000; Ono et al., 2002). Our inability to reveal the shape recognition might reflect the ecological situation of chicks, which do not depend on the food shape for selective foraging. Another study of visual behavior in the quail chicks (Hayashi et al., 2001) suggested the chick's capability to discriminate conspecific hatchlings by fine plumage patterns; biological motion might be another cue as has been shown in imprinting (Regolin et al., 2000).

Reinforcement color discrimination task

Housed in an experimental chamber, the subject chick was presented with a bead (Fig. 1A). The bead was protruded from a hole on a wall for a short period of time (2-4 sec cue-period). The bead was colored either in red, green or blue. When a red bead was presented, chick was required to peck at the bead, and food reward was subsequently delivered after a short delay (2–4 sec reward-period after a 1-sec delay). Red was thus associated with a delayed reward via pecking as operant (rewarded GO). When a green bead was presented, on the other hand, chick was required NOT to peck in order to be rewarded (rewarded NOGO). When a blue bead was presented, chick learned not to peck, because reward was not delivered irrespectively of whether the chick pecked or not (non-rewarded NOGO). This is the basic configuration of the task designed and developed by Yanagihara et al. (2001). In this task, we can clearly dissociate the overall procedure into distinct phases, i.e., perception of color, recall of association memory, execution of operant pecking, anticipation of reward during the delay, recognition of food item, execution of food pecking, and finally ingestion of food.

Fig. 1

Behavioral paradigms developed for the study of cognitive processes underlying foraging behaviors in the domestic chick. A: In the single choice task, the subject chick was presented with a single cue bead, and required either to peck or to stay not pecking. The chick must memorize the association between the color cue (red, green or blue) and the required operant (peck or not peck), as well as the association between the cue and the consequence (i.e., whether a reward is delivered or not). In this manner, sub-processes (such as cue recognition, reward anticipation, and operant behavior) could be experimentally dissociated (Yanagihara et al., 2001). B: In the concurrent choice task, the subject chick was presented with a pair of cue beads, and required to peck either one of them. The consequences (quantity and proximity of rewards) depend on the choice. The chick must memorize the associations as in the case of single choice task. At execution, the chick should recall the associated rewards, anticipate the consequence for each choice, compare them, and finally make a choice. In this manner, anticipation of the temporal proximity of the reward could be experimentally dissociated from that of the reward quantity (Izawa et al., 2003).

Single neuron as a “pin-hole”

By a miniature micro-drive mounted on the chick skull together with FET-input buffer amplifiers, we obtained stable extracellular recording of action potentials (spikes) from single neurons continuously for up to 6–9 hr. But, what could a neuron tell us? What do we know by analyzing spiking behaviors of a single neuron, that is truly a “microscopic” entity among millions of similar cells in the whole brain? A simple rationale behind the single-unit analysis could be that we observe the whole brain system through the neuron as a “pin-hole.” Assume that a neuron is connected with a network. We search for positive correlation of the neuronal firing with various behavioral events, and fortunately find a link. For example, the neuron fires in response to a stimulus (light or buzzer) that is given in advance to delivery of reward (food or water). One interpretation is that the neuron codes the “memory-based anticipation of the forthcoming reward.” If we could dissect out the neuron under study and put it in a culture dish, however, the neuron might generate a regular pattern of spikes in isolation, but the spiking would tell us nothing about “anticipation” any longer. Without that neuron, on the other hand, rest of the whole brain would “anticipate” the reward by the associated stimuli, due to redundant organization of the brain. The link between spikes and the code is not an attribute to the neuron; instead, it is an attribute to the whole brain system. For the observer, the neuron operates as a “pin-hole,” and the projected “image” represents the whole relevant process viewed through that neuron. One recorded neuron produces one “image,” and thousand simultaneous neuronal recordings give rise to thousand “images” of the single brain. Thus, our job is to synthesize the brain performance from these thousand of “images.”

Memory correlates in IMHV

Neuronal “pin-hole images” of memory should meet at least the following criteria. First, changes in the neuronal spikes (excitatory or inhibitory) should occur in response to the presentation of associated stimuli in a specific manner. Second, the responses should emerge only after relevant training, showing a good parallelism with the memory retention. Neuronal spikes that meet these criteria have been found in the chick brain in several different regions including the IMHV. Using the imprinting paradigm, it has been established that IMHV contains neuronal correlates of the imprinting memory (Nicol et al., 1995, 1998; Horn et al., 2001). Some IMHV neurons responded specifically to a combination of color AND shape, whereas others showed generalized responses to color OR shape. These authors argued that the IMHV neurons principally represent stored visual features of the imprinting object.

Code of attention in IMHV

In our reinforcement learning paradigm, on the other hand, IMHV neurons responded to wider range of objects, such as rewarding colors as well as novel colors. When habituated, presentation of a familiar color failed to elicit any responses (E-I Izawa et al., 2000). Most probably, these IMHV neurons are related to the chick's subjective “attention,” or what appears conspicuous to the chick. Note that the “attentions” should be generated only after specific experience with the rewarding colors. Similarly, the novelty responses should appear only after the chick had experienced, memorized and recalled a finite number of colors. In this sense, these responses represent the memory, and our interpretation of the IMHV as “attention-generator” fits the memory trace hypothesis. A possibility is not excluded that the memory trace resides somewhere outside of the IMHV, and the IMHV neurons responded secondarily. To localize the memory trace, therefore, we should systematically survey the brain regions that are interconnected with the IMHV; these candidate regions include, e.g., visual Wulst in the dorsal pallium, Arch (arch-istriatum), and LPO.

Anticipation code and “paradox” in LPO

Survey of the task-related neuronal activities in the LPO revealed two important populations of neurons. One group of neurons fired specifically to the visual cues associated with the reward, i.e., those responded in the cue- / delay-periods in both of the rewarded GO and rewarded NOGO trials, but not in the non-rewarded NOGO trials (Yanagihara et al. 2001; E-I Izawa and T Matsushima unpublished). Most probably, these neurons code memory-based anticipation of the forthcoming food reward. Another group of neurons fired when the chick actually gained a reward; a subset of these neurons fired irrespective of whether food or water was gained. Neurons of the second group might represent chicks' subjective evaluation of the gained reward.

What do these codes (anticipation and evaluation) do in the behavioral execution? The evaluation code could be responsible for the formation of novel cue-reward associations. If this were the case, localized LPO lesion should result in an acquisition failure (anterograde amnesia). Otherwise, the anticipation code could be responsible either for selective execution of color-selective pecking. If this were the case, LPO lesion should result in a recall error (retrograde amnesia). In a series of lesion experiments (Izawa et al., 2001, 2002, 2003), we analyzed the effects of pre-training and post-training lesions on a variety of learning paradigms in the domestic chicks; i.e., filial imprinting, passive avoidance learning, water-reinforced color discrimination task, food-reinforced GO-NOGO task, and food-reinforced concurrent choice task. For the imprinting, both of pre- and post-training lesions caused no effects. For the reinforcement learning, similar LPO lesions caused severe deficiency in the acquisition, whereas the learned associates were recalled without difficulties. Therefore, the evaluation code in LPO could actually have a role. For the GONOGO color discrimination task, similarly, the LPO lesion caused anterograde amnesia, but no retrograde amnesia. Here arises a “paradox.” The anticipation code is formed in LPO after the training, however, the LPO does not seem to be required for execution of selective pecking. Without the memorized code, how could chicks execute the correct operant pecking?

Impulsiveness and behavioral tolerance

One possible way to account for the “paradox” is to assume that the anticipation code in LPO is responsible for some other functions than behavioral execution per se. Alternative account for the “paradox,” though not exclusive, is that the anticipation code is multiply represented in various regions of the brain, and the lesion localized in LPO failed to interfere with the link between the color and pecking.

In concurrent choice task (Fig. 1B), post-training LPO lesion actually had an effect. In this task, chicks learned to peck one of two simultaneously presented beads, e.g., red and yellow. Red bead was associated with a large reward (6 pellets of millet grain), and yellow with a small reward (1 pellet). Naturally, chicks learned to peck the red bead to gain the large reward. The choice differed when the large reward was delivered with a delay of 1–3 sec. When the red bead was associated with a large but delayed reward (delay time of 3 sec), the chick learned to choose the yellow bead to gain the small reward. For a delay of 1–2 sec, chicks proved to be patient enough to wait for the large reward, just staring at the still empty food tray. Chemical lesion of LPO (particularly the posterior LPO) ablated the chick-sized patience, thus unmasking the underlying impulsiveness; no amnesia accompanied the lesion. With the anticipation code in LPO neurons, the future gain would be “guaranteed” so that the chick reasonably suppressed the impulsive action for immediate reward. In other words, past experiences yield an internal representation of the future reward in LPO, and the represented reward makes the chick behaviorally tolerant. At the neuronal level, we can further assume that the anticipation-coding neurons in LPO could be responsible specifically for temporal proximity, rather than quantity and quality of the reward; the quantity and quality of anticipated reward should be represented somewhere else in the chick brain. Note that the system for the patient choice is highly adaptive, because the net gain (i.e., total amount of food obtained) could be optimized, thus serving for a rapid growth and a higher chance of survival (see Table 1).

Alternative code of anticipation in AIv

Archistriatum located in the ventro-lateral telencephalon might be in charge. In particular, the AIv region (a ventral subdivision of intermediate archistriatum) has reciprocal connections with IMHV, and a massive efferent projection to LPO (Székely et al., 1994), thus could play an important role as relay center for the memory formation (Davies et al., 1997; Csillag, 1999). Actually, localized lesions to the Arch are reported to prevent memory formation in passive avoidance learning (Lownders and Davies, 1994) as well as in imprinting (Lownders et al., 1994). The “paradox” of LPO functions (see above) might thus be explained by assuming an alternative code of cue-reward association in the Arch. Actually, single-unit recordings revealed a population of Arch neurons that responded specifically to the cues associated with reward (Aoki et al., 2002, 2003). We should examine whether these Arch neurons code aspects of the anticipation, i.e., quality and / or quality of the forthcoming reward, rather than temporal proximity.

“Observation learning”: a social transmission

Chicks could learn also by observation. In addition to the own experiences of pecking and tasting as described above, the pecking preference can be socially transmitted from hens to day-old chicks (Suboski and Bartashunas, 1984). Even a motor-driven arrow-shaped paper model, that moved its taper pointing to a colored bead, could tell a chick which object to peck. The chick subsequently pecked at the “instructed” bead object, even after the arrow-operation was removed. Authors argued that “information about the visual characteristics of food objects” could be transmitted from hens to chicks by the same process. Similar transmission of pecking selectivity is reported in the one-trial passive avoidance (Johnston et al., 1998). Just by observing another individual pecking at a bitter bead, and subsequently showing disgust responses, day-old subject chicks learned not to peck at the same bead when tested afterwards. This finding is reminiscent of the finding in monkeys, in which a lasting phobia of snakes developed by observing another individual's fearful reactions to a snake (Mineka, Davidson, Cook and Carr, 1984; cited by Mazur, 2002). Beside the well-documented effects of social context (being observed by other individuals) on the re-cashing behavior in scrub jays (Emery and Clayton, 2001), chicks might also be endowed with a high ability to actively learn by observations. Development of a novel paradigm tractable for system neuroscience will enable us to penetrate into many interesting issues, such as how chicks observe others, how chicks convert the observed events into own behavioral rules, and what neural mechanisms are responsible for the conversion.

Deviations from optimal foraging: “naïve curiosity,” “contra-freeloading,” “Concorde fallacy,” and “altruism”

Chicks might be wise enough to actively “earn” information at the expense of immediate material benefits. In our controlled laboratory condition, week-old chicks are trained and tested under a limited diet so that the chick's motive toward food reward is maximized. Consequently, chicks quickly learn the association between cue colors and reward quantity, so that chicks reliably choose a color associated with a larger reward (Izawa et al., 2003). In this context, chicks behave in accordance with the most normative theory of the optimal foraging (Alcock, 2001) with a slight modification that anticipated reward in the future should weigh proportionately less than the immediate gain. Internal representation of the anticipated future plays a critical role.

The situation somewhat differs in day-old chicks. They are much more curious, pecking non-selectively at a variety of conspicuous objects they encounter. Within the initial 3–4 days post-hatch, chicks survive by the yolk reserve and do not depend on food. During this period, chicks have to make up an internal directory of edible foods and non-edible objects of similar but distinct appearance such as gravels or ground debris. “Naïve curiosity,” or an eagerness for information in the limited post-hatch period could play a critical role, serving a biological basis for the passive avoidance learning and the reinforcement tasks.

Similar deviation from the immediate optimization can be found in adult birds, which often work (i.e., pay behavioral “cost”) for food even when the same food items can be freely available; a process known as “contra-freeloading” (Inglis et al., 1997). The “contra-freeloading” has been reported in a variety of vertebrate species, including fish (Betta splendens), pigeons, domestic chicks, crows, star-lings, rats, monkeys, chimpanzee, and humans. In this context, it is argued that animals have “a hunger for information,” and a more information gain could offset the extra cost to be paid now, so long as the immediate need for food is not so great.

In European starlings, it is further reported that the cost that had been paid for gaining food reward increased the preference in choice condition (Kacelnik and Marsh, 2002), in a clear contrast to the consequence predicted by the optimization theory. The authors claim that they can relate their finding of the behavioral “perversity” in birds to a phenomenon known as “Concorde fallacy,” in which a behavioral choice is biased toward a recipient of big efforts in previous history, just as the maladaptive investments by developers to the supersonic airplanes Concorde that simply did not pay. Though it is difficult to separate the effects of investment in the past and the effects of anticipated gain in the future, a plausible explanation is that the past record of investment is a reliable measure for estimation of future gains in most of the ecologically realistic circumstances, and the fallacy could represent a maladaptive side effect.

Some cases of “altruism” could constitute still another example of deviation from the optimal foraging. When an indirect fitness gain is available, animals often invest material benefits to genetically related individuals as has been demonstrated in the Florida scrub jays (Woolenfenden 1974, cited by Wilson, 1975). The choice by helpers in this context is the one between giving food to others and ingesting it by oneself. We can assume a similar proximate mechanism for the “altruistic” choice, to the one found in the anticipation codes of chick LPO. The scrub jay helpers could suppress the option of own ingesting, probably after developing an internal representation of the benefits available by the alternative option of giving. Future researches by system neuroscience might be successful in revealing the internal representation, a mental representation comparable to our ethical self-control or the Freudian super-ego.

In summary, evolution of these behavioral variations such as “naïve curiosity,” “contra-freeloading,” “Concorde fallacy,” and “altruism” should be examined, in concert with the accounts by behavioral ecology, toward understanding the responsible brain mechanisms as targets of the selection pressures. Definitely, the telencephalic structures (limbic system and striatal complex) involved in cognitive processes (memory, evaluation, anticipation, and decision making) should be the sites for the future researches. These processes could be understood as deviations from the gain optimization, rather than assuming distinct centers of “instincts.” The issue of “animal mind” could be agued most fruitfully, if approaches of the system neuro-science are thus synthesized with the evolutionary perspectives.