Who Needs Emotions The Brain Meets the Robot - Fellous & Arbib Part 6 ppt

On the neural side, the criteria for inclusion of brain areas in the limbic system remain undefined, and evidence that any limbic area e.g., the amygdala, which we will discuss below, ho

is a mammalian specialization: other vertebrates have a primordial cortex, but only mammals were believed to have a neocortex Since thinking, rea-soning, memory, and problem solving are especially well developed in mam-mals, particularly in humans and other primates that have relatively more neocortical tissue, it was argued that these cognitive processes must be mediated by the neocortex and not by the old cortex or other brain areas In contrast, the old cortex and related subcortical ganglia form the limbic sys-tem, which was said to mediate the evolutionarily older aspects of mental life and behavior, our emotions In this way, cognition came to be thought

of as the business of the neocortex and emotions of the limbic system The limbic system theory ran into trouble when it was discovered, in the mid-1950s, that damage to the hippocampus, the centerpiece of the lim-bic system, led to severe deficits in a distinctly cognitive function, episodic long-term memory (Scoville & Milner, 1957) This was incompatible with the original idea that the primitive architecture of the limbic system, espe-cially of the hippocampus, was poorly suited to participate in cognitive functions (MacLean, 1949, 1952) Subsequently, in the late 1960s, it was discovered that the equivalent of mammalian neocortex is present, though rudimentary, in nonmammalian vertebrates (Nauta & Karten, 1970) As a result, the old/new cortex distinction broke down, challenging the evolu-tionary basis of the assignment of emotion to the limbic system and cogni-tion to the neocortex (Swanson, 1983)

The limbic system itself has been a moving target Within a few years after the inception of the theory, it expanded from the original notion of

“old cortex” and related subcortical forebrain nuclei to include some areas

of the midbrain and even some regions of the neocortex Several attempts have been made to salvage the limbic system by defining it more precisely (Livingston & Escobar, 1971; Isaacson, 1982; Swanson, 1983) Neverthe-less, after half a century of debate and discussion, there are still no agreed-upon criteria that can be used to decide which areas of the brain belong to the limbic system Some have suggested that the concept be abandoned (LeDoux, 1987, 1991; Kotter & Meyer, 1992)

In spite of these difficulties, the limbic system continues to survive, both

as an anatomical concept and as an explanation of emotions, in textbooks, research articles, and scientific lectures This is in part attributable to the fact that both the anatomical concept and the emotional function it was supposed to mediate were defined so vaguely as to be irrefutable For ex-ample, in most discussions of how the limbic system mediates emotion, the

meaning of the term emotion is presumed to be something akin to the com-mon English-language use of the term, which is to say feelings However, the common English use of emotion is at best a poor theoretical notion, for

emo-tion is a rich and complex theoretical concept with many subtle aspects, some

of which are nonintuitive and thus inconsistent with the common use of the term (Ekman & Davidson, 1994; LeDoux, 1996) On the neural side, the criteria for inclusion of brain areas in the limbic system remain undefined, and evidence that any limbic area (e.g., the amygdala, which we will discuss below), however defined, contributes to any aspect of any emotion has been claimed to validate the whole concept Mountains of data on the role of limbic areas in emotion exist, but there is still very little understanding of how our emotions might be the product of the limbic system

Particularly troubling is the fact that one cannot predict, on the basis of the original limbic theory of emotion or any of its descendants, how specific aspects of emotion work in the brain The explanations are all post hoc Nowhere is this more apparent than in recent work using functional imag-ing to study emotions in the human brain Whenever a so-called emotional task is used and a limbic area activated, the activation is explained by refer-ence to the fact that limbic areas mediate emotion When a limbic area is activated in a cognitive task, it is often assumed that there must have been some emotional undertone to the task We are, in other words, at a point where the limbic theory has become a self-contained circularity Deference

to the concept is inhibiting creative thought about how mental life is medi-ated by the brain

Although the limbic system theory is inadequate as an explanation of the specific brain circuits of emotion, MacLean’s original ideas are quite interesting in the context of a general evolutionary explanation of emotion and the brain In particular, the notion that emotions involve relatively primi-tive circuits that are conserved throughout mammalian evolution seems right

on target Further, the idea that cognitive processes might involve other cir-cuits and might function relatively independently of emotional circir-cuits, at least in some circumstances, also seems correct These general functional ideas are worth retaining, even if we abandon the limbic system as a structural theory of the emotional brain They also may be key in other areas of inves-tigation of emotion, such as artificial intelligence

ESCAPING THE LIMBIC SYSTEM LEGACY: FEAR CIRCUITS

The limbic system theory failed in part because it attempted to account for all emotions at once and, in so doing, did not adequately account for any one emotion A more fruitful strategy is to take the opposite approach and study one emotion in detail Our own approach has focused on the study of fear, but the basic principles that have been uncovered about the fear system are likely to be applicable to other systems Different brain circuits may be involved in different emotion functions, but the relation of specific emotional

processing circuits to sensory, cognitive, motor, and other systems is likely

to be similar across emotion categories Some progress has also been made

in understanding emotions other than fear, as will be discussed below The neural system underlying fear has been studied especially in the context of the behavioral paradigm called “fear conditioning” (Blanchard, Blanchard, & Fial, 1970; Davis, 1992; Kapp, Whalen, Supple, & Pascoe, 1992; LeDoux, 1996, 2000; Fanselow & LeDoux, 1999) In this work, the fear system has been treated as a set of processing circuits that detect and respond

to danger, rather than as a mechanism through which subjective states of fear are experienced Measurable correlates of fear include blood pressure changes, freezing responses, and release of pituitary–adrenal stress hormones Through such measurements, fear is operationalized, or made experimen-tally tractable Some limbic areas turn out to be involved in the fear system, but the exact brain areas and the nature of their involvement would never have been predicted by the limbic system theory This operationalization of emotion may also lead to interesting work in robotics The understanding of the processing circuits that detect and respond to danger can be used to design new types of sensor, effector, and controlling device that together would make up an “operationally fearful” autonomous robot The general question

of the role of fear and its complex interactions with cognition and with other emotional circuits can then be addressed explicitly in the fully controlled and measurable environment of the robot and can potentially give insight into the role of fear in humans and other animals

Before describing research on fear in detail, several other approaches to the study of emotion and the brain that will not be discussed further should

be mentioned One involves stimulus–reward association learning (Aggleton

& Mishkin, 1986; Everitt & Robbins, 1992; Gaffan, 1992; Ono & Nishijo, 1992; Rolls, 1998), another involves the role of septo–hippocampal circuits

in anxiety (Gray, 1982), and still another involves distinct hypothalamic and brain-stem circuits for several different emotions (Panksepp, 1998; Siegel, Roeling, Gregg, & Kruk 1999)

What Is Fear Conditioning?

Since Pavlov (1927), it has been known that an initially neutral stimulus (a conditioned stimulus, or CS) can acquire affective properties upon repeated temporal pairings with a biologically significant event (the unconditioned stimulus, or US) As the CS–US relation is learned, innate physiological and behavioral responses come under the control of the CS (Fig 4.1) For example,

if a rat is given a tone CS followed by an electric shock US, after a few tone–

shock pairings (one is often sufficient), defensive responses (responses that

typi-cally occur in the presence of danger) will be elicited by the tone alone Ex-amples of species-typical defensive responses that are brought under the con-trol of the CS include behaviors such as freezing in rodents and autonomic (e.g., heart rate, blood pressure) and endocrine (e.g., hormone release) re-sponses, as well as alterations in pain sensitivity (hypoanalgesia) and reflex expression (fear-potentiated startle and eye blink responses) This form of conditioning works throughout the phyla, having been observed in flies, worms, snails, fish, pigeons, rabbits, rats, cats, dogs, monkeys, and humans

Research from several laboratories combined in the 1980s to paint a relatively simple and remarkably clear picture of the neuroanatomy of fear conditioning (Davis, 1992; Kapp, Whalen, Supple, & Pascoe 1992; LeDoux, 1992; Fanselow & Gale, 2003) In such studies, the CS and US are typically

an audible tone and a foot shock, and the responses measured include freez-ing It was shown that fear conditioning is mediated by the transmission of information about the CS and US to a small almond-shaped area (the

Figure 4.1 (A) Fear conditioning involves the presentation of a noxious

unconditioned stimulus (US), such as footshock, at the end of the occurrence

of a neutral conditioned stimulus (CS), such as a tone (B) After conditioning,

the CS elicits a wide range of behavioral and physiological responses that characteristically occur when an animal encounters a threatening or fear-arousing stimulus Thus, a rat that has been fear-conditioned to a tone will express the same responses to a CS as to a natural threat (e.g., a cat) (Adapted

Conditioned stimulus

(tone)

Unconditioned stimulus

(footshock)

Time

Natural threat

or Conditioned stimulus

Defensive behavior Autonomic arousal Hypoalgesia Reflex potentiation Stress hormones

A

B

amygdala) and the control of fear reactions by way of output projections from the amygdala to behavioral, autonomic, and endocrine response control sys-tems located in a collection of nuclei, altogether referred to as the “brain stem.” We briefly describe below the input and output pathways, as well as the connections within the amygdala The focus will be on findings from rodents and other small mammals as most of the work on fear conditioning has involved these species

The amygdala consists of approximately 12 different regions, each of which can be further divided into several subregions Although a number of different schemes have been used to label amygdala areas (Krettek & Price, 1978; Amaral, Price, Pitkanen, & Carmichael, 1992), the scheme adopted

by Amaral et al (1992) for the primate brain and applied to the rat brain by Pitkanen et al (1997) will be followed here The areas of most relevance to fear conditioning include the following nuclei: lateral (LA), basal (B), acces-sory basal (AB), central (CE), and intercalated (IC), as well as connections between them Studies in several species, including rats, cats, and primates, are in close agreement about the connections of LA, B, AB, and CE (Amaral, Price, Pitkanen, & Carmichael, 1992; Paré, Smith, & Paré, 1995; Pitkanen, Savander, & LeDoux, 1997; Paré, Royer, Smith, & Lang, 2003) In brief,

LA projects to B, AB, and CE and both B and AB also project to CE; IC is also an intermediate step between LA/B and CE However, it is important

to recognize that the connections of these areas are organized at the level of subnuclei within each region rather than at the level of the nuclei them-selves (Pitkanen, Savander, & LeDoux, 1997) For simplicity, though, we will for the most part focus on nuclei rather than subnuclei

The pathways through which CS inputs reach the amygdala have been studied extensively in recent years Much of the work has involved the au-ditory modality, which is focused on here Auau-ditory and other sensory in-puts to the amygdala terminate mainly in LA (Amaral, Price, Pitkanen, & Carmichael, 1992; Mascagni, McDonald, & Coleman, 1993; Romanski & LeDoux, 1993; McDonald, 1998), and damage to LA interferes with fear conditioning to an acoustic CS (LeDoux, Cicchetti, Xagoraris, & Romanski, 1990) Auditory inputs to LA come from both the auditory portion of the thalamus (a brain center considered to be a point of convergence of the perceptual senses en route to the rest of the brain) and auditory cortex, where complex sound interpretation is achieved (LeDoux, Cicchetti, Xagoraris, & Romanski, 1990; Mascagni, McDonald, & Coleman, 1993; Romanski & LeDoux, 1993) Fear conditioning to a simple auditory CS can be mediated

by either of these pathways (Romanski & LeDoux, 1992) (Fig 4.2) It ap-pears that the projection to LA from the auditory cortex is involved with a more complex auditory stimulus pattern (Jarrell et al., 1987), but the exact conditions that require the cortex are poorly understood (Armony & LeDoux,

1997) Although some lesion studies have questioned the ability of the tha-lamic pathway to mediate conditioning (Shi & Davis, 1999), recordings from single neurons show that the cortical pathway conditions more slowly over trials than the thalamic pathway (Quirk, Armony, & LeDoux, 1997), thus indicating that the association between CS and US in the amygdala occurs initially through the thalamic pathway Recent functional magnetic resonance imaging (fMRI) studies in humans have found that the amygdala shows activity changes during conditioning that correlate with activity in the thala-mus but not the cortex (Morris, Ohman, & Dolan, 1999), further empha-sizing the importance of the direct thalamo-amygdala pathway

In addition to expressing fear responses to the CS, rats exhibit these when returned to the chamber in which the tone and shock were paired or a chamber

in which shocks occur alone This is called “contextual fear conditioning,” where context refers to the various visual and olfactory aspects of the cham-ber, and requires both the amygdala and hippocampus, a brain structure know to enable long-term memories (Kim & Fanselow, 1992; Phillips & LeDoux, 1992; Maren, Aharonov, & Fanselow, 1997; Frankland et al., 1998) Areas of the hippocampus project to B and AB in the amygdala (Canteras & Swanson, 1992), and damage to these areas interferes with contextual conditioning (Maren & Holt, 2000) Hippocampal projections

to B and AB thus seem to be involved in contextual conditioning (for a

Figure 4.2 The neural pathways involved in fear conditioning are well characterized When the conditioned stimulus (CS) is acoustic, the pathways involve transmission to the lateral nucleus of the amygdala (LA) from auditory processing areas in the thalamus and auditory cortex LA, in turn, projects to the central nucleus of the amygdala (CE), which controls the expression of fear responses by way of projections to brain-stem areas

controlling the autonomic nervous system, the production of hormones, and the appropriate behavior

CE

Auditory Cortex

Behavior

Autonomic system

Hormones

CS (tone)

LA

Thalamus

comparison of the amygdala pathways involved in conditioning to a tone

CS and to a context, see Fig 4.3)

Given that LA is the site of termination of pathways carrying acoustic

CS inputs, it is important to ask whether US inputs might also reach this area and potentially lead to CS–US association Thalamic areas that receive afferents from the spinothalamic tract (LeDoux et al., 1987) project to LA (LeDoux, Farb, & Ruggiero, 1990) (Fig 4.3) Further, cells in LA are re-sponsive to nociceptive stimulation, and some of the same cells respond to auditory inputs as well (Romanski & LeDoux, 1993) Thus, the substrate for conditioning exists in LA

Cortical areas that process somatosensory stimuli, including nociceptive stimuli, also project to LA and some other amygdala nuclei (Turner & Zimmer, 1984; McDonald, 1998) Recent behavioral studies show that conditioning can be mediated by US inputs to the amygdala from either thalamic or

corti-Figure 4.3 (A) Conditioning to a tone involves projections from the auditory

system to the lateral nucleus of the amygdala (LA) and from LA to the

central nucleus of the amygdala (CE) (B) In contrast, conditioning to the

apparatus and other contextual cues present when the conditioned stimulus and unconditioned stimulus are paired involves the representation of the context by the hippocampus and the communication between the hippocam-pus and the basal (B) and accessory basal (AB) nuclei of the amygdala, which

in turn project to CE As for tone conditioning, CE controls the expression of the responses

B / AB LA

CE

Auditory stimulus (tone)

Brainstem

Fear reactions

Brainstem

Fear reactions

B / AB

LA

CE

Hippocampus

Contextual stimulus (cage)

A

B

cal areas (Shi & Davis, 1999), a finding that parallels the conclusions above concerning CS inputs

The AB amygdala receives inputs from the posterior thalamic area (LeDoux, Farb, & Ruggiero, 1990), which is a terminal region of the spino-thalamic tract (LeDoux et al., 1987) While AB does not receive CS in-puts from auditory systems, it does receive inin-puts from the hippocampus (Canteras & Swanson, 1992) The hippocampus, as described above, is nec-essary for forming a representation of the context, and these contextual representations, transmitted from the hippocampus to AB, may be modi-fied by the US inputs to the AB

The CE receives nociceptive inputs from the parabrachial area (Bernard

& Besson, 1990) and directly from the spinal cord (Burstein & Potrebic, 1993) Although CE does not receive inputs from sensory areas processing acoustic CS, it is a direct recipient of inputs from LA, B, and AB Also, US inputs to CE could be involved in higher-order integration For example, representations created by CS–US convergence in LA or context–US con-vergence in AB, after transfer to CE, might converge with and be further modified by nociceptive inputs to CE

Information about a simple CS (e.g., as a tone paired with shock) is di-rected toward CE (where response execution is initiated) by way of path-ways that originate in LA While LA projects to CE directly, and by way of

B and AB, the direct projection from LA to CE seems to be sufficient since lesions of B and AB have no effect on simple fear conditioning to a tone (Killcross, Robbins, & Everitt, 1997) LA and B also project to CE via IC (Paré & Smith, 1993)

The CE projects to brain-stem areas that control the expression of fear responses (LeDoux, Iwata, Cicchetti, & Reis, 1988; Davis, 1992; Kapp, Whalen, Supple, & Pascoe, 1992) It is thus not surprising that damage to CE interferes with the expression of conditioned fear responses (Hitchcock & Davis, 1986; Iwata et al., 1986; Van de Kar, Piechowski, Rittenhouse, & Gray, 1991; Gentile et al., 1986) In contrast, damage to areas that CE projects to selectively interrupts the expression of individual responses For example, damage to the lateral hypothalamus affects blood pressure but not freezing responses, and damage to the periaqueductal gray interferes with freezing but not blood pressure responses (LeDoux, Iwata, Cicchetti, & Reis, 1988) Simi-larly, damage to the bed nucleus of the stria terminalis has no effect on either blood pressure or freezing responses (LeDoux, Iwata, Cicchetti, & Reis, 1988) but disrupts the conditioned release of pituitary–adrenal stress hormones (Van

de Kar, Piechowski, Rittenhouse, & Gray, 1991) Because CE receives inputs from LA, B, and AB (Pitkanen, Savander, & LeDoux, 1997), it is in a position

to mediate the expression of conditioned fear responses elicited by both acous-tic and contextual CSs (Fig 4.3)

Is the Amygdala Necessary?

In spite of a wealth of data implicating the amygdala in fear conditioning, some authors have suggested that the amygdala is not a site of US–CS association

or storage during fear conditioning (Cahill & McGaugh, 1998; McGaugh, 2000; McGaugh & Izquierdo, 2000; McGaugh, McIntyre, & Power, 2002; McIntyre, Power, Roozendaal, & McGaugh, 2003) They argue instead that the amygdala modulates memories that are formed elsewhere It is clear that there are multiple memory systems in the brain (McDonald & White, 1993; Squire, Knowlton, & Musen, 1993; Suzuki & Eichenbaum, 2000; Eichenbaum, 2001) and that the amygdala does indeed modulate memories formed in other sys-tems, such as declarative or explicit memories formed through hippocampal circuits or habit memories formed through striatal circuits (Packard, Cahill,

& McGaugh, 1994) However, evidence for a role of the amygdala in modu-lation should not be confused with evidence against a role in US–CS associa-tion That the amygdala is indeed important for learning is suggested by studies showing that inactivation of the amygdala during learning prevents learning from taking place (Muller, Corodimas, Fridel, & LeDoux, 1997) Further, if the inactivation occurs immediately after training, then there is no effect on subsequent memory (Wilensky, Schafe, & LeDoux, 1999), showing that the effects of pretraining treatment are on learning and not on processes that occur

after learning Thus, in addition to storing implicit memories about dangerous situations in its own circuits, the amygdala modulates the formation of explicit

memories in circuits of the hippocampus and related areas

THE HUMAN AMYGDALA AND

COGNITIVE–EMOTIONAL INTERACTIONS

We now turn to studies on the roles of the human amygdala Deficits in the perception of the emotional meaning of faces, especially fearful faces, have been found in humans with amygdala damage (Adolphs et al., 1996; Stone

et al., 2003) Similar results were reported for detection of the emotional tone of voices (Scott et al., 1997) Further, damage to the amygdala (Bechara

et al., 1995) or areas of the temporal lobe including the amygdala (LaBar

et al., 1998) produced deficits in fear conditioning in humans Also, dam-age to the hippocampus in humans, as in rats, disrupts fear conditioning to contextual cues (Anagnostaras, Gale, & Fanselow, 2001) Functional imag-ing studies have shown that the amygdala is activated more strongly in the presence of fearful and angry faces than happy ones (Breiter et al., 1996) and that subliminal presentations of such stimuli lead to stronger activations than freely seen stimuli (Whalen et al., 1998) Fear conditioning also leads

to increases in amygdala activity, as measured by fMRI (LaBar et al., 1998; Buchel & Dolan, 2000), and these effects also occur to subliminal stimuli (Morris et al., 1999) Additionally, when the activity of the amygdala dur-ing fear conditiondur-ing is cross-correlated with the activity in other regions of the brain, the strongest relations are seen with subcortical (thalamic and collicular) rather than cortical areas, further emphasizing the importance of the direct thalamic–amygdala pathway in the human brain (Morris, Ohman,

& Dolan, 1999) Work in humans has further implicated the amygdala in social interactions (Hart et al., 2000; Phelps et al., 2000) Other aspects of emotion and the human brain are reviewed elsewhere (Davidson & Irwin, 1999; Critchley, Mathias, & Dolan, 2002; Dolan & Vuilleumier, 2003) There is growing enthusiasm for the notion that fear-learning processes similar to those occurring in fear-conditioning experiments might indeed be

an important factor in certain human anxiety disorders For example, fear-conditioning models of posttraumatic stress disorder (PTSD) and panic dis-order (Goddard & Charney, 1997; Rauch et al., 2000) have been proposed

by researchers in these fields

Earlier in the 20th century, the notion that conditioned fear contributes

to phobias and related fear disorders was fairly popular However, this idea fell out of favor because laboratory fear conditioning seemed to produce easily extinguishable fear, whereas clinical fear is difficult to treat Fear disorders involve a special kind of learning, called “prepared learning,” where the CS

is biologically significant rather than neutral (de Silva, Rachman, & Seligman, 1977; Ohman, 1992) While preparedness may indeed contribute, there is another factor to consider In studies of rats, easily extinguished fear could

be converted into difficult to extinguish fear with damage to the medial prefrontal cortex (Morgan, Romanski, & LeDoux, 1993) This suggested that alterations in the organization of the medial prefrontal regions might pre-dispose certain people in some circumstances (e.g., stressful situations) to learn in a way that is difficult to extinguish (treat) under normal circum-stances These changes could come about because of genetic or experiential factors or some combination Recent imaging studies have shown amygdala alterations in PTSD, panic disorders, and depression (Price, 1999; Davidson, Pizzagalli, Nitschke, & Putnam, 2002; Anand & Shekhar, 2003; Drevets, 2003; Rauch, Shin, & Wright, 2003; Wright et al., 2003)

One of the key issues for the coming years is to integrate research on emotion and cognition As a step in this direction, we consider how fear processing by the amygdala is influenced by and can influence the percep-tual, attentional, and memory functions of the cortex

The amygdala receives inputs from cortical sensory-processing regions

of each sensory modality and projects back to these as well (Amaral, Price, Pitkanen, & Carmichael, 1992; McDonald, 1998) These projections allow