Abnormalities of neural circuitry in Alzheimer's disease

At first encounter, the basic elements that enable the nervous system to execute its functions appear exceedingly simple. These include a transmitting component, which is usually an axon, a point of transmission, which is subserved by the synaptic specialization, and a receptive component, represented by the cell body and dendrites. These elements make up a simple neural "circuit." This simplicity, however, vanishes quickly when the multiple circuits that underlie channeling of information through the nervous system are considered collectively. Each neuron receives input from a large number of axons and, in turn, transmits signals to many other neurons. Neural transmission takes place electrically, through gap junctions, or chemically, through the deployment of a bewildering variety of neurotransmitters and modulating peptides. Neurotransmitters, in turn, produce their postsynaptic effects through interaction with a multiplicity of receptors. In some instances, transmission bypasses the synapse completely, occurring through passive diffusion of the transmitter. Finally, processing of information in the nervous system is accomplished through the utilization of a network of axons and synapses involving many waystations. Viewed in this light, the neural circuitry involved in the production of the simplest behavior becomes exceedingly complex.

Neurodegenerative disorders, including Alzheimer's disease (AD), produce their clinical symptoms by disrupting, to various degrees, the normal circuitry of the brain. The role of the scientist in relation to these disorders is to discover which specific circuits are vulnerable to the neurodegenerative insult and the reasons for such vulnerability. To be of clinical relevance, the vulnerable circuits must be related to the clinical symptoms. In AD, the major clinical symptoms occur in the cognitive and behavioral domains, implicating the cerebral cortex as a major site of pathology. The earliest and most prominent cognitive abnormality in this disorder is loss of memory. A large body of evidence indicates that the specific cortical circuits involved in memory processing are selectively vulnerable to pathology in AD. This review, presents current knowledge of the general circuitry of the primate cerebral cortex, followed by a summary of available evidence on specific circuits vulnerable to AD pathology.

Circuitry of the primate cerebral cortex. On the basis of cytoarchitectural characteristics and functional affiliations, the cerebral cortex of the primate brain can be divided into five cortical types or zones^1,2 (see Mesulam² for review). These include the core limbic, paralimbic, multimodal (heteromodal) association, unimodal association, and sensorimotor cortical areas. The core limbic structures are the most primitive type of cortex and have the least differentiated lamination, containing two or three layers of neurons. There are two core limbic cortical zones in the primate brain: the hippocampus and the piriform (primary olfactory) cortex. Paralimbic areas represent zones of cortical transition between the poorly differentiated core limbic and the six-layered isocortical zones. Accordingly, paralimbic cortical areas contain more and better-differentiated layers than the core limbic zones but do not contain the six layers characteristic of isocortex. The paralimbic zones in the primate brain include the parahippocampal gyrus (including the entorhinal cortex), the cingulate cortex, the temporal pole, the caudal orbitofrontal cortex, and the insula. The multimodal association areas, which include the prefrontal cortex, the inferior parietal lobule, and perhaps the banks of the superior temporal sulcus, have the definite appearance of six-layered isocortex and, as their designation implies, respond to and process information from two or more sensory modalities. Each of the unimodal association areas, on the other hand, processes information from a single sensory modality. Therefore, each primary sensory cortical area is flanked by an association area of the same name. Unimodal association areas can be distinguished from multimodal areas by their increased laminar differentiation, including clear sublamination of layers III and V and increased granularization of layers II and IV. Finally, the sensorimotor cortical zones represent perhaps the most differentiated types of cortex and allow sensory input into the cortex and motor output from the cortex.

Our knowledge of the cortical connectivity of the primate brain is derived, in great part, from studies in nonhuman primates. The parcellation of the primate cortex presented above is also useful in understanding the pattern of cortical connections. This pattern is shown infigure 1. In general, areas comprising one type of cortex have their primary connections with the cortical areas, that are one level above or below them in terms of cytoarchitectonic differentiation.² For example, the primary vertical connections of paralimbic areas are with core limbic zones on the one hand and multimodal association areas on the other.^3-5 Similarly, the multimodal association areas have their primary connections with the paralimbic and unimodal association areas.² In addition, various cortical areas of the same type share extensive horizontal connections. For example, all paralimbic areas have extensive horizontal interconnections. The horizontal connections are well developed in the multimodal association, paralimbic, and limbic zones but constitute a minor component in the primary sensorimotor and unimodal association areas.^2,4,5 It should be kept in mind that this schema applies to major connections among cortical areas. There are cortical connections that violate this general pattern, but these are minor connections.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Schematic representation of various cortical types and their pattern of connectivity. Cortical types are listed in the order of increasing cytoarchitectonic development from bottom to top. The vertical lines indicate major connections among the various cortical types. Thus, the major vertical connections of each zone are with areas immediately above and below it in terms of cytoarchitectonic differentiation. Most of these connections are reciprocal. Diagonal lines represent cortical input from subcortical nuclei with diffuse projections, such as the cholinergic neurons of the basal forebrain. The input from these subcortical nuclei reaches virtually all zones of cortex.

This pattern of cortical connectivity allows integration of input from the external world with information concerning the internal condition of the organism so that appropriate responses can be executed. Information concerning the external world is received by the primary sensory areas via thalamic relay nuclei. After an initial processing, this sensory input is channeled to the appropriate unimodal association area for further elaboration. The core limbic regions, on the other hand, are in touch with the internal environment (e.g., needs, drives) through their extensive connections with the hypothalamus (which has been called the head ganglion of the internal milieu) and with subcortical components of the limbic system such as the amygdala, the septal region, and the substantia innominata.^2,6 Information concerning the internal milieu is further processed in paralimbic zones. The first opportunity for the integration of information about the external world and the internal environment is afforded by the multimodal association areas. Because of this integrative capacity, the multimodal association areas of the human brain are major contributors to higher cognitive functions. Finally, the multimodal association zones communicate with the supplementary and primary motor cortices to allow an appropriate response. As figure 1 implies, the vertical access to information at various levels of cortical organization is bi-directional. Thus, core limbic structures have access to processed sensory information from multimodal association areas via paralimbic zones. Similarly, sensorimotor areas have access to processed information concerning the internal environment through multimodal association areas via unimodal association zones. In short, the pattern of cortical connectivity in the primate brain allows channeling of information along specific neural circuits or networks of circuits. The uninterrupted channeling of information through these circuits is the basis of normal cognitive function.

The primate cerebral cortex is the recipient of a second type of input.^2,7 This input originates, in its entirety, outside of cortex in a number of subcortical regions (figure 1). These include dopaminergic neurons of the ventral tegmental area, noradrenergic neurons of the locus ceruleus, cholinergic neurons of the pontomesencephalic reticular formation, serotonergic neurons of brainstem raphe nuclei, lateral and medial hypothalamus, intralaminar thalamic nuclei, and cholinergic neurons of the basal forebrain. Unlike the subcortical input from thalamic sensory relay nuclei, which channel information to specific cortical areas, the projections from each of these subcortical regions is diffuse, reaching virtually the entire cortical mantle. Therefore, the activity of these nuclei is capable of influencing information processing in many cortical areas simultaneously. For this reason, it has been suggested that input from these subcortical nuclei may function to set the general"state" of cortical circuitry, in contrast to other cortical connections that channel information through particular circuits.²

Pathologic hallmarks of AD. In the first report on the disease that now bears his name, Alois Alzheimer⁸ described two types of lesions in the brain of his patient: "tangled bundle of fibrils" and"miliary foci resulting from the deposit of a unique substance." The terms commonly used today to designate these lesions are the neurofibrillary tangle and the senile plaque, respectively. The plaque is a complex structure found in the neuropil and consists of amyloid (Aβ), abnormal neurites, and glial cells.⁹ Several types of plaques have been described, and it is believed that these represent maturational stages of a single pathologic process. According to this hypothesis, amyloid is first deposited in the form of a diffuse plaque. Gradually, this amyloid is transformed into fibrils, which are believed to be toxic to neurons and disruptive to neuronal processes present in the neuropil.¹⁰ Still later, dystrophic neurites become associated with the plaque, presumably representing degeneration of neuronal components damaged by amyloid.

Tangles are intracellular accumulations of cytoskeletal elements in the cytoplasm and are made of paired helical filaments.⁹ A major component of tangles is abnormally phosphorylated τ (PH-τ, a microtubule-associated protein), and antibodies against this element can be used to stain tangles.¹¹ PH-τ is also a component of plaque neurites and neuropil threads, which are believed to represent degenerating processes (dendrites and perhaps axons) of neurons with tangles.¹² Therefore, all parts of a neuron are vulnerable to cytoskeletal pathology in AD. Tangles damage neurons by disrupting transport of various cellular components and by displacing cytoplasmic elements, thus leading to degeneration of the neurons in which they are formed.

A wealth of recent evidence indicates that Aβ, which accumulates in plaques, may be a major contributor to the pathology of AD.^13,14 Aβ deposition in plaques, however, appears to represent an early pathologic event, because it also occurs in aged individuals with or without mild cognitive disturbance.^15-17 Cytoskeletal abnormality (i.e., tangles, plaque neurites, neuropil threads) and neuronal loss, on the other hand, are probably later events in the cascade of AD pathology, which coincide with the clinical manifestation of dementia and are correlated with dementia severity.^18,19 Therefore, the disruption of neural circuits in AD is most convincingly explained by the cytoskeletal pathology and neuronal loss observed in this disease.

General pattern of cortical pathology in AD. At advanced stages of the disease process, which is the point at which the majority of AD brains come to autopsy, most of the cerebral cortex and many subcortical areas contain large numbers of plaques and tangles.^19-21 Plaque density is usually high and is relatively homogeneous among most cortical areas. Within the paralimbic and especially the core limbic and some sensorimotor areas, relatively lower densities of plaques are found compared with other cortical areas.^19-21 Cortical tangle density, on the other hand, displays considerable regional variation. Core limbic and paralimbic areas contain the highest density of tangles in the AD cortex. The next highest density is found in multimodal association areas, followed by unimodal association zones. Sensorimotor areas contain only a low density of tangles.^20,21 Neuropil threads and, to a lesser extent, the neuritic component of plaques are distributed in parallel with tangles in the above cortical regions. Although tangles can be found without neuronal loss, most cortical areas with a high density of tangles also display significant loss of neurons.^22,23 In addition, tangle-bearing regions of cortex display substantial loss of synapses^24,25 and decreases in neuronal dendritic extent.^22,26 An inevitable consequence of this pathology is the disruption of neural circuits and isolation of affected areas from the rest of the cortex. Disruption of neural circuitry in such a large number of cortical areas helps explain the cognitive nature of abnormalities in AD.

Cortical circuitry in AD is further disrupted by the withdrawal of the influence of subcortical nuclei with diffuse cortical projections.⁷ Although tangle formation and neuronal loss are observed in a number of subcortical structures, the nuclei with diffuse cortical projections are more severely affected.²⁷ Some of these nuclei, such as the cholinergic neurons of the basal forebrain and the raphe nuclei, contain a very high density of tangles.^27-29

Cortical circuits selectively vulnerable to Alzheimer pathology. Loss of memory is the earliest clinical abnormality and the most salient cognitive deficit in AD. By implication, the neural systems that subserve learning and memory should display selective, early, and substantial vulnerability to AD-type pathology. Pathologic observations in AD brains have provided overwhelming support for this implication. Two networks of circuits that have been shown to be involved in various aspects of memory processing exhibit such selective vulnerability in AD. One consists of the pathways that connect the hippocampus and the entorhinal region with the rest of the cortex. These pathways belong to the type of circuit that subserves "channel" functions. The other is the cholinergic system of the basal forebrain, which is a subcortical region with diffuse cortical projections. Here the normal anatomy of these circuits is described, followed by a summary of evidence on their selective vulnerability in AD and a discussion of the functional implications of this vulnerability.

Hippocampus and entorhinal cortex. The hippocampus and entorhinal cortex are related structures located in the medial aspect of the temporal lobe.^2,6 The hippocampus is a core limbic structure and is therefore cytoarchitectonically less differentiated than most other cortical areas. It consists of the dentate gyrus (DG), the Ammon's horn [cornu Ammonis (CA)], and the subiculum⁶(figure 2A). The DG is composed of three layers: the inner polymorphic layer, the granule cell layer, and the cell-free molecular layer (figure 2B). Four distinct sectors can be identified within the CA field. The CA4 sector is surrounded by the DG and is followed distally by the CA3, CA2, and, the largest sector, CA1(figures 2B and C). The CA1-CA4 sectors consist of one major cell-rich layer, which contains pyramidal neurons (hence referred to as the pyramidal cell layer), and two or three other layers, which are composed primarily of axons and dendrites. The thickness of the pyramidal cell layer and the density of neurons within it help differentiate the various CA sectors from one another. The subiculum follows CA1 and is not clearly laminated. The zone of transition between CA1 and subiculum is called the prosubiculum (figure 2G).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Cytoarchitecture of the human hippocampus(A-C,G), entorhinal cortex (D-F), and cholinergic neurons of the basal forebrain (H). A-G are photomicrographs of Nissl-stained sections. (A) Low-power photomicrograph of the hippocampus. All sectors of the hippocampus are present in this micrograph. DG = dentate gyrus; CA1-CA4 = the four sectors of Ammon's horn; ProS = prosubiculum; Sub = subiculum. (B) Higher-power photomicrograph showing the dentate gyrus and CA3-CA4 of hippocampus. G1 = granule cell layer of dentate gyrus; ml = molecular layer of dentate gyrus. The polymorphic layer of dentate gyrus is located immediately above the granule cell layer in this micrograph. (C) CA1-CA3 sectors of Ammon's horn. Only the initial portions of CA1 are shown. Pyl = pyramidal cell layer; ml = molecular cell layer of Ammon's horn. (D) Low-power photomicrograph of entorhinal cortex and adjacent structures. Ento= entorhinal cortex; Amy = amygdala; Pri = perirhinal cortex; CS = collateral sulcus. (E) Higher-power micrograph of the section shown in D, allowing the identification of six layers (1-6) in the entorhinal cortex. Arrows point to layer 2 stellate cell islands. (F) High-power micrograph of stellate cell islands of layer 2 in the entorhinal cortex. (G) Section through the hippocampus, depicting the subiculum and adjacent regions. Arrows point to borders of prosubiculum with the subiculum and the CA1. (H) Choline acetyltransferase-immunoreactive cholinergic neurons in the basal forebrain(nucleus basalis of Meynert; Ch4).

The entorhinal cortex is a paralimbic area and is therefore cytoarchitectonically more differentiated than the hippocampus but less so compared with isocortical zones^2,30(figures 2D and E). It is situated in the anterior aspects of the parahippocampal gyrus, adjacent to the amygdala and the anterior portions of the hippocampus, and is demarcated inferiorly by the collateral sulcus. The transition between entorhinal cortex and temporal isocortex occurs in the perirhinal cortex, located deep within the banks of the collateral sulcus. The entorhinal cortex contains six identifiable layers(figure 2E), most of which are different from isocortical layers.³⁰ Layer I contains few neurons. Layer II is characterized by islands of large stellate cells (figures 2E and F). Layer III is thick and sparsely populated with pyramidal neurons. Layer IV is cell-free and is therefore termed the lamina dissecans. Layer V is densely populated by polymorphic neurons, and layer VI contains diffusely distributed polymorphic neurons.

Consistent with the pattern of cortical connectivity discussed earlier, virtually all cortical input to the hippocampus originates in the adjacent paralimbic area, the entorhinal cortex (figure 3). Therefore, the hippocampus has no direct access to input from primary sensory areas or to information processed in association cortical areas. The entorhinal cortex is interconnected with other paralimbic cortical zones and, more substantially, with the perirhinal and posterior parahippocampal cortices.^3,31,32 The latter two areas, in turn, receive input from paralimbic zones such as the insula, and from unimodal and particularly multimodal association areas such as the posterior parietal cortex, prefrontal cortex, and association cortical areas in the temporal lobe.^3,33,34 The integrated information, which is channeled to the entorhinal cortex, is further processed within this structure and is then conveyed to the hippocampus. The input to the hippocampus originates primarily in layers II (stellate cell islands) and III and secondarily in layers V and VI of the entorhinal cortex.^3,31,32 This input terminates on the dendrites of granule cells in the outer two-thirds of the molecular layer of the DG and, less prominently, on the dendrites of pyramidal neurons in the molecular layer in CA1-CA3 and the subiculum.^6,35 The hippocampus also receives input from the hypothalamus and subcortical components of the limbic system, such as the septum and amygdala.⁶

• Download figure
• Open in new tab
• Download powerpoint

Figure 3. Diagram depicting cortical input to the hippocampus. Input from multimodal (and to a smaller extent unimodal) association areas and many paralimbic regions converge on the entorhinal cortex. The entorhinal cortex, in turn, projects prominently to the hippocampus (via the perforant path). The hippocampus has no access to cortical input except through the entorhinal cortex. Much of the hippocampal output to other cortical areas also travels through the entorhinal cortex.

Information processing in the hippocampus takes place through tightly wired circuits (see Rosene and Van Hoesen⁶ for details). Axons of the granule cell layer of DG project to the CA4 and polymorphic layer of DG and, more prominently, to CA3. The former two regions, in turn, project back to the inner one-third of the molecular layer of DG. Axon collaterals from CA3 and probably CA2 project prominently to CA1 and, to a lesser extent locally, to CA2-3. Finally, a major pathway carries information from CA1 to the subiculum.

Hippocampal output originates primarily from the CA1, prosubiculum, and subiculum.⁶ A major extrinsic projection originating in all three areas carries information to the entorhinal, perirhinal, and posterior parahippocampal cortex and to other paralimbic cortical zones such as the orbitofrontal, cingulate, and retrosplenial cortex. A second extrinsic projection originates in the subiculum and terminates in the mammillary bodies in the hypothalamus. Two other substantial outputs of the hippocampus are to the septum via the subiculum and the CA subfields and to the amygdala via the subiculum, CA1, and particularly the prosubiculum.

In summary, cortical input to the hippocampus is channeled through the entorhinal cortex and, after intrinsic processing through tightly wired circuits, hippocampal output is channeled back to the entorhinal cortex and other paralimbic regions. The hippocampus is also reciprocally interconnected with the hypothalamus and with a number of subcortical components of the limbic system.

A large body of evidence indicates that certain classes of neurons in the entorhinal cortex and hippocampus are selectively vulnerable to tangle formation.^36,37 In the cerebral cortex, neurofibrillary tangles first appear in the entorhinal cortex.^38-40 These early tangles are observed, primarily within the stellate cells of layer II (figures 4A and B). From this restricted locus, tangle formation appears to spread to involve progressively more of the cerebral cortex.^39,40 The first stages of this spread are marked by an increase in the number of tangles in the entorhinal cortex, the appearance of tangles in the pyramidal neurons of CA1 and subiculum (figures 4C-F), and increased appearance of tangle-bearing neurons in other limbic and paralimbic regions.³⁹ Later, the number of tangles in the above structures is increased and tangles appear in neocortical regions, particularly in association areas. Still later, the density of tangles shows an increase in all of the above areas to produce widespread tangle distribution in the end stages of AD. Even at the late stages of the disease, however, the entorhinal cortex and hippocampus appear to display among the highest densities of tangles compared to other cortical regions.^20,21 (figures 4A-F). In addition to susceptibility to tangle formation, the entorhinal cortex, the CA1, and the subiculum display early and substantial accumulation of neuropil threads, neuronal loss, decrease in dendritic extent, and loss of synapses.^41-47 Early AD pathology therefore appears to disrupt the circuitry responsible for both hippocampal input and output.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4. Cytoskeletal pathology in the entorhinal cortex (A,B), hippocampus (C-F), and cholinergic neurons of the basal forebrain (G) in Alzheimer's disease (AD). These sites are vulnerable to early and severe pathology in AD. (A) Hyperphosphorylated τ(PH-τ)-immunoreactive structures in the entorhinal cortex of a mildly demented AD brain. Many tangles are present, especially in layers 2, 3, and 5. Most of the stellate neurons of layer 2 contain tangles. (B) Higher-power micrograph of the section of entorhinal cortex shown in A. Straight arrows point to tangles in stellate cells of layer 2 islands, many of which are affected. Curved arrows point to tangles in smaller pyramidal neurons in layer 3. This pattern of pathology disrupts cortical input to the hippocampus that is conveyed via the entorhinal cortex. Note that a high density of neuropil threads (small arrows) is also present in this area. (C) PH-τ-positive tangles in the CA1 pyramidal neurons of the hippocampus. Small arrows point to PH-τ positive neurites in plaques. (D) PH-τ-stained section showing relatively heavy concentration of tangles in CA1 of the hippocampus and virtual sparing of the dentate gyrus (DG) and the rest of the CA field. (E) Higher-power photomicrograph of CA1 showing the higher density of PH-τ-positive tangles. Note that a high density of neuropil threads (small arrows) is also present in this area. (F) PH-τ-positive neurofibrillary tangles and neuropil threads (small arrows) in the subiculum (Sub). The severe cytoskeletal pathology in CA1 and Sub disrupts output from the hippocampus to its targets. (G) Thioflavin S-positive tangles within the cholinergic neurons of the basal forebrain (the Ch4 group). Many neuropil threads (small arrows) are intermingled with the tangles.

At present, the reasons for the selective vulnerability of the hippocampus and the entorhinal cortex to AD pathology remain unknown.

Cholinergic system of the basal forebrain. The mammalian cerebral cortex contains a dense plexus of cholinergic axons.⁷ In the primate brain, the source of this input is extrinsic to the cortex, originating from a large population of neurons in the basal forebrain^48,49 (figure 2H). The cholinergic neurons of the basal forebrain can be divided into several groups on the basis of their anatomic location and their preferred cortical targets, as deduced from studies in the monkey. The cholinergic neurons of the medial septum [cholinergic cell group 1 (Ch1)] and the vertical limb of the diagonal band of Broca (Ch2) provide the major cholinergic innervation of the hippocampus, the Ch3 neurons provide the major cholinergic innervation of the olfactory bulb, and the Ch4 neurons provide the major cholinergic innervation of the entire cortical mantle and the amygdala. In the human brain, cortical cholinergic innervation is regionally variable.^7,21 The highest density of cholinergic axons is found in core limbic areas such as the amygdala and hippocampus. Paralimbic cortical areas exhibit the next highest density of cholinergic fibers. The unimodal and multimodal association areas contain an intermediate density of cholinergic fibers, and the primary visual cortex (area 17) contains the lowest density. The ACh released from the terminals of cortical cholinergic axons produces its effects through interaction with muscarinic or nicotinic receptors at cholinergic synapses or via passive diffusion.⁷

One of the major and most consistent neurotransmitter abnormalities in AD is a loss of cortical cholinergic markers.^7,50-52 The AD cortex displays a major depletion of its cholinergic axons. The loss of these axons is not uniform but exhibits regional variations.⁵³ The areas with the greatest loss of cholinergic innervation are all in the temporal lobe and include the entorhinal cortex and the hippocampus. The frontal and parietal association areas, as well as the insula and temporal pole, show an intermediate magnitude of loss. The anterior cingulate gyrus, primary motor, primary somatosensory, and primary visual cortex display a minor loss of cholinergic fibers. Consistent with the loss of cortical cholinergic axons, the cholinergic neurons of the basal forebrain from which these axons emanate also display a major loss in AD.^7,54,55 In virtually all AD brains, a large population of tangles is observed in the Ch1-Ch4 cell groups^27,56,57(figure 4G).

Like the entorhinal cortex, the basal forebrain cholinergic system appears to be selectively vulnerable to the cytoskeletal pathology of AD. A substantial body of evidence indicates that the loss of cortical cholinergic innervation in AD is larger, appears more widespread, and occurs earlier than the loss of other cortically projecting neuronal systems.^7,29,58,59 Furthermore, the Ch1-Ch4 cell group is one of the first subcortical areas to display neurofibrillary tangles. Therefore, in mildly demented cases, or in normal cases that show some tangles within the entorhinal cortex, the cholinergic neurons of the basal forebrain almost always contain a variable number of tangles.^60-62

Recent evidence provides a possible reason for the selective vulnerability of the basal forebrain cholinergic system in AD. In the primate, but not in the rodent, the overwhelming majority of the Ch1-Ch4 neurons contain the calcium binding protein calbindin-D28k (CalB).⁶³ In the human brain these neurons display a selective loss of CalB in the course of normal aging.^64,65 The age-related loss of CalB is selective, in that other neurochemicals in the Ch1-Ch4 neurons are preserved and CalB is not lost from other neurons in the aged brain. CalB binds calcium with high affinity and is therefore able to regulate intracellular calcium levels. The loss of CalB would deprive the aging cholinergic neurons of one mechanism for balancing intracellular calcium and may result in high intracellular calcium concentrations that are lethal to neurons. In this manner, the age-related loss of CalB would leave the basal forebrain cholinergic neurons vulnerable to any pathologic insults with the potential to increase intracellular calcium levels, including the pathology of AD.

Consequences of selective vulnerability. The early and severe pathology in the entorhinal cortex, CA1, and subiculum discussed above leads to a virtual isolation of the hippocampus.^36,37 Tangle formation and neuronal loss in the entorhinal cortex, combined with pathology in other paralimbic areas, deprive the hippocampus of much of its cortical input and access to information concerning the external world. Damage to the CA1 and subiculum prevents outflow of information from the hippocampus. Therefore, the neuronal circuitry that enables the hippocampus to execute its normal functions is severely disrupted in AD.

It is now well established that the hippocampus is intimately involved in the neural processing of memory. Specifically, the hippocampus and structures with which it is reciprocally connected are believed to be essential for establishing long-term memory for facts and events.^66,67 In nonhuman primates, lesions of the hippocampus produce severe deficits in learning and memory.^66,68-70 In humans, lesions in the hippocampus result in amnesia.^66,67 Destruction of structures with hippocampal interconnections, such as the entorhinal, perirhinal, and posterior parahippocampal cortices, amygdala, and the mammillary bodies in the hypothalamus, have also been shown to interfere with memory processing in primates.^34,66,71-73 Some of these structures, such as the entorhinal cortex and amygdala, are vulnerable to severe pathology in AD. On the basis of the above evidence, it has been suggested that the relatively early isolation of the hippocampus and the pathology of structures that are related to it comprise one of the main substrates for the severe memory deficit observed in AD patients.^36,37

The basal forebrain cholinergic system also appears to be involved in the neural processing of learning and memory as well as attention. With few exceptions,^74,75 interference with the basal forebrain cholinergic system interferes with learning and memory. Lesions of the cholinergic cells of the basal forebrain deplete the cortex of its cholinergic innervation and impair learning and memory in a number of animal species, including primates.^76-79 The cholinergic basis of this deficit was demonstrated through the reversibility of memory loss by cholinergic agonists.^80-82 In humans, cholinergic antagonists such as scopolamine have been shown to interfere with learning in young volunteers and to cause memory deficits.⁸³ Recent evidence indicates that the basal forebrain cholinergic system may contribute to learning and memory by influencing the ability of the organism to pay attention to relevant stimuli in its environment.^74,75,84 On the basis of the evidence presented here, it is reasonable to assume that the early loss of the basal forebrain cholinergic projection to cortex provides a second substrate for the memory deficit in AD.

In summary, the neural circuits that are selectively vulnerable to AD pathology are those involved in learning and memory. This vulnerability is a probable reason for the clinical memory loss that constitutes the most salient and early symptom in this disorder. The high correlation observed between pathology in these circuits and memory loss, as well as other cognitive symptoms in AD, lends further support to this contention.^{18,39,40,51,52,85}

Therapeutic implications. At present there are no therapeutic strategies that address the selective vulnerability of the hippocampus and related structures to AD pathology. Similar to the rest of the cerebral cortex, the major transmitter used in the hippocampal pathways is glutamate.^18,39,40 Many of the input, output, and intrinsic neurons of the hippocampus are rich in glutamate.^68,86,87 Furthermore, the hippocampus and the entorhinal cortex contain a high density of several types of glutamate receptors.^88-90 In AD, alterations in glutamate and its receptors have been described in the above structures and pathways.^91-96 A major problem with application of free glutamate or glutamate receptor agonists is that these agents can be extremely toxic to neurons. For this reason, glutamatergic transmitter replacement therapy in AD would be difficult at best. It therefore, appears that the major therapeutic approach in relation to hippocampal vulnerability would be one that attempts to arrest the formation of pathologic lesions and neuronal loss. At present, no such therapeutic approaches exist.

In contrast to the hippocampus, the basal forebrain cholinergic system has been a target of intense therapeutic attention in AD. Most efforts in this area have concentrated on the development and application of acetylcholinesterase (AChE) inhibitors. The rationale for this approach is to inhibit ACh degradation by AChE and thus increase the available pool of this transmitter. Many trials of AChE inhibitors have been undertaken. A major problem has been the side effects, including toxicity, of some AChE inhibitors.^97,98 Despite this problem, most trials have demonstrated improvements in cognitive performance in AD patients after application of AChE inhibitors.^99-102 The magnitude of this improvement, however, has been small, and major changes in cognitive abilities have been reported in relatively few patients. In AD brains, AChE is also found in the overwhelming majority of plaques and tangles and may participate in the genesis or deleterious effects of these lesions.^103,104 Therefore, a secondary benefit of AChE inhibitors in AD may be inhibition of this enzyme in plaques and tangles. A second approach in cholinergic therapy in AD is concentrated on the application of selective agonists of muscarinic and nicotinic cholinergic receptors.^105-108 This effort, however, is relatively new and will need to be proved effective for future use.

A major problem in relation to any therapeutic approach directed toward the hippocampus or the basal forebrain cholinergic system is the early vulnerability of these areas to AD pathology. By the time the clinical symptoms appear, both of these structures show major and severe pathology. For therapies to be effective, early diagnosis of AD is necessary. Therefore, future efforts at therapy must be combined with an equally intense effort at developing tools capable of early diagnosis of AD, preferably at a preclinical stage before the onset of cognitive symptoms.

Acknowledgments

The preparation of this review was made possible in part by grants from the National Institute of Aging (AG10282) and the Bayer Corporation. We thank Daniel Kuznetsov for expert technical assistance.