Biol Res 28: 59-72 (1995) 59 in the cortex of the primate

JOAQUIN M FUSTER

Department of Psychiatry, School of Medicine, University of California at Los Angeles, Los Angeles, CA, USA

Memory is viewed as hierarchical and distributed in primary and association areas of cerebral cortex. Different memory neural networks are interconnected at various levels in this hierarchy, sharing neurons and connections. All memory is essentially associative in its generation, structure and retrieval. External and internal stimuli, to which we attend by virtue of their biological relevance or for other reason, can at any time activate ("turn on ") the neuronal network to which they belong by previous association. This is the basis of knowledge and remembering. The reverberation in recurrent circuits may keep the network in an active state, that is, serving behavior, and consciousness. Monkey neuropsychological and electrophysiological data, and human tomographic (brain metabolism) evidence are presented supporting these concepts.

Key terms: association cortex, associative memory, prefrontal cortex.

INTRODUCTION GENERAL PRINCIPLES

In the first place, I will try to present an I will start by stating some general principles overall vision of the role of the cerebral of memory that derive from neuropsycho­ cortex in memory, a vision which, although logical and neurophysiological work in the somewhat personal and perhaps too am­ last years. Some of these principles are still bitious, harmonizes with the evidence that in an embryonic stage, reason for which we through many years we have been accu­ cannot blindly accept them. However, for the mulating in my laboratory. In second place, I sake of brevity, they can be accepted at least will try to suggest a working plan, an agenda as working hypotheses and as premises for for future research, a guide to shed some the experiments that I shall try to summarize light on the still obscure but fascinating and below: complex subject of memory. More precisely, here it will be discussed 1. Memory is a widely distributed func­ (1) how cortical memory is formed; (2) tion. All neural systems have memory. Each where it is formed; and (3) how it is ac­ component of the central nervous system, tivated. In relation to the third point, particu­ each system or subsystem of the brain has its lar attention will be given to one aspect of memory, memory whose contents strictly cortical dynamics about which we have some depends on the information processed by it; experience in our laboratory. It concerns thus, there is in sensory the dynamic structure of associative net­ systems, , auditory memory, works and their activation during recognition haptic memory, etc, ... motor memory in mo­ as well as in the temporal retention of mem­ tor systems, visceral memory in visceral con­ ory. trol systems, etc. In each of these systems,

Correspondence to: Joaquin M Fuster. Department of Psychiatry. UCLA School of Medicine. 760 Westwood Plaza, Los Angeles, CA 90024-1759. USA. Fax: (1-310) 825-6792 60 Biol Res 28: 59-72 (1995) memory forms an integral part in the pro­ formed by association, is maintained by as­ cessing of the information in which the sys­ sociation, and is evoked, remembered, by tem specializes; in fact, representation and association. processing share the same neuronal networks 4. The structure and contents of cortical within each system. Therefore, in neurophys- memory are hierarchically organized, as cor­ iological terms it does not make much sense responds to the organization of the informa­ to speak of systems of memory, but it does tion processing systems constituting these make sense to speak of memory of systems. networks. This hierarchical organization is This is a Copernican turn that I believe is adjusted, at least in a crude manner, to the worth making in our way of thinking if we phylogenetic and ontogenetic order that, as intend to get to some understanding of the we know, is a hierarchical development. To neurobiological basis of memory. understand this better, it is worth observing 2. Systems that interest us are cortical the ontogenetic map of human cortex. systems, and therefore we will be speaking Figure 1 presents the ontogenetic map of the memory of the cerebral cortex, as according to Flechsig (1901) and Von Bonin cortical memory. This fits everything, that is, (1950). Numbers do not indicate here dif­ all classes of memory that are investigated in ferent architectonic areas, but the order in modern neuropsychology. In the perceptual which the different cortical areas become sectors of associative cortex, at different myelinated, that is, the order in which their levels in its hierarchy, we find from down to extrinsic and intrinsic fibers develop their top unimodal sensory memory, multimodal myelin sheaths. Somehow there is a certain sensory memory, , semantic correlation between myelinization and the memory and conceptual memory. Cortical other indexes of ontogenetic maturation memory is therefore the basis of what we call around and after birth, and the map thus cognitive memory, the base of knowledge approximately reflects the maturation of and personal experience. On the other hand, neocortical systems and the functionally in the associative sectors of the frontal lobe, hierarchical development of these systems. we find the plans, action schemes and pro­ This development and its hierarchy them­ grams in all aspects of motility (skeletal, selves result from the -also hierarchical- phy­ ocular, phonetic, etc), from the more abstract logenetic evolution of the cortex. to the more concrete, all plans and programs Thus, at the lowest level of cortical hier­ constituting what we call motor memory. archy, I find the primary sensory and motor Moreover, there in the frontal lobe, we find areas, here in black and with the lowest the mechanisms of organization of the ac­ numbers (Fig 1). These are presumably the tions in the temporal domain in order to first ones to develop. Soon after birth, the implement those plans and motor programs. organization and connectivity of these areas 3. A cortical memory, of any kind, the is complete, and their functions are prac­ "engram", so to speak, is a more or less ex­ tically identical to those of the adult organ­ tended neuronal network of interconnected ism, although it is also true that before they neurons. Depending on its complexity and its must undergo some critical postnatal periods conceptual or semantic contents, this net­ of functional exercise. These areas could be work may be circumscribed to a specific considered as the anatomical basis of what I cortical area or can be extended to different call phyletic memory, or species memory areas with association tentacles of varying (Fuster, 1994). This would be a form of ele­ length and convolutedness. Some of these mentary memory, common to all the association tentacles cross from one hemi­ members of a species, that has developed sphere to the other through the corpus callo- through innumerable generations and by the sum. Thus, in its neurobiological aspect, all same mechanisms of temporal coincidence as memory is a fundamentally associative struc­ those involved in the generation of individual ture, defined by the neuronal assemblies memory, of which we will speak below. The constituting it and by their connections. term "phyletic memory", although seeming a Similarly, in its psychological aspect, all little esoteric and arbitrary, does not need to memory is an associative phenomenon. It is alarm anyone, since I use it to characterize Biol Res 28: 59-72 (1995) 61

Fig 1. Myelogenetic map of the human cortex. something that is perfectly well established. urally idiosyncratic information that the indi­ It is nothing more than a different term for vidual accumulates along its life. Some the sensory and motor devices we are born caveats are necessary before going on: with. Thus, the neuronal networks of primary 1. What I just said does not imply that the cortex, with their proverbial modules and structure of primary areas comes to us al­ columns, would constitute the basis of the ready specified and unchangeable at birth. In cortical hierarchy, predisposed to the first place, there are critical periods in "re-cognize" what the species already knows, their development, to which I have alluded, that is, the most elementary sensory and mo­ where use is necessary for the definitive func­ tor attributes. Beyond and over those primary tional maturation of these areas. These would areas in the connective and developmental be like training periods of phyletic memory, trajectory, are the areas labeled in gray (Fig needed by the organism in order to freely 1), and then in white, constituting the asso­ practice it. In second place, it has been well ciation cortex and conforming the personal demonstrated that these primary areas retain memory networks. These networks are form­ their plasticity even in the adult, a fact that ed there to represent and process the nat­ is apparent in the spontaneous or induced 62 Biol Res 28: 59-72 (1995) rehabilitation of their functions after certain post-rolandic cortex, with its large associa­ lesions (Merzenich and Kaas, 1982). tive region, is involved not only in the pro­ 2. This does not mean either that the cessing of perception but in perceptual mem­ substrate of associative memory comes to ory, including all those memory forms that the world to be made. On the contrary, all its are acquired by the senses (declarative, ep­ cytoarchitectonic and neurochemical struc­ isodic, semantic, etc). Pre-rolandic cortex, on ture seems already genetically determined, at the other hand, is mainly involved in the pro­ least in the monkey. What probably remains cessing and representation of action (includ­ to be done is the synaptic infrastructure at the ing the so-called ""). finer levels, that is, the proteic basis deter­ The cytoarchitectonic map of Figure 2 mining the variability and facility of trans­ (Brodmann) shows the two components, mission of preexisting synapses. It is there perceptual and motor, of the anatomical where individual memory will take its place. substrate of language, with their adjacent 3. In any case, the transitions between pri­ areas, housing perceptual and motor memory mary and associative cortices, that is, be­ for the highest and more exclusive cognitive tween phyletic and individual memory, are activity of man. gradual in all their respects. Both of them keep an intimate relationship, with no clear border between them, neither at a physiologi­ MEMORY FORMATION cal nor at a psychological level. In fact, there are no seams between sensation and percep­ The mechanisms of memory formation are tion, between sensory and perceptual mem­ not precisely known yet, but we do know ory. some of the general principles of this pro­ The ontogenetic map points to another cess, which I will refer to even if in a sum­ general division between the two basic mem­ marized manner. In the first place, it is ory categories we have been talking about: known that in order to form new memory the

Broca's region primary motor cortex

primary auditory cortex Wernicke's region

Fig 2. Brodmann's architectonic map. Language areas marked by shading. Biol Res 28: 59-72 (1995) 63 functional integrity of the hippocampus is networks and the memory they sustain. I needed, which is a sector of phylogenetically only need to add that in the process also par- primitive cortex. The hippocampus is recip- ticipates in some way the amygdala, and that rocally well connected with all associative both, hippocampus and amygdala, participate cortical regions (Fig 3). It is through these not only in the formation but also in the connections, by a mechanism that we do not activation, evocation and remembering of know well as yet, that the hippocampus de- cortical memory. termines the formation, or I would better say What is it that specifically happens in the the expansion of cortical memory networks. cortex, for a new memory to be formed? What supposedly happens is that through a Here we have to introduce the principles pre­ synaptic facilitation mechanism, like long- posed by Hebb more than forty years ago term potentiation, and by mediation of (Hebb, 1949), especially the principle of certain glutaminergic transmitters and some synchronous convergence, or of temporal substances such as nitric oxide, the hippo- coincidence, that has been widely demon- campus facilitates certain changes in the strated in primitive organisms (Carew et al, synaptic structure of neocortical neurons and 1984), such as the mollusk, by Kandel and with that the establishment of cortical neural colleagues (Fig 4). However, it must be

Fig 3. Interconnections between hippocampus and neocortical areas. 64 Biol Res 28: 59-72 (1995)

HEBB S RULE illustrates the main elements of neuronal connectivity in associative areas of the cere­ bral cortex: convergence, divergence and re­ currence. Two stimuli coincide in time, facil­ itating the mutual synaptic contacts between the neuronal assemblies that they represent. Pre- and post-synaptic coincidence When the stimuli disappear, the mnemonic track they leave is precisely in these same

ZA assemblies that, together, linked by potenti­ ated synapses (in white) constitute the mem­ ory network. This memory is established and defined by its contacts and its new relations. And thus it stays, latent, until any one of the stimuli that formed it arrives and by itself reactivates the whole network, re-evoking HEBBS "SECOND RULE" the memory. In this way, by interactions between cor­ tex and limbic system, whose mechanisms we do not yet understand, memory networks Pre-synaptic can be formed by auto-organization, without synchronous other principles than coincidence and use. convergence The paths of memory become established in the association cortex, in an autonomous way and through use.

VISION

VISION TOUCH TOUCH Fig 5. Schematic diagram of memory formation in neural networks. Active neurons in black. On top, temporal coincidence of two visual stimuli (1 ), remaining facilitated synapses (2), and reactivation of the memory network by only one of the stimuli (3). Below (4, 5 and 6), similar processes for stimuli of two different modalities. memory, have been subject of many studies other laboratories in USA, France, Japan and in our laboratory by use of the reversible Scandinavia. lesion method, by selective cooling of cor­ Below, I will present some examples of tical areas. I will not deal with this issue the activity of cortical cells during retention since this is not the place to do it. of short-term memory. Figure 6 indicates What I will discuss, although only briefly, some behavioral paradigms of short-term are some functional phenomena of the ac­ memory. Each trial consists of: (1) the pre­ tivation of memory networks at a cellular sentation of a sensory stimulus that the ani­ level in the monkey and at a level of cortical mal must remember; (2) a delay period; and, metabolism in man. at the end of it, (3) a motor act according to It was probably in our laboratory where the nature of the stimulus. It must be kept in for the first time the cellular phenomena of mind that in the trained animal all stimuli are active memory in association cortices were known and therefore they are part of long- observed: first in the frontal (Fuster and term memory. The task simply requires the Alexander, 1971), then in the inferotemporal successive activation, in successive trials, of (Fuster and Jervey, 1981) and more recently memory of one or another stimulus. Each is in parietal cortices (Koch and Fuster, 1989). a piece of established memory, of long-term In the first cortices, prefrontal and infero­ memory that at the moment mobilizes the temporal, the phenomenon has been poste­ whole memory, activating the whole net­ riorly replicated and observed in several work, including the associations of the 66 Biol Res 28: 59-72 (1995)

B

CUE DELAY CHOICE

DR O O O ® O O O ® <

DA O o o (W) (W) o o ® ®

DMS ® O O o o o o ® ©

Time

MACACA MULAT

Fig 6. A. classical delayed response task. B. Indirect-method delay tasks on automated apparatus. C. Monkey's cytoarchitectonic map; lesions of area FD impair performance in all delay tasks. stimulus with the place, with motor memory, temporal cortex (Fuster and Jervey, 1981), with what the monkey has to do, and with the since this cortex is so to speak the center of memory of the reward if he does the task visual memory (Fig 7). However, not every­ right. All this is part of the mnemonic net­ thing is activation. The neuron of Figure 8 is work activated by association when the stim­ inhibited by the stimulus, be it red or green. ulus appears. If the stimulus, the color that the monkey has After what I have said about the network's to remember, is red, then during the delay extension, and about the many things that are period the neuron activates considerably and associated with the stimulus, we should keeps active during most of the delay period. suppose that while the monkey remembers It is as if, at the stimulus arrival, the neuron it and prepares for the motor response, wide­ was for a moment left in a "tabula rasa" state spread neuronal populations are activated in in order to best receive and retain what it is the posterior perceptual cortex, in frontal specific for, which is the color red, doing this cortex, and even in the hippocampus, which of course not by itself but with the help of its is also involved in recognition. This is what companions in the neural network. is observed recording units in short-term With my colleagues of the University of memory tasks. California at San Diego, David Zipser and If the stimulus. that the monkey must collaborators (Zipser et al, 1993), we are remember is visual, then the most active investigating with empirical models the neurons must naturally be those of the infero- mechanisms of short-term active memory. Biol Res 28: 59-72 (1995) 67

60

GREEN SAMPLE TRIALS

RED SAMPLE TRIALS

YELLOW SAMPLE TRIALS

BLUE SAMPLE TRIALS

I I I I I L_ -J 1 I—I—I I I—I—I I I I I L_l I I I I I I I

10 -I—I-

BASELINE DELAY POST-CHOICE

TIME (SEC) Fig 7. Frequency histograms of the activity of an inferotemporal cell (location marked by dot in brain diagram at right) during delayed matching to sample with four colors. S, sample period; M, match period. Note differences in cell discharge during sample stimulus, during choice (M), and, most important, during memorization of sample color.

Among the several computational models haptein, to touch, palpate), we must observe that we have developed, there is one whose the activity of neurons in tactile associative "units" simulate with extraordinary similarity memory during retention tasks of objects the behavior of inferotemporal neurons dur­ perceived by manual handling. In fact, we ing short-term memory. The model's archi­ also find memory cells there, that is, neurons tecture is essentially based in recurrent cir­ that seem to retain tactile information during cuits (Fig 9). In the trained model, and using the delay period in a short-term memory task the adequate parameters, one finds units that (Koch and Fuster, 1989; Zhou and Fuster, scarcely differ from the real ones (Fig 10). 1992). What is most surprising is that here In any case, it must be kept in mind that the we find them not only in association areas model does not explain how things really such as area 5, but very near the entry of the happen in the brain. What the model tells somatic system, in primary sensory areas us is that short-term active memory is per­ (areas 1, 2 and 3). The difference with the fectly compatible with the activation of recur­ visual system is noteworthy, since there we rent circuits in a neuronal network with pre­ could not find memory cells until well into existing synaptic weights, which harmonizes the associative system, in the inferotemporal with what we have just said about the rela­ areas. The difference may be explained by tion between established and active memory. the fact that the tactile system processes in­ In order to determine what happens with formation mainly in series, while the visual memory activation in a different sensory system does it in parallel. The haptic system modality, haptic (tactile, from the Greek works by temporal integration. 68 Biol Res 28: 59-72 (1995)

R

• •amatisi I BBaaiiaiaMaaiiiB imi IIIMB taiii BIBIBIB n laminami IBIBI i uni 11 I il in «alani laiaaBiaaaaiu ai aaaiai iBiaini ib inaiai ii niiiaaiBBiBi anni n i mil II II BIIBIIIIII I laMIMIMHMI I I BBBBBHiaaaiaBiiiaaBBa im ill I aaiBBaaBBBB iiiiiaiB lam • mi IIIIIIIIIIIIIIIIIBI •iiniiiHii •iniaaBBBBiiai laiiiaaainiaiiiii IIIBBMIIIII IIIIIBIBIIIM II im I •minai I ibi I in ii inalili • I mil liliali ni mi •mi • min maw hi mmrni imi BiiiBiBBi HI I I • im I niaana •ill l • mi a mini iiaanaai i iHamiiiui mi in nun um I aiiinaaai aaauniaaai••• aaaiiiH a aiaiBBBBii ••!•« il in I I IB mi a in IBIi min II I ni iaai ••iiiniii ii inaia iiai ii •iMBBaiiianiBMMai n • mm na IIBBHII iiBBiaBBBiiiiiii la nail aiBBiB mi mail n iiiiiiina •BBBBBaa^ aii iBBia I hi ia a il I ai HBHiMiiiiiiii Mai auiiniiiiamiBHiiiiiBi »um I iMiMMMii an a i Balaam III I mi I IM I mill ill inn II in I anBiiiMiM

im 11 aaa imiaiii naiiiaaii I aiaaBiaaaiaaaaMMiiiiii iniiiBiiBaaaa 11 II II I IBB a aiMia ini aia an I lana i laaBiaHiiBaaBBaaaii I I I aaui• a• linn IIBBBBBI MI BMniai amimi lau IBB in na aaa imi 11 ama I in iiiiiMMiiiii it ••• II in ii nil • I in iiiaiaaiiiBBBaiiMiiMaiiiiaiaMM I mill BBBI IBBB BI IB am aiaa BIIIBBHIBIBBMII ia i IIBIIII Haan I MM iiiiiMBiiiiaiiiaBiiBBBaiBiiBaiiiii I in • I • a I • I minium HI lllllll III II •Malllllll I • NIM mi II laiiiiiaaiiiiBiiiiiiiBBiiiiBaBiiii ••nan I BB IBM III IIIIIIIII • I laiiiaiBBiBBii in II Maaiaaaaiaiaimn I BBM II naiBiiiBaaaaBBaBBBBMBaaaaiiiaaiiMiiBi I iniiiBBBMiii MiMaaaa •• II mi mi laiaaaaaaaiMi II in IBB •I • I I mam aaMMMaiiMaaiiMBBMMi iniiM I ni a a Balani I aiaiaiiiiii I aaai tail a an n un in in n Ban ia in iMiaaii in naiaaiaaii laiBMBi aia II I aai I iiinaiiaiiaaaMM IB i II II in iaa I • am I n in mil niiaa nil a la BBM aan BBMI IMI BBMIIIB II I aiiHMaiaaiBiaia I a a limn I I in a iminai ami mi aiiani MBiiaaaiaaiiaaBBBaB IBMMIIMIB I BiiiiBBBBian BIIIBBBI aaaajHiaaiiiiiiBiiiniiaii a aaa il li un 11MB 11 BBaaaiiiM MIBB Baaiaaai llIII lalaa limi ••mal

20 • RED • GREEN Choice

ä io H 00

25 30 Seconds

Fig 8. Rasters and frequency histograms of the activity of an inferotemporal cell during delayed matching (paradigm below). Note inhibition at sample and match, and differential activation during memorization of red. oft)

What matters, in last instance, is that sensory information that is waning, while the tactile perception, more than visual percep­ latter ones accelerate, as if anticipating and tion, needs a mechanism of short-term predicting movement. Through a visuomotor memory already in the early processing task using colors, where each color is asso­ stages. It is as if phyletic memory had al­ ciated to a certain probability of manual mo­ ready its own short-term memory in primary vement, Javier Quintana and I (Quintana and areas. Fuster, 1992) were able to show that the cells Now we will see some examples of active that presumably code for the motor memory memory in the sectors of frontal associative increase their firing in proportion to the prob­ cortex, where not only motor memory is ability with which the monkey can predict represented but is also where behavior is or­ the motor act that he will have to perform ganized in the temporal axis (Fuster, 1989). (Fig 11). These cells seem to participate not There we find two main types of memory only in the network representing the act but cells: those of the sensory type, that code in the network that prepares it, which un­ information of any sensory modality for the doubtedly is to some extent the same. motor act that will take place, and those of Lastly, I want to show that things are the motor type, coding this act. During the probably not very different in man, at least delay period, the formers decrease their regarding the role of the prefrontal cortex. firing, as if looking backwards in time to the With my associates in Los Angeles Biol Res 28:59-72 (1995) 69

Short Term Memory Model

-\ Info Red Exp

High Model

TRAINING Green Kxp

Io» Model

TIME (STEPS)

Fig 10. Simulation by model in Figure 9 of neuron in Fig 9. Scheme of the architecture and "training" features of Figure 8. a computer model of active short-term memory.

MOTOR-COUPLED CELLS (N=15)

100%-predictable direction

CUE-COUPLED CELLS R2 = 0.910 (N=9) o I .

20 i R2 = 0.832 2 4 6 8 10 12 UJ 15

z 10 3.0 75%-predlctable direction

QQ 5 "v. 2.5 R2 = 0.790 Ì 0 6 8 10 12

1.5 t DELAY (Sec.) t c R 1.0 4 6 8 10 12

t DELAY (Sec.) t c R Fig 11. Activity of prefrontal cells during the delay of a delay task with probabilistic relationship between stimulus (color) and motor response (C, cue; R, response). At left, cells related to color; at right, cells related to motor response. Note that motor-coupled cells show gradual increase of firing as the motor response approaches, and that the slope of that increase is related to the probability with which the monkey can predict the direction of response. 70 Biol Res 28: 59-72 (1995)

(Schwartz et al, 1995), we have been inves­ neurons, tomographic (PET) images of the tigating metabolic activity of cerebral cortex brain were obtained. Many areas were acti­ during visual short-term memory. Here, vated by the tasks. What most interests us instead of using colors, as in the monkey, we here is that prefrontal cortex is vividly acti­ used more complex stimuli, such as abstract vated, especially during the memory task. If drawings (Fig 12). The task is as follows. A the metabolic activations produced by the trial begins with an alert signal. Then, an two tasks are subtracted, a map such as that abstract image, the sample, appears. Follows on Figure 13 results, showing that during the the delay period, the period of active memory task there is a higher activation of memory, and finally a second image appears. Brodmann's areas 9, 10 and 46. Motor and If this second image is identical to the sam­ premotor areas are also relatively activated. ple, the subject presses a button with one This is probably due to the fact that even if hand; if not, he presses another button with both tasks require the same number of motor the other hand. There is also a control task, responses with the two hands, the subjects identical to the first (the same stimuli, the spend more time preparing a response in the same motor responses), with the difference memory task than in the control. that there memory of the image is not re­ quired: both stimuli appear at almost the same time, with minimal delay. CONCLUSIONS Before executing the two tasks, radio­ active glucose was injected to 18 volunteers, Certainly, the data I have presented offer a and subsequently, during both tasks and dur­ partial vision of cortical memory. However, ing the uptake of the isotope by the central at least they give us an idea of the principles

Si

'Ready" (400 msec)

DELAY (1500 msec) /DMS8.0 sec\ VIMS: 0.1 sec/ Fig 12. Delayed matching to sample task for the human (see text for description). Biol Res 28: 59-72 (1995) 71

every sense. This means that and their networks are not strictly stratified but they overlap among them at several levels. In fact, it cannot be otherwise considering that all memories and their networks share neurons and connections. This concept is fundamental in order to clear the existing confusion in the neuropsychological tax­ onomy of memory. 3. Above phyletic memory -if we accept the concept- is unimodal memory, and above it multimodal, and then episodic memory, which is a kind of associative memory, as all the others, but with space and time associa­ tions; and over them is , and above it, conceptual memory. All of them are associative. What happens is that, as we go up, knowledge, memory and the network are all anchored in a larger number of connections, that is, associations. With it, they become more resistant to aging and injury. 4. With aging and cortical injury, the memory of an episode is lost, but not that of the name or of the concept; of the category, but not of the procedure. This is not because semantic, episodic and procedural memories are in different places in the brain. It is simply because some are better anchored by Fig 13. Areas in which activation in memorization (memory task) exceeds activation in control (no-memory task). Black associations than others. There is no reason area (dorsolateral prefrontal) is activated the most during to suppose that the memory to play chess, for visual memorization. example, occupies a separated and different cortex from the memory of the day and place of distribution and operation of cortical we learned to play it. But, because of their memory. We only need to underline these lability and specificity, place and time asso­ principles in the wider context of human ciations are more vulnerable and susceptible memory. to be forgotten. 1. As I mentioned, the vision of individual 5. Some external or internal stimuli to memory that guides our research is hierar­ which we attend by virtue of their biological chical. For sure, the data on memory struc­ relevance or for whatever reason, may, at any ture and dynamics I have referred to are lim­ given moment activate the cortical network, ited to what happens at relatively low levels they can so to speak, "turn it on" (to use in the hierarchy, at the unimodal perceptual Braitenberg's [1978] term). This is the memory level and the manual motor mem­ essence of recognition or remembering. If ory level. My position here is that what the network needs to stay active to dial the happens at these low levels of memory and telephone number, to complete the arithmetic associative cortex is paradigmatic, qualita­ operation or the discourse, or, as in the case tively similar to what happens at higher of our monkey, to choose between red and levels. green, there must be a mechanism, probably 2. As we ascend in the hierarchy of cor­ of reverberation in the network, as our exper­ tical memory, networks gain not only in iments and models indicate, maintaining the hierarchical level, that is, complexity, but excitation of the network over a certain level also in extension, that is, in diffusion in until the organism achieves its goal. 72 Biol Res 28: 59-72 (1995)

REFERENCES FUSTER JM, JERVEY J (1981) Inferotemporal neurons distinguish and retain behaviorally relevant features of BONIN G von (1950) Essay on the Cerebral Cortex. Spring­ visual stimuli. Science 212: 952-955 field, 111: Charles C Thomas HEBB DO (1949) The Organization of Behavior. New York: BRAITENBERG V (1978) Cell assemblies in the cerebral John Wiley and Sons cortex. In: HEIM R, PALM G (eds) Theoretical Ap­ KOCH KW, FUSTER JM (1989) Unit activity in monkey proaches to Complex Systems. Berlin: Springer- parietal cortex related to haptic perception and tempo­ Verlag, pp 171-188 ral memory. Exp Brain Res 76: 292-306 CAREW TJ, HAWKINS RD, ABRAMS TW, KANDEL E MERZENICH MM, KAAS JH (1982) Reorganization of (1984) A test of Hebb's postulate at identified synapses mammalian somatosensory cortex following peripheral which mediate in Aplysia. J nerve injury. Trends Neurosci 5: 434-436 Neurosci4: 1217-1224 QUINTANA J, FUSTER JM (1992) Mnemonic and pre­ FLECHSIG P (1901) Developmental (Myelogenetic) dictive functions of cortical neurons in a memory task. localisation of the cerebral cortex in the human subject. Neuroreport 3: 721-724 Lancet 2: 1027-1029 SCHWARTZ BE, HALGREN E, FUSTER JM, SIMPKINS FUSTER JM (1989) The Prefrontal Cortex: Anatomy, F, GEE M, MANDELKERN M (1994) Cortical meta­ Physiology, and Neuropsychology of the Frontal Lobe. bolic activation in humans during a visual memory New York: Raven task. Cereb Cortex (in press) FUSTER JM (1994) Memory in the Cerebral Cortex: An ZHOU Y, FUSTER JM (1992) Unit discharge in monkey's Empirical Approach to Neural Networks in the parietal cortex during perception and mnemonic retention Human and Nonhuman Primate. Cambridge, MA: of tactile features. Soc Neurosci Abstr 18: 706 (Abstract) MIT Press ZIPSER D, KEHOE B, LITTLEWORT G, FUSTER JM FUSTER JM, ALEXANDER GE (1971) Neuron activity (1993) A spiking model of short-term active memory. J related to short-term memory. Science 173: 652-654 Neurosci 13: 3406-342