Introduction
Knowledge or experiences are voluntarily recalled from memory by reactivation of their neural representations in the cerebral association cortex.  Three questions are central in understanding this process:
  (1) Where are the mnemonic representations coded and how are they organized?
  (2) Which neural processes create the representation?
  (3) What is the mechanism that underlies reactivation of the representation on demand of voluntary recall?  Neuropsychological studies in humans have suggested that long-term declarative memory is stored in the neocortical association area, which is also engaged in sensory perception [1,2].  When the temporal lobe of epileptic patients was electrically stimulated via cortical surface electrodes, they sometimes reported recollection of past perceptual experiences, or ‘experiential response’ [3] (but also see [2]).  This recollection suggests that artificial electrical input to the putative memory storehouse might reactivate the ‘brain’s record of auditory and visual experience’ [3].  Recent experimental studies in humans and non-human primates moved beyond those classical clinical observations and revealed clearer and solider views on the issues, particularly regarding the memory system for visual objects. 

The inferior temporal (IT) cortex of monkeys, the final stage of the ventral processing stream devoted to object vision, has long been assumed to serve as the storehouse of visual long-term memory [2,4,5].  Using single-unit recordings in monkeys performing visual memory tasks, the neuronal correlates of long-term memory have been identified in the anterior ventral part of the IT cortex [5,6].  Specifically, in a visual stimulus-stimulus association task, a class of neurons exhibited significantly correlated visual responses to arbitrarily assigned picture pairs (‘pair-coding neurons’) [7,8], demonstrating that IT neurons can establish new linkage between different stimuli that have meaningful connections. 

Therefore, two complementary approaches are of interest in answering the above three questions.  First, the analysis of single-unit responses would aid in firmly establishing the roles of higher-order visual representations (even beyond the IT cortex) as a part of dynamical network (Fig.1a).  Characterization of bottom-up and top-down signals might also afford a clue to the executive roles of the prefrontal cortex.  Second, the well-analysed neural representations in the monkey cortex should be compared with those in the humans cortex that have recently been revealed by functional magnetic resonance imaging (fMRI) [9,10] and/or by transcranial magnetic stimulation (TMS).  The comparison would provide powerful evidence that couples active neural codes with conscious experiences.

 

Figure 1
(a)  Primate memory system for object vision.  Long-term representations are encoded through interaction of feature analysis and memory control processes.  Representations can be reactivated not only through bottom-up analysis of retinal image but also by top-down control from the prefrontal cortex.  (b)  Physiological evidence for the existence of top-down signals from the prefrontal cortex to the inferior temporal cortex (top-down, blue; bottom-up, black).  Single inferior temporal neurons are robustly activated by the top-down signal without bottom-up visual inputs.  In the spike density functions, thick lines show responses to the optimal cue, whereas thin lines show responses to a null cue.  Onset of the top-down response (arrowhead) is 105 ms later than that of the bottom-up response (double arrowhead). (Reproduced from [25]) 

Creating the mnemonic representation

Lesion studies in primates have implicated the IT cortex in long-term memory storage of visual objects [2,4].  Neuronal correlates of associative long-term memory in the IT cortex were first reported by Miyashita [6, 11] and Sakai & Miyashita [7].  Their single-unit recording experiments identified two mnemonic properties of IT neurons.  First, the stimulus-selectivity of IT neurons can be acquired through learning in adulthood.  Second, the activity of IT neurons can link the representations of temporally associated but geometrically unrelated stimuli.  Modifiability of the stimulus-selectivity was soon supported by Logothetis et al. [12] and by Kobatake, Wang and Tanaka [13].  In contrast, the association effect upon IT neurons, the second part of the findings by Miyashita and colleagues, required more time to gain support.  Guided by a theoretical conjecture that the modification of response properties for temporally associated stimuli might also be a mechanism for learning to associate different three-dimensional views of the same object [5,14], Logothetis et al. found that TE neurons indeed responded more commonly to different three-dimensional views of the same object than would be expected by chance [12].  However, two other groups [15,16], using two different association tasks, failed to find evidence of IT cell participation in association learning.  Recently, Erickson and Desimone provided positive evidence for the association effects upon IT cells [17].  They recorded neurons in the perirhinal area of the IT cortex from monkeys during performance of a visual cue-choice association task, and found the correlation between responses to a familiar cue and choice stimuli to be highly significant.  They also reported that the learning effect was observed only for stimulus pairs that had been associated together for at least two days of training.  Therefore, it has now been clarified that IT neurons have the ability to establish new linkages between different stimuli that have arbitrary but cognitively meaningful connections. 

 Activating the representation in the temporal cortex on demand

In spite of the classical clinical observation that electric stimulation to the temporal lobe produced ‘experiential responses’ [3], there have been only a few lines of direct evidence supporting the notion of ‘reactivation of neural representations’ during memory retrieval.  A neuronal correlate of the reactivation process was first reported as ‘pair-recall neurons’ by Sakai and Miyashita [7].  A subsequent study [18] devised a new modified paired associate task in which the necessity for memory retrieval and its initiation time were controlled by a colour switch, independent of the cue stimulus presentation.  Single-unit recordings in the monkey performing this task revealed that IT neurons selective to a memorized object were dynamically activated at the time of memory retrieval of that object, while suppressed at the time of retrieval of other objects.  A similar prospective delay activity, an activity predicting the sought target rather than the cue stimulus itself, was also found in a GO/No-Go type task for stimulus pairs that had been associated together for at least two days of training [17].  With these accumulating results, it has been an outstanding question to ask what the neural network is that drives the memory retrieval machinery in the IT cortex.  A recent advancement is discussed in the next section. 

Figure 2
Which information is carried via top-down signals?  (a)  A stimulus-stimulus association task.  Twenty cue-pictures are randomly sorted into five categories.  Each of the four cues in one category specifies a common choice.  (b)  Category-selective delay activity of an inferior temporal neuron. Delay activities were raised for all cues in Category I, but not for any cues in Category V (rastergrams, top-down condition). Choice responses are also strongest for Category I and weakest for Category V.  Spike density functions show averaged activities across four cues in Category I (thick) and in Category V (thin) for both conditions (top-down, blue; bottom-up, black). (c)  Spike density functions of the top-down response for five categories, as well as  the choice responses, shown by a pseudo-colour coding.  Note that the category-selective delay activity can predict choice selection.  (Reproduced from [25]).

 Top-down activation through fronto-temporal feedback pathway

A candidate for a component of the neural circuit for top-down activation is the prefrontal cortex.  The prefrontal cortex has been implicated in various executive processes [19], and its contribution in mnemonic functions, especially in episodic memory and working memory, is repeatedly demonstrated by neuroimaging studies (for review, [20]).  Although prefrontal lesions do not usually result in severe amnesia [19], some neuropsychological observations indicate that the prefrontal cortex may play a role in strategic control of memory retrieval [21,22].  On the other hand, the capacity for interhemispheric transfer through the anterior corpus callosum, a key structure interconnecting prefrontal cortices, has been positively highlight executive processes undertaken by the human prefrontal cortex [23].  Inspired by these clinical observations, we introduced the posterior-split-brain paradigm into the associative memory task in monkeys [24].  Long-term memory acquired through stimulus-stimulus association did not interhemispherically transfer via the anterior corpus callosum.  Nonetheless, when a visual cue was presented to one hemisphere, the anterior callosum could instruct the other hemisphere to retrieve the correct stimulus specified by the cue.  Therefore, although visual long-term memory is stored in the temporal cortex, memory retrieval is under the executive control of the prefrontal cortex. 

In spite of predictions from these behavioural experiments, no neuronal correlate of top-down signal from the prefrontal cortex to IT cortex was detected.  The first direct evidence has been demonstrated by Tomita et al. [25] by conducting single-unit recording in posterior-split-brain monkeys.  In the absence of bottom-up visual inputs, single IT neurons were robustly activated by the top-down signal (Fig. 1b), which conveyed information on semantic categorization imposed by visual stimulus-stimulus association (Fig. 2).  Behavioural performance was severely impaired concomitantly with a loss of the top-down signal.  Control experiments confirmed that the signal was transmitted not through a subcortical but through a fronto-temporal cortical pathway. 

Tomita et al. also demonstrated that the top-down signal had longer latency by around 100 ms than the bottom-up signal (Fig. 1b).  The longer latency is most likely ascribed to multi-synaptic conduction delay reflecting the signal transformation within the prefrontal cortices.  During this delay, the prefrontal cortex might prompt to maintain cue-related information, seek out and retrieve the relevant target in the IT cortex, and verify it.  Human PET and fMRI studies in memory retrieval tasks have shown that prefrontal cortex is engaged in retrieval attempt, and contextual monitoring or memory judgement. (e.g., [26- -29]).  A recent report [30] on prospective coding of prefrontal neurons similar to that found in IT neurons [7,18] also supported this view.  Understanding into the signal transformation within the prefrontal cortex would help to bring out the mechanism of top-down control in memory retrieval. 

 Monkey fMRI as an approach to the causal role of brain activation

Because most neuroimaging studies in humans depend upon correlation between cognitive processes and brain activities, to approach causal relationship between them needs participation of other methods.  To verify behavioural significance of brain activities detected by neuroimaging studies, neuropsychological studies on memory retrieval were performed on patients with focal brain lesions (e.g., [31- -33]).  By combining PET and a repetitive transcranial magnetic stimulation (rTMS), Kosslyn et al. examined the contribution of area 17 in visual imagery [34].  Area 17 was activated in imagery task, and moreover, performance of the task was impaired after rTMS to the activated area.  Converging of these complementary methods should, therefore, be a promising direction. To know neural bases of cognitive functions in depth, however, would require studies using monkeys as an experimental model, because most of the detailed anatomical knowledge of the cortex and its functional properties derives from studies in monkeys. 

Recently, fMRI studies in macaque monkeys have been emerging [35- -39].  Stefanacci et al. and Dubowitz et al. [35,36], using fMRI, obtained the first BOLD images of the monkey visual cortex activated by non-retinotopic stimuli.  By mapping topographical organization of the primary and secondary somatosensory cortices, we showed the feasibility of monkey fMRI for distinguishing between adjacent functionally distinct regions [37].  Logothetis and colleagues [39], designing a custom MR machine with a vertical magnet (4.7 T), goes further in imaging technique and in spatial resolution at submillimetre.  By using these techniques, they demonstrated distinct retinotopic activation patterns in the primary visual area.  They also showed activation of superior temporal sulcus, frontal cortex, and amygdala in response to images of monkey faces.  Further studies will direct toward knowing the causal relation of activation and behaviour by combined fMRI and electrophysiology or lesion studies. 

 Conclusion

The role of IT cortex in higher-order visual representation of objects has been established by recent studies in monkeys.  Associative long-term memory is stored in the IT cortex, of which neurons can acquire the stimulus-selectivity through learning and link representations of temporally associated stimuli.  Recently the mnemonic representation was found to be activated on demand of retrieval by the top-down signal from the prefrontal cortex.  The executive control, as a function of the prefrontal cortex, has been decomposed to subprocesses, such as selection and decision making, and their individual neural bases are proceeding to be identified.  Further research into the signal transformation within the prefrontal cortex would elucidate the machinery of top-down control in memory retrieval. 

Numerous neuroimaging studies are carried out to clarify cognitive functions of the human brain.  Because most neuroimaging studies rely upon correlation between cognitive processes and brain activations, conjunction with other complementary methods, such as neuropsychological studies and transcranial magnetic stimulation, should be promoted to clarify behavioural significance of observed brain activities.  Emerging fMRI of the monkey brain, we hope, will direct toward knowing the causal relation of activated brain areas and cognitive functions in cooperation with electrophysiology or lesion studies.  Monkey fMRI also allow us to directly compare the brain activity of humans and monkeys with the same modality.  Since most of the detailed knowledge of anatomy and function of the cortex has come from studies in monkeys, sharing the same method would, in turn, advance understanding in neural organization of human cognitive functions.

  References  

1.   Milner B. Right temporal-lobe contribution to visual perception and visual memory. In: Vision, Memory and the Temporal Lobe. Edited by Iwai E and Mishkin M. New York: Elsevier; 1990: 43-53.
2.   Squire LR: Memory and brain. New York: Oxford University Press; 1987.
3.   Penfield W, Perot P: The brain's record of auditory and visual experience. Brain 1963, 86: 595-696.
4.   Mishkin M: A memory system in the monkey. Philos Trans R Soc Lond B Biol Sci 1982, 298: 83-95.
5.   Miyashita Y: Inferior temporal cortex: where visual perception meets memory. Annu Rev Neurosci 1993, 16: 245-63.
6.   Miyashita Y: Neuronal correlate of visual associative long-term memory in the primate temporal cortex. Nature 1988, 335: 817-20.
7.   Sakai K, Miyashita Y: Neural organization for the long-term memory of paired associates. Nature 1991, 354: 152-5.
8. Higuchi, S. and Miyashita, Y.:Formation of mnemonic neuronal responses to visual paired  associates in inferotemporal cortex is impaired by perirhinal and entorhinal lesions. Proc. Natl. Acad. Sci. USA 1996,93: 739-43.
9.   Tootell RB, Hadjikhani NK, Mendola JD, Marrett S, Dale AM: From retinotopy to recognition: fMRI in human visual cortex. Trends Cognit Sci 1999, 2:174-183.
10. Konishi, S., Nakajima, K., Uchida, I., Kikyo, H., Kameyama, M. and Miyashita, Y.:Common  inhibitory mechanism in human inferior prefrontal cortex revealed by event-related  functional MRI. Brain 1999, 122: 981-91.
11.   Miyashita Y, Chang HS: Neuronal correlate of pictorial short-term memory in the primate temporal cortex. Nature 1988, 331: 68-70.
12.   Logothetis NK, Pauls J, Poggio T: Shape representation in the inferior temporal cortex of monkeys. Curr Biol 1995, 5: 552-63.
13.   Kobatake E, Wang G, Tanaka K: Effects of shape-discrimination training on the selectivity of inferotemporal cells in adult monkeys. J Neurophysiol 1998, 80: 324-30.
14.   Stryker MP: Temporal associations. Nature 1991, 354: 108-9.
15.   Sobotka S, Ringo JL: Investigation of long-term recognition and association memory in unit responses from inferotemporal cortex. Exp Brain Res 1993, 96: 28-38.
16.   Gochin PM, Colombo M, Dorfman GA, Gerstein GL, Gross CG: Neural ensemble coding in inferior temporal cortex. J Neurophysiol 1994, 71: 2325-37.
17.   Erickson CA, Desimone R: Responses of macaque perirhinal neurons during and after visual stimulus association learning. J Neurosci 1999, 19: 10404-16.
18. Naya, Y., Sakai, K. and Miyashita, Y.: Activity of primate inferotemporal neurons related  to a sought target in pair-association task.Proc. Natl. Acad. Sci. USA 1996,93: 2664-9.
19.   Stuss DT, Benson DF: The frontal lobes. New York: Raven Press, 1986.
20.   Fletcher PC, Frith CD, Rugg MD: The functional neuroanatomy of episodic memory. Trends Neurosci 1997, 20: 213-8.
21.   Incisa della Rocchetta A, Milner B: Strategic search and retrieval inhibition: the role of the frontal lobes.Nuropsychologia 1993, 31: 503-24.
22.   Moscovitch M, Melo B: Strategic retrieval and the frontal lobes: evidence from confabulation and amnesia. Neuropsychologia 1997, 35: 1017-34.
23.   Gazzaniga MS: Principles of human brain organization derived from split-brain studies. Neuron 1995, 14: 217-28.
24.   Hasegawa, I., Fukushima, T., Ihara, T. and Miyashita, Y.: Callosal window between  prefrontal cortices : cognitive interaction to retrieve long-term memory.Science 1998,281: 814-8.
25.  Tomita, H., Ohbayashi, M., Nakahara, K., Hasegawa, I. and Miyashita, Y. : Top-down signal  originating from the prefrontal cortex for memory retrieval.Nature 1999,401:699-703.
26.   Fletcher PC, Shallice T, Frith CD, Frackowiak RSJ, Dolan RJ: The functional roles of prefrontal cortex in episodic memory. II. Retrieval. Brain 1998, 121: 1249-56.
27.   Wagner AD, Desmond JE, Glover GH, Gabrieli JD: Prefrontal cortex and recognition memory. Functional-MRI evidence for context-dependent retrieval processes. Brain 1998, 121: 1985-2002.
28.   Buckner RL, Koutstaal W, Schacter DL, Dale AM, Rotte M, Rosen BR: Functional-anatomic study of episodic retrieval. II. Selective averaging of event-related fMRI trials to test the retrieval success hypothesis. Neuroimage 1998, 7: 163-75.
29.   Maguire EA, Frith CD, Morris RGM: The functional neuroanatomy of comprehension and memory: the importance of prior knowledge. Brain 1999, 122: 1839-50.
30.   Rainer G, Rao SC, Miller EK: Prospective coding for objects in primate prefrontal cortex. J Neurosci 1999, 19: 5493-505.
31.   Swick D, Knight RT: Is prefrontal cortex involved in cued recall? A neuropsychological test of PET findings. Neuropsychologia 1996, 34: 1019-28.
32.   Kroll NE, Markowitsch HJ, Knight RT, von Cramon DY: Retrieval of old memories: the temporofrontal hypothesis. Brain 1997, 120: 1377-99.
33.   Thompson-Schill SL, Swick D, Farah MJ, D'Esposito M, Kan IP, Knight RT: Verb generation in patients with focal frontal lesions: a neuropsychological test of neuroimaging findings. Proc Natl Acad Sci U S A 1998, 95: 15855-60.
34.   Kosslyn SM, Pascual-Leone A, Felician O, Camposano S, Keenan JP, Thompson WL, Ganis G, Sukel KE, Alpert NM: The role of area 17 in visual imagery: convergent evidence from PET and rTMS. Science 1999, 284: 167-70.
35.   Stefanacci L, Reber P, Costanza J, Wong E, Buxton R, Zola S, Squire L, Albright T: fMRI of monkey visual cortex. Neuron 1998, 20: 1051-7.
36.   Dubowitz DJ, Chen DY, Atkinson DJ, Grieve KL, Gillikin B, Bradley WG, Jr., Andersen RA: Functional magnetic resonance imaging in macaque cortex. Neuroreport 1998, 9: 2213-8.
37.   Hayashi T, Konishi S, Hasegawa I, Miyashita Y: Mapping of somatosensory cortices with functional magnetic resonance imaging in anaesthetized macaque monkeys. Eur J Neurosci 1999, 11: 4451-6.
38.   Disbrow E, Roberts TP, Slutsky D, Krubitzer L: The use of fMRI for determining the topographic organization of cortical fields in human and nonhuman primates. Brain Res 1999, 829: 167-73.
39.   Logothetis NK, Guggenberger H, Peled S, Pauls J: Functional imaging of the monkey brain. Nat Neurosci 1999, 2: 555-62.