Backward Spreading of Memory-Retrieval Signal in the Primate Temporal Cortex

See allHide authors and affiliations

Science  26 Jan 2001:
Vol. 291, Issue 5504, pp. 661-664
DOI: 10.1126/science.291.5504.661

This article has a correction. Please see:


Bidirectional signaling between neocortex and limbic cortex has been hypothesized to contribute to the retrieval of long-term memory. We tested this hypothesis by comparing the time courses of perceptual and memory-retrieval signals in two neighboring areas in temporal cortex, area TE (TE) and perirhinal cortex (PRh), while monkeys were performing a visual pair-association task. Perceptual signal reached TE before PRh, confirming its forward propagation. In contrast, memory-retrieval signal appeared earlier in PRh, and TE neurons were then gradually recruited to represent the sought target. A reasonable interpretation of this finding is that the rich backward fiber projections from PRh to TE may underlie the activation of TE neurons that represent a visual object retrieved from long-term memory.

Encoding and retrieval of declarative memory depends on the integrity and interaction between the neocortex and the medial temporal lobe system (1, 2). The inferior temporal (IT) cortex, which serves as the storehouse of visual long-term memory (3–10), consists of two cytoarchitectonically distinct but mutually interconnected areas (11, 12): area TE (TE) and the perirhinal cortex (PRh). TE is located at the final stage of the ventral visual pathway (Fig. 1A) (13, 14), whereas PRh is a limbic polymodal association area (1, 2). Forward flow of visual information from TE to area 36 (A36), an immediate adjoining region in PRh, is thought to serve the memory-encoding process (5, 6, 15, 16). Recently, we found that memory neurons are more abundant in PRh (16), and that a neurotrophin, BDNF (brain-derived neurotrophic factor), was selectively induced in PRh during memory formation (17), further supporting the hypothesis of memory storage by neural circuit reorganization. However, the function of the rich backward projection from A36 to TE has not been examined. On the basis of previous observations that IT neurons are dynamically activated by the necessity for memory recall in monkeys (18–20), we hypothesized that the backward projection participates in memory-retrieval processes.

Figure 1

(A) Left panel: Lateral view of a macaque brain. TE is located at the final processing stage of the ventral visual pathway. A36 is thought to be a part of the medial temporal lobe memory system. V4, visual area 4; TEO, area TEO. Right panel: Role of the backward connection from A36 to TE. A36 receives forward visual signal from TE. (B) Sequence of events in a trial of the PA task. Fixation points and cue stimuli were presented at the center of a video monitor. Choice stimuli were presented randomly in two of four positions on the video monitor. (C) Location of recording sites in TE (red) and A36 (green). Left panel: Ventral view of a monkey brain (anterior at the right). Right panel: A part of the coronal cross section (dorsal at the top) indicated by a horizontal line on the ventral view. Scale bars, 10 mm.

Monkeys were trained in the pair-association (PA) task (21), which requires retrieval of a target from long-term memory (Fig. 1B), and extracellular spike discharges of single neurons were recorded from A36 and TE (Fig. 1C) (22). The responses of a representative A36 neuron, with stimulus-selective delay activity related to the sought target specified by a cue stimulus (18, 19), are shown in Fig. 2, A to C. One stimulus elicited the strongest response during the cue period, and the response continued into the delay period (Fig. 2A, upper panel). In the trial when the paired associate of this cue-optimal stimulus was used as a cue, the same cell started to respond during the cue period without an initial perceptual response and maintained a tonic activity until choice stimuli were presented (Fig. 2A, lower panel). The paired associate elicited the highest delay activity among the stimuli (Fig. 2B). We refer to this type of activity as target-related (19). In TE, we also found neurons exhibiting the target-related delay activity (Fig. 2, D to F). However, the time course of the delay activity was different from that for the A36 neuron shown in Fig. 2, A to C: Although the paired associate eventually elicited the highest delay activity (Fig. 2E, lower panel), the onset of the target-related tonic activity was later than that of the A36 neuron (Fig. 2D, lower panel). We examined the time course of the target-related delay activity of each neuron by considering responses to all cue stimuli: The partial correlation coefficients of instantaneous firing rates at time t for each cue stimulus were calculated with the visual responses to its paired-associate stimulus (pair-recall index; PRI) (23–25). In Fig. 2, C and F, PRI(t)s for both of the neurons are plotted as a function of time; the PRI(t) for the A36 neuron started to increase earlier than that for the TE neuron.

Figure 2

Neuronal activity related to memory retrieval during the PA task, as shown by a single cell in A36 (A toC) and in TE (D to F). For the raster displays [(A) and (D)], spike density functions (SDFs) were aligned at the cue onset in trials with the cue-optimal stimulus as a cue (upper panel) and in trials with its paired associate as a cue (lower panel). In the SDFs, black lines indicate responses to the cue-optimal stimulus (upper panel) or its paired associate (lower panel), and gray lines indicate mean responses to all 24 stimuli. In (B) and (E), mean discharge rates during the cue (upper panel) and delay (lower panel) periods are shown for each cue presentation (mean ± SEM). Twelve pairs of stimuli are labeled on the abscissa. The open and filled bars in pair 1 refer to the responses to the stimulus 1 and 1′, respectively. A stimulus-selective delay activity was closely coupled with a strong cue response to its paired associate. In (C) and (F), temporal dynamics of response correlation are shown; the values of the pair-recall index (PRI) are plotted against the time axis and are fitted with Weibull functions (solid lines) (29). The vertical lines, intersecting the best-fit Weibull functions, indicate the transition times (TRTs). The shaded areas indicate the transition durations (TRDs). (G and H) Temporal dynamics of averaged PRI(t) for the population of the stimulus-selective neurons. Mean values of PRI(t) were plotted every 100 ms for A36 neurons [(G), green] and TE neurons [(H), red] (filled circle, total; open diamond, monkey 1; open square, monkey 2; open triangle, monkey 3). Thick lines (green and red, respectively) indicate the best-fit Weibull functions for the population-averaged PRI(t) in the two areas (A36, TRT, 181 ms, TRD, 76 ms; TE, 493 ms, 602 ms). Thin lines, same but for the neurons whose PRI(t) increased above the 5% significance level (A36, 197 ms, 69 ms; TE, 472 ms, 625 ms).

In total, 516 visually responsive cells were recorded from A36 (123 cells) and TE (393 cells), and 418 of 516 cells were cue selective (97 cells in A36, 321 cells in TE). Of the 418 cells, 114 (45 cells in A36, 69 cells in TE) showed significant stimulus-selective activity [analysis of variance (ANOVA), P < 0.01] during the delay as well as during the cue period (26) and are referred to here as stimulus-selective neurons. First, we examined the perceptual signal and found that the latencies of visual response for these neurons in TE were significantly shorter than those in A36 (TE, median 77 ms; A36, median 89 ms; Kolmogorov-Smirnov test,P < 0.05) (27). Second, we examined the mnemonic signal. The time courses of the average PRI(t) across the population of stimulus-selective neurons in A36 and TE, respectively (Fig. 2, G and H), significantly differed between the two areas (repeated-measures ANOVA, P< 0.0001), which was also confirmed in all the animals (P< 0.0001 in monkeys 1 and 3, P < 0.002 in monkey 2) (28). The PRI(t) for the A36 neurons began to increase within the cue period and developed with a rapid time course. The PRI(t) for the TE neurons increased slowly and reached a plateau at the middle of the delay period. The slow, gradual increase of the population-averaged PRI(t) in the TE neurons could be due to the slow development of the PRI curve for single neurons, or to the wide distribution of the onset time for the PRI(t) increase.

We thus determined two parameters that characterized the time course of PRI(t) for each single neuron on the basis of the best-fit Weibull function (Fig. 2, C and F) (29–31). Transition time (TRT) was defined as the period from the cue onset to the instant when the Weibull function reached 50% of its full increase. Transition duration (TRD) was defined as the duration between the instants when the function reached 10% and 90% of its full increase. The parameters were definable for the stimulus-selective neurons with significantly increased PRIs (A36, n = 20; TE, n = 29) (29) and were compared between the two areas (Fig. 3, A and B). The TRT values for the A36 neurons were significantly shorter than those for the TE neurons (A36, median 206 ms; TE, median 570 ms; Kolmogorov-Smirnov test, P < 0.005; Fig. 3A) (32). Moreover, the shorter 70% of TRT values of the A36 neurons were distributed in the range of 138 to 304 ms (i.e., within a 166-ms time window), whereas those of the TE neurons widely ranged from 161 ms to 653 ms (i.e., within a 492-ms time window). The distributions of the TRD values did not differ between the two areas (A36, median 115 ms; TE, median 145 ms; P > 0.8; Fig. 3B) (32). These results indicate that the gradual increase in the population-averaged PRI(t) curve for the TE neurons was due to the wide distribution of TRT values for single neurons, and not due to the longer TRDs.

Figure 3

Time courses of PRI(t) for single neurons. (A and B) Cumulative frequency histograms of the TRT (A) and the TRD (B) for A36 (green) and TE (red) neurons. TRTs for A36 neurons were significantly shorter than those for TE neurons (asterisk, Kolmogorov-Smirnov test, P < 0.005).

We examined the time course of visual and memory-retrieval signals in two subareas of IT cortex while monkeys attended to and retrieved the paired associate of the cue stimulus from long-term memory. The visual signal reached TE before it reached A36. In contrast, the memory-retrieval signal emerged earlier in A36 (median TRT, 206 ms), and TE neurons were then gradually recruited to represent the sought target (median TRT, 570 ms), although there was some overlap in the distributions of TRTs between the two areas. Interestingly, TRD did not differ between the two areas, suggesting that the neural dynamics for the growth of the memory-retrieval signal was similar in A36 and TE (33).

Previous studies have provided some anatomical and behavioral evidence on the nature of the memory-retrieval signals observed in the two areas, although there is as yet no direct evidence concerning which brain region effects the choice decision in the task (30,31). First, area TE receives numerous backward fiber projections from A36 (1, 2, 12, 34). Second, two research groups have demonstrated a dissociation between the effects of damage to PRh (i.e., A36 plus A35) and damage to TE (35, 36). They suggested, that PRh is engaged in mnemonic processing and/or in processing stored knowledge of objects, whereas TE functions specifically in perceptual processing and/or processing structural attributes of objects. Therefore, it is a reasonable interpretation of the difference in TRT distributions in the two areas, although it is not a logical requirement, that the mnemonic signal of the target is spreading backward from A36 to TE. We cannot exclude the possibility that the delayed activations of TE neurons were generated from fast changes in other TE neurons. Another interpretation is that the delayed activations were triggered from other areas such as the prefrontal cortex. Previously, we demonstrated a top-down memory-retrieval signal from prefrontal cortex to IT cortex (37, 38). It remains to be clarified whether the memory-retrieval signal that TE neurons represent originates from a frontotemporal top-down signal for voluntary recall, from a limbic-neocortical backward signal for automatic recall, or from both sources depending on the demand.

  • * These authors contributed equally to this report.

  • To whom correspondence should be addressed. E-mail: yasushi_miyashita{at}


Stay Connected to Science

Navigate This Article