Spontaneous Changes of Neocortical Code for Associative Memory During Consolidation

See allHide authors and affiliations

Science  07 Nov 2008:
Vol. 322, Issue 5903, pp. 960-963
DOI: 10.1126/science.1161299


After learning, the medial prefrontal cortex (mPFC) gradually comes to modulate the expression of memories that initially depended on the hippocampus. We show that during this consolidation period, neural firing in the mPFC becomes selective for the acquired memories. After acquisition of memory associations, neuron populations in the mPFC of rats developed sustained activity during the interval between two paired stimuli, but reduced activity during the corresponding interval between two unpaired stimuli. These new patterns developed over a period of several weeks after learning, with and without continued conditioning trials. Thus, in agreement with a central tenet of consolidation theory, acquired associations initiate subsequent, gradual processes that result in lasting changes of the mPFC's code, without continued training.

The hippocampus is necessary for rapid association among elements of an event (14), and is initially also critical for retrieval of these associations; however, its necessity for retrieval is time-limited (1, 5). In trace eyeblink conditioning, the medial prefrontal cortex (mPFC) becomes necessary for retrieval of associations as they become independent of the hippocampus (6), a process that requires intact N-methyl-d-aspartate (NMDA) receptor function in mPFC (7). If, as these results suggest, memory is gradually consolidated in a network encompassing mPFC, then mPFC neurons should become selective for learned associations with a similar time course.

We recorded from cells in the deep layers of prelimbic mPFC of rats, during a conditional associative learning task (table S1). Four rats were trained on a context-dependent association between a neutral tone [conditioned stimulus (CS)] and a mild shock to the eyelid [unconditioned stimulus (US)] (Fig. 1A). When the CS and US were paired in a fixed temporal pattern (Paired), the rats gradually expressed eyeblinks to the CS [conditioned response (CR), monitored by eyelid electromyogram] over ∼10 days, and the frequency of CR expression was near asymptote throughout the recording sessions (Fig. 1B). In contrast, rats did not express CRs to the same CS when it was unpaired with the US (Pseudo) or presented alone (CS in box A and B). Selective neural activity for acquired associations was quantified with a discrimination function based on the difference in the neurons' responsiveness during the 500-ms interval after the CS (8) (Fig. 1C). The mPFC becomes necessary for retrieval 2 weeks or more after acquisition (6, 7). Consistent with this time course, the selective neural activity for the association increased from the late stage of acquisition and reached a peak during the second week of overtraining (table S2A).

Fig. 1.

Context-dependent acquisition of memory associations. (A) Rats were exposed to an environment in which an auditory CS was paired with eyelid stimulation US (Paired) and a separate environment in which the CS and US were unpaired (Pseudo), with each condition bordered by rest periods. During the first 20 trials in both conditions, the CS was delivered without the US. (B) Percentage of trials in which rats exhibited CRs increased in Paired, but not Pseudo, condition across weeks of learning (Early and Late) and overtraining (Post 1w to Post 6w; mean ± SEM from four rats; Spont.: spontaneous eyeblink frequency). (C) The mPFC becomes necessary for retrieval 2 weeks or more after acquisition [arrows (6)]. This process requires NMDA receptor function in the mPFC [shaded area (7)]. With a similar time course, neuron activity in the mPFC became selective for acquired associations (mean ± SEM from four rats). Selective activity was quantified on the basis of difference in firing-rate changes during the post-CS interval between Paired and Pseudo condition. Only the neurons that showed firing-rate differences between conditions were included.

We compared each neuron's firing rate during the CS or trace interval (interval between the CS and US) to its baseline activity (8). About 60% of neurons responded to the CS and/or trace interval during at least one condition throughout acquisition and overtraining. About 25% of neurons showed a different response pattern during Paired than during Pseudo and CS in box A (Behavioral context, Fig. 2B). Because they discriminated paired stimuli from unpaired stimuli, these neurons may encode acquired associations. Alternatively, they may have changed firing rates in conjunction with eyelid movements during Paired (i.e., CR), which were absent during Pseudo and CS in box A; however, this is unlikely, because firing rates of only ∼5% of neurons were significantly correlated with CR amplitude or duration (fig.S1, A and B). Some neurons were sensitive to spatial context (i.e., the conditioning boxes) (∼5%, Spatial context, Fig. 2C) or to whether the CS was presented alone or with the US (∼5%, Shock context, Fig. 2D). Others showed the same response pattern in all four conditions (∼35%, Non-selective, Fig. 2A). About 40% of neurons changed their baseline rates during Paired compared to Pseudo, suggesting that their activity was selective for the context per se or for different background behaviors in the two contexts. No changes in the ensemble pattern of these response categories were observed across two stages of acquisition and six stages of overtraining (Fig. 2E).

Fig. 2.

Response categories of activity in single neurons. (A to D) Although some neurons showed the same stimulus-related neural firing patterns regardless of context (A), others were found with firing rates that depended on behavioral context [rates in Paired were different from rates in Pseudo, and from rates in CS in box A condition (B)], spatial context [rates in Paired were different from rates in Pseudo, but not from rates in CS in box A condition (C)], or shock context [rates in Paired were different from rates in CS in box A, but not from rates in Pseudo condition (D)]. Raster plots and peristimulus time histograms show firing rates during a ±1500-ms period around the CS onset (1-ms bins, smoothed with 50-ms hamming window; pink: CS in box A; red: Paired; cyan: CS in box B; blue: Pseudo). The black filled bar indicates a blackout period because of the US artifact. (E) There were no changes in the proportion of cells in each category across stages of learning and overtraining. This was true when examining the activity during the pre-CS period (left), the CS (middle), or the trace interval (right).

To examine how the activity patterns of single neurons generalized to the population and how the population response patterns changed during acquisition and overtraining, we normalized the activity of each neuron and used the averages within the same response type (excitatory or inhibitory, fig. S2) (8) as a measure of population response pattern (Fig. 3). At the beginning of acquisition (Early), the excitatory response pattern during Paired was similar to the pattern during Pseudo; however, as the rats acquired CRs, response patterns in the trace interval during Paired became separated from those in the corresponding interval during Pseudo (Fig. 3, A and C, right, and table S2B). During Paired, a sustained response to the trace interval was observed from the early stage of acquisition and was maintained for a 6-week period of overtraining (Fig. 3C, right, and table S2C). In contrast, during Pseudo, the response to the corresponding interval gradually weakened during overtraining (Fig. 3C, right, and table S2D). The excitatory response to the CS was similar between Paired and Pseudo and gradually weakened during overtraining (Fig. 3C, left, and table S2E). Similarly, the response of neurons that exhibited inhibition in the interval between the stimuli was stronger during Paired than during Pseudo from the beginning of acquisition to the final week of overtraining (Fig. 3, B and D, right, and table S2F). The inhibitory response to the CS was similar between Paired and Pseudo during acquisition and overtraining (Fig. 3, B and D, left, and table S2G). Population excitatory and inhibitory responses were weak and transient to the CS when the CS was presented alone throughout all of the stages (Fig. 3, A and B).

Fig. 3.

Plasticity in firing-rate patterns of mPFC neural populations. (A) Z-score values for neuron responses to the CS and trace interval (i.e., change from baseline during a 1800-ms window around the CS, truncated to 8 and –8 for illustration purposes) were averaged across all excitatory neurons, and this measure of population activity is plotted over time within trials (x axis) across days (y axis, ascending from bottom to top; averaged percent CRs are plotted to the left). During acquisition and overtraining, mPFC neurons exhibited sustained excitatory activity in the trace interval during Paired condition (left). During overtraining, this sustained activity reduced during Pseudo condition (right), and was absent in trials with the CS alone (lower panels). (B) Medial prefrontal neurons with inhibitory responses exhibited a sustained response in the trace interval specifically during Paired. (C) Neural activity in the population of cells with excitatory responses (averaged z scores) did not differ between Paired and Pseudo conditions during the CS (left), but were different between conditions during the trace interval after CRs were acquired (mean ± SEM from four rats). (D) Neurons exhibiting inhibitory responses differentiated between Paired and Pseudo trace interval before CRs were fully acquired.

To clarify whether these new firing rate patterns resulted from repetitive overtraining or a spontaneous neural process subsequent to the initial acquisition of the memory, we recorded from three additional rats (rats 5 to 7) during reconditioning (Retention) after a 6-week, training-free “consolidation period” (Fig. 4B). Although the percentage of CRs was reduced in the first retention session, it was significantly higher during Paired than during Pseudo (table S2H) and spontaneous eyeblink frequency (table S2I). Moreover, the rats relearned the CR faster than during the initial conditioning: The number of sessions required to reach 60% CRs was 4.7 ± 0.4, whereas during initial acquisition, 10.7 ± 0.8 sessions were required (table S2J). The mPFC population showed a sustained excitatory response to the CS during Paired, but only a transient response to the CS during Pseudo (Fig. 4D, top). These activity patterns were similar to the patterns during the final (sixth) week of overtraining (Post 6w, Fig. 4C, top right), but not during learning (Late, Fig. 4C, top left). The population inhibitory response during Paired was larger than it was during Pseudo (Fig. 4D, bottom), in a manner similar to that of the patterns during Late (Fig. 4C, bottom left) and Post 6w (Fig. 4C, bottom right). Further control experiments (fig. S3) confirmed that differential activity between the paired stimuli and the unpaired stimuli could not be attributed to the fixed order of conditions or to instability of neural activity.

Fig. 4.

Change of neural activity patterns with time, without overtraining. (A) Mean percentage of trials with CRs increased specifically during Paired condition in the overtraining group (mean ± SEM from four rats). (B) The consolidation group maintained higher percent CRs in Paired as compared with Pseudo condition after a 6- to 8-week period without exposure to the conditioning environments (mean ± SEM from three rats; Spont.: spontaneous eyeblink frequency). (C) In the overtraining group, excitatory firing-rate patterns differentiated between Paired and Pseudo conditions during the sixth week of overtraining (Post 6w, right), but not during the late stage of acquisition (Late, left). (D) In the consolidation group, population firing-rate patterns during reconditioning were similar to those observed after overtraining. Plots show averaged firing-rate patterns during ±1500-ms time windows around the CS onset, binned at 1 ms, for neurons with excitatory responses (top) and inhibitory responses (bottom).

In addition to changes in firing rate, information about memory associations may also be encoded in relative spike-timing patterns of multiple neurons. The template-matching method (9, 10) showed that population spike-timing patterns gradually became selective for associations as a consequence of overtraining, but not simply as a consequence of the passage of time (figs. S4 and S5).

Consolidation of memory is presumed to involve gradual reorganization of cortical networks (2, 6, 1116). Our findings suggest that this consolidation process may be directly reflected in the activity pattern of the mPFC neurons. As the rats acquired the conditional association, the firing rates of some mPFC neurons became selective for the acquired associations, by exhibiting sustained activity during the interval between the paired stimuli, and correspondingly reduced activity during the interval between the unpaired stimuli. Considering that a proportion of single neurons with selective responses to the paired stimuli did not increase across sessions of acquisition or overtraining, a specific set of mPFC neurons encodes the difference between paired and unpaired stimuli, and changes in the magnitude of their differential activity, rather than changes in the number of differentiating neurons, are responsible for the gradual development of activity selective for memory associations. In addition, some neurons changed their baseline firing rate depending on the context (Fig. 2, C to E), indicating that acquired associations modify not only neural responses to presented stimuli, but also responses to the context.

Population firing-rate activity selective for memory associations was observed from the late stage of acquisition to 6 weeks of overtraining. Similar patterns developed after a 6-week period without any conditioning. This slow time course of firing-rate changes agrees with a previously observed time window in which the mPFC becomes important for retrieval (6). One possible mechanism of these lasting changes may be that the weights of intra- and inter-area connections in the mPFC are gradually redistributed by multiple rounds of synaptic modification (7). Such synaptic modification might be triggered by the replay of task-related neural activity patterns during sleep (17). In contrast, hippocampal neurons change firing-rate patterns at the onset of learning (18, 19). Therefore, the hippocampus may direct the mPFC to refine its memory-related neural activity, as proposed in standard consolidation theory (2, 15, 16, 20, 21).

Memories of associations between various elements of an event are presumed to be distributed over wide areas of cortex. The hippocampus may store a memory index for a unique array of neocortical modules representing each experiential event (22). Direct cortico-cortical connections that are gradually established during consolidation may render the hippocampal index codes for the original memory no longer necessary; alternatively, for certain forms of memory, the mPFC might take over the linking function from the hippocampus by storing a similar, but perhaps more efficient, index code (23). The observed selective activity for context-dependent, cross-modal associations in the mPFC is a critical prerequisite for the mPFC to function as a storage site of memory indices for consolidated memory. The role of the mPFC in the retrieval of consolidated memory may thus parallel the putative role of the hippocampus in retrieving recently formed memories, by completing the intercortical neural pattern from a partial cue. This role may be related to the involvement of human mPFC in memory retrieval (23, 24) or retrieval judgment, such as a feeling of knowing (2528).

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S6

Tables S1 to S6


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article