Prefrontal Neuronal Activities During Active Retrieval of Information From Long-Term Memory
Abstract
Abstract. Single-neuron studies performed in the primate prefrontal cortex (PFC) revealed that retaining information in working memory (WM) is associated with sustained firing during the delay period in a match-to-sample task. On the other hand, single-neuron studies using a pair-association task have shown that retrieving information from long-term memory (LTM) is related to two kinds of neural activities: decreasing activity representing information linked to the sample stimulus and increasing activity predicting information for the forthcoming matching stimulus. To further examine neuronal behavioral patterns during LTM retrieval, we used a partial correlation coefficient analysis to analyze single-neuron activities in the PFC while monkeys performed the visual pair-association task. Results showed that, for most of the task-related neurons, firing activity depicted information from the sample stimulus. Nevertheless, some neurons showed an opposite pattern, this is, increasing activity during the delay period, possibly indicating a prospective memory coding from LTM. Interestingly, both activities seem to be present at different degrees as the delay period progresses. Together, these results unveil a new aspect of PFC neurons when retrieving unseen information from LTM.
Single-neuron studies performed in the prefrontal cortex (PFC) of non-human primates have shown that PFC neurons exhibit sustained activities during the delay period in a match-to-sample task. These delay activities have been considered the neural correlates of working memory (WM). Therefore, constant firing during the delay period reflects a neural mechanism for holding information actively for a short period of time (Funahashi et al., 1989; Funahashi & Andreau, 2013; Klingberg, 2010; Constantinidis et al., 2018). Nevertheless, the fact that this delay activity persists from tens of milliseconds to seconds (after a neuron integrates its synaptic inputs) allows us to hypothesize that during this period, more complex network dynamics (Amit & Brunel, 1997; Compte et al., 2000) or intrinsic cellular properties (Zylberberg & Strowbridge, 2017) could be taking place. Variations of the delay’s magnitude over time were largely overlooked owing to analyses that averaged population activity (Funahashi et al., 1989; Rainer et al., 1999).
Other than WM memory processes, inferior temporal cortex (ITC) neurons also showed an increased firing rate along the delay period during the retrieval of information from long-term memory (LTM). In their seminal work, Sakai and Miyashita (1991) trained monkeys to perform a delayed pair association (PA) task in which a first stimulus (sample) was followed by a delay period, and then a matching stimulus (its pair associate) or a distractor (a non-related stimulus) was presented. They found neurons that exhibited a gradually increasing activity during the delay period only when the paired associate of the sample stimulus generated strong firing responses during sample presentation, too (the cell’s preference for the stimulus is observed during the sample period and the delay period activity is viewed as an expectation to perceive it). They referred to these cells as “pair-recall” neurons. Therefore, it has been hypothesized that the magnitude of the delay activity between the sample and the matching stimulus is determined by the neuron’s selectivity for that particular pair. Even though the nature of these pair-recall neurons is not yet clarified, it was suggested that it might be related to the memory retrieval dynamics in its interaction with PFC (Miyashita, 2004).
In the PFC, neurons exhibited two patterns of activities at a population level while monkeys performed a PA task. The first pattern is characterized by a strong firing activity right after the sample presentation, which diminishes toward the end of the delay period. This pattern was called sensory-related activity (Christophel et al., 2017; Rainer et al., 1999) or retrospective activity (Erickson & Desimone, 1999), and it is thought to represent the information of the sample stimulus during the delay period. The second pattern of activity refers to a gradually increasing firing rate which is maximal towards the end of the delay period. This activity is similar to the “pair recall” neurons. It has been proposed that this activity represents prospective information (Rainer et al., 1999; Erickson & Desimone, 1999), depicting information related to the paired associate of the sample stimulus.
Based on these data, we could argue that during the PA performances, PFC neurons seem to temporarily hold sample information or retrieve relevant information from the LTM during the delay period. But why does the activity fade or increase throughout the delay? What is the exact function of delay activity in the retrieval of associative memories? If the association between stimuli is separated by a delay, then a likely function of the sensory-related activity would be to hold sample stimulus information until the matching/distractor stimulus is presented, but why would it vanish in the process? On the other hand, prospective delay activity may be the neural representation of the expected stimulus guided by memory. But when and how does that information start to develop? Unfortunately, these questions remain unanswered.
The time length of the delay period seems to be a key variable in understanding neural activity in memory retrieval. In their now classic work, Kojima and Goldman-Rakic (1982) found that delay activity differed from typically sustained firing when the delay was lengthened from 2 to 8 s in a spatial delayed response task. About 33% of the delayed-related PFC neurons changed their latency, duration, peak position, and/or pattern of firing as the delay lengthened. They interpreted that activity as “anticipatory” or “expectancy” related to the matching stimulus. Therefore, a long delay period (as the 5 s used in Sakai & Miyashita) would be preferable to get more detailed information on changes in the firing rate during memory retrieval.
Although delay-period activities representing either retrospective or prospective information were observed at a population level, it is not clear whether this would also be the case if the activity is analyzed individually. A growing number of current studies started to focus on the “changing” activity from a single unit while a task is performed. According to them, single-unit activity represents more complex information than simple sustained or prospective activity (Gupta et al., 2020). For example, during a PA task, the delay period activity could depict both processes simultaneously or altered both retrieval strategies.
In the present study, we trained a monkey in a PA task with a 5 s delay period and analyzed single-neuron activities from the PFC. To solve the task, the monkey should remember the pair associated with the sample stimulus. It could either keep cue information in WM or retrieve information from LTM or both activities intermixed. To unveil the meaning of neuronal activity underlying these processes, we used a suitable analysis to study sample-holding and prospective-code activity developed by Naya et al. (2003) and apply it to PFC neurons. Results would show a glimpse of the complexity behind the single-unit activity of memory retrieval.
Methods
Subjects and Apparatus
One adult Japanese monkey (Macaca fuscata, KU-46, 8 kg) was used as a subject. The monkey was housed in an individual home cage. The monkey’s intake of water was restricted in the home cage. However, the monkey was given its daily requirement of water in the laboratory as a reward. When the monkey obtained insufficient water in the laboratory, additional water and fruit were given in its home cage to ensure that it remained healthy.
The instruments used for this experiment and surgical procedures were the same as in our previous experiment and described in detail in our previous report (Andreau & Funahashi, 2011). Before the start of the experiment, MRI (magnetic resonance imaging) pictures of the monkey’s brain were taken to determine the location of the recording area in the prefrontal cortex (the cortex within and surrounding the principal sulcus). Under aseptic conditions, we performed two surgeries. The monkey was first immobilized by the intramuscular injection of ketamine and then anesthetized by an intravenous injection of pentobarbital sodium (max 25 mg/kg body weight). Using the technique described by Judge et al. (1980), the eye coil was placed under the conjunctiva of the right eye of the monkey. In addition, a head-holder made of stainless steel was placed on the skull to restrict the monkey’s head movements during the experiment. To secure the head-holder, several stainless steel bolts were implanted in the skull. The connector for the eye coil, the head holder, and the stainless steel bolts was fixed to the skull with dental acrylic. During and after the surgery, the monkey was treated with antibiotics (cephalosporin, Astellas Pharma, Tokyo, Japan) to prevent infection. After training was complete, we performed the second surgery to place a recording cylinder on the skull. The procedures were similar to those used in the first surgery. To record single-neuron activity, we made a small hole (20 mm in diameter) in the skull over the center of the recording area using a trephine and attached a stainless steel recording cylinder. The stereotaxic coordinates of the center position of the recording cylinder (32 mm anterior from the interaural axis and 16 mm lateral from the midline) were determined based on MRI pictures of the monkey’s brain. Several stainless steel bolts were implanted in the skull and fixed by dental acrylic with the recording cylinder. During and after the surgery, the monkey was again treated with antibiotics to prevent infection. The monkey was given a full amount of food and water until fully recovered from the surgery. We performed this experiment according to the guideline based on the Guide for the Care and Use of Laboratory Primates published by the Primate Research Institute, Kyoto University. The experiment was approved by the Animal Research Committee at the Graduate School of Human and Environmental Studies, Kyoto University.
Pair-Association (PA) Task
The monkey was trained to perform a PA task. Details of this task were described in Andreau and Funahashi (2011). Briefly, after the monkey pressed a lever, a fixation spot appeared at the center of the monitor. The monkey was required to look at this fixation spot. After the 1-s fixation period, one sample stimulus was presented for 0.5 s over the fixation spot. The sample stimulus was randomly selected from the 24 visual stimuli prepared for the PA task. After the sample presentation, the monkey was required to look at the fixation spot during the 5-s delay-1 period. At the end of delay-1, another visual stimulus was presented on the monitor (decision period). The stimulus was either a visual stimulus that was a correct pair of the sample stimulus (matching stimulus) or any of the remaining 22 stimuli that were not paired to the sample stimulus (distractor stimulus). If the matching stimulus was presented, the monkey was required to release the lever within 0.5 s to get a juice reward (Go condition). If a distractor stimulus was presented, the monkey was required to continue to press the lever until the end of the 1-s delay-2 period (No-Go condition). When the delay-2 was over, the matching stimulus (forced-choice period) appeared, and the monkey was required to release the lever within 0.5 s to get the reward (Figure 1). Even though the critical decision was made during the decision period, the force choice period was added to be sure that the monkey was paying attention. We arranged Go and No-Go trials in such a way that the number of Go trials was nearly equal to the number of No-Go trials within one experimental session. Any of the 24 stimuli prepared for the PA task could appear as a sample stimulus in an A-B and B-A fashion. Trials were defined as an error when the monkey either broke fixation or released the lever during the delay period or when the monkey made an incorrect response during the decision period or the forced-choice period.
The task was trained through shaping in which the monkey learned several steps: (1) press the lever to get reward, (2) press the lever and look at the sample stimulus and release the lever to get reward, (3) press the lever, look at the sample stimulus and keep pressing the lever after the sample stimulus disappears and a second stimulus (pair associate) is presented and then release the lever to get reward, (4) release the lever only when a specific second stimulus appears (pair associate) and not when any other second stimulus (distractor) appears. The learning of the whole task with the 24 pairs took around 4 months to be completed. The criterion for learning was a proportion correct of 80% or above in both Go and No-Go conditions.
Recording and Analyzing Single-Neuron Activity
Single-neuron activity was recorded by glass-coated elgiloy microelectrodes (0.5–2.0 MΩ at 1 kHz). The raw activity was amplified using an amplifier (DAM80, WPI, Sarasota, FL) and monitored visually on an oscilloscope (SS-7802, Iwatsu, Tokyo, Japan) and by an audio monitor. We isolated single-neuron activity using a window discriminator (DIS-1, BAK Electronics, Mount Airy, MD). The output from the window discriminator was read out and stored in a computer by TEMPO (Reflective Computing, Olympia, WA).
The present study focused on neural activity during the delay-1 period of those neurons showing selective activity for particular sample pictures. Stored single-neuron data were analyzed offline using custom-made programs with MATLAB (MathWorks, Natick, MA). For statistical analysis, we also used the statistical software R 3.6.0 (R Core Team, 2019). To determine whether the activity was task-related or not, we first calculated a baseline discharge rate for each neuron. The baseline discharge rate was defined as the mean discharge rate during a 400-ms period before the sample stimulus onset. Then, we calculate discharge rates during the sample period for each trial. We defined sample-period activity as the discharge rate during a 200-ms period from 100 to 300 ms after the sample stimulus presentation. We performed a Wilcoxon’s signed-rank test to determine whether a neuron exhibited significant sample-period activity or not. If discharge rates during the sample period were significantly different (p < .05) from the baseline discharge rate, we defined that the neuron had sample period activity. To determine whether or not neurons with sample-period activity exhibited stimulus selectivity, we compared the mean discharge rates in response to 24 sample stimuli by one-way analysis of variance (ANOVA) and later by Tukey’s honestly significant difference (HSD) test as a post hoc analysis. If a statistically significant difference (p < .01) was present among the responses to the 24 sample stimuli, we considered that the neuron exhibited stimulus selectivity in sample-period activity. Using this analysis, we also determined which stimulus produced the maximum sample-period activity and which stimulus produced second maximum sample-period activity. There are two main reasons to choose sample period activity for the data analysis. The most important one is that during the sample period, the subject perceives the stimulus in a “pure” manner. This means that the cell’s response is more likely specific to the stimulus and is not contaminated with other underlying cognitive or motor processes. On the other hand, during the decision period, the firing rate produced by the paired-associate could be mixed with several processes required to solve the task (e.g., monkey’s reward expectations, decision making, motor preparation, the memory of other stimuli, etc.). The second reason is that most of the previous studies examining pair selectivity in single neurons analyzed the sample period activity to the point that the formula used in the present paper was taken from Naya et al. (2003) who also analyzed the sample period activity to depict sensory-related and prospective activity.
When visually inspecting these stimulus selective neurons, we found a very clear pattern of delay-1 period activity directly related to the sample-period activity. To roughly estimate the temporal pattern of delay-1 period activity, we first divided the delay-1 period into two parts: delay-1 first epoch (D1FE) and delay-1 second epoch (D1SE). We then were able to observe delay period activities resembling both sensory related and prospective coding activity (Figure 2).
Delay-1 Period Analysis
The originality of our approach was the use of a longer delay-1 period (5 s) than previous single unit LTM retrieval studies using the PA task. It would be as if we can make the 1 s analysis performed by Rainer et al. (1999) over five periods of delay. This analysis would give us substantial information regarding a long delay period similar to the one in which pair recall neurons were found (Sakai & Miyashita, 1991). Moreover, we draw on the analysis made by Naya et al. (2003) to enrich the information obtained from the recorded neurons. The first step of the analysis was to divide our 5 s delay-1 period into 5 parts of 1 s each: Dly1 = 0–1 s, Dly2 = 1–2 s, Dly3 = 2–3 s, Dly4 = 3–4 s, and Dly5 = 4–5 s (Figure 3).
The next step was to apply a formula developed by Naya et al. (2003) to each one of these subdivisions of the delay-1 period. The rationale behind this analysis was that since this formula calculates an index of the delay-1 activity associated with the sample (sample-holding-index) or associated with the recall of the pair associate (pair-recall-index), we could know if the single-unit activity observed during a given part of the delay was at the serve of keeping visual information active or retrieve studied LTM information.
For that purpose, the delay-1 divisions of the set of 24 sample stimuli were denoted as a 24-dimensional vector D: [d1, …, d24] or Dp [dp(1), …, dp(24)]. The sample-period responses were denoted as S: [s1, …, s24] or Sp [sp(1), …, sp(24)], where the ith and p(i)th stimuli belong to a pair. To define the sample-holding index (SHI) and the pair-recall index (PRI), partial correlation coefficients of D with the sample-period responses to the corresponding sample stimuli S and the paired associate stimuli Sp were calculated using standard formulae (Erickson & Desimone, 1999; Movshon & Newsome, 1996; Naya et al., 2003):
where 〈A|B〉 indicates a simple correlation coefficient between A and B. Note that 〈S|D〉 = 〈Sp|Dp〉 and 〈Sp|D〉 = 〈S|Dp〉. SHI and PRI values are bounded by −1 and 1. If a single neuron in a population showed the pattern of stimulus selectivity during the delay period that was independent of the pattern of stimulus selectivity during the sample period, the mean values of the PRI and the SHI for the neuronal population would be expected to approach to zero as the number of neurons in the population increased.
Results
Behavioral Performance
In the present task, the monkey was required to decide whether the presented stimulus was a matching stimulus or a distractor stimulus (as compared to the sample) during the decision period. To make a correct decision, the monkey needed to retrieve specific information from LTM during the delay-1 period. The average percentage of correct responses was 97% for Go trials and 94% for No-Go trials. This tendency was observed throughout the recording sessions. These results indicate that monkey KU-46 performed the PA task correctly and solved the task using the learned association of visual stimuli. The mean latency of lever release was 343 ± 19 (SD) ms in the Go trial (the decision period) and 220 ± 49 (SD) ms in the No-Go trial (forced-choice period). This difference in the mean latency of lever release was statistically significant for the average of the sessions corresponding to the neurons selected (Student’s t-test, t(24) = 17.17, p < .01).
Sample-Period Activity
We recorded the activities of a total of 172 neurons from the PFC of monkey KU-46 (163 from the left hemisphere and 9 from the right hemisphere). Among the neurons analyzed, 60 showed statistically significant sample-period activity. The magnitude and temporal pattern of sample-period activity differed within each neuron, indicating that PFC neurons respond selectively to visual stimuli. To further analyze the characteristics of the stimulus selectivity observed in sample-period activity, we selected neurons for which the activity was recorded in at least 5 trials per stimulus (meaning at least 120 trials per session). Of 60 neurons with a sample-period activity that were recorded, 43 fulfilled this criterion. We then examined the stimulus selectivity of sample-period activity by comparing the mean discharge rates of sample-period activity among 24 stimuli using one-way ANOVA (p < .01) and Tukey’s HSD test as a post hoc test. As a result, 25 neurons were classified as neurons having stimulus selectivity. We based our delay-1 period analysis on those 25 sample selective neurons. To analyze delay-1 period activity, we first subtracted baseline activity from the sample period and each of the 5 parts of the delay (to normalize the activity to the baseline). After the subtraction, we applied the formula taken from Naya et al. (2003). Therefore, we obtained 5 SHI values (expressing the way in which each part of the delay period is representing the information related to the sample period activity) and 5 PRI values (expressing the way in which each part of the delay period represents the information related to the pair associate) for each of the 25 visual selective neurons. The differences between SHI and PRI would give us information regarding the meaning of the information during that section of the delay-1 period. For example: if PRI > SHI, that activity is related to the pair associate (prospecting coding type), if SHI > PRI, that activity is related to the sample information (sample holding type), and finally, PRI = SHI would mean ambiguous activity. The results can be seen in Table 1 and Figure A1 in the Appendix.
Results from Table 1 and Figure A1 show a clear tendency of the delay-1 period in all its subdivisions to depict activity related to the sample stimulus (SHI) compared to prospecting coding (PRI). When we plot the average of SHI and PRI values on a single graph, we can observe an interesting pattern (Figure 4). While SHI index values decrease along the subsequent delay periods, PRI index values show the opposite pattern. This is, their values increase along with the delay-1 period subdivisions. In this way, index values for SHI are bigger during Dly1 as compared to Dly5, and index values for PRI are smaller during Dly1 as compared to Dly5. Actually, 83% of the analyzed neurons showed a decreasing pattern for SHI, whereas 67% of the neurons showed the opposite pattern for PRI values.
To further test the strength of that difference, we performed a one-way ANOVA on all five SHI and PRI values. Results showed that there was a significant difference in the SHI group, F(4) = 3.02, p < .05. Tukey HSD post hoc test showed that index SHI1 was significantly higher (more stimulus related) than index SHI5, p < .05. On the other hand, the PRI group showed no significant differences.
Discussion
Studies of the neural correlates of memory processes at a single unit level have reported a sustained discharge rate from PFC neurons during a delay period in a delayed match to sample task (Funahashi et al., 1989; Miller et al., 1996; Takeda & Funahashi, 2002). Single unit activity related to WM has also been found in the ITC during short-term memory tasks (Miller et al., 1993).
Interestingly, studies of neuronal activity during LTM retrieval are scarce. In 1991, Sakai and Miyashita described a “pair-recall” neurons in ITC during a delay period in a PA task, but the nature of that activity remains unclear. For example, Naya et al. (2003) found that neurons in one region of the ITC (called area TE) mostly represented a prospective coding activity while another region (known as A36) showed a similar pattern to TE neurons plus sample holding activity. Moreover, previous single-unit studies of memory retrieval analyzing activity from ITC highlighted the importance of investigating the origin and meaning of that activity since the ITC is regarded as the storehouse of visual information (Miyashita, 2004; Naya et al., 2003). Regarding PFC neurons, sample holding and prospective coding activities were found in a PA task at a population level and with a short delay period (e.g., 1 s in Rainer et al., 1999).
Here, we analyzed single unit activity in PFC during a LTM retrieval task with a 5 s delay period. The combination of all these variables allowed us to further explore the delay activity based on approaches used for ITC neurons (Naya et al., 2003). We based our analysis on neurons with stimulus selectivity. It is a fact that the number of cells analyzed (n = 25) constitutes a limitation to the results. Nevertheless, usually, task-related neurons found in single-cell studies do not represent a big proportion (“pair recall” neurons were found only in 11% of all recorded neurons in Sakai & Miyashita, 1991) compared to all the cells recorded. This happens because the recording approach is to record any neuron with clear activity disregarding if it is task-related or not. Even though our results are far from being conclusive, they could definitely provide interesting insights into the actual theories on how neurons work to create and retrieve information from LTM. In this way, this paper might help future research on this topic, and it may contribute to new discoveries in the field.
For the present study, we found that, when dividing a 5 s delay into 5 parts of 1 s, each part shows different values in a specific index which calculates how the delay activity represents the sample stimulus information (SHI index) or to what extent that same delay period represents the information related to the pair associate of the sample (PRI index). Results showed that for most of the neurons, all delay parts reflect sample stimulus information (SHI stronger than PRI values). When observing the pattern of each of the 25 neurons individually, we noticed that (except for neuron L06501) SHI is always bigger than PRI at Dly1 period. On the other hand, 9 neurons (36%) show the opposite pattern at Dly5 period. From the Dly2 to Dly4 periods, the most frequent pattern is SHI bigger than PRI along with the delay. This is true for 10 neurons (40%). For the remaining 15 neurons (60%), all sorts of combinations can be seen (Figure A1). These mixed combinations could be an indication of the complexity of the communication between several neural circuits located in different brain regions. The present paper constitutes an attempt to understand the hidden meaning behind the vanishing or increasing activity present during the delay period in PA tasks. The action potentials might be depicting this changing information related to the neural circuitry involved in memory retrieval, therefore, the recorded firing rate represents different purposes as the task goes forward. It would be reasonable to hypothesize that memory-related processes go back and forth from PFC to posterior cortices since information regarding modality-specific WM interacts with general purpose WM. Interestingly, on the average of the 25 neurons analyzed, SHI activity seems to diminish as the delay stretches in time (Figure 4). Being significantly bigger during the first part of the delay as compared to the last part of the delay. This might suggest that the first part of the delay period strongly represents information related to the sample stimulus, but as the delay approaches the pair associated with that sample, the activity fades away. Furthermore, the PRI index showed the opposite pattern as compared to the SHI index. This is an increase in the value as the delay period comes to an end. This pattern might signal the fact that delay activity is turning from representing sample information to start representing prospective information retrieved from LTM.
If the neural representation of the sample is active during the presentation of the matching stimulus, this simultaneous activation of different populations could reactivate Hebbian connectivity created during learning. If the firing rate is vanishing toward the end of the delay, this could mean that its goal was already achieved. Now, activity related to the unseen pair associate needs to be activated and maintained until the matching stimulus presentation. This relay might take place at different moments during the delay period (see Figure A1). Importantly, even though our results are based on the pair-related activity taken from the sample period, perhaps a thorough analysis of the decision or matching period activities could also contribute to defining a neuron’s stimulus preference.
Recent research has confirmed that single-unit activity interpretation might not be as straightforward as it seems (Gupta et al., 2020). For example, transmitted information could be influenced by the computation performed by the postsynaptic partners (Felleman & Van Essen, 1991). Also, the alternation between activity patterns corresponding to different items could be used as a strategy to enhance the brain processing capacity (Caruso et al., 2018).
Tomita et al. (1999) found that, in order to retrieve LTM information from ITC, a top-down signal must be excerpted from the PFC. Therefore, PFC orchestrates the active LTM retrieval by sending these signals and receiving information from ITC (mainly when talking about visual information in monkey studies). The importance of the PFC in LTM processes is supported by neuropsychological and neuroimaging studies (Andreau, Idesis, et al., 2020; Andreau, Torres Batán, et al., 2020; Blumenfeld & Ranganath, 2007; Buckner et al., 1999; Nolde et al., 1998). It is known that modality-specific WM is a mechanism for temporarily storing and processing one domain of information (e.g., sensory), while a general-purpose WM monitors and controls the activities of the brain areas that perform modality-specific WM (Funahashi & Andreau, 2013). The PFC is the brain region in charge of this general-purpose WM, and the way in which it modulates activity in different cortical areas is through top-down signals. Therefore, it is possible that activity found during the delay period in ITC is related to modality-specific WM, whereas PFC activity during the same period represents general-purpose WM, the constant interaction between these two processes would lead to a complex firing pattern which requires further analysis to be unveiled.
Typically, increased activity during the delay period would indicate searching for information from LTM, whereas a decreasing activity would indicate information regarding the cue stimulus being kept active (Rainer et al., 1999). Our results might indicate that it is not as clear as it seems. By virtue of the introduction of a 5 s delay period, we were able to divide a delay period into 5 parts of 1 s each and analyze them independently and individually. This possibility gives us richer information as the delay period extends through time. At first glance, it looks like the delay activity reflects the information regarding the sample stimulus, but as the delay extends through time, that activity seems to reflect less and less of the sample information. At the same time, the activity depicting the pair associated with the sample stimulus starts to increase with the delay, possibly depicting a prospective coding. What is interesting is that these two patterns can be observed in the same data, meaning that both cue holding and prospective coding activities could coexist even at a single neuron level, showing the complexity involved in a cognitive process, as is LTM retrieval. Further research might clarify the nature of the neural activity related to LTM retrieval.
References
(1997). Model of global spontaneous activity and local structured activity during delay periods in the cerebral cortex. Cerebral Cortex, 7(3), 237–252. https://doi.org/10.1093/cercor/7.3.237
(2011). Primate prefrontal neurons encode the association of paired visual stimuli during the pair-association task. Brain and Cognition, 76(1), 58–69. https://doi.org/10.1016/j.bandc.2011.03.001
(2020). Unraveling the electrophysiological activity behind recognition memory. Journal of Psychophysiology, 35(2), 76–88. https://doi.org/10.1027/0269-8803/a000263
(2020). The time course of associative memory retrieval for pictorial stimulus: An exploratory event-related potential study. Psychology & Neuroscience, 13(1), 32–50. https://doi.org/10.1037/pne0000186
(2007). Prefrontal cortex and long-term memory encoding: An integrative review of findings from neuropsychology and neuroimaging. The Neuroscientist, 13(3), 280–291. https://doi.org/10.1177/1073858407299290
(1999). Frontal cortex contributes to human memory formation. Nature Neuroscience, 2(4), 311–314. https://doi.org/10.1038/7221
(2018). Single neurons may encode simultaneous stimuli by switching between activity patterns. Nature Communications, 9(1), 1–16. https://doi.org/10.1038/s41467-018-05121-8
(2017). The distributed nature of working memory. Trends in Cognitive Sciences, 21(2), 111–124. https://doi.org/10.1016/j.tics.2016.12.007
(2000). Synaptic mechanisms and network dynamics underlying spatial working memory in a cortical network model. Cerebral Cortex, 10, 910–923. https://doi.org/10.1093/cercor/10.9.910
(2018). Persistent spiking activity underlies working memory. Journal of Neuroscience, 38(32), 7020–7028. https://doi.org/10.1523/JNEUROSCI.2486-17.2018
(1999). Responses of macaque perirhinal neurons during and after visual stimulus association learning. Journal of Neuroscience, 19(23), 10404–10416. https://doi.org/10.1523/JNEUROSCI.19-23-10404.1999
(1991). Distributed hierarchical processing in the primate cerebral cortex. Cerebral Cortex, 1(1), 1–47. https://doi.org/10.1093/cercor/1.1.1-a
(2013). Prefrontal cortex and neural mechanisms of executive function. Journal of Physiology-Paris, 107(6), 471–482. https://doi.org/10.1016/j.jphysparis.2013.05.001
(1989). Mnemonic coding of visual space in the monkey’s dorsolateral prefrontal cortex. Journal of Neurophysiology, 61(2), 331–349. https://doi.org/10.1152/jn.1989.61.2.331
(2021). A hybrid multi-criteria decision-making approach to order natural gas consuming countries. Independent Journal of Management & Production, 12(2), 691–710. https://doi.org/10.14807/ijmp.v12i2.1287
(2020). A single-neuron: Current trends and future prospects. Cells, 9(6), Article
1528 . https://doi.org/10.3390/cells9061528(1980). Implantation of magnetic search coils for measurement of eye position: An improved method. Vision Research, 20, 535–538. https://doi.org/10.1016/0042-6989(80)90128-5
(2010). Training and plasticity of working memory. Trends in Cognitive Sciences, 14(7), 317–324. https://doi.org/10.1016/j.tics.2010.05.002
(1982). Delay-related activity of prefrontal neurons in rhesus monkeys performing delayed response. Brain Research, 248(1), 43–50. https://doi.org/10.1016/0006-8993(82)91145-3
(1996). Neural mechanisms of visual working memory in prefrontal cortex of the macaque. Journal of Neuroscience, 16(6), 5154–5167. https://doi.org/10.1523/JNEUROSCI.16-16-05154.1996
(1993). Activity of neurons in anterior inferior temporal cortex during a short-term memory task. Journal of Neuroscience, 13, 1460–1478. https://doi.org/10.1523/JNEUROSCI.13-04-01460.1993
(2004). Cognitive memory: Cellular and network machineries and their top-down control. Science, 306, 435–440. https://doi.org/10.1126/science.110186
(1996). Visual response properties of striate cortical neurons projecting to area MT in macaque monkeys. Journal of Neuroscience, 16(23), 7733–7741. https://doi.org/10.1523/JNEUROSCI.16-23-07733.1996
(2003). Delay‐period activities in two subdivisions of monkey inferotemporal cortex during pair association memory task. European Journal of Neuroscience, 18(10), 2915–2918. https://doi.org/10.1111/j.1460-9568.2003.03020.x
(1998). Left prefrontal activation during episodic remembering: An event-related fMRI study. NeuroReport, 9(15), 3509–3514. https://doi.org/10.1097/00001756-199810260-00032
(1999). Prospective coding for objects in primate prefrontal cortex. Journal of Neuroscience, 19(13), 5493–5505. https://doi.org/10.1523/JNEUROSCI.19-13-05493.1999
. (2019). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/
(1991). Neural organization for the long-term memory of paired associates. Nature, 354(6349), 152–155. https://doi.org/10.1038/354152a0
(2002). Prefrontal task-related activity representing visual cue location or saccade direction in spatial working memory tasks. Journal of Neurophysiology, 87(1), 567–588. https://doi.org/10.1152/jn.00249.2001
(1999). Top-down signal from prefrontal cortex in executive control of memory retrieval. Nature, 401(6754), 699–703. https://doi.org/10.1038/44372
(2017). Mechanisms of persistent activity in cortical circuits: Possible neural substrates for working memory. Annual Reviews of Neuroscience, 40, 603–627. https://doi.org/10.1146/annurev-neuro-070815-014006
Appendix
