prefrontal cortex short-term potentiation model for working memory

Posted comment on ´A Spiking Working Memory Model Based on Hebbian Short-Term Potentiation` by F. Fiebig and A. Lansner and published in Journal of Neuroscience 37(1) p. 83:


Researchers hypothesize that the encoding and maintenance of informational items in working memory requires persistent elevated activity of neural networks in the prefrontal cortex (PFC). The models involve short-term, non-associative synaptic plasticity (non-Hebbian) producing an active buffer with periodic reactivations refreshing the decaying synaptic firing hence, retaining the memory. However, the models cannot explain encoding of novel associations since this type of learning, according to Fiebig and Lansner is presynaptic, which implies all the outgoing synapses of the active neuron are enhanced. The authors of this article, Fiebig and Lansner, re-examined experiments where there are single informational units during delay periods and showed that synaptic activity of the prefrontal cortex area was instead more variable. They found discrete gamma bursts of activity associated with the information of multiple items in working memory. As a result, Fiebig and Lansner suggested a recently identified fast-expressing form of Hebbian synaptic plasticity (called associative Short-term Potentiation – STP) was involved. STP occurs after brief high frequency bursts and decays not with time, but in an activity dependent manner. Therefore, the authors suggested that memory reactivation occurs in discrete oscillatory bursts rather than with sustained neuronal activity.

For their experiments on the cortex, Fiebig and Lansner used a NEST and a computational model of the cortical layer 2/3 network consisting of 16 hyper-columns (HC) with 5760 pyramid cells and 384 inhibitory basket cells. Each HC had 24 basket cells and the pyramid cells were divided into 12 functional multi-columns (MC) of 30 pyramidal cells each. The HCs were laid out on a hexagonal grid. Fiebig and Lansner computed the axonal delay between the presynaptic neuron (termed i) and post-synaptic (termed j) based on known conduction speeds and distances between the MCs. The pyramidal cells of the HC exhibit lateral AMPA mediated excitatory projections to the basket cells with a calculated connection probability which they designated ppb. The pyramidal cells received inhibitory feedback via the GABA mechanism from basket cells with a connection probability defined as pBp. The pyramidal cells also formed NMDA and AMPA connections within the HCs and across them with a connection probability termed ppp. The entire model was described by the authors as demonstrating appropriate plastic neuronal connections and featured a total of over 13 million synapses between pyramidal cells, over 100,000 excitatory connections from pyramidal cells to basket cells in their respective HCs as well as many inhibitory connections. Fiebig and Lansner used a modified AdEX IAF neuron model with a spike frequency adaptation. This was compatible with a custom made BCPNN synapse model in NEST through the addition of the intrinsic excitability current they designated Ibetaj. AMPA and NMDA synapses were modeled with a spike-based version of the Bayesian Confidence Propagation Neural Network (BCPNN) learning rule so that presynaptic and postsynaptic rates could be calculated as well as co-activation.

Memory performance was assessed by cued and free recall word tests. For the latter, pattern activation was counted and for the former, half of each cued pattern was counted and then checked to see whether the pattern was fully activated afterwards. The authors also performed behavioural studies of 2 types: Study A used the Betula Study of a large battery of cognitive tests including study and immediate free recall of a word list; and in Study B the cued recall used was the word recall study of Gershberg and Shimamura (1994).

In the case of the test of single item memory encoding and performance of the free recall task (termed Demonstration 1) and carried out using a delayed match to sample task (delayed free recall with one item), Fiebig and Lansner found that the ground state for the first second was characterised by low-rate (1.7HZ),  irregular, asynchronous firing of pyramidal cells whereas local basket cells were found to often spike together even if not synchronizing the firing activity globally. Targeted stimulation from 1-2 secs onwards of one MC in each HC led to increases of firing in the stimulated pyramidal population which led to rapid bursting of the local basket cells. These then inhibited all the neurons in their HC resulting in lower firing of non-specific cells. The authors interpreted this as the firing network counterbalancing the increased activity in some MCs by decreasing the firing of neighbouring populations. Approx. 1 sec after the beginning of the stimulation brief spontaneous reactivations of the originally stimulated MCs began.  These reactivations were found to be gamma oscillation bursts and synchronization of the firing cells was attributed to fast feedback inhibition and short connection delays in the excitatory associative connections. The specific firing rate of the pattern activated cells was approx. 25Hz and stable for the duration of reactivation. After 120msecs, the evoked firing was terminated due to synaptic depression, but spontaneous reactivations were still observed resulting in a pattern of repeated spontaneous attractor reactivations in discrete oscillatory bursts similar to Lundqvist`s  view, but based on Hebbian STP as a result of the new learning. The non-stimulated pyramidal population showed decreased firing rate both during and after stimulus.

Fiebig and Lansner also performed experiments to show how the network learned and could simultaneously store larger numbers of items indicative of multiple item working memory. This list learning without intermittent replay task was designated Demonstration 2. In the first 20 secs the network ground state consisted of background plus evoked irregular spiking of pyramidal cells and triggering of basket cells. The network was then stimulated with 12 patterns of 1 second each with intervals of 500msecs. Unspecific background activity decreased as a result of competing neural event such as the working memory task with distractor task and attention diverted to abolish active maintenance. The authors found that as learning progressed, the network encoded the properties of the structured input. A linear relationship between how recently a pattern had been trained and the excitability of the relevant pyramidal member neurons was found since recently active neurons featured a less negative bias current than neurons that had been silent for a longer time. Memory performance during the free recall phase was measured by tracking the autonomous attractor reactivations and Fiebig and Lansner found that reactivations of recently trained attractors dominated (although only the last 5 patterns) whereas earlier patterns were reactivated by cued recall.

In Demonstration 3, Fiebig and Lansner examined the PSPs of successful attractor memory activation. They found that patterns were excited to spike at 40HZ via one of 3 separate scenarios: one type of presynaptic neuron in the same MC; a presynaptic neuron in another MC; or a basket cell. They averaged several hundred recorded post-synaptic traces to give isolated PSP test results and found that synapses may depress so the peri-stimulus PSP magnitude depended on the duration of the 40 Hz input (25 ms interspike interval). At the ground state near −67 mV, EPSP amplitudes were initially large, but quickly depressed. Strong inhibition incurred from presynaptic basket cells and these were found not to depress.

Fiebig and Lansner also performed two Simulation studies. In Simulation Study 1, multi-item working memory was tested using list learning with intermittent replay. Twelve items were presented one every two seconds, but the background activity rate was not manipulated by the authors so reactivation could occur during the inter-stimulus interval. Therefore, early patterns were observed and could strengthen during the intermittent reactivations occurring in the learning period. This was attributed to ´memory refresh` or short-term memory consolidation. Free recall testing produced greater results with 5-8 patterns recalled and it was found that primacy and recency rules applied. The authors found that the mean firing rates of patterns eventually recalled increased from 1.3HZ at baseline to 2.7HZ after learning. In Simulation Study 2, cued recall in word list learning was examined. Here, Fiebig and Lansner found a recency effect. However, the weak middle position that exhibited a free recall probability of only 20-30% could be improved by cueing recall. Recall success then rose to 80%.

In summary, Fiebig and Lansner in their experiments supported  a functional cortical working memory model where there is fast Hebbian synaptic plasticity of pyramidal cells (STP) representing the stimulated cells with reactivations of a gamma oscillation burst nature accompanied by fast, basket cell mediated feedback inhibition. Hence, it appears that the encoding, maintenance and reactivation of working memory information occurs from limited sustained activity and then discrete oscillatory bursts rather than persistent activity of the relevant cells. The authors were able to expand their work by showing that verbal memory learning and recall could be shown by their model and was limited to 5-8 patterns recalled. An increase in frequency from 1.3HZ to 2.7HZ was shown with this learning. Primacy and recency effects were also demonstrated in both free and cued recall. Therefore, the authors concluded that fast Hebbian short term potentiation is a key mechanism in working memory.


What makes this article interesting is that it reaffirms the view that brain waves can indicate the type of functioning that a brain area is carrying out at the immediate time, or the state of the connectivity between areas at that time. Brain waves are seen because of a measurable level of synchronized firing of neurons and the changes in them with sleep states has been known for a long time. Later research has been associating the different types generally with different cognitive functions in the awake state, eg. thinking with beta oscillations, meditation with alpha, and even on a more smaller scale, synchronised waves of firing of particular cells within an area, eg. the theta clock-spiking cells of the hippocampus described in a previous post (March 2017 – Zhang and colleagues).

The article described in this post involved the authors looking at brain waves associated with what they described as working memory (WM). They concluded that working memory requires fast Hebbian synaptic plasticity of pyramidal cells (STP) of cortical layer 2/3 from stimulated cells with reactivations of a gamma oscillation burst nature accompanied by fast, basket cell mediated feedback inhibition. In this way, encoding, maintenance and reactivation of working memory information appeared to occur from limited sustained activity and then discrete oscillatory bursts rather than persistent activity of the relevant cells. However, my view is that the activation observed in their experiments cannot be described as working memory in its true sense, but can be instead attributed to non-complex learning and recall of information. This is because no manipulation of inputted information or recalled information was required by the nature of the experiments used. The experiments used instead demanded that word information was learnt and recalled exactly as learnt independent of whether the experiment was structured so that recall was free or cued. Therefore, the memory mechanisms required for such a task would be visual information input, learning (the necessary sensory store formation, short-term store formation and long-term store formation) and recall without processing (Salt, 2011) ie. neuronal firing of cell assemblies representing the coded information. Since manipulation of neither inputted material or recalled material was required, then the working memory capability could not in my view be said to be involved.

However, even if working memory was not involved, the experiments do show the brain wave patterns of the cortical layer 2/3 cells during the learning and recall stages of verbal lists (the basis of Fiebig and Lansner`s experiments). Such cognitive requirements demand the cooperation and connectivity of multiple brain areas and mechanisms. In the case of the information input, both visual and attentional pathways are activated and neuronal cell assemblies are built at the appropriate cortical area at the highest hierarchical level in order to represent electrically the stimulus. Connectivity between firing neurons and the patterns formed continue until one of three things happen: the stimulus is stopped (stimulus interval) as would be the case if a stimulus is continually or repetitively presented; the external stimulus is stopped, but firing is continued by internally induced means involving the PFC and/or hippocampus (theoretical, but observed in some cases for both areas); or the normal biochemical firing limits of the cell are reached (cell refractory period) and firing ceases until the cell biochemically recovers. The presentation of another word is unlikely to produce the same neuronal cell assembly and therefore, firing of a new grouping begins and continues in the same vein as the first. According to the authors of the article, firing of the ´old` assembly also continues intermittently as shown by the bursts of activity observed. This is what the authors say provide conditions that shift the electrical representation from sensory store/short term memory to long-term memory and not persistent cell firing as previously thought. This corresponds to the theoretical Option 2 described above where the external stimulus is stopped (the stimulus is now the new one ie. the new word being presented), but firing is continued by internally induced means involving the PFC and/or hippocampus. Biochemically, whether Option 1 or 2 occurs, means that there is a shift from short to long term memory and the short term physiological changes seen in each firing neuronal cell, eg. the rise in internal calcium ion concentration, change in pyruvate kinase activity cause long-term changes, eg. gene transcription alterations, higher receptor number. The overall effect is that the firing propensity of those cells within that group (ie. the cells of that group are more sensitive to firing if the same input is given) is increased and this represents the Hebbian plasticity cited by the authors.  Firing of areas is inhibited by GABAergic excitation as seen in other cortical and hippocampal systems.

As indicated above, in my view, intermittent firing could be explained by the firing of the entorhinal cortex and certain areas of the hippocampus (DG dentate gyrus leading to CA1, CA3) which can occur independently of external stimulus and possibly create the conditions required to shift the short-term memory formation to long-term memory formation in the absence of real-time stimulus repetition. Firing of PFC cells independent of external stimulus is also shown to occur and this would link in with its working memory and visuomotor functions for example.

The second part of Fiebig and Lansner`s experiments and the proof of successful learning is the recall of the word when asked. Recall can be free or cued. Free recall means that firing of the neuronal cell assembly representing one word will prompt the firing of another and this is brought about by the temporal connections (although in this case strictly speaking order) between one functional cell assembly and the next. We know ourselves that associating one word with another for whatever reason will increase the chance of remembering. Cued recall can be easier and hence, recall performance as the authors showed is greater because a characteristic of the event is used to prompt the firing of the whole assembly and perhaps others through association.

Fiebig and Lansner in their article link the firing of neuronal cell assemblies in the PFC with working memory performance although their experiments looked at cortical layer 2/3 firing with no definitive brain area given and working memory may not be strictly what was being observed. The cortical cells used could have come from other brain areas associated with the cognitive functions required, eg. post-parietal cortex known for item maintenance and manipulation of information, visuomotor areas. If samples were from the PFC then firing could be attributed to this area`s attentional function rather than working memory since PFC pyramidal cells are said to be multi-tasking (working memory and attention – work by Messenger). The authors looked at the firing of cell assemblies representing the words and found columnar cell assemblies consisting of excitatory pyramidal cells and inhibitory cells stimulated by GABA. Since they thought that long-term potentiation (LTP), which is the signal of Hebbian plasticity, requires a longer time and working memory is fast they interpreted their findings as LTP translating as long-term storage as consistent with biochemical explanation and working memory, the sensory store or short-term memory version. Fiebig and Lansner suggested that different early forms of LTP such as STP (short-term potentiation – un-stimulated, no presynaptic spikes) and E-LTP (NMDA dependent, independent of protein synthesis, possible presynaptic transmitter release, AMPA R phosphorylation by CaM-CaMKII receptor insertions, potassium from postsynaptic NMDA R activation as retrograde messenger for presynaptic STP induction) were responsible. In this case it is likely that E-LTP represents these early changes observed in the neuron as a result of firing continuously after neurotransmitter release and stimulation of the post-synaptic membrane. STP may represent the later stages as gene transcription occurs before the  post-synaptic changes in AMPA R number takes place traditionally associated with LTP. This is maybe why it needs no presynaptic stimulation since gene transcription changes have already been instigated.

Therefore, the biochemical changes taking place during the learning and recall processes have been explained, but Fiebig and Lansner used brain waves to demonstrate the functioning of the neurons. They found a ground state which consisted of low rate (1.7HZ), irregular firing of the pyramid cells plus basket cells of the chosen columns firing in synchronous bursts that were not global. One to two seconds after the presentation of the stimulus the authors found an increase in the population of stimulated pyramidal cells as expected and rapid bursting of basket cells which lead to the inhibition of the firing of the cells. Bursts of gamma wave reactivation occurred. After 120 seconds then there was a decrease in evoked stimulation due to neural firing depression because of the cellular refractory periods. Other researchers support the observation of bursts of gamma activity associated with working memory and that persistent activity is not necessary for short-term retention, but a general increase in overall activity for neurons fired in the working memory pattern.  This is different to other views where persistent activity is required for single item delayed match to sample tasks. The authors suggested that working memory manifests as multiple forms of firing activity eg bursting, persistent, fast neural sequences akin to synchronous firing chains and phase relationships.

In order to interpret the brain wave patterns observed by Fiebig and Lansner in their experiments, we have to look at the type of brain waves observed and their hypothesized associated biochemical function. The lowest frequency (1/2 to 3Hz), highest amplitude wave of synchronized neural activity is designated the delta wave and this is seen in Stage 3 (or 4 depending on how the stages are defined) of sleep. This stage of sleep (deep, delta or slow wave sleep – SWS) is characterized by its low frequency (0.5-4HZ) waves with large amplitude and appearance of some sleep spindles. It represents coordination of interregional cortical communication and acetylcholine receptors appear to be important. Sleep-active neurons located in the ventrolateral preoptic nucleus (VLPO) play a crucial role in the induction and maintenance (glucose effect on VLPO) of this stage and also are important in memory formation. During slow wave sleep, both GABAergic neurons of the nucleus reticularis thalami (NRT) and thalamocortical (TC) neurons discharge high-frequency bursts of action potentials brought about by low-threshold calcium spikes due to T-type Ca2+ channel activation. Sharp-wave/ripple (SPW/R) complexes, which are short episodes of increased activity with superimposed high-frequency oscillations of the hippocampus are also involved in this stage plus the neuronal replay of previous behaviour. In my view, the delta waves observed in this sleep state could represent the firing of the GABAergic neurons mediating inhibition of excitation whereas the sharp -wave/ripple (SPW/R) complexes observed in the hippocampus could be the equivalent of the gamma waves seen in the awake state and representing the neuronal cell assembly connectivity.

In contrast to the delta state, theta brain waves with a frequency of 4-7HZ can exist as ´background` activity or as bursts such as the theta clock-spiking cell activity. Theta wave activity is found in the PFC, sensory cortex, hippocampus and cingulate cortex and appears to be needed for a wide range of cognitive functions, eg. memory, perception, consciousness, working memory with high frequency stimulation, recall (in mPFC and EC), learning (in EC and hippocampus), visual memory (in V4 visual cortex and PFC), spatial memory (in hippocampus), and temporal coordination (in PFC). It is this last function which in my opinion holds the key to the overlying purpose of the theta brain oscillation, ie. that it demonstrates temporal coordination including order. This is achieved through bursts of activity when observed in areas dominated by other firing eg. alpha or beta waves, or when constant, eg. as in the case of the theta clock-spiking cells. The mechanism causing this type of intermittent firing could be via GABAergic interneurons in the hippocampus which switch off firing in connected areas, hence giving a timing to firing in a manner similar to Morse code, ie. dot (firing) dot dash (not firing/switched off firing). Evidence for this view comes from spatial navigation, the existence of hippocampus and PFC coupling, the observation that it is at its highest amplitude in the DG, and that its presence in the cingulate cortex reflects the amount of irrelevant information perceived in an event. The view is also supported by the appearance of theta brain waves during the sleep stages. Theta brain waves arise in Stage 2 as unsynchronized beta and gamma waves reduce, and as conscious awareness fades. The stage includes the occasional sleep spindle (8-14Hz) generated by the thalamic pacemaker which lasts half a second plus high amplitude K complex waves of the theta type with short negative high voltage peaks, followed by a slower positive complex and then a final negative peak with each complex lasting 1-2 mins. These serve to protect sleep and suppress responses to outside stimuli as well as to aid in sleep-based memory consolidation and information processing.   Theta waves represent the deep sleep of the NREM  and is shown by synchronous activity in the thalamocortical network (thalamus to M1, SMA, PFC, V1, auditory cortex), modulated by inhibitory inputs from the thalamic reticular nucleus (TRN) which is itself modulated by GABAergic neurons from the lateral hypothalamus. This provides support for the view that theta waves portray background activity as visual stimuli are gone and memory consolidation and information processing occurs through timing and GABA interneuron functioning. The question as to why the neurons are better serviced by burst activity and not continuous activity can be answered by looking at the biochemical mechanisms involved. Continuous activity would result in the area suffering ´electrical shut-down` due to the cells experiencing simultaneous absolute refractory periods and therefore, burst activity is a natural switching off, but not to this point. This could mean that the firing activity of the neuronal cell assembly is stronger since timed together through phase locking and more capable of firing more frequently than if each cell is activated to biochemical ´exhaustion`.

The alpha brain wave (8-12HZ) occurs in the quiet, waking state. It is found when the subject is relaxed, during meditation, fantasizing, daydreaming, learning (alpha activity is said to occur for pre-stimulus activity – Myers) and with spontaneous firing which can be alpha or beta in nature. It is observed in the occipital, parietal, and anterior areas and is responsible for item maintenance (post parietal and occipital), episodic memory recall of visual information and in recall theta-alpha (4-13HZ) oscillations are said to bind the hippocampus, PFC and striatum. It is this latter function that may indicate in my view the importance of alpha brain waves. A demonstrable connectivity between these areas may allow information to be linked together through simultaneous firing of neuronal cells within a group and associated with other groups (perhaps of different areas) within a definite time period. This view is supported by: the role of alpha waves in daydreaming/fantasizing (internal recall without stimulus from outside), learning (internal repetition – explains observations of gamma waves for cell assembly building with alpha waves); appearance at the end of Stage 1 of sleep as the individual relaxes; the observation that alpha activity inhibits disruptive distracting information in recall; and alpha amplitude varies with memory maintenance and updating demands (increased alpha post cue is associated with high relevant load).

The brain wave oscillation with higher frequency than alpha is the beta wave with a frequency of 13-25HZ and appears when awake in general thinking, analysis, talking and learning and specifically episodic memory recall of visual information, attention, motor control, perception, and awareness. It is observed in all activated cortical areas, the sub-thalamus nucleus, basal ganglia, hippocampus and motor cortex. This type of brain wave like alpha in my view represents the active firing of neurons of the cortex representing memory recall, working memory, attention and perception. This also explains its appearance in Stage 5 of REM sleep and importance to procedural memory and learning. Here it is likely to be involved in the strengthening of the neuronal assemblies by recall of features whilst in the dream-state.

Of all the brain wave types, the most interesting is probably the gamma type which often exists as 30-80 HZ bursts. These are observed in nearly all brain areas including sensorimotor, frontoparietal, PFC and mPFC (visuomotor), parietal cortex and cingulate cortex and are linked to neuronal connections and groupings, perception, attention, memory (long-term and short-term), awareness, visuomotor associations, and spatial memory (firing patterns during wakefulness correlate greatly to spatial patterns observed in sleep during gamma wave activity phases). In my view, the gamma wave activity is linked to the phase of interneuron activity and is responsible for the firing of cell group assemblies. Gamma bursts likely represent the connectivity between the PFC and other areas such as visuomotor areas, or PFC and hippocampus. In the latter, it exists with theta connectivity in spatial memory between ventral/medial PFC areas and the hippocampus. In this case too, a functionally columnar network of recurrently connected excitatory and inhibitory neural populations is seen. Firing of the hippocampal and PFC cells link objects together as required for spatial memory (eg. object and location). Gamma oscillations could also represent the links between different areas of the PFC a view supported by Brovelli who showed that performance of visuomotor associations was characterized by an increase in gamma power oscillations and functional connectivity over the sensorimotor and frontoparietal network, in addition to medial prefrontal areas. The superior parietal area plays a driving role in the network, exerting Granger causality on the dorsal premotor area. Premotor areas act as a relay from the parietal to medial prefrontal cortices, which plays a receiving role in the network. Further analysis shows that visuomotor mappings reflect the coordination of multiple subnetworks with strong overlapping firing of motor and frontoparietal areas. In the case, of working memory, gamma bursts are observed in the post-parietal cortex responsible for the manipulation of material plus item maintenance and the lateral occipital cortex responsible for item maintenance. Different areas of the PFC are involved in different types of WM as shown by Pasternak (work on senses) and Soto (work on TMS).

Therefore, from the above explanation of brain waves and how they can change with area and function we can see that Fiebig and Lansner looked at pyramidal and basket cells from the cortical layer 2/3 of an indefined area, but undergoing a specific task, eg.  learning and recall. Their observation of the theta waves is justified by their suggested role as ´background` activity or as bursts such as the theta clock-spiking cell activity and their involvement in perception, awareness, working memory, recall (in mPFC and EC), learning (in EC and hippocampus), visual memory (in V4 and PFC), spatial memory (in hippocampus), and temporal coordination (in PFC). It is likely that in the experiments carried out here the theta oscillations as bursts provide the temporal coordination required for the input, storage and recall of the word information with the intermittent firing via the inhibitory GABAergic interneurons of the cellular columns and possibly through hippocampus and PFC connectivity. It is likely that the authors would have seen lots of other areas working during the course of the experiment in seeing a word, learning a word, recalling a word and would have seen other brain waves in play if examined.

Therefore, what makes this article interesting is the insight it gives us on the mechanisms in play during learning and recall for one particular area and specific cell types. These mechanisms are not limited to the biochemical exchanges of ions or the transmission of chemical signals for example, but include how these biochemical actions manifest as measurable synchronized cell assemblies and cognitive function. What is also clear is that fine analysis of experimental results is required and that we cannot assume that the mechanisms and neuronal connectivity patterns that occur in one area can be given as doctrine for others. As always believed nature is complex and the understanding of the brain, a master of changeability and adaptability, requires a flexible, exacting scientific approach if it is to be ever completely grasped.

Since we`re talking about the topic …………

……working memory training programs are said to increase working memory performance. (They could mean attentional system positive differences, but still an overall increase in working memory performance is observed). Therefore, can we assume that priming or training programs will not affect the timing and systems of firing seen by the authors in a repeat of these experiments, but may affect the strength of firing since increased myelination and increased attention are said to be the beneficial results of working memory training?

……age differences have been reported for learning and recall linked to the increased amount of irrelevant information input and learning associated with rising age. Can we assume that older age groups exhibit the same patterns of brain wave activity shown by Fiebig and Lansner, or does age have an effect on the levels, or timing of the gamma bursts? Administration of a drug that temporarily prevents the opening of potassium channels in the PFC leads to the restoration of working memory performance associated with age. Would changes in gamma activity mirror the administration?



This entry was posted in brain waves, neuronal firing, short-term potentiation, Uncategorized, working memory and tagged , , , . Bookmark the permalink.