Semblance Hypothesis

FAQs

Why do we need a hypothesis?

We need to discover the mechanism of operation of the nervous system that provides internal sensations of various higher brain functions such as perception, memory and consciousness. Without knowing this, we will not be able to understand the system. In this context, hypothesis development is essential to understand how internal sensations are induced in a system with nearly 1011 neurons and 1015 synapses. A single counter-example of proof against a hypothesis can then be used as sufficient reason to modify or reject it. According to Karl Popper, a philosopher of science, a hypothesis must be falsifiable; i.e. it must at least in principle be possible to make an observation that would disprove the proposition as false, even if one has not actually (yet) made that observation (Popper 1965). Once such an observation is made, it will lead to rejection of the hypothesis. However, even with the rejection of a hypothesis, we are likely to make some conclusions that will aid in the development of new and better hypotheses.

Nervous system is being studied by several faculties of sciences at various levels – biochemical, cellular, electrophysiological, systems, behavioural, imaging. In order to explain all these features, the solution must be a unique one. In this regard, a testable hypothesis is highly valuable. Even though the internal sensations cannot be directly examined, we can circumvent the difficulty. If a simple unique solution can be derived to explain all the findings made at various levels, then this solution must be right (This is similar to viewing an unknown variable in an equation within a solvable system of linear equations where the values of all the other variables are known). This motivated developing a hypothesis for nervous system functions. The hypothesis can then be verified in biological systems by a) searching for the predictions that can be made from the hypothesis, b) examining comparable circuit features for different sensations in remote species of animals, and c) postdictive examination of several previous findings. Once verified, it can be further studied by the gold standard test of replicating the mechanism in engineered systems. This approach will truly enable us to undertake a cost-effective research work in the right direction.

What function should we begin examining to build the hypothesis?

Learning and memory are the best functions to study the nervous system operations. This is because we can 1) induce changes in the nervous system during associative learning that can be verified, 2) induce first-person internal sensation of retrieved memories at physiological time-scales, 3) carry out loss of function studies, 4) test whether the hypothesis can be extended to understand consolidation of memories, perception and consciousness, 5) replicate in engineered systems to test for the formation of the first-person inner sensation of memory, and 6) use the very large amount of already collected data to verify the hypothesis being built at its various stages. For example, the following questions can be addressed. a) What parallel cellular changes are taking place during testing for long-term potentiation (LTP) with a regular stimulus and retrieval of memories? b) How LTP can get correlated with the surrogate markers of behavioural motor activities indicative of the induction of internal sensation of memory?

What is the difference between single synapse strengthening hypothesis and semblance hypothesis?

From Hebb’s postulates, it was derived that synaptic plasticity changes the strengths of single synapses during learning. If two stimuli are associatively learned, then the synapses along their paths are expected to undergo plasticity changes. However, it is not yet known how arrival of one of the stimuli (cue stimulus) that propagates through its path utilizes the changes in synaptic strength to induce memory of the associatively learned second item. In other words, it lacks a mechanistic explanation that can support replication of the mechanism in an engineered system. In this context, the present hypothesis was developed from asking the question "At the time of memory retrieval, when one of the sensory stimuli (the cue stimulus) moves through its pathway, how can it induce inner sensation of memory of the associatively learned sensory stimulus (that moved through a second pathway at the time of learning) and also generate behavioral motor activity reminiscent of the associatively learned second stimulus?"

Based on the semblance hypothesis, when an associative learning takes place between two sensory stimuli, there should be certain changes at the locations where these stimuli converge (for example, hippocampus in spatial memory or amygdala in fear memory). This hypothesis examined the interaction between the synapses of the associatively learned stimuli at locations of their convergence. At a later time, when one of the stimuli (cue stimulus) arrives at the locations of convergence of the two sensory stimuli, the cue stimulus should be able to induce internal sensation of the memory of the associatively learned second stimulus. Therefore, semblance hypothesis focused on identifying the locus of interaction between the two neuronal pathways and more specifically, the sub-synaptic locations between which this interaction can lead to learning-induced changes from which the cue stimulus can induce inner sensations of memory of the second stimulus. In this approach, it is not possible to use neuronal firing due to several reasons explained in the answer to the next question.

What are the issues with studying neuronal firing (axonal spike) in understanding higher brain functions?

Neural network studies have been carried out for the last fifty years and are finding severe difficulties in solving the nervous system. The problems can be explained as follows.

1) Investigations during the last 15 years have shown that in addition to axonal spikes (neuronal firing or action potential), there are spiking potentials occurring at the dendrites (Antic et al. 2010; Moore et al. 2017). Spikes are the summated (summed up) potentials occurring at a localized region. The purpose of the axonal spike is to propagate the potentials towards all the axonal terminals of the neurons. However, we have to still discover the function of the dendritic spikes. Only by directing our studies to interconnect as many observations as possible, we will be able to find the functional attributes of dendritic spikes that will help us to solve the system. 

2) The number of input connections (postsynapses or postsynaptic terminals or dendritic spines) vary widely among the neurons. It ranges from one (passive conductance of potentials between the initial orders of neurons of the visual pathway) to approximately 5,600 (as in a monkey’s visual cortex) to 60,000 (as in a monkey’s motor cortex) (Cragg 1967). Most often, arrival of a tiny fraction of inputs is sufficient to fire a neuron. For example, a pyramidal neuron that receives tens of thousands of inputs can fire an action potential by spatial summation (summation at the same time) of 40 EPSPs at the axonal hillock that arrives from any combination of 40 inputs (Basic electrophysiology; Palmer et al. 2014). Please note that temporal summation of even less than 40 EPSPs can induce an action potential. The combinatorial probability of the number of sets of synapses whose activation can give rise to the firing of a neuron is enormously high. This makes an action potential non-specific with regards to its inputs.

3) Thirdly, postsynaptic potentials contributing to both sub- and supra-threshold activation of a neuron do not contribute to the neuronal firing. Therefore, if there are mechanisms for inducing internal sensations occurring at the unaccounted synapses, they will get ignored if neuronal firing alone is examined. For example, let us take one pyramidal neuron (excitatory neuron) with 25,000 inputs (dendritic spines). If 3600 inputs (dendritic spines) are activated simultaneously (due to their synaptic activation) during an action, only one action potential will be elicited. Simultaneous arrival of 40 inputs at the axonal hillock is enough to induce that action potential. This means (3600 - 40) = 3560 EPSPs get wasted without having any functional use. Is this advantageous to the system? For the purpose of this discussion, let us assume that 40 EPSPs can fire a neuron. In this context, any set of inputs of less than 40 EPSPs that do not lead to the generation of action potential is also getting wasted. In what context evolution would have conserved this mechanism? The input redundancy may be a possible mechanism to achieve common set of outputs for operating the limited set of combinations of muscles in the body for achieving behavioral activities to survive in the environment. However, in the context that we are still searching for a mechanism of induction of first-person internal sensations, reminiscent of the external stimuli in their absence, it is required to examine possible mechanisms occurring at the input level. This is necessary to avoid ignoring any valuable operational mechanism occurring at the input level.

4) Postsynaptic potentials induced at the dendritic spines located at remote locations on the dendritic tree (for example, pyramidal neurons with long apical dendritic tree) has to travel long distances to reach towards the axon hillock to summate above the threshold for triggering the action potential. They degrade significantly as they reach the axon hillock (Spruston 2008). Therefore, contributions of these potentials to neuronal firing get reduced and vary depending on the distance they have to travel and the dendritic diameter. This naturally leads to the question "Why would these potentials get conserved?" Except in conditions where they contribute to the nth EPSP required to trigger the action potential, it is very likely that they are providing functions independent of the neuronal firing. Therefore, we have to think about a mechanism other than neuronal firing.

5) Since EPSPs get degraded as the distance from the dendritic spine to the soma increases, in reality EPSPs from more than 40 dendritic spines will be needed to fire a neuron. For example, let us assume that inputs from 100 spines need to arrive at the axon hillock for one event of firing of a given neuron. Let us also assume that this pyramidal neuron has 30,000 dendritic spines (inputs or postsynaptic terminals). If EPSPs arriving from nearly 100 of its dendritic spines can fire that neuron, then nearly [3x104! ÷ (100! x (3x104! – 100!))]  4.68x10289 sets of combinations of inputs can fire that neuron. If we consider that a pyramidal neuron has only 3,000 dendritic spines, then the set of combinations will reduce to 1.04x10189. This means that a gigantic number of combinations of inputs can cause the same neuronal firing. Therefore, when we see a neuron firing (axonal spike) (in vivo, at physiological conditions), it is not at all specific with respect to its inputs. Understanding this redundancy in inputs that lead to firing of a neuron is of paramount importance in undertaking experiments that artifically fire neurons and the interpretating their results.

6) Many times, several neurons are held at subthreshold activation. It means that they will be receiving less than 40 postsynaptic potentials all the time, just short of few potentials for triggering an action potential (neuronal firing). Neurons located at higher orders than those that are firing in an oscillating fashion (reasons for these oscillating type of neuronal firing need explanation, especially the horizontal component of the oscillations – which are explained by the present hypothesis) are mostly held at a range of subthreshold values. For example, 38 or 39 inputs arriving at higher order neurons will not lead to the firing of those neurons. These sub-threshold-activated neurons require only 1 or 2 inputs to cause their firing. Therefore, when we see these neurons firing, these neuronal firings have to be interpreted completely differently.

All the above findings show that studies using neuronal firing and networks of firing-neurons do not examine specific mechanisms that are likely to take place at the level of the inputs (dendritic spines). In addition, when it comes to the need for explaining the first-person internal sensations of higher brain functions, the current studies examining the third-person observations are a dimension away (third-person v/s first-person) from where we need to reach.

So what does a neuronal firing mean with respect to its inputs? From the above paragraphs, we have seen examples of conditions in which a neuron held at its baseline state can get fired by either 3600 inputs or just 1 input. In what context evolution would have conserved this mechanism? It may be a possible mechanism to achieve common set of outputs for operating the limited set of combinations of muscles in the body for achieving behavioral activities to survive in the environment. In the context that we are still searching for a mechanism of induction of first-person internal sensations, reminiscent of that are induced by the external stimuli (in the latter's absence), it is required to examine possible mechanisms occurring at the input level.  In the context of input redundancy in firing a neuron, this will avoid ignoring any valuable operational mechanism occurring at the input level. This will allow us to address the question from the previous subtitle "Where is the ideal location for convergence to occur that will allow the cue stimulus to induce internal sensation of the associatively learned second stimulus?" without ignoring the specificity of inputs brought by the cue stimulus. It is reasonable to expect interactive changes occurring at the input levels of the neurons at locations of convergence of associatively learned stimuli. This is examined in the new hypothesis.

How can the information from fMRI studies be used to understand the operational mechanism?

From the Table 1 on the first page of this website, it can be seen that one of the requirements of the operational mechanism is that it should take place at physiological time scales. Since blood oxygenation level dependent (BOLD) signals take place following a delay of up to 4 seconds (normal synaptic delay is only 1 to 2 milliseconds; both associative learning and memory retrieval can take place in milliseconds) following neuronal activities at the same location (Fig.2 in Monti et al., 2010; Figs.2-5 in Murayama et al., 2010), it does not provide information regarding the normal operational mechanism. However, when the actual mechanism of operation is known, it should be able to provide an explanation why oxygen is released at those locations following a time delay. In other words, the hypothesized mechanism should be able to accommodate a proper explanation for the BOLD signals.

What are the current challenges in memory research and how can we overcome them? 

Memories are virtual internal sensations at the time of memory retrieval. The behavioral motor activities observed along with it should be considered as surrogate markers indicative of memory retrieval. Strong correlation between the experimental finding of long term potentiation (LTP) and the surrogate behavioral motor activities at the time of memory retrieval have been observed. However, alone, LTP has certain limitations. LTP takes at least 30 seconds (Gustafsson and Wigström, 1990) and even more than a minute to reach it peak level of induction, which does not match with the physiological time-scales of changes occurring during associative learning. LTP was reported as lacking sufficiency to be the mechanism of memory storage (Shors and Matzel, 1997; Martin et al., 2000; Piorazi and Mel, 2001). Furthermore, several reported correspondences of LTP temporal phases do not correspond with that of memory phases (Abbas et al., 2015). In spite of these, the correlation between the behavioral markers of memory with LTP (excluding the time-scale issues) has some hidden facts that can provide a valuable piece of the puzzle towards understanding the cellular changes occurring during associative learning. In this context, it becomes necessary that the true mechanism of formation of first-person internal sensation of retrieved memories should be able to explain how LTP is related to memory.

Challenges in understanding the mechanistic changes during associative learning that enables cue-induced internal sensation of retrieved memory and its related effects on the observations in the field of psychology have been discussed (Gallistel and Balsam, 2014; Edelman, 2012). The challenges become manageable when the frame of reference of examination of the higher brain functions is changed from the third-person to the first-person.

What are the general requirements of a hypothesis of memory?

It should theoretically be able to explain the following features.

-
Ability to learn at physiological time-scales (milliseconds), following which memory can be retrieved

- Retrieval of memory at physiological time-scales (milliseconds)

- Provision for unlimited memory life-times (Rubin and Fusi, 2007)

- Ease of learning a related task

- Disuse reduction in memory

- Instant access to very large memory stores (Abbott, 2008)  

- Should have provision for a mechanism for retaining specificity of memory retrieval                                              

- Functional integration & operation of hippocampal new neurons in learning and memory & its possible role in consolidation of memory

- Transfer of the basic units of memory for a different learning and retrieval event (Dahlin et al., 2008)

- Ability to explain observed correlations between LTP & behavior motor activities indicative of formation of inner sensation of memory

- Explain observations, following the removal of hippocampi, of retrivel of memories of what was learned prior to 8-10 years ago

- Ability to explain internal sensation of perception at least as a framework

- Ability to explain internal sensation of consciousness at least as a framework

- Mechanism within the system to generate hypothesis (Abbott, 2008)

- Ability to explain some of the features of mental disorders (as a "loss of function" of normal operational mechanism)

A hypothesis that can provide a broad framework incorporating the above features needs to be built and tested theoretically followed by experimental approaches to confirm the basic structural changes taking place both during associative learning and memory retrieva
l. The operational mechanism should take place within the synaptically-connected neuronal circuitry.

Explain semblance hypothesis in simple words?

Since an infinite number of memories are expected to get generated using a finite number of neuronal processes, it is reasonable to assume that memory is formed from unitary mechanisms and their natural computation is occurring at physiological time-scales. Since the nature of retrieved memory changes as we keep changing the cue stimulus slightly, it indicates that the changes in cue stimulus is capable of inducing specific units of internal sensation. The induction of internal sensation occurring at physiological time scales require explanation for a feasible cellular mechanism. Since the induced first-person internal sensation is virtual in nature, the aim of the hypothesis building was to examine the system for specific properties and mechanism that can induce such a function. The mechanism should operate by a simple mechanism and should be operating universally to explain similar functions in members of different species of animals. When the hypothesis was developed, care was given to make sure that it fits very well with the constraints offered by large number of findings given in Table 1 on the front page.

The derivation of the hypothesis has two major stages. Each stage consists of few steps that are numbered.

Stage I

This stage can be approached by different methods. Here, two methods of approach are given.

Method 1:

1. For the purpose of derivation of the hypothesis, memory is viewed as a virtual inner sensation of a sensory stimulus since the sensory inputs from the item memorized is not present during the retrieval of memory.

2. We store thousands of memories. A specific internal or external cue stimulus is required to retrieve a specific memory.

3. Let us now conduct an imaginary experiment. Let us look at a yellow-colored pen. While looking at it, let us assume that a specific set of 105 synapses (out of the total 1015 synapses in our brain) were activated at different orders of neurons (1st order being the order close to the sensory level). If we can specifically stimulate the set of those specific 105 synapses, we can reasonably assume that we are likely to memorize/ visualize that yellow-colored pen.

4. How can we activate a specific set of 105 synapses out of the total 1015 synapses? Saying in a different way, how can we selectively activate each of those 105 specific synapses from a set of 1015 synapses for retrieving the memory immediately after associative learning? If we know how we can activate one of those 105 specific synapses that identify the item to be memorized, then we can extend the same mechanism to all the 105 synapses.

5. Alternatively, we can address the issue in a modified way. What is the minimum requirement that satisfies activation of a synapse? Activation of the postsynaptic terminal (dendritic spine or spine) can be taken as the equivalent of activating a synapse since the activation of a postsynaptic terminal takes place after the arrival of an action potential at its presynaptic terminal.

6. Since there is no sensory stimulus available from the item to be memorized, we cannot anticipate any action potential reaching at the presynaptic terminals of the routes through which it is supposed to propagate. Therefore, we need to activate the postsynaptic terminals of the synapses through which stimulus from the item (whose memory is to get retrieved) had passed before. This should occur in the absence of arrival of any action potentials at those presynaptic terminals during memory retrieval.

7. Activation of a postsynaptic terminal without the arrival of an action potential at its presynaptic terminal (at the synaptic level) can represent the idea of evoking a virtual sensation of a sensory stimulus (at the systems/behavioral level). In other words, the cue stimulus is expected to activate a specific set of postsynaptic terminals that can evoke virtual sensation of a sensory stimulus arriving from the learned item. After the learning, can the cue stimulus activate the postsynaptic terminals of the synapses of the path through which the learned item had propagated before?

8. The above arguments get further support by the fact that lateral entry of activity of certain areas of the brain either artificially or by pathological conditions can induce virtual internal sensations in the form of hallucinations with a compelling sense of reality (Selimbeyoglu and Parvizi, 2010).

9. At this point, we come across with two key questions. 1) Can we activate the postsynaptic terminal of a synapse in the absence of the arrival of an action potential at the presynaptic terminal? 2) How can we choose to activate those 105 specific postsynaptic terminals from the total 1015 synapses for specific activation immediately after associative learning? What we have is a specific cue stimulus that activates a specific set of synapses. We can now arrive at a simple question at the synaptic level: “How can we activate a specific set of 105 postsynaptic terminals that would otherwise be activated by the item whose memory is to be retrieved in the presence of the activation of the specific set of synapses by the cue stimulus?

10. Let us assume that the cue stimulus evokes activation (depolarization) of the postsynaptic terminals through which activity from the learned item pass through. Then, it is reasonable to argue that some of the synapses through which activity spreads from the cue stimulus should be physically close enough to some of the postsynaptic terminals through which activity from the learned item passed through at the time of learning. At the time of memory retrieval, a mechanism should exist that can cause spread of activity from the synapses of the cue stimulus to the postsynaptic terminals of the item whose memories are retrieved (Figure 1).

                                                                                  
Neuroscience and Artificial Intelligence
Figure 1. Illustration of the hypothesized depolarization spread during retrieval of memory. During retrieval, the cue stimulus reaching presynaptic terminal A depolarizes its postsynaptic membrane B, and the depolarization spreads to postsynaptic membrane D. This can only happen, provided there is a functional LINK between the postsynaptic terminals B and D. Therefore, we can assume that a functional LINK is required to be formed between postsynaptic terminals B and D during learning.

Method 2:

1. Let us imagine that two sensory stimuli, namely stimulus 1 and stimulus 2 undergoes associative learning. At a later time when stimulus 1 (cue stimulus) arrives, it is expected to induce internal sensation of memory of the second stimulus 2. For this to happen, it is necessary that some changes should occur at the locations of convergence of stimulus 1 and stimulus 2 at the time of learning. (Note that hippocampus known as an area of the brain associated with learning and memory receives inputs from all the different sensory modalities after 3 to 5 orders of neurons from the sensory receptor level). Now, let us examine what changes should be occurring at the location of convergence between two sensory stimuli at the time of learning. What should be the critical change occurring during learning between the synapses activated by the stimulus 1 and stimulus 2? Between what locations of the synapses that these changes should take place? The interaction should take place between those sub-synaptic locations that will enable retrieval of memory of the second stimulus when the first stimulus arrives and vice versa. In this regard, interaction taking place between the postsynaptic terminals of the stimulus 1 and stimulus 2 is suitable. (This was arrived by examining different sub-synaptic areas for their properties that endows them to generate units of internal sensation by trial and error method. This is described in section II). The interaction between the postsynaptic terminals was named as inter-postsynaptic functional LINK (Figure 2). The term "functional" is used to indicate that the formation of the LINK is a function of the activities arriving at the postsynaptic terminals activated by the stimulus 1 and stimulus 2 during associative learning. At the time of memory retrieval, reactivation of the inter-postsynaptic functional LINK is a function of arrival of activity from either stimuli at one of their corresponding postsynaptic terminals. The term LINK is written in capital letters to indicate that it is the key element of the hypothesis.

                                                                                  IPL formation          

Figure 2. Illustration showing the formation of hypothesized functional LINK between the two postsynaptic membranes B and D during associative learning between stimulus 1 and stimulus 2.

2. The inter-postsynaptic functional LINKs formed during associative learning can be of different types:

a.
Those that are formed by removal of water of hydration between the postsynaptic terminals, which will allow abutting of the membranes. This requires very high energy and will lead to rapid reversal of the functional LINK. This can provide sufficient learning-induced changes that can last only for a short period of time responsible for working memory.

b. Strong interaction between the postsynaptic terminals can lead to reversible partial hemifusion between the postsynaptic terminals. This can explain retention of learning-induced mechanism for more time.

c. Further interaction can lead to reversible complete hemifusion between the postsynaptic terminals that will enable its retention for much more time.

d. If the complete hemifusion can be retained for some time, it is likely that the stabilizing mechanisms can result in long-term maintenance of this.

3. Inter-postsynaptic functional LINKs may be viewed as biological parallels of
K-lines proposed by Marvin Minsky (Minsky, 1980) who was one of the founders of MITs Artificial Intelligence program.

4. Now let us examine the effect of arrival of the stimulus during memory retrieval. Let stimulus 1 arrive as a cue stimulus (See Figure 3). It arrives at the synapse A-B. The postsynaptic potential at B propagates through the inter-postsynaptic functional LINKs and reach towards the postsynaptic terminal D. As discussed in Method 1, the arrival of the stimulus 1 (cue stimulus) happens only infrequently. Therefore, when the second postsynaptic terminal D is depolarized incidentally in the absence of the arrival of an action potential at its corresponding presynaptic terminal C, then the postsynaptic terminal D is expected to get the cellular hallucination that it is receiving sensory inputs through its presynaptic terminal C, resulting in “semblance". This can induce units of virtual inner sensation of memories at the time of memory retrieval and can meet the expectations of a mechanism for memory (Minsky, 1980), if there is a specific operational logic at this location. We will explain the finding of an operational logic in Stage II. Before this, let us examine few properties of inter-postsynaptic functional LINKs.

                                                                    a)   Reactivation of an IPL            b)  Vector components

Figure 3. a) During retrieval, the cue stimulus reaching presynaptic terminal A depolarizes its postsynaptic membrane B, re-activates the inter-postsynaptic functional LINK. In this manner, depolarization spreads to postsynaptic membrane D evoking cellular hallucination at the postsynaptic terminal D of the arrival of sensory stimuli at its presynaptic terminal C. This is named semblance. b) The propagation of potentials at the synapse A-B and through the inter-postsynaptic LINK B-D provides ionic changes in the extracellular matrix space that contribute vector components for the oscillating extracellular potentials.

5. The propagation of potentials through the synapse A-B and the IPL B-D provide vector components that are responsible for contributing to the oscillating extracellular potentials (Fig.3b).

6. When the related learning events continue, one of the postsynaptic terminals that already took part in a previous learning event (either B or D in the Figure 3) will be used to form functional LINKs with the postsynaptic terminals of the neighboring synapses (seen as additional postsynaptic terminals on the right side of the postsynaptic terminal D in the left panel, Figure 4). As this process continues, it will result in the formation of islets of LINKed (LINKable/ re-activatible during retrieval) postsynaptic terminals (right panel, Figure 4).

                                                               NI to AI

Figure 4. Left panel: Illustration showing formation of islets of LINKed postsynaptic terminals. Continued learning events following the initial learning event can lead to the formation of multiple inter-postsynaptic LINKs between the involved postsynaptic terminals (dendritic spine heads). Only two presynaptic terminals (A and C) and two postsynaptic terminals (B and D) are marked. Assume that there are several postsynaptic terminals arranged in a horizontal plane. The dotted line shows a cross-section across the inter-LINKed postsynaptic terminals. Right panel: A hypothetical cross-sectional view of LINKed postsynaptic terminals of the synapses in one horizontal plane in a brain region (see the horizontal dotted line across the postsynaptic membranes in the left panel); in this illustration, we imagine that all the postsynaptic membranes are in the same plane. Postsynaptic membranes are shown in small dark circles (broken arrow). When learning occurs, functional LINKs between activated postsynaptic terminals can be established. Continued learning using any of those synapses will increase the number of interconnected postsynaptic membranes forming islets of functionally LINKed postsynaptic terminals (solid arrow). Multiple LINKs between the postsynaptic terminals in an islet can cause spread of postsynaptic potentials across the islet. The individual islets are expected to be functionally separate from each other.

To which neurons do the postsynaptic terminals that inter-LINK belong to?

Since the mean inter-spine distance is even larger than the mean spine diameter (Konur et al., 2013), the inter-LINKing postsynaptic terminals should belong to different neurons. This is also essential to maintain the specific outputs associated with each of the associatively-learned sensory inputs. This is expected to be the general rule. There could be exceptions; for example, when granule neuron axonal terminals continuously form synapses with the fixed number of dendritic spines of a CA3 neuron.

Stage II

In the next stage, the basic units of semblances occurring at the functionally inter-LINKed postsynaptic terminal are derived (Figure 5). We need to answer two questions. 1) How can a cellular hallucination (semblance) get induced at the inter-LINKed postsynaptic terminal D that was previously activated by the item whose memory needs to get retrieved? 2) What is the sensory content of this hallucination?

1. What is the logic behind the generation of cellular hallucination (semblance)?

Semblance is the mechanism by which virtual internal sensations are being created. Searching for a cellular location where such a mechanism can be formed resulted in arriving at the requirement for inter-postsynaptic functional LINK. In Figure 3, when cue stimulus arrives at the postsynaptic terminal B and re-activate the inter-postsynaptic functional LINK, it activates the postsynaptic terminal D. What makes the postsynaptic terminal to have a cellular hallucination (semblance) that it is receiving activity from its own presynaptic terminal C? The logic can be explained as follows. By default, postsynaptic terminal D is normally activated by its presynaptic terminal C. To make sure that this is the case, it appears that the Mother Nature has designed an excellent method. There is continuous quantal release of neurotransmitter molecules from the synaptic vesicles of the presynaptic terminal C even during periods of rest (and sleep). These provide regular arrival of miniature potentials at the postsynaptic terminals. The combined effect of all these potentials is represented by the miniature excitatory postsynaptic potentials (mEPSPs or “minis”). The fact that it is not possible to completely block mEPSPs “even in experimental conditions” indicates that it is a highly conserved default operation of the nervous system. Another necessary condition is the maintenance of oscillatory neuronal activity. The finding that electrical stimulation of the visual cortex produces a visual percept (phosphene) only when high-frequency gamma oscillations are induced in the temporo-parietal junction (Beauchamp et al., 2012) emphasizes the role of oscillating neuronal activity as a system requirement for semblance formation for creating internal sensations. The lateral spread of activity through the inter-postsynaptic functional LINKs can contribute towards the horizontal component of the oscillating potentials and the synaptic potentials between vertically oriented neurons in the cortex can provide the vertical component. Since inter-postsynaptic spread of potentials occur perpendicular to the trans-synaptic spread of potentials, this general feature can explain the wave form of oscillating potentials in all other regions in the nervous system, especially where sensory inputs converge.

2. What is the sensory content of the cellular hallucination (semblance)?

Cue stimulus activates postsynaptic terminal B that leads to re-activation of inter-postsynaptic functional LINK and activates the postsynaptic terminal D that was previously activated by the item whose memory is getting retrieved now (see Figure 5). At postsynaptic terminal D, this leads to semblance of activity arriving from the sensory receptors through neuron Z. Neuron Z is normally depolarized by activating a set of axonal terminals of the neurons in order 4 that synapse to neuron Z’s dendritic spines (postsynaptic terminals). The spatial summation of nearly 40 or the temporal summation of less than 40 EPSPs (from nearly 40 postsynaptic terminals (dendritic spines) out of the nearly 4×104 postsynaptic terminals of each neuron) triggers an action potential at neuron Z’s axon hillock (Note that the number of postsynaptic terminals (dendritic spines) for a neuron varies. In the hippocampus, we expect that the excitatory neurons have postsynaptic terminals in the order of 104). 

In the same way, the neurons in set {Y} in turn receive synaptic transmissions and spread of activity through functional LINKs from a set of neurons {X} in neuronal order 3. By continuing the extrapolation in a retrograde fashion towards the sensory level, it will be possible to determine the set of sensory receptors {SR} whose activation could theoretically cause the activation of postsynaptic terminal D. Dimensions of internal sensations resulting from the lateral activation of postsynaptic terminal D can be understood from the nature of the sensory stimulus that can activate sensory receptors in the set {SR}. It is likely that activation of subsets of a minimum number of sensory receptors from {SR} (example, {sr1}, {sr2}, and {sr3} (in Figure 5) is sufficient to activate postsynaptic terminal D. Therefore, a hypothetical packet of minimum sensory stimuli called “semblion” capable of activating one of the above subsets of sensory receptors that can activate postsynaptic terminal D is hypothesized as the basic unit of internal sensation of memory.

                                               
   Induction of semblances
                                                                             
Figure 5. Schematic representation of sensory elements induced during the activation of a synapse or a neuron. The gray circles represent neurons. The numbers on the left side of the neuronal orders denote their position in relation to the sensory receptors. Neuron Z is shown in neuronal order 8. During memory retrieval, a cue-stimulus (marked by asterisk) reaching presynaptic terminal A depolarizes its postsynaptic membrane B and the resulting EPSP at postsynaptic terminal B re-activates the functional LINK that activates postsynaptic membrane D (Mechanisms other than depolarization are also considered (Vadakkan, 2010)). When postsynaptic membrane D is depolarized, it evokes the cellular hallucination of an action potential reaching its presynaptic terminal C. This is called synaptic semblance. Note that presynaptic terminal C belongs to the neuron Z. Either synaptic semblance occurring at postsynaptic terminal D or random activation of neuron Z produces the hallucination that it is receiving inputs from the set of neurons {Y} that synapse to it. The set of neurons {Y} are activated by the activation of the set of neurons {X}. The set of neurons {X} in turn are activated by the set of neurons in the neuronal order above it. (Recurrent collaterals and projection neurons can also activate a higher order neuron. For simplicity these are not shown). Continuing this extrapolation towards the sensory level identifies a set of sensory receptors {SR}. It can be seen that stimulation of subsets of sensory receptor sets {sr1}, {sr2}, and {sr3} from the set {SR} may be capable of independently activating neuron Z. The dimensions of hypothetical packets of sensory stimuli capable of activating the sensory receptor sets {sr1}, {sr2}, and {sr3} are called semblions 1, 2 and 3 respectively. These semblions are viewed as the basic building blocks of the virtual internal sensations of memory. A cue stimulus can cause postsynaptic terminal D to hallucinate about any of the semblances 1, 2, 3 or an integral of them. Activation of postsynaptic terminal D by the cue stimulus can lead to the virtual internal sensation of semblions 1, 2, 3 or an integral of them. The method of integrating the semblions that match can with the internal sensations induced by the cue stimulus with that of the item whose memory is retrieved can be determined by computational studies. Note that the potentials through the synapse and IPL contribute vector components to the oscillating extracellular potentials (marked by the waveform) (Modified from Vadakkan, 2011).

As the cue stimulus passes through different functional LINKs, it evokes large number of semblances as explained above. Once these possible semblions are identified, their integration can be carried out to obtain net semblance that matches the sensory characteristics of the item whose memory is retrieved. Attempts to match the different computational products from the semblions with that of the sensory stimuli from the item whose memories are retrieved will lead to the discovery of the algorithm for neural computations for memory retrieval. The net semblance can exceed more than the threshold without any effect on the retrieved memory. As the functional LINKs get re-activated during memory retrieval, the expected spread of excitatory postsynaptic potential (EPSP) that occurs through some of these functional LINKs can be crucial in adding to the existing sub-threshold EPSP at the axonal hillocks of some neurons that are routinely activated by the oscillatory neuronal activities in the hippocampus and cortex as well as from baseline sensory activities arriving at many neurons. Since the number of functional LINKs continues to change (due to continued associative learning) over the life-span of the nervous system, the characteristic features of the semblions are also expected to change gradually. This will lead to gradual changes in the net semblances for memory. Related learning can increase the number of LINKed postsynaptic terminals and increase semblance for memory. Absence of retrieval of a specific memory, lack of repetition of learning or lack of related learning will reduce the number of re-activatible inter-postsynaptic functional LINKs and will reduce semblance for retrieval of a specific memory. Along with the induction of semblances, the reactivation of inter-postsynaptic LINKs can also provide additional potentials to the inter-LINKed postsynaptic terminal that can lead to firing of the latter’s neuron if it is kept at subthreshold activated level (Figure 6).

                              

Figure 6. Diagram showing the formation of internal sensations and fine control of the motor activation by a cue stimulus. Oscillating neuronal activity results in the activation of many downstream neurons. They can be kept tonically inhibited under resting conditions (not shown) to subthreshold levels such that they can be disinhibited at the arrival of one or a few excitatory postsynaptic potentials (EPSPs). There were two associative learning events that occurred previously with the cue stimuli. The first one was with items 1 and 2. After this first step of associative learning, the cue stimulus was retrieving memories of items 1 and 2. Note the reactivation of a sparse inter-postsynaptic functional LINK in the cortex. Along with retrieving memory of the second item, cue stimulus also evokes a motor response using the motor neuron. At a later time, the same cue stimulus had undergone a second associative learning event with item 3. Following this second learning event, the cue stimulus evoked internal sensations (semblances) of learned items 1, 2 and 3. However, as the semblance for item 3 was evoked, it also resulted in an inhibition of the motor activity (note the output from postsynaptic terminal D3 providing inhibitory potentials to the upper motor neuron). This type of an event is an example of the behavioral inhibition occurring at the frontal cortices. Complexities of the internal sensations can be based on the nature of the cue stimulus, previous associative learning, and the type of the nervous system. Reward-induced associative learning may be facilitated by dopamine-induced enlargement of dendritic spines (Yagishita et al., 2014) that promotes possible inter-postsynaptic membrane hemifusion and its stabilization for a long period of time. Also note that the cue stimulus reactivates inter-postsynaptic functional LINKs at other cortical areas to evoke memories for learned item 1. Since the inter-postsynaptic functional LINKs are transient and need reinforcement for long-term persistence, the induction of a minimum number of inter-postsynaptic functional LINKs alone may not maintain the effect of learning for a long period of time. In the hippocampus, the reactivation of inter-postsynaptic functional LINKs in response to spatial stimuli is expected to induce semblances for memories associated with that space and the EPSPs arriving through the inter-postsynaptic LINK induce firing of subthreshold-activated CA1 neurons (place cells). This explains how spatial memories are associated with place cell firing. Formation of circuits in this manner can explain the induction of internal sensations along with simultaneous behavioral motor action. Note the formation of a sparse inter-postsynaptic functional LINK at the cortex, which can contribute to specificity of retrieved memory (for a more complex path of its formation, see figure 9 in Vadakkan, 2015b). EPSP: excitatory postsynaptic potential. (n)th EPSP: the last EPSP necessary to achieve threshold EPSP to generate an action potential. Each motor action will evoke certain sensory stimulus in the form of proprioception that will act as a feedback stimulus to the system confirming that the motor action was executed. N: Excitatory neuron; IN: Inhibitory neuron. A and C: Presynaptic terminals; B and D: Postsynaptic terminals. Red line between B and D: Inter-postsynaptic LINK. (+) stimulation; (-) inhibition (Modified from Vadakkan, 2015b).

What is the nature of inter-postsynaptic functional LINK?

Different mechanisms for the formation of inter-postsynaptic LINKs are possible and are required to explain formation of internal sensations of other higher brain functions that operate at different time-scales. These different types of inter-postsynaptic LINKs with varying half-lives are suitable to explain perception, working, short- and long-term memories. A description of some of them are given in Figure 7.

                                                  AI from neuroscience

Figure 7. Different types of reversible inter-postsynaptic functional LINKs. A) Two abutted synapses A–B and C–D. Presynaptic terminals A and C are shown with synaptic vesicles (in blue color). Postsynaptic terminals (dendritic spines or spines) B and D have membrane-bound vesicles marked V containing subunits of AMPA receptor inside them. Action potential arrives at presynaptic terminal A releasing a volley of neurotransmitters from many synaptic vesicles inducing an excitatory postsynaptic potential (EPSP) at postsynaptic terminal B. From the presynaptic terminal C, one vesicle is shown to release its contents to the synaptic cleft. This quantal release is a continuous process (even during rest) that leads to generation of very small potentials on postsynaptic membrane D. Note the presence of a hydrophilic region separating postsynaptic terminals B and D. When action potential arrives at presynaptic terminal A, it activates synapse AB and generates an EPSP at postsynaptic terminal B. The hydrophilic region prevents any type of interaction between postsynaptic terminals B and D. Very high energy is required for excluding the inter-postsynaptic hydrophilic region (Martens and McMahon 2008). B) Membrane expansion occurring at physiological time-scales can provide sufficient energy to exclude the inter-postsynaptic hydrophilic region allowing close contact between the postsynaptic membranes at this region. This forms a transient inter-postsynaptic LINK that lasts only for a short period of time. During this short period of time, a cue stimulus-generated action potential arriving at synapse AB reactivates this inter-postsynaptic functional LINK and spreads to postsynaptic terminal D and induces units of internal sensation at the inter-LINKed postsynaptic terminal D. This can explain working memory. C) Diagram showing formation of a partial inter-postsynaptic membrane hemifusion. These vesicles contain glutamate receptor subtype 1 (GluA1). Activity arriving at the synapse can lead to exocytosis of vesicles containing AMPA GluA1 receptor-subunits abutted to the cell membranes and expansion of the postsynaptic membrane at physiological time-scales. During exocytosis, the vesicle membrane gets incorporated into the postsynaptic membrane at locations of exocytosis making this region of the membrane highly re-organizable. This matches with the location where AMPA receptor subunits were shown to concentrate at the extra-synaptic locations extending at least 25nm beyond the synaptic specialization (Jacob and Weinberg 2014). Note the interaction between the outer layers of membranes of the postsynaptic terminals. Depending on the lipid membrane composition, the process of close contact between the membranes described in the above section B) can get converted to a partial hemifusion state. D) Stage of partial hemifusion can progress to complete hemifusion. The reversible partial and complete hemifusions are short-lived and can explain the necessary learning-induced changes responsible for short-term memory. Some of the hemifusion changes can get stabilized for different lengths of time. For example, insertion of a transmembrane protein across the hemifused segment can maintain the inter-postsynaptic LINK until this protein gets removed. These changes can be responsible for long-term memory. E) Dopamine is known to facilitate motivation-promoted learning. In this diagram dopaminergic input to postsynaptic terminal B results in its expansion, which will augment inter-postsynaptic LINK formation. This can explain the action of dopamine on learning. Furthermore, it can sustain the hemifused LINK for long period of time, which may facilitate its stabilization. F) Hemifusion can advance to a complete fusion state in pathological conditions and it depends on several factors. Fusion of the postsynaptic terminals between two different neurons can lead to cytoplasmic content mixing and cytotoxic cell response. These include dendritic spine loss and eventually triggering of apoptosis leading to neurodegenerative changes. Note that excessive dopamine can lead to excessive expansion of the postsynaptic membrane and can lead to membrane fusion if other factors that resist this get compromised. Rm: membrane segment marked in Turkish blue shows area where membrane reorganization occurs (Figure modified from Vadakkan KI (2015a, b).

Are there any experimental evidence supporting the presence of the inter-postsynaptic functional LINK?

New technologies are required to test for the presence of the close contact between the membranes by hydration exclusion (Figure 7B) in vivo. Another mechanism of inter-postsynaptic functional LINK is the reversible inter-postsynaptic membrane hemi-fusion. If this is correct, then examination of the membrane bilayers at locations where postsynaptic areas are close together is an opportunity to test the hypothesis. It is also true that at locations where (sensory) inputs converge, the extracellular matrix space is very minimal as observed by routing electron microscopic (EM) examination of these regions. At these locations, abutted postsynaptic membranes are expected to be seen. However, there are some hurdles. First, the membrane hemi-fusions are reversible. However, locations within the hippocampus that has already undergone many associative learning, stabilization of these hemi-fused areas (most probably by the insertion of trans-membrane proteins) are expected. Secondly, only a very small area of membrane hemi-fusion is required for the functional effect of the formation of inter-postsynaptic functional LINK. Since the area of the postsynaptic membrane surface that has to be examined for such small areas of membrane hemi-fusion is very large, dedicated EM studies by taking serial sections spanning an entire postsynaptic terminal is required.

Alternatively, examination of large number of electron microscopic pictures of the hippocampal regions taken for other purposes can be tried. The limitation of this is the lack of resolution of the electron microscopic pictures to visualize the membrane double layer. In a recent EM work (Figure 8) with good resolution, it is possible to observe closely abutted areas suggesting that they may lack inter-membrane extracellular matrix space. Since dehydration during the tissue processing contribute to these observations, inter-membrane close contacts with hydration exclusion need to be verified using new in vivo techniques. In the above figure, another finding is very striking. There are areas of two layers of hemi-fused membrane for short distances instead of four layers of the two abutting postsynaptic membranes. These are very unlikely to be caused by rotation of the membranes or changes during processing of the tissue. These short spans of reduced number of layers is what is expected by the hemi-fusion process and provide support for the hypothesis until further verifications are carried out. Multiple fused spine heads on a single spine neck seen on dendritic excrescences at the CA3 dendritic tree (Amaral and Dent, 1981; Chicurel and Harris, 1992; Frotscher et al., 1991) is a possible structural modification evolving from long-standing inter-postsynaptic functional LINKs.

                                                                     AI from Neurons

Figure 8. This is Figure 4D from Burette A.C, Lesperance T, Crum J, Martone M, Volkmann N, Ellisman M.H, and Weinberg RJ (2012) Electron Tomographic Analysis of Synaptic Ultrastructure. Journal of Comparative Neurology 520 (12): 2697-2711. This figure is modified by inserting a red arrow. The red arrow points towards a likely inter-postsynaptic area with only 2 layers of membrane instead of the expected 4 layers. Even though tissue distortions during tissue processing and folded membrane are possibilities, such changes that can span for distances of only 100 nm is very unlikely. This observation indicates the possibility that it is an area of inter-postsynaptic membrane hemi-fusion. It needs further dedicated studies for verification. The green arrow points to a likely location where the close contact between membranes is visible. Since some of the cell processes are likely astrocytic pedocytes, dedicated studies are required to verify these observations. Scale bar = 100nm.

Why didn't we discover these IPLs until now?

First, there was no reason to search for a mechanism of exclusion of water of hydration at the two inter-neuronal inter-spine regions that are activated during associative learning. Secondly, no dedicated studies were carried out to image the lipid bilayers of an entire dendritic spine. Since the IPLs are expected to take place at regions of 10nm length, ultra-structural details of entire spine membranes is necessary.

It seems that all the above steps used third-person observations. Where is the examination from a first-person frame of reference?

In Figure 5, the steps needed in finding out the sensory content of the cellular hallucination induced at the postsynaptic terminal D involves examination form a first-person frame of reference. It requires searching backwards from the postsynaptic terminal D towards the sensory receptor level to find out the subset of minimum sensory receptors whose stimulation can activate the postsynaptic terminal D. The minimum sensory stimuli required to activate this subset of sensory receptors constitute the semblion, which is the basic unit of internal sensation. The backward extrapolation from the postsynaptic terminal D towards the sensory receptor level to find out the packets of sensory stimuli is an implicit process taking place during the internal sensations of all the higher brain functions. In this examination, we observe the packets of sensory stimuli (content of the unit of internal sensation) from a first-person frame of reference.

How can we explain long term potentiation (LTP) in terms of the semblance hypothesis?

The semblance hypothesis was derived to explain plausible synaptic changes occurring during learning suitable for evoking virtual inner sensation of a sensory stimulus during memory retrieval. The operational principle of the formation of semblances resulting in memories is completely different from that of LTP; however, the formation of inter-postsynaptic LINKs can be viewed as a common denominator in both semblance hypothesis and LTP induction (has yet to be confirmed). Explanation of semblance formation through inter-postsynaptic membrane functional LINKs can fill the gaps in our findings of correlation between memory and LTP and can explain why it has led to large number of debates. One general argument is that any hypothesis of memory should be able to explain the relationship between LTP and the surrogate behavioral motor activity indicative of memory retrieval.

Previous experiments have shown that spatial learning becomes impaired after saturation of LTP (Moser et al., 1998). Later experiments have shown specific inter-relationship between LTP and surrogate markers of memory retrieval (Whitlock et al., 2006). In this work it was shown that one-trial inhibitory avoidance learning in rats produced the same changes in hippocampal glutamate receptors as the induction of LTP with high-frequency stimulation. This study showed that learning-induced synaptic potentiation occludes high-frequency stimulation-induced LTP. Based on the findings in this work, a plausible reasoning for the relationship between LTP and memory through the semblance hypothesis can be done as follows.

a. Learning first followed by LTP induction
:

According to the semblance hypothesis, prior learning events in a caged environment would have already made many islets of LINKed postsynaptic terminals (dendritic spines) in the hippocampi of the rats. Since associative learning opportunities are finite during caged life, we can expect a slow expansion (by LINKing more postsynaptic terminals with additional related learning events) of discrete islets of LINKed postsynaptic terminals as the rats grow up. When rats undergo avoidance learning (a novel instance of associative learning), we can expect the formation of functional LINKs between two or more islets of functional LINKs that are already present in the animal. Even though this is particularly important in this experimental context, it will also hold true in any novel associative learning.

In experiments using inhibitory avoidance testing (Whitlock et al., 2006), not all the recording electrodes recorded an increase in field excitatory postsynaptic potential (fEPSP) slope, indicating that ionic changes at the locations of the tips of these electrodes (CA1 dendritic tree) required to produce an increase in fEPSP slope did not take place. However, among those electrodes that recorded an increase in fEPSP slope after inhibitory avoidance learning, a sufficient number of Shaffer-CA1 synapses were potentiated. Let I and II stand for two islets of functionally LINKed postsynaptic terminals that were already present in the animal before the avoidance learning session. During learning, it is likely that LINKs were formed between the islets (islets of LINKed postsynaptic terminals) I and II. This will generate a sudden increase in the size of an islet of LINKed postsynaptic terminals to nearly two-fold, forming a mega-islet of LINKed postsynaptic terminals (Figure 9).

                                                                             Neurons, synapses and AI

Figure 9. Illustration explaining the basis of long term potentiation (LTP) based on the present hypothesis. Illustration shows potential LINKable site between islets of postsynaptic terminals (dendritic spines) (please see Figure 4 for details of the islets; they are visualized by hypothetical cross-sectional view through functionally LINKed postsynaptic terminals) that belong to two different CA1 neurons. During an associative learning, LINK formed between the postsynaptic terminals (marked with asterisks) of islets 1 and 2 (large circles) can lead to the formation of a mega-islet that can continue to contribute to the LTP recorded from the recording electrode as explained in the text. Position of the stimulating electrode is at the Schaffer collaterals. Shaffer collateral from the CA3 neurons synapse to the dendritic spines (postsynaptic terminals) of the CA1 neurons. Many of these postsynaptic terminals are functionally LINKed to form islets in an animal (see Figure 4 for details of the islets of functional LINKs). Here two such islets I and II (large circles) are shown. One of the postsynaptic terminals from each of the islets I and II is shown to continue towards the soma of the CA1 neurons. Activation of any one of the postsynaptic terminal within an islet will result in EPSP spread towards the somas of the CA1 neuron. The islets are formed between postsynaptic terminals that are concurrently activated during previous associative learning. During an associative learning of a novel item or during induction of LTP (note the position of the stimulating electrode is at the Schaffer collaterals), a new functional LINK may form between the postsynaptic terminals (marked asterisks) of islets I and II. This can lead to the formation of a mega-islet combining the two islets. This can contribute to the LTP recorded from the recording electrode as explained in the text.

Activation of a postsynaptic terminal of this mega-islet of LINKed postsynaptic terminals can cause spread of depolarization between its postsynaptic terminals. Since a subset of postsynaptic terminals in the mega-islet already LINKed to one of the dendritic spines (postsynaptic membrane) on the dendritic tree of one CA1 neuron, multiple EPSPs from this subset will reach the main dendrite of a CA1 neuron simultaneously. This results in a summated EPSP at this dendritic location sufficient to produce a corresponding increase in current sink in the extracellular matrix. Immediately following the associative learning event, a proportion of sensory inputs reaching the animal for a long duration of time is likely to activate the postsynaptic terminals of this mega-islet, leading to prolonged activation of the main dendrites of the above CA1 neuron (until the CA1 neuron begins homeostatic mechanisms to reduce this prolonged and increased EPSP generation). The extracellular signal recorded from the apical dendrites of a population of pyramidal neurons in the stratum radiatum of the CA1 region in response to Schaffer collateral stimulation, namely the field EPSP, will now show an increase in amplitude and contribute to an increase in fEPSP slope for a long duration of time (LTP). This learning-induced LTP can occlude further LTP induction.

b. LTP induction first followed by learning
:

The occlusion process explained in the study (Whitlock et al., 2006) can be considered a bidirectional process, meaning that the induction of LTP in a sufficient number of synapses that are involved in inhibitory avoidance learning will prevent consequent avoidance learning. It is likely that hundreds of axons of the CA3 neurons in the Schaffer collateral pathway are activated by high-frequency stimulation (LTP induction), activating the postsynaptic terminals (dendritic spines) of a CA1 neuron. During this process, many postsynaptic terminals can get functionally LINKed due to the simultaneous activation of closely placed postsynaptic terminals by high-frequency stimulation (assuming that sufficient oxygenation state is present during this process). Some of these LINKs will occur between the islets of already LINKed postsynaptic terminals, leading to the generation of mega-islets. Following this, the activation of one or more postsynaptic terminals by a regular stimulus (not high frequency) can lead to the spread of depolarization between the postsynaptic terminals within the mega-islet. Since one or a small subset of postsynaptic terminals in the mega-islet originates from the dendritic tree of a single CA1 neuron, multiple EPSPs from these postsynaptic terminals can reach one dendrite of a CA1 neuron simultaneously. This results in an increase in the EPSP at these dendritic locations, leading to LTP. This artificially-induced LTP can occlude further learning-induced LTP.

If we can artificially induce LTP in a large number of fibers that includes those that are critical for the learning, then the animal may not be able to successfully retrieve specific memories after a new associative learning using those synapses following the LTP induction. This means that the animal cannot retrieve the specific memories; i. e., when a cue stimulus tries to retrieve a memory using these synapses, the induced depolarization spreads across all those postsynaptic terminals that are LINKed by the LTP induction. The retrieval using a specific cue now induces synaptic semblances at all those LINKed postsynaptic terminals in the mega-islet, some of which were non-specifically LINKed during the LTP induction. Activation of those non-specific postsynaptic terminals will also lead to the activation of non-specific neurons, leading to the induction of non-specific network semblances that are not related to the learned item. In other words, the expected specificity of semblance for the learned item gets diluted by the large amount of non-specific semblances, preventing specific memory retrieva
l.

The following diagram (Fig. 10) demonstrates the similarities between the cellular processes in LTP following induction and internal sensation of retrieved memory following associative learning.

                                 Comparison between LTP and memory

Figure 10. Illustration showing the structural mechanism of formation of internal sensation of memory and its relationship with a possible mechanism of LTP. A) During memory retrieval, a cue-stimulus reaching presynaptic terminal A depolarizes its postsynaptic terminal B, re-activates the hemi-fused inter-postsynaptic membrane and activates postsynaptic terminal D, evoking a cellular hallucination of arrival of sensory inputs at  the latter's presynaptic terminal C. In normal conditions, an action potential reaches presynaptic terminal C when the CA3 neuron is activated. Sensory identity of the semblance of activity occurring at the postsynaptic terminal D consists of inputs from the set of neurons {Y} that synapse to the CA3 neuron. The set of neurons {Y} are normally activated by inputs from a set of lower order neurons {X}. The set of neurons {X} in turn are activated by a further large set of its lower order neurons {W}. Continuing this extrapolation toward the sensory level identifies a set of sensory receptors {SR}. {sr1}, {sr2}, and {sr3} are subsets of {SR} and are capable of independently activating the CA3 neuron. Hypothetical packets of sensory stimuli activating sensory receptor sets {sr1}, {sr2}, and {sr3} are called semblions 1, 2, and 3, respectively. The activation of the postsynaptic terminal D by the cue stimulus can lead to the virtual internal sensation of semblions 1, 2, 3 or an integral of them. A CA1 neuron (place cell in the context of spatial memory) is shown to receive sub-threshold excitatory postsynaptic potential (EPSP) from oscillating neuronal activities of its lower order neurons. Cue stimulus-induced activation of postsynaptic terminal D reaches the soma of its neuron in the CA1 region. If the CA1 neuron receives a baseline summated EPSP short of one EPSP to trigger an action potential, then the additional EPSP arriving from the postsynaptic terminal D can add to sub-threshold EPSP, inducing an action potential in the CA1 neuron, resulting in its concurrent activation during memory retrieval; this CA1 neuron will not otherwise be activated in the absence of prior associative learning. This can explain place cell (CA1neuron) firing occurring concurrently with spatial memory retrieval. Bottom Panel: Cross-section through the postsynaptic terminals showing a newly formed functionally LINKed postsynaptic terminals B and D during associative learning. Three other islets are also shown. B) Stimulation of the Schaffer collateral induces LTP by inducing postsynaptic membrane hemi-fusion between postsynaptic terminals that belong to islets of postsynaptic terminals B-D and F-H-J-L forming a mega-islet B-D-F-H-J-L. A regular stimulus at the stimulating electrode has now an increased probability of reaching the recording electrode through the large number of hemi-fused postsynaptic membranes within the large mega-islet, showing a potentiated effect when recorded from the CA1 neuron. Neuronal orders from 1 to 6 are numbered from the sensory receptors. Bottom Panel: Cross-section of an area containing the newly formed mega-islet of functionally LINKed postsynaptic terminals B-D-F-H-J-L formed during LTP induction. Two other islets are also shown. {SR}, Set of sensory receptors; {sr}, subset of sensory receptors. If LTP-induced mega-islets include postsynaptic terminals B and D, it reduces the specificity of retrieved memories in retrieving memories since spread of activity through different non-specific postsynaptic terminals of the islet induces non-specific semblances (From Vadakkan (2012).

The hypothesis has used one key assumption that internal sensation is induced at a specific location by a specific mechanism. Why should this be correct?

In order to build a hypothesis, some assumption has to be made in the beginning. If one assumption can consistently substantiate all the nervous system functions, then the probability for that assumption to be correct is high. This is similar to solving a system of linear equations having a unique solution. When only one variable remains unknown, then using its different relations with other variables the value of the unknown variable can be found mathematically. Alternatively, one is allowed to assign different values that unknown variable and use trial and error methods to solve the system. Similarly, in the case of a biological system where only one variable of internal sensation remains unknown, large number of known variables and their relationships with the unknown variable can be used by trial and error methods to solve the system. In deriving semblance hypothesis, induction of semblances as a system property was assumed to take place at the inter-LINKed postsynaptic terminal (dendritic spine) by the reactivation of the inter-postsynaptic functional LINK due to compelling reasons such as 1) some form of depolarization is always taking place at the postsynaptic terminal continuously, 2) the miniature EPSP generation cannot be blocked completely by any natural or synthetic chemicals on earth, 3) the formation of the inter-postsynaptic LINK can be achieved as a function of simultaneous activation of the abutted postsynaptic terminals during associative learning, 4) induction of semblance can then be derived as a function of lateral activation of the inter-postsynaptic LINK, 5) presence of different types of inter-postsynaptic functional LINKs provides suitability to explain different higher brain functions with varying duration of their existence, 6) it is possible to stabilize the functional LINK, providing ability to retain ability to retrieve memory of associatively learned items or events for different duration of time, 7) the lateral spread of activity through the inter-postsynaptic functional LINK contributes to the horizontal component of the oscillating potentials which is a requirement for inducing the system property of internal sensations, 8) semblance is a virtual property that suits to explain the virtual internal sensations of various higher brain functions, 9) semblance is a first-person property induced within the system towards which only the owner of the nervous system has access. All these fitting conditions make induction of semblance as an appropriate assumption. The possibilities for a spectrum of changes that can be formed during the generation of inter-postsynaptic LINK (see Figure 7) and their maintenance for a wide range of time periods shows the exact features that one would expect from the basic operational mechanism. Inter-postsynaptic functional LINK mechanism can operate in agreement with all the constraints offered by the findings listed in the Table 1 on the front page of this web site. Due to these reasons, the hypothesized mechanism is likely to be correct. 

This work has explained induction of units of internal sensation of memory. How can it explain perception and consciousness?

We first examined memory to derive the mechanism due to the advantages of examining this function.  Changes can be induced during learning and these changes are expected to be used to induce memories. Since these expected changes can be hypothesized, they can be tested to verify the hypothesis. This led to the derivation of inter-postsynaptic functional LINK formation and induction of units of internal sensations at the inter-LINKed spines as the basic operational mechanism. It is reasonable to expect that the basic mechanism of induction of units of internal has shared properties with the internal sensations of both perception and consciousness. Slight modification of the process of induction of units of internal sensations is expected to occur both during perception and in the operational mechanism of consciousness. In the case of perception, a real time process of induction of units of internal sensations has to be explained that can explain large number of known properties of perception. This was carried out to explain visual perception (Vadakkan, 2015c). In the case of consciousness, it has to explain a) why large number of units of internal sensations are getting induced while the animal is at rest, b) what is the net semblance formed by these units of internal sensations, and c) how it form a background matrix upon which internal sensations in response to specific cue stimuli can be efficiently induced (Vadakkan, 2010).

What is the basic logic behind this work?

Let us imagine that there is a solvable system of linear equations. This means that there are few equations that contain several variables and these equations form a system, meaning that they are all inter-related. If we know the values of all the variables except one, then we will be able to find out the value of that unknown variable using simple mathematical methods. But if we examine very carefully, we will see that the relationship of the unknown variable with the known variables within the equations can guide us to understand what the value of that unknown variable is. The above relationships are constraints that allow us to understand the value of the unknown variable. So, instead of using mathematical methods, we can also find the value of the unknown variable by trial and error method.

In a similar manner, nervous system has very large number of variables that are observed as different findings at various levels (molecular, cellular, inter-cellular, electrophysiological, systems, behavioral, and imaging). We already know all those findings and we have made large number of correlations between several findings already. Now we have one unknown variable, which is the generation of internal sensations within the mind. We are afraid to use it, since we don’t know how it is getting generated. Let us now use this variable in our findings. Here is an example. In addition to observing the behavior alone, let us include the fact that during memory retrieval there is internal sensation of memory. Now, internal sensation is the only unknown variable within a large number of findings within the system. Now, we can use trial and error methods by using all the constraints offered by findings from different levels (Given in Table 1 on the front page) to arrive at the mechanism of generation of internal sensations. The only difficulty is that we need to examine very large number of observations from different levels to arrive at the solution and fine-tune it. Semblance hypothesis has used this method.

Are there any convincing evidences to show that this hypothesis is correct?

Foremost, present hypothesis has viewed memories in their true sense as first-person internal sensations. The system has background properties for the induction of units of virtual internal sensations. The continuous depolarization of the spine heads (by both quantal release and EPSPs induced by intermittent arrival of action potentials at their presynaptic terminals by sensory stimuli from environment) sets the background state. Associative learning between two stimuli is expected to generate an inter-postsynaptic (inter-spine) LINK (IPL) between the spines that belong to two different neurons. In the above-explained background state, if one of the stimuli (cue stimulus) can incidentally reactivate the IPL then it is expected to spark a cellular hallucination (semblance) at the inter-LINKed spine of receiving stimulus from the second stimulus. This provides a mechanism for the generation of virtual, first-person internal sensation of memory at physiological time-scales. We can make an extrapolation from the inter-LINKed spine towards the sensory receptors to identify the minimum sensory stimuli required to activate that inter-LINKed spine, which forms units of internal sensation. This unique combination of background state of the system and the mechanism for the generation of virtual first-person internal sensation matches with the expectations of a system that generates mind. This provides the most convincing evidence.

Now, if we look at the above mechanism carefully, we can see that the ability to induce hallucination (that constitutes first-person internal sensation) mandates that the dominant state of the system should be that the depolarization of the inter-LINKed spine head occurs by the arrival of activity from its presynaptic terminal. Therefore, lateral activation by the cue stimulus that depolarizes the inter-LINKed spine head from a lateral direction for retrieval of memory cannot occur continuously beyond a certain length of time. After certain period of time, the system has to enter into a state of sleep (that prevents lateral activations by cue stimuli) following which the system resets itself back to the above dominant state. This provides an explanation for the substantive nature of sleep to the extent that the system will cease to function if it is not allowed to sleep for a few days. This provides another convincing evidence.

The above derived mechanism agrees with all the constraints offered by very large number of observations from different levels. Now, some of the convincing evidences that were found at later stages of investigation are the following. a) Most learning-induced changes will reverse back quickly as the animal moves through the environment, explaining working memory. The proportion of what remains is very less and they remain for varying periods of time for short- and long-term memories. In this regard, we expect that most of the learning-induced mechanism should be able to reverse back quickly. It should leave a small proportion of learning-induced changes to last for varying periods of time. So if we have arrived at the actual mechanism, then we should be able to observe changes to explain the above. By examining Figure 7 in this page, it can be seen that the formation of inter-postsynaptic functional LINKs by exclusion of water of hydration (Fig.7B) requires huge amount of energy and it reverses back quickly. Only a small proportion of these LINKs can form partial and complete hemifusions (Fig.7C,D) and will last for different periods of time. This forms a perfect fit with what we expect from learning-induced changes. This perfect fit shows that this mechanism is inevitable. b) As a continuation of what we found just now, there should be a mechanism for stabilization of learning-induced change for long period of time. Since the stage of complete hemifusion (Fig.7D) can be stabilized by different methods, it provides a suitable mechanism. In addition, when the newly formed inter-LINKed spines become part of an islet of inter-LINKed spines, it will be able to get both activated more frequently enabling its long-term maintenance. c) Furthermore, since dopamine cause spine enlargement (Fig.7E), it augments inter-postsynaptic functional LINK formation and its stabilization. This can result in retention of memories for long period of time. This is another example for a perfect fit. d) Several correlations were found between the ability to learn and induce LTP. The derived mechanism has explained all those correlations and in addition explained some of the remaining uncorrelated observations in the field. This ability of the derived work provides further convincing evidence. e) Ability to provide a framework of a mechanism for perception, explaining various features of visual perception and finding a comparable circuitry for olfactory perception in a remote species Drosophila provides another convincing evidence. f) Ability to provide a framework for internal sensation of consciousness and how anesthetic agents can lead to loss of consciousness provides another evidence. g) Ability to explain large number of common features of neurodegenerative disorders as a loss of function of the normal operational mechanism is another convincing evidence.

What are the implications of this work?

This work informs that the nervous system has two types of circuitries that operate in unison with each other. 1) A functional circuitry that operates through the formation, reactivation and reversal of IPLs. 2) A structural circuitry that operates through the synapses. Since firing of a single neuron can occur when it receives a fraction of its inputs (see section titled "What are the issues with studying neuronal firing (axonal spike") in understanding higher brain functions?" above), several neurons are kept at sub-threshold activation level under the influence of inhibitory neurons (Hangya et al., 2014; Karnani et al., 2014). In this context, arrival of a small fraction of potentials through the IPLs will be able to fire some of those neurons for effective downstream end organ activation. Based on the IPL mechanism, internal sensations of various higher brain functions take place at the inter-neuronal inter-spine level. Since a very minute fraction of inputs can lead to firing of a neuron, it will not be possible to use neuronal firing to understand how internal sensations of various higher brain functions that include motivation, memory, perception, consciousness, aversion, reward, pleasure, anxiety, stress, fear, intentionality, hunger, thirst, and pain. From the derivation of the mechanism, we have seen that computation of units of internal sensations induced at a specific sets of inter-LINKed spines are essential for the generation of specific internal sensations associated with all the above higher brain functions. By artificially altering neuronal firing at the neuronal orders at a level lower than that of the inter-LINKed spines, the internal sensations of most of the higher brain functions can be altered. But to understand the mechanism of generation of internal sensations of a higher brain function, we need to understand the induction of units of internal sensations by the IPL mechanism and their computation.

Why should this hypothesis be correct?

Science always seeks truth. In order to understand certain phenomenon, it is necessary to examine it from multiple angles and make sure that our understanding is correct by all means. When we say that we do not understand the brain, what we really meant are two things. a) we cannot inter-connect findings from different levels, and b) we do not understand how the first-person internal sensations in the mind are being generated. So our task is to find out the still undiscovered key operational mechanism for the generation of internal sensations that is also expected to inter-connect findings from different levels. In this context, we can use all the observations that we have already made from various levels and list the constraints offered by all of them (Given in Table 1 on the front page). From those constraints, it is clear that the basic operational mechanism has to operate within the limited amount of freedom that it has. In other words, the constraints dictate what the solution is. The constraints offered by Nature are the guideposts that we can use to reach the solution. Wherever we reach, it should be the solution. So we need to use all these constraints together and reach at the solution. Since a) the present work strictly adhered to these principles all throughout, b) large number of constraints were used to arrive at the solution, and c) large number of constraints were used to verify the solution, the derived solution is expected to be correct. The solution is expected to be a very unique one. At the same time, it is also expected to be a simple solution and universally present in all species of animals. The derived mechanism matches with these qualities. Now, we can further verify the hypothesis by asking more questions. Making attempts to disprove the hypothesis is also part of this process. These can be continued until we become totally satisfied.

What if this hypothesis is wrong?

This is a constant feeling during any hypothesis building. But if the hypothesis is correct, then the degree of this feeling will reduce with time. There are two ways that a hypothesis of nervous system functions can go wrong. 1) Let us imagine that the nervous system consists of a large jigsaw puzzle in multiple dimensions and that we are trying to solve it using its parts. The most important thing is that we need to collect all the pieces of the puzzle from multiple levels and bring them on to the table. Since we will only solve for what we have on the table, missing few pieces from few levels can lead to a wrong hypothesis. 2) Let us imagine that all the pieces of the puzzle are of the same color. So direct matching of the pieces is the only way to confirm whether we are putting the pieces of the puzzle correctly. In this exercise, we may be able to put together large number of pieces correctly, until we figure out that the remaining pieces won’t fit into the remaining slots. A good jigsaw puzzle will lead to these situations. Now, we need to dismantle everything and start building it again. Is this hypothesis reaching a stage described as above? Did it bring all the parts on to the table for assembling? Since this hypothesis has incorporated as many pieces of the puzzle as possible from multiple levels and was guided by the tight constraints offered by these findings (given in Table 1 on the front page), these possibilities are expected to be very less. But, there is still a possibility that it is wrong. This is the reason why there is an open invitation to everyone for falsifying this hypothesis. Alternatively, if it is wrong, then we should be able to build another hypothesis with more compelling explanations and inter-connectable features.

References

Abbas AK, Villers A, Ris L (2015) Temporal phases of long-term potentiation (LTP):myth or fact? Rev. Neurosci. 26(5):507-46 PubMed

Abbott LF (2008) Theoretical neuroscience rising. Neuron 60(3):489-95 PubMed

Amaral DG, Dent JA (1981) Development of the mossy fibers of the dentate gyrus: I. A light and electron microscopic study of the mossy fibers and their expansions. J Comp Neurol. 195:51-86 PubMed

Antic SD, Zhou WL, Moore AR, Short SM, Ikonomu KD (2010) The decade of the dendritic NMDA spike. J Neurosci Res 88:2991–3001 Pubmed

Beauchamp MS, Sun P, Baum SH, Tolias AS, Yoshor D (2012) Electrocorticography links human temporoparietal junction to visual perception. Nature Neuroscience 15(7)957-9 PubMed

Burette AC, Lesperance T, Crum J, Martone M, Volkmann N, Ellisman MH, Weinberg RJ (2012) Electron Tomographic Analysis of Synaptic Ultrastructure. Journal of Comparative Neurology 520 (12): 2697-2711 PubMed

Chicurel ME, Harris KM (1992) Three-dimensional analysis of the structure and composition of CA3 branched dendritic spines and their synaptic relationships with mossy fiber boutons in the rat hippocampus. J Comp Neurol. 325:169-82 PubMed

Cragg BG (1967) The density of synapses and neurones in the motor and visual areas of the cerebral cortex. J Anat 101(Pt 4):639-654 PubMed

Edelman S (2012) Six challenges to theoretical and philosophical psychology. Front Psychol. 3:219. PubMed

Frotscher ML, Seress L, Schwerdtfeger WK, Buhl E (1991) The mossy cells of the fascia dentata: a comparative study of their fine structure and synaptic connections in rodents and primates. J Comp Neurol. 312:145-63 PubMed

Gallistel CR, Balsam PD (2014) Time to rethink the neural mechanisms of learning and memory. Neurobiol Learn Mem. 108:136–144 PubMed

Gustafsson B, Wigström H (1990) Long-term potentiation in the hippocampal CA1 region, its induction and early temporal development, Prog. Brain Res. 83:223–232  PubMed

Hangya B, Pi H-J, Kvitsiani D, Ranade SP, Kepecs A (2014) From circuit motifs to computations: mapping the behavioral repertoire of cortical interneurons. Curr. Opin. Neurobiol. 26, 117–124  PubMed

Jacob AL, Weinberg RJ (2014) The organization of AMPA receptor subunits at the postsynaptic membrane. Hippocampus. doi:10.1002/hipo.22404 PubMed

Karnani MM, Agetsuma M, Yuste R (2014) A blanket of inhibition: functional inferences from dense inhibitory connectivity. Curr Opin Neurobiol. 26:96-102. PubMed

Konur S, Rabinowitz D, Fenstermaker VL, Yuste R (2003) Systematic regulation of spine sizes and densities in pyramidal neurons. J Neurobiol. 56(2):95-112. PubMed

Martens S, McMahon HT (2008) Mechanisms of membrane fusion: disparate players and common principles. Nat Rev Mol Cell Biol 9(7):543–556 PubMed

Martin SJ, Grimwood PD, Morris RG (2000) Synaptic plasticity and memory: an evaluation of the hypothesis. Ann Rev Neurosci 23:649-711 PubMed

Minsky M (1980) K-lines: a theory of memory. Cognitive Science 4: 117–133 Article

Moore JJ, Ravassard PM, Ho D, Acharya L, Kees AL, Vuong C, Mehta MR (2017) Dynamics of cortical dendritic membrane potential and spikes in freely behaving rats. Science doi: 10.1126/science.aaj1497. PubMed

Monti MM, Vanhaudenhuyse A, Coleman MR, Boly M, Pickard JD, Tshibanda L, Owen AM, Laureys S (2010) Willful modulation of brain activity in disorders of consciousness. N Engl J Med. 362 (7): 579-589. PubMed

Moser EI, Krobert KA, Moser MB, Morris RG. Impaired spatial learning after saturation of long-term potentiation. Science. 1998 Sep 25;281(5385):2038-42. PubMed

Murayama Y, Biebetamann F, Meinecke FC, Muller KR, Augath M, Oeltermann A, Logothetis NK (2010) Relationship between neural and hemodynamic signals during spontaneous activity studied with temporal kernel CCA. Magn Reson Imaging. 28(8):1095-103. PubMed.

Palmer LM, Shai AS, Reeve JE, Anderson HL, Paulsen O, Larkum ME (2014) NMDA spikes enhance action potential generation during sensory input. Nat Neurosci 17(3):383-390 PubMed

Piorazi P, Mel BW (2001) Impact of active dendrites and structural plasticity on the memory capacity of neural tissue. Neuron 29:779-96. PubMed

Popper C (1965) The logic of scientific discovery. ISBN:0-415-27843-0 Book

Rubin DD, Fusi S (2007) Long memory lifetimes require complex synapses and limited sparseness. Front. Comput Neurosci. 1:7. PubMed

Selimbeyoglu A, Parvizi J (2010) Electrical stimulation of the human brain: perceptual and behavioural phenomena reported in the old and new literature. Front. Human Neuroscience 4: 46 PubMed

Shors TJ, Matzel LD (1997) Long-term potentiaion: What's learning got to do with it? Behavioral and brain sciences 20:597-655 PubMed

Spruston N (2008) Pyramidal neurons: dendritic structure and synaptic integration. Nat Rev Neurosci 9(3):206-221 PubMed

Vadakkan KI (2010) Framework of consciousness from semblance of activity at functionally LINKed postsynaptic membranes. Front Psychol. 1:168. doi: 10.3389/fpsyg.2010.00168. PubMed

Vadakkan KI (2011) Processing semblances induced through inter-postsynaptic functional LINKs, presumed biological parallels of K-lines proposed for building artificial intelligence. Frontiers in Neuroengineering 4:8 PubMed

Vadakkan KI (2012a) The nature of "internal sensations" of higher brain functions may be derived from the design rules for artificial machines that can produce them. Journal of Biological Engineering. 5;6(1):21 PubMed

Vadakkan KI (2012b) A structure-function mechanism for schizophrenia. Frontiers in Psychiatry. 28;3:108 PubMed

Vadakkan KI (2013) A supplementary circuit rule-set for the neuronal wiring. Frontiers in Human Neuroscience. 1;7:170 PubMed

Vadakkan KI (2015a) A pressure-reversible cellular mechanism of general anesthetics capable of altering a possible mechanism for consciousness. SpringerPlus 4:485 PubMed

Vadakkan KI (2015b) The functional role of all postsynaptic potentials examined from a first-person frame of reference. Rev Neurosci. doi: 10.1515/revneuro-2015-0036 PubMed

Vadakkan KI (2015c) A framework for the first-person internal sensation of visual perception in mammals and a comparable circuitry for olfactory perception in Drosophila. Springerplus. 2015 Dec 30;4:833. doi: 10.1186/s40064-015-1568-4. PubMed