Frequently Asked Questions

Why is a testable hypothesis necessary?


We are tasked with identifying a unifying operational mechanism for a system composed of approximately 1011 neurons interconnected by around 1015 synapses. This system exhibits a vast array of disparate features that are not easily reconciled within a single framework. Complicating this challenge further is its most significant characteristic: the generation of internal sensations associated with higher brain functions such as perception, memory, and consciousness. Thus, we must develop a hypothesis capable of integrating these diverse features into a coherent explanation—one that also yields testable predictions.


As emphasized by philosopher of science Karl Popper (1965), a scientific hypothesis must be falsifiable—that is, there must be, in principle, a conceivable observation that could refute it. Even in the absence of such a falsifying observation, the hypothesis remains provisional until tested. A single counterexample is sufficient to necessitate the revision or rejection of the hypothesis. However, even the rejection of a hypothesis can provide valuable insights, guiding the refinement of existing theories and the formulation of new, more robust hypotheses.


The nervous system is studied across multiple scientific disciplines and at various levels of organization—including biochemical, cellular, electrophysiological, systems, behavioral, and imaging approaches. To account for all these diverse features, the underlying solution must be a unique and unifying one. In this context, a testable hypothesis becomes especially valuable. While internal sensations cannot be directly observed, this challenge can be addressed indirectly. If a simple, unified solution can be derived that explains findings across all levels of investigation, then that solution is likely to be correct. This logic is analogous to solving for an unknown variable in a system of linear equations where all other variables are already known.


This reasoning served as the motivation to develop a hypothesis for the functions of the nervous system. The proposed hypothesis can then be evaluated within biological systems through several approaches: (a) postdictive analysis, assessing whether findings reported after the hypothesis can be explained by the proposed mechanism; (b) identifying novel predictions that arise from the hypothesis; and (c) examining whether similar circuit features exist across different species for comparable sensory functions. Once validated, the hypothesis can undergo the gold-standard test of replication in engineered systems. This comprehensive approach has the potential to direct neuroscience research more efficiently and cost-effectively toward uncovering the fundamental principles of brain function.

Why is it important to study the formation of first-person inner sensations? Is it possible to understand the brain without examining this aspect?

Every organ in the body is structured to perform specific, well-defined functions. For instance, the heart is designed to pump blood to other organs, and every cell within the heart contributes to this function in some way. Studying the heart without considering its pumping action would be incomplete. Similarly, it is not feasible to study the kidneys without addressing their role in filtration.


When it comes to the brain, its most fundamental and distinctive function is the generation of first-person inner sensations associated with various cognitive processes, such as perception, memory, and thought. Additionally, it governs motor actions that are influenced by survival needs—needs that are, in turn, determined by the subjective experiences of decision-making. For this reason, it is inadequate to study the brain while overlooking its most defining feature: the generation of the inner experiences we refer to as the “mind.”


The brain’s wiring has evolved specifically to produce robust first-person inner sensations, which are critical for the survival of animals. However, scientific inquiry has often neglected the possible anatomical locations and mechanisms responsible for generating these inner sensations—perhaps due to a lack of confidence in our ability to verify them empirically. Yet, by disregarding the pathways involved in producing first-person experiences, we risk overlooking major circuit connections that are fundamental to brain function.


Why haven’t we shown interest in understanding the mechanism that generates first-person properties?


Third-person experimenters currently lack the tools necessary to directly investigate first-person inner sensations, creating a significant barrier to straightforward scientific approaches. Compounding this challenge is the complexity of the nervous system itself, which exhibits an enormous array of features that are being studied independently across various subfields of brain science. Integrating the constraints and insights from all these disciplines into a unified framework is an immense and daunting task. These factors contribute to a pervasive sense of apprehension—an impending fear of failure—among individual researchers attempting to tackle this problem alone. Furthermore, the absence of a coordinated, interdisciplinary consortium focused on understanding the neural basis of first-person experiences has limited progress in this critical area.


Which function should we start examining in order 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 in 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 memor


Learning and memory are ideal functions for studying the operations of the nervous system due to several key advantages. First, 1) changes in the nervous system can be induced during associative learning, and these changes can be empirically verified. Second, 2) we can induce first-person internal sensations through the retrieval of memories on physiological timescales. Third, 3) loss-of-function studies can be conducted to assess the role of specific neural circuits. Fourth, 4) the hypothesis can be tested for its applicability to the consolidation of memories, perception, and consciousness. Fifth, 5) replication in engineered systems can be performed to examine the formation of first-person inner sensations related to memory. Finally, 6) the extensive body of existing data can be leveraged to verify the hypothesis at various stages of development.


Several critical questions can be explored within this framework. For instance: a) What cellular changes occur in parallel during the induction of long-term potentiation (LTP), and how are these changes related to the retrieval of memories? b) How can LTP be correlated with surrogate markers of behavioral motor activities that signal the generation of the internal sensation of memory?

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


Hebb’s postulates suggest that synaptic plasticity strengthens synapses during learning, particularly when two stimuli are associatively learned. However, it remains unclear how the cue stimulus, upon re-encounter, uses these synaptic changes to evoke the memory of the second stimulus. Most studies focus on synaptic changes during learning, often relying on observed behavior during memory retrieval to interpret the process, making it difficult to directly link these changes to behavior and provide a mechanistic explanation for memory.


While synaptic strength changes have been conceptualized as weight adjustments at the junctions between neurons in neural networks, this approach is borrowed from the development of efficient artificial systems. In such systems, weights are adjusted through backpropagation from the postsynaptic neuron to the presynaptic neuron to achieve optimal efficiency. However, this adjustment process is not feasible in the nervous system, where the conduction of signals in chemical synapses is strictly unilateral, flowing from the presynaptic to the postsynaptic terminal.


It is important to note that Hebb’s postulates have guided research thus far, resulting in a wealth of observations. However, the difficulty of obtaining a mechanistic explanation for the first-person internal sensation of memory—based on learning-induced synaptic changes that can be replicated in engineered systems—compels us to re-examine Hebb's postulates. This process involves identifying their limitations and formulating a new postulate. In this context, the present hypothesis was developed by asking the following question: "At the time of memory retrieval, when a sensory stimulus (the cue stimulus) propagates through its path, how can it induce the inner sensation of memory for the associatively learned second stimulus (which traversed a different path during learning) and also trigger behavioral motor activity reminiscent of that second stimulus?"


According to the semblance hypothesis, associative learning between two sensory stimuli leads to specific changes at the sites where these stimuli converge—such as the hippocampus in the case of spatial memory, or the amygdala in fear memory. This hypothesis investigates the interactions between the synapses of the associatively learned stimuli at their points of convergence. Later, when one of these stimuli (the cue stimulus) is encountered again, it is expected to activate these convergence sites and induce the internal sensation of the memory of the second stimulus—within the physiological timescale of milliseconds. As a result, the semblance hypothesis focuses on identifying the precise loci of interaction between the two neuronal pathways, and more specifically, the sub-synaptic components belonging to each pathway where learning-induced changes occur. These changes are hypothesized to enable the cue stimulus to generate the inner sensation of the associated memory. Importantly, this approach does not rely on neuronal firing as a primary mechanism for memory retrieval, for reasons that are explained in the response to the next question.

What are the limitations of studying neuronal firing (somatic spike) in understanding higher brain functions?


Studies that use behavior as a surrogate marker for memory have identified specific sets of neurons that fire during both learning and memory retrieval. For instance, in fear conditioning experiments, exposure to a cue stimulus after learning leads to the activation of additional neurons in the lateral amygdala, compared to those activated by the same stimulus before learning (Schoenbaum et al., 1998; Tye et al., 2008). Furthermore, experimental manipulation of neuronal firing has helped identify distinct neuronal ensembles active during learning and memory retrieval (Tonegawa et al., 2015; Josselyn and Tonegawa, 2020). Despite these findings, neural network research over the past fifty years has encountered significant challenges in replicating the mechanisms of the nervous system. It is hoped that the discovery of a replicable learning mechanism capable of generating memories will shed light on how neuronal ensembles are activated during these processes. Given the critical need to understand memory as a first-person inner experience, it is essential to examine the conditions under which neurons fire. The following findings, which highlight key gaps in our understanding, warrant urgent attention.


1) Research over the past 15 years has revealed that, in addition to axonal action potentials (classically considered as neuronal firing), spiking activity also occurs within dendrites (Antic et al., 2010; Moore et al., 2017). Spikes represent the instantaneous summation of localized electrical potentials. While the purpose of axonal spikes—to transmit signals toward the axon terminals and downstream neurons—is well understood, the functional significance of dendritic spikes remains largely unclear. Uncovering their role is crucial, and it is likely that this can only be achieved by integrating a broad range of experimental observations. A comprehensive, systems-level approach is therefore necessary to reveal the functional attributes of dendritic spikes and their potential role in the mechanisms underlying learning and memory.

2) The number of input connections (postsynaptic terminals, or dendritic spines) varies widely among neurons—from as few as one, as seen in the initial stages of the visual pathway (where passive conductance occurs), to approximately 5,600 in the monkey’s visual cortex, and up to 60,000 in the monkey’s motor cortex (Cragg, 1967). Despite this high number of inputs, the arrival of a relatively small subset is often sufficient to trigger neuronal firing. Early experiments suggested that spatial summation of around 40 excitatory postsynaptic potentials (EPSPs) at the soma could initiate an action potential. More recent modeling studies have shown that pyramidal neurons with tens of thousands of inputs can be activated by the spatial summation of approximately 140 randomly distributed EPSPs at the axon hillock (Palmer et al., 2014; Eyal et al., 2018). While in some cases 40–50 strong EPSPs located close to the soma may suffice, the value of 140 will be used for further discussion. It is also important to note that temporal summation of fewer EPSPs can similarly induce firing. The vast combinatorial possibilities of synaptic inputs that can lead to action potential generation render the output spike non-specific with respect to the exact input configuration.

3) Postsynaptic potentials that contribute to both subthreshold and suprathreshold activation of a neuron do not individually determine whether the neuron fires. As a result, if mechanisms for inducing internal sensations are occurring at these unaccounted synapses, they may be overlooked if one only examines neuronal firing. For instance, consider a pyramidal neuron with approximately 25,000 dendritic spines. During an action, suppose 3,600 of these inputs are activated simultaneously. If only 140 of them summate at the axon hillock to trigger an action potential, the remaining 3,460 excitatory postsynaptic potentials (EPSPs) do not contribute to the firing and are effectively "ignored" in conventional analyses. This raises a critical question: Is this apparent redundancy functionally meaningful, and if so, how?


Assuming 140 EPSPs are sufficient to generate a spike, any input set with fewer than 140 EPSPs that fails to cause firing is also functionally silent in terms of action potential output. Why would evolution preserve such a seemingly inefficient design? One possibility is that this input redundancy enables the nervous system to generate consistent outputs for operating the limited combinations of muscle groups required for survival-related behaviors.


Furthermore, this raises another key question: When a cue stimulus activates one of a neuron's inputs—either subthreshold or suprathreshold—without altering the neuron's firing state, can that stimulus still contribute to internal processing? If so, this could suggest a mechanism by which the brain uses subthreshold synaptic activity to encode the first-person internal sensation of memory, even in the absence of overt firing. Given that we are still seeking a mechanistic explanation for how memories are internally experienced, it becomes essential to explore processes occurring at the input (synaptic) level, independent of whether the neuron fires.

 
4) Postsynaptic potentials (PSPs) generated at dendritic spines located far from the axon hillock—particularly in neurons with extensive dendritic trees like pyramidal neurons—must travel long distances to reach the axon hillock, where they may contribute to action potential generation. However, as these potentials propagate, they degrade significantly (Spruston, 2008), with the extent of attenuation depending on both the distance traveled and the diameter of the dendritic branch. This degradation diminishes their contribution to neuronal firing, raising an important question: Why would evolution conserve a system in which signals are predictably weakened and potentially rendered ineffective for spike generation?


A likely answer is that these distal PSPs serve functions independent of triggering action potentials—except in rare cases where they constitute the final PSP necessary to reach the firing threshold. Since attenuation leads to information loss, it is inefficient to rely solely on spike generation for processing specific inputs. This suggests the necessity for a local operational mechanism—occurring near the site of input at the postsynaptic spine—that preserves input specificity and avoids degradation-related loss.


Such a mechanism would allow each specific synaptic input to contribute meaningfully to a brain function (such as memory specificity) without requiring participation in an action potential. It supports the idea that important computations and representational transformations could occur at the level of dendritic processing, preserving input-specific information even when not contributing to overt spiking. This line of reasoning reinforces the need to look beyond neuronal firing and examine local synaptic dynamics as a plausible substrate for generating first-person internal sensations, such as memory.


5) Since excitatory postsynaptic potentials (EPSPs) degrade as they travel from the dendritic spine to the soma, it is realistic to assume that roughly 140 dendritic inputs must be co-activated to summate and fire a pyramidal neuron. Let us consider a pyramidal neuron with 10,000 dendritic spines (postsynaptic terminals). The number of unique combinations of 140 inputs that can summate to generate an action potential is approximately: [1x104! ÷ (140! x (1x104! – 140!))] ≈ 2.79x10318 Even if we assume a more conservative estimate—say, the neuron has only 3,000 dendritic spines—the number of possible 140-input combinations is still astronomically large: [3x103! ÷ (140! x (3x103! – 140!))] ≈ 1.72x10224 To put this into perspective, the number of atoms in the observable universe is estimated to be around 1082. These calculations only represent combinations of exactly 140 inputs. In reality, neurons may fire with varying numbers of inputs (e.g., from 141 to 10,000), and each input count will contribute additional combinations, making the total number of effective input combinations even larger. 


This extreme degeneracy—where countless combinations of inputs can yield the same neuronal firing—has profound implications. It means that a single spike from a neuron does not uniquely correspond to a specific input pattern. Therefore, interpreting a spike as a marker for a particular cognitive or perceptual event is fundamentally limited. This undermines attempts to directly correlate neural firing with specific internal sensations or higher brain functions unless the underlying input-level mechanisms are understood.

Recognizing this degeneracy is crucial in both neuroscience and artificial intelligence, as it challenges the assumption that specific spikes encode specific meanings. It also underscores the need to investigate local subthreshold synaptic mechanisms—possibly preceding or independent of action potential generation—as the basis for specific internal sensations like memory or perception.


6) Many neurons are frequently maintained in a state of sub-threshold activation, meaning they consistently receive fewer than 140 excitatory postsynaptic potentials (EPSPs)—just below the threshold required to trigger an action potential. Higher-order neurons, particularly those situated downstream from neurons exhibiting oscillatory firing patterns (whose mechanisms, especially the horizontal component of oscillations, are addressed in the present hypothesis), are typically held within this sub-threshold range. For instance, if a neuron receives 138 or 139 inputs, it remains inactive. However, such neurons need only one or two additional inputs to reach the threshold and fire. Consequently, when these higher-order neurons do fire, their activity must be interpreted in a fundamentally different context.

All the above findings demonstrate that studies focusing on neuronal firing and networks of firing neurons do not address the specific mechanisms likely occurring at the level of inputs, such as dendritic spines. Furthermore, when it comes to explaining the first-person internal sensations associated with higher brain functions, current research based on third-person observations is still a step removed from the goal we aim to achieve.

Neuronal firing reflects the integration of multiple inputs, with the neuron reaching a threshold to produce an output. The ability to be activated by either a large number of inputs or a single input suggests an evolutionary mechanism designed to generate a common set of outputs for coordinating muscle movements essential for survival. In exploring mechanisms that induce first-person internal sensations, it’s important to consider potential processes at the input level. Specifically, in cases of input redundancy, we must investigate how inputs interact at convergence sites of associatively learned stimuli. This approach addresses the question: "Where can the cue stimulus induce the sensation of the associated memory?" while accounting for the specificity of the inputs involved. The new hypothesis explores such interactive changes at these convergence points.


Since inner sensations cannot be accessed by third-person observers, how can we study them?


To study phenomena beyond the reach of our sensory systems, we must adopt methodological principles similar to those used in physics to investigate particles and fields that are also inaccessible to direct observation. The core idea parallels a fundamental concept in mathematics: solving a system of linear equations with a unique solution. In such cases, the constraints imposed by the equations guide the process toward the correct solution. Crucially, one must consider all equations in the system to arrive at that solution.


Likewise, by integrating constraints derived from findings across multiple levels of the nervous system (see Table 2 on the first page of this website), it becomes possible to formulate a solution that accounts for the generation and integration of units of internal sensation at physiological timescales. Once such a solution is established, it can be tested by examining its ability to explain known (postdictive) findings. If successful, the next step involves making predictions that can be empirically verified—a standard approach in physics for uncovering new knowledge. This work follows a similar methodology to explore the location and mechanisms underlying the generation of internal sensations, progressing step by step to understand their operational basis.

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

Table 2 on the homepage of this website highlights that one key requirement of the operational mechanism is that it must function at physiological timescales. Blood-oxygen-level-dependent (BOLD) signals, however, are known to initiate slowly, typically peaking around four seconds after neural activity occurs in a given brain region (see Fig. 2 in Monti et al., 2010; Figs. 2–5 in Murayama et al., 2010). Due to this delay, BOLD signals do not directly reflect the normal, real-time functioning of the brain’s operational mechanisms. However, once the actual mechanism is identified, it should also be capable of explaining why oxygen is released at those locations with such a delay. In other words, any proposed mechanism must account for and be consistent with the observed characteristics of BOLD signals.


Which higher brain function can be studied to explore the generation of first-person inner sensations occurring within the matching time scale of milliseconds?


Among various brain functions, memory offers a unique advantage: it allows experimental investigation both by associatively training the system to induce learning-related changes and by examining how these changes might be utilized during memory retrieval. Notably, since no observable cellular changes occur during memory retrieval, it is likely that retrieval involves the passive reactivation of a change induced during learning.


To advance our understanding, it is essential to redefine memories. Memories can be described as first-person, virtual internal sensations of an item in its absence, triggered either by a cue stimulus or spontaneously. Traditionally, memories have been classified as working, short-term, or long-term, based on the time interval between learning and retrieval. Since the qualia—the subjective internal experience—of these retrieved memories are nearly identical across these categories, it is reasonable to hypothesize that:

a) a specific cellular mechanism is activated during learning, and b) the reactivation of learning-induced changes, retained for varying durations, accounts for the different types of memory classified by retrieval timing.


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


Memories are virtual internal sensations that occur at the time of retrieval. The accompanying behavioral motor responses should be viewed as surrogate markers that indicate memory retrieval, rather than as the memory itself. A strong correlation has been observed between long-term potentiation (LTP)—an experimental finding—and these behavioral markers. However, LTP alone presents certain limitations. Notably, LTP requires at least 20 to 30 seconds (Gustafsson and Wigström, 1990), and in some cases over a minute (Escobar and Derrick, 2007) to reach its peak induction, a timescale that does not align with the rapid changes occurring during associative learning. Moreover, LTP has been viewed as insufficient to serve as the sole mechanism for memory storage (Shors and Matzel, 1997; Martin et al., 2000; Piorazi and Mel, 2001). Several studies have also shown that the temporal phases of LTP do not correspond well with those of memory formation and retrieval (Abbas et al., 2015).


Despite these discrepancies—particularly regarding timing—the observed correlation between LTP and behavioral indicators of memory may still hold important clues. It suggests that LTP could reflect aspects of the underlying cellular changes associated with associative learning. Therefore, any proposed mechanism for the formation of first-person internal sensations during memory retrieval must also be capable of explaining the observed relationship between LTP and memory.


The challenges involved in understanding the mechanistic changes that occur during associative learning—particularly those that enable cue-induced internal sensations of retrieved memories and their related psychological effects—have been extensively discussed (Gallistel and Balsam, 2014; Edelman, 2012). These challenges, however, become more tractable once a method is developed to bridge the gap between third-person observable findings and the first-person frame of reference. 

What are the general requirements for a memory hypothesis?

Any proposed solution must account for all observed features of the system across its various levels. The hypothesis should be able to meet the constraints imposed by the full range of empirical findings—see Table 2 on the homepage of this website for a comprehensive list. Given that learning and memory offer an ideal domain for such investigation, it is essential to develop a mechanistic understanding of memory. Since perceiving a stimulus in its absence is, by definition, a hallucination, memories can be conceptualized as cue-induced, cue-specific hallucinations (Minsky, 1980).


This raises a critical question: can we identify a learning mechanism capable of inducing virtual, first-person internal sensations—essentially, memories experienced as cue-induced hallucinations? This question forms the foundation of the semblance hypothesis, initially introduced in a book published in 2007, with revised editions appearing in 2008 and 2010. For the hypothesis to be scientifically valid, it must generate testable predictions that can be empirically verified.


How much time does it take for the learning mechanism to occur?


Humans possess the remarkable ability to associate multiple pairs of sensory stimuli in rapid succession, often within a time frame as short as one second. The learning-induced changes resulting from these rapid associations can later be used to retrieve the corresponding memories. This suggests that the underlying learning mechanism can be completed within a sub-second timescale—on the order of milliseconds. The duration for which these learning-induced changes persist determines when the associated memories can be retrieved, whether shortly after learning or much later. Further details are available in a Preprint

What is the most crucial step to success?

Associative learning induces changes within milliseconds—a physiological timescale. These rapid changes enable a cue stimulus to retrieve the first-person internal sensation of memory, also within milliseconds. If the changes that occur during this brief learning window persist in the system, they can support memory retrieval even after extended periods, forming the basis of long-term memory. Therefore, it is essential to focus on the neural changes that occur during this millisecond window, as they likely represent the core of the nervous system’s operational mechanisms. All delayed molecular changes observed after learning are likely secondary, modulating or supporting the primary changes established during the initial milliseconds.

How can a set of constraints guide us toward a solution?

To solve the complex problem of understanding the nervous system, we must unify a vast number of constraints derived from various levels of its function. This challenge is akin to solving a system of linear equations, where relationships between variables guide us to a unique solution. In mathematics, efficient methods are developed to find solutions quickly, though the core insights often precede formal equations. Likewise, in neuroscience, we must apply a step-by-step approach, using trial and error to derive a solution that explains findings across multiple levels of the system. Given the uniqueness of the solution, this method will likely lead us to the correct answer.


Existing neuroscience research often focuses on findings from a limited subset of the system due to specialization and publication constraints. Traditional theories, like synaptic plasticity, which suggest that changes in synaptic connections drive memory and learning, no longer suffice to explain the complexity observed in current research. A comprehensive solution must integrate findings from all levels of the system. Reaching the correct solution requires considering all variables involved. We must be open to accepting tentative solutions that explain all the findings and verify them through rigorous testing. The hypothesis should be treated as provisional until it can withstand extensive triangulation. Once no alternative explanations remain viable, the solution can be accepted as the best available.


To solve the system, we must integrate findings from diverse levels—systems, behavior, first-person inner sensations, electrophysiological, cellular, and biochemical. The critical missing piece is the variable of first-person inner sensations, which uniquely distinguishes the nervous system. Without considering this, we cannot understand how higher brain functions like perception and memory are generated. Current research relies on behavior to infer inner sensations, but this approach fails to account for the first-person perspective. We need to define the relationship between inner sensations and behavior. For example, if a drug blocks memory retrieval, we must determine whether it prevents the generation of the memory's inner sensation or blocks its connection to behavior. By focusing on how inner sensations are generated and how they relate to behavior, we aim to identify the neural pathways involved in learning and memory retrieval. The solution we propose must align with prior experimental findings and operate within physiological time scales. Verification of hypotheses should be immediate and rigorous, ensuring the accuracy and robustness of any proposed mechanisms.


Should we consider the generation of unitary mechanisms and their integration to produce internal sensations?


In systems of the body that must generate a vast number of outputs (products) using limited resources, these systems often rely on the power of combinatorial effects through unitary mechanisms. A prime example of this is the generation of nearly 1011 specific antibodies to recognize a vast array of potential antigenic molecules from the environment. This is achieved via a combinatorial mechanism that utilizes a finite number of variable (V), joining (J), and sometimes diversity (D) gene segments (Tonegawa, 1983; Janeway et al., 2001). This is possible due to the common ability of different DNA segments to undergo recombination. Another example is the production of a vast array of protein molecules using just 20 different amino acids. This is made possible by the common property of amino acids to form peptide bonds at their ends. Similarly, these amino acids are synthesized from just four different nucleotides, each capable of forming phosphodiester bonds at their ends.


In summary, when a system needs to generate an exceptionally large number of outputs using finite resources, it is highly likely that the system has adopted a combinatorial mechanism. Given that an infinite number of memories are expected to be generated by a finite number of neuronal processes, it is reasonable to assume that memory formation also follows a unitary mechanism, with natural integration occurring at physiological time scales. In this context, it is likely that the system utilizes unitary mechanisms with common properties that allow them to bond together. For memory, it is reasonable to expect the generation of units of internal sensations that integrate to form the internal sensation of memory in response to specific cue stimuli.


When investigating this process, we must assume that a unitary mechanism is in operation. Our search should focus on identifying a learning mechanism capable of inducing these units of internal sensation, as well as a mechanism that integrates them to generate the internal sensation of memory.

Can you explain the derivation of the semblance hypothesis in simple terms?

As the nature of retrieved memories shifts with slight changes to the cue stimulus, this suggests that such variations can trigger the computation of specific units of internal sensation. Understanding how these sensations emerge within physiological timescales requires identifying a plausible cellular mechanism. Since first-person internal sensations are inherently virtual, the aim of the hypothesis was to investigate the system for specific properties and mechanisms capable of producing such experiences. This mechanism should be simple, universally applicable, and able to account for similar functions across different animal species. In developing the hypothesis, careful consideration was given to ensure consistency with the wide range of findings presented in Table 2 on the front page.

The development of the hypothesis involves two major stages, each comprising a series of numbered steps.

Stage I

The initial stage aims to identify mechanisms underlying the changes that occur during associative learning—changes that can later facilitate the retrieval of first-person internal sensations associated with memory. There are multiple ways to approach this problem; here, two methods are presented. Both approaches require meeting certain conditions, which are based on the following assumptions: a) Generating an infinite variety of internal sensations from a finite number of neuronal processes necessitates a combinatorial mechanism built upon unitary processes; and b) The system must support the binding of these units of internal sensation. To enable this, there must be a mechanism that maintains the coherence of the structure-function units—a property referred to as binding.

Method 1:


1. For the purpose of hypothesis development, memory is conceptualized as a virtual internal sensation of a sensory stimulus, given that the actual sensory input from the memorized item is absent during memory retrieval.


2. Humans store thousands of memories, and the retrieval of any specific memory typically requires a corresponding internal or external cue stimulus.


3. Consider a thought experiment: imagine looking at a yellow-colored pen. During this perception, let us assume that a specific set of 105 synapses—selected from the brain’s approximately 1015 synapses—is activated across different neuronal levels (with first-order neurons being closest to the sensory input). If it were possible to selectively stimulate this precise set of 105 synapses, it is reasonable to assume that one would internally visualize or recall the yellow-colored pen.


4. How can a specific set of 105 synapses be selectively activated from among the brain’s approximately 1015 synapses? Put differently, how can each of these 105 synapses be precisely reactivated immediately following associative learning, in order to retrieve the corresponding memory? If we can determine a mechanism for selectively activating just one of these synapses that corresponds to the item being memorized, the same principle could, in theory, be applied to the remaining synapses in the set.


5. An alternative approach is to reframe the problem by asking: what is the minimal condition required for synaptic activation? Since activation of a postsynaptic terminal (such as a dendritic spine) occurs following the arrival of an action potential at its presynaptic terminal, we can consider postsynaptic activation to be functionally equivalent to synaptic activation.


6. In the absence of the original sensory stimulus from the memorized item, we cannot expect action potentials to arrive at the presynaptic terminals along the pathways previously engaged during perception. Therefore, to retrieve the memory, we must find a way to activate the postsynaptic terminals of the synapses that were previously involved—without the arrival of action potentials at their presynaptic terminals during retrieval


7. The activation of a postsynaptic terminal in the absence of an action potential at its corresponding presynaptic terminal may serve as a synaptic-level representation of a virtual sensory experience at the systems or behavioral level. In this framework, the cue stimulus is expected to trigger activity in a specific set of postsynaptic terminals, thereby evoking the internal sensation of a sensory input associated with the learned item. This raises a critical question: following learning, can a cue stimulus activate the postsynaptic terminals of the synaptic pathways through which the original stimulus had previously propagated?


8. This line of reasoning is supported by evidence that artificial or pathological activation of certain brain regions can induce vivid internal sensations—hallucinations—with a compelling sense of reality (Selimbeyoglu and Parvizi, 2010). Such phenomena demonstrate that internally generated activity can give rise to virtual sensory experiences without corresponding external stimuli.


9. At this point, two central questions emerge: Is it possible to activate a postsynaptic terminal without the arrival of an action potential at its presynaptic terminal? How can we selectively activate the same 105 postsynaptic terminals—out of approximately 1015 synapses—that were involved in encoding the memory, immediately after associative learning? Given that we possess only a specific cue stimulus that activates a distinct set of synapses, we arrive at a fundamental question at the synaptic level: How can the activation of a specific set of synapses by a cue stimulus lead to the selective activation of the same 105 postsynaptic terminals that would normally be activated by the original stimulus, thereby enabling memory retrieval?


10. Let us assume that the cue stimulus leads to the activation (depolarization) of the postsynaptic terminals through which signals from the learned item originally propagated. In that case, it is reasonable to propose that some synapses activated by the cue stimulus must be physically proximate to the postsynaptic terminals involved during the initial learning. For memory retrieval to occur, there must exist a mechanism that enables the spread of activity from the synapses activated by the cue stimulus to those postsynaptic terminals associated with the learned item (see Fig. 1).



Figure 1. Illustration of the hypothesized depolarization spread during memory retrieval. During retrieval, the cue stimulus activates presynaptic terminal A, leading to the depolarization of its corresponding postsynaptic membrane B. This depolarization is proposed to spread to another postsynaptic membrane, D. Such a spread is possible only if a functional LINK exists between postsynaptic terminals B and D. Consequently, it can be inferred that a functional LINK must be established between these postsynaptic terminals during the learning process.


Method 2:


Imagine that two sensory stimuli, stimulus 1 and stimulus 2, undergo associative learning. Later, when stimulus 1 (the cue stimulus) is presented, it is expected to evoke the internal sensation of the memory associated with stimulus 2. For this to occur, changes must take place at the convergence points where stimulus 1 and stimulus 2 meet during the learning process. (It is worth noting that the hippocampus, a brain region closely linked to learning and memory, receives inputs from various sensory modalities after three to five neuronal orders from the sensory receptor level).


Now, let us explore what changes occur at the location of convergence between these two sensory stimuli during learning. What critical changes must occur between the synapses activated by stimulus 1 and stimulus 2? Where specifically should these changes take place in the synaptic network? The interaction should occur between the sub-synaptic locations that enable memory retrieval of the second stimulus when the first stimulus arrives, and vice versa. In this context, the interaction between the postsynaptic terminals of stimulus 1 and stimulus 2 is particularly relevant. (This conclusion was reached by systematically examining various sub-synaptic areas to identify properties that would allow them to generate units of internal sensation through a trial-and-error method, as discussed in Section II).


The interaction between these postsynaptic terminals is referred to as the inter-postsynaptic functional LINK (IPL) (Fig. 2). The term "functional" emphasizes that the formation of the LINK is a result of the activities arriving at the postsynaptic terminals activated by stimulus 1 and stimulus 2 during associative learning. During memory retrieval, reactivation of the inter-postsynaptic functional LINK is triggered by the arrival of activity from either stimulus at its corresponding postsynaptic terminal. The term LINK is written in capital letters to signify that it is a critical component of the hypothesis.



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


Various types of inter-postsynaptic functional links formed during associative learning.


The inter-postsynaptic functional LINKS formed during associative learning can take various forms:


a. Water removal between the postsynaptic terminals: In this case, the removal of hydration water between the postsynaptic membranes allows them to abut each other. This process requires a high amount of energy and results in a rapid reversal of the functional LINK. Such changes are typically short-lived and are associated with the mechanisms underlying working memory.

b. Partial hemifusion between postsynaptic terminals: A strong interaction between the postsynaptic terminals can lead to a reversible partial hemifusion. This type of interaction explains the retention of learning-induced mechanisms for a longer period of time compared to the previous type.

c. Complete hemifusion between postsynaptic terminals: Further interaction can lead to a reversible complete hemifusion between the postsynaptic terminals. This allows for even longer retention of the learning-induced mechanism.

d. Long-term retention through stabilization: If complete hemifusion is maintained for a certain period, it is likely that stabilizing mechanisms will enable the long-term maintenance of this connection, providing lasting changes to the system.


Inter-LINKing spines are expected to belong to different neurons


At this point, it is crucial to understand the origin of the spines that are being inter-LINKed. According to studies based on the synaptic plasticity hypothesis, synapses on the adjacent spines on a dendritic branch interact with one another to generate clustered plasticity (Govindarajan et al., 2006; Stuart and Spruston, 2015; Bloss et al., 2018). These interactions are thought to be responsible for the computation necessary for brain functions.


However, in contrast to these hypotheses, the present work approaches the issue differently. The inter-LINKs formed during associative learning are expected to generate first-person internal sensations at physiological timescales during memory retrieval. Additionally, these links are anticipated to trigger motor activity corresponding to the retrieved memory. This leads to the question: “To which neuron or neurons should the inter-LINKed spines belong in order to maintain specific outputs associated with each of the associatively learned sensory inputs?”


The immediate answer is that these spines should belong to different neurons (Fig. 3). Furthermore, since the mean distance between spines is typically greater than the mean diameter of the spines themselves (Konur et al., 2003), it follows that the inter-LINKing postsynaptic terminals must belong to distinct neurons. This is expected to be a general rule. However, there could be exceptions, such as when axonal terminals from newly formed granule neurons synapse with a fixed number of dendritic spines on a CA3 neuron (Fig. 4).



Figure 3. A) Sensory stimuli 1 and 2 activate sensory receptors, and these activities propagate through several orders of neurons. For associative learning to occur between these two sensory stimuli, it is expected that they will converge at some point along their respective pathways. Let us assume that each stimulus is associated with a specific motor output (as demonstrated by the fact that the last order of neurons sends outputs to muscle fibers). At the location of convergence, neuronal processes are expected to generate a functional LINK between the pathways of these two stimuli. Following associative learning, each stimulus is anticipated to trigger a motor response corresponding to the second stimulus, along with the generation of the internal sensation associated with it. The critical question then becomes: Where can the inter-postsynaptic functional LINK (IPL) occur between the two paths? B) Can sensory inputs reach two closely located spines on a single neuron? While an inter-postsynaptic functional LINK (IPL) could theoretically form between these spines, there are practical challenges with such an arrangement: a) The outputs of the associatively learned sensory stimuli 1 and 2 would need to pass through the same neuron (Neuron N). This creates a problem, as it would not allow for the motor responses expected in a conditioning paradigm, where distinct motor outputs are typically associated with each sensory input. b) Additionally, the observation that the average inter-spine distance exceeds the average spine diameter (Kamme, 2003) suggests that any inter-spine interaction must occur through the dendritic branch. Since there are no known electrically isolated pathways between spines along the dendritic branch, this arrangement lacks the capability to provide a universal operational mechanism for such interactions. C) This leads to the critical question: "When one sensory input arrives at one spine of a neuron, what should be the pathway through which the second sensory stimulus arrives in order to fulfill the following requirements: a) A learning mechanism that can induce the internal sensation of memory, and b) Output activity that generates a motor response reminiscent of the arrival of the second stimulus?" D) The next feasible configuration involves an interaction between spines on neurons that belong to separate neurons. The figure illustrates the convergence of stimulus 1 and stimulus 2 onto the spines of two distinct neurons (denoted as N, with each neuron marked in a different color), each providing separate motor outputs. If these spines are closely abutted and their interaction during learning generates an inter-spine mechanism that can be reactivated by one of the stimuli (post-learning) to evoke the internal sensation of the second stimulus, then this configuration could serve as a viable candidate for explaining memory retrieval. Furthermore, if the inter-spine mechanism exhibits different half-lives to account for short- and long-term memory retention, this would strengthen its suitability. Additionally, it is anticipated that a strong connection exists between the inter-spine interaction and a narrow range of frequencies of oscillating extracellular potentials. In this context: S: sensory stimulus. N: neuron. The inter-postsynaptic functional LINKs can be conceptualized as biological equivalents of the K-lines proposed by Minsky (Minsky, 1980).

.


Figure 4. Figures illustrate the importance of inferring that the nearest spine to a given spine on the dendritic branch of a neuron is most likely to belong to a different dendrite, which is probably from another neuron. A) Golgi staining reveals a dendritic branch with spines, which are the inputs to a neuron. The output terminals of the preceding neurons that synapse with these spines are not stained by the Golgi method, suggesting their presence adjacent to the spines. This assumption is supported by electron microscopy, as shown in Figure B, where both the dendritic spines and output terminals can be observed. In this context, nearly any 140 such inputs arriving at the neuronal axon hillock will fire the neuron, leading to the propagation of a signal to all of its output terminals. B) The electron microscopic image demonstrates the crowded nature of neuronal processes and other cells within the neural environment. The extracellular matrix, which exists between neuronal processes and glial cells, serves as an insulating medium that prevents the spread of signals between different neurons that are not connected. Notably, this space is very thin. Arrow: The arrow in Figure A points to a spine, which is further shown in Figure B. The spine in Figure B is marked by a postsynaptic density (PSD), indicated by a dense dark area. Adjacent to the spine is a presynaptic terminal containing synaptic vesicles (for further clarity, refer to Figure 9). As observed in Figure A, the mean inter-spine distance exceeds the mean spine diameter. This suggests that the nearest spine to any given spine on a dendritic branch likely belongs to another dendrite, which, in turn, most likely belongs to a different neuron. If brain function is to arise through spine-to-spine interactions, and if output from a different neuron is required (as indicated by classical conditioning experiments), then the nearest spine to a given spine on a dendritic branch is most probably located on a dendrite belonging to another neuron (Note: No scale bars are provided in the figures). 


Inter-postsynaptic functional LINKs can be regarded as the biological equivalents of K-lines

K-lines were proposed by Minsky (1980) as the key operational changes that occur during associative learning. These changes are thought to play a crucial role in facilitating the necessary functions during memory retrieval. The concept of K-lines emerged from Minsky's efforts to understand natural intelligence and translate these insights into engineered systems.

Stage II

In the next stage, we derive the basic units of semblances occurring at the functionally inter-LINKed postsynaptic terminals. To examine the effect of stimulus arrival during memory retrieval, let us consider stimulus 1 arriving as the cue stimulus (Figure 5). It reaches synapse A-B, where the postsynaptic potential at terminal B propagates through the inter-postsynaptic functional LINKs and eventually reaches postsynaptic terminal D. As discussed in Method 1, the arrival of stimulus 1 (the cue stimulus) occurs infrequently. Therefore, when postsynaptic terminal D is depolarized incidentally—without the arrival of an action potential at its corresponding presynaptic terminal C—terminal D is expected to experience a form of "cellular hallucination." In this state, terminal D perceives that it is receiving sensory inputs through its presynaptic terminal C, resulting in a “semblance.” This semblance can give rise to units of virtual inner sensation, corresponding to the memory being retrieved. This mechanism aligns with Minsky's (1980) expectations for memory processes, provided that a specific operational logic is in place at this location. Before exploring the operational logic required for the generation of internal sensations, two important questions need to be addressed: 1) How can a cellular hallucination (semblance) be induced at the inter-LINKed postsynaptic terminal D, which was previously activated by the item whose memory is to be retrieved? 2) What is the sensory content of this hallucination?

.


Figure 5. Extracellular ionic changes during synaptic transmission and depolarization propagation along the IPL contributing vector components to oscillating extracellular potentials. A) During memory retrieval, the cue stimulus reaches presynaptic terminal A, leading to the depolarization of its postsynaptic membrane B. This process re-activates the inter-postsynaptic functional LINK. As a result, depolarization propagates to postsynaptic membrane D, causing a cellular hallucination at postsynaptic terminal D, where it perceives the arrival of sensory stimuli through its presynaptic terminal C. This phenomenon is referred to as "semblance." B) Synaptic transmission in synapse A-B and propagation of depolarization through the inter-postsynaptic LINK B-D generate ionic changes within the extracellular matrix space. These changes contribute vector components that drive oscillating extracellular potentials.


Synaptic transmission through synapse A-B and the inter-postsynaptic LINK B-D generates vector components that contribute to the oscillating extracellular potentials (Figure 5B).

As the learning events continue, one of the postsynaptic terminals that was previously involved in an earlier learning event (either B or D in Figure 5) will be used to form functional LINKS with the postsynaptic terminals of neighboring synapses. These additional postsynaptic terminals are shown on the right side of postsynaptic terminal D in the left panel of Figure 6. As this process progresses, it leads to the formation of clusters of inter-LINKed postsynaptic terminals, which can be reactivated during memory retrieval. These clusters, or "islets," are illustrated in the right panel of Figure 6.



Figure 6. Left Panel: An illustration depicting the formation of islets of LINKed postsynaptic terminals. As learning events continue beyond the initial learning phase, multiple inter-postsynaptic LINKS can form between the involved postsynaptic terminals (dendritic spine heads). In this example, only two presynaptic terminals (A and C) and two postsynaptic terminals (B and D) are marked. Several other postsynaptic terminals are arranged in a horizontal plane. The dotted line represents a cross-section through the inter-LINKed postsynaptic terminals. Right Panel: A hypothetical cross-sectional view of LINKed postsynaptic terminals from synapses in one horizontal plane of a brain region (corresponding to the horizontal dotted line across the postsynaptic membranes in the left panel). In this illustration, all postsynaptic membranes are assumed to be in the same plane. Postsynaptic membranes are represented as small dark circles (broken arrow). During learning, functional LINKs between activated postsynaptic terminals are established. Continued learning involving any of these synapses increases the number of interconnected postsynaptic membranes, forming islets of functionally LINKed postsynaptic terminals (solid arrow). Multiple LINKS within an islet allow for the spread of postsynaptic potentials across the islet. These individual islets are expected to functionally operate independently from one another.


It is anticipated that the fundamental units of semblances occur at the functionally inter-LINKed postsynaptic terminal (see Fig. 5). To fully understand this process, two key questions must be addressed: 1) How is a cellular hallucination (semblance) induced at the inter-LINKed postsynaptic terminal D, which was previously activated by the item whose memory is to be retrieved? 2) What is the sensory content of this hallucination?

What is the logic behind the generation of cellular hallucinations (semblance)?


Semblance refers to the mechanism by which virtual internal sensations are generated. The search for a cellular site capable of supporting such a mechanism led to the identification of the inter-postsynaptic functional LINK as a necessary requirement. As illustrated in Figure 5, when a cue stimulus arrives at postsynaptic terminal B and reactivates the inter-postsynaptic functional LINK, it results in the activation of postsynaptic terminal D. This raises a critical question: What causes postsynaptic terminal D to experience a cellular-level hallucination (semblance), as if it were receiving input from its own presynaptic terminal C?


The logic can be outlined as follows: under normal conditions, postsynaptic terminal D is activated by presynaptic terminal C. To ensure the fidelity of this relationship, it appears that nature has devised a robust mechanism. Even at rest (including during sleep), presynaptic terminal C continuously releases neurotransmitter molecules in quantal fashion from synaptic vesicles. This ongoing release generates regular miniature synaptic potentials at postsynaptic terminal D. The cumulative effect of these is reflected in miniature excitatory postsynaptic potentials (mEPSPs or “minis”). The fact that mEPSPs cannot be completely blocked—even under experimental conditions—suggests that this is a highly conserved, default function of the nervous system.


Another essential condition for semblance formation is the presence of oscillatory neuronal activity. Notably, Beauchamp et al. (2012) demonstrated 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. This finding underscores the importance of oscillatory activity as a systemic requirement for generating internal sensations through semblance.


The lateral spread of activity via inter-postsynaptic functional LINKs likely contributes to the horizontal component of these oscillations, while the vertical component may arise from synaptic activity between vertically arranged neurons in the cortex. Since inter-postsynaptic spread of potentials occurs perpendicularly to the traditional trans-synaptic flow, this structural and functional organization may explain the waveform of oscillatory potentials observed all across the cortex. 


What is meant by "tricking" the inter-LINKed spine into hallucinating?


The inter-LINKed spine heads—like all other postsynaptic terminals—are continuously depolarized by quantally released neurotransmitter molecules, even during sleep. This persistent background activity establishes the dominant state of the system. It is this state that enables a laterally arriving depolarization via the inter-postsynaptic LINK (IPL) to trick the inter-LINKed spine into perceiving that it is receiving sensory input from its own presynaptic terminal—thus creating a cellular-level hallucination. To get a clear grasp of "tricking a system to hallucinate" effect, it is necessary to prime our brains with the following examples. The following conditions maintain certain dominant states that facilitate induction of internal misrepresentations.

1. How pickpockets exploit sensory dominance?  A compelling real-life example of this principle can be seen in the tactics employed by pickpockets. When I was in sixth grade, we read The Adventures of Tom Sawyer by Mark Twain. In the story, Tom Sawyer and a group of boys receive training in pickpocketing, which involves diverting the victim’s attention. The strategy involves introducing alternative sensory stimuli or waiting for naturally occurring ones to obscure the act of theft. Figure 7 illustrates how the continuous, natural movement of the gluteal muscles during walking creates a dominant sensory state. This state enables the pickpocket to ensure that the sensation of the wallet being taken is perceived as a normal, unremarkable part of the routine movement.


A modern version of this phenomenon can be observed in a video demonstration (Video). The video shows that pickpockets are especially successful when the victim is walking up or down stairs. In such situations, the brain is already processing regular sensory input from movement. This ongoing, familiar background activity creates a dominant sensory state. As a result, when the pickpocket makes contact, the brain interprets it as part of the expected stimuli—effectively hallucinating that nothing unusual is happening. This example highlights how maintaining a dominant state of background sensory activity can trick the nervous system into misinterpreting new inputs, allowing a deceptive event to go unnoticed.

.


Figure 7. A cartoon illustrating a pickpocket stealing from a walking individual. The dominant sensory input from the natural movements of the gluteal region causes any additional stimuli in that area—such as the act of pickpocketing—to be misinterpreted as part of the routine motion. This misperception allows the theft to go unnoticed (Image source: Wikipedia; original source not identified).


2. Masking sensory input during injections in animals: Another real-world example of exploiting the dominant sensory state can be seen in veterinary practice. A common technique for administering injections to animals involves first patting the intended injection site, followed by tapping the area repeatedly with moderate intensity. By doing this—say, tapping ten times—the tapping becomes the expected, dominant sensory input at that location. When the needle is inserted on the eleventh contact, the animal is less likely to notice it, as the nervous system interprets the sensation as yet another tap (Video).  The key takeaway is that in order to induce this kind of perceptual "hallucination"—where a painful or unusual stimulus is misinterpreted as something benign—one must first establish a dominant background sensory state. Within this context, the injection is perceived not as a distinct or threatening event, but simply as a continuation of the familiar stimulus.


The two examples provided may not be perfect, but they offer useful insight into how a system can be tricked into hallucinating—given that a dominant sensory state is maintained. This raises an important question: how does the nervous system come to operate in this manner?


Upon closer examination, we find that synapses continuously release neurotransmitters in a quantal fashion from presynaptic vesicles, which depolarize the postsynaptic spine heads, including those of the inter-LINKed spines. There are no known toxins on Earth that can completely block this quantal release (and, for our sake, let’s hope we don’t encounter any such toxins from other planets or moons!). In a system where synaptic junctions are engaged in constant quantal release, it is possible that, during the early stages of evolution, two spines (postsynaptic terminals) might have accidentally abutted, forming an inter-postsynaptic link (IPL) when two simultaneous stimuli arrived. This formation would have allowed one of the stimuli to propagate postsynaptic potentials, depolarizing the inter-LINKed spine laterally. As a result, the inter-LINKed spine would be tricked into hallucinating that it was receiving sensory input from the environment via its own presynaptic terminal.


This mechanism likely provided a survival advantage to early animals, particularly when the fastest or first-arriving stimulus (such as light, or sound from a hidden source) reached the nervous system. Over time, this property would have been refined and passed down through generations, eventually becoming a fundamental operational feature of the nervous system. Article

It is important to emphasize the role of maintaining a dominant state of the system in order to trick the nervous system into hallucinating sensory content from an associatively learned second item. In the example of injecting a cow, tapping the injection site ten times allows the needle to be inserted on the eleventh tap without the animal noticing. To administer a second injection at the same location, the tapping must be repeated for another ten times, effectively resetting the system’s sensory background state. This ensures that the animal’s nervous system perceives the needle as just another regular tap, rather than an invasive stimulus. While this may not be a perfect analogy, it captures the essential idea. This concept shares similarities with sleep in animals. Throughout the evolutionary development of nervous systems on Earth, where day and night cycles occur, sleep offered a crucial opportunity for resetting the system. During sleep, the continuous quantal release of neurotransmitter molecules depolarizing the spine heads established a dominant, stable state of the system. This process highlights the substantive—not merely indispensable—nature of sleep in maintaining the proper functioning of the nervous system. Article

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


Cellular hallucination refers to the perception of a stimulus in its absence. As Minsky (1980) described, memories can be seen as cue-directed, cue-specific hallucinations. To understand the sensory content of a cellular hallucination, one must examine the site where it is induced. For instance, if two stimuli converge at a particular neural location during learning, and later only one of them is presented, that location may generate a cellular hallucination representing the sensory features of the absent stimulus. For such a hallucination to occur, two conditions must be met: first, certain background features must be present at the convergence site; second, there must be an inherent system property capable of producing the hallucination. To grasp the qualia—the subjective sensory qualities—of this hallucinated content, we must extrapolate from the convergence site back toward the level of the sensory receptors (Figure 9). 

.


Figure 9. Schematic representation of sensory content of first-person property during reactivation of an inter-postsynaptic functional LINK (IPL). The grey 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 5. During memory retrieval, a cue-stimulus reaching presynaptic terminal A depolarizes its postsynaptic membrane B. The resulting excitatory postsynaptic potential (EPSP) re-activates the IPL that activates postsynaptic membrane D from its side, evoking a hallucination that the latter is being depolarized by its presynaptic terminal C. This generates a semblance of arrival of a sensory stimulus from the environment at the presynaptic terminal C. This can be viewed as the basic operational mechanism for the first-person inner sensation. An extrapolation from presynaptic terminal C can be carried out as follows. Presynaptic terminal C belongs to the neuron Z that in turn receives inputs from the set of neurons {Y}. 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 may not be necessary to stimulate the entire receptor set {SR} to stimulate the neuron Z. Stimulation of either one of the subsets of sensory receptor sets {sr1}, {sr2}, or {sr3} of the set {SR} may be sufficient to independently activate neuron Z. The dimensions of hypothetical packets of sensory stimuli capable of activating the sensory receptor subsets {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. In the figure, a cue stimulus can cause postsynaptic terminal D to hallucinate about any of the semblions 1, 2, 3 or an integral of them. Since the cue stimulus can reactivate large number of IPLs, it can generate large number of semblions. There is a computation occurring between the semblions that provide the final qualia of inner sensation evoked in response to a specific cue stimulus. Memory is generated only when the brain operates in a narrow range of frequency of oscillating extracellular potentials. Note that synaptic transmission and propagation of potentials across the IPLs in near perpendicular directions contribute vector components to generate the oscillating extracellular potentials (marked by the waveform) (Modified from Vadakkan, 2011).


The dimensions of internal sensations arising from the lateral activation of postsynaptic terminal D can be inferred from the nature of the sensory stimuli capable of activating the sensory receptors in the set {SR}. It is likely that the activation of specific subsets within this set—such as {sr1}, {sr2}, or {sr3} (see Figs. 9 & 10)—is sufficient to trigger the activation of postsynaptic terminal D. Based on this, we hypothesize the existence of a minimal packet of sensory stimuli, termed a semblion, which is capable of activating one of these subsets of sensory receptors. The semblion thus represents the basic unit of internal sensation associated with memory retrieval.

.


Figure 10. An alternate description is shown in the figure below. A) What can spark a unit of internal sensation when stimulus1 (S1) arrives at one spine of the inter-postsynaptic functional LINK (IPL) that was formed with the spine of another neuron at the time of learning? The background conditions at the inter-LINKed second spine is that a) the spine head is getting continuously depolarized by the quantal release of neurotransmitter molecules from its presynaptic terminal all the time, which is shown by small vertical lines in the figure, and b) large postsynaptic potential generated by the intermittent arrival of a volley of neurotransmitter molecules when an action potential arrives at its presynaptic terminal (shown by a large vertical line). B) The activation of the inter-LINKed spine from a lateral direction sparks a hallucination that it is receiving a sensory input from the environment through its presynaptic terminal. By making a retrograde extrapolation from the inter-LINKed second spine's presynaptic terminal, we can identify the sensory receptors from where the activity can arrive. Everything here on wards is not associated with neurotransmission. It is virtual in nature. Even though any set of 140 inputs arriving from locations close to the soma (or nearly 140 random inputs arriving from anywhere from the dendritic tree) out of tens of thousands of its inputs (marked in the figure as 8000 - 30,000) can fire the neuron N, the retrograde extrapolation should include all the inputs of neuron N. C) Continuing this process to the level of the sensory receptors identifies a large set of sensory receptors {SR}. From this, a large number of sets of minimum sensory stimuli whose activation can activate subsets of the large sensory receptor set {SR} can be found. D) This extrapolation is continued towards the lower orders of neurons until it reaches the level of the sensory receptors. This will identify all the sensory receptors. The content of the hallucination occurring at the inter-LINKed second spine is about the sensory stimuli stimulating these sensory receptors. Here, we have to think again and ask, "Is it necessary to stimulate all these receptors for an action potential to arrive at the presynaptic terminal of the inter-LINKed spine in real life?" This need not be necessary. In fact, activation of a small subset of these receptors will be able to generate an action potential of the presynaptic terminal's neuron. In other words, content of hallucination at the inter-LINKed spine can be of a sensory stimulus that can activate a fraction of sensory receptors {SR} that are drawn as round dense areas on the sensory receptor (SR) layer in the figures D and E. The content of hallucination can be a hypothetical packet of minimum sensory stimuli activating a minimum set of sensory receptors. E) At the top of this picture a set of minimum sensory stimuli that forms the content of the hallucination (internal sensation in the absence of arrival of a sensory stimulus), which is called a "semblion" is shown above one of the sensory receptor subsets.


Comparing the various integrated products of these semblions with the actual sensory features of the item whose memory is being retrieved can provide critical insights into the neural algorithm underlying memory retrieval. Notably, the net semblance can surpass a certain threshold without compromising the accuracy of the retrieved memory. Because the number of inter-LINKed spines is continuously modified through associative learning throughout life, the characteristics of the associated semblions are also expected to change gradually. This leads to a progressive refinement of the net semblances for memory.


Beyond inducing semblances, the reactivation of inter-postsynaptic LINKs (IPLs) can also supply additional excitatory potentials to the connected spines. If these potentials are sufficient to bring the postsynaptic neurons to firing threshold, the neurons can fire. When motor neurons are involved, this neuronal activity can result in motor outputs such as speech or behavior (Fig.11).

.



Figure 11. 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. A cue stimulus undergoes associative learning with item 1 and item 2. Following this, the cue stimulus retrieves memories of items 1 and 2. Along with retrieving memory item 2, the cue stimulus also evokes a motor response through the motor neuron. Later, the same cue stimulus undergoes a second associative learning with item 3. Following this, the cue stimulus evokes internal sensations (semblances) of learned items 1, 2 and 3. However, as the semblance for item 3 was evoked, it results in the inhibition of the motor activity. This is an example of behavioral inhibition in the frontal cortex. Reward-induced associative learning may be facilitated by dopamine-induced enlargement of dendritic spines (Yagishita et al., 2014) that favors IPL formation and stabilization. In the hippocampus, reactivation of certain IPL in response to specific locations can lead to firing of certain subthreshold-activated CA1 neurons (place cells). This explains how spatial memories are associated with place cell firing. EPSP: excitatory postsynaptic potential. nth 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 the inter-postsynaptic functional LINK (IPL)?

Multiple mechanisms for the formation of inter-postsynaptic LINKs (IPLs) are likely and necessary to account for the generation of internal sensations underlying various higher brain functions, which operate across different time scales. These distinct types of IPLs, each with varying half-lives, provide a plausible framework for explaining perception, as well as working, short-term, and long-term memory. An overview of some of these mechanisms is presented in Figure 12.

.



Figure 12. Different types of reversible inter-postsynaptic functional LINKs (IPLs). 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 into the synaptic cleft. This quantal release is a continuous process (even during rest) that leads to the generation of very small potentials on postsynaptic membrane D. Note the presence of a hydrophilic region separating postsynaptic terminals B and D. When an action potential arrives at presynaptic terminal A, it activates synapse A–B 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 timescales can provide sufficient energy to exclude the inter-postsynaptic hydrophilic region, allowing close contact between the postsynaptic membranes in 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 A–B reactivates this IPL 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 timescales. 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 up to 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 trans-membrane protein across the hemifused segment can maintain the IPL 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 that results in latter's expansion, which will augment IPL formation. This can explain the action of dopamine on learning. Furthermore, it can sustain the hemifused LINK for a 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, 2015a, b).


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

New technologies are required to detect close membrane-to-membrane contact through hydration exclusion in vivo (Fig. 12B). To investigate reversible hemifusion between adjacent spine membranes at sites of converging sensory inputs—where the extracellular matrix space is minimal—advanced live imaging techniques are necessary to test this hypothesis. Since only a very small portion of the spine membrane surface is sufficient for inter-postsynaptic LINK (IPL) formation, comprehensive electron microscopy (EM) studies employing serial sectioning across entire postsynaptic terminals may help identify stabilized hemifused IPLs. In hippocampal regions that have undergone extensive associative learning, a high number of stabilized hemifused areas are expected, likely maintained by the insertion of transmembrane proteins. Furthermore, dendritic excrescences in the CA3 region can be examined for evidence of hemifused inter-spine membrane structures.

Alternatively, electron microscopy (EM) images of hippocampal regions from previous, unrelated studies can be re-examined for relevant structural clues. However, many earlier EM images lack the resolution needed to clearly visualize the membrane bilayers. In a recent high-resolution EM study (Fig. 13), closely apposed membrane areas can be seen, suggesting a possible absence of inter-membrane extracellular matrix (ECM) space. Since tissue dehydration during sample processing can contribute to such appearances, confirming true inter-membrane contacts with hydration exclusion will require new in vivo imaging techniques. In Fig. 13, regions displaying two membrane layers—rather than the expected four—at sites where spines are in close apposition may indicate hemifusion. However, such appearances could also result from membrane rotation or dehydration artifacts during tissue processing. Despite these limitations, these observations offer promising leads for continued investigation. The presence of multiple fused spine heads sharing a single spine neck—particularly in dendritic excrescences of the CA3 region, as reported by Amaral and Dent (1981), Chicurel and Harris (1992), and Frotscher et al. (1991)—may represent a structural adaptation stemming from long-standing IPLs.

.



Figure 13. This is a modified version of Figure 4D from Burette et al. (2012), Electron Tomographic Analysis of Synaptic Ultrastructure, Journal of Comparative Neurology, 520(12):2697–2711. In this version, two arrows—one red and one green—have been added for illustrative purposes. The red arrow indicates a region likely representing an inter-postsynaptic area with only two visible membrane layers, instead of the four typically expected when two separate spines are in proximity. This region is suspected to be a site of inter-spine hemifusion. Notably, if the structure marked by the red arrow is indeed a dendritic spine, it does not exhibit a postsynaptic density (PSD) in this particular section. However, it is reasonable to expect that the spine might display a PSD at another location, where it forms a synapse with a presynaptic terminal—possibly from a different neuron—thus supporting the possibility of inter-neuronal inter-spine hemifusion. The green arrow points to another area of close membrane apposition, potentially representing a site of partial hemifusion. While tissue distortions or membrane folding during processing cannot be ruled out, it is unlikely that such artifacts would consistently produce closely apposed membranes over distances as short as ~100 nm. These findings suggest the need for dedicated studies to confirm the presence and significance of such structures. Furthermore, as some of the cellular processes in the region may belong to astrocytic pedicles, additional investigation is required to distinguish neuronal from glial elements. The scale bar represents 100 nm. Unlike traditional EM techniques that typically use ~5 µm thick sections, this study employed ~120 nm sections, significantly increasing the likelihood of detecting suspected hemifused regions of comparable length. Importantly, because hemifusion is hypothesized to occur along linear trajectories, the detection of such a structural pattern in a random tomographic section strongly suggests a high probability of its general presence. This is consistent with the proposed existence of islets of inter-LINKed spines (see Fig. 5 on this page), further emphasizing the need for targeted studies to validate these findings.


Why have we not discovered these inter-postsynaptic functional links (IPLs) until now?

First, there has historically been no impetus to investigate mechanisms for the exclusion of water of hydration at inter-neuronal inter-spine regions activated during associative learning. Second, no dedicated studies have been conducted to image the complete lipid bilayer architecture of an entire dendritic spine and its interactions with adjacent spines from different neurons, with the aim of examining the formation and reversal of inter-postsynaptic functional LINKs (IPLs). Different IPL interactions may involve: (a) the exclusion of water of hydration between membranes, (b) the formation of partial or complete inter-spine hemifusion, and (c) the formation of inter-spine fusion under pathological conditions. Although these mechanisms may seem straightforward conceptually, their investigation is a substantial technical challenge. Given that IPLs are expected to span lengths of only ~10 nm, capturing the necessary ultrastructural detail requires high-resolution imaging of the entire spine membrane.

It appears that all the steps above rely on third-person observations. Where does the examination from a first-person frame of reference fit in?

In Figure 9, the steps involved in determining the sensory content of a cellular hallucination induced at postsynaptic terminal D require examination from a first-person frame of reference. This process entails a backward search from postsynaptic terminal D toward the sensory receptor level to identify the minimal subset of sensory receptors whose activation can trigger terminal D. The minimal sensory stimuli capable of activating this subset of receptors constitute the semblion—the basic unit of internal sensation. This backward extrapolation, from the site of postsynaptic activation to the sensory receptor level, is an implicit process underlying internal sensations across all higher brain functions. Through this approach, we conceptualize the packets of sensory stimuli—the content of the unit of internal sensation—from a first-person experiential perspective.


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

For a more detailed description, see published article

The semblance hypothesis was developed to explain the plausible synaptic changes that occur during learning—specifically, those capable of evoking a virtual internal sensation of a sensory stimulus during memory retrieval. The formation of inter-postsynaptic functional LINKs (IPLs) appears to be a common denominator between learning and long-term potentiation (LTP) induction, although this connection has yet to be conclusively established. IPLs possess the unique property of generating units of first-person internal sensations, or semblances, as previously described. Verifying semblance formation at inter-LINKed spines could bridge critical gaps in our understanding of the relationship between the ability to learn and the ability to induce LTP, potentially resolving longstanding debates in the field. Demonstrating a correlation between the ability to exhibit surrogate behavioral motor activity indicative of memory retrieval (a proxy for learning) and the ability to induce LTP would represent a major advancement in neuroscience.

Previous experiments have demonstrated that spatial learning becomes impaired following the saturation of long-term potentiation (LTP) (Moser et al., 1998). A later study revealed a specific interrelationship between LTP and surrogate markers of memory retrieval (Whitlock et al., 2006). Here, using one-trial inhibitory avoidance learning in rats, it was shown that learning-induced synaptic potentiation occludes high frequency stimulation (HFS)-induced LTP. Based on these findings, the semblance hypothesis provides a plausible explanation for the relationship between LTP and memory, as outlined below.

a. Learning first, followed by LTP induction:

According to the semblance hypothesis, prior learning events in a controlled environment would have already led to the formation of many islets of linked postsynaptic terminals (dendritic spines) in the hippocampi of the rats. Since opportunities for associative learning are limited in a caged setting, we can expect a gradual expansion of these discrete islets of linked postsynaptic terminals as the rats mature, with additional learning events further linking postsynaptic terminals. When the rats undergo avoidance learning—a novel instance of associative learning—we can anticipate the formation of functional links between two or more pre-existing islets of linked postsynaptic terminals. While this is particularly relevant in the experimental context, it is also applicable to any novel associative learning experience.

In experiments using inhibitory avoidance testing (Whitlock et al., 2006), not all recording electrodes detected an increase in the slope of the field excitatory postsynaptic potential (fEPSP). This suggests that the ionic changes necessary to produce such an increase did not occur at the electrode tips, which were located in the CA1 dendritic region. However, for those electrodes that did record an increase in fEPSP slope following inhibitory avoidance learning, a sufficient number of Schaffer collateral–CA1 synapses were potentiated. Let us denote two pre-existing islets of functionally LINKed postsynaptic terminals (IILPs) as 1 and 2. During the learning process, it is likely that new LINKs are formed between these two IILPs. This would result in a rapid expansion of the network, effectively doubling the size of the islet and forming a mega-IILPs (see Fig. 14).

.



Figure 14. This illustration depicts the proposed mechanism of long-term potentiation (LTP) based on the current hypothesis. It shows two islets of inter-LINKed spines (IILPs), represented in two different colors, each belonging to a different CA1 neuron (see Fig. 5 for an illustration of IILPs). One spine from each IILP is positioned abutted to the other, representing a potential site for LINK formation between the two IILPs. The stimulating electrode is placed at the Schaffer collaterals, which originate from CA3 neurons and form synapses on the dendritic spines of CA1 neurons. During associative learning or LTP induction, the formation of a LINK between the two abutted spines (indicated by asterisks) from IILPs 1 and 2 can result in the creation of a mega-IILP. This larger islet can contribute to the enhancement of synaptic responses recorded by the recording electrode, thereby reflecting LTP.


Now, let us consider a scenario in which a new associative learning event causes depolarization of the abutted spines belonging to IILPs 1 and 2. This depolarization can lead to the formation of an inter-postsynaptic functional LINK (IPL) between the spines, effectively connecting the two IILPs and resulting in the formation of a mega-IILP. This structural reorganization can influence both memory retrieval and the induction of LTP.


First, any cue stimulus arriving at one of the inter-LINKed spines within the mega-IILP can evoke units of internal sensations (semblances) across all of the inter-LINKed spines in the network. This can enhance the net semblance contributing to memory, enabling the retrieval of multiple previously associated memories that were involved in the formation of the original IILPs. Second, extracellular recordings from the apical dendrites of pyramidal neurons in the stratum radiatum of the CA1 region, in response to Schaffer collateral stimulation, will show an increase in the amplitude of the field excitatory postsynaptic potential (fEPSP). A sustained increase in the fEPSP slope is referred to as LTP. Notably, this learning-induced LTP can occlude the induction of additional LTP, highlighting a potential upper limit of potentiation.


b. LTP induction first, followed by learning:

The occlusion process described by Whitlock et al. (2006) can be interpreted as bidirectional: the induction of LTP in a sufficient number of synapses involved in inhibitory avoidance learning can, in turn, prevent the acquisition of subsequent avoidance learning. During high-frequency stimulation used to induce LTP, it is likely that hundreds of CA3 axons in the Schaffer collateral pathway are activated, thereby stimulating numerous dendritic spines on a CA1 neuron. In this process, many of these spines may become functionally inter-LINKed. Some of these newly formed LINKs may occur between pre-existing IILPs, resulting in the formation of mega-IILPs. Once mega-IILPs are formed, activation of a single spine by a regular stimulus can lead to the spread of depolarization across all the inter-LINKed spines within the mega-IILP. This convergence of excitatory postsynaptic potentials (EPSPs) onto a single CA1 neuron can amplify the overall postsynaptic response, producing a larger EPSP. This enhanced response underlies the long-term potentiation (LTP) observed.


When LTP is experimentally induced, a large number of abutted spines become inter-LINKed in the synaptic region. This includes spines that are crucial for encoding future learning events. As a result, it may become difficult—or even impossible—to teach the animal new information following LTP induction. This difficulty manifests as an inability to retrieve specific memories. In such cases, when a cue stimulus attempts to trigger memory retrieval by reactivating specific IPLs and eliciting targeted semblances on particular inter-LINKed spines, the induced depolarization instead spreads indiscriminately across all the non-specifically inter-LINKed spines formed during prior LTP induction. Consequently, the specificity of the semblance signal associated with the originally learned item becomes diluted by a multitude of irrelevant semblances, thereby impairing accurate memory retrieval.


Figure 15 illustrates the similarities between the cellular mechanisms underlying LTP following its induction and those involved in the internal sensation of memory retrieval after associative learning.

.



Figure 15. Structural mechanism of internal sensation formation in memory and Its relationship to a potential mechanism of LTP. A) Memory Retrieval and Semblance Generation: During memory retrieval, a cue stimulus activates presynaptic terminal A, which depolarizes its corresponding postsynaptic spine B. This reactivates an inter-postsynaptic functional LINK (IPL), leading to the depolarization of an inter-LINKed spine D. The result is a cellular hallucination—a virtual internal sensation—as though sensory input has arrived at spine D’s presynaptic terminal C. Under normal physiological conditions, terminal C receives action potentials via activation of a CA3 neuron. The sensory identity of the semblance experienced at spine D corresponds to the activation of a set of neurons {Y}, which synapse onto the CA3 neuron. These neurons {Y} are, in turn, activated by inputs from a lower-order neuronal set {X}, which itself is driven by a broader set of upstream neurons {W}. Tracing this hierarchical structure toward the sensory periphery leads to a set of sensory receptors {SR}. Subsets of these receptors—{sr1}, {sr2}, and {sr3}—can independently activate the CA3 neuron. Hypothetical sensory input patterns that stimulate receptor subsets {sr1}, {sr2}, and {sr3} are referred to as semblions 1, 2, and 3, respectively. Thus, activation of spine D by the cue stimulus can generate a virtual internal sensation corresponding to one or more of these semblions, or their combined representation. A CA1 neuron—considered a place cell in the context of spatial memory—is depicted as receiving subthreshold excitatory postsynaptic potentials (EPSPs) from oscillatory activity among its lower-order neurons. When the cue-induced activation of spine D contributes an additional EPSP, it may push the summated signal past the action potential threshold, causing the CA1 neuron to fire during memory retrieval. This neuron would not otherwise fire in the absence of prior associative learning. This mechanism provides a plausible explanation for the activation of place cells during the recall of spatial memories. Bottom panel: Cross-sectional view of spines B and D, which have formed an IPL during associative learning. Three additional islets of inter-LINKed spines (IILPs) are also illustrated. B) LTP Induction and Mega-IILP Formation: High-frequency stimulation of the Schaffer collaterals induces LTP by promoting postsynaptic membrane hemi-fusion between spines belonging to different IILPs. Specifically, hemi-fusion occurs between spines in the IILPs B–D and F–H–J–L, resulting in the formation of a mega-IILP (B–D–F–H–J–L). As a result, a standard stimulus delivered via the stimulating electrode now has an increased likelihood of reaching the recording electrode due to enhanced electrical conductivity through the hemi-fused spines. This increased efficacy is recorded as a potentiated response from the CA1 neuron. Bottom panel: Cross-sectional view of the area showing the newly formed mega-IILP consisting of spines B–D–F–H–J–L. Two additional IILPs are also shown. Neuronal orders are labeled from 1 to 6, beginning with the sensory receptors.If the LTP-induced mega-IILP includes spines B and D, which are critical for prior associative learning, the specificity of memory retrieval may be compromised. This occurs because stimulus-induced activity spreads nonspecifically across the newly hemi-fused spines of the IILPs, generating irrelevant semblances that interfere with the retrieval of precise, learned associations. Abbreviations: {SR}, set of sensory receptors; {sr}, subset of sensory receptors; IPL, inter-postsynaptic functional LINK; IILP, islet of inter-LINKed spines (Adapted from Vadakkan, 2013).


The hypothesis relies on the key assumption that internal sensation is induced at a specific location by a specific mechanism. Why should this assumption be correct?


To construct a viable hypothesis, it is necessary to begin with a foundational assumption. If a single assumption consistently accounts for all known functions of the nervous system, the likelihood of its correctness is high. This approach is analogous to solving a system of linear equations that has a unique solution: when only one variable remains unknown, its value can be mathematically determined by analyzing its relationships with the other variables. Alternatively, one can assign various trial values to the unknown and use a process of elimination to arrive at the correct solution. In the case of the nervous system, where the internal sensation remains the sole unknown variable, a large number of known variables and their interrelationships can be leveraged through iterative trial-and-error methods to deduce the unknown. In the formulation of the semblance hypothesis, it was assumed—based on compelling evidence—that the induction of semblance as a system property occurs at inter-LINKed postsynaptic terminals (dendritic spines) via reactivation of inter-postsynaptic functional LINKs. This assumption is supported by several key observations:


1) Continuous depolarization consistently occurs at the spine head region. 2) A large number of spines from different neurons are abutted closely, with minimal extracellular matrix (ECM) separating them, allowing for potential learning-induced modifications through the opening of connections at these sites. 3) Miniature EPSP generation cannot be completely inhibited by any known natural or synthetic chemical. 4) Inter-postsynaptic LINKs can form as a result of the simultaneous activation of adjacent postsynaptic terminals during associative learning. 5) The induction of semblance can then emerge as a function of lateral activation across these LINKS, effectively ‘tricking’ the system into generating a hallucination-like internal experience. 6) The presence of various types of inter-postsynaptic functional LINKs, each with different lifespans, supports a mechanism for encoding memories with varying durations. 7) These LINKS can be stabilized, enabling the retention and retrieval of memories over diverse time scales. 8) Lateral activity spread through the inter-postsynaptic LINKs contributes to the horizontal component of oscillatory potentials, which must be maintained within a narrow frequency range for cognitive processing. 9) Semblance, as a virtual property, is well-suited to explain the subjective nature of internal sensations underlying higher brain functions. 10) Semblance is a first-person experience accessible only to the owner of the nervous system, aligning with the inherently private nature of consciousness.


Collectively, these supporting factors make the assumption of semblance induction a strong candidate for explaining internal sensation. The range of structural and temporal changes that can occur during inter-postsynaptic LINK formation (as illustrated in Fig. 12), and their potential to persist across various time scales, reflects exactly the kinds of features expected from a fundamental operational mechanism of the nervous system. Moreover, the inter-postsynaptic functional LINK model aligns with all constraints outlined in Table 2 on the front page of this website. Given these considerations, the proposed mechanism is likely to be valid.


This work has explained the induction of units of internal sensation in memory. How can it account for perception and consciousness?

Learning and memory were studied primarily because they offer a practical entry point for understanding brain function. Learning induces observable changes, and these changes are presumed to play a role in memory formation. Since such changes can be hypothesized, they also provide an opportunity for empirical testing, enabling hypothesis validation. This approach led to the formulation of a mechanism involving the formation of inter-postsynaptic functional LINKs and the induction of discrete units of internal sensation at inter-LINKed dendritic spines—proposed as the fundamental operations underlying memory.


It is reasonable to assume that the mechanism responsible for generating these internal sensations shares core features with those underlying perception and consciousness. Slight modifications to the induction process likely give rise to the internal sensations experienced during both perception and conscious awareness. For perception, the mechanism must account for real-time induction of internal sensations in a manner that aligns with the wide range of known perceptual phenomena. This framework was applied specifically to explain visual perception (Vadakkan, 2015c).


In the case of consciousness, the model must address three key questions a) Why are numerous units of internal sensations induced even when the organism is at rest? b) Is there any benefit in integrating these inner sensations into a single non-specific inner sensation? c) What is the net semblance generated by these internally induced sensations? d) How does this semblance form a background matrix that facilitates the efficient induction of stimulus-specific internal sensations in response to external cues? (Vadakkan, 2010).


What is the underlying logic behind this work?

Imagine a solvable system of linear equations: a set of interrelated equations containing multiple variables. If all but one variable are known, we can determine the unknown variable using straightforward mathematical methods. However, by carefully examining how this unknown variable relates to the known ones, we can also deduce its value using logical inference or a trial-and-error approach, guided by the constraints imposed by the system. Similarly, the nervous system presents a vast set of known variables—observed across molecular, cellular, intercellular, electrophysiological, systems-level, behavioral, and imaging studies. These findings have already been extensively correlated. Among them, one critical variable remains unknown: the mechanism by which internal sensations—such as the subjective feeling of memory—are generated. Despite its central importance, this variable is often ignored because its mechanism is not yet understood.


Let us now include internal sensations in our framework. For instance, instead of focusing solely on observable behavior during memory retrieval, we acknowledge the accompanying internal sensation of remembering. With this, internal sensation becomes the only unknown variable within a system defined by numerous well-established findings. Using a trial-and-error approach, we can now apply the constraints provided by existing data (such as those listed in Table 2 on the homepage) to deduce the underlying mechanism responsible for internal sensations. This is precisely the approach taken by the semblance hypothesis, which systematically incorporates internal sensation into the analysis and refines its model by integrating observations across different levels of the nervous system.


Is there convincing evidence to support the validity of this hypothesis?

To study a system that generates first-person internal sensations—experiences inherently inaccessible to third-person observation—it is necessary to apply methods from mathematics and physics, which are designed to analyze phenomena beyond the reach of direct sensory perception. In this context, the present work derives a solution based on the underlying principles used to solve systems of linear equations, a strategy that also aligns with the broader scientific goal of unification. This solution has been employed to triangulate findings across multiple levels of analysis (Munafò and Smith, 2018), thereby reinforcing its validity. The resulting framework now offers a foundational explanation of the first principles governing the system’s operations.

The central hypothesis of this work regards memories in their true nature—as first-person internal sensations. The derived solution reveals a set of background properties necessary for the induction of discrete units of virtual internal sensation. Continuous depolarization of dendritic spine heads—mediated by both quantal neurotransmitter release and excitatory postsynaptic potentials (EPSPs) triggered by the intermittent arrival of action potentials at presynaptic terminals in response to environmental stimuli—establishes a baseline or background state. During associative learning, it is proposed that a functional link, termed an inter-postsynaptic (or inter-spine) LINK (IPL), forms between spines belonging to different neurons. In the context of this background state, if one of the associated stimuli (the cue stimulus) reactivates the IPL, it is expected to induce a localized, internally generated activation at the linked spine—effectively a "cellular hallucination" or semblance—that mimics the reception of the second stimulus. This mechanism aligns with Marvin Minsky’s (1980) conceptualization of memory as a cellular-level hallucination. Such a process could offer a plausible explanation for the emergence of virtual, first-person internal sensations of memory within physiological time scales.


Furthermore, by tracing the pathway from an inter-LINKed spine back to the associated sensory receptors, it becomes possible to determine the minimal sensory input required to trigger that internal sensation—thus defining units of internal experience. The convergence of several key features supports the plausibility of this mechanism: (a) the presence of a distinct background state, (b) the potential for IPL formation with varying lifespans, (c) the strategic positioning of IPLs enabling cue stimuli to reactivate them, (d) the physiological feasibility of inducing virtual internal sensations, and (e) the orthogonal propagation of potentials through synapses and IPLs, contributing vector components to oscillating extracellular fields whose frequency in a narrow range determines the system functions.


A closer examination of the proposed mechanism reveals that the system’s ability to generate cellular hallucinations—which underlie first-person internal sensations—requires that its dominant operational state be defined by the depolarization of inter-LINKed spine heads through activity arriving from their respective presynaptic terminals. In this context, lateral activation of an inter-LINKed spine by a cue stimulus—via the IPL, rather than the presynaptic terminal—can only transiently induce a cellular hallucination (i.e., the experience of memory). In other words, as long as presynaptic depolarization remains the dominant form of activation, the system can be effectively "tricked" into generating a cellular hallucination when the cue stimulus causes lateral depolarization through the IPL.


The higher the ratio of spine depolarizations driven by presynaptic neurotransmitter release to those driven by lateral activation through IPLs, the more reliably the system can interpret lateral activations as genuine memory experiences. However, in order to maintain presynaptic neurotransmitter release as the dominant mode of activation, the system must tightly regulate the duration of lateral activations through IPLs to ensure functional stability. To preserve this balance, the system periodically enters a sleep state, during which lateral activations triggered by cue stimuli are suppressed. This reset mechanism restores the baseline conditions necessary for the effective induction of cellular hallucinations that constitute first-person internal sensations. This understanding provides a functional explanation for the necessity of sleep: without it, the system's ability to generate coherent internal sensations, such as memories, would progressively degrade. The critical role of sleep in resetting the system to its optimal state serves as further compelling evidence in support of the proposed framework.


The mechanism derived above aligns with the constraints imposed by a wide array of observations across multiple levels of analysis (see Table 2 on the homepage). Several lines of compelling evidence have emerged through continued examination of this hypothesis:


a) Most learning-induced changes are transient and reverse rapidly as an animal interacts with its environment, which corresponds with the concept of working memory. Only a small subset of these changes persists for longer durations, supporting short- and long-term memory formation. Therefore, any accurate mechanistic explanation of memory must account for both rapid reversibility and selective persistence. Examination of Figure 12 reveals that the formation of inter-postsynaptic functional LINKs through the exclusion of water of hydration (Fig.12B) demands significant energy and reverses quickly—consistent with the fleeting nature of working memory. A smaller fraction of these LINKs proceeds to form partial or complete hemifusions (Figs.12C,D), which can persist over varying durations. This pattern mirrors the expected behavior of learning-induced changes, indicating a strong alignment between the proposed mechanism and empirical observations. b) Building on this, a viable mechanism must also explain how certain learning-induced changes are stabilized for long-term retention. The complete hemifusion stage (Fig.12D) can be maintained through multiple stabilization mechanisms. Furthermore, when newly formed inter-LINKed spines become part of a larger islet of inter-LINKed spines, their increased likelihood of activation promotes long-term maintenance, offering a plausible substrate for memory consolidation. c) Dopamine, known to cause spine enlargement (Fig.12E), facilitates the formation and stabilization of inter-postsynaptic functional LINKs. This dopaminergic effect supports long-term memory retention, representing yet another point of convergence between empirical findings and the proposed mechanism. d) Numerous studies have reported a correlation between the capacity to learn and the induction of long-term potentiation (LTP). The proposed mechanism accounts for these correlations and also offers explanations for previously uncorrelated observations in the literature (Vadakkan, 2019), further reinforcing its validity. e) The mechanism provides a unified framework for understanding perception, including detailed features of visual perception, and identifies analogous olfactory circuitry in Drosophila, a distantly related species. This cross-species applicability adds another layer of support. f) The framework also extends to internal sensations of consciousness, offering an explanation for how anesthetic agents disrupt consciousness by interfering with the underlying mechanism—thus aligning with empirical observations of anesthesia’s effects. g) Lastly, the mechanism accounts for various common features of neurodegenerative disorders by interpreting them as disruptions or losses in the normal functional operations of the system.


The ability of a single mechanism to account for a wide range of seemingly disparate findings strongly substantiates the validity of the hypothesis.


Could you provide a commentary on this work within the context of connectomic studies?

The brain’s functions include receiving sensory information, consciously interpreting some of it, generating motor responses based on previous associative learning, and storing newly acquired information. What type of connectome could integrate all of these functions? Why have we not yet fully understood the nervous system, and what are we missing? A key aspect of brain function is the generation of first-person inner sensations—such as perception, memory, and consciousness—as properties of the mind. Investigating this requires a novel approach, distinct from current anatomical, molecular biological, and electrophysiological methods, while still adhering to their foundational principles. To determine the necessary approaches, a hypothesis must be developed that can explain both first-person internal sensations and third-person observations across various levels of analysis. Since many higher brain functions are grounded in first-person internal sensations, the primary focus of our search for neural circuits should be to explain these sensations. In other words, the working hypothesis should provide a bridge between cellular and electrophysiological properties and the resulting virtual internal sensations.


The semblance hypothesis was proposed to address these goals. The hypothesis introduces a new connection—inter-postsynaptic functional LINKS (IPLs)—to the synaptically connected nervous system. This novel addition to the functional connectome provides a fresh brain circuitry model (Fig. 16). The connectome presented in this work offers an explanation for the wide array of functions observed across the nervous system by various branches of brain science (see Table 2 on the homepage). The core cellular principle underlying the first-person internal sensations of higher brain functions likely shares a common cellular mechanism. The concept of inter-postsynaptic functional LINKS and the induction of semblance provides a mechanistic foundation for these processes. At this stage it is necessary to a) verify its operation in the nervous system, and b) conduct the gold-standard test by replicating the mechanism in engineered systems, which would allow us to convert internal sensations into measurable outputs for third-person understanding.

.



Figure 16. Comparison between synaptically-connected circuitry (Left) and inter-postsynaptic functional LINK (IPL)-mediated wiring (Right). Left Panel: Conventional synaptically-connected neuronal circuit diagram. In this setup, there is a single synaptic connection between neurons N1 and N2. When neuron N1 is activated, it generates an excitatory postsynaptic potential (EPSP) at postsynaptic membrane B. If neuron N2 is simultaneously receiving EPSPs from other neurons, and the sum of these EPSPs is one short of the threshold required for spatial summation to trigger an action potential, the EPSP from neuron N1 will contribute to the firing of neuron N2. The role of the EPSP from neuron N1 in temporal summation—where the timing of EPSPs also matters for triggering an action potential—should also be considered. In isolation, a single EPSP or a small train of EPSPs reaching postsynapse B may not be sufficient to induce neuron N2's action potential. Right Panel: Wiring diagram based on the semblance hypothesis. In this framework, activation of neuron N1 triggers the inter-postsynaptic functional LINKS (IPLs) within the islet of inter-LINKed spines (IILPs). The reactivation of postsynapse B, which is part of neuron N2, generates an EPSP that enables neuron N2 to fire an action potential, much like the threshold conditions in the conventional wiring diagram. Additionally, the EPSPs propagate to other hemi-fused postsynapses (D, F, H, J, L) depending on the extent of the spread through the IILPs, potentially reaching their respective neuronal somata. According to supplementary rules, a total of six postsynapses are reactivated here, compared to just one in conventional synaptic transmission (as shown in the left panel). This increased reactivation boosts the likelihood of sub-threshold neurons firing, as it brings them closer to the threshold for activation. For example, neuron N6 continuously receives (n−1) EPSPs, just shy of the necessary EPSP count to induce spatial or temporal summation and trigger an action potential. The arrival of the nth EPSP via the IILPs enables neuron N6 to reach the threshold and fire an action potential (highlighted in red). If neuron N6 is a motor neuron, it can evoke motor activity in sync with the reactivation of inter-LINKed spines B, D, F, H, J, and L. These inter-LINKed spines also facilitate the formation of internal sensations through the induction of semblions, provided they are located in areas of oscillatory neuronal activity. All neurons marked in red receive sufficient EPSPs to fire action potentials. Importantly, the lateral spread of activity through the IPLs contributes to the horizontal vector of oscillatory neuronal activity observed in regions like the cortex and hippocampus, indicated by the red wave passing through the IILPs (Modified from Vadakkan KI, 2013).


What are the implications of this work?

The nervous system comprises two interdependent circuitries: (1) a functional one involving the formation, reactivation, and reversal of inter-postsynaptic LINKs (IPLs), and (2) a structural one operating through synapses. Because neurons can fire with only partial input (see section "What are the issues with studying neuronal firing (axonal spike) in understanding higher brain functions?"), many are maintained in a sub-threshold state by inhibitory neurons (Hangya et al., 2014; Karnani et al., 2014). Small inputs via IPLs can push some of these neurons to fire, enabling downstream activation. According to the IPL mechanism, internal sensations underlying higher brain functions arise at the inter-spine level between neurons. The integration of IPL-induced units of internal sensations at specific inter-LINKed spines is crucial for generating these functions—including motivation, memory, perception, consciousness, aversion, reward, pleasure, anxiety, stress, fear, intentionality, hunger, thirst, and pain. While modulating neuronal firing below the level of inter-LINKed spines can alter these sensations, understanding their origin requires testing the IPL-based hypothesis of induction and integration of internal sensation units.


Should we consider this hypothesis to be correct?


Science seeks truth through rigorous, multi-perspective examination. This work adopts the highest standard—replication of the mechanism in an engineered system—as its benchmark for validation, prompting comprehensive scrutiny and anticipation of challenges. Our limited understanding of the brain stems from two gaps: (a) the inability to integrate findings across scales, and (b) the lack of a mechanistic explanation for first-person internal sensations such as perception, memory, and consciousness. Addressing this requires identifying a core operational mechanism that explains internal sensations while integrating diverse neuroscientific data. This work compiles empirical constraints from multiple system levels (Table 2 on the homepage), which serve as nature-imposed guideposts narrowing the solution space. Any valid mechanism must satisfy all these constraints.


Accordingly, this study: (a) applies constraint-based reasoning, (b) derives the proposed mechanism from extensive empirical constraints, and (c) validates it against the same. Its strength lies in methodological rigor, predictive power, and consistency with principles of simplicity, universality, and evolutionary conservation. Ongoing scrutiny—via questioning, testing, and attempts at falsification—is essential to uphold scientific integrity. In summary, this work offers a strong mechanistic candidate for higher brain functions by fulfilling key scientific criteria: 1) explains first-person internal sensations; 2) unifies diverse findings; 3) yields testable predictions; 4) accounts for sleep; 5) supports engineered replication; and 6) remains falsifiable. Nonetheless, independent empirical validation—particularly confirming IPLs and semblance induction at inter-LINKed spines—is vital for strengthening confidence.


Can you provide an example of molecular evidence that supports the validity of this hypothesis?

As Richard Feynman once said, “When you have put a lot of ideas together to make an elaborate theory, you want to make sure, when explaining what it fits, that those things it fits are not just the things that gave you the idea for the theory; but that the finished theory makes something else come out right, in addition.” This principle is critical for validating a genuine discovery. If the semblance hypothesis is correct, it should lead to novel, unanticipated predictions that are later confirmed—several of which are already evident. One such example is at the molecular level. Figure 12 outlines the mechanism of IPL formation, which begins with the interaction of the outer membranes of dendritic spines from different neurons. These interactions can be transiently stabilized, suggesting the need for molecular mechanisms capable of operating on timescales consistent with known brain functions.


Membrane fusion is energetically demanding, requiring mechanisms to overcome high-energy barriers (Rand & Parsegian, 1984; Harrison, 2015). Intracellular proteins such as SNAREs (soluble NSF attachment protein receptors) facilitate this by providing energy to bring membranes into close proximity, neutralize repulsive forces, and overcome curvature-induced energy barriers, thereby initiating hemifusion (Kozlovsky et al., 2004; Martens & McMahon, 2008; Oelkers et al., 2016; Hernandez et al., 2012). SNAREs also generate force to tightly appose membranes (Jahn & Scheller, 2006), forming characteristic hemifusion intermediates (Lu et al., 2005; Liu et al., 2008). These properties suggest a plausible mechanism for transient IPL formation via hemifusion between adjacent spine heads—especially relevant given that inter-spine distances exceed average spine head diameters (Konur et al., 2003).


While SNAREs are well-characterized in presynaptic terminals, their presence in dendritic spines implies additional roles. Their known ability to mediate hemifusion supports the feasibility of inter-spine interactions between neurons. After rapid IPL formation during associative learning, reinforcement likely involves delayed processes. One such process is SNARE-mediated fusion of AMPAR-containing vesicles with the spine membrane (Lu et al., 2001; Kennedy et al., 2010), potentially contributing membrane material to lateral spine regions and stabilizing hemifusion. Supporting this, fear conditioning has been shown to drive AMPAR insertion into the postsynaptic membrane of lateral amygdala neurons within hours, with memory impaired when this insertion is blocked in as few as 10–20% of neurons (Rumpel et al., 2005). Furthermore, AMPA GluR1 subunits have been found up to 25 nm from the synaptic cleft (Jacob & Weinberg, 2014), suggesting lateral spine head regions as probable sites for both vesicle exocytosis and inter-spine fusion. These molecular observations, many of which emerged after the hypothesis was developed, offer independent support for the IPL mechanism—underscoring its predictive power and alignment with Feynman’s standard for scientific discovery.


What are the chances of developing a correct hypothesis?


A colleague once asked a question that others may also wonder: Given the odds, how did you arrive at a correct hypothesis? Statistically, the chances are indeed low. In truth, I treated hypothesis-building as a disciplined hobby. I immersed myself in learning how to construct and critically evaluate hypotheses that could plausibly explain nervous system function across levels. I focused on one idea at a time, allowing each to mature over months. Over time, their weaknesses—especially logical inconsistencies in explaining neural mechanisms—became clear. Each failed hypothesis taught me valuable lessons and improved my approach. With each iteration, the ideas held up longer under scrutiny. My fifth hypothesis, which lasted nearly a year, was based on charge transfer along DNA—an appealing concept, particularly since neurons don't divide. Yet it ultimately failed to meet critical criteria. The semblance hypothesis was my sixth attempt.


Was it possible to easily arrive at the details of the solution?

Although the core framework of the hypothesis and the idea of the inter-postsynaptic functional LINK (IPL) emerged in 2007, it took nearly four more years to uncover the mechanism responsible for triggering units of internal sensation at the IPL. Because the nervous system operates within a narrow band of oscillating extracellular potentials—crucial for learning and memory—the mechanism had to align with this constraint. The combination of unidirectional synaptic transmission and perpendicular IPL-based depolarization suggested the generation of vector components contributing to these oscillations, pointing to the IPL mechanism as a promising candidate. Motivated by this, I focused on how a cue stimulus could spark an internal sensation at an interlinked spine. I sensed I was missing something fundamental—“missing the trees for the forest”—and revisited the problem repeatedly. When I published a paper on consciousness in 2010, I still hadn’t identified the IPL mechanism, but I was convinced that a simple, evolutionarily plausible and elusive process must exist—one that required cross-disciplinary reasoning to uncover.


I persisted, aiming to find a mechanism that could operate on the millisecond timescale. I had to get the background dominant state just right for the lateral activation of an interlinked spine to “trick it into hallucinating.”  This was the hardest, yet most exhilarating concept to grasp. By late 2010, I finally assembled the operational mechanism and published it in early 2011 in Processing semblances.... The discovery took four years of persistence—but every moment was worth it.

What if this hypothesis is incorrect?

Uncertainty is natural after proposing a new hypothesis. But if the hypothesis is correct, that uncertainty tends to fade over time—because while there are countless ways a hypothesis can fail, there’s typically only one path for it to be right. The core challenge is to identify that path: a mechanism that coherently accounts for all observed functions across multiple levels of the nervous system. Two main types of error can derail this process. First, consider the nervous system as a complex, multi-dimensional jigsaw puzzle. The essential first step is to gather all relevant pieces from various levels. Missing any can lead to a distorted hypothesis. Second, even with all pieces present, if they’re indistinguishable—like puzzle pieces of the same color—we’re forced to rely only on shape. We might assemble 99% of the puzzle, only to discover a mismatch that reveals a foundational error. Many of us have experienced this: sometimes, the only solution is to dismantle everything and start over.


So, is this hypothesis ready for that level of scrutiny? Has it assembled all the necessary pieces? This framework was built by integrating non-redundant findings across multiple levels, guided by strict constraints (see Table 2 on the front page). While the likelihood of a fundamental flaw is low, it isn’t zero. That’s why this work includes an open call for falsification. If any observation emerges that the hypothesis cannot explain, it must be discarded—and replaced by one that can. Should someone reveal a fatal flaw in this framework, I will start a new page to document it openly for all to examine.


What is the motivation behind this work?


My deep interest in understanding the nervous system led me to become fully immersed in its challenges.


1) The nervous system produces first-person internal sensations—virtual experiences that drive third-person observable outputs like speech and behavior. Studying only these outputs is insufficient; we must also understand how these internal sensations arise and how they connect to motor actions.


2) A solution likely exists; our difficulty lies in perceiving it. While we're trained to update beliefs through prior knowledge (as in Bayes' rule), solving foundational problems may require discarding priors and adopting a fresh, outsider's perspective. This path is risky: unconventional work struggles to secure funding, is often undervalued without it, and faces hurdles at publication. Still, if empirical constraints lead to a testable hypothesis, it deserves serious consideration and experimental validation.


3) Suppose we arrive at a solution, X. If X can explain all observed features across levels of the system, it must be treated as a candidate mechanism.


4) Because internal sensations are virtual, the solution must stem from a first principle capable of generating testable predictions.


5) History shows that nature's mechanisms are often simple. It’s likely that the answer lies in something subtle we've repeatedly overlooked.


Solving this system demands constraint-based reasoning across multiple levels—an inherently slow and iterative process, especially in early stages. Without sustained, unrestricted funding, such progress would be nearly impossible. Yet the urgency is clear: without understanding how the healthy brain functions, we cannot effectively prevent or treat many neurological and psychiatric disorders. This urgent need is what ultimately motivated me to pursue this work.


What is the main conclusion of this work?

The Semblance Hypothesis presents a testable theoretical discovery: a primary neuronal circuitry embedded within synaptically connected networks, which has until now evaded our attention. This circuitry, mediated by the inter-postsynaptic functional LINK (IPL), is outlined in the present work. Additionally, this study proposes a mechanism for the induction of internal sensation units when IPLs are reactivated. According to the findings, reactivation of IPLs generates vector components of oscillating extracellular potentials, with their frequency closely linked to the system's efficiency in inducing internal sensations. Further investigation is required to verify the presence of a spectrum of changes responsible for IPLs and the induction of internal sensations. 

What steps do we need to take in our experiments to verify the findings of this work?

Current experiments focus on examining spines located on a dendritic branch of a neuron to understand higher brain functions, often correlating behavioral markers indicative of internal sensation generation associated with these functions. In these studies, we observe spine enlargement, reductions in spine size, and even spine elimination. However, these studies often overlook the immediate neighborhood of the spines on the dendritic branch—specifically, whether interactions may occur with intervening spines from different neurons. Given that the average inter-spine distance on a dendritic branch is greater than the average spine diameter (Konur et al., 2003), spines on a neuron are likely to be in close proximity to spines from other neurons. Additionally, electron microscopic images from cortical regions (such as Figures 4, & 13 on this page) reveal that the extracellular matrix (ECM) space between adjacent spines is minimal. Investigating the interaction between two adjacent spines from different neurons during higher brain functions, such as learning, at physiological timescales of milliseconds, could significantly enhance our understanding of inter-postsynaptic functional LINKs (IPLs) and their roles, as described in this work. Only changes occurring at the millisecond timescales of learning or memory retrieval are relevant. 

Why is first-person neuroscience necessary?

The brain generates first-person inner sensations, a phenomenon we have yet to study effectively. This presents a unique challenge, one that drives many of us to pursue neuroscience: the desire to confront difficult problems, fail, and ultimately succeed. To address this challenge, we must approach it with rigor and face all the complexities it presents. Recent trends, such as pharmaceutical companies moving away from drug development for neurological and psychiatric disorders (Wegener & Rujescu, 2013; Burke, 2014; Mehta et al., 2017), highlight the urgency of finding the right approach. The failure of drug trials has caused a loss of confidence and stalled progress in developing effective treatments. This setback underscores the need for a new perspective.


We must ask: Why do these drug trials fail, and what must we do differently? A closer look reveals that current studies often rely on surrogate markers—such as speech output and behavior—to study higher brain functions. However, we are not yet examining the mechanisms responsible for first-person experiences like memory. To design effective treatments for conditions like memory loss or hallucinations, we need a scientific understanding of how these internal sensations are generated. The key to addressing these issues lies in understanding the brain’s normal operational mechanisms. To do so, we must develop a specialized field of first-person neuroscience, which will differ from current third-person approaches in fundamental ways, as outlined in Table 1.


Third person neuroscience
First-person neuroscience0

Studies conducted at various levels rely on third-person observations. Examples include:


Biochemical Findings: Gene expression and the action of protein molecules.


Cellular Changes: Outgrowth of neuronal processes, formation of new neurons, their connections, and neuronal firing.


Electrophysiological Changes: Alterations in AMPA and NMDA receptor currents, changes in postsynaptic potentials, and shifts in voltage-dependent calcium currents.


System Changes: Oscillating potentials recorded using surface or extracellular electrodes.


Imaging findings: Changes in signals in fMRI, changes in neuronal ensembles that fire during a higher brain function.


Behavioral Changes: Speech and motor actions that provide sensory input to third-person experimenters, offering insights into the formation of first-person internal sensations.

The first-person scientific approach focuses on studying the mechanism behind the induction of first-person internal sensations, which are currently regarded as emergent properties. The main challenge in this approach lies in the access problem. What is needed are new methods and tools to overcome the difficulties posed by this challenge.


Since third-person experimenters cannot access first-person properties, the methods to address this issue involve several critical steps: 1) Hypothesize a feasible mechanism that explains the key points at which internal sensations can emerge, under specific conditions. This mechanism should incorporate all the elements necessary to satisfy the requirements of findings made across various levels by different fields of neuroscience. 2) The hypothesized mechanism must operate in conjunction with known circuit properties and be capable of explaining diverse nervous system functions. 3) Using the proposed mechanism, develop a circuit to conduct a gold-standard test by replicating the mechanism in engineered systems. At this stage, it is essential to identify the key points and conditions where units of internal sensation emerge as a system property. 3) Develop methods to capture these emergent properties and convert them into suitable readouts for third-person experimenters. This step is feasible, as we are designing the engineered system.

Table 1. Key differences between current third person approaches and potential first-person approaches for understanding the operation of the nervous system:

What steps are necessary to develop a first-person neuroscience?

Some steps can be taken immediately. Neuroscience experiments examining first-person functions, such as consciousness, perception, and memory, should explicitly state whether the study is focused on a surrogate marker for the first-person property or is directly investigating the first-person property itself. This distinction will help raise awareness about the need to conduct studies on first-person properties.


It is crucial to develop methods to study these first-person properties, which will require contributions from many researchers. We have demonstrated success in addressing abstract concepts in the past. For instance, we invented numbers, which do not exist in a physical sense, and later introduced negative numbers and plotted graphs within virtual spaces. This has allowed us to extend our thinking into realms that our sensory systems cannot directly perceive. Similarly, although our sensory systems cannot detect the Earth's rotation at nearly 1670 km/h, we first hypothesized this movement using theoretical methods (by applying appropriate reference frames) and later confirmed it by observing the Earth from space. It is reasonable to argue that we can develop a comparable approach to understand first-person properties. By establishing a solid foundational framework, we should become more adept at studying and engaging with first-person inner sensations.

How would it change the current approach to studying the nervous system?

It is important to recognize that the primary function of the nervous system is to generate inner sensations associated with various higher brain functions. The ability of nervous systems to spark inner sensations of memory—of sensory stimuli that have either arrived late or not at all, but were previously associated with the first or fastest cue stimulus when the item was close to the animal—provided a significant survival advantage to both predators and prey. From that point onward, muscle power (in both predators and prey) became secondary in importance. Variations in the mechanisms generating these inner sensations are likely to have driven the evolution of different species. Further refinement of these mechanisms eventually enabled reasoning and hypothesis-building abilities in animals. Therefore, the evolution of the nervous system has been largely shaped by the need to fine-tune inner sensations to enhance survival. Given this context, it is reasonable to view the circuits responsible for generating and integrating inner sensations as the primary circuitry of the nervous system. Confirming these findings on an emergent basis will help guide our efforts to understand the nervous system in the correct direction.

How does the nervous system differ from an electronic circuit board and its components?

A printed circuit board (PCB) is a non-conductive plate that electrically connects components via conductive tracks etched onto its surface. In the brain, a similar role is performed by the very thin, often negligible extracellular matrix (ECM) that exists between cellular processes (Fig. 17). The IPL mechanism suggests that during each new learning event, a new connection is formed between spines from different dendrites at the convergence points of two sensory stimuli. This is achieved by removing the insulating ECM between the spines over an area smaller than 10 nm². 




Figure 17. Difference between a printed circuit board (PCB) and the brain. In a PCB, electrical paths that connect electronic components are separated from each other by large area of non-conductive (insulating) material. But in the brain, neuronal processes are separated by very thin (& often negligible) insulating medium of extracellular matrix (ECM). Left side: A printed circuit board made of a non-conducting plate on which conductive tracks are etched to connect the circuit components. Note that the surface area of non-conducting plate that does not have the conductive tracks is roughly more than 80% of the surface area of this plate. Right side: An electron microscopic image from the brain cortex. Note that neuronal & glial cell processes occupy most of the surface area with only very negligible insulating ECM space in between them. Note that while acting as an insulting medium, ECM also has two additional functions. 1) Acts as a buffer zone that facilitates ion flux across membranes. 2) Brain functions occur only in a narrow range of frequency of oscillations of potentials within ECM that spans throughout the cortex. It is to be noted that the negligible ECM has to function very faithfully as an insulating medium without causing spread of depolarization to non-targeted neuronal processes. According to the IPL mechanism, the negligible ECM has an added advantage to etch IPLs between abutted spines. Note that even though it may seem easy for forming an IPL between abutted spines, very high energy is required to displace the hydration water between two lipid membranes (Cohen and Melikyan, 2004; Martens and McMahon, 2008). Furthermore, since the repulsive “hydration force” increases steeply when distance between the two bilayers reduces below 20 Å, fusion between two membranes becomes a very high energy requiring process (Rand and Parsegian, 1984; Harrison, 2015). However, it is anticipated that robust molecular mechanisms are present to overcome this energy barrier.


What does the presence of the thin ECM prompt us to consider?

The extracellular matrix (ECM) acts as a buffer zone for ions, facilitating ion flux across membranes during the propagation of depolarization. It also forms part of the oscillating extracellular potentials, which are maintained within a narrow frequency range essential for brain functions. The most notable feature of the ECM in the cortex is the extremely thin space it occupies (see Fig. 14). However, it serves a critical role as a robust insulating medium, due to the high energy required to establish electrical connectivity (Rand and Parsegian, 1984; Cohen and Melikyan, 2004; Martens and McMahon, 2008; Harrison, 2015) between neuronal processes by displacing fluid ECM between them. This provides a functional advantage. If learning induces lipid membrane changes on millisecond time-scales that overcome the energy barrier, establishing inter-neuronal electrical continuity could become a powerful learning mechanism. According to the current work, the IPL mechanism offers this advantage. Additionally, studies using artificial membranes (Leikin, 1987) suggest that the area of inter-spine hemifusion is likely limited to approximately 10 nm² or less. Astrocytic pedicels are present at only about 50% of synapses, and they occupy only 50% of the perisynaptic space (Ventura & Harris, 1999). Despite this limitation, the remaining surface area of the ECM where IPLs can be established is vast, which greatly benefits the system’s operation. The high energy requirement for establishing physical interactions between lipid membranes ensures that there will be no non-specific interactions leading to electrical continuity between neuronal processes. This also suggests that learning must trigger a biological mechanism to overcome the high energy requirement within milliseconds. Experimental evidence points to the role of SNARE proteins and complexin (see Vadakkan, 2019) in overcoming this energy barrier.


SNARE proteins are known to provide energy to bring membranes together, overcoming repulsive charges and addressing energy barriers related to curvature deformations during hemifusion between abutted membranes (Oelkers et al., 2016). They also generate the force needed to pull the membranes together as tightly as possible (Hernandez et al., 2012). By initiating the fusion process and supplying the necessary energy (Jahn and Scheller, 2006), SNARE proteins facilitate the formation of characteristic hemifusion intermediates (Lu et al., 2005; Giraudo et al., 2005; Liu et al., 2008). These properties underscore the significance of SNARE proteins in forming hemifusion intermediates between the lateral regions of spine heads. Additionally, the protein complexin, present within postsynaptic terminals (Ahmad et al., 2012), is known to interact with the neuronal SNARE core complex to halt fusion at the hemifusion stage (Schaub et al., 2006).


The next question is: "Between which two membranes do hemifusion occur in postsynaptic terminals (dendritic spines)?" In the presynaptic terminal, several synaptic vesicles are docked to the membrane facing the synapse, so one might assume they are hemifused with the presynaptic membrane. However, dendritic spines lack reports of vesicles being docked to their membranes. So, the question arises: "Where do SNARE and complexin act to generate hemifusion intermediates in the spines?" Detailed electron microscopic and nanometer-scale real-time studies are needed to test for inter-spine interactions, ranging from simple membrane contacts to partial or complete hemifusion between spines belonging to different dendrites. Based on the present work, it is expected that inter-spine interactions occur between abutted spines where sensory stimuli converge during learning (or when they are stimulated together) on millisecond time-scales. Since working memory typically lasts only a few seconds, most of these interactions are expected to reverse quickly. However, as animals have already learned many items and events in their environment associatively, it is reasonable to expect several stable hemifused regions between spines at locations of stimulus convergence, making them detectable. Dendritic excrescences on the dendrites of CA3 neurons provide clues to this phenomenon and can be examined to verify the presence of hemifused membranes between spines within them.


When an animal navigates its environment, it encounters a variety of new stimuli, resulting in the formation of numerous IPLs at the convergence points of these signals. Interestingly, most IPLs will reverse within a few seconds, which is expected since the majority of our day-to-day memories are working memories that last only a short time. Given that SNARE-mediated vesicle fusion at presynaptic terminals occurs within milliseconds, it is anticipated that SNARE-mediated IPL formation will follow a similar timescale.


What is the solution to the black box problem presented on the front page of this website (Figs. 18 & 19)?




Figure 19. Proposed solution to the Black Box problem (Refer to Figures 1 & 2 on the homepage). A) During the learning process, if stimuli St1 and St2 arrive at the dendritic spines of neurons GN and RN, respectively—and if these spines are closely abutted—they may undergo structural interactions that result in the formation of a new electrical connection (a novel channel) between them. In the cortical environment, neuronal processes, including dendritic spines, are densely packed with minimal extracellular matrix (see Figures 4 & 13 on this page), a condition that favors the formation of such inter-spine channels between neurons. B) Following learning, stimulus St1 alone can elicit an excitatory postsynaptic potential (EPSP) in neuron RN via the newly formed inter-neuronal inter-spine connection (represented by a black arrow between the spines of GN and RN), ultimately causing RN to fire. This mechanism is proposed to fulfill the criteria for a classical conditioning paradigm, as illustrated in Figure 1 on the homepage. In the vicinity of this new channel (black arrow), it is necessary to identify a unique biophysical property capable of generating first-person inner sensations. The semblance hypothesis addresses this need by identifying a specific property at this location that can give rise to discrete units of subjective experience, referred to as semblances.



Figure 20. This illustration reveals the internal structure of the black box depicted on the homepage of this website. The black box problem—specifically, how the conditioned stimulus (CS) comes to exhibit features of both the CS and the unconditioned stimulus (US) after learning—can be resolved only if a link is established between the dendritic spines of neurons N5 and N10. This link must be formed in such a way that, after learning, the arrival of stimulus 1 alone is sufficient to trigger the firing of both N5 and N10. At this stage, it is reasonable to propose that this link is also associated with a unique property capable of generating units of first-person inner sensations. Identifying and explaining this hidden mechanism logically is essential—and this is where the concept of semblance for stimulus 2 comes into play. The detailed mechanism is outlined in the sections above.


What is required to fully decipher the contents of the black box mentioned above?

Further resolution of the dendritic spine structure is necessary, and this should be achieved without compromising the integrity of the surrounding extracellular matrix. Preserving this volume is crucial for identifying the formation and reversal of the most common inter-postsynaptic functional links (IPLs), which are thought to underlie working memory (see Fig. 12B). To fully understand the derived first principles and the algorithm responsible for integrating units of internal sensations, collaboration with foundational scientific disciplines—including physics, mathematics, computer science, and electronic engineering—is essential.


How does the artificial intelligence community currently view first-person properties?


Members of the AI community have begun to question how neuroscientists approach the task of providing a mechanistic explanation for cognition. They are eager for neuroscience to uncover the specific mechanisms by which the brain generates its functions. For example, see the article "Neurotechnology is Critical for AI Alignment" (cvitkovic.net), which highlights this growing interest.


References


Abbas AK, Villers A, Ris L (2015) Temporal phases of long-term potentiation (LTP): myth or fact? Rev. Neurosci. 26(5):507-546. PubMed

Abbott LF (2008) Theoretical neuroscience rising. Neuron 60(3):489-495. PubMed

Ahmad M, Polepalli JS, Goswami D, Yang X, Kaeser-Woo YJ, Südhof TC, Malenka RC (2012) Postsynaptic complexin controls AMPA receptor exocytosis during LTP. Neuron 73, 260–267. PubMed

Akerman S, Goadsby PJ (2005) Topiramate inhibits cortical spreading depression in rat and cat: impact in migraine aura. Neuroreport. 16(12):1383–1387. 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

Ayata C, Lauritzen M (2015) Spreading Depression, Spreading Depolarizations, and the Cerebral Vasculature. Physiol Rev. 95(3):953–993. Article

Beauchamp MS, Sun P, Baum SH, Tolias AS, Yoshor D (2012) Electrocorticography links human temporoparietal junction to visual perception. Nature Neuroscience 15(7)957-959. PubMed

Becker WJ (2020) Botulinum Toxin in the Treatment of Headache. Toxins (Basel). 12(12):803. Article

Bloss EB, Cembrowski MS, Karsh B, Colonell J, Fetter RD, Spruston N (2018) Single excitatory axons form clustered synapses onto CA1 pyramidal cell dendrites. Nat Neurosci. doi: 10.1038/s41593-018-0084-6. 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

Burke M (2014) Why Alzheimer's drugs keep failing. Scientific American. Article

Calabresi et al., (2014) Direct and indirect pathways of basal ganglia: a critical reappraisal. Nature Neuroscience 17(8):1022-1030 Article

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. 325:169-182. PubMed

Cohen A, Burns B, Goadsby PJ (2009) High-flow oxygen for treatment of cluster headache: a randomized trial. JAMA. 302(22):2451–2457. Article

Cohen FS, Melikyan GB (2004) The energetics of membrane fusion from binding, through hemifusion, pore formation, and pore enlargement. J. Membr. Biol. 199: 1–14. PubMed 

Collard & Gellman (2001) Pathophysiology, clinical manifestations, and prevention of ischemia-reperfusion injury. Anesthesiology. 94: 1133 Article

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

Dahlin E, Neely AS, Larsson A, Bäckman L, Nyberg L (2008) Transfer of learning after updating training mediated by the striatum. Science. 320(5882):1510-1512. PubMed

Dreier JP, Reiffurth C (2015) The stroke-migraine depolarization continuum. Neuron. 86(4):902–922. Article

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

Escobar ML, Derrick B (2007) Long-term potentiation and depression as putative mechanisms for memory formation, in: F. Bermudez-Rattoni (Ed.), Neural Plasticity and Memory, from Genes to Brain Imaging, CRC Press, Boca Raton, Chapter 2. PubMed

Eyal G, Verhoog MB, Testa-Silva G, Deitcher Y, Benavides-Piccione R, DeFelipe J, de Kock CPJ, Mansvelder HD, Segev I (2018) Human cortical pyramidal neurons: From spines to spikes via models. Front. Cell Neurosci.12:181. 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. 312:145-163. PubMed

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

Giraudo CG, Hu C, You D, Slovic AM, Mosharov EV, Sulzer D, Melia TJ, Rothman JE (2005) SNAREs can promote complete fusion and hemifusion as alternative outcomes. J. Cell Biol. 170, 249–260. PubMed

Govindarajan A, Kelleher RJ, Tonegawa S (2006) A clustered plasticity model of long-term memory engrams. Nat Rev Neurosci. 7(7):575-583. 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

Harrison SC. (2015) Viral membrane fusion. Virology 479-480:498–507. PubMed

Hernandez JM, Stein A, Behrmann E, Riedel D, Cypionka A, Farsi Z, Walla PJ, Raunser S, Jahn R (2012) Membrane fusion intermediates via directional and full assembly of the SNARE complex. Science 336(6088):1581–1584. PubMed

Hodgson SE, Harding AM, Bourke EM, Taylor DM, Greene SL (2021) A prospective, randomized, double-blind trial of intravenous chlorpromazine versus intravenous prochlorperazine for the treatment of acute migraine in adults presenting to the emergency department. Headache. 61(4):603-611. Article

Jacob AL, Weinberg RJ (2014) The organization of AMPA receptor subunits at the postsynaptic membrane. Hippocampus. 25(7):798-812. PubMed

Jahn R, Scheller RH (2006) SNAREs--engines for membrane fusion. Nat Rev Mol Cell Biol 7(9):631–643. PubMed

Janeway Jr CA, Travers P, Walport M, Shlomchik MJ (2002) Immunobiology: the immune system in health and disease. 5th edition. The generation of diversit in immunoglobulins. New York: Garland Science. ISBN 0-8153-3642-X. PubMed

Josselyn SA, Tonegawa S (2020) Memory engrams: Recalling the past and imagining the future.  Science. 367(6473):eaaw4325. PubMed

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

Kennedy MJ, Davison IG, Robinson CG, Ehlers MD (2020) Syntaxin-4 defines a domain for activity-dependent exocytosis in dendritic spines. Cell 141:524–35. 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

Koyama T, Kato K, Tanaka YZ, Mikami A (2001) Anterior cingulate activity during pain-avoidance and reward tasks in monkeys. Neurosci Res. 39(4):421–430. Article

Kozlovsky Y, Efrat A, Siegel DP, Kozlov MM (2004) Stalk phase formation: effects of dehydration and saddle splay modulus. Biophys J 87(4):2508–21. PubMed

Kung TA, Guyuron B, Cederna PS (2011) Migraine surgery: a plastic surgery solution for refractory migraine headache. Plast Reconstr Surg. 127(1):181–189. Article

Leikin SL, Kozlov MM, Chernomordik LV, Markin V, Chizmadzhev YA (1987) Membrane fusion: overcoming of the hydration barrier and local restructuring. J. Theor. Biol. 129, 411–425. PubMed

Liu T, Wang T, Chapman ER, Weisshaar JC (2008) Productive hemifusion intermediates in fast vesicle fusion driven by neuronal SNAREs. Biophys J 94(4):1303–1314. PubMed

Lu W, Man H, Ju W, Trimble WS, MacDonald JF, Wang YT (2001) Activation of synaptic NMDA receptors induces membrane insertion of new AMPA receptors and LTP in cultured hippocampal neurons. Neuron 29(1):243–254. PubMed

Lu X, Zhang F, McNew JA, Shin YK (2005) Membrane fusion induced by neuronal SNAREs transits through hemifusion. J Biol Chem 280(34):30538–30541. 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

McMahon HT, Missler M, Li C, Südhof TC (1995) Complexins: cytosolic proteins that regulate SNAP receptor function. Cell 83, 111–119. PubMed

Meador et al., (2009) Cognitive function at 3 years of age after fetal exposure to antiepileptic drugs. N Engl J Med. 360(16):1597-605. Article

Mehta D, Jackson R, Paul G, Shi J, Sabbagh M (2017) Why do trials for Alzheimer's disease drugs keep failing? A discontinued drug perspective for 2010-2015. Expert Opinion on Investigational Drugs 26(6):735-739. Article

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

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

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

Moser EI, Krobert KA, Moser MB, Morris RG (1998) Impaired spatial learning after saturation of long-term potentiation. Science. 281(5385):2038-2042. PubMed

Munafò MR, Smith GD (2018) Repeating experiments is not enough. Nature 553:399. 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):1095-1103. PubMed.

Nagy AJ, Gandhi S, Bhola R, Goadsby PJ (2011) Intravenous dihydroergotamine for inpatient management of refractory primary headaches. Neurology. 77:1827–1832. Article

O'Donnell P, Grace AA (1993) Dopaminergic modulation of dye coupling between neurons in the core and shell regions of the NAc. J Neurosci 13(8): 3456-3471 Article

Oelkers M, Witt H, Halder P, Jahn R, Janshoff A (2016) SNARE-mediated membrane fusion trajectories derived from force-clamp experiments. Proc Natl Acad Sci U S A 113(46):13051–13056. PubMed

Onn SP, Grace AA (1994) Dye coupling between rat striatal neurons recorded in vivo: compartmental organization and modulation by dopamine. J Neurophysiol 71(5): 1917-1934. Article

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 29:779-796 PubMed

Piovesan EJ, Kowacs PA, Tatsui CE, Lange MC, Ribas LC, Werneck LC (2001) Referred pain after painful stimulation of the greater occipital nerve in humans: evidence of convergence of cervical afferences on trigeminal nuclei. Cephalalgia. 21(2):107–109. Article

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

Rand RP, Parsegian VA (1984) Physical force considerations in model and biological membranes. Can J Biochem Cell Biol 62(8):752–759. PubMed

Robbins MS, Kuruvilla D, Blumenfeld A, Charleston L 4th, Sorrell M, Robertson CE, Grosberg BM, Bender SD, Napchan U, Ashkenazi A (2014) Trigger point injections for headache disorders: expert consensus methodology and narrative review. Headache. 54(9):1441–59. Article

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

Rumpel S, LeDoux J, Zador A, Malinow R (2005) Postsynaptic receptor trafficking underlying a form of associative learning. Science 308(5718):83–88. PubMed

Saldanha IJ, Cao W, Bhuma MR, Konnyu KJ, Adam GP, Mehta S, Zullo AR, Chen KK, Roth JL, Balk EM (2021) Management of primary headaches during pregnancy, postpartum, and breastfeeding: A systematic review. Headache. 61(1):11–43. Article

Schaub JR, Lu X, Doneske B, Shin YK, McNew JA. (2006) Hemifusion arrest by complexin is relieved by Ca2+-synaptotagmin I. Nat. Struct. Mol. Biol. 13, 748–750. PubMed

Schneider T, Erben U, Moos V (2017) Peripheral T-Cell Reactivity to Heat Shock Protein 70 and Its Cofactor GrpE from Tropheryma whipplei Is Reduced in Patients with Classical Whipple's Disease. PubMed

Schoenbaum G, Chiba AA, Gallagher M (1998) Orbitofrontal cortex and basolateral amygdala encode expected outcomes during learning. Nat Neurosci.1(2):155-159. 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

Stuart GJ, Spruston N (2015) Dendritic integration: 60 years of progress. Nat Neurosci. 18(12):1713-1721. PubMed

Tonegawa S (1983) Somatic generation of antibody diversity. Nature 302: 575-581. Article; Review1, Review2

Tonegawa S, Pignatelli M, Roy DS, Ryan TJ (2015) Memory engram storage and retrieval. Curr Opin Neurobiol. 35:101-109. PubMed

Trotta L, Weigt K, Schinnerling K, Geelhaar-Karsch A, Oelkers G, Biagi F, Corazza GR, Allers K, Infect Immun. 85(8):e00363-17. PubMed

Tye KM, Stuber GD, de Ridder B, Bonci A, Janak PH (2008) Rapid strengthening of thalamo-amygdala synapses mediates cue-reward learning. Nature 453, 1253–1257 PubMed

Unekawa M, Tomita Y, Toriumi H, Suzuki N (2012) Suppressive effect of chronic peroral topiramate on potassium-induced cortical spreading depression in rats. Cephalalgia. 32(7):518–527. Article

Vadakkan K.I (2021a) Neurological disorders of COVID-19 can be explained in terms of both "loss and gain of function" states of a solution for the nervous system. Brain Circulation. Article

Vadakkan K.I (2021b) Framework for internal sensation of pleasure using constraints from disparate findings in nucleus accumbens. World Journal of Psychiatry. Article.

Vadakkan KI (2007) Semblance of activity at the shared post-synapses and extracellular matrices - A structure function hypothesis of memory. ISBN:978-0-5954-7002-0 Download

Vadakkan KI (2010) Framework of consciousness from semblance of activity at functionally LINKed postsynaptic membranes. Front Psychol. 1:168. Article

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 (2012c) A structure function mechanism for schizhophrenia. Front. Psychiatry. Article

Vadakkan KI (2013) A supplementary circuit rule-set for the neuronal wiring. Front Hum Neurosci. 7:170. Article

Vadakkan KI (2015a) A framework for the first-person internal sensation of visual perception in mammals and a comparable circuitry for olfactory perception in Drosophila. Springerplus. 4:833. Article

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 (2016) The functional role of all postsynaptic potentials examined from a first-person frame of reference. Reviews in the Neurosciences (Explained how a neuronal soma is flanked by a large number of internal sensory processing units and their relationship with neuronal firing). Article

Vadakkan KI (2016b) Neurodegenerative disorders share common features of "loss of function" states of a proposed mechanism of nervous system functions. Biomedicine and Pharmacotherapy. Article

Vadakkan KI (2016c) Rapid chain-generation of inter-postsynaptic functional LINKs can trigger seizure generation: Evidence for potential interconnections from pathology to behavior. Epilepsy & Behavior. 59: 28-41. Article

Vadakkan KI (2018) A learning mechanism completed in milliseconds and capable of transitioning to stabilizable forms can generate working, short and long-term memories - A verifiable mechanism. Peerj Preprints Article

Vadakkan KI (2019) From cells to sensations: A window to the physics of mind. Phys Life Rev. 31:44-78. PubMed

Vadakkan KI (2020) A derived mechanism of nervous system functions explains aging-related neurodegeneration as a gradual loss of an evolutionary adaptation. Curr Aging Sci. 13(2):136-152. Article

Vadakkan KI (2021c) Golgi staining of neurons: Oxidation-state dependent spread of chemical reaction identifies a testable property of the connectome. OSF Preprints

Vaz AP, Inati SK, Brunel N, Zaghloul KA (2019) Coupled ripple oscillations between the medial temporal lobe and neocortex retrieve human memory. Science 363:975-978. Article

Ventura R, Harris KM (1999) Three-dimensional relationships between hippocampal synapses and astrocytes. J. Neurosci. 19 (16):6897–906. PubMed

Wegener G, Rujescu D (2013) The current development of CNS drug research. International Journal of Neuropsychopharmacology. 16(7):1687-1693. PubMed

Whitlock JR, Heynen AJ, Shuler MG, Bear MF (2006) Learning induces long-term potentiation in the hippocampus. Science. 313(5790):1093-1097. PubMed

Yagishita S, Hayashi-Takagi A, Ellis-Davies GC, Urakubo H, Ishii S, Kasai H (2014) A critical time window for dopamine actions on the structural plasticity of dendritic spines. Science. 345:1616–1620. Article

Zimmerman CA, Leib DE, Knight ZA (2017) Neural circuits underlying thirst and fluid homeostasis. Nat Rev Neurosci. 18(8):459-469. Article