Findings after the hypothesis & their new explanations

It is necessary to examine findings from different laboratories to test whether their experimental results can be interpreted in terms of formation of inter-postsynaptic functional LINKs (IPLs) derived by the semblance hypothesis.

A. In physiological conditions/ artificial systems

Barcoding of episodic memories in the hippocampus of a food-campus of a food-caching bird. Chettih SN, Mackevicius EL, Hale S, Aronov D (2024). Cell. 187(8):1922–1935.e20. Article

Hippocampal neurons in chickadees show patterns of activity of CA1 neurons that are specific to the locations of hidden food resulting from retrieval of specific memories but are independent of place fields (set of CA1 neurons that fire when the bird reaches a place) for the same locations. In other words, one set of CA1 neurons fire during memory retrieval associated with reaching a place and another set of CA1 neurons fire when the bird reaches a particular location. Authors call these different sets of CA1 neurons that fire as barcodes.

It was previously shown that place cells (hippocampal CA1 neurons) (O’Keefe and Dostrovsky, 1971) are also associated with different higher brain functions such as trajectory of motion (Frank et a., 2000); location and memory in different contexts (Pastalkova et al., 2008), reward (Gauthier and Tank, 2018), auditory frequency in cognitive maps (Aronov et al., 2017), and abstract learned variables (Schuck and Niv, 2019; Park et al., 2020). In humans, visual images led to firing of sparsely located hippocampal neurons among a larger population of non-responsive neurons (Waydo et al., 2006). Firing of different sets of sparse hippocampal pyramidal neurons during different brain functions necessitates a mechanistic explanation. There is a combinatorial mechanism operating at the input levels of these CA1 neurons. To arrive at this mechanism, it is possible to view the following.

1. The different brain functions associated with separate operations operate at a lower level than the CA1 neurons – this can be at the synaptic levels between CA3 & CA1 &/or CA1 & CA3, &/or CA3 & granule neurons &/or CA3 & entorhinal neurons.

2. Since any set of nearly 140 input signals can fire a cortical pyramidal neuron, pyramidal neurons having thousands of input terminals can be fired by a gigantic combination of input signals arriving through those input terminals (Palmer et al. 2014; Eyal et al., 2018).

3. Reaching a place is associated with first-person inner sensations of that place. Memory of food located in a specific place is associated with first-person inner sensation of food in a particular location.

4. Firing of sets of sparsely located CA1 neurons indicates that specificity to any function is achieved by firing of certain CA1 neurons being held at subthreshold activation states.

These lead to the question, “What type of a mechanism occurring at the lower level, presumably at the input level of CA1 neurons that can generate both first-person inner sensations and firing of different sets of CA1 neurons?” Semblance hypothesis provides a solution to this. Semblance hypothesis describes an operational mechanism at the initial locations of the input terminals (see the FAQ section on the website). To summarize, according to the semblance hypothesis, learning generates inter-neuronal inter-spine LINKs between spines that belong to different neurons at the locations of convergence of pathways through which associatively learned stimuli propagate. Two inter-LINKed spines can inter-LINK with more spines of different neurons during additional associative learning events to form an islet of inter-LINKed spines (Fig.1).


Figure 1. One islet of inter-LINKed spines having 32 spines (dendritic spines) that belong to 32 different neurons are shown. Due to space limitations only neurons (N1, N2, N3 and N4) of 4 inter-LINKed spines are shown in the picture. Depending on the location of the inter-LINKed spine where the input signals from a cue stimulus arrive in an islet of inter-LINKed spine, its neighboring inter-LINKed spines will be depolarized. This also depends on whether the neighboring spines are receiving inhibitory inputs at the same time. Along with this, depolarization propagates from the inter-LINKed spines to their neurons. If the neuron is being held at sub-threshold activate level short of the potentials that are arriving, then the neuron will fire.

Note that spines of one neuron can be part of different islets of inter-LINKed spines (Fig.2). Chettih et al.,’s work (Chettih et al., 2024) shows that at the time of memory retrieval of a place or the location, different sets of neurons fire. According to semblance hypothesis, during different learning events, converging pathways of associated input signals from inter-LINKed spines. Different spines of a neuron can inter-LINK with spines that belong to different neurons. Since an inter-LINKed spine can inter-LINK with more than one spine, it leads to the formation of islets of inter-LINKed spines. When this continues, spines of a neuron will be part of different islets of inter-LINKed spines, which will give a picture shown in Fig.2.

Figure 2. Spines of a neuron are part of different islets of inter-LINKed spines. Here, three islets of inter-LINKed spines are shown. Note that one spine of the neuron N1 on the leftmost side is part of the leftmost islet of inter-LINKed spines (having 19 inter- LINKed spines. Also, another spine of the same neuron N1 is part of the middle islet of inter-LINKed spines (having 31 inter-LINKed spines). Similarly, neuron N4 has one spine as part of the middle islet of inter-LINKed spines. It also has another spine that is part of another inter-LINKed spines (with only 8 inter-LINKed spines). When a cue stimulus arrives at an inter-LINKed spine in an islet of inter-LINKed spines, then units of first-person inner sensations is expected to be generated. Along with this, potentials from the inter-LINKed spines reach their neurons. If addition of these potentials can allow the net potential to cross the threshold for firing an action potential, then that neuron fires.

According to the operational mechanism derived by the semblance hypothesis, first-person inner sensation of a memory of food or a place occurs at the inter-LINKed spines within the islets of inter-LINKed spines. Furthermore, firing of postsynaptic neurons depends on amount of depolarization arriving at these sub-threshold activated neurons. Based on these, findings in Chettih et al.,’s work (Chettih et al., 2024) can be explained in terms of specific sets of input signals arriving at different sets of islets of inter-LINKed spines leading to firing of different sets of neurons that are being held at sub-threshold activation levels. Each higher brain function in which hippocampal CA1 neurons are involved, it leads to firing of different sets of neurons that in turn lead to the appearance of a barcoding of firing neurons.

Chettih SN, Mackevicius EL, Hale S, Aronov D (2024) Barcoding of episodic memories in the hippocampus of a food-campus of a food-caching bird. Cell. 187(8):1922–1935.e20. Article

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

Frank LM, Brown EN, Wilson M (2000) Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27:169–178. Article

Gauthier JL, Tank DW (2018 ) A dedicated population for reward coding in the hippocampus. Neuron. 99(1):179–193.e7. Article

O’Keefe J, Dostrovsky J (1971) The hippocampus as a spatial map. Preliminary evidence from unit activity

 in the freely-moving rat. Brain Res. 34:171175. 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

Park SA, Miller DS, Nili H, Ranganath C, Boorman ED (2020) Map Making: Constructing, Combining, and Inferring on Abstract Cognitive Maps. Neuron 107:1226–1238.e8. Article

Pastalkova E, Itskov V, Amarasingham A, Buzsáki G (2008) Internally generated cell assembly sequences in the rat hippocampus. Science. 321(5894):1322–1327. Article

Schuck NW, Niv Y (2019) Sequential replay of nonspatial task states in the human hippocampus. Science 364:eaaw5181. Article

Waydo S, Kraskov A, Quian Quiroga R, Fried I, Koch C (2006) Sparse representation in the human medial temporal lobe. The Journal of neuroscience. 26:10232–10234. Article

A ubiquitous spectrolaminar motif of local field potential power across the primate cortex. Mendoza-Halliday et al., (2024) Nat. Neurosci. doi: 10.1038/s41593-023-01554-7. Article

During the development of cerebral cortex, there is a particular pattern in which neuronal cells move and settle in the cortical layers. This has a profound influence on the inter-neuronal inter-spine interactions and hence the waveform of oscillating extracellular potentials. At one stage of development, neuronal cells from the periventricular zone move up along the vertically oriented processes of the radial glia towards the pial surface (Marín-Padilla, 1998). The neurons that reach the subpial region anchor their processes to the marginal zone close to the pia and then descend towards the direction of the ventricular zone area (Fig.1). As the new neurons arrive at their destination, they continue to settle one above the other. Thus, the first set of neurons becomes the sixth neuronal layer of the cortex. This is followed by the fifth neuronal layer and so on. Layer 1 cortical neurons that are mostly GABAergic send horizontal processes interconnecting several postsynaptic terminals of apical tufts. Because of the subpial anchoring of the apical dendritic tufts for all the neurons of all the layers, dendritic arbors of all cortical pyramidal neurons overlap each other more densely in cortical layer 2. The overlap will be less in layer 3; further reduced in layer 4 and so on (Fig.2). Maximum number of synapses and hence inter-neuronal inter-spine interactions are expected to occur in layer 2 providing large number of components of oscillating extracellular potentials in layer 2. Furthermore, large number of thalamocortical inputs also contribute large number of components for these oscillations. Hence, the net effect is expected to increase the frequency of oscillating extracellular potentials in layer 2 when two differential electrodes (extracellular) are placed at two locations in the cortex. This provides an explanation for the observation of high frequency gamma waveforms of oscillating extracellular potentials in layers 2 & 3, and comparatively low frequency alpha and beta waves in deeper layers in the work of Mendoza-Halliday et al.,. This also acts as an indirect retrodictive finding that matches with the semblance hypothesis.


Figure 1. Stages of neuronal migration and arrangement of neurons in different neocortical neuronal orders. Both figure panels are views of the vertical section through the cortex. A: Progenitor neuroepithelial cells in the fetal ventricular zone proliferate, and the newly formed daughter cells migrate along the processes of the radial glia towards the superficial layer of the cortical plate. The new cells develop processes, and the dendrites are anchored to the extracellular matrix structural proteins at this region. As new cells arrive at the superficial layer, the older ones get pushed towards the direction of the ventricular zone. However, their apical dendrites remain anchored to the superficial layer. Since the dendrites are already anchored to the superficial layer, the main dendritic stem elongates. This continuous process results in the displacement of the oldest cell layer, namely layer 6 located close to the ventricular zone and the last arrived cells to remain at the most superficial layer as neuronal order 1. B: Figure showing the dendritic trees of neurons that belong to different neuronal orders (numbered 1 to 6). Note that the dendrites of neurons of almost all the neuronal orders anchor at the subpial region, making this region rich in dendritic spines. As the dendritic spine density is very high at the cortical layers 1 and 2, the role of this arrangement contributing to the interspine interactions and oscillatory waveform of the cortical surface-recorded potentials are explained in the following sections (figure modified from Vadakkan, 2015).


Figure 2. Anatomy of pyramidal neurons and locations of spike generations. A: Diagram showing different locations of neuronal processes that are capable of producing regenerative potentials (spikes). These include apical tuft, apical trunk, oblique, basal, and axon. Synapses in distal dendrites produce EPSP of amplitude more than 10 mV; whereas those proximal to the soma produce EPSP amplitude of 0.2–0.3 mV. The EPSPs from distal dendrites attenuate from nearly 10 millivolts (mV) to nearly 0.014 mV (more than 900-fold attenuation) as they reach the soma (Spruston, 2008). B: Diagram showing the locations of spike generation and inhibitory mechanisms to regulate spike propagation. In the apical tuft, oblique, and basal dendrites, several dendritic conductance contributed by regenerative NMDA receptor current trigger dendritic plateau potentials with a rapid initial sodium spikelet followed by a plateau phase that collapses abruptly (Schiller et al., 2000). At the apical trunk, calcium spikes are generated. Axonal spikes (action potentials) are generated at the axon initial segment. Inhibitory inputs can regulate the net potential reaching the soma. Both recurrent collateral and thalamo–cortical inputs control the generation and propagation of the spikes at different locations. Layer 1 cortical neurons that are mostly GABAergic send horizontal processes interconnecting several postsynaptic terminals of apical tufts. Representative traces of different spikes are shown (figure modified from Vadakkan, 2015).

Mendoza-Halliday D, Major AJ, Lee N, Lichtenfeld MJ, Carlson B, Mitchell B, Meng PD, Xiong YS, Westerberg JA, Jia X, Johnston KD, Selvanayagam J, Everling S, Maier A, Desimone R, Miller EK, Bastos AM (2024) A ubiquitous spectrolaminar motif of local field potential power across the primate cortex. Nat. Neurosci. doi: 10.1038/s41593-023-01554-7. PubMed

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

Vadakkan KI (2015) A pressure-reversible cellular mechanism of general anesthetics capable of altering a possible mechanism for consciousness. SpringerPlus 4:485. doi: 10.1186/s40064-015-1283-1. PubMed

Schiller J, Major G, Koester HJ, Schiller Y (2000) NMDA spikes in basal dendrites of cortical pyramidal neurons. Nature 2000;404:285289. PubMed

Marín-Padilla M (1998) Cajal–Retzius cells and the development of the neocortex. Trends Neurosci. 21(2):64–71. PubMed

Vadakkan KI (2016) Rapid chain generation of interpostsynaptic functional LINKs can trigger seizure generation: Evidence for potential interconnections from pathology to behavior. Epilepsy Behav. 59:28–41. PubMed

Hotspots of dendritic spine turnover facilitate clustered spine addition and learning and memory. Frank AC, Huang S, Zhou M, Gdalyahu A, Kastellakis G, Silva TK, Lu E, Wen X, Poirazi P, Trachtenberg JT, Silva AJ. Nat Commun. 2018 Jan 29;9(1):422. PubMed

See supplementary figure 8 in the above article. Learning leads to loss of spines & formation of new spines at those regions (spine turnover). Why should spines get lost? Based on the semblance hypothesis, learning leads to inter-neuronal inter-spine interaction leading to inter-postsynaptic functional LINKs (IPLs) (see figure 8 in FAQ section of this website). Inter-spine fusion is at the extreme end of this spectrum of changes. The nature of IPLs depends on several factors. These include a) nature of fatty acids in the phospholipid molecules that form spine membranes, and b) intensity of stimuli that affect propagation of signals towards the IPLs. If IPL formation leads to an extreme change of inter-neuronal inter-spine fusion, then it will lead to mixing of the contents of cytoplasm of two neurons. Since even adjacent neurons of a similar type vary in their gene expression/protein content (Kamme et al., 2003; Cembrowski et al., 2016), it is reasonable to expect cellular mechanisms for closing the fusion pore. If it is not possible, then the neurons will trigger mechanisms to remove the spines. This can explain spine loss. As a homeostatic mechanism, the involved neurons can be expected to produce new spines using phospholipids that resist inter-spine fusion. Thus, the basic operational mechanism of semblance hypothesis can be extended to provide a mechanistic explanation for spine turnover during learning.

Cembrowski MS, Bachman JL, Wang L, Sugino K, Shields BC, Spruston N (2016) Spatial Gene-Expression Gradients Underlie Prominent Heterogeneity of CA1 Pyramidal Neurons. Neuron. 89(2):351-368. PubMed]]

Frank AC, Huang S, Zhou M, Gdalyahu A, Kastellakis G, Silva TK, Lu E, Wen X, Poirazi P, Trachtenberg JT, Silva AJ (2018) Hotspots of dendritic spine turnover facilitate clustered spine addition and learning and memory. Nat Commun. 29;9(1):422. PubMed

Kamme F, Salunga R, Yu J, Tran DT, Zhu J, Luo L, Bittner A, Guo HQ, Miller N, Wan J, Erlander M (2003) Single-cell microarray analysis in hippocampus CA1: demonstration and validation of cellular heterogeneity. J Neurosci. 23(9):3607-3615. PubMed

Synapse-specific representation of the identity of overlapping memory engrams. Abdou K, Shehata M, Choko K, Nishizono H, Matsuo M, Muramatsu SI, Inokuchi K (2018) Science. 360(6394):1227-1231. (Unfortunately, I saw this paper only on the 20th March 2023. So, posting this explanation very late). Article

Related article: Entorhinal cortex directs learning-related changes in CA1 representations (Grienberger and Magee, 2022) Nature. November, doi: 10.1038/s41586-022-05378-6. Article

Semblance hypothesis has provided a mechanistic explanation for both memory (inner sensation of features of the item/event whose memory is being retrieved) and motor action reminiscent of arrival of that item/event. Work by Abdou et al., has two implicit assumptions. One is that engram cells interconnect between memories and secondly, synapse-specific plasticity ensures the identity and storage of individual memories. The findings in this work need mechanistic explanations for both behavioral motor action (withdrawal of foot) and, if possible, for inner sensation of memory of foot shock. Following is an explanation of how the findings in this paper can be explained in terms of semblance hypothesis. Alternatively speaking, findings in this paper allow us to look at semblance hypothesis from a new angle.

In fear conditioning experiments, two stimuli are associated. Out of this one (foot shock) generates a motor response (foot withdrawal). The other one has no motor response. The one with motor response is called unconditioned stimulus (US). The one that does not trigger any motor response on its own is called conditioned stimulus (CS). Authors associated between US and CS. After learning when CS arrives, motor action in response to the US (that occurred prior to learning) takes place.

At this juncture, we would like to find an explicit answer to the questions a) where associative learning is getting stored and how does memories get retrieved, and b) how behavioral motor activity reminiscent of arrival of a stimulus whose memory is being retrieved. We should understand them with such a clarity that we can explain them to an engineer who wants to replicate the mechanism in an engineered system. For this, the issue can be simplified as given in figure 1.

CS arrives through the auditory cortex (AC) & medial geniculate body. Medial division of the medial geniculate body (MGm or MGN) receives both auditory and somatosensory inputs (LeDoux et al., 1987; Bordi and LeDoux, 1994) and project to LA (LeDoux et al., 1990). Furthermore, it was verified that the US (foot shock) activates lateral amygdala neurons (Lanuza et al., 2008). Medical geniculae body is only a relay station for both auditory and foot shock stimuli without any interactions between them at this level. So, we need to imagine that AC-LA pathway is the auditory stimulus pathway and MGm-LA (MGN-LA) pathway is the footshock pathway. At the same time, we have to keep in mind that MGm (MGN) is the path through which AC connects to the LA. LA neurons receive inputs from both AC (sound) and MGm (foot shock). All the diagrams here refer to MGm path as MGN.


Figure 1. Conditioned stimulus (CS) arrives through neurons of the auditory cortex (ACN). Unconditioned stimulus (US) arrives through neurons of the medical geniculate nucleus (MGN). They synapse with different spines on a lateral amygdala (LA) neuron. All three of the above neurons are referred to as engram neurons. The question is "After associative learning, when the CS alone arrives, how does it trigger motor response (foot withdrawal) as if it is receiving a foot shock?" To answer this, it is necessary to show at least some evidence for an interaction between the spines of LA that synapse with stimuli arriving through both ACN and MGN in the figure.

The first question is, “What is the mechanistic explanation for the firing of lateral amygdala (LA) neuron when US comes after the associative learning between US and CS?” Authors provide and implicit explanation that plasticity changes occur at the spines of LA neuron on which inputs from CS and US synapse. A mechanistic explanation needs to meet 2 requirements. 1) How does arrival of CS alone cause firing of LA neuron reminiscent of arrival of US? 2) How does arrival of CS generate an internal sensation of arrival of US? If a single explanation can provide answers to both these two questions, then there is a good chance that it can be found correct after verification. At this time, readers can have more questions. First, why can't the inputs from two associatively learned item be shown synapsing to two neighboring spines on one LA neuron as follows (Figure 2)?


Figure 2. Inputs from two associatively learned stimuli arrive and synapse on to two neighboring spines on one LA neuron. Can this explain associative learning mechanism?

Looking at figure 2, we can ask the question, "Can it provide an explanation?" An interpretation of clustered plasticity along with synaptic tagging was explained previously by Govindarajan et al., 2006. The problems with this model are 1) there is no evidence for the formation of an electric cable between two neighboring spines, 2) there are no evidence for the formation of such a cable through the extracellular matrix space, 3) no evidence for the formation of specific tag molecules that can form in physiological timescales of milliseconds to explain function.

So, let us examine the underlying issue more closely by asking some questions.

1) How does arrival of CS cause firing of LA neuron reminiscent of arrival of US? An engineer will want to see at least the formation of a cable between the spines to which CS (ACN neuron) and US (MGN neuron) synapses. But there is no evidence for an electrical cable between them. Synaptic tagging was put forward as a mechanism (Frey and Morris, 1998). But sufficient number of specific tag molecules that can form and act at physiological timescales are not found. The solution must provide a mechanistic explanation.

2) Since there is no evidence for a long-lasting direct interaction between two spines that receive CS and US on an LA neuron, the configuration given Figure 2 is not compatible with the actual mechanism of learning. So, is there an alternative?

3) We can also ask, "How does arrival of CS generate an internal sensation of memory of arrival of US?" An ideal solution for the first question is expected to answer this question as well. Currently, we are not searching for a mechanism that generates inner sensations of memory due to several reasons that we can only assume. But an engineer who wants to replicate the mechanism will need to see a blueprint that explains a mechanism for this.  

So, the question is, "How to move forward to provide an explanation for both behavioral motor action reminiscent of memory retrieval and the very process of memory retrieval itself?" Both these are intricately connected parts of a single mechanism. Hence, we need to find such a mechanism that will also allow us to explain how the brain can store very large number of associated memories that are associated with firing of neuronal ensembles. Results from the present paper demands that the mechanism should be able to explain how memories are stored apparently using the same neuron in such way that they can be distinguished from each other.

This needs a renewed approach. First, let us see how the dendritic tree is organized. Are the tree branches of a neuron similar to the tree branches of a tree in a forest? Tree branches of a tree in a forest usually don't overlap with each other so that the leaves can get maximum sunlight. But the dendritic arbor of neighboring neurons overlap intensely so that we cannot separate the arbor of one neuron from another one (Figure 3).