In my Model of Memory from neuron to Memory System, How Memory Might Work I suggest that Marr’s definition of Neocortical Memory as being a self-classifying content addressable memory, modified by my Dual Mode Cortex paper suggests an architectural role for the cerebral cortex as main memory.

But all that requires that Marr be essentially correct about the function of the top 3 layers of cortical tissue. Marr has been soundly criticized for presenting a 4 layer solution to a six layer problem, but I have found that there is a tissue type very close to Marr’s model called Three Layer Allocortical Tissue, which might reflect the role he has suggested, and noted that other types of Allocortical Agranular, and Isocortical tissue have virtually the same type of top three layers, and have hypothesized that this means that the deeper tissue layers are an add-on or development based on the original three layer architecture.

In an attempt to dig a little deeper, I tried to find information on the Micro-architecture of the top three layers, and found first Eccles, and then
Braak to help me put together the pieces.

First of all we know that the layering system is based on classical interpretations of staining studies. There is some problem with classifying where the layers begin and end, if only because the brains organization is so soft in so many ways. Probably we will find as new staining techniques appear that cortex tissues should have more layers, the classification of sublayers such as IIIc is an indication of how we are trying to keep the six layer structure, while finding deeper complexity and so adapting our six layer structure to a wider number of sub-layers.

This is one of the problems with the anatomical concepts, is that until we begin to really understand what is going on, within the cortex, we won’t really know where to place the boundaries, and how many layers to denote.

Layers 2/3 are often called the Superficial Layers, as if they are somehow less important than the deeper layers. Perhaps this is because they tend to be more homogeneous across the cortex. You might think that it is a mistake for Layer 1 to be ignored, but in fact, Laminae I as it is sometimes called, was shown in Eccles, and Braak to be mostly myelinated fiber, with only sparse neural elements.

The main interest we have in Laminae I, is where the connections that are represented by the fibers connect, and according to Eccles they have multiple sources. It is a mistake however to assume that this is a general rule, and that laminae I is uniform in its connectivity. The sharp reduction of Laminae I fibers found in Associative Areas, suggests that mapping the connectivity across different areas of the cortex, might result in a map of some interest. When combined with my model, it might indicate whether there is support for the models assumptions about how processing is done in the brain, simply by confirming relationships between different segments of the brain at different stages of processing, as for instance indicated by the Core/Belt/Associative area processing center model.

Laminae II and III are often clumped together, suggesting that there might be a neural circuit that operates in these two layers making them into some sort of functional element. Some researchers in computational Neuroscience are discussing the idea that the two layer circuit is some sort of self-correcting storage element. Either offering a negative feedback loop, or a positive feedback loop, that essentially reinforces previously stored patterns of storage for similar stimuli.

Marr’s suggestion, which I think has merit is that this storage layer acts as a content addressable memory, I mean that in the sense that a stimulus, is analyzed by parallel evaluation against either past storage, or predictions based on past storage (the negative/positive feedback loops) in order to reinforce recognition of stimuli that repeat. The output is essentially a mapping of the cortex, that indicates every stimulus, that might be triggered by a particular environmental state, all presented in parallel. I call this a data field, to distinguish it against “Memories” which would be discrete clusters of stimuli that are somehow linked together.

The primary difference between layer 2 and layer 3 seems to lie in the target for the axons of the pyramidal neurons that dominate the layers. Layer 2 seems to be locally linked while Layer 3 axons penetrate to the deeper levels of the brain, suggesting that Layer2 neurons are probably the feedback components and Layer 3 are the outputs for the circuit.

Superficially we have defined the circuit, and we know what it does, but are we correct? We are not sure, because we know that Laminae III is more complex than its layer declaration indicates. Further look at the architectonics is warranted, and indeed when we look at the micro-structure of the layers, we see that there are sub-layers within Laminae III that are not well dealt with by the model. sublayers a and b, seem consistent, but sublayer IIIc is obviously important because of the changes in this layer that we find over multiple sites when we study the staining studies.

It is interesting to note that IIIc is often linked to Laminae IV, which suggests a blurring of laminae that might indicate problems with the six layer model. Also, there is sometimes evidence for layer swapping in the sense that Laminae IIIc might transfer to Laminae I or II in transitional zones between different micro-architectures. What could this possibly mean?

What, I hypothesize if the different sub-layers do not affect the process so much, as the target of the neural outputs. Laminae IIIa would target one zone of the brain, target IIIb another, and Laminae IIIc another.

Working from this hypothesis, I looked at the Laminae IIIc sublayer in Braaks Architectonics of the human telencephalic Cortex and found evidence of myelinated fibers connecting adjacent areas of the brain in IIIc. This suggested to me that IIIc was a Local Transport Layer. Which suggested that the difference between IIIc in the core and IIIc in the belt was probably whether the signal was being sent or recieved. The larger IIIc pyramids in the Core, and the fact that I expected flow from the Core to the Belt, suggested that the size of the pyramidal cell, might be a function of the distance it transported the signal. The Pyramids that transmit would be larger than the pyramids that receive. Possibly because in order to extend their processes over a distance, they had to start out as bipolar cells and grow the processes before differentiating into pyramidal cells, to store data. As a result of this two stage growth, they would tend to “Grow” more than the other less far reaching cells and thus, their size would be larger.

The sheer difference in size between the smallest pyramidal cells, and the gigantic ones might be related to how long the cells remained in the bipolar stage. Perhaps the soma size of a bipolar cell, is some factor of the distance they have to build the processes that connect the neuron to its target neurons, possibly because of the support role the soma has for producing the energy needed to sustain the longer processes. In any case a hypothetical connection between size and distance is possible.

If so, then the Belt pyramidal cells in IIIc being smaller indicate that the distance they send their signals is reduced in relation to the distance that the Core pyramids in IIIc do. But, what possible use is the ability to send the output from the core, to the output layer of the Belt? It might make sense to transfer the output from laminae IIIc to laminae I and feed it through the processing of the superficial layers in a new location, but having two different versions of the same output, one intermixed with the belt output, seemed a bit strange, until I realized that the outputs of laminae III, make up the stimulus that forms Neural Groups. In other words neural groups are dependent on the content of Laminae III. If this was true, transferring data between sections of the brain at laminae III would be useful, in changing the nature of the Neural Group Outputs.

In my memory model, I have suggested that these first three layers, define an essentially implicit memory system, that operates as if it were a content addressable memory. But if the core/belt connection changes the nature of the implicit output at the Neural Group level, then at that level we might get recognition of simularities between the primary percpetion areas in the core, and the secondary perception areas in the belt. this can’t be thought of as a normal implicit memory function even though the content is still implicit.