Our traditional idea of how the brain works is based on the neuron: it fires action potentials, which travel along the axon and, when the reach the synapses, the receiving neuron performs a calculation that results in the decision when (or whether) to fire its own action potential. Thus, the brain, from a thinking point of view, is viewed as a network of neurons each performing its own calculation.
This view, which I'm going to call the axon-centric view, is simplistic in many ways, and two recent papers add to it, pointing up the ways in which the glial cells of the brain participate in ongoing calculation as well as performing their more traditional support functions.
In the traditional view, the brain was made up neurons, which processed data, and glial cells, which performed support functions, including "surround neurons and hold them in place, to supply nutrients and oxygen to neurons, to insulate one neuron from another, and to destroy pathogens and remove dead neurons." (Wiki). In recent years, this picture has been breaking down.2, 24, 25, 31, 32
Glia are generally divided into two broad classes, microglia and macroglia. Microglia are part of the immune system, specialized macrophages, and probably don't participate in information handling. Macroglia are present in both the peripheral and central nervous systems, in different types. For purposes of analyzing the brain, which is part of the central nervous system (CNS), we will ignore the types not present there. ...
Traditionally, there were four types of glia in the CNS: astrocytes, oligodendrocytes, ependymal cells, and radial glia. Of these, the one type that's most important to the developing revolution in our ideas are those cells called astrocytes.2 It turns out that there are at least two types of cell (at least) subsumed under this name.24, 25, 31, 32 One, which retains the name of astrocyte, takes up neurotransmitters released by neurons (and glial cells), aids in osmoregulation,10 controls circulation in the brain,1, 31 and generally appears to provide support for the neurons and other types of glia.
The other type have several names, including NG2 glial cells,23 Synantocytes,25 oligodendrocyte precursor glia,26 NG2(+) Progenitor Cells,28, 29, 33 PDGFRA/NG2 glia,30 and NG2-glia,32 which last name will be used here (because it's short). While they are similar to astrocytes, there are distinct differences:
Astrocytes have asymmetrically radiating processes that consist of several primary thick processes, from which emanate multiple smaller collateral branching secondary processes, giving them a bushy appearance (Fig. 1C; [refs]). By contrast, NG2-glia have a centrally located soma, from which extend several long, slender primary processes, which bifurcate two or more times, but less extensively than astrocytes, to form a symmetrical process field (Fig. 1D). Astrocyte processes often end in bulbous swellings, or terminal end-feet, which form the vascular and pial glia limitans (Fig. 1A), whereas NG2-glial processes taper to an end and do not appear to contribute to the glia limitans (Fig. 1B; [refs]). Although both NG2-glia and astrocytes extend processes to nodes of Ranvier in white matter ([refs]) and synapses in grey matter, their geometric relationship to these neuronal elements is different. Thus, although astrocytes and NG2-glia bear a superficial resemblance, they are distinguished by their different process arborizations. This will reflect fundamental differences in the way these two glial cell populations interact with other elements in the neural network.32 [The figure number matches with here.]
Figure 1: Astrocytes and NG2-glia. A and C are astrocytes, B and D are NG2-glia. Click on image to see full image. (From Ref 32.)
Both types of glia are closely integrated with the nervous system, receiving information from action potentials via synapses22 (which, only a few years ago were thought to be limited to neurons), and returning control of neuron activity through release of neurotransmitters and other modulators. Both, then, demonstrate the potential for considerable intelligent activity, contributing to the overall intelligence of the brain.
Astrocytes probably (IMO) are limited, or mostly so, to maintaining the supplies of energy and necessary metabolites. They receive action potentials,3, 6 which allows them to closely and quickly monitor general activity and increase circulation in response, even before the neurons and NG2-glia have reduced their supply of ATP.21 They appear to be linked in a network among themselves,2, 5 allowing them to communicate their needs without interfering with the higher-level calculations of the brain. (They may also serve to regulate overall neural activity in response to reduced supplies of oxygen or food.1, 4, 8)
NG2-glia appear to have several functions, but one of the most exciting things about them is that they seem to be able to fire action potentials.33 Their cell membranes, like those of the dendrites of neurons, have all the necessary channels and receptors to perform real-time electrical calculations in the same way as neural dendrites. They have also demonstrated the ability to learn through long term potentiation.27
In one of the two new papers mentioned above,23 Dividing glial cells maintain differentiated properties including complex morphology and functional synapses (by Woo-Ping Gea, Wei Zhoub, Qingming Luob, Lily Yeh Jana, and Yuh Nung Jan), we find that these cells are capable of maintaining their full function even while dividing. Evidently, they maintain a network of processes, which would be called dendrites if they were neurons, with the soma pretty much being a support structure. When the network grows too big, the soma divides, with each daughter soma getting part of the processes.
Figure 2: NG2-glia cell division while retaining differentiated morphology. Click on image to see original caption, and enlargeable image. (From Ref 23.)
Dividing NG2-glia also retain the ability to fire action potentials, as well as receiving synaptic inputs from neurons.23 Presumably, they continue to perform their full function, including retaining any elements of long term potentiation or depression contained in their synapses.27
The second paper,30 PDGFRA/NG2 glia generate myelinating oligodendrocytes and piriform projection neurons in adult mice (by Leanne E Rivers, Kaylene M Young, Matteo Rizzi, Françoise Jamen, Konstantina Psachoulia, Anna Wade, Nicoletta Kessaris & William D Richardson), shows evidence that this type of cell can mature into either oligodendrocytes or neurons. Oligodendrocytes are responsible for the insulation of the axons, wrapping around approximately 1 mm of each of up to 50 axons within their reach, and forming the myelin sheath.
Although the precise type of neuron formed by maturing cells hasn't been determined, the very fact that cells of this type can change into neurons is very important. We actually don't know whether the cells that do this maturation are the same as those that perform neuron-like activities, there appear to be two separate types of NG2-glia, spiking and non-spiking.26 It may very well be that the "spiking" type have actually differentiated, while the "non-spiking" type may be doing the maturing. Of course, very few differentiated cell types remain capable of division, as even the "spiking" type do.23
Another possibility is that the "non-spiking" type may be able to mature into either oligodendrocytes or "spiking" types, while the "spiking" types may be able to grow an axon and turn into neurons. There's much we don't know yet about these cells.33
Although the NG2-glia appear to be able to fire (or at least propagate) action potentials, they probably don't in vivo. Rather, they probably act much like dendrites, which are also capable of propagating action potentials, but are usually thought not to fire them. (Rather, they will back-propagate action potentials fired by the axon hillock.18)
What's important about both dendrites and NG2-glia isn't so much their ability to propagate action potentials, as that their entire cell membranes are capable of "intelligent" manipulation of the voltage across it. Let's start by looking at the action potential.
Action Potentials and Ion Channels
While there are many ion channels involved in controlling the voltage across the cell membrane, the only type we really need to worry about for action potentials is voltage-gated sodium channels. These are channels that sometimes allow sodium ions to pass through the cell membrane, which they will do because the concentration of sodium ions outside the cell is very much higher than inside. When and how much they open depends, among other things, on the voltage across the membrane.
A normal neuron will have a voltage of around -60 to -80mV (millivolts), in a direction that tends to push the sodium ions (which are positive) into the cell (the same direction as the concentration is pushing). When the voltage falls to around -55mV, the primary type of gate will open for a millisecond or so, after which it will close and rest for several milliseconds. It won't be able to open again until the voltage is somewhere between -55 and around -10mV. Meanwhile, the sodium current has caused the voltage to swing past zero to around +20mV. (The high concentration of sodium outside the cell allows it to drive the ions "uphill" electrically.)
When one part of the cell membrane is "depolarized" in this fashion, the voltage near it is also depressed. Thus, if the voltage is at zero at one point, it might be at -20mV 10 microns (μm) away, and -40mV 20μm away, and -60mV 30μm, and so on. Notice that somewhere between 20μm and 30μm, it has passed the threshold for the ion channels, which means that they are open, allowing a current that drives the voltage further down. This will produce a wave of voltage drop along the membrane, which is what the action potential is.
After the action potential has passed, and the gates have closed (see above), the voltage is recovered by diffusion of ions towards and away from the membrane, the opening of other gates (primarily potassium), and a set of pumps that push the ions back to their resting state. These pumps are mostly powered by the sodium gradient, except for the sodium/potassium pump that maintains it, which is powered by ATP. (See my Wiring the Cell for Power for more on the cell's power systems.)
In the laboratory, action potentials are usually fired by applying a voltage to the cell membrane at some point, but how are they fired in vivo? Although I know of no research to back me on this, IMO the cell creates "hot spots", where a voltage greater than -55mV is capable of opening the ion channels. How is this accomplished? Most ion channels come in a large variety of types: there are nine types of primary (alpha, α) subunits of the voltage-gated sodium channel, which can operate on their own but usually associate with secondary (beta, β) subunits, of which there are at least 4 types. In addition, there are other types of modulation that can be applied. If a bunch of channels that are modified to open at a voltage greater than -55mV, say -60mV, are concentrated in one place, then if the voltage at that place falls below that value the action potential fires. In almost all neurons, the "hot spot" will be on the axon hillock.
What modifies the voltage on the soma and axon hillock? Well, every time an action potential arrives at a synapse, the pre-synaptic part of the synapse releases a bunch of neurotransmitters, which cross the synaptic cleft and interact with receptors on the post-synaptic part, usually called the post-synaptic density (PSD).
Figure 3: Synapse. (From Wiki.)
Some of these receptors are "ligand gated ion channels", which can allow currents of sodium, chloride, potassium, and calcium in response to neurotransmitters. These currents, in turn, modify the voltage at the synapse, creating a voltage wave that spreads along the cell membrane, that is the membrane of the dendritic arbor. The voltage difference created by any action potential at any synapse decays in the manner described above, even if it never reaches the threshold for an action potential. Similarly, all the other synapses on that post-synaptic cell from that axon (if any) create their local currents, at other points in the arbor. These voltage differences decay in both space and time, the current fading out as the neurotransmitters are taken up or diffuse out of the synapse. While this is happening, other action potentials, in other axons, are adding their own contribution, causing the distribution of voltage across the membrane of the entire arbor to vary in an extremely complex manner.17 Note that depending on the type of ions involved in the current, it can be depolarizing (tending to lower the voltage towards the threshold) or hyperpolarizing (tending to raise the voltage even farther away from the threshold).
If the cell membrane away from the synapses were passive, with no change in current in response to voltage changes, the voltage distribution at any point could be calculated using passive cable theory.38, 40 However, membranes outside the synapses are well supplied with voltage-gated ion channels,11, 13 which create their own non-linear reactions to voltage changes,16, 17 of which the action potential is only one example, and IMO not the most important. In dendrites, the result is an incredibly complex, non-linear, response to the synaptic inputs, which contains the potential for enormous calculating capacity, much higher than the point calculations performed by synapses.
This is especially true because many of these ion channels are not only sensitive to voltage, but their voltage-sensitivity can be modified by a wide variety of different neuro-modulators,15 many of which are released by axon arbors in response to action potentials. In addition, both astrocytes and NG2-glia have been shown to release chemicals that function as neuro-modulators, in response to their own cellular circumstances.1, 8, 9, 11, 12, 13
Neuro-Glial Calculations Beyond Action Potentials
The fact that astrocytes can release these neuro-modulators, partly in response to electrical signals received from action potentials in axons,3 demonstrates that action potentials within a specific cell aren't necessary for that cell to perform its release. Unless and until NG2-glia have been shown to have action potentials in vivo, we can assume the same to be true of them.
Although it hasn't been proven yet, we can certainly theorize that this release is partly voltage-dependent, probably through the use of calcium currents stimulated by voltages that don't reach the threshold for action potentials. (I'll note that calcium currents are almost always associated with the release of vesicles full of neuro-transmitters and -modulators,8, 9 because of the energy required. See my How Smart is the Cell: Part IV: Local Intelligence, especially The Calcium System, for further discussion.) Certainly calcium channels can be stimulated by a variety of sub-threshold voltages.20 This means that both dendritic networks and NG2-glial networks are tied together in an extremely complex network of electro-chemical analog calculations,28 with far greater potential calculating ability than even the collection of dendrites that mediates the high-speed action potential-type communication.
We might think of the brain, then, as an extremely large organization, with offices spread out through an entire densely populated country, but without telephones or other modern communications; sort of the way things were in the 1800's. Most communication has to take place via letters in the post, or face-to-face communication. However, there is a telegraph network to provide high-speed transmission of a very small subset of the total information involved.
Similarly, the vast majority of calculation that goes into human intelligence takes place at the level of the network of dendrites and NG2-glia, with the whole system of axons, dendrites, and action potentials only carrying a tiny subset of the total information over long distances. This is especially important considering that the human brain has a much higher proportion of glial matter than our relatives.34, 35, 36, 37
The Membrane-Centric Approach
This, in turn, suggests that our overall approach to understanding the brain has been far too axon centric, there needs to be a shift to a more membrane-centric approach to understanding how the brain creates intelligence.
Another point to keep in mind is the differences between various regions of the brain. Although brain cells are divided into broad classes: pyramidal cells, interneurons, astrocytes, NG2-glia, and so on, the cells in each region of the brain may well behave differently. Each cell "knows" where it is, it had to in order to create the right connections, which vary depending on which region of the brain, even which region of the neocortex, they're in. Not only that, but in each layer of the neocortex the cells make different types of connections, so they must know that as well. If they know enough to make connections differently, they know enough to express different patterns of ion channels and vesicles at appropriate locations within the membrane.14 Thus, whenever research finds some capability in some type of cell in one part of the brain, we know that same capability may be used anywhere (or everywhere) else, but we don't know if it will be used in the same way. And when research fails to find some mechanism used in one part of the brain, that certainly doesn't mean it isn't used in other parts.
This adds great complexity to the membrane-centric approach, since it's not only the elements of the action potential and synaptic communication that may be specific to each region, but the way in which the neuro-glial network operates. For instance, astrocytes have been investigated in several regions, and while they have the ability to respond to neuro-transmitters released by axons, no sign of action potentials have been found in them. But that doesn't mean that the astrocytes in some other regions of the brain haven't evolved this capacity, presumably in response to a need for faster coordination of blood flow.
In my view, the membrane-centric approach, as it is implemented in research, will provide tremendous insight into the functionality of the brain. The opportunities for new research, and exciting findings, are enormous. It'll be interesting to see what turns up.
Ge, W., Zhou, W., Luo, Q., Jan, L., & Jan, Y. (2009). Dividing glial cells maintain differentiated properties including complex morphology and functional synapses Proceedings of the National Academy of Sciences, 106 (1), 328-333 DOI: 10.1073/pnas.0811353106
Rivers, L., Young, K., Rizzi, M., Jamen, F., Psachoulia, K., Wade, A., Kessaris, N., & Richardson, W. (2008). PDGFRA/NG2 glia generate myelinating oligodendrocytes and piriform projection neurons in adult mice Nature Neuroscience, 11 (12), 1392-1401 DOI: 10.1038/nn.2220
1. Astrocyte Control of Synaptic Transmission and Neurovascular Coupling
(I've included only the links referenced in this leader.)
2. Astrocytes, from brain glue to communication elements: the revolution continues
3. NMDA Receptors Mediate Neuron-to-Glia Signaling in Mouse Cortical Astrocytes
4. Synapse-Specific Expression of Functional Presynaptic NMDA Receptors in Rat Somatosensory Cortex
5. The astroglial syncytium summary only
6. P2X1 and P2X5 Subunits Form the Functional P2X Receptor in Mouse Cortical Astrocytes
7. Inhibition of the ATP-gated P2X7 receptor promotes axonal growth and branching in cultured hippocampal neurons
8. Glutamate-induced Exocytosis of Glutamate from Astrocytes
9. Fast Subplasma Membrane Ca2+ Transients Control Exo-Endocytosis of Synaptic-Like Microvesicles in Astrocytes
10. Substantia nigra osmoregulation: taurine and ATP involvement
11. Tonic activation of NMDA receptors by ambient glutamate of non-synaptic origin in the rat hippocampus
12. Regulation of Synaptic Transmission by Ambient Extracellular Glutamate
13. Glutamate Transporters Regulate Extrasynaptic NMDA Receptor Modulation of Kv2.1 Potassium Channels
14. Potassium Channel Phosphorylation in Excitable Cells: Providing Dynamic Functional Variability to a Diverse Family of Ion Channels
15. Dendritic Excitability and Synaptic Plasticity
16. Spine Neck Plasticity Controls Postsynaptic Calcium Signals through Electrical Compartmentalization
17. Timing and Location of Synaptic Inputs Determine Modes of Subthreshold Integration in Striatal Medium Spiny Neurons
18. Differential Excitability and Modulation of Striatal Medium Spiny Neuron Dendrites
19. G-Protein-Coupled Receptor Modulation of Striatal CaV1.3 L-Type Ca2+ Channels Is Dependent on a Shank-Binding Domain
20. Modulation of Voltage- and Ca2+-dependent Gating of CaV1.3 L-type Calcium Channels by Alternative Splicing of a C-terminal Regulatory Domain
21. Rapid Astrocyte Calcium Signals Correlate with Neuronal Activity and Onset of the Hemodynamic Response In Vivo
22. Synaptic transmission onto hippocampal glial cells with hGFAP promoter activity
23. Dividing glial cells maintain differentiated properties including complex morphology and functional synapses
24. Polydendrocytes: NG2 Cells with Many Roles in Development and Repair of the CNS abstract only
25. Synantocytes: the fifth element
26. Spiking and non-spiking classes of oligodendrocyte precursor glia in CNS white matter
27. Long-Term Potentiation of Neuron-Glia Synapses Mediated by Ca2+-Permeable AMPA Receptors requires free registration
28. Satellite NG2 Progenitor Cells Share Common Glutamatergic Inputs with Associated Interneurons in the Mouse Dentate Gyrus RB?
29. Vesicular release of glutamate from unmyelinated axons in white matter
30. PDGFRA/NG2 glia generate myelinating oligodendrocytes and piriform projection neurons in adult mice
31. Astrocyte-mediated control of cerebral blood flow
32. Astrocytes and NG2-glia: what's in a name?
33. Synapses on NG2-expressing progenitors in the brain: multiple functions?
34. The scalable mammalian brain: emergent distributions of glia and neurons
35. Evolution of increased glia–neuron ratios in the human frontal cortex
36. Cellular scaling rules for primate brains
37. Increased Cortical Expression of Two Synaptogenic Thrombospondins in Human Brain Evolution
38. Distinctive Roles for Dendrites in Neuronal Computation
39. The synaptic organization of the brain by Gordon M. Shepherd
40. Linear Electrical Circuit Theory