1. Introduction

Neurons have long been taken to be the relevant functional units in the brain. Neurons are specialized to fire action potentials in response to other neurons. They are discrete, well-defined biological entities, producing what look like paradigmatic signals. Textbooks present information as literally “flowing through” neurons, from the dendrites that collect signals from other cells, through the cell body, which integrates the signals and generates an outgoing signal that propagates down the axon and crosses the synapse to activate the next set of neurons in the network. Bullock et al. (Reference Bullock, Bennett, Johnston, Josephson, Marder and Fields2005) have described this picture as the “Neuron Doctrine.” As a model, the neurocentric picture has been useful for science; but as with other stripped-down models, it has the potential to mislead as well.

The received view—that brain activity is essentially constituted by signaling between its functional components—can be interpreted in two ways. On one interpretation, it is signaling that is important, perhaps because we have independent theoretical reasons to think that the brain processes information. Thus, we should focus on the signals, wherever they may lead. I will argue that if we commit to action potentials as the relevant signals, we should end up taking synapses rather than neurons to be the relevant functional units.

On another interpretation, it is the neurons that are important, and so we should take whatever neurons do, en masse, as constitutive of brain function. But neurons are complicated cells for which spiking may be means to other ends, rather than their sole function. If we focus on the full repertoire of neuronal activities, the signaling-as-spiking model of brain function may seem less compelling.

To start with the first interpretation, let us look at a standard description of an action potential in more detail. Action potentials result from the activation of voltage-sensitive ion channels in the cell membrane by depolarization resulting from synaptic activation. They sweep through the cell, leaving traces of their passing in the cell. They activate synapses as they go, which then in turn help to initiate more action potentials in other cells.

Notice that action potentials literally travel within cells, rather than between them. Neurotransmitters are what travel between cells. This suggests a revision in the neuron doctrine: suppose that synapses are the real functional units—insofar as they are the ones sending and receiving the signals. In fact, such a revision may seem hardly worth mentioning if we think that synapses are special only because they allow neurons to do what makes neurons special—namely, propagate action potentials. It appears to be entirely compatible with the neuron doctrine.

2. Synapses as Functional Units

The idea of the synapse as the functional unit in the brain will seem like more of a departure if we reverse the usual order of priority. Suppose we come to think that what makes neurons special is just that they harbor synapses—and that without the synapse, the neuron is just another cell type in the brain, engaged in the usual biological drudgery. With this figure-ground shift, the brain starts to look different.

Imagine the brain not as a nest of intertwined neurons, but instead as a vast population of synapses, suspended in a cellular matrix, loosely allied into shifting confederations, connected by cellular highways that serve to conduct their signals, while also serving as supply routes that provide metabolic and material support for synaptic function.

Such a vision is not so far-fetched. The very term “interneuron,” used loosely to describe cells that form local connections (and in fact constitute the majority of neurons in the brain), suggests that the purpose of the neuron is to connect to others. The difference is that now we see the neuron as supporting the synapse, rather than the synapse serving the neuron.

In textbooks, the synapse is often portrayed as a mere glitch in the propagation of action potentials: evolution’s stopgap means of compensating for the handicap of the nervous system being made of discrete cells rather than some kind of reticular syncytium better suited to signal transmission. In the peripheral nervous system, there may be some truth to this—there we find synapses that seem to be specialized to transmit signals with high reliability.

The situation in the central nervous system is different. Firings do not reliably evoke firings there, and synapses display heterogeneous response profiles over time. Explaining the response of a synapse to any given signal requires appeal to the internal structure and workings of the synapse. One way to understand this lack of reliability is as suboptimality in the system. But another way is to see the heterogeneous synaptic response profiles as the consequence of functional filtering of incoming signals. This second interpretation would fit with the notion of the brain as processing (e.g., transforming, combining, filtering) signals in accordance with changing circumstances, rather than merely relaying or transmitting them. That is, synapses would be understood as internally sophisticated/differentiated mechanisms for processing signals.

Synapses also form respectably cohesive units. The typical synapse is composed of a presynaptic terminal (from one neuron), a postsynaptic terminal (from another neuron), and the astrocytic process that ensheathes the two. The three parts are stable only in this configuration—remove one part, and the others quickly disappear as well. When synapses are pruned, microglia remove the whole synapse as a unit. The parts are not found in isolation—terminals that don’t receive input tend to move away to more active territory, or else get reabsorbed.

The synapse is highly functionally specialized. Much of the regulation is locally controlled. Synapses have their own calcium reservoirs. Some protein synthesis happens at the synapse, as mRNAs are shipped to the synapse, not just fully formed proteins. Receptor cycling happens at the synapse, as does packaging of various signaling complexes that travel to the nucleus to effect plasticity at longer timescales.

Synapses are the target of drug action from antidepressants to stimulants to anesthetics. They are a major site of astrocytic modulation. On our best theories, synapses and synaptic change underlie learning and memory. Without the synapse, the neuron might still be electrically excitable, but it would not show adaptive plasticity or the ability to rewire itself. It would be more like a cardiac pacemaker cell, which also shows electric excitability but is not usually implicated in information processing or signaling.

Furthermore, the strength (and perhaps even survival) of synapses is to some extent determined locally. In addition to Hebbian and anti-Hebbian mechanisms, which change synaptic strength depending on whether the synapse receives input that is coordinated in the right way (or wrong way) with its neighbors on the same cell, there are a variety of other kinds of locally regulated synaptic plasticity (Turrigiano Reference Turrigiano2012). Which cell a synapse sits on is not always sufficient to determine its fate.

On the neurocentric view, the synapse is like the sensor that tells the cell what is happening in its vicinity. On the synapse-centric view, the cell is part of the environment of the synapse, shaped to serve its purposes.

3. Beyond the Neuron Doctrine

Not only are synapses better functional units than they might have seemed at first glance, but neurons are also less good than might have originally been supposed. Because they are spatially spread out, neurons are not always functionally cohesive. New research on glia also suggests that neurons are not as functionally unique as we might have thought.

On the first point, it has long been thought that functional individuation may diverge from cellular individuation (as argued, e.g., by London and Häusser Reference London and Häusser2005). Some processes are sufficiently remote (isolated electrotonically) from other processes on the same cell that they in effect operate independently. Conversely, gap junctions, while relatively rare in the mammalian nervous system, connect geographically adjacent compartments from distinct cells more effectively than distant parts of the same cell that are connected to each other.

But perhaps the main challenge to the neuron doctrine comes from findings about the important role of glia in the brain, especially astrocytes. Much of what neurons do is influenced by glial activity. It might seem that neurons are dramatically different from glia, given that the most strikingly distinctive feature of neurons is their ability to produce action potentials. But action potentials are specialized for fast, faithful transmission over large distances, whereas a great deal of functional activity in the brain happens at much smaller spatial scales and slower timescales. The point made earlier about synapses applies as well to whole cells: motor and sensory neurons in the peripheral nervous system are very reliable. Cortical neurons in the brain, by contrast, are much less so, and this is just what we should expect from processors of information, rather than mere relays of it.

What neurons most often do is modulate other cells. Modulation is important because, as noted earlier, processing information involves the transformation of signals, rather than merely copying and passing them along. Modulation is also important for the brain plasticity seen in learning, which happens at slower timescales. Glia turn out to be active players in the latter types of functional activity.

In the central nervous system, projection neurons—big pyramidal cells that connect different brain regions over relatively large distances (say, millimeters or centimeters)—form a small minority of the neurons in the mammalian brain. The remaining neurons are interneurons, mostly inhibitory, that exert a variety of local modulatory effects—that is, they change the timing and the responsiveness of the cell to excitatory input, at spatial scales orders of magnitude smaller than in the peripheral nervous system. Without modulatory neurons, it is very difficult to get neural networks (in simulation and in real life) to do anything interesting.

If, as we expect, neural variability is functional, then that functionality is the product of not just the activity of neuronal neighbors but also other local features of their extracellular environment. Being situated in the brain means that neurons are exposed to biophysical factors such as temperature, metabolic variables such as oxygen levels and extracellular osmolarity, and a vast array of signaling molecules, including nitric oxide, hormones, and neurotransmitters. Many of the signaling molecules that neurons respond to have nonneuronal sources. They come from glia, and they come from the bloodstream. That is, neurons are not the only source of modulation. The environment matters—unsurprisingly, since the importance of contextual effects is familiar from other areas of biology—and a key part of that environment is dictated by other (nonneuronal) cells.

For many years, glia were assumed to be mere structural and metabolic support for neurons. But more recently, they have been found to perform many interesting functions that are not metabolic—for example, they are involved in synaptic plasticity and modulate synaptic strength. Astrocytes are the best-studied class of glia. They have been shown in the past 10 years to share many functional similarities with neurons, despite not producing action potentials. Like neurons, they exhibit complex internal calcium dynamics that are not yet well understood. They are responsible for the reuptake of neurotransmitter at the synapse, which means that they control how long a synapse is active and how sensitive it is. They are necessary for hippocampal plasticity, which is currently thought to be a primary site for learning and memory. Like neurons, astrocytes are electrically polarized, and they release extracellular signaling molecules that affect both neurons and their own activity.

As complex and important as all these activities are, they do take place at a slower timescale than neural spiking. And so it might be argued that relative to neurons, glia still play only a supportive role. But to accept this is to deny the importance of modulation at timescales as rapid as a second—that is, it already assumes that the only signals fast enough to matter are action potentials, and that transmitting spikes is what the brain really does. But as suggested in the previous section, taking action potentials to be primary leads us away from a cell-centric picture. In fact, taking the synapse-centric view also helps to neatly accommodate many of the findings about the significance of astrocytes. This is because the synapse is a tripartite structure—owing its function in equal part to presynaptic, postsynaptic, and astrocytic components (Araque et al. Reference Araque, Parpura, Sanzgiri and Haydon1999; Haydon and Carmignoto Reference Haydon and Carmignoto2006; Allen and Barres Reference Allen and Barres2009).

The neuron doctrine leaves out a great deal of the brain that we know to be functionally relevant. The emerging scientific picture reminds us that the brain is composed of not just a network of neurons but a complex of many cells, including astrocytes, microglia, the endothelial and smooth muscle cells of the blood vessels, and oligodendrocytes. These cells share sensitivity to the same kinds of molecules, expressing many of the same receptors. All of these play a role, and in fact, the neurons are outnumbered by their nonneuronal neighbors. Given the relevance of so many other players, the rationale for prioritizing neurons as the functional units in the brain seems less clear.

4. The Function of Signaling

Let me now return to something that I have glossed over until now. What exactly is it to be a signaling unit? By what criteria should we evaluate the aptness of a candidate for the signaling role? If signaling is the function of the brain, and signaling is what it is to process information, and information processing is what makes the brain special, then we ought to have a clear idea of what signaling actually comes to.

Talk of information processing and signaling is deeply embedded in neuroscience, but it is also often vague and perhaps metaphorical. One reason to think that these notions are doing genuine work is that they highlight the communicative aspect of brain function—that is, that which enables parts of the brain to coordinate their activities with each other and the needs of the organism relative to its environmental situation. Use of the terms “information” and “signaling” both licenses and exemplifies the distinction already assumed in neuroscience between those aspects of the system that deal in information and those that merely engage in noninformational biological interactions. If we can understand these terms more precisely, we can better evaluate their theoretical role (if any).

To provide a formulation of what is distinctive about signaling networks that distinguishes them from causal networks more generally, I would like to introduce the framework of the Lewis-Skyrms signaling game (e.g., as described in Godfrey-Smith Reference Godfrey-Smith2012). In such a game, there are identified senders and receivers, with signals acting as intermediaries. Each signal corresponds to a range of possible states of the world, and depending on the state, one of a range of possible acts is produced by the receiver. When the receiver chooses the best act for the state of the world that actually obtains, both sender and receiver benefit, and not otherwise. What makes the signals informational is, first, that signals are a reliable indicator of relevant states of the world, and second, that the senders and receivers do better as a result of signaling than they would otherwise. The content of the signals being sent and received—“processed”—is then given by the states of the world in which the signal in question is appropriately sent.Footnote ¹ We might take “processing” itself to be the transformation of the content of a signal between subsequent pairs of senders.

How well do synapses fit the role of senders and receivers in the brain? For synapses to be good senders, they must be sensitive to their environment. They must be able to distinguish between a range of environmental states and send messages that differentiate between the states that evoke them. As synapses are sensitive to both incoming action potentials coming down the axon and modulatory signals from the environment, they seem like reasonable candidates for senders.

For synapses to be good receivers, they must act on their environment in a way that generates positive returns to both senders and receivers. Their acts include contributing to the generation of onward action potentials by producing a graded signal—either hyperpolarizing or depolarizing. In addition to their ionotropic contribution, synapses send a vast array of slower metabotropic signals that regulate activity both near the synapse itself and back at the cell body and nucleus. One way for synapses to generate positive returns is through their ability to command resources from astrocytes, which are both a direct source of energy sources such as lactate and raw materials such as neurotransmitters and an indirect means of obtaining metabolic support through their regulation of local blood flow.

Depending on which units in the brain we take to be the signalers, different kinds of world states will be indicated by the signals. This could be spelled out in different ways (e.g., Cao Reference Cao2012). From a local perspective, synapses pick up on whether an action potential is present or not at its presynaptic component. Neurons, on the other hand, pick up on the concentration of neurotransmitters and other modulatory molecules over many (∼10,000) distinct spatial areas, integrating over them in ways that depend precisely on their timing, relative location, and the actual morphological structure of the neuron. Similarly, most neurons in the central nervous system cause the release of neurotransmitters and other modulatory molecules over hundreds of active sites. At least for now, too little is known about astrocytic function to say whether the many processes of a single cell work together or separately to modulate neurotransmitter levels at different synapses attached to the same cell, or to what extent that modulatory function interacts with the astrocytic modulation of neurons at nonsynaptic sites.

Thus, synapses, neurons, and glia are responsive to different kinds of world states, individuated at different spatial and temporal scales. Their activities are correlated in complex ways, but whatever contents we ascribe to their respective signals will be quite different. If we think that we already have a good handle on what the contents of brain signals should be, that will place top-down theoretical constraints on which brain components should count as functional units. If contents are states of the world outside the brain, then it seems unlikely that single cells (of any kind) or synapses alone could be the right kinds of receivers for those signals. If contents are states of the world inside the brain, the action potential received by a synapse will carry very different information from the pattern of neurotransmitters received over a neuron’s dendritic arbor.

The Lewis-Skyrms signaling framework might push us toward pluralism about the kinds of signalers present. Perhaps there are many kinds of functional units in the brain. It also seems to imply some genuine indeterminacy about the content of any given signal. That is a surprising implication, given the determinate contents that are often assigned to patterns of activation by neurophysiologists (e.g., many examples in Rieke et al. Reference Rieke, Warland, van Steveninck and Bialek1999).

One objection to thinking of synapses as signalers is that even though synapses “act” in some minimal sense, they do not directly initiate action potentials. And inhibitory synapses act by damping down the effects of excitatory influences to prevent spiking that would otherwise have happened. In either direction, synapses change the conductance of the neural membrane through the opening of ion channels. When the synaptic potentials evoked are excitatory, it is ultimately the axon hillock that initiates the action potential after sufficient depolarization from a multitude of synapses accumulates. But one might also say that the synapse uses the axon hillock to send its signal, just as individual voters use the electoral system to send votes. It takes a lot of votes to make anything happen, but each voter is signaling nonetheless.

Another response is to revise the picture to a more complex (but perhaps more realistic and useful) one in which the cell body is a sender in its own right. Hence, we can envision two networks, sending signals to and through each other—a network of cell bodies that act as senders, paired with a network of synapses that act as receivers.

We might retreat further—perhaps synapses are not as good candidates to be either senders or receivers as we had originally hoped. But if we zoom back out to the cellular level at which neurons are signalers, sending transmitter molecules back and forth, then so too are glia, as well as many cells outside the nervous system. There will be nothing particularly distinctive about the particular mechanism of neural signaling, even though the functional role of neurons in the organism is still unique. Cells coordinate their activities throughout the body by means of ligands (of which neurotransmitters are just a subset), over both short and long distances. In that case, we might wonder whether the characterization of brain activity as signaling (and thus informational) is really the most apt way to understand the unique function of the brain, rather than a common function shared by many of the cells in a multicellular organism, and not particularly cognitive or special. Perhaps looking at the details will instead push us back toward a view of brain activity that is messier, richer, and more obviously biological than the relatively spare signaling model.

5. Conclusion

It might be useful to compare the neuron doctrine with another textbook staple, the central dogma of molecular biology, which says that information flows from DNA to RNA to protein. To the extent that there is an implicit “and that’s all” aspect to the story that it tells, the central dogma remains influential beyond what it explicitly says (even well after the discovery of retroviruses demonstrated that there were exceptions to the story that it told). It is not so hard to imagine how taking DNA to be the source helps elevate it to a kind of uncaused causer. It encouraged a reductive program that was perhaps most bluntly expressed in the early days of the Human Genome Project (HGP). For example, it was promised that HGP would provide a complete blueprint for humans in both health and disease.Footnote ² There were extensive critiques of such reductive gene-centrism (e.g., Oyama Reference Oyama1986; Griffiths Reference Griffiths2001). A decade after the conclusion of HGP, the rhetoric has changed—now it seems that it was never seriously supposed that a full DNA sequence would be sufficient to specify all the relevant features of the organism.

Returning to the neuron doctrine, if the brain’s highly specialized components are most aptly characterized as signalers, then the obvious conclusion is that brain function is just “signals running around a very complicated network” (Skyrms Reference Skyrms2010, 3). And indeed it is the received view that the brain is best understood as an information processing network, in which neurons are the functional units. This view draws much of its motivation from the appealing simplicity of the neuron doctrine.

We should examine the received view more carefully, to make sure that the motivating features come from what is happening in the brain, rather than an overly simplistic model of it. What empirical work in the past 20 years has shown is that accounting for nonneuronal contributions to brain function complicates the model of the brain as a signaling network.

One response is to hang on to the signaling model by privileging synapses (with both neuronal and nonneural components) as the functional units, acting as both senders and receivers. Another response might be to democratize functional participation, allowing astrocytes and other glia to be functional units as well. That would de-emphasize the action potential as the signal of interest, allowing other forms of interaction within and between cells in the brain.

A consequence of adopting the more inclusive model in which nonneuronal cells play key functional roles in the brain is that we might be pushed away from the original idea that the central function of the brain is to propagate signals. We may have independent reasons to suppose that information processing and signaling are a good model of brain function, but we cannot rely on the intuitive appeal of the neuron doctrine alone.

Instead, the brain may come to look more like the rest of the body—a diverse collection of biological cells exerting biological effects on each other via biological mechanisms that are not unique to the nervous system. In particular, the effects of these cells on each other (and their own activities) start to seem less efficiently informational and more messily biological. That is a relevant possibility, and one that deserves more attention.