Barry W. Connors
Neurons receive, combine, transform, store, and send information
Neurons have arguably the most complex job of any cell in the body. Consequently, they have an elaborate morphology and physiology. Each neuron is an intricate computing device. A single neuron may receive chemical input from tens of thousands of other neurons. It then combines these myriad signals into a much simpler set of electrical changes across its cellular membrane. The neuron subsequently transforms these ionic transmembrane changes according to rules determined by its particular shape and electrical properties and transmits a single new message through its axon, which itself may contact and inform hundreds of other neurons. Under the right circumstances, neurons also possess the property of memory; some of the information coursing its synapses may be stored for periods as long as years.
This general scheme of neuronal function applies to most neurons in the vertebrate nervous system. However, the scheme is endlessly variable. For example, each region of the brain has several major classes of neurons, and each of these classes has a physiology adapted to perform specific and unique functions. In this chapter, the general principles of neuronal function are outlined, and the almost unlimited variability contained within the general schema is discussed.
Neural information flows from dendrite to soma to axon to synapse
Numerous dendrites converge on a central soma, or cell body, from which a single axon emerges and branches multiple times (see Fig. 10-1). Each branch culminates at a presynaptic terminal that contacts another cell. In most neurons, dendrites are the principal synaptic input sites, although synapses may also be found on the soma, on the axon hillock (the region of the soma neighboring the axon), or even directly on the axons. In some primary sensory neurons, the dendrites themselves are transducers of environmental energy. Regardless of their source, signals—in the form of voltage changes across the membrane—typically flow from dendrites to soma to axon and finally to synapses on the next set of cells.
Excitatory input to a neuron usually generates an inward flow of positive charge (i.e., an inward current) across the dendritic membrane. Because the interior of a resting neuron is polarized negatively with respect to the external environment, this inward current, which makes the membrane voltage more positive (i.e., less negative), is said to depolarize the cell. Conversely, inhibitory input to a neuron usually generates an outward current and, thus, hyperpolarization.
If the neuron receives its input from a neighboring cell through a chemical synapse, neurotransmitters trigger currents by activating ion channels. If the cell is a sensory neuron, environmental stimuli (e.g., chemicals, light, mechanical deformation) activate ion channels and produce a flow of current. The change in membrane potential (Vm) caused by the flow of charge is called a postsynaptic potential (PSP) if it is generated at the postsynaptic membrane by a neurotransmitter and a receptor potential if it is generated at a sensory nerve ending by an external stimulus. In the case of synaptic transmission, the postsynaptic Vm changes may be either positive or negative. If the neurotransmitter is excitatory and produces a depolarizing PSP, we refer to the PSP as an excitatory postsynaptic potential (EPSP) (see Chapter 8). On the other hand, if the neurotransmitter is inhibitory and produces a hyperpolarizing PSP, the PSP is an inhibitory postsynaptic potential (IPSP). In all cases, the stimulus produces a Vm change that may be graded from small to large, depending on the strength or quantity of input stimuli (see Chapter 7). Stronger sensory stimuli generate larger receptor potentials; similarly, more synapses activated together generate larger PSPs. A graded response is one form of neural coding whereby the size and duration of the input are encoded as the size and duration of the change in the dendritic Vm.
The synaptic (or receptor) potentials generated at the ends of a dendrite are communicated to the soma, but not usually without substantial attenuation of the signal (Fig. 12-1A). Extended cellular processes such as dendrites behave like leaky electrical cables (see Chapter 7). As a consequence, dendritic potentials usually decline in amplitude before reaching the soma. As an EPSP reaches the soma, it may also combine with EPSPs arriving by other dendrites on the cell; this behavior is a type of spatial summation and can lead to EPSPs that are substantially larger than those generated by any single synapse (Fig. 12-1B, C). Temporal summation occurs when EPSPs arrive rapidly in succession; when the first EPSP has not yet dissipated, a subsequent EPSP tends to add its amplitude to the residual of the preceding EPSP (Fig. 12-1D).
Figure 12-1 Spatial versus temporal summation of excitatory postsynaptic potentials (EPSPs).
The tendency for synaptic and receptor potentials to diminish with distance along a dendrite puts significant limitations on their signaling abilities. If nothing else happened, these depolarizing potentials would simply dwindle back to the resting membrane potential as they spread through the soma and down into the axon. At best, this passive signal might be carried a few millimeters, clearly inadequate for wiggling a toe when the axon of the motor neuron stretching from the spinal cord to the foot might be 1000 mm long. Some amplification is therefore necessary for certain inputs to generate effective signals to and from the central nervous system (CNS). Amplification is provided in the form of regenerating action potentials. If the Vm change in the soma is large enough to reach the threshold voltage (see Chapter 7), the depolarization may trigger one or more action potentials between the soma and axon, as shown in Figure 12-1B to D. Action potentials are large, rapid fluctuations in Vm. As described in Chapter 7, an action potential is an efficient, rapid, and reliable way to carry a signal over long distances. However, notice that generation of action potentials entails another transformation of neuronal information: the neuron converts the graded-voltage code of the dendrites (i.e., the PSPs) to a temporal code of action potentials in the axon.
Action potentials are fixed in amplitude, not graded, and have uniform shape. So how is information encoded by action potentials? This question has no simple answer and is still hotly debated. Because one axonal spike looks like another (with slight exceptions), neurons can vary only the number of spikes and their timing. For a single axon, information may be encoded by the average rate of action potential firing, the total number of action potentials, their temporal pattern, or some combination of these mechanisms. Figure 12-1 illustrates that as the synaptic potential in the soma increases in size, the resultant action potentials occur more frequently, and the burst of action potentials in the axon lasts longer. Notice also that by the time the signal has propagated well down the axon, the transformation has become complete—the graded potential has waned and vanished, whereas the action potentials have retained their size, number, and temporal pattern. The final output of the neuron is entirely encoded in these action potentials. When action potentials reach axonal terminals, they may trigger the release of a neurotransmitter at the next set of synapses, and the cycle begins again.
SIGNAL CONDUCTION IN DENDRITES
The word dendrite is derived from the Greek word dendron for “tree,” and indeed some dendrites resemble the branches or roots of an oak tree. Inspired by trees, no doubt, the anatomist Camillo Golgi suggested in 1886 that the function of dendrites is to collect nutrients for the neuron. The truth is analogous but more interesting: dendrites arborize through a volume of brain tissue so that they can collect information in the form of synaptic input. The dendrites of different types of neurons exhibit a great diversity of shapes. Dendrites are often extensive, accounting for up to 99% of a neuron’s membrane. The dendrites of a single neuron may receive as many as 200,000 synaptic inputs. The electrical and biochemical properties of dendrites are quite variable from cell to cell, and they have a profound influence on the transfer of information from synapse to soma.
Dendrites attenuate synaptic potentials
Dendrites tend to be long and thin. Their cytoplasm has relatively low electrical resistivity, and their membrane has relatively high resistivity. These are the properties of a leaky electrical cable, which is the premise for cable theory(see Chapter 7). Leaky cables are like leaky garden hoses; if ionic current (or water) enters at one end, the fraction of it that exits at the other end depends on the number of channels (or holes) in the cable (hose). A good hose has no holes and all the water makes it through, but most dendrites have a considerable number of channels that serve as leaks for ionic current (see Fig. 7-22).
Cable theory predicts how much current flows down the length of the dendrite through the cytoplasm and how much of it leaks out of the dendrite across the membrane. As summarized in Table 7-3, we can express the leakiness of the membrane by the resistance per unit area of dendritic membrane (specific membrane resistance, Rm), which can vary widely among neurons. The intracellular resistance per cross-sectional area of dendrite (specific resistivity of the cytoplasm, Ri) is also important in determining current flow inasmuch as a very resistive cytoplasm forces more current to flow out across the membrane rather than down the axis of the dendrite. Another important factor is cable diameter; thick dendrites let more current flow toward the soma than thin dendrites do. Figure 7-22C illustrates the consequences of a point source of steady current flowing into a leaky, uniform, infinitely long cable made of purely passive membrane. The transmembrane voltage generated by the current falls off exponentially with distance from the site of current injection. The steepness with which the voltage falls off is defined by the length constant (λ; see Chapter 7), which is the distance over which a steady voltage decays by a factor of 1/e(~37%). Estimates of the parameter values vary widely, but for brain neurons at rest, reasonable numbers are ~50,000 Ω · cm2 for Rm and 200 Ω · cm for Ri. If the radius of the dendrite (a) is 1 μm (10−4 cm), we can estimate the length constant of a dendrite by applying Equation 7-8.
Because dendrite diameters vary greatly, λ should also vary greatly. For example, assuming the same cellular properties, a thin dendrite with a radius of 0.1 μm would have a λ of only 354 μm, whereas a thick one with a radius of 5 μm would have a λ of 2500 μm. Thus, the graded signal spreads farther in a thick dendrite.
Real dendrites are certainly not infinitely long, uniform, and unbranched, nor do they have passive membranes. Thus, quantitative analysis of realistic dendrites is complex. Sharp termination of a dendrite decreases attenuation because current cannot escape farther down the cable. Branching increases attenuation because current has more paths to follow. Gradually expanding to an increased diameter progressively increases λ and thus decreases attenuation. Real membranes are never completely passive because all have voltage-gated channels, and therefore their Rm values can change as a function of voltage. Finally, in the working brain, cable properties are not constant but may vary dynamically with ongoing brain activity. For example, as the general level of synaptic input to a neuron rises (which might happen when a brain region is actively engaged in a task), more membrane channels will open and thus Rmwill drop as a function of time, with consequent shortening of dendritic length constants. However, all these caveats do not alter the fundamental qualitative conclusion: voltage signals are attenuated as they travel down a dendrite.
So far, we have described only how a dendrite might attenuate a sustained voltage change. Indeed, the usual definition of length constant applies only to a steady-state voltage shift. An important complication is that the signal attenuation along a cable depends on the frequency components of that signal—how rapidly voltage changes over time. When Vm varies over time, some current is lost to membrane capacitance (see Chapter 6), and less current is carried along the dendrite downstream from the source of the current. Because action potentials and EPSPs entail rapid changes in Vm, with the fastest of them rising and falling within a few milliseconds, they are attenuated much more strongly than the steady-state λ implies. If Vm varies in time, we can define a λ that depends on signal frequency (λAC, where AC stands for alternating current). When signal frequency is zero (i.e., Vm is steady), λ = λAC. However, as frequency increases, λAC may fall sharply. Thus, dendrites attenuate high-frequency (i.e., rapidly changing) signals more than low-frequency or steady signals. Another way to express this concept is that most dendrites tend to be low-pass filters in that they let slowly changing signals pass more easily than rapidly changing ones.
Figure 12-2A shows how an EPSP propagates along two different dendrites with very different length constants: when the dendrites have a longer λ, a larger signal arrives at the axon hillock. How do leaky dendrites manage to communicate a useful synaptic signal to the soma? The problem is solved in two ways. The first solution deals with the passive properties of the dendrite membrane. The length (l) of dendrites tends to be relatively small in comparison to their λ; thus, none extend more than one or two steady-state length constants (i.e., the l/λ ratio is smaller than 1). One way that dendrites achieve a small l/λ ratio is to have a combination of diameter and Rm that gives them a large λ. Another way is that dendrites are not infinitely long cables but “terminated” cables. Figure 12-2B shows that a signal is attenuated more in an infinitely long cable (curve a) than in a terminated cable whose length (l) is equal to λ (curve b). The attenuation of a purely passive cable would be even less if the terminated cable had a λ 10-fold greater than l (curve c). Recall that in our example in Figure 12-2A, such a 10-fold difference in λ underlies the difference in the amplitudes of the EPSPs arriving at the axon hillock.
Figure 12-2 Effect of λ on propagation of an EPSP to two different axons. A, The neuron at the top fires an action potential that reaches the left and right neurons below, each at a single synapse. The EPSPs are identical. However, the left neuron has a thin dendrite and therefore a small length constant (λ = 0.1 mm). As a result, the signal is almost completely attenuated by the time it reaches the axon hillock, and there is no action potential. In the right neuron, the dendrite is thicker and therefore has a larger length constant (λ = 1 mm). As a result, the signal that reaches the axon hillock is large enough to trigger an action potential. B, The graph shows four theoretical plots of the decay of voltage (logarithmic plot) along a dendritic cable. The voltage is expressed as a fraction of maximal voltage. The length along the cable is normalized for the length constant (λ). Thus, an l/λ of 1.0 corresponds to one length constant along the dendrite. Curve a: If the cable is infinitely long and passive, the voltage decays exponentially with increasing length, so that the semilog plot is linear. Curve b: If the cable is terminated at a length that is equal to one length constant, then voltage decays less steeply. Curve c: If the cable is terminated at a length that is equal to 10% of the length constant, the voltage decays even less steeply. Curve d: If the membrane is not passive but has a slow voltage-gated conductance, the dendritic attenuation will be much smaller. (Data from Jack JJB, Noble D, Tsien RW: Electrical Current Flow in Excitable Cells. Oxford: Oxford University Press, 1975.)
The second solution to the attenuation problem is to endow dendrites with voltage-gated ion channels (see Chapter 7) that enhance the signal more than would be expected in a purely “passive” system (curve d). We discuss the properties of such “active” cables in the next section.
Dendritic membranes have voltage-gated ion channels
All mammalian dendrites have voltage-gated ion channels that influence their signaling properties. Dendritic characteristics vary from cell to cell, and the principles of dendritic signaling are studied intensively. Most dendrites have a relatively low density of voltage-gated channels (see Chapter 7) that may amplify, or boost, synaptic signals by adding additional inward current as the signals propagate from distal dendrites toward the soma. We have already introduced the principle of an active cable in curve d of Figure 12-2B. If the membrane has voltage-gated channels that are able to carry more inward current (usually Na+ or Ca2+) under depolarized conditions, a sufficiently strong EPSP would drive Vm into the activation range of the voltage-gated channels. These voltage-gated channels would open, and their additional inward current would add to that generated initially by the synaptic channels. Thus, the synaptic signal would fall off much less steeply than in a passive dendrite. Voltage-gated channels can be distributed all along the dendrite and thus amplify the signal along the entire dendritic length, or they can be clustered at particular sites. In either case, voltage-gated channels can boost the synaptic signal considerably, even if the densities of channels are far too low to generate action potentials.
An even more dramatic solution, used by a few dendrites, is to have such a high density of voltage-gated ion channels that they can produce action potentials, just as axons can. One of the best documented examples is the Purkinje cell, which is the large output neuron of the cerebellum. As Rodolfo Llinás and colleagues have shown, when the dendrites of Purkinje cells are stimulated strongly, they can generate large, sharp action potentials that are mediated by voltage-gated Ca2+ channels (Fig. 12-3). Such Ca2+ spikes can sometimes propagate toward—or even into—the soma, but these Ca2+ action potentials do not continue down the axon. Instead, they may trigger fast Na+-dependent action potentials that are generated by voltage-gated Na+ channels in the soma and initial segment. The Na+ spikes carry the signal along the axon in the conventional way, and those in the soma are considerably quicker and larger than the dendritic Ca2+ spikes. The faster Na+ spikes do not propagate too far backward into the dendritic tree because the rapid time course of the Na+ spike is strongly attenuated by the inherent filtering properties of the dendrites (i.e., the λAC is smaller for the rapid frequencies of the Na+ action potentials than for the slower Ca2+ action potentials). The dendrites of certain other neurons of the CNS, including some pyramidal cells of the cerebral cortex, can also generate spikes that are dependent on Ca2+, Na+, or both.
Figure 12-3 Ca2+ action potentials (in dendrites of Purkinje cells). Usually, dendrites do not fire action potentials; however, in these Purkinje cells of the cerebellum (left panel), the high density of voltage-gated Ca2+ channels in the dendrites allows the generation of slow dendritic Ca2+ spikes (records a, b,and c on the right), which propagate all the way to the axon soma. In the axon soma, these Ca2+ action potentials trigger fast Na+ action potentials (record d on the right). Moreover, the fast Na+ spikes back propagate into the dendritic tree but are attenuated. Thus, these fast Na+ spikes appear as small spikes in the proximal dendrites (record c) and even smaller blips in the midlevel dendrites (record b). (Data from Llinás R, Sugimori M: Electrophysiological properties of in vitro Purkinje cell dendrites in mammalian cerebellar slices. J Physiol 1981; 305:197-213.)
Dendritic action potentials, when they exist at all, tend to be slower and weaker than those in axons. The reason is probably that one of the functions of dendrites is to collect and to integrate information from a large number of synapses (often thousands). If each synapse were capable of triggering an action potential, there would be little opportunity for most of the input to have a meaningful influence on a neuron’s output. The cell’s dynamic range would be truncated; that is, a very small number of active synapses would bring the neuron to its maximum firing rate. However, if dendrites are only weakly excitable, the problem of signal attenuation along dendritic cables can be solved while still allowing the cell to generate an output (i.e., the axonal firing rate) that is indicative of the proportion of its synapses that are active.
Another advantage of voltage-gated channels in dendrites may be the selective boosting of high-frequency synaptic input. Recall that passive dendrites attenuate signals of high frequency more than those of low frequency. However, if dendrites possess the appropriate voltage-gated channels, they will be better able to communicate high-frequency synaptic input.
CONTROL OF SPIKING PATTERNS IN THE SOMA
Electrical signals from dendrites converge and summate at the soma. Although action potentials themselves often appear first at the nearby axon hillock and initial segment of the axon, the large variety of ion channels in the soma and proximal dendritic membranes is critically important in determining and modulating the temporal patterns of action potentials that ultimately course down the axon.
Neurons can transform a simple input into a variety of output patterns
Neurophysiologists have sampled the electrical properties of many different types of neurons in the nervous system, and one general conclusion seems safe: no two types behave the same. The variability begins with the shape and height of individual action potentials. Most neurons within the CNS generate action potentials in the conventional way, as described in Chapter 7 (see Fig. 7-4). Fast, voltage-gated Na+ channels generate a strong inward current that depolarizes the membrane from rest, usually in the range of −60 to −80 mV, to a peak that is usually between +10 and +40 mV. This depolarization represents the upstroke of the action potential. The Na+ channels then quickly inactivate and close, and certain K+ channels (often voltage-gated, delayed outward rectifier channels) open and thus cause Vm to fall and terminate the spike. However, many neurons have somewhat different spike-generating mechanisms and produce spikes with a range of shapes. Although a fast Na+ current invariably drives the fast upstroke of neuronal action potentials, an additional fast Ca2+ current can frequently occur and, if it is large enough, broaden the spike duration. The greatest variability occurs in the repolarization phase. Many neurons are repolarized by other voltage-gated K+ currents in addition to the delayed outward rectifier K+ current, and some also have a K+current carried by channels that are rapidly activated by the combination of membrane depolarization and a rise in [Ca2+]i (see Chapter 7). (See Note: Shapes of Action Potentials in Various Neurons)
More dramatic variations occur in the repetitive spiking patterns of neurons, observed when the duration of a stimulus is long. One way to illustrate this principle is to apply a simple, continuous stimulus (a current pulse, for example) to each neuron and to measure its output (the number and pattern of action potentials fired at its soma). The current pulse is the equivalent of a steady, strong input of excitatory synaptic currents. The transformation from stimulus input to spiking output can take many different forms. Some examples are shown in Figure 12-4, which illustrates recordings from three types of neurons in the cerebral cortex. In response to a sustained current stimulus, some cells generate a rapid train of action potentials that do not adapt (Fig. 12-4A); that is, the spikes occur at a regular interval throughout the current pulse. Other cells fire rapidly at first but then adapt strongly (Fig. 12-4B); that is, the spikes gradually become less frequent during the current pulse. Some cells fire a burst of action potentials and then stop firing altogether, and still others generate rhythmic bursts of action potentials that continue as long as the stimulus (Fig. 12-4C). These varied behaviors are not arbitrary but are characteristic of each neuron type, and they are as distinctive as each cell’s morphology. They are also an intrinsic property of each neuron; that is, a neuron’s fundamental firing pattern is determined by the membrane properties of the cell and does not require fluctuations in synaptic input. Of course, synaptic input may also impose particular firing patterns on a neuron. When a neuron is operating in situ, its firing patterns are determined by the interaction of its intrinsic membrane properties and synaptic input.
Figure 12-4 Spiking patterns.
Rhythmically bursting cells are particularly interesting and occur in a variety of places in the brain. As described in Chapter 16, they may participate in the central circuits that generate rhythmic motor output for behavior such as locomotion and respiration. Cells that secrete peptide neurohormones, such as the magnocellular neurons of the hypothalamus, are also often characterized by rhythmic bursting behavior. These cells release either arginine vasopressin or oxytocin and help control water retention and lactation, respectively (see Chapters 40 and 56). Rhythmic bursting is a more effective stimulus for the synaptic release of peptides than are tonic patterns of action potential. It may be that the bursting patterns and the Ca2+ currents that help drive them can elicit the relatively high [Ca2+]i necessary to trigger the exocytosis of peptide-containing vesicles (see Chapter 8). One additional role of rhythmically bursting neurons is to help drive the synchronous oscillations of neural activity in forebrain circuits (i.e., thalamus and cortex) during certain behavioral states, particularly sleep.
Although it has been difficult to prove, the diverse electrical properties of neurons are probably adapted to each cell’s particular functions. For example, in the first stage of auditory processing in the brain, cranial nerve VIII axons from the cochlea innervate several types of neurons in the cochlear nucleus of the brainstem. The axons provide similar synaptic drive to each type of neuron, yet the output from each neuron type is distinctly different. Some are tuned to precise timing and respond only to the onset of the stimulus and ignore anything else; some respond, pause, and respond again; others chop the ongoing stimulus into a more rhythmic output; and still other cells transform the input very little. The mechanisms for this range of transformations include both synaptic circuitry and the diverse membrane properties of each cell type. The different properties allow some cells to be particularly well tuned to specific features of the stimulus—its onset, duration, or amplitude modulation—and they can then communicate this signal to the appropriate auditory nuclei for more complex processing.
Intrinsic firing patterns are determined by a variety of ion currents with relatively slow kinetics
What determines the variety of spiking patterns in each type of neuron, and why do neurons differ in their intrinsic patterns? The key is a large set of ion channel types that have variable and often relatively slow kinetics compared with the quick Na+ and K+ channels that shape the spike. For a discussion of the properties of such channels, see Chapter 7. Each neuron expresses a different complement of these slow channels and has a unique spatial arrangement of them on its dendrites, soma, and axon initial segment. The channels are gated primarily by membrane voltage and [Ca2+]i, and a neuron’s ultimate spiking pattern is determined by the net effects of the slow currents that it generates. We illustrate three examples of systems that have been studied in detail.
1. A neuron with only fast, voltage-gated Na+ channels and delayed rectifier K+ channels will generate repetitive spikes when it is presented with a long stimulus. The pattern of those spikes will be quite regular over time, as for the cerebral cortical interneuron that we have already seen in Figure 12-4A.
2. If the neuron also has another set of K+ channels that activate only very slowly, the spiking pattern becomes more time dependent: the spiking frequency may initially be very high, but it adapts to progressively lower rates as a slow K+ current turns on to counteract the stimulus, as shown for the small pyramidal cell in Figure 12-4B. The strength and rate of adaptation depend strongly on the number and properties of the fast and slow K+ channels.
3. A neuron, by exploiting the interplay between two or more voltage-gated currents, can generate spontaneous rhythmic bursting—as in the case of the large pyramidal neuron in Figure 12-4C—even without ongoing synaptic activity to drive it.
Axons are specialized for rapid, reliable, and efficient transmission of electrical signals
At first glance, the job of the axon seems mundane compared with the complex computational functions of synapses, dendrites, and somata. After all, the axon has the relatively simple job of carrying the computed signal—a sequence of action potentials—from one place in the brain to another without changing it significantly. Some axons are thin, unmyelinated, and slow; these properties are sufficient to achieve their functions. However, the axon can be exquisitely optimized, with myelin and nodes of Ranvier, for fast and reliable saltatory conduction of action potentials over very long distances (see Chapter 7). Consider the sensory endings in the skin of your foot, which must send their signals to your lumbar spinal cord 1 meter away (see Fig. 10-11B). The axon of such a sensory cell transmits its message in just a few tens of milliseconds! As we see in our discussion of spinal reflexes in Chapter 16, axons of similar length carry signals in the opposite direction, from your spinal cord to the muscles within your feet, and they do it even faster than the sensory axons. Axons within the CNS can also be very long; examples include the corticospinal axons that originate in the cerebral cortex and terminate in the lumbar spinal cord. Alternatively, many central axons are quite short, only tens of micrometers in length, and they transmit their messages locally between neurons. The spinal interneuron between a sensory neuron and a motor neuron (see Fig. 10-11B) is an example. Some axons target their signal precisely, from one soma to only a few other cells, whereas others may branch profusely to target thousands of postsynaptic cells.
Different parts of the brain have different signaling needs, and their axons are adapted to the local requirements. Nevertheless, the primary function of all axons is to carry electrical signals, in the form of action potentials, from one place to other places and to do so rapidly, efficiently, and reliably. Without myelinated axons, the large, complex brains necessary to control warm, fast mammalian bodies could not exist. For unmyelinated axons to conduct action potentials sufficiently fast for many purposes, their diameters would have to be so large that the axons alone would take up far too much space and use impossibly large amounts of energy.
Action potentials are usually initiated at the initial segment
The soma, axon hillock, and initial segment of the axon together serve as a kind of focal point in most neurons. The many graded synaptic potentials carried by numerous dendrites converge at the soma and generate one electrical signal. During the 1950s, Sir John Eccles and colleagues used glass microelectrodes to probe the details of this process in spinal motor neurons. Because it appeared that the threshold for action potentials in the initial segment was only ~10 mV above the resting potential, whereas the threshold in the soma was closer to 30 mV above resting potential, they concluded that the neuron’s action potentials would be first triggered at the initial segment. Direct evidence now indicates that spikes may begin at the initial segment, at least for some neurons. EPSPs evoked in the dendrites propagate down to and through the soma and trigger an action potential in the initial segment (Fig. 12-5A). The action potential then propagates in two directions: forward (orthodromically) into the axon, with no loss of amplitude, and backward into the soma and dendrites, with strong attenuation, as we saw earlier in Figure 12-3. Orthodromic propagation carries the signal to the next set of neurons. The function of backward propagation is not completely understood. It is possible that backwardly propagating spikes trigger biochemical changes in the neuron’s dendrites and synapses, and they may have a role in synaptic plasticity. (See Note: Sir John Carew Eccles)
Figure 12-5 Simultaneous recording of action potentials from different parts of a neuron. A, In this hypothetical experiment, an excitatory synapse on a dendrite is stimulated, and the response near that dendrite is recorded in the soma and at the initial segment. The EPSP attenuates in the soma and the initial segment, but the EPSP is large enough to trigger an action potential at the initial segment. B, The threshold is high (−35 mV) in regions of the neuron that have few Na+ channels but starts to fall rather steeply in the hillock and initial segment. Typically, a stimulus of sufficient strength triggers an action potential at the initial segment. C, The density of Na+ channels is high only at the initial segment and at each node of Ranvier.
The axon achieves a uniquely low threshold in its initial segment (Fig. 12-5B) by packing voltage-gated Na+ channels at a remarkably high density (in comparison to the soma and dendrites) in the initial segment (Fig. 12-5C). The Na+ channel density in the axon initial segment is ~2000 channels/μm2, ~1000-fold higher than in the membrane of the soma and dendrites.
Conduction velocity of a myelinated axon increases linearly with diameter
The larger the diameter of an axon, the faster its conduction velocity, other things remaining equal. However, conduction velocity is usually far greater in myelinated axons than it is in unmyelinated axons (see Chapter 7). Thus, a myelinated axon 10 μm in diameter conducts impulses at about the same velocity as an unmyelinated axon ~500 μm in diameter. Myelination confers not only substantial speed advantages but also advantages in efficiency. Almost 2500 of the 10-μm myelinated axons can pack into the volume occupied by one 500-μm axon!
Unmyelinated axons still have a role in vertebrates. At diameters below ~1 μm, unmyelinated axons in the peripheral nervous system (PNS) conduct more rapidly than myelinated ones do. In a testament to evolutionary frugality, the thinnest axons of the peripheral sensory nerves, called C fibers, are ~1 μm wide or less, and all are unmyelinated. Axons larger than ~1 μm in diameter are all myelinated (Table 12-1). Every axon has its biological price: the largest axons obviously take up the most room and are the most expensive to synthesize and to maintain metabolically. The largest, swiftest axons are therefore used sparingly. They are used only to carry sensory information about the most rapidly changing stimuli over the longest distances (e.g., stretch receptors in muscle, mechanoreceptors in tendons and skin), or they are used to control finely coordinated contractions of muscles. The thinnest, slowest C fibers are mainly sensory axons related to chronic pain and temperature, for which the speed of the message is not as critical.
Table 12-1 Classes of Peripheral Sensory and Motor Axons, by Size and Conduction Velocity
The relationship between form and function for axons in the CNS is less obvious than it is in the periphery, in part because it is more difficult to identify each axon’s function. Interestingly, in the brain and spinal cord, the critical diameter for the myelination transition may be smaller than in the periphery. Many central myelinated axons are as thin as 0.2 μm. At the other extreme, very few myelinated central axons are larger than 4 μm in diameter.
The myelinated axon membrane has a variety of ion channels that may contribute to its normal and pathological function. Of primary importance is the voltage-gated Na+ channel, which provides the rapidly activating and inactivating inward current that yields the action potential. Nine isoforms of the α subunit of the voltage-gated Na+ channel exist (see Table 7-1). In normal central axons, it is specifically Nav1.6 channels that populate the nodes of Ranvier at a density of 1000 to 2000 channels/μm2 (Fig. 12-5). The same axonal membrane in the internodal regions, under the myelin, has fewer than 25 channels/μm2(versus between 2 and 200 channels/μm2 in unmyelinated axons). The dramatically different distribution of channels between nodal and internodal membrane has important implications for conduction along pathologically demyelinated axons (see later). K+ channels are relatively less important in myelinated axons than they are in most other excitable membranes. Very few of these channels are present in the nodal membrane, and fast K+ currents contribute little to repolarization of the action potential in mature myelinated axons. This diminished role for K+ channels may be a cost-cutting adaptation because the absence of K+ currents decreases the metabolic expense of a single action potential by ~40%. However, some K+ channels are located in the axonal membrane under the myelin, particularly in the paranodal region. The function of these K+ channels is unclear; they may set the resting Vm of the internodes and help stabilize the firing properties of the axon.
Demyelinated axons conduct action potentials slowly, unreliably, or not at all
Numerous clinical disorders selectively damage or destroy myelin sheaths and leave the axonal membranes intact but bare. These demyelinating diseases may affect either peripheral or central axons and can lead to severely impaired conduction (Fig. 12-6). The most common demyelinating disease of the CNS is multiple sclerosis (see the box titled Demyelinating Diseases), a progressive disorder characterized by distributed zones of demyelination in the brain and spinal cord. Among the demyelinating diseases of peripheral nerves is Landry-Guillain-Barré syndrome, which is an inflammatory disorder that may rapidly incapacitate but often ends in substantial recovery. The specific clinical signs of these disorders vary and depend on the particular sets of axons affected.
Figure 12-6 Demyelination in the CNS. A cross section of a spinal root from a dystrophic mouse shows amyelinated bundles of axons (thin borders) that are surrounded by myelinated fibers (thick borders). V, lumen of blood vessel. (From Rosenbluth J: In Waxman SG, Kocsis JD, Stys PK [eds]: The Axon: Structure, Function, and Pathophysiology. New York: Oxford University Press, 1995.)
In a normal, myelinated axon, the action currents generated at a node can effectively charge the adjacent node and bring it to threshold within ~20 μs (Fig. 12-7A) because myelin serves to increase the resistance and to reduce the capacitance of the pathways between the axoplasm and the extracellular fluid (see Chapter 7). The current flowing across each node is actually 5-fold to 7-fold higher than necessary to initiate an action potential at the adjacent node. Removal of the insulating myelin, however, means that the same nodal action current is distributed across a much longer, leakier, higher capacitance stretch of axonal membrane (Fig. 12-7B). Several consequences are possible. Compared with normal conduction, conduction in a demyelinated axon may continue, but at a lower velocity, if the demyelination is not too severe (Fig. 12-7B, record 1). In experimental studies, the internodal conduction time through demyelinated fibers can be as slow as 500 μs, 25 times longer than normal. The ability of axons to transmit high-frequency trains of impulses may also be impaired (record 2). Extensive demyelination of an axon causes total blockade of conduction (record 3). Clinical studies indicate that the blockade of action potentials is more closely related to symptoms than is the simple slowing of conduction. Demyelinated axons can also become the source of spontaneous, ectopically generated action potentials because of changes in their intrinsic excitability (record 4) or mechanosensitivity (record 5). Moreover, the signal from one demyelinated axon can excite an adjacent demyelinated axon and induce crosstalk (Fig. 12-7C), which may cause action potentials to be conducted in both directions in the adjacent axon.
Figure 12-7 Demyelination. A, Four action potentials at the leftmost node of Ranvier are conducted—unchanged in amplitude and frequency—to the nodes that are farther to the right. B, Action potentials propagate from a normally myelinated region into a demyelinated area. The axial action current that was generated at the last healthy node is distributed across a long region of bare or partially myelinated axonal membrane, with its decreased resistivity and increased capacitance. Thus, the current density at the two affected nodes is greatly reduced. This panel shows five consequences of demyelination: (1) decreased velocity, which is manifested as longer delays for the arrival of the train of spikes; (2) frequency-related block, which does not pass high-frequency trains—as a result, some of the spikes are missing distally; (3) total blockade; (4) ectopic impulse generation—even though there is no input in the proximal portion of the axon, action potentials arise spontaneously beyond the lesion; and (5) increased mechanosensitivity—ectopic action potentials arise by mechanical stimulation. C, If the demyelination affects two adjacent axons, action potentials in one (the top axon in this case) cause action potentials to propagate in both directions in the adjacent axon.
Multiple sclerosis (MS), the most common demyelinating disease of the CNS, affects more than 350,000 Americans alone. MS is an autoimmune disease directed against the myelin or oligodendrocytes and is thus a purely CNS disease. The trigger is unclear. Some have proposed a viral antigen (including canine distemper virus), but many other theories abound, none of which has been clearly proved. The viral antigen theory is based at least in part on classic studies that concluded that MS has a greater incidence in temperate regions than in the tropics and that one’s risk depends on where one spent one’s childhood, not adulthood. According to this view, the “incubation” period, or the delay between exposure to an as yet unspecified environmental factor and onset of the first attack, is 10 to 25 years. However, some contemporary neurologists doubt the statistical significance of the temperate/tropic data. No good evidence has been presented for a viral etiology yet. MS is twice as common in women than in men. Susceptibility to MS has also been linked to a number of genetic alleles of the HLA system.
The signs and symptoms of MS are protean, but common features include the following:
1. An optic neuritis may cause decreased visual acuity.
2. An internuclear ophthalmoplegia results in an inability of the eyes to adduct on lateral conjugate gaze, reflecting a lesion in the medial longitudinal fasciculus, a white matter tract. The result is double vision.
3. The Lhermitte sign is an electrical sensation that shoots down the back and into the legs when the neck is flexed.
Elevated IgG levels are typically found in the cerebrospinal fluid, and electrophoresis reveals oligoclonal bands in the IgG region. Magnetic resonance imaging can demonstrate the lesions in MS. Evoked potential studies, which measure nerve fiber conduction in various sensory pathways, are abnormal in almost all established cases and in about two thirds of suspected cases of MS.
MS is defined clinically as a disease that involves demyelination of the CNS with episodes separated in both time and space—at least two episodes of demyelination involving at least two separated regions of the CNS. Remissions and relapses are characteristic of many patients with MS. Thus, episodes of demyelination may be separated by intervals during which the symptoms and signs improve or even completely resolve. It is only the minority of patients who inexorably progress to the point of disability.
An exacerbation is due to the occurrence of active inflammation of a white matter tract in the CNS. A remission then occurs when the inflammation subsides and the axons that have now been demyelinated recover some of their function and are able to conduct action potentials through the area of myelin damage. However, the pathologically demyelinated fibers are not normal. Among the molecular changes that occur, Nav1.2 channels may replace Nav1.6 channels in demyelinated axons. Conduction is often barely adequate under normal circumstances and may become inadequate under stressful situations, such as illness, emotional stress, and exhaustion. Under these circumstances, symptoms will reappear. Such reappearance of symptoms must be distinguished clinically from a new exacerbation. In the reappearance of symptoms, it is an old area of damage that has become dysfunctional. In the new exacerbation, it is a new area of damage caused by new inflammation. The treatments of these two causes of symptoms are very different.
The reasons for remissions and exacerbations are not entirely clear. One explanation focuses on the impaired conduction in the pathologically demyelinated fibers. If conduction is just barely adequate under normal circumstances, it may become inadequate under certain stressful situations, such as illness, emotional stress, and exhaustion; only then will symptoms become evident. A change in temperature is a classic example of circumstances changing conduction. Conduction through demyelinated axons is exquisitely sensitive to temperature, but the direction of the effect is counterintuitive: conduction is “safer” at low temperature. Lower temperatures slow the gating kinetics of Na+ channels, thereby lengthening the duration of the action potentials and slowing conduction velocity. A decrease of just 2°C increases the duration of an action potential by ~20%. The augmented currents of the broader spike are better able to charge the increased capacitance of the axonal membrane of a demyelinated region, thus making it more likely that the threshold for continued conduction will be reached. When the body temperature of a patient with MS is elevated, for example, by fever or immersion in a hot bath, conduction is compromised and neurological deficits may appear. Conversely, the slightly increased impulse duration gained by cooling of the body may be sufficient to restore conduction through a demyelinated region. Thus, MS patients may experience improved nerve function in cold versus hot environments. These temperature effects on symptoms are explained nicely by the physiology of neural conduction and probably account for evanescent changes in symptoms in many patients. However, the temperature effects do not explain exacerbations and remissions, which are due to appearance and disappearance of inflammation. (See Note: Temperature Dependence of Axonal Conduction)
Certain drugs can also prolong the duration of action potentials and facilitate conduction through demyelinated axons. Demyelination may expose the voltage-gated K+ channels, which are normally hidden under the myelin. Activation of outwardly rectifying K+ channels can further impair spike production in demyelinated axons. However, the K+ channels also become accessible to pharmacological agents, such as the class of K+ channel blockers called aminopyridines. Because fast K+ channels may contribute to the repolarization phase of an action potential, blocking of these K+ channels can prolong the spike and facilitate its propagation through demyelinated regions. Clinical trials of aminopyridines showed some symptomatic relief in patients with MS, but the drugs have not proved effective enough for routine use.
At present, standard treatment of MS exacerbations includes immunosuppressive agents. The most common is interferon beta, which is one of the class of interferons that have antiviral, antiproliferative, and immune-modulating activity. This drug reduces the number and severity of exacerbations in patients with mild or moderate relapsing and remitting MS and may also help patients with chronic progressive MS. Many new drugs are currently under investigation to help prevent the autoimmune attack, to reduce the damage during an attack, or to improve function in already demyelinated regions.
Another demyelinating disease of the CNS is progressive multifocal leukoencephalopathy, a fatal illness common in patients with acquired immunodeficiency disease and caused by opportunistic infection of the oligodendroglia (myelin-producing cells) by a strain of human papovavirus. Another disorder, central pontine myelinolysis, is caused by too rapid correction of hyponatremia; the sudden osmotic fluctuations appear to produce demyelination.
The most common demyelinating disease of the PNS is Landry-Guillain-Barré syndrome. After a respiratory or other viral infection, an ascending neurological syndrome develops that is characterized by the onset of weakness, leading to paralysis of the legs and subsequent involvement of the hands and arms. In severe cases, the paralysis involves the nerves feeding the brainstem, and patients may lose the ability to swallow or even to breathe, which requires mechanical ventilation for a time. This initial stage of Landry-Guillain-Barré syndrome reaches a plateau in several weeks and then gradually resolves. Most patients recover completely as long as the physician recognizes the syndrome and initiates supportive treatment. Untreated, this syndrome is often fatal.
The pathological lesions of Landry-Guillain-Barré syndrome consist of lymphocytic infiltration of the myelin sheath produced by Schwann cells in the PNS. The result is segmental demyelination. These patients generally recover fully because the PNS has the ability to remyelinate itself. Axons of the CNS do not remyelinate to any significant extent, which explains why complete recovery in patients with MS is not as common as in these patients.
Books and Reviews
Ashcroft FM: Ion Channels and Disease. New York: Academic Press; 1999.
Connors BW, Gutnick MJ: Intrinsic firing patterns of diverse neocortical neurons. Trends Neurosci. 1990; 13:99-104.
Johnston D, Wu SM-S: Foundations of Cellular Neurophysiology. Cambridge, MA: MIT Press, 1994.
London M, Häusser M: Dendritic computation. Annu Rev Neurosci 2005; 28:503-532.
Waxman SG, Craner MJ, Black JA: Na+ channel expression along axons in multiple sclerosis and its models. Trends Pharmacol Sci. 2004; 25:584-591.
Zamvil SS, Steinman L: Diverse targets for intervention during inflammatory and neurodegenerative phases of multiple sclerosis. Neuron 2003; 38:685-688.
Huguenard JR, Hamill OP, Prince DA: Sodium channels in dendrites of rat cortical pyramidal neurons. Proc Natl Acad Sci USA 1989; 86:2473-2477.
Khaliq ZM, Raman IM: Axonal propagation of simple and complex spikes in cerebellar Purkinje neurons. J Neurosci 2005; 25:454-463.
Larkum ME, Zhu JJ, Sakmann B: Dendritic mechanisms underlying the coupling of the dendritic with the axonal action potential initiation zone of adult rat layer 5 pyramidal neurons. J Physiol 2001; 533:447-466.
Llinás R, Nicholson C: Electrophysiological properties of dendrites and somata in alligator Purkinje cells. J Neurophysiol 1971; 34:534-551.
Magee JC, Johnston D: Synaptic activation of voltage-gated channels in the dendrites of hippocampal pyramidal neurons. Science 1995; 268:301-304.
Swensen AM, Bean BP: Ionic mechanisms of burst firing in dissociated Purkinje neurons. J Neurosci 2003; 23:9650-9663.