Structural tuning and plasticity of the axon initial segment in auditory neurons

Abstract

The axon initial segment (AIS) that separates axonal and somato-dendritic compartments is a highly specialised neuronal structure enriched with voltage-gated Na⁺ channels and functions as the site of spike initiation in neurons. The AIS was once thought to be uniform and static in structure, but has been found to be organised in a manner specific to the function of individual neurons and to exhibit plasticity with changes in synaptic inputs. Such structural specialisations are found in the avian auditory system. In the nucleus magnocellularis (NM), which is involved in a precise relay of timing information, the length of the AIS differs depending on sound frequency and increases with decreasing frequencies to accommodate frequency-specific variations in synaptic inputs. In the nucleus laminaris, which integrates the timing information from both NMs for sound localisation, the length and the location of the AIS vary depending on sound frequency: AISs are shorter and more remote for higher frequency. Furthermore, the AISs of NM neurons elongate to increase their excitability when synaptic inputs are removed by cochlea ablation, suggesting their contribution to the homeostatic control of neural activity. These structural tunings and plasticities of the AIS are thus indispensable for the function of the auditory circuits in both normal and pathological conditions.

Hiroshi Kuba did his PhD with Hrunori Ohmori in Kyoto University. Following an academic career in Kyoto University, where he was also a visiting fellow of Larry Trussell's lab in Oregon Health Science University, he became a professor in the Cell Physiology department at Nagoya University in 2011. He is interested in understanding the neural mechanisms of sound localisation, and studying synaptic integration and neuronal excitability control in brainstem auditory circuits using in vitro patch-clamp recording, multi-photon imaging, computer simulation, and immunohistochemistry.

Introduction

The axon initial segment (AIS) is a non-myelinated region of the axon that separates the axonal and somato-dendritic compartments (Palay et al. 1968) and contributes to the initiation of action potentials in neurons (Araki & Otani, 1955; Coombs et al. 1957; Khaliq & Raman, 2006; Palmer & Stuart, 2006; Meeks & Mennerick, 2007; Shu et al. 2007a; Atherton et al. 2008; Kole et al. 2008; Fleidervish et al. 2010; Foust et al. 2010; Schmidt-Hieber & Bischofberger, 2010; Popovic et al. 2011). The function of the AIS is primarily achieved by a molecular complex composed of various proteins, including cytoskeletal and membrane scaffold proteins, cell adhesion molecules and ion channels (for review, see Ogawa & Rasband, 2008). One of these proteins, the voltage-gated Na⁺ (Nav) channel, is specifically anchored at this site by ankyrinG, lowers the threshold of action potentials and is a key contributor to the electrogenesis of the AIS (Kordeli et al. 1995; Zhou et al. 1998; Jenkins & Bennett, 2001; Pan et al. 2006; Dzhashiashvili et al. 2007; Hedstrom et al. 2008). In addition, the small diameter of the AIS and its segregation from the soma minimise the conductive and capacitive load, further promoting the efficacy of spike initiation at this site.

Recently, it was proposed that the AIS is more than a site of spike initiation and is involved in a delicate control of neuronal output that includes the temporal patterns, numbers, sizes and shapes of action potentials (for review, see Bean, 2007). Indeed, the AIS is variable in its composition and the localisation of its ion channels among different neuronal types (Van Wart et al. 2007; Lorincz & Nusser, 2008; for review, see Clark et al. 2010). More strikingly, it has been suggested that the AIS also varies in its structure, including length and location, in a cell-specific manner and can be modulated by synaptic activity. The structure of the AIS has a strong impact on neuronal excitability, suggesting the importance of these structural specialisations of the AIS to defining the function of individual neurons.

Because several detailed reviews have recently appeared on the molecular composition and signal processing of the AIS (see Rasband, 2010; Debanne et al. 2011; Bender & Trussell, 2012; Kole & Stuart, 2012), in this review I will focus on the structure of the AIS and give particular attention to the findings in the avian auditory system. I will show that the structure of the AIS is nicely linked to the specific function of each neuron, in terms of temporal coding, sound source localisation, and the homeostatic control of neuronal activity. First, tuning-frequency-dependent variations in AIS structure and their significance in auditory function are shown. Second, the homeostatic nature of the AIS in these neurons is described. Through the discussion, I will emphasise how the structure of the AIS is tuned to synaptic inputs and optimises the function of neural circuits.

Structural tuning of the AIS

The importance of structural differentiation in shaping the signal processing of neurons has been well documented in dendrites (for review, see Gulledge et al. 2005). However, these differentiations are not limited to dendrites; they also occur in the AIS. Here, as unique examples of such structural tuning of the AIS, I introduce the AISs of the nucleus magnocellularis (NM) and the nucleus laminaris (NL) of chickens; these nuclei are the second and third order nuclei in the avian auditory system, respectively (Fig. 1A; Kuba et al. 2006; Kuba & Ohmori, 2009). Importantly, neurons in these nuclei are well defined in terms of their synaptic inputs and physiological roles in the auditory system (see below for details). Furthermore, they respond to a specific frequency of sound (characteristic frequency, CF) and are arranged tonotopically (Fig. 1B) from the rostro-medial high-frequency region (CF of ∼4 kHz) to the caudo-lateral low-frequency region (CF of ∼0.2 kHz) in chickens (Rubel & Parks, 1975). This prior knowledge base enables us to interpret the roles of the AIS in relation to the information and function of the neurons.

Figure 1. The avian auditory system.

A, coronal section of the chicken brainstem. In birds, auditory signals are carried by the VIIIth nerve to the nucleus magnocellularis (NM) and relayed bilaterally to the nucleus laminaris (NL) where interaural time differences are computed for azimuthal sound localisation. B, topographic representation of sound frequency (tonotopy) in the auditory pathway of chickens. Neurons with high characteristic frequencies (CFs) are located rostro-medially, while those with low CFs are located caudo-laterally in the NM and NL (Rubel & Parks, 1975). R, rostral; L, lateral; CG, cochlea ganglion. Note that NM neurons are adendritic, while NL neurons differ in the number and length of dendrites along the tonotopic axis; dendritic numbers decrease, and lengths increase, toward low CFs (Jhaveri & Morest, 1982; Smith & Rubel, 1979). C, the temporal information of sound is represented by the timing of spikes that are precisely locked to a specific phase of the sound (phase lock) in the auditory pathway.

AIS tuning in NM neurons

NM neurons receive synaptic inputs from the auditory nerve, generate precisely timed spikes at a specific phase of the sound (Fig. 1C) and thereby convey temporal information about the sound to the NL (for review, see Trussell, 1999). This precision is achieved in part by distinct patterns of synaptic inputs and the geometry of the AIS (Fig. 2). In particular, the timing fluctuations (jitter) of auditory nerve activities increase ∼3-fold at low CFs compared to high CFs (Hill et al. 1989; Köppl, 1997; Fukui et al. 2006); NM neurons compensate for this variation by changing the pattern of synaptic inputs as a function of CF (Carr & Boudreau, 1991; Köppl, 1994; Fukui & Ohmori, 2004). High-CF NM neurons receive only a few end-bulb terminals, each of which mediates a sufficiently large synaptic input to drive a spike. Thus, high-CF NM neurons exhibit high-fidelity, one-to-one synaptic transmission that preserves the presynaptic spike timing. On the other hand, low-CF NM neurons receive small synaptic inputs from multiple bouton terminals and integrate these inputs to generate spikes, which reduces the jitter of postsynaptic spiking and compensates for the large presynaptic jitter at low CFs. This is because having a large number of individually weak inputs allows postsynaptic spike firing to follow the average timing of a population and minimises the influence of any one aberrantly timed input (Joris et al. 1994).

Figure 2. Tonotopic specialisations of AIS in NM.

A, distribution of Nav channels (red) at the AISs (arrowheads) of high- and low-CF NM neurons (from Kuba & Ohmori, 2009). B, scatter plot of the distance to length relationship of AISs in the NM showing that the AISs become longer as the CF of neurons decreases. C, CF-specific specialisations of synaptic inputs and AIS geometry in the NM. NM neurons with high-CFs receive a large synaptic input from an end-bulb terminal to drive a spike, while those with low-CFs summate temporally dispersed small inputs from multiple bouton terminals. The bottom box shows schematic drawings of excitatory postsynaptic potentials (EPSPs) in the NM. The slow rising phase of converging EPSPs in low-CF NM neurons (blue) inactivates Na⁺ current at the AIS, while the rising phase is rapid for the giant EPSP in high-CF NM neurons (red) (see text for details).

To accommodate these specialisations of synaptic inputs, NM neurons differ in the length of the AIS depending on the CF but do not differ with respect to the location of the AIS (10 μm from the soma); the length of the AIS is 2-fold greater in low-CF NM neurons (20 μm) than in high-CF NM neurons (10 μm), implying that the low-CF NM neurons have a greater accumulation of Nav channels at the AIS (Fig. 2; Kuba & Ohmori, 2009). This is a highly strategic difference because large Na⁺ currents enable the low-CF NM neurons to compensate for the Na⁺ current inactivation that occurs during the slow depolarisation of summed synaptic potentials. In contrast, a reduced Na⁺ current is clearly able to generate a spike with high precision in response to the single giant input in the high-CF NM neurons. Thus, the length of AIS changes with the number and size of synaptic inputs and ensures reliable and precise firings in NM neurons. Notably, similar tuning-frequency-dependent enhancement of temporal coding is reported in the spherical bushy cells of the mammalian anteroventral cochlear nucleus (Joris et al. 1994), a homologue of avian NM neurons, suggesting that corresponding variations in the synaptic input patterns and structure of the AIS may also exist in the mammalian auditory system.

AIS tuning in NL neurons

NL neurons perform a different computation; they are the coincidence detectors for synaptic inputs projecting from the left and right NM. These neurons are also tonotopically arranged and play a critical role in computing interaural time differences (the differences in the arrival time of sounds between the two ears) for the physiological mechanism of sound localisation (for reviews, see Konishi, 2003; Kuba, 2007). The AISs of NL neurons vary in location relative to the soma and length as a function of CF (Kuba et al. 2006); the location is farther from the soma (up to 80 μm) and the length shorter (10 μm) in high-CF NL neurons, while the AISs of low-CF neurons are close (10 μm) and long (25 μm) (Fig. 3). Thus, the distance of the AIS from the soma is negatively correlated with the length of the AIS in NL neurons. Computer simulations have revealed that this correlation is optimal for achieving high neuronal excitability throughout the tonotopic axis; a large distance reduces the effects of the large capacitive and conductive loads of the soma on the AIS, thereby enabling short AISs to elicit spikes with small Na⁺ conductances. Then, why are short and distant AISs rather than the long and close AISs present in high-CF NL neurons?

Figure 3. Tonotopic specialisations of the AIS in the NL.

A, distribution of Nav channels (red) in the AISs (arrowheads) of high- and low-CF NL neurons (from Kuba et al. 2006). AISs are shorter and farther from the soma in high-CF NL neurons. B, scatter plot showing a negative correlation between the distance from the soma and the length of AISs in the NL. C, CF-specific specialisations of synaptic inputs and AIS geometry in the NL. NL neurons receive multiple small bouton terminals irrespective of CF, but the frequency of synaptic inputs is much higher in neurons with higher CFs (see text for details). The bottom box shows schematic drawings of EPSPs in the NL. Temporal summation of EPSPs, which could inactivate Na⁺ current, occurs in high-CF NL neurons (red) but not in low-CF NL neurons (blue). To prevent this from occurring, the AIS is moved away from the soma in high-CF NL neurons.

This question can be answered by taking into account the CF-specific patterns of synaptic inputs to the NL. NL neurons are innervated by multiple NM neurons, whose activities are precisely phase-locked to stimulus sounds, implying that the frequencies of synaptic inputs to each NL neuron are within a narrow bandwidth that corresponds to its CF (Slee et al. 2010; Funabiki et al. 2012). Accordingly, high-CF NL neurons receive high-frequency synaptic inputs (up to ∼4 kHz in chickens) that are subject to temporal summation and cause a large depolarisation at the soma. This depolarisation inactivates Na⁺ currents and decreases the membrane excitability of the AISs located close to the soma. Thus, distant locations of AISs are particularly important for high-CF NL neurons to maintain their excitability by attenuating depolarisation through an electrotonic decrement along the axon. In contrast, as temporal summation is not large for low-frequency synaptic inputs, the closer locations of the AISs are effective in low-CF NL neurons. Thus, the location and the length of the AIS nicely match the frequency of synaptic inputs, ensuring accurate coincidence detection in NL neurons.

Similar cell-type-specific variations in the location of the AIS have been found in retinal ganglion neurons; the AISs of direction-selective cells are more distally located than those of other types of cells (Fried et al. 2009); this finding may also support the idea that the distribution of AISs is optimally tuned to the specific function of in each neuron.

Types and localisation of ion channels in the AIS of NM and NL neurons

The type and localisation of ion channels in the AIS are also critical in defining the efficacy of spike generation. Multiple ion channels, such as Nav channels, and voltage-gated K⁺ (Kv) and Ca²⁺ (Cav) channels, are found at the AISs of different types of neurons (for reviews, see Debanne et al. 2011; Bender & Trussell, 2012; Kole & Stuart, 2012). Among these, Nav1.6 and Kv1.2 have been identified at the AIS of NM and NL neurons (Kuba et al. 2006; Kuba & Ohmori, 2009). These channels are the most common subtypes of Nav and Kv channels at the AIS (Dodson et al. 2002; Inda et al. 2006; Kole et al. 2007; Shu et al. 2007b; Van Wart et al. 2007; Duflocq et al. 2008; Goldberg et al. 2008; Lorincz & Nusser, 2008, 2010; Royeck et al. 2008; Hu et al. 2009; Kress et al. 2010). Nav1.6 has a negative voltage dependence for activation and is therefore effective in lowering the spike threshold (Colbert & Pan, 2002; Kole et al. 2008), while the rapid and low-threshold activation of Kv1.2 can accelerate membrane responses at the AIS (Dodson et al. 2002; Kole et al. 2007; Shu et al. 2007b). Thus, the expression of these channels should be particularly crucial for ensuring the reliable and precise spike generation of these neurons. So, are there any variations in the expression of these channels within the AIS and along the tonotopic axis?

It is well demonstrated that, in various neurons, ion channels at the AIS show distinct cell-specific patterns of expression that include their types and localisation and characterise the signal processing of individual neurons (for reviews, see Clark et al. 2010; Debanne et al. 2011; Bender & Trussell, 2012; Kole & Stuart, 2012). However, this is not the case in NM and NL neurons; Nav1.6 and Kv1.2 are uniformly distributed within the AIS irrespective of CF (Kuba et al. 2006; Kuba & Ohmori, 2009), suggesting that the AIS is structurally, but not biophysically, differentiated along the tonotopic axes of these nuclei. Nevertheless, further studies on the expressions of other channels at the AIS are necessary to support this conclusion.

Structural plasticity of the AIS

The tuning of AIS structure to the patterns of synaptic inputs in NM and NL neurons suggests that the AIS structure might be re-tuned when the synaptic inputs to a neuron change. This re-tuning has indeed been found in chicken NM neurons.

AIS plasticity in NM neurons

In the NM, deprivation of synaptic inputs by removal of the cochlea increases the length of AISs in the deprived side by more than 50%, while other features of the AIS, such as the density of Nav channels and the location and width of the AIS, are not altered (Fig. 4; Kuba et al. 2010). Notably, auditory deprivation abolished the tonotopic variation in the length of AISs that is normally observed in the NM, supporting the notion that synaptic activity is crucial for the structural tuning of AISs. This increase in AIS length results from the reduction of presynaptic activity rather than damage to the cochlea itself because attenuation of sound conduction at the tympanic membrane or columella (middle ear bone) also causes elongation that parallels the level of attenuation. Auditory deprivation increases the amplitude of Na⁺ currents in the axon without changing their properties, such as the voltage dependence of inactivation, TTX sensitivity, or the persistent component, suggesting that elongation of the AIS increases the number of Nav channels of the same subtype. Accordingly, the spike threshold is reduced, and spontaneous firing appears in some neurons after auditory deprivation. Thus, auditory deprivation increases the length of the AIS, thereby enhancing the excitability of NM neurons. These findings indicate that elongation of the AIS most likely works as a homeostatic mechanism that compensates for the loss of presynaptic activity and may maintain the activity of central auditory circuits after hearing loss (Marder & Prinz, 2002; Turrigiano & Nelson, 2004). Importantly, elongation of the AIS has been observed in mature animals, including one-month-old chickens, suggesting that the homeostatic regulation of the AIS is not limited to the developmental period.

Figure 4. Homeostatic plasticity of the AIS in the NM.

A, auditory deprivation elongates the distribution of AISs (arrowheads) in the high-CF region of the NM (from Kuba et al. 2010). B, scatter plot showing an increase in the length of AIS without changes in the distance of the AIS from the soma after auditory deprivation. C, homeostatic regulation of neuronal activity in avian auditory circuits. Deprivation of auditory inputs elongates the AIS and increases the excitability of NM neurons, which may restore the firing activity in the circuit.

Cell-type specific phenotypes of AIS plasticity

Plasticity of the AIS occurs as changes in the location as well as the length of the AIS. In hippocampal pyramidal neurons in culture, an increase in neuronal activity causes a distal shift of the AIS, which reduces the excitability of the neurons (Grubb & Burrone, 2010). Interestingly, plastic changes were not observed at the AISs of inhibitory interneurons in the same preparation, indicating that AIS plasticity occurs in a cell-type-specific manner, and its phenotype differs depending on neuronal type and brain region.

Plasticity of the AIS has also been suggested in other pathological conditions. The length of the AIS increases in a mouse model of Angelman syndrome (Kaphzan et al. 2011), while ischaemic injury disrupts AIS structure through proteolysis of cytoskeletal proteins, which may contribute to the prevention of over-excitation of neurons in an ischaemic region (Schafer et al. 2009).

Mechanism of AIS plasticity

It is well known that the induction of synaptic plasticity is rapid, occurring within minutes (Bliss & Lomo, 1973). In contrast, AIS plasticity occurs over days (Grubb et al. 2011). This slow time course of AIS plasticity might be due to the process of reorganising the protein complex that composes the AIS, including cytoskeletal and cell adhesion molecules (Ogawa & Rasband, 2008). Shifts in AIS location are sensitive to blockers of T-type and L-type Cav channels, indicating an involvement of [Ca²⁺]_i in the process (Grubb & Burrone, 2010). Location shifts of the AIS should require modulation of both proximal and distal ends of the AIS, while modulation limited to the distal end should be sufficient for elongation of the AIS. Therefore, interesting remaining questions include whether the shift and the elongation of the AIS share the same mechanisms that involve [Ca²⁺]_i and what determines the phenotype of AIS plasticity.

The molecular mechanisms downstream of Ca²⁺ are not known, but they might involve phosphorylation of proteins. In cultured hippocampal neurons, protein kinase CK2 (casein kinase 2) increases the affinity of Nav1 to ankyrinG and promotes the accumulation of Nav channels at the AIS (Bréchet et al. 2008), while protein kinase Cdk5 (cyclin-dependent kinase 5) promotes the targeting of Kv1 channels to the axonal membrane (Vacher et al. 2011). Recently, Cdk5 was found to dose-dependently shorten the length of an AIS-like structure in Drosophila (Trunova et al. 2011). Whether similar mechanisms are operating in vertebrates is an important issue to be examined.

Perspective

I have shown that in the NM the length of the AIS varies depending on the number and size of synaptic inputs, and this variation improves temporal coding. In contrast, in the NL, both the location and the length of the AIS differ depending on the frequency of synaptic inputs, and these differences ensure accurate coincidence detection. These observations indicate that the distribution of AISs is effectively coupled with synaptic inputs to optimise the function of neural circuits. Further studies of the structural specialisations of the AISs in other neurons will facilitate an understanding of the roles of the AIS in neuronal computation. Whether similar specialisations exist in the mammalian system is also an important issue to be examined.

I have also shown that, in NM neurons, the AIS is the site of structural plasticity and contributes to the homeostatic regulation of neuronal activity. However, many questions remain regarding the characteristics, mechanisms and significance of this AIS plasticity: do the densities and types of channels other than Nav change during plasticity? Does AIS plasticity occur in NL neurons? If so, what is the phenotype of this plasticity? Is the myelin adjacent to the AIS affected during this plasticity? What is the role of the homeostatic regulation of neuronal activity by this plasticity? The AIS plasticity that has been found thus far occurs only in pathological conditions in vivo (Schafer et al. 2009; Kuba et al. 2010; Kaphzan et al. 2011). Therefore, it is also important to determine whether AIS plasticity occurs in more physiological conditions.

As the AIS has a strong impact on neuronal activity and its deficit causes several neurological diseases (for reviews, see Rasband, 2010; Wimmer et al. 2010), unveiling the structural tuning and plasticity of the AIS will strengthen our knowledge of the regulation of the activity and function of neural circuits in both physiological and pathological conditions.

Glossary

AIS

axon initial segment

Cav channels

voltage-gated Ca²⁺ channels

Cdk5

cyclin-dependent kinase 5

CF

characteristic frequency

EPSPs

excitatory postsynaptic potentials

Kv channels

voltage-gated K⁺ channels

Nav channels

voltage-gated Na⁺ channels

NL

nucleus laminaris

NM

nucleus magnocellularis