SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Appendix

The site at which action potentials initiate within the terminal region of unmyelinated sensory axons has not been resolved. Combining recordings of nerve terminal impulses (NTIs) and collision analysis, the site of action potential initiation in guinea-pig corneal cold receptors was determined. For most receptors (77%), initiation mapped to a point in the time domain that was closer to the nerve terminal than to the site of electrical stimulation at the back of the eye. Guinea-pig corneal cold receptors are Aδ-neurones that lose their myelin sheath at the point where they enter the cornea, and therefore their axons conduct more slowly within the cornea. Allowing for this inhomogeneity in conduction speed, the resulting spatial estimates of action potential initiation sites correlated with changes in NTI shape predicted by simulation of action potentials initiating within a nerve terminal. In some receptors, more than one NTI shape was observed. Simulations of NTI shape suggest that the origin of differing NTI shapes result from action potentials initiating at different, spatially discrete, locations within the nerve terminal. Importantly, the relative incidence of NTI shapes resulting from action potential initiation close to the nerve termination increased during warming when nerve activity decreased, indicating that the favoured site of action potential initiation shifts toward the nerve terminal when it hyperpolarizes. This finding can be explained by a hyperpolarization-induced relief of Na+ channel inactivation in the nerve terminal. The results provide direct evidence that the molecular entities responsible for stimulus transduction and action potential initiation reside in parallel with one another in the unmyelinated nerve terminals of cold receptors.

Sensory encoding in the peripheral nerve terminals of sensory neurons comprises two distinct processes (Katz, 1950; Hodgkin, 1938): an initial transduction of the stimulus energy into a receptor current and the subsequent transformation of this current into an action potential discharge. For cold-sensitive neurons, transduction is likely to involve several molecular entities including the menthol-sensitive, cooling-activated transient receptor potential melastatin 8 (TRPM8) channel (McKemy et al. 2002; Peier et al. 2002; Reid et al. 2002), heat-activated two-pore domain K+ channels (e.g. TREK-1 K+ channels; Maingret et al. 2000; Reid & Flonta, 2001; Viana et al. 2002) and a Na+,K+-ATPase-mediated hyperpolarizing current that is proportional to temperature (Pierau et al. 1974). The receptor current that ensues is transformed into action potentials by voltage-gated Na+ channels, the kinetics and availability of which are also affected by temperature. The extent to which the processes of transduction and transformation interact electrically within the nerve terminal determines the encoding of the stimulus and is consequently of fundamental importance to an understanding of cold receptor function.

In principle, the site at which action potentials initiate is determined by the spatial distribution of voltage-gated Na+ channels in the nerve terminal. The conventional view is that the receptive region of axonal membrane at the terminal is devoid of voltage-gated Na+ channels and therefore that the site of sensory signal transduction and the site of action potential initiation are spatially separate. However, extracellular recordings suggest that action potentials may be initiated and conducted within the unmyelinated nerve ending of Pacinian corpuscle afferents (Hunt & Takeuchi, 1962; Ozeki & Sato, 1964; Nishi & Sato, 1966) and frog muscle spindle afferents (Ito et al. 1974), and this is supported by immunohistochemical evidence of Na+ channels in the nerve terminal region of the Pacinian corpuscle (Pawson & Bolanowski, 2002). Despite the apparent presence of Na+ channels in the receptive nerve terminal membrane, action potential initiation is nevertheless thought to occur at the first node of Ranvier (Loewenstein & Rathkamp, 1958) where the Na+ channel density is high (Conti et al. 1976; Neumcke & Stampfli, 1982) and the membrane capacity is reduced by Schwann cell myelination.

In the guinea-pig cornea, cold receptors are likely to be Aδ-neurones and extracellular recording from their nerve terminals has revealed that they possess both tetrodotoxin (TTX)-sensitive and TTX-resistant Na+ channels at densities sufficient to support a regenerative Na+ current in the terminal (Carr et al. 2002). However, in the cornea, Aδ-axons lose their myelination at the limbus where they enter the cornea and have a long unmyelinated axon termination, so it is unlikely that action potentials are initiated at their first node of Ranvier. At present there is no evidence of clustering of Na+ channels at sites along the unmyelinated axons within the cornea (Black & Waxman, 2002) nor of morphological specializations that would favour action potential initiation at specific axonal sites. This leaves the question as to where action potentials are initiated in unmyelinated nerve terminals unresolved.

In the present study, the initiation site of action potentials generating ongoing nerve terminal impulses (NTIs) recorded extracellularly from cold-sensitive neurones in the guinea-pig cornea was determined using collision analysis. Collision of the ongoing orthodromic action potentials with antidromic action potentials evoked by electrical stimulation at the back of the eye was used to map the site of origin of the ongoing activity. In the majority of cold receptors, ongoing action potentials were initiated at points that mapped close to the site of recording. Consistent with this finding, differences in the shape of ongoing and electrically evoked NTIs were comparable with those predicted by computer simulation for initiation sites electrotonically close to the site of recording. Indeed, in some cases, the shape of the ongoing NTIs could be simulated by action potential initiation within half a length constant of the sensory nerve terminal. In these cases, Na+ current at the nerve ending contributed to action potential genesis demonstrating directly that the process of sensory signal transduction and the regenerative processes producing action potentials exist in parallel with one another in cold receptor nerve terminals.

Methods

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Appendix

All experimental procedures conformed to the Australian National Health and Medical Research Council guidelines for animal experimentation and were approved by the University of New South Wales Animal Care and Ethics Committee.

Tissue preparation

Guinea-pigs of both sexes and in the weight range 200–400 g were used. Animals were anaesthetized with sodium pentobarbitone (100 mg kg−1i.p.) and killed by exsanguination. Both eyes were dissected free from their orbits and isolated along with a short length of optic nerve and associated ciliary nerves. Eyes were mounted in a recording chamber and superfused at 5 ml min−1 with physiological saline of the following composition (mm): Na+ 151; K+ 4.7; Ca2+ 2; Mg2+ 1.2; Cl 144.5; H2PO3 1.3; HCO3 16.3; glucose 7.8. This solution was gassed with 95% O2–5% CO2 to pH 7.25. Under control conditions, the temperature of the bathing solution was maintained at 31–32.5°C. A thermocouple placed in close apposition to the surface of the cornea was used to monitor bath temperature. The optic nerve and associated ciliary nerves were drawn into a suction-stimulating electrode. The ciliary nerves were stimulated electrically with a constant voltage stimulator (pulse width 0.1–0.5 ms, 5–30 V).

To record electrical activity from sensory nerve terminals, a glass-recording electrode (tip diameter ∼50 μm) filled with physiological saline was applied to the surface of the corneal epithelium with slight suction (Brock et al. 1998). An Ag–AgCl electrode in the recording chamber served as the indifferent electrode. Electrical activity was amplified (1000×, Geneclamp 500, Axon Instruments), filtered (high pass 1 Hz, low pass 10 kHz; 432 Wavetek), digitized at 20 kHz (Powerlab, ADInstruments) and stored to disk using the recording software Chart (ADInstruments). Recordings were only made from sites on the cornea where the NTIs were readily distinguished from the noise (approx. 10 μV peak to peak). At many sites on the corneal surface, electrically evoked or ongoing electrical activity was either absent or too small to be analysed.

Receptor identification

The data presented were collected at recording sites where the electrical activity originated from a single nerve terminal. At these sites, electrical stimulation of the ciliary nerves evoked a single all-or-none NTI at the site of recording and the orthodromically propagating action potentials producing ongoing NTIs could be collided with antidromically propagated, electrically evoked NTIs (see Fig. 1B). Only NTIs that were defined as originating in cold receptors were analysed (see Brock et al. 1998). Cold receptors had relatively high levels of ongoing NTI discharge (4–15 Hz) which occurred in a rhythmic manner and this activity was decreased by warming and increased by cooling the solution superfusing the cornea.

image

Figure 1. Determination of the site of action potential initiation in the time domain by collision analysis From their site of initiation in sensory nerve terminals, action potentials propagate both ortho- and anti-dromically (upper trace, A). The antidromic action potential was detected at the axon terminal as a nerve terminal impulse (NTI; centre trace, A) and used to trigger electrical stimulation of the ciliary nerves at the back of the eye at variable delay (B, 20 overlaid traces at each delay interval). The experimental configuration with electrical stimulation of the ciliary nerves at the back of the eye and recordings from nerve terminals at the corneal surface is shown in the inset in A. The longest delay at which the electrically evoked action potential collides with an orthodromic action potential was determined; i.e. the longest delay interval at which the electrical stimulus failed to evoke a stimulus locked NTI (between 8.8 and 7.7 ms in B). The collision interval (tC; middle trace, A) determined in this way comprises both the time it takes the orthodromic action potential to propagate from its location in the axon at the time the NTI is detected to the site of electrical stimulation at the back of the eye and the absolute refractory period of the axon at the site of electrical stimulation (lower trace, A). The absolute refractory period (tR) was estimated by monitoring NTI responses to the second of a pair of electrical stimuli applied to the ciliary nerves (C), the interval between the pair of stimuli being varied to establish longest interval at which the second stimulus failed to evoke an NTI. For both the collision and paired pulse data, the probability of electrically evoking an NTI as a function of stimulus delay was fitted with a Boltzmann function and the point of inflection (i.e. where 50% of electrical stimuli failed to evoke an NTI) was taken as the collision interval (tC) and absolute refractory period (tR) respectively (D). Subtraction of the absolute refractory period from the collision interval (D) gives the virtual interval (tV;eqn (1)) from which the position of action potential initiation in the time domain can be determined (tI; eqn (2)). For comparisons between individual receptors, values in the time domain were normalised to the latency of the electrically evoked NTI (tL).

Download figure to PowerPoint

Collision analysis

Collision techniques were used to (1) establish that ongoing and electrically evoked NTIs originated in the same axon and (2) to map the site(s) of initiation of action potentials producing the ongoing NTIs (see below and Results). Ongoing NTIs were detected using a gated amplitude discriminator (Neurolog 200, Digitimer, UK) that produced a TTL pulse for each event detected. TTL pulses were used to trigger electrical stimulation of the ciliary nerves at the back of the eye at variable delay. To reduce the effect of activity-induced changes in axonal conduction velocity, electrical stimuli were triggered from ongoing NTIs at a maximum rate of 1 Hz.

Determination of collision interval

The term collision interval is used here to describe the empirically derived delay interval for electrical stimulation to evoke an NTI in 50% of trials. This interval was determined by varying the delay between the occurrence of an ongoing NTI at the surface of the cornea and electrical stimulation of the ciliary nerves at the back of the eye (Fig. 1B). For each stimulus delay, the probability of recording an electrically evoked NTI at the surface of the cornea was determined from 20–30 stimulation trials and the resulting data points were fitted with a Boltzmann function (see Fig. 1D). The fit parameter corresponding to the point of inflection was taken as the collision interval.

For action potentials initiating at a point in the axon that is closer, in the time domain, to the site of electrical stimulation at the back of the eye than it is to the terminal (i.e in the proximal half of the axon), occlusion of the electrically evoked NTI will only occur when the electrical stimulus is delivered before the occurrence of an ongoing NTI (i.e. at negative collision intervals, see Fig. 2). This was achieved by taking advantage of the rhythmic firing of NTIs in cold receptors and setting the delay between the time of occurrence of an ongoing NTI and electrical stimulation to an interval slightly shorter than the average period of ongoing NTI activity. For example, if NTIs occur at a mean frequency of 10 Hz, then stimulating at delays slightly less than the mean inter-impulse interval, i.e. between 90 and 100 ms, will, on occasion, result in an ongoing NTI occurring in the period between the time of electrical stimulation at the back of the eye and the detection of the resultant antidromic NTI at the surface of the cornea (see Fig. 2C).

image

Figure 2. Determination of the site of action potential initiation by collision analysis for a single cold receptor with a negative collision interval A negative collision interval is observed for action potentials that are initiated at a site in the time domain that is closer to the point of electrical stimulation at the back of the eye than it is to the nerve termination. Consequently, electrical stimulation at the back of the eye when the ongoing NTI is detected (i.e. at zero delay) always evokes an action potential that conducts to the nerve terminal producing a stimulus-locked NTI (A, 20 overlaid traces at each delay interval). In these cases, collision of the electrically evoked action potential with the orthodromically conducted ongoing action potential only occurs when electrical stimulation precedes the occurrence of an ongoing NTI (i.e. at negative collision intervals). This was achieved by setting the delay between the occurrence of ongoing NTIs and electrical stimulation at the back of the eye to values close to the interval between successive ongoing NTIs (C). In C, the interval between ongoing NTIs was approximately 78 ms and by varying the delay between NTI detection and electrical stimulation about this value, it was possible to obtain recordings in which the ongoing NTI occurred after the electrical stimulus (the lower panels in C show the same traces on an expanded time base). The upper overlaid traces (black) show trials in which electrical stimuli evoked NTIs whereas the lower overlaid traces (grey) show trials in which electrical stimuli failed to evoke NTIs. Ongoing NTIs that occurred at intervals >2.8 ms after the electrical stimulus occluded the electrically evoked NTI (grey traces, C). The negative collision interval determined in this manner still comprises the absolute refractory period (tR) of the axon at the point of stimulation. In B, subtraction of the absolute refractory period from the collision interval (eqn (1)) gives the virtual interval (tV), from which the initiation interval (tI) can be calculated (eqn (2)). For comparisons between individual receptors, values in the time domain were normalised to the latency of the electrically evoked NTI (tL). The flat line segment immediately following the stimulus in all traces shown in A is due to saturation of the amplifier. A change in the stimulus artefact with time resulted in most traces shown in C occurring without saturation. For determination of the collision interval, only the traces where an ongoing NTI could be identified between the time of stimulation and the expected time of occurrence of the electrically evoked NTI were used.

Download figure to PowerPoint

The collision interval establishes the position of the orthodromically propagating action potential relative to the point of electrical stimulation at the time of occurrence of the ongoing NTI at the surface of the cornea (see Fig. 1A). Accordingly, for positive collision intervals the action potential is propagating towards, but not yet at, the site of electrical stimulation when the NTI is detected whereas for negative collision intervals the action potential has already reached or propagated beyond this point when the NTI is detected. In both cases, the empirically derived collision interval includes the absolute refractory period of the axon at the site of electrical stimulation and to establish the position of the orthodromically propagating action potential this needs to be subtracted.

Determination of the absolute refractory period

To estimate the absolute refractory period of the axon, NTI responses to paired electrical stimuli were studied (see Fig. 1C). Electrical stimulation was triggered by the occurrence of ongoing NTIs such that the first of the paired electrical stimuli was delivered 20 ms after the occurrence of an ongoing NTI. This was done to prevent the occurrence of an ongoing NTI during the paired stimulation period. The interval between the paired stimuli was varied to establish the interval at which NTIs evoked by the second stimulus of the pair were recorded at the surface of the eye in 50% of trails (Fig. 1C and D). Since axonal diameter and therefore conduction velocity may well vary along the length of the axon (see Discussion), the absolute refractory period estimated in this way is likely to be representative of the slowest conducting region of axon. The slowest conducting region of axon is unlikely to be that at the site of electrical stimulation at the back of the eye and accordingly this measure of absolute refractory period probably over estimates that at the site of electrical stimulation but it is the only estimate that we were able to ascertain.

Mapping the site of ongoing action potential initiation in the time domain

The site of ongoing action potential initiation cannot be determined by any single empirical measure determined here but it can be inferred by simple calculation as illustrated in the example shown in Fig. 1. For this analysis it is assumed that conduction velocity is uniform along the whole length of the axon. Under this assumption, the latency of the electrically evoked NTI (tL) provides an empirical measure of the total time required for an impulse to travel antidromically from the site of electrical stimulation at the back of the eye to the recording site at the corneal surface. Figure 1A shows a schematic example where the evoked latency is unity and the ongoing action potential initiates at a point 0.75 time units distal to the point of electrical stimulation at the back of the eye (Fig. 1A, upper trace). As this action potential will be conducted both orthodromically and antidromically from its point of initiation, the orthodromically propagating action potential will be 0.5 time units along the axon when the NTI is observed at the nerve terminal (Fig. 1A, centre trace). This position is designated the virtual interval (tV) and is the difference between the empirically determined collision interval (tC) and the absolute refractory period (tR):

  • image(1)

the site of action potential initiation (tI) is then given by:

  • image(2)

Most NTIs were diphasic (positive–negative) and approximate the first derivative of the change in membrane potential at the site of recording (see Brock et al. 2001). For positive–negative NTIs, the point for determination of impulse arrival at the nerve terminal was taken to be the positive peak. Action potentials initiating very close to the site of recording produced NTIs with an initial negative-going component and for these signals the negative peak was used to determine the time of impulse arrival at the terminal. For both NTI configurations, the chosen time point is expected to occur shortly following the initiation of the regenerative Na+ current (see Fig. 4A).

image

Figure 4. Comparison of ongoing NTI shapes with those simulated by initiation of action potentials at discrete locations relative to a sealed-end nerve terminal NTIs were simulated as the sum of capacitive and ionic currents in the sealed-end terminal compartment of a length of axon with Hodgkin–Huxley conductances. Action potentials were initiated by applying constant current to compartments at various distances from the nerve terminal (dashed lines, A). In A, membrane potential along the axon is shown at the time of the first (positive or negative) peak of the NTI (upper traces in the right column show simulated NTIs). This time point was used empirically to determine the time of occurrence of ongoing NTIs and, for both positive–negative- and negative–positive-shaped NTIs, occurs just after the Na+ current is initiated (the lower traces in the right column show the Na+ current). In B, the amplitude and polarity of the first (circles) and second (triangles) peaks of the NTI change depending upon the site of action potential initiation (the filled markers indicate values for the NTIs shown in A). In C and D, the ratio between the positive peak amplitude of positive–negative-shaped ongoing NTIs and electrically evoked NTIs is plotted as a function of the site of action potential initiation determined in the time domain (C) and when transformed into the space domain (D). Averages of at least 30 NTIs were used for the determination of positive peak amplitude (filled markers correspond to the NTI insets in D). The dashed lines in C and D indicate constrained exponential fits to the dataset.

Download figure to PowerPoint

Simulation of nerve terminal impulses

NTIs were simulated using custom-written routines in Igor Pro (Wavemetrics, Lake Oswego, Oregon, USA). The ordinary differential equations of Hodgkin & Huxley (1952) describing active axonal propagation were integrated in both length and time dimensions using the forward Euler method with integration steps of 0.1 length constants and 5 μs, respectively. Action potentials were generated by applying constant current to a given compartment. NTIs are the sum of ionic and capacitive currents in the sealed-end terminal compartment. Axonal diameter was taken as 1 μm, specific membrane capacitance as 1 μF cm−2 and specific axoplasmic resistance as 150 Ω cm−1 resulting in a length constant of ca 700 μm.

Imaging TRPM8GFP axons in the mouse cornea

To ascertain the nerve terminal branching pattern of corneal cold receptors we have investigated their structure in corneas from Trpm8GFP transgenic mice (Takashima et al. 2007). Images from whole-mount corneas were obtained with a LSM710 (Zeiss, Germany) confocal microscope fitted with an oil immersion objective (60×) and using an excitation wavelength of 488 nm.

Data analysis

Group values for population descriptors are quoted as mean ±s.d. Curve fitting was performed in IgorPro which uses the Levenberg–Marquadt algorithm for least-squares minimization.

Results

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Appendix

Mapping the site of impulse initiation

For the 26 cold-sensitive receptors in which collision analysis was performed, the mean conduction latency for electrically evoked NTIs was 8.70 ± 3.14 ms and the mean absolute refractory period determined by paired-pulse stimulation of the ciliary nerves at the back of the eye was 3.45 ± 1.29 ms. Collision techniques were used to verify that ongoing and electrically evoked NTIs originated in the same nerve terminal. An example collision experiment is illustrated in panels B and C of Fig. 1. In panel B, ongoing NTIs (shown to the left of the stimulus artefact) were used to trigger electrical stimulation of the ciliary nerves at the back of the eye at delays between 8.4 and 7.7 ms. At the longest delay (8.4 ms), electrical stimuli reliably evoked a stimulus locked NTI (shown to the right of the stimulus artefact). At a delay of 8 ms, electrical stimulation intermittently evoked NTIs whereas at a delay of 7.7 ms electrical stimuli failed to evoke NTIs. The absence of an electrically evoked NTI indicates collision of the ongoing orthodromic action potential with the electrically evoked antidromic action potential. In panel C, the determination of the absolute refractory period of the axon is shown. In this case, two electrical stimuli were delivered at inter-pulse intervals between 3.5 and 2.5 ms and the likelihood that the second stimulus evoked an NTI was determined.

In 20 receptors, the collision interval was positive, meaning that the electrically evoked NTI was occluded by electrical stimulation at some time after the occurrence of an ongoing NTI (Fig. 1). In the remaining six receptors, occlusion of the electrically evoked NTI could only be demonstrated for electrical stimuli delivered before the occurrence of an ongoing NTI (i.e. at negative collision intervals; see Fig. 2A and C). In these cases, the ongoing action potential must arise at a position along the axon from which it takes longer for the orthodromic action potential to travel to the nerve terminal than it does for the antidromic action potential to reach the point of electrical stimulation at the back of the eye (see Fig. 2). These proximally located sites could potentially map a branch point that gives rise to a co-lateral axonal branch within which the ongoing action potentials were initiated.

For individual receptors, the action potential initiation site for ongoing NTIs was derived according to eqns (1) and (2) using the empirically derived values for the collision interval and the absolute refractory period (see Fig. 1A and Methods). The resulting estimate of the action potential initiation site is therefore a time point relative to the time it takes the electrically evoked action potential to conduct from the point of stimulation at the back of the eye to the point of recording at the terminal (Fig. 3A). Initiation sites determined in the time domain do not correspond directly with spatial locations because axonal conduction is not uniform over the segment of axon, being influenced by axonal geometry (e.g. tapering, beading and branching) and myelination. At present it is not possible to account for the effects of changes in axonal geometry and instead we have only attempted to estimate the effect of the most prominent feature contributing to the spatial inhomogeneity in axonal conduction velocity, namely myelination.

image

Figure 3. Distribution of action potential initiation sites mapped in the time domain and transformed into the space domain As corneal cold receptors are most probably Aδ-neurones with a long unmyelinated nerve terminal axon within the cornea, initiation sites determined in the time domain for 26 receptors (A) were transformed according to eqn (4) to yield estimates of their equivalent location in the space domain (C). The transformation between the two domains is depicted in B. For each receptor, unity in the normalised time domain is the latency of the electrically evoked NTI. Unity in the space domain is defined by unity in the time domain.

Download figure to PowerPoint

The cornea is innervated by trigeminal neurons with either thinly myelinated (Aδ) or unmyelinated (C) axons; however, at their transition into the cornea at the limbus, Aδ-axons loose their myelin and have a long unmyelinated nerve termination (Zander & Weddell, 1951). For Aδ-axons innervating the cornea, conduction velocity is therefore expected to be slower in the unmyelinated portion of the axon within the cornea and faster in the thinly myelinated parent axon proximal to the limbus. Cold receptors innervating the guinea-pig cornea have an average conduction velocity of >2 m s−1 at 37°C (Carr et al. 2003) when determined over the distance between the site of electrical stimulation at the back of the eye and the site of recording at the corneal surface. It is therefore likely that they are Aδ-neurones. Consequently, a simple two-compartment axon was assumed in order to estimate the effect of this inhomogeneity in conduction velocity on the spatial site of action potential initiation. To transform initiation sites determined in the time domain into the space domain, the fractional length of myelinated axon (f) and the conduction velocity ratio between myelinated and unmyelinated segments (ν) were used to define a normalising factor (α):

  • image(3)

This allows initiation sites in the normalised time domain to be transformed into a normalised space domain:

  • image(4)

where inline image and inline image with xI and xL being the initiation site and unity, respectively, in the space domain.

This transformation is depicted graphically in Fig. 3B and the resulting initiation sites in the space domain are shown in Fig. 3C assuming (1) that the myelinated portion of the axon comprises the proximal three quarters of the total axonal length between the recording site at the corneal surface and the stimulation site at the back of the eye (i.e. f= 0.75) and (2) that the conduction velocity in the myelinated portion is five times greater than that in the unmyelinated portion (i.e. ν= 5). This transformation of action potential initiation sites in the time domain to the space domain is based on several simplifying assumptions and is therefore unlikely to be accurate. However, the transformation is likely to provide a more realistic estimate of the spatial location of action potential initiation sites than those determined in time.

NTI shape correlates with the action potential initiation site

We have previously reported that the amplitude of the initial positive component of ongoing NTIs is often smaller than that of NTIs evoked by electrical stimulation in the same axon (Brock et al. 2001). Modelling NTI shape (see Methods for details) as a function of action potential initiation site (i.e. the site of current injection) predicts that NTI shape changes in the manner shown in Fig. 4A. Action potentials that are initiated at sites greater than approximately 3 length constants from the terminal produce diphasic (positive–negative) NTIs that have the same amplitude and time course. Consistent with this prediction, all electrically evoked NTIs, which are produced by action potentials that are initiated at a site that is on average 14 mm from the recording site, have a positive–negative diphasic configuration. For action potential initiation sites located between approximately 3 and 0.6 length constants from the nerve terminal, the resulting NTI is similarly diphasic (positive–negative) but the amplitude of the initial positive component decreases as the site of initiation is moved closer to the terminal (Fig. 4A and B). This change in NTI size, resulting from action potential initiation relatively close to the site of recording, could potentially explain why in many receptors the positive amplitude of the ongoing NTIs is smaller than that of the electrically evoked NTIs (Brock et al. 2001). To examine this possibility, the ratio between the amplitude of the positive component of ongoing and electrically evoked NTIs (Fig. 4D, insets) was plotted as a function of the estimated site of ongoing action potential initiation in the time and space domain (n= 23 receptors, Fig. 4C and D). Consistent with simulated NTIs (Fig. 4B), this ratio decreases mono-exponentially as a function of the estimated proximity of the action potential initiation site to the nerve terminal.

Locally generated action potentials

For action potentials initiating at sites within 0.5 length constants of the terminal the modelled NTIs have a distinctive inverted diphasic configuration, consisting of an initial negative component reflecting the locally generated inward Na+ current followed by a positive component reflecting a net outward current (Fig. 4A and B; see Discussion). Of the data collected during a 5 year period (2001–2006) for 430 cold receptors, in 37 receptors the ongoing nerve activity was comprised of both conventional positive–negative NTIs as well as inverted NTIs with an initial negative component (see examples in Figs 5, 6 and 8). In 8 of these 37 receptors, where the incidence of inverted NTIs was relatively high, collision analysis confirmed (1) that both NTI shapes originated from within the same axon and (2) that the action potentials generating the inverted NTIs initiated at a location in the axon closer to the sensory nerve terminal (i.e. the site of recording) than those action potentials generating positive–negative NTIs. In the example shown in Fig. 5, collision analysis placed the site of action potential initiation at 0.929 normalized time units for the positive–negative NTI (0.071 time units from the terminal) and at 0.952 normalized time units for the inverted NTI (0.048 time units from the nerve terminal). In this experiment, and in the other seven experiments where the action potential initiation site for both shapes of NTI was mapped, the amplitude of the positive component of the positive–negative NTIs was smaller than that of the electrically evoked NTIs (see inset in Fig. 5B), indicating that the action potentials generating these signals were also initiated at sites that were electronically close to the nerve terminal.

image

Figure 5. The shape of ongoing NTIs correlates with the site of action potential initiation determined by collision analysis within individual cold receptors In A, the ongoing activity at 32°C was comprised of two distinct NTI shapes: a conventional positive–negative-shaped NTI (left panel) and a negative–positive-shaped NTI (right panel). Twenty overlaid traces are shown for each stimulus delay interval and in these traces the stimulation artefact has been removed by subtraction. The action potentials generating both ongoing NTI shapes could be collided with the action potentials generating the electrically evoked NTI (i.e. they could occlude electrically evoked NTIs), indicating that they both originated in the same neurone. In B, collision analysis showed that the initiation site of action potentials generating positive–negative NTIs mapped to a site in the time domain that was further from the nerve terminal than that generating negative–positive NTIs. The left inset shows averages of both the ongoing (continuous line) and electrically evoked (broken line) positive–negative NTIs overlaid and the right inset shows an average of the negative–positive NTIs. The NTI insets within B are averages of 50 individual traces.

Download figure to PowerPoint

image

Figure 6. Independence of ongoing NTI rhythm and NTI shape In cold receptors with both positive–negative and negative–positive NTI shapes, the rhythm of ongoing NTI discharge was independent of NTI shape at the control temperature (∼32°C). In the example shown, NTIs occurred at intervals of ∼130 ms irrespective of whether they had a positive–negative (+) or negative–positive (•) shape. The NTIs inset within the figure are averages of 50 individual traces.

Download figure to PowerPoint

image

Figure 8. Changes in the relative incidence of positive–negative and negative–positive NTI shapes in a single cold receptorproduced by heating and cooling In the example shown, ongoing action potentials generating both positive–negative (A) and negative–positive (B) NTI shapes at the control temperature (∼32°C) could be collided with the action potentials generating electrically evoked NTIs, confirming that they originated in the same receptor. In A and B, the upper overlaid traces show stimulus delay intervals that triggered electrically evoked NTIs whereas the lower overlaid traces show stimulus delay intervals that failed to trigger electrically evoked NTIs. C–E, graphs showing the temperature (C), the frequency of NTI discharge (D) and the relative incidence of negative–positive NTIs during successive 10 s periods (E). F, averaged NTIs recorded during the periods indicated in D. At the control temperature, both shapes of NTI were recorded but the incidence of negative–positive NTIs was very low (D, E and F, period 1). During heating, when the nerve activity decreases, the incidence of positive–negative NTIs falls to zero and all NTIs take on a negative–positive waveform (D, E and F, period 2). In contrast, shortly after initiating cooling (D), when nerve activity increases, the incidence of negative–positive NTIs falls to zero and only positive–negative NTIs are recorded (D, E and F, period 3). During reheating, only positive–negative NTIs are also recorded (D, E and F, period 4). In F, each trace is the average of at least 20 NTIs.

Download figure to PowerPoint

In all cold receptors with both conventional and inverted NTI shapes, the rhythm of ongoing NTI discharge was independent of NTI shape, further supporting the proposal that the different shaped NTIs arose from a single neuron. An example is shown in Fig. 6 where NTIs occurred at intervals distributed about 130 ms irrespective of their shape.

TRPM8GFP axons in the mouse cornea

Cold-sensitive trigeminal neurones isolated from guinea-pig respond to cooling stimuli and to menthol in a manner comparable to cold-sensitive trigeminal neurones isolated from mouse (Madrid et al. 2006). However, it has not yet proved possible to selectively label cold-sensitive nerve terminals in the guinea-pig cornea. For this reason we have used corneas from transgenic mice selectively expressing eGFP in axons expressing TRPM8 (Takashima et al. 2007) to establish the morphology of cold-sensitive nerve terminals. In the mouse corneal stroma, TRPM8GFP axons form a ramified network within which axons can be seen to branch (Fig. 7A, arrow). Individual axons ascend from this network to enter the corneal epithelium where they are varicose and may course alone over distances of at least 100 μm before ascending to the outermost layers of the epithelium and terminating with an enlarged end bulb (Fig. 7B). Branching of TRPM8GFP axons within the epithelium of the mouse corneas was not observed. This arrangement is similar to that reported for the terminals of Aδ TRPM8GFP neurones in mice teeth, where the most terminal branch point is within the pulp and the terminating axons can extend for over 100 μm into the caniculi within the dentin (Takashima et al. 2007). This morphology makes it likely that the action potential initiation sites determined in cold receptors in the guinea-pig cornea which map to sites electrotonically very close to the nerve ending (see Fig. 3) are within the most terminal axon branch.

image

Figure 7. Transient receptor potential melastatin 8 (TRPM8GFP) axons in mouse cornea Each image was generated by compressing stacks of 10 individual confocal images taken with a z-step of 0.8 μm. In the corneal stroma (A), there is an extensive ramified network of TRPM8GFP axons within which individual axons can be seen to branch (arrow). In the epithelial layers of the cornea (B), individual TRPM8GFP axons are varicose and course alone for several hundreds of micrometers before ascending to the external layers of the corneal epithelium and terminating with an enlarged end-bulb. The scale bar in each panel represents 20 μm.

Download figure to PowerPoint

Temperature-induced changes in NTI shape

The shape of ongoing NTIs recorded at the same ambient temperature in cold receptors changes depending on whether the receptor is being heated or cooled (Carr et al. 2003). NTIs recorded during heating are larger in amplitude and faster in time course than those recorded during cooling. We have demonstrated previously that while the nerve terminals of cold receptors possess sufficient Na+ channels to support a regenerative Na+ current (Carr et al. 2002), local Na+ channel blockade does not normally change the configuration of NTIs recorded at the static holding temperature (Brock et al. 2001; Carr et al. 2003). To explain these observations, we suggested that a considerable portion of the Na+ channels present in the nerve terminal are in an inactivated state and that NTI shape is determined by electrotonic invasion of the nerve terminal from a point more proximal in the axon where the action potential fails to propagate actively. If this were the case, the heating-induced increase in amplitude and speeding in time course of the NTI can be explained by hyperpolarization of the nerve terminal which relieves Na+ channel inactivation proximal to the nerve ending allowing active propagation further into the terminal. This explanation only applies for action potentials that back-propagate into the nerve terminal producing positive–negative NTIs. For receptors where both positive–negative NTIs and inverted NTIs are recorded, heating-induced relief of Na+ channel inactivation within the nerve terminal might be expected to favour action potential initiation at sites closer to the nerve termination and thereby increase the likelihood of inverted NTIs.

Evidence that this can indeed occur was found in some receptors where both positive–negative- and negative–positive-shaped NTIs were recorded at the control temperature. During heating and/or at the start of cooling following heating, the frequency of occurrence of negative–positive-shaped NTIs increased relative to that of positive–negative-shaped NTIs (see Fig. 8). For six receptors, there was a period during heating when all ongoing NTIs had an initial negative component. Figure 8 shows an example where under control conditions most ongoing NTIs had a positive–negative shape although an occasional negative–positive-shaped NTI was recorded. Both shapes of NTI collided with the electrically evoked NTI, indicating both that they originated in the same axon and also that at least two action potential initiation sites were present. Shortly after the start of heating, the incidence of negative–positive-shaped NTIs increased and for most of the period of heating and at the start of cooling from ∼38°C, only NTIs of this configuration were recorded (Fig. 8C–F). Only positive–negative-shaped NTIs were recorded during the remaining period of cooling and when the receptor was reheated to the control temperature. Although not always as dramatic as this example, NTIs with an initial negative component appeared during heating and/or at the start of cooling following heating in 47 cold receptors that had only positive–negative NTIs at the control temperature. These findings suggest that the site of action potential initiation in the unmyelinated nerve terminals of cold receptors is labile and that the site of initiation can move toward the nerve terminal when the temperature is increased.

Discussion

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Appendix

Establishing the site at which the action potential initiates within the terminal region of unmyelinated sensory axons is fundamental to an understanding of their function. Cold-sensitive axons provide a suitable model in which to examine action potential initiation in unmyelinated terminals because they have an ongoing discharge of action potentials and in the guinea-pig cornea it is possible to make extracellular recordings from their sensory nerve terminals (Brock et al. 1998, 2001). The site of action potential genesis was determined in corneal cold-sensitive receptors by establishing the interval over which ongoing action potentials collided with activity evoked by electrical stimulation of the ciliary nerves at the back of the eye. In three quarters of the receptors, the origin of activity in the time domain mapped to sites in the axon closer to the nerve terminal than they were to the site of electrical stimulation. As indicated in the Results, we believe that guinea-pig corneal cold receptors are Aδ-neurones that lose their myelin sheath close to the point where they enter the cornea. The loss of myelin will slow action potential propagation within the cornea, an effect that would move action potential initiation sites mapped in the time domain to sites closer to the site of recording at the nerve termination in the space domain (see Fig. 3). In accord with this suggestion, in ca 40% of receptors, the proximity of the initiation site to the nerve ending can be further inferred from differences between the shape of ongoing NTIs and those evoked by electrical stimulation of the ciliary nerve back of the eye (see Fig. 4). In the remaining one quarter of the corneal cold receptors, the origin of activity mapped to sites in the time domain that are closer to the site of electrical stimulation at the back of the eye than they are to the nerve terminal. Such sites are likely to be axonal branch points, which are known to occur close to the point where Aδ-axons innervating the cornea loose their myelin sheath at the corneo-scleral junction (Zander & Weddell, 1951). In these cases, we believe the ongoing action potentials are initiated in the receptive nerve terminal(s) of an axon branch and propagate by axon reflex antidromically to the recording site.

The accuracy of estimates of the site of action potential initiation determined using collision analysis would also be affected by directional asymmetry in axonal conduction velocity, with estimates being erroneously skewed towards the terminal if antidromic conduction (i.e. toward the nerve termination) were more rapid. Axonal branching is expected to exert the most prominent effect in this respect and along their course axons within the cornea display a moderate degree of branching (Zander & Weddell, 1951; Whitear, 1960; MacIver & Tanelian, 1993). Theoretically, at divergent branch points, the decrease in diameter of each daughter branch (Quilliam, 1956) may produce an impedance mismatch that would render conduction in the antidromic direction more rapid than that in the orthodromic direction (Goldstein & Rall, 1974). In absolute terms, each branch point at which the change in axonal diameter does not exceed a halving in the daughter branches would only be expected to produce a small lead of ca 100 μs for antidromically conducted action potentials (Manor et al. 1991). Functional (Waddell et al. 1989; Peng et al. 1999) and anatomical (Thiel, 1957; Banks et al. 1982; Heppelmann et al. 1990) evidence both suggest minimal tapering along myelinated and unmyelinated axons. At the axon termination, however, it would be predicted that there is a reduction in axon diameter and theoretically this could also slightly speed conduction in the antidromic direction relative to that in the orthodromic direction (Toth & Crunelli, 1998). It is not possible to account for these differences in orthodromic and antidromic conduction velocity but it is reasonable to assume that the errors in estimates of initiation site attributable to either axonal branching or tapering of the terminal portion of axon are small.

The NTI signal recorded extracellularly from nerve terminals is proportional to net membrane current, with positive- and negative-going signals representing outward and inward current respectively (Smith, 1988). The most commonly recorded NTI shape can be adequately simulated by an action potential propagating into the terminal and consists of an initial net outward current followed by a smaller net inward current (Carr et al. 2003). An NTI of this general shape results from antidromic invasion of the nerve terminal by an action potential initiated at a site that is electronically remote from the nerve terminal, such as that resulting from electrical stimulation of the ciliary nerves at the back of the eye. For ongoing NTIs with this conventional shape, the amplitude of the initial positive component is typically smaller than that of electrically evoked NTIs in the same nerve terminal (Brock et al. 2001). It was previously suggested that the smaller amplitude of ongoing NTIs resulted from the nerve terminal being depolarized during action potential invasion (Brock et al. 2001). Simulations of NTI shape indicate, however, that for action potentials initiating between approximately 0.6 and 3 length constants from the nerve terminal the positive amplitude of the NTI decreases monotonically as the site of initiation moves closer to the nerve termination (Fig. 4). We have previously reported that on average ongoing NTIs are smaller than their electrically evoked counterparts in cold-sensitive receptors in the cornea (Brock et al. 2001), which is consistent with our observation that action potential initiation often occurs very close to the site of recording at the nerve terminal.

Ongoing NTIs with an initial negative-going component (i.e. inward current) observed in some receptors are particularly interesting because simulations suggest that NTIs of this configuration arise from initiation of action potentials within about half a length constant of the nerve termination (Fig. 4). For these negative–positive NTIs, the initial net inward current is dominated by Na+ channel opening during action potential genesis and the later net outward current is produced as the action potential propagates away from its point of initiation. Collision analysis confirmed that these inverted NTIs result from action potentials that initiate very close to the nerve termination. Furthermore, within a single nerve terminal, ongoing NTIs with an inverted shape always occurred together with NTIs of conventional shape and collision mapping indicated that action potentials producing inverted NTI shapes initiated at a site closer to the nerve termination than those generating conventionally shaped NTIs (Fig. 5). The presence of multiple ongoing NTI shapes in a single axon termination indicates that there can be multiple sites of action potential initiation within the nerve terminal axon and that the point of initiation is labile, shifting from impulse to impulse. While this could be interpreted as indicating that action potential initiation can occur anywhere within the nerve terminal, collision analysis indicates that each NTI shape has a relatively narrow domain over which collision occurs (see Figs 1D, 2B and 5B), demonstrating that there are favoured sites of action potential initiation. It is possible that these sites are located at points along the unmyelinated axon where voltage-gated Na+ channels are clustered (Hildebrand & Waxman, 1983) as has recently been reported for the en passant boutons along the axons of mossy fibre neurons in the hippocampus of the rat (Engel & Jonas, 2005).

Sensory neurones with thinly myelinated (Aδ) and unmyelinated (C) axons often form a branched nerve terminal arbour in the periphery (Heppelmann et al. 1990, 1995; MacIver & Tanelian, 1993). The functional correlate of axonal branching is a structured receptive field composed of discrete regions of high sensitivity, figuratively designated ‘hot spots’ at which the threshold for evoking an action potential that propagates to the parent axon is relatively low (Bessou & Perl, 1969; Schmidt et al. 2002). The receptive fields of cutaneous cold-sensitive receptors can be composed of multiple discrete ‘cold spots’ at which cold stimuli increase nerve activity (Duclaux & Kenshalo, 1973). However, when two cold receptive spots were simultaneously cooled, the activity in the parent axon was the same as that evoked by cooling the most sensitive spot, i.e. the cold receptive spot generating the highest rate of action potential discharge occludes the activity originating at the other receptive spot. This suggests that each cold-sensitive spot is the termination of a separate axon branch and that each branch encodes action potentials autonomously. This implies that electrotonic interaction between the individual action potential encoders is small. Consistent with this suggestion, when one cold receptive spot was heated in receptors with multiple receptive spots, the ongoing activity was little affected presumably because activity originating at the other receptive sites where the temperature did not change maintained the ongoing activity (Duclaux & Kenshalo, 1973).

At the normal holding temperature (∼32°C), corneal cold receptors, like cold receptors in other tissues, have an ongoing discharge of nerve impulses that occur in a rhythmic manner, comprising both single impulses as well as short bursts of impulses at relatively fixed intervals (see Carr et al. 2003). In receptors with multiple cold receptive spots, it would be predicted that the ongoing nerve activity originates at the receptive site where the membrane potential reaches the threshold for action potential initiation first, i.e. the dominant pacemaker. In accord with this suggestion, in cornea, all the ongoing action potentials recorded appeared to originate within a single spatially restricted region of the nerve terminal arbour that was either located relatively close to the site of recording or, most likely, at the termination of a co-lateral axon branch. Furthermore, in the collision analyses, it was always possible to find a collision interval that completely occluded electrically evoked NTIs, indicating that all action potentials generating ongoing NTIs also propagated centrally irrespective of their site of origin within the nerve terminal arbour.

The cell bodies of cold-sensitive sensory neurons in dorsal root and trigeminal ganglia hyperpolarize and depolarize, respectively, during warming and cooling (Reid et al. 2002; Viana et al. 2002). On the basis of changes in NTI shape recorded at the same temperature during heating and cooling, we have previously suggested that this also occurs at the sensory nerve terminals of corneal cold receptors (Carr et al. 2003). During cooling, the shape of the ongoing positive–negative NTIs is more like that expected for an action potential invading the nerve termination electrotonically. Assuming that NTIs of this shape are produced by action potentials that propagate antidromically from their point of genesis into the nerve terminal (see above), the changes in NTI shape seen during cooling would be consistent with a depolarization-induced inactivation of Na+ channels reducing their availability to support active propagation in the nerve termination. Conversely, warming alters NTI shape in a manner consistent with an increased contribution of Na+ channels to action potentials propagating into the nerve terminal, an effect that can be accounted for by hyperpolarization of the nerve terminal reducing the number of voltage-inactivated Na+ channels. Here we report that in receptors where NTIs of both conventional and inverted (negative–positive) shape are recorded, heating increases the relative frequency of occurrence of inverted NTIs (Fig. 8). Furthermore, in approximately 10% of receptors for which only conventional shaped NTIs where recorded at ∼32°C, heating caused the appearance of inverted NTIs. As inverted NTIs represent initiation close to the nerve termination, these findings are also consistent with a hyperpolarization of the nerve terminal during warming increasing the number of Na+ channels available for activation in the nerve terminal axon and thereby increasing the likelihood that action potentials are initiated at sites located very close to the site of recording at the nerve termination.

The sensory nerve terminals of guinea-pig corneal cold-sensitive neurons possess both TTX-sensitive (TTX-S) and TTX-resistant (TTX-R) Na+ channels (Brock et al. 1998, 2001; Carr et al. 2002). However, TTX-R Na+ channels alone are able to support the initiation of action potentials in cold receptors (Brock et al. 1998). In most cold receptors, under control conditions, blockade of both TTX-S and TTX-R Na+ channels selectively at the nerve terminal with the local anaesthetic, lignocaine, does not change the shape of positive–negative-shaped NTIs (Brock et al. 2001). However, hyperpolarization of the nerve terminal with extracellularly applied current reveals a large inward Na+ current that is abolished by lignocaine but is largely resistant to TTX (Carr et al. 2002). These findings demonstrate that, under control conditions, Na+ channels within the nerve terminal are substantially inactivated and that, when hyperpolarized, TTX-R Na+ channels make a dominant contribution to the Na+ current elicited when the action potential invades the nerve termination. Therefore, when warmed, the increased likelihood of action potentials being initiated at sites located very close to the nerve terminal is most probably due to relief of TTX-R Na+ channel inactivation.

In Trpm8GFP transgenic mice, axons expressing eGFP branch within the stromal nerve plexus but do not branch after entering the corneal epithelium within which they typically course for over 100 μm before terminating. Assuming a similar arrangement in the guinea-pig cornea, the length constant of the nerve terminal axons would need to be in the order of hundreds of micrometers if individual terminals of corneal cold receptors were to interact electrotonically. To estimate the length constant necessitates some assumptions regarding axon size as well as the specific resistivity of both the axoplasm and axolemma. In cornea, the axons terminating in the superficial layer of the corneal epithelium have diameters of 0.25 μm or less but the terminating axons are varicose (see Belmonte & Gallar, 1996). We have therefore assumed an average axon diameter of 0.5 μm. The specific resistivity of axoplasm has been determined in a variety of preparations and values in the range 50–100 Ω cm are consistently reported (Jack et al. 1975). Estimates of the specific resistivity of the membrane for mammalian sensory neurons vary considerably. For C-fibre neurones, values determined from somal recordings range from 2–7 kΩ cm2 (cat nodose ganglion: Gallego & Eyzaguirre, 1978; cat DRG: Brown et al. 1981) to 50 kΩ cm2 (rat DRG: Villiere & McLachlan, 1996) while for Aδ-fibre neurones values range between 2 kΩ cm2 (rat nodose ganglion: Gallego & Eyzaguirre, 1978) and 30 kΩ cm2 (rat DRG: Villiere & McLachlan, 1996). For specific membrane resistivities between 2–30 kΩ cm2, a specific axoplasmic resistivity of 100 Ω cm and an axonal diameter of 0.5 μm, the axonal length constant ranges between ca 220 and 860 μm. A length constant of this magnitude suggests that action potentials generating positive–negative-shaped NTIs may initiate at a site where sensory stimulus-induced changes in membrane potential generated in the nerve terminals of adjoining nerve terminal branches are integrated. This point may be proximal to the most distal axon branch and therefore within the stroma. However, it is also possible that action potential initiation occurs within the terminating axon branch as evidenced by the occurrence of negative–positive NTIs. Indeed, because we are only recording from a single nerve terminal, it is possible that action potentials are always initiated within a terminating axon and propagate by axon reflex to the site of recording generating a positive–negative-shaped NTI, in which case the low incidence of recording from nerve terminals with inverted shaped NTIs may simply reflect the likelihood of recording from the nerve terminal branch within which action potentials are being initiated. In either case, the estimate of axonal length constant relative to the morphology of cold-sensitive axons in the cornea suggests that sensory-induced changes in membrane potential from individual nerve terminal branches can interact within the nerve terminal arbour.

The recording of an action potential during its initiation provides evidence that the functional substrates responsible for sensory transduction and action potential initiation exist in parallel with one another within the nerve terminal region of unmyelinated sensory nerve terminals. This arrangement implies that in cold-sensitive nerve terminals, changes in membrane potential resulting from the transduction of temperature not only drive the initiation of action potentials but also affect directly the number of Na+ channels available for activation within the nerve terminal region. As a consequence impulse encoding during hyperpolarization is inextricably associated with a relief of Na+ inactivation and therefore a tendency for the site of action potential initiation to shift towards the nerve termination. Likewise during depolarization the likelihood of action potential initiation at sites more proximal to the axon termination increases.

References

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Appendix

Appendix

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. References
  7. Appendix

Acknowledgements

We are very grateful to Swetlana Nikolajewa, Peter Grafe and James Fallon for valuable discussions. This work was supported by the Australian Research Council.