Myelination speeds conduction of the nerve impulse, enhancing cognitive power. Changes of white matter structure contribute to learning, and are often assumed to reflect an altered number of myelin wraps. We now show that, in rat optic nerve and cerebral cortical axons, the node of Ranvier length varies over a 4.4-fold and 8.7-fold range respectively and that variation of the node length is much less along axons than between axons. Modelling predicts that these node length differences will alter conduction speed by ~20%, similar to the changes produced by altering the number of myelin wraps or the internode length. For a given change of conduction speed, the membrane area change needed at the node is >270-fold less than that needed in the myelin sheath. Thus, axon-specific adjustment of node of Ranvier length is potentially an energy-efficient and rapid mechanism for tuning the arrival time of information in the CNS.

DOI: http://dx.doi.org/10.7554/eLife.23329.001

eLife digest

Information is transmitted around the nervous system as electrical signals passing along nerve cells. A fatty substance called myelin, which is wrapped around the nerve cells, increases the speed with which the signals travel along the nerve cells. This allows us to think and move faster than we would otherwise be able to do.

The electrical signals start at small “nodes” between areas of myelin wrapping. Originally it was thought that we learn things mainly as a result of changes in the strength of connections between nerve cells, but recently it has been proposed that changes in myelin wrapping could also contribute to learning.

Arancibia-Cárcamo, Ford, Cossell et al. investigated how much node structure varies in rat nerve cells, and whether differences in the length of nodes can fine-tune the activity of the nervous system. The experiments show that rat nerve cells do indeed have nodes with a range of different lengths. Calculations show that this could result in electrical signals moving at different speeds through different nerve cells.

These findings raise the possibility that nerve cells actively alter the length of their nodes in order to alter their signal speed. The next step is to try to show experimentally that this happens during learning in animals.

DOI: http://dx.doi.org/10.7554/eLife.23329.002

Main text


During development or learning, an adjustment of myelin thickness or internode length may be used to tune the conduction speed of myelinated axons (Fields, 2008; Tomassy et al., 2014; Ullén, 2009; Kimura and Itami, 2009). This can promote synchronous neuronal firing (Lang and Rosenbluth, 2003; Sugihara et al., 1993), make impulse propagation time less dependent on the spatial trajectory of the axon transmitting information between areas (Salami et al., 2003), or adjust propagation delays to mediate sound localization (Carr and Konishi, 1990; McAlpine and Grothe, 2003; Seidl et al., 2010; Ford et al., 2015). Magnetic resonance imaging of humans and cellular studies of rodents suggest that myelination can increase while learning motor tasks such as piano playing (Bengtsson et al., 2005), juggling (Scholz et al., 2009), reaching (Sampaio-Baptista et al., 2013) and running (McKenzie et al., 2014).

Although most interest has focussed on the effect of changes of myelin thickness or internode length, the node of Ranvier is another potential determinant of action potential conduction speed. Increasing the length of the node will increase the node capacitance and the axial resistance for current flow into the internode, which will both decrease conduction speed. On the other hand, a greater length could increase the number of voltage-gated Na+ channels at the node (if the channel density is constant), which may increase conduction speed. Given the potential influence of node length on conduction speed, we quantified heterogeneity of the length of the node of Ranvier in the white matter of the rat optic nerve and corpus callosum, and in the grey matter of rat cerebral cortex. Computer modelling was then used to explore the effects on conduction speed of the range of node lengths observed.


Variation of node length in the optic nerve and cortex

We first measured the length of nodes of Ranvier and axon diameter in adult rat optic nerve using both confocal and serial electron microscopy (Figure 1, Figure 1—figure supplement 1). Using electron microscopy (EM) we found the mean node length was 1.08 ± 0.02 µm (mean ± s.e.m., n = 46 nodes). Node lengths varied 2-fold, from 0.7 to 1.4 µm with a standard deviation of 0.15 µm (Figure 1—figure supplement 1). Node lengths were not significantly correlated with axon diameter at the node, which had a mean value of 0.80 ± 0.03 µm (standard deviation 0.19 µm; Figure 1—figure supplement 1, the slope of the regression line is not significantly different from zero, p=0.46). Thus, the different node lengths observed did not simply reflect axons being of different sizes.

In contrast to EM, confocal microscopy allowed us to measure a greater number of nodes in different parts of the CNS. In addition, given that oligodendrogenesis and myelination continues well into adulthood (Dimou et al., 2008; Young et al., 2013), confocal microscopy of antibody-labelled nodes allowed us to distinguish developing nodes from mature nodes, ensuring that any differences in size observed were not due to different developmental stages. Mature nodes of Ranvier were identified from their NaV1.6 staining (a marker of mature nodes: Boiko et al., 2001; Kaplan et al., 2001) flanked by Caspr-labelled paranodes (Figure 1A), and node length was measured from the Caspr intensity profile across the node (Figure 1B, see Materials and methods). We first assessed whether, using confocal microscopy (Figure 1C), we could observe variability of node lengths in the rat optic nerve similar to that found when using EM. We again found node length variability (Figure 1C,E,F,H), the mean node length was not significantly different (p=0.06) from that obtained with EM (1.02 ± 0.02 μm, standard deviation 0.29 µm, n = 164, Figure 1E), and there was again no significant dependence on either node diameter (Figure 1H, p=0.14) or axon diameter at the paranode (measured as the diameter of the Caspr labelling: p=0.89). However, the node length range was slightly broader, covering a 4.4-fold range (Figure 1F,H), perhaps due to the ~4 fold greater number of nodes quantified.

In the grey matter of the adult cerebral cortex (layer V of motor cortex) an even larger, 8.7-fold, range of node lengths was observed (Figure 1D,E,G,H), from 0.43 to 3.72 µm (mean value ± s.e.m 1.50 ± 0.05, standard deviation 0.58 µm, n = 158). Again there was no significant dependence on node diameter (Figure 1H, p=0.42), the mean value of which was 0.64 ± 0.01 µm (n = 158, standard deviation 0.14 µm), or on axon diameter at the paranode (p=0.98).

Thus, variability of node length is a general feature of myelinated axons. The greater variability of node length observed in the cortex than in the optic nerve has parallels with the far greater variability of myelination seen in the adult cortex (Tomassy et al., 2014) and could reflect tuning of individual axon conduction speeds to meet information processing needs.

To investigate whether this node length variability was accompanied by a similar variability in sodium channel number, we summed the intensity of NaV1.6 staining over individual nodes in cortex and compared this with node length (Figure 1I). We found that there is a linear correlation between summed NaV1.6 staining and node length (p=1.2×10−15) indicating that the number of sodium channels at nodes is approximately proportional to node length (cf. Rios et al., 2003), consistent with larger nodes being generated by the insertion of more NaV-containing membrane. However, at any given sodium channel labelling intensity there was still a large variation in node lengths, suggesting that sodium channel density is not absolutely constant, and that it may be possible to vary node length in a manner independent of sodium channel number.

Node lengths vary far more between axons than along axons

If adjustment of node length is used to tune conduction speed, one might expect all the nodes along one axon to have similar lengths (e.g. all long or all short). To assess whether the variability in node lengths mainly occurs along axons or between axons we iontophoretically injected a fluorescent dye into the cortex of adult rats (see Materials and methods). This was taken up into axons and diffused along them, allowing us to measure the lengths of up to 13 successive nodes (mean 6.7 ± 0.8 nodes in 18 axons) along single fluorescently labelled axons in the corpus callosum (Figure 2A). Remarkably, along individual axons (Figure 2B,C), the distribution of node lengths was much narrower than that observed over all callosal axons examined (Figure 2C), with a 48.8% ± 3.5% lower coefficient of variation (s.d./mean, Figure 2D, p=1.1×10−10, one sample t-test, n = 18 axons). Importantly, even when excluding axons with mean node lengths >2.25 µm (which were found less frequently, as predicted by the distribution in Figure 1G–H), the coefficient of variation for single axons was 38.8% ± 4.7% lower than that observed over all axons (p=6.6×10−7). Node length was not correlated with internode length (p=0.1, Figure 2E).

Figure 2.
Download figureOpen in new tabFigure 2. Node of Ranvier lengths are correlated along an axon.

(A) Composite confocal image of a single axon in the corpus callosum iontophoretically labelled with tetramethylrhodamine dextran (red). Three consecutive nodes of Ranvier are highlighted and shown in high resolution images. Nodes of Ranvier are identified as NaV1.6 positive clusters (green) flanked by Caspr positive paranodes (blue). (B) Successive node lengths along three example axons with different mean node lengths. The mean node length for each axon is plotted as a dashed line. (C) Distributions of node lengths of 18 individual axons (in 0.5 µm bins, centered on the median node length for each axon) show that the variability along each axon is much less than the variability between axons. (D) Mean coefficient of variation for node lengths along 18 individual axons and the overall coefficient of variation for all axons examined. (E) Mean internode length for each axon plotted against the mean node length for that axon. Each axon is represented by a different colour, and that colour is maintained for panels B, C and E. Regression line slope is not significantly different from zero (p=0.1).

DOI: http://dx.doi.org/10.7554/eLife.23329.005

Thus, node lengths are similar along axons but differ significantly between axons. This raises the possibility that individual axons consistently adjust their node length to tune conduction speed.

Predicted effects of node length variation on conduction speed

To examine the consequences of nodes of Ranvier having different lengths, we simulated action potential propagation in optic nerve and cortical grey matter myelinated axons, as described in the Materials and methods. The differential equations of the model were derived and solved as in Halter and Clark (1991). Details of the parameters used are summarised in Table 1. The conduction speeds predicted for the mean node lengths observed (2.95 m/s for the optic nerve and 2.61 m/s for the cortex) were within the range of values observed experimentally in the adult rat optic nerve (2.5–15 m/s: Foster et al. (1982); Sefton and Swinburn (1964); Sjöström et al. (1985)) and for different classes of rat cortical grey matter output axons (1.8–5.9 m/s: Kelly et al., 2001).

Table 1.

Electrical and geometrical parameters of the models.

DOI: http://dx.doi.org/10.7554/eLife.23329.006





Nodal Na+ conductance*




Nodal K+ conductance*




Nodal persistent Na+ conductance*




Leakage conductance












Myelin membrane conductance




Axon membrane capacitance












Myelin membrane capacitance†,‡




Axoplasmic resistivity




Periaxonal resistivity




Resting potential




Leakage potential




Na+ reversal potential




K+ reversal potential




Node diameter

Optic nerve









Node length

Optic nerve









Paranode length

Optic nerve









Paranodal effective periaxonal space

Optic nerve









Internodal axon diameter

Optic nerve









Internodal periaxonal space




G ratio

Optic nerve







Number of myelin wraps

Optic nerve







Internode length

Optic nerve









  • *Values for standard node length: 1.02 µm in optic nerve, 1.50 µm in cortex; these are constant for simulations with fixed nodal conductance density, but scaled inversely with node length for simulations where number of nodal channels is kept constant.

  • Membrane capacitance values are from Gentet et al. (2000).

  • Figures are per myelin membrane. There are two membranes per myelin lamella.

Our data suggest a positive correlation between the number of NaV1.6 channels and node length (indicating a fixed channel density), but also raise the possibility of node length varying in a manner independent of channel number (Figure 1I). We therefore modelled two extreme situations, for both the optic nerve and the cortical axons studied: either the density of nodal ion channels was assumed to be constant (so the number of ion channels increases in proportion to node length), or the number of ion channels at the node was held constant at the values assumed for the mean node length observed (so the density of channels varies inversely with node length).

Figure 3A and B show that, when the number of channels was held constant at each node, the predicted conduction speed falls with increasing node length (dashed curves). This occurs for two reasons: the increase in node length increases the nodal capacitance (so each node takes longer to charge), and the intracellular axial resistance to current flow from the node into the internode is increased. The changes in conduction speed for the optic nerve are shown in Figure 3A (the range of measured node lengths is shown for comparison). Increasing the node length from its mean value of 1.02 µm to the largest value observed (2.2 µm) is predicted to decrease the conduction speed by 6.5%, while decreasing the node length to the smallest value measured (0.5 µm) increases the speed by 3.2% (giving a speed that is 10.3% larger than at a length of 2.2 µm). For cortical axons (Figure 3B) the predicted changes are larger, partly because, with a 1.5-fold longer node and a 1.7-fold shorter internode length, the nodal membrane contributes a larger fraction of the total membrane capacitance (14% in cortical axons versus 8% in optic nerve axons). The node length variation observed in rat cortex (Figure 1G–H and 0.43 µm to 3.7 µm) results in conduction speeds that are 11.6% slower (for the longest node) and 7% faster (for the shortest node) than the speed for the mean node length of 1.5 µm. Thus, altering node length from 3.7 to 0.43 µm would increase the speed by approximately 21%.

Figure 3.
Download figureOpen in new tabFigure 3. Predicted effect on conduction speed of different node lengths.

(A–F) Calculated conduction speed as a function of node length for axons in (A) the optic nerve and (B–F) the cortical grey matter. For panels A–F, simulations were carried out assuming either that the density of ion channels at the node is constant as the node length is changed (solid lines), or that the number of ion channels is kept constant (dashed lines) at the value assumed for the mean node length. (C–D) Simulations for cortex as in B but examining the effect of altering internode length (INL, given by each curve). (E–F) Calculated dependence of conduction speed of cortical axons on internode length for different assumed node lengths (NL). The observed range of each abscissa parameter is indicated on the graphs. (G). Change in membrane area needed, in the myelin sheath (myelin wrap) or node of Ranvier (node length, NL), to change the conduction speed by 8.6% in optic nerve or 10.5% in cortical axons, when nodal ion channel density is held constant (note logarithmic scale).

DOI: http://dx.doi.org/10.7554/eLife.23329.007

When the nodal ion channel density is kept constant another factor affects the predicted conduction speed, in addition to the change of capacitance and axial resistance at the node: as the node length is decreased the reduction in the number of ion channels present leads to a decrease of conduction speed. Consequently, the plot of speed against node length shows a maximum (solid curves in Figure 3A–B: note that, above this maximum, increasing node length decreases speed for the reasons stated above, despite the increase in number of sodium channels at the node). Decreasing the optic nerve node length from its mean value of 1.02 µm to the lowest length observed (0.5 µm) is predicted to decrease conduction speed by 14.1%, while increasing node length to the value generating the maximum conduction speed (1.7 µm) increases the speed by 3.3% (Figure 3A), to a value that is 20.2% higher than at a length of 0.5 µm. Similarly, for the cortex, decreasing the node length from the mean value of 1.5 µm to the smallest observed value of 0.43 µm decreases the conduction velocity by 19.6% while increasing the length slightly to 1.7 µm increases the speed by 0.2% to a value 24.6% larger than at a node length of 0.43 µm (Figure 3B).

Although the node length is similar at successive nodes along individual axons (Figure 2B), some variation does exist (with a mean standard deviation of 0.25 ± 0.06 µm in 18 axons), so we assessed the effect of this by simulating a situation where alternate nodes had a length 0.25 µm shorter and 0.25 µm longer than the mean node length. For example, for a mean node length of 1.0 µm, alternate nodes had lengths of 0.75 µm and 1.25 µm. When the number of channels was held constant at each node, the approximately linear relationship between conduction velocity and node length (Figure 3A,B) resulted in a slightly faster propagation through shorter nodes and a slightly slower propagation through longer nodes, the effects of which cancel out with no overall effect on conduction velocity. When the nodal ion channel density was kept constant, the concave-downwards dependence of speed on node length in Figure 3A–B resulted in marginally slower conduction speeds (<2% reduced) with variable length nodes, compared to when all the nodes along a particular axon had exactly the same length.

It has recently been reported that internode lengths may vary significantly in the cortex (Tomassy et al., 2014; Chong et al., 2012). We measured internode lengths in dye-filled axons in rat layer V cortex by measuring the distance between NaV1.6 positive nodes. Similar to previous studies we found a large variability in the length of 30 internodes, the shortest being 27 µm and the largest 154 µm, with a mean value of 82.7 ± 6.3 µm. Given this large variation, we examined how internode length (assumed, for simplicity, to be the same for all internodes along an axon) affects the tuning of conduction speed by changes of node length (Figure 3C–F). For cortical axons (with a constant nodal sodium channel density), the peak of the dependence of conduction speed on node length is displaced to longer node lengths when the internode length is increased (Figure 3C), and in an axon with internode lengths of 154 µm the range of node lengths measured could generate changes in conduction speed of up to 42%. When the number of channels was held constant at each node, the dependence of conduction speed on node length was greater in axons with short internodes (27 µm), resulting in changes in conduction speed of up to 38% over the range of measured node lengths (Figure 3D). Changes of node length also affect the predicted dependence of conduction speed on internode length (Figure 3E,F). This relationship rises with internode length at low values of internode length, because more of the axon is myelinated, but decreases at large internode lengths as the spread of depolarization between nodes becomes less efficient. This relationship shows a sharper peak for shorter node lengths (Figure 3E,F).

We assessed how powerful node length changes can be for tuning conduction speed, compared to altering the amount of myelin around the axon. With the node length and internode length set at the mean values observed and the nodal conductances constant, if the number of wraps of myelin is decreased by 1 (from the normal 7 to 6 for the optic nerve and from the normal 5 to 4 for the cortex), the conduction speed is predicted to decrease by 8.6% for the optic nerve and 10.5% for the cortex (for simplicity no change of paranode length was assumed for these simulations, although the paranode may be shorter with less myelin wraps). For comparison, a similar speed decrease is predicted (with no change of wrap number) if the nodal NaV density is reduced by 26% in the optic nerve or 34% in the cortex, or if the internode length is increased by 36% in the optic nerve or 74% in the cortex. These same speed decreases can be achieved with no change of myelin wrapping or ion channel density by decreasing the node length from 1.02 µm to 0.625 µm in the optic nerve, or decreasing it from 1.5 µm to 0.635 µm in cortical axons. (Whether the internode needs to be lengthened by the same amount, to maintain axon length, and the consequences of this, are considered in the Simulations section of the Materials and methods). Strikingly, to produce these speed changes, the change of membrane area needed (shown in Figure 3G) when altering node length is 1006-fold less in the optic nerve, and 273-fold less in the cortex, than that needed when altering the myelin sheath. Thus, tuning conduction speed by altering node length is far more efficient than altering myelination when considering the amount of lipid and protein synthesis or breakdown needed, and would probably also be faster.


By combining anatomical measurements with computational modelling we have established a proof of principle for the idea that node of Ranvier length may be adjusted to tune axon conduction speed, and hence alter action potential arrival time. This differs from previous ideas on white matter plasticity (which have focussed on the effect of changes of myelin thickness and internode length (Fields, 2008; Ullén, 2009)), and demonstrates that the geometry of the node of Ranvier is also a crucial determinant of action potential conduction speed. Even at constant axon diameter, we found that node length displayed a surprising variability, both in the optic nerve and in the grey matter of the cortex (Figure 1). Remarkably, this node length variation was largely between different axons, while the nodes on any given axon tended to have similar lengths (Figure 2). The range of node lengths observed is sufficient to produce large variations in action potential conduction speed (Figure 3), comparable to those produced by adding or removing a wrap of myelin. These data suggest that axons may be able to adjust their node lengths in order to tune their conduction speed, and thus the arrival time of the information that they transmit.

There is now evidence of oligodendrogenesis and myelination in the adult CNS (Dimou et al., 2008; Young et al., 2013) which presumably results in the formation of new nodes of Ranvier. To avoid these, we studied only nodes that expressed the mature node marker NaV1.6 (Boiko et al., 2001; Kaplan et al., 2001), and the fact that the variability in node length was observed across a large fraction of the nodes measured makes it improbable that this variation is solely due to nodes being at different developmental stages. The node length was approximately proportional to the amount of sodium channel labelling at the node (Figure 1I), suggesting that node length may mainly be adjusted by the insertion or removal of membrane containing sodium channels (although conceivably it is possible to vary node length in a manner independent of sodium channel trafficking by endo- or exocytosis of vesicles lacking sodium channels in their membrane).

The fact that node lengths are similar over long distances along an axon (Figure 2B) raises three mechanistic questions. First, when the node length is set to a different mean value in different axons, by what mechanism is the nodal ion channel density controlled (Figure 1I)? Second, what signal regulates node length, in order to adjust the arrival time of action potentials at the end of the axon? Conceivably a signal could be passed back along the axon from a postsynaptic cell by dynein-based motors, as occurs for BMP signalling from postsynaptic cells to the nuclei of presynaptic neurons (Smith et al., 2012). Third, what local molecular mechanism regulates the length of each node, how accurately can this be controlled, and is the internode length shortened when the node is elongated (to preserve overall axon length)? Node length regulation may involve modifying the paranodal cell adhesion between the myelin and the axon, mediated by the molecules Caspr, neurofascin 155 and contactin, as well as altering ankyrin G mediated scaffolding within the axon that locates voltage-gated Na+ channels at the node (Arancibia-Carcamo and Attwell, 2014). Interestingly, nodal amyloid precursor protein has been proposed as a regulator of node length (Xu et al., 2014), prompting the speculation that changes in the processing of this molecule could alter node length in Alzheimer’s disease.

Computer simulations of the propagation of action potentials along myelinated axons (Figure 3) show that rather small changes in node length can produce quite significant changes of conduction speed. The range of node lengths seen in the optic nerve (0.5–2.2 µm) can alter the conduction speed by up to 20% (for a constant nodal channel density in Figure 3A), while the range seen in the cortex (0.43–3.7 µm) can produce a larger change of up to 25% (for a constant density of nodal channels in Figure 3B) in axons with 82 µm long internodes, or 38% for axons with 27 µm long internodes (Figure 3D). The effect of altering node length is larger in cortical axons than in the optic nerve, partly because the 1.5-fold longer nodes contribute a larger fraction of the total axon capacitance in the cortex where the internodes are also 1.7-fold shorter.

Our data and simulations suggest that modulation of node length could be a viable strategy for adjusting the propagation time of action potentials to meet information processing needs. For an intercortical callosal axon of length 1 cm (in rat) or 6 cm (in human), with the properties that we assume for our simulations, a 20% decrease of axon conduction speed (from the value occurring for the mean observed node length) would increase the time needed to propagate information between the cortices from 3.8 to 4.8 ms in rat and from 23 to 29 ms in humans. Such modulation has been suggested to occur during chronic stress and major depression (Miyata et al., 2016) which shorten the node, as well as high frequency action potential activity (Huff et al., 2011; Trigo and Smith, 2015), acoustic over-exposure (Tagoe et al., 2014), pathological release of glutamate (Fu et al., 2009) and hypoxic conditions (Reimer et al., 2011), all of which lengthen the node. Altering node length offers the advantage that very small changes of membrane area, which could easily be produced rapidly by exocytosis or endocytosis at the node, produce large changes of conduction speed. In comparison, to produce the same speed changes by altering the number of myelin wraps requires the energetically expensive (Harris and Attwell, 2012), and probably more time consuming, synthesis or disassembly of a membrane area that is 273–1006 fold larger. In practice, both mechanisms might be used on different time scales.

Materials and methods

Electron microscopy

For the optic nerve, 3 male (8–10 weeks old) Sprague-Dawley rats were anaesthetised and perfused through the heart with fixative containing 2.5% glutaraldehyde and 2% paraformaldehyde in 0.1 M cacodylate buffer. The optic nerves were dissected out, post-fixed with 1% OsO4 in 0.1 M cacodylate buffer, embedded in EPON and polymerised. Serial ultrathin (70 nm) sections, perpendicular to the nerve’s long axis, were cut on an ultramicrotome and picked up on an osmium-coated glass slide. Back-scattered images were obtained on a scanning electron microscope (Hitachi SU8010) with a working distance of 2 mm, 1–1.5 kV accelerating voltage, scan speed of 40 or 80 s, and a typical pixel size of 1.65 nm at x30,000 magnification, and were analysed with ImageJ (FIJI). Node length (assessed from the number of sections containing the node) and mean axon diameter at the node (assessed as axon perimeter/π) were measured (uncorrected for tissue shrinkage during fixation).

Tracer injections and node labelling

Four male 8–10 week old rats were anaesthetized with isoflurane and killed by cervical dislocation in accordance with United Kingdom animal experimentation regulations. After decapitation the brain was carefully dissected from the skull and 1 mm thick coronal slices containing the corpus callosum were obtained from the forebrain (from 4 to 8 mm rostral of the olfactory bulb) using a tissue cutter block. A 10% solution of tetramethylrhodamine dextran (MW 3000, Invitrogen, Paisley, UK) was iontophoretically injected into the cortical grey matter. Thereafter slices were incubated in oxygenated aCSF containing (in mM) 124 NaCl, 26 NaHCO3, 1 NaH2PO4, 2.5 KCl, 2 MgCl2, 2 CaCl2, 10 glucose, bubbled with 95% O2/5% CO2 for 2 hr at room temperature to allow for diffusion of the tracer. After incubation slices were immersion fixed in PFA and resliced at 80–100 µm for subsequent immunohistochemical labelling of nodal (NaV1.6) and paranodal (Caspr) marker proteins.

Immunohistochemistry and confocal microscopy

Optic nerves and 4 mm thick coronal sections of fronto-parietal (motor) cortex (from 4 to 8 mm rostral of the olfactory bulb) from brains of 4 male (8–10 week old) Sprague-Dawley rats were either perfusion or immersion-fixed in 4% paraformaldehyde in PBS. Fixed tissue was then cut into 50 µm slices using a Leica vibratome VT1200S or, for NaV1.6 density experiments, cut into 10 µm sections using a cryostat. Slices were blocked and permeabilised in 10% horse serum and 0.5% Triton X-100 in PBS. Immunofluorescence labelling was performed over 3 days with the following primary antibodies: rabbit anti-NaV1.6 (Alomone, 1:500); mouse anti-Caspr clone K65/35 (Neuromab, UC Davies, 1:100). Slices were then washed extensively (3 × 20 min) and incubated overnight with secondary antibodies: anti-rabbit AlexaFluor488 (Invitrogen, 1:500), anti-mouse Dy-Light 647 (Jackson Immunoresearch, 1:500). Slices were then washed 3 × 10 min in PBS and mounted with Dako Fluorescent Mounting Medium. Slices were viewed using an LSM700 or LSM780 confocal microscope using a 63x (NA 1.4) oil immersion lens, and images were acquired with LSM software with the pinhole set to 1 Airy unit for the Caspr signal, resulting in an optical slice of 0.8 µm. Pixel size was 39.7 nm for Figure 1A–H and 99.2 nm for Figure 1I and 52.7 nm for node measurements in Figure 2 and 263.6 nm for internode measurements in Figure 2.

Image analysis

For node length analysis, confocal images were analysed using ImageJ software. Images were background subtracted and only nodes that lay approximately parallel to the plane of section (i.e. displayed nodal NaV1.6 labelling with flanking Caspr-labelled paranodes all within a single 0.8 µm optical slice) were selected. Measuring the angle of the axon to the plane of the slice for a subset of 10 randomly chosen axons showed that the apparent node length measured in this way underestimated the actual node length by only 1.7% ± 0.6%. A maximum intensity projection was generated of the sections in which Caspr labelling was present for a particular node (up to five interleaved confocal slices at 0.38 µm intervals, with a maximum stack thickness of 2.32 µm), and a line intensity profile (the thickness of which was slightly less than the Caspr labelling thickness) was drawn spanning both Caspr-labelled paranodes. The size of the node was then calculated using a MATLAB (The MathWorks, Inc.) script which measures the distance between the half maximum intensity for each paranode. Node diameter, paranode length and axon diameter were measured using the line tool in ImageJ over NaV1.6 staining (for node diameter) and over the Caspr staining (for axon diameter and paranode length). NaV1.6 staining was summed over the nodal area to obtain a parameter assumed to be proportional to sodium channel number. Internode length was measured in three dimensions in FIJI using the simple neurite tracer plugin (Longair et al., 2011). Data were not corrected for tissue shrinkage during fixation.


To simulate action potential propagation along myelinated axons, we implemented, in MATLAB, model C of Richardson et al. (2000) (at 37°C, with the unphysiologically low membrane capacitance of Richardson et al. (2000) corrected to a normal value). The differential equations of the model were derived and solved as in Halter and Clark (1991). In brief, the axon is divided into compartments representing the node, paranode and internode. For each time step, current flow across the axonal or total myelin membrane is calculated from the values of voltage (and its rate of change), and the membrane capacitance and membrane conductances present per unit length (simultaneously solving the differential equations that define activation and inactivation of the voltage-gated currents present at the node), and intracellular and periaxonal axial current flow are calculated from the intracellular or periaxonal resistance per unit length and the gradient of intracellular or periaxonal voltage. Details of the parameters used are summarised in Table 1. The MATLAB code used can be obtained immediately on request from the authors; it will be written up and documented as a resource for free access from GitHub by August 1st 2017.

Simulations were carried out as in Bakiri et al. (2011) except that the periaxonal space under the myelin was included (51 nodes were simulated and conduction speed was measured between nodes 20 and 30). This model includes fast and persistent Na+, and slow K+, voltage-gated channels at the node, but omits voltage-gated K+ channels at the juxtaparanode (which are little activated because the 100 mV voltage change of the action potential is distributed across the 11–15 membranes of the 5–7 myelin wraps and the axon, implying only a 7–9 mV voltage change across the axonal membrane). For simplicity, the node length was usually assumed to be the same at all nodes on the axon, i.e. we ignored the variability in node length along the same axon described in Figure 2. The node diameter was set to the mean value measured experimentally, i.e. 0.73 µm (in 164 nodes) for the optic nerve, and 0.64 µm (in 158 nodes) for cortex. The region between two nodes, 139.3 µm long for the optic nerve (Butt et al., 1994) and 81.7 µm long for the cortical axons (the mean value measured in layer V from 30 internodes, see main text), was kept constant when node length was varied (both for simplicity, and because there was no correlation of node length and internode length: Figure 2E). This internodal region was divided (along its length) into 66 and 86 compartments for the optic nerve and the cortex, respectively, the end 2.11 and 1.90 µm parts of which represent the paranode where the myelin attaches to the axon (values measured in 164 and 158 nodes, respectively, from the length of the Caspr labelling; the number of compartments used has to be large for the simulation to be accurate, and needs to be chosen so that an integral number of compartments can represent the paranodal junction; the number was adjusted appropriately when simulating different internode lengths). The internodal axon diameter is larger than the diameter at the node (Halter and Clark, 1991; Berthold and Rydmark, 1983), although this difference is a much smaller percentage for small than for large axons (Rydmark and Berthold, 1983). The internodal and paranodal axon diameters were set to 0.82 µm and 0.73 µm for the optic nerve and cortex, respectively (mean values obtained from 164 axons in optic nerve and 158 cortical axons from the diameter of the paranodal Caspr labelling), so that the node diameters were 88% and 86% of the internodal axon diameters respectively. Apart from at the paranodes, the internodal axon was assumed to be surrounded by a periaxonal space of thickness 15 nm (Robertson, 1959; Mierzwa et al., 2010; Möbius et al., 2016), and to have a g ratio (axon diameter/myelin diameter) of 0.79 in the optic nerve and 0.8 in the cortex (Sugimoto et al., 1984; Oorschot et al., 2013; to obtain an integral number of myelin wraps these values were adjusted slightly, to 0.78 and 0.81 respectively). This led to the optic nerve and cortical axons having 7 and 5 myelin wraps, respectively assuming a myelin wrap periodicity of 15.6 nm (Agrawal et al., 2009; Harris and Attwell, 2012). The periaxonal space at the paranode, because of the structure of the attachment of the myelin to the axon at the paranode, is thought to comprise (Mierzwa et al., 2010) a pathway of cross sectional area A = 170 nm2, which spirals around the axon (from the node to the periaxonal space of the internode) for a total distance of approximately D = π.d.Nwraps where Nwraps is the number of myelin wraps and d is the axon diameter. A periaxonal space of width w, along a paranode of length L, would have the same resistance as this pathway ifL/(π.d.w)=D/Aorw=A.L/(π.D.d)=A.L./[(π.d)2.Nwraps]

The effective value of w used to model this spiral pathway for the optic nerve and cortical axons was thus 0.0077 nm and 0.0123 nm respectively.

In the main text we present calculations showing that a given change of conduction speed can be produced far more efficiently (in terms of the change of membrane area needed) by shortening of the node than by adding another wrap of myelin. Those calculations ignore the possibility that, when the node is shortened, the internode needs to be lengthened by the same amount in order to maintain the axon length. It is unclear whether the sub-micron node length changes postulated in our calculation would actually require remodelling of the adjacent myelin sheath – conceivably slackness in the somewhat non-straight internode would allow the change of node length to be accommodated without a change of internode length, and furthermore node length might change by an eversion of the paranodal loops closest to the node without any other significant change to the myelin sheath (reviewed by Arancibia-Carcamo and Attwell, 2014). Thus, small changes in nodal length might well occur without major remodelling of the myelin sheath. Nevertheless, if one assumes that node shortening by X µm absolutely does require an X µm elongation of the myelin sheath, then more membrane changes are needed than are accounted for in our simple calculation. One can show mathematically that the sheath membrane area increase is larger than the area decrease at the node by a factor of 2x(number of myelin wraps)x(mean radius of wraps)/(node radius), which is roughly 18.4 for the optic nerve and 13 for the cortex. Accounting for these area changes (and noting that, in this situation, there is no change of total axon length) would reduce the ratio of the membrane area changes needed to produce a given speed change (when adding a layer of myelin to the sheath versus changing the node length) from 1006-fold to 55-fold for the optic nerve and from 273-fold to 21-fold for the cortex, but these ratios remain impressively large, and so the energetic argument favouring speed tuning by alteration of the node length still holds.


Data are shown as mean±s.e.m. Comparisons are via 2-tailed Student’s t-tests unless otherwise stated. Assessment of whether the slope of linear regressions differed significantly from zero was obtained using the t-statistic for the slope.



Supported by a Wellcome Trust Senior Investigator Award to DA, a Wellcome Trust PhD studentship to LC, a Marie Curie fellowship to MCF, and JSPS grants (24650181 and 25245069) to KT. We thank Boris Barbour, Beverley Clark, Renaud Jolivet, Josef Kittler, Anna Krasnow and Angus Silver for comments on the manuscript.

Decision letter

Klaus-Armin Nave, Reviewing editor, Max Planck Institute for Experimental Medicine,, Germany

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: a previous version of this study was rejected after peer review, but the authors submitted for reconsideration. The first decision letter after peer review is shown below.]

Thank you for choosing to send your work entitled "Ranvier node length and myelinated axon conduction speed" for consideration at eLife. Your full submission has been evaluated by Eve Marder (Senior Editor), a Reviewing Editor, and two peer reviewers, and the decision was reached after discussions between the reviewers. Based on our discussions and the individual reviews below, we regret to inform you that we are forced to reject this version of the manuscript.

The reviewers were quite conflicted because they really were taken by the aims and goals of the paper, and thought it is potentially important. Nonetheless, they were troubled by a variety of technical issues that are perhaps captured by their feeling the manuscript suffered by comparing "apples and oranges". Virtually all of the reviewers' comments could be dealt with, but eLife has a strong policy to not require extensive new experimental work when allowing a revision. Therefore, when reviewers really ask for extensive experimental work (work that we expect would take more than a month to do), we are forced to reject the manuscript. This frees the authors to take their manuscript elsewhere if they choose, as it is no longer in consideration at eLife. If, however, you feel in retrospect that the reviewers are, on balance, correct, and if you can deal with these critiques, we would be willing to entertain a new submission, which would be treated as such, at some time in the future.

Reviewer #1:

This brief manuscript combines an assessment of the variation in length of nodes of Ranvier (the distance separating adjacent internodes) along myelinated axons in the rat CNS with modeling to estimate what effect these variations might have on the conduction velocity of action potentials. The extent and character of myelin is regulated by neuronal activity during development, and there has is renewed interest in the possibility that myelination may be modified throughout life. The effects of altered myelination could be exerted at many levels – myelinated versus unmyelinated, thickness of myelin, length of internodes, and length of nodes, not to mention changes in NaV and Kv densities along axons. However, little is known about the variations in these parameters in the adult CNS. Defining these variations, exploring how they are affected by life experience and determining their effects on information processing are important goals. Here the authors show that there is greater variation in node length along myelinated axons in brain than in optic nerve, and through modeling show that this variation in brain would be expected to have substantial effects on conduction velocity.

1) The study uses two different methods to determine the length of nodes – EM data from optic nerve and immunofluorescence for brain. It isn't clear why different methods were used for the two preparations. Some assurance should be provided that the differences seen are not due to the use of different methodologies. In particular, the authors should estimate what error might be induced by analyzing nodes that are in different orientations in the brain samples and indicate how orientation was determined. It would be helpful if a dozen examples spanning the range of node lengths were illustrated in Figure 1.

2) The study compares tissue from animals that were different ages – 8-12 weeks for the optic nerve and P30 for the brain. It isn't clear why different ages were examined. Oligodendrogenesis and myelination are still ongoing at P30 in brain, raising the possibility that the variation in node length reflects an intermediate developmental state. As internodes grow together to form nodes, some of the longer nodes could represent internodes captured at this stage. As the development of myelin varies across cortical layers and in different regions of the cortex, the authors should provide information about the regions of the cortex that were examined.

3) As NaV density is a crucial parameter, it would be valuable to examine NaV immunoreactivity at these nodes. Although not quantitative, this could provide some indication of whether NaV density or NaV number is held constant. Also, NaV staining would help clarify if the longest nodes are merely two hemi-nodes prior to consolidation.

4) If node lengths varied from 0.3 to 4 μm in the cortex, it appears that only a partial dataset is shown in Figure 1D or that the x-axis is mis-labeled.

5) The plot on Figure 3A should extend from.4 -1.4, because this is the range of node lengths observed, or at least, this region should be highlighted in the graph.

6) It seems most relevant to highlight the effect that the observed changes in node length would have on conduction velocity in Figure 3. In particular, although the range is larger in brain, the distribution in Figure 2A suggests that ~90% of nodes fall within 0.5 to 2 um, which the modeling suggests would have minimal effect on conduction velocity.

7) The prediction about the length of successive nodes on the same axon should be validated by examining node length along individual axons using available serial EM (TEM or SEM) datasets.

Reviewer #2:

Arancibia-Carcamo et al., submit "Ranvier node length and myelinated axon conduction speed" for consideration at eLife. Numerous recent studies have indicated that myelin in the central nervous system can be generated and modulated well in to adult life. The manuscript by Arancibia-Carcamo et al., serves as a timely reminder that modulation of myelin is not the only way in which the functional output of myelinated axons can be modulated and fine-tuned. The study is primarily a modeling based study built on the observations that nodes of Ranvier vary in length (independently of axon/ node diameter) in both the optic nerve, and to a greater extent, the cerebral cortex. The main predictions of the model are that relatively subtle changes in node of Ranvier length can profoundly affect conduction speed, and that such nodal changes are likely very energy efficient compared to the changes in myelin thickness that would be required to elicit similar effects on speed. This core conclusion/prediction is certainly provocative and interesting, and provides a simple hypothesis that will require experimental investigation in future experiments out-with the scope of this manuscript. However, the manuscript itself needs significantly more detail in parts before it is ready for publication, and could some additional considerations could be integrated into the model to provide more depth. The study is, in my opinion, of sufficient general interest to the readership of eLife.

1) Ultimately there are two main concepts here, that changing nodal length can affect conduction speed and that this may be a far more energy efficient way to fine tune function than modulating myelination. The first concept has been delineated nicely given the parameters applied, but the second dealt with a little more cursorily (by simply pointing out that different membrane areas would be involved in remodelling a node versus myelin). Could the authors consider the following points in their model?

A) What would the effect of conduction be on simply changing ion channel density at the node and keeping node length the same? One could imagine that the targeting of channels to nodes of a similar size might represent an even more straightforward manipulation that required no change in membrane area at all. How could this effect conduction speed?

B) Could the authors take into account the consequences of changing node length along the length of the axon and take these into consideration with respect to energy requirements (in which the corresponding author is also expert)? If, for example, an average myelin sheath length in the cortex is say 29um+ 1um for the node and the axon is 3mm long then there would be 100 myelin sheaths along its length. If a 1um node were to increase in length to 3um then 200um of space along the axon would be required to accommodate the longer nodes. This would still require significant remodelling of myelin along the axon, either through small changes in many sheaths or the removal of a small number of sheaths and subsequent readjustment.

I don't know if the authors can easily integrate such considerations into their model and account for relevant energy requirements, but the point is that the simple comparison of membrane area seems quite simplistic, particularly given that this is a core conceptual point of the study.

2) The diameter of the axon is constricted at the node and in particular at the paranodal axon-glial junction. The authors have not really dealt with this point. They mention that values as to the relative axonal sizes have been determined by "unpublished observations of 10-12 axons," but it is not clear how many nodes this reflects, nor more importantly what the variability in internode to node diameter is. Perhaps this is a very important parameter that could profoundly affect functional output and is it being overlooked here. More details and consideration will be important, particularly as at least one previous study has investigated how constriction of diameter at the node affects conduction.

3) There are essentially no details relating to the model itself in the manuscript. Input values for various parameters are provided and one sentence referring to modifications of previous models described in two references. Without getting in to the details of the Richardson et al. paper in particular readers of eLife will be left quite in the dark about the details of the model being used and indeed of the alternatives. The pros and cons of different models have not been discussed, and so it is impossible to review the modelling aspect per se. The authors should provide the details of each model, at least as an appendix or as supplementary data, giving due recognition to other sources as required?

This is clearly not essential, but, if appropriate, would it be possible for the authors to generate an online interface so that the community could take advantage of the model? This could be an incredible resource.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for resubmitting your work entitled "Node of Ranvier length and myelinated axon conduction speed" for further consideration at eLife. Your article has been favorably evaluated by Eve Marder (Senior Editor) and three reviewers, one of whom is a member of our Board of Reviewing Editors.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below:

The essential experimental findings of this resubmitted manuscript are that in CNS nodal length varies over an approximate 4.4 to 8.7-fold range, but is more consistent along individual axons. Little correlation is seen between nodal length and width, which is taken as evidence that nodal length can be regulated at an axon-specific level independently of axonal diameter. Computer modeling predicts that the range of nodal lengths observed will have a biologically significant effect on conduction (in the order of ~20-38%). The authors therefore propose that regulation of nodal length is a viable method for fine-tuning conduction velocity. They make a compelling case for it being a less energy/resource demanding form of plasticity than others currently under investigation (e.g., myelin thickness, de novo myelination). The paper is thought-provoking and of sufficient interest for the readership of eLife.

In general the manuscript is considerably improved, being more detailed and streamlined. The authors have addressed the initial comments in detail. Much of the in vivo relevance relies on the strength of the new data where individual axons have been labelled and along which node length has been measured to determine relative variability along and between axons. This is a very nice approach and the authors should be commended for this revision. Although many of the concepts proposed will require extensive experimental investigation, this study in and of itself is important in that it provides numerous key hypotheses that can and will be tested in the future.

1) A new reviewer (on the second round) notes: I am a little reluctant to shift the goalposts on the authors who seem to have responded fairly well to the first set of reviewer comments. Nevertheless, I feel that the language of the manuscript sometimes goes beyond what is demonstrated (e.g., "Thus, axon-specific adjustment of node of Ranvier length is an energy-efficient and potentially rapid mechanism for tuning the arrival time of information in the CNS"). It is not shown that nodal length is modified during plasticity, or even that it varies between relevant axons in naturally occurring systems where adaptations need to be made to ensure synchrony (such as in auditory brainstem circuits, where the authors have previously studied internode length and axonal diameter). In the absence of such data, it should be made very clear in the Abstract that this paper identifies regulation of nodal length as a potential mechanism for neuroplasticity only. Indeed, the authors must make sure that also the superficial reader will not confuse fact and fiction. "Fact" and well documented is the observation that nodes are more similar in size along an individual axon than the nodes of different but otherwise comparable axons. Still "fiction" is the idea that nodes can change in size (with predictable consequences on axonal conduction velocity) as a physiological mechanism, by which neurons adapt axonal conduction to the demands of a larger neuronal network. For comparison, had the authors been the first to discover that axons differ in caliber – which of course is well known – we could have the same discussion. In fact, it might be easier for neurons to fine-tune their axonal calibers (e.g. by Akt/mTOR signaling) than to remodel nodal domains individually in a concerted effort with oligodendrocytes. Thus, there was consensus of all reviewers that the language needs to make much clearer (also in title and Abstract) that the main conclusions that make the paper so interesting are hypothetical/ theoretical.

2) Figure 2: One wonders how the relative variability along versus between axons would look without inclusion of the two axons with longer nodal lengths ("outliers", late developmental stages? See below). One worries to what extent the increased variance between axons is driven by these two axons. Perhaps the authors can simply address this comment without further information, but the presentation and annotation/ explanation of Figure 2C is difficult to interpret. If the removal of the two axons with the larger nodes removes the differences between versus within axons then it would be reassuring to see a slightly larger number of axons such that one didn't worry about any outlier or sampling artifacts. How representative are the 3 axons shown in Figure 2B? Can the authors address this issue by analysis of additional axons (in the hope of catching more that have nodal lengths in the 2-3μm range) or by a separate analysis that excludes these two axons)?

3) In response to the concern that nodes may be still be immature at P30, the authors repeated their analysis with cortices at age 8-10 weeks. However, cortical myelination may not be finished by 2 months either, because even at age 6 months (NG2CreERT2-based) OPC lineage tracing experiments revealed newly generated oligodendrocytes (Dimou et al., J Neurosci. 2008) that presumably make myelin with immature nodes of Ranvier. This should be discussed.

DOI: http://dx.doi.org/10.7554/eLife.23329.008

Author response


If your username is different from your full name, we require you to identify yourself within the comment itself. Comments are checked by a moderator (and/or an eLife editor) before they appear. Comments should be constructive, relevant to the article, conform to our terms and conditions, and include any pertinent competing interests.