Superresolution imaging reveals activity-dependent plasticity of axon morphology linked to changes in action potential conduction velocity

We adapted a custom-built STED microscope for live-cell imaging of GFP-labeled axons in organotypic brain slices, using an oil objective, providing a spatial resolution of around 50 nm close to the coverslip (<10 μm; SI Appendix, Fig. S1). This approach (Fig. 1 A–C) revealed a large morphological diversity for axons of CA3 pyramidal neurons [median axon diameter = 203 nm (range = 58–446 nm; n = 26,574 measurements over ∼550 µm axon length, excluding boutons; R² = 0.99, from 63 axon segments and 14 slices; Fig. 1D); median bouton size area = 0.53 µm² (range = 0.2–2.2 µm²; n = 319 synaptic boutons; R² = 0.98, from 63 axon segments and 14 slices; Fig. 1E); see SI Appendix, Fig. S2 for details on morphometric analysis], in agreement with previous measurements based on electron microscopy (10). Axon diameters varied more across than within axon segments (CV_across = 18.1%; CV_within = 6.1 ± 2.8%; P < 0.001; n = 52 axon segments from 24 slices; one sample t test; Fig. 1F), which may reflect consistent size differences between individual axons.

Time-lapse STED imaging showed that axons were largely stable during baseline conditions [for axon diameters, P > 0.54 (n = 113 individual 1-µm-long segments); for boutons, P > 0.40 (n = 63 boutons from five slices); Kruskal-Wallis tests; Fig. 1 G–I], in line with previous reports based on conventional light microscopy (4, 12).

Confocal image analysis of myelin basic protein immunostaining revealed that the vast majority of GFP-labeled CA3 axons were unmyelinated in the CA1 and CA3 areas (myelin basic protein-positive fraction: 0.08 ± 0.07%, from three slices; Fig. 1 J and K and SI Appendix, Fig. S3), which is consistent with previous reports in the literature (11, 13).

To test whether axon morphology can be regulated by neuronal activity, we applied high-frequency electrical stimulation (HFS) to the area of labeled CA3 neurons. Shortly after stimulation, synaptic boutons were enlarged relative to baseline levels (at 2 min: +20.7 ± 3.0% corresponding to a ∼33% increase in bouton volume; P < 0.001; n = 147 boutons from five slices; one-way ANOVA; Fig. 2 A and B). However, this enlargement was mostly transient, decaying within minutes to slightly above baseline levels [at 8 min: +10.1 ± 2.5% (P = 0.006); at 14 min: +4.3 ± 2.4% (P = 0.72); Fig. 2B].

In contrast, axon diameters did not exhibit any statistically significant structural changes during the same period after HFS (at 2 min: +0.4 ± 1.9%; at 14 min: +3.6 ± 1.9%; P > 0.37; n = 148 1-μm-long shaft segments; Fig. 2 A and C).

We then patched individual CA3 pyramidal neurons with Atto 488 dye in the patch pipette (SI Appendix, Fig. S4A) and measured the structural changes in boutons in response to APs elicited by brief current injections in whole-cell current clamp mode. Confocal image analysis showed that AP firing led to a rapid enlargement of bouton size (1 min after the stimulation: +12.3 ± 3.2%; P < 0.001; SI Appendix, Fig. S4 B and C), which increased with the number of elicited APs (R² = 0.4; SI Appendix, Fig. S4 D and E). In the presence of cadmium in the bath solution (Cd²⁺; 100 μM), which blocks voltage-gated Ca²⁺ entry and synaptic release, the bouton changes were suppressed, even though AP firing was unaffected (−0.9 ± 1.1%; P > 0.79; SI Appendix, Fig. S4 F and G).

To evaluate the persistence of the structural changes in boutons and axon shafts, we carried out additional STED time-lapse imaging experiments over a longer period (up to 60 min in total; Fig. 3A). Although bouton size remained slightly above baseline levels until the end of the observation period (at 45 min: +5.6 ± 1.4%; P < 0.001; Fig. 3B), axon shafts increased gradually over tens of minutes (at 33 min: +4.3 ± 0.9%; at 45 min: +5.2 ± 0.7%; P < 0.001; Fig. 3C). Moreover, HFS induced robust long-term potentiation (LTP) at CA3/CA1 synapses (+65 ± 16%; P < 0.001; 30–40 min after HFS; n = 8 slices; two-sample t test; Fig. 3D; recorded concurrently with the experiments shown in Fig. 3 B and C). In control experiments (without HFS), neither structure showed any significant size changes (Figs. 2 B and C and 3 B and C).

Importantly, these average values underestimate the actual effect size because some of the imaged axons may not have been fully or not at all stimulated by the HFS, yet they were included in the analysis (SI Appendix, Fig. S5). In line with this, a sizeable fraction (>20%) of axon segments showed much larger increases in diameter (>10%, corresponding to a ∼21% increase in cross-sectional area), whereas others appeared not to undergo any structural changes at all (SI Appendix, Fig. S6).

We applied (2R)-amino-5-phosphonovaleric acid (APV) to test whether these structural effects depended on NMDA receptor activation. Although they were prevented on the long-term [at 45 min: bouton size: +1.4 ± 1.3% (n = 109 boutons; P = 0.95); axon diameters: +2.1 ± 0.6% (P = 0.28; n = 317 1-μm-long shaft segments); Fig. 3 B and C] and LTP was blocked (30-40 min after HFS, P > 0.18; n = 8 slices; Fig. 3D), the initial enlargement of synaptic boutons was not blocked by the drug (at 2 min: +21.9 ± 2.8%; P < 0.001; n = 148 boutons; Fig. 2B).

In addition, we also used a more physiological pattern of hippocampal activity (theta-burst stimulation) as a current injection protocol in whole-cell patch-clamp mode. Theta-burst stimulation-induced patterns of AP firing in the CA3 neurons led to similar morphological effects on CA3 boutons and axon diameters, in terms of both direction and time course (SI Appendix, Fig. S7).

Given the influence of axon diameter on the axial resistance of axons, we wondered whether these morphological changes might affect AP conduction velocity. To explore this possibility, we used compartmental modeling to simulate AP propagation along unmyelinated axons as a function of morphological changes in boutons and axon shafts.

Increasing axon diameters accelerated AP conduction along the axons in the model (Fig. 4 A and C and SI Appendix, Fig. S8), which is in line with cable theory (6, 7). Wider axons have less internal electrical resistance, and thus longer space constants, which effectively facilitates the electrotonic spread of the AP. Interestingly, enlarging synaptic boutons affected AP conduction velocity in exactly the opposite way (Fig. 4 B and C). This effect can be explained biophysically by the extra time it takes for propagating APs to charge up membranes of enlarged synaptic boutons. For axon shafts, the axial resistance is inversely proportional to the square of the radius (∼1/r²), whereas the capacitance of the axon scales linearly with radius (∼r), explaining why the accelerating effect on AP conduction wins out in the axon shafts. In contrast, changes in bouton size mostly affect the capacitance of the boutons and, to a much lesser extent, the axial resistance of the axons, which accounts for the slowing down effect predicted by the simulations.

In contrast to the effects on conduction velocity, the waveform of the AP was barely affected by any changes in morphology (SI Appendix, Fig. S9). This is probably because of the local regenerative nature of the AP, which acts as a reset that counteracts any potential build-up of changes in AP waveform during AP propagation. Moreover, the simulation results did not depend on whether the number or density of voltage-gated Na⁺ and K⁺ channels were kept constant (SI Appendix, Fig. S10).

We then simulated AP propagation based on the time course and magnitude of the morphological changes we experimentally observed in the axons (Fig. 4D). The simulations predict biphasic effects on AP conduction velocity, showing an initial increase in AP conduction delay followed by a sustained reduction (Fig. 4E).

To estimate the cumulative effects on AP propagation by changes in conduction speed along the axon, we calculated the changes in conduction latency as a function of axon length, which can be more than 22 mm for Schaffer collateral axons (14). As thin unmyelinated fibers conduct APs rather slowly (< 0.2 m/s) (15), the cumulative changes in latency grew into the range of several milliseconds over relatively short AP travel distances (< 10 mm; Fig. 4 F and G). Depending on whether the structural changes affected the boutons or axon shafts, the latency either increased or decreased. Notably, the magnitude of the structural changes where the simulated delays became sizable was within the experimentally observed range.

To test the model predictions on AP conduction velocity, we performed electrophysiological measurements. We measured AP latency in response to antidromic stimulation of CA3 afferents by whole-cell patch-clamp recordings (Fig. 5A) before and after application of the HFS protocol. In parallel, we recorded field potentials in CA1 stratum radiatum to confirm successful LTP induction. We checked that the latency measurements were not contaminated by synaptic responses by blocking glutamatergic transmission (SI Appendix, Fig. S11), which silenced excitatory field potentials (fEPSPs) in the CA1 area, but not the antidromic responses in the CA3 neurons (Fig. 5B). Although AP latency was highly reproducible for a given neuron, it varied substantially across different neurons (CV_within = 1.8 ± 0.1%; CV_across = 30.5%; P < 0.001; n = 29 cells from 29 slices; one-sample t test; Fig. 5 C and D).

Immediately after HFS, antidromic responses in CA3 pyramidal neurons were substantially delayed (range = −92 to +562 μs; +3.3 ± 1.5%; P < 0.025, n = 10 cells; one-tailed paired t test, Fig. 5 E and F). However, AP latency was increased only for a few minutes and then decreased below baseline levels 15 min after HFS, until the end of the ∼45 min of recordings (range = −326 to +29 μs; −2.2 ± 0.77% at 30–40 min; P < 0.02; n = 10 cells; Fig. 5F). At the same time, LTP was robustly expressed throughout this period (+81 ± 22%; P < 0.006; n = 10 slices; two-sample t test; Fig. 5G). Applying APV, we also observed an initial increase in AP latency right after HFS, but did not observe any delayed decrease (range = −49 to +60 µs; −0.12 ± 0.42%; n = 8 cells; P > 0.28; Fig. 5 F and H), mirroring the effects of APV on the time course of the morphological changes.

This qualitative agreement between the model predictions (Fig. 4E) and the direction and time course of the electrophysiological changes (Fig. 5F) supports the view that the effects on AP latency were, at least in part, caused by the activity-induced changes in axonal morphology, likely reflecting a biophysical consequence of an altered electrotonic structure of the axon. As the electrophysiological responses were measured over relatively short distances (<1 mm), the observed changes in conduction delay were accordingly small (<0.5 ms).

What physical mechanisms could account for the size increases? Physical stretching of the axolemma is unlikely to contribute much to the observed size increases because biological membranes exhibit modest elastic properties (24). In contrast, rapid redistribution and/or de novo synthesis of membrane are more likely to provide the source of the activity-dependent net enlargement of the axons.

The time course of the morphological changes suggests a scenario whereby a massive exocytosis of neurotransmitter vesicles induced by strong AP firing caused an initial increase in bouton size, which in turn led to delayed widening of the axons as some of the newly added membrane diffused away from the boutons to the adjacent axon shafts.

The pharmacological experiments support this hypothesis because Cd²⁺, which blocks synaptic release, abolished the initial enlargement of the boutons, whereas APV did not affect the bouton changes. In contrast, APV blocked the long-term morphological and functional effects, indicating that their consolidation requires NMDA receptor/Ca²⁺-signaling. Moreover, a recent electron microscopy study showed NMDAR-dependent long-term depletion of axonal vesicles after LTP induction (25).

The axonal swelling may also reflect an increase in the number of transport vesicles in the axon that have been shown to replenish the synaptic vesicle pool (26) or to deliver protein material to nascent synapses (27). However, the addition of such vesicles would make the intra-axonal space more crowded, and thus counteract the effect of axonal widening on AP conduction velocity.

Changes in AP conduction latency may affect several important aspects of hippocampal and cortical physiology, which are sensitive to (sub)millisecond timing variations. They may be important for spike-timing dependent plasticity, as the time window for supralinear summation of calcium caused by dendritic spikes and synaptic events is on the order of a few milliseconds and determines whether synapses get potentiated or depressed (28). A change in the exact time of arrival of excitatory inputs may also affect the dynamic balance of excitatory and inhibitory drive, which controls postsynaptic spiking (29) and network oscillations (30). Because the effects scale with axon length, the transfer of information to distant targets via long-range fibers may be particularly affected by changes in spike velocity, shaping the correlation and integration time constants for neuronal activity across different brain areas (31). Although these speculations apply in general to positive and negative changes in conduction delay, their directionality, in addition to their magnitude, may very well determine the nature and extent of the downstream functional consequences.

The morphological changes may also affect the fidelity of AP conduction by changing the electrical impedance at axon branch points, where AP propagation failures might occur (32–35). In addition to these electrical effects, changes in axon diameter may influence the spread of signaling molecules along the axon, imposing a variable diffusion barrier that could regulate the degree of biochemical compartmentalization of presynaptic boutons, akin to the situation for the neck of dendritic spines (36). Finally, the morphological changes may perturb the structural integrity of the boutons and the nanodomains that underlie tight excitation-release coupling; for example, by altering the spatiotemporal profile of presynaptic Ca²⁺ transients or the access of the vesicular pool to the active zone, which may affect synaptic release probability or short-term plasticity.

Organotypic hippocampal slices (Gähwiler type) were dissected from 5–7-d-old wild-type mice and cultured 3–5 wk in a roller drum at 35 °C, as previously described (37). Experimental procedures were in accordance with the European Union and CNRS UMR 5297 institutional guidelines for the care and use of laboratory animals (Council directive 2010/63/EU) and approved by the Committee of Ethics of Bordeaux (no. 50120198-A).

fEPSPs were recorded from the CA1 stratum radiatum, using glass microelectrodes (tip resistance, 3–5 MΩ) filled with artificial cerebrospinal fluid (ACSF). To increase the overlap between the labeled and stimulated fibers, a monopolar stimulating electrode (3-5 MΩ filled with ACSF) was placed in the center of the labeled CA3 cell bodies. Brief current pulses (0.2 ms) were delivered from a stimulus isolator (AMPI; Science Products). Stimulus strength (ranging between 10 and 35 µA) was adjusted to elicit fEPSPs at 50% of their maximal amplitude. fEPSPs were recorded using a patch clamp amplifier (Multiclamp 700B; Molecular Devices) with a gain of 20× and a Bessel filter set at 2 kHz. A test pulse was recorded every 15 s throughout the recordings. The stimulus protocol consisted of three trains of 100 pulses delivered at 100 Hz and 20 s apart to induce LTP and was started after recording a baseline of 20 min. The initial slopes of the fEPSPs were determined with Clampfit 10.1 (Molecular Devices). CA3 pyramidal neurons were recorded in current clamp mode with a dynamic current injection correction to maintain the resting membrane at −70 mV. The gain of the patch clamp amplifier was set at 1×, and a Bessel filter with a cutoff frequency of 2 kHz was applied. The data were digitized at a sampling rate of 250 kHz. To elicit antidromic APs, a glass electrode back-filled with ACSF was positioned in the distal part of the CA1 area, and brief current pulses (0.1 ms, 25–75 µA) were delivered from a stimulus isolator. The electrode was moved toward the CA3 area until an antidromic AP could be recorded reliably at the soma. Recordings with access resistances larger than 25 MΩ or holding currents greater than −100 pA were excluded from the analysis. fEPSP recordings in the CA1 stratum radiatum were performed concurrently, using the same stimulation electrode.

We used a home-built STED microscope based on an inverted microscope (DMI 6000 CS Trino; Leica), using a galvanometric beam scanner (Yanus IV; TILL Photonics) and a high numerical aperture objective lens (PL APO, 100×, oil, NA 1.4; Leica), as described in detail previously (38).

Image acquisition was controlled by the software Imspector (A. Schönle, MPI for Biophysical Chemistry). Image acquisition parameters were chosen to minimize photodamage and photobleaching. All images were acquired with time-averaged powers of <6 µW for excitation and <8 mW for STED (measured at the back aperture of the objective). Image stacks were acquired with a voxel size of 19.5 nm (xy), 375 nm (z) and a dwell time of 15 µs (SI Appendix, Fig. S12). The imaging depth was controlled by a piezo z-focusing device (Physikinstrumente) and was maximally 15 µm below tissue surface. For the short-term experiments, z-stacks of 40 × 40 × 3 µm (x, y, and z) were acquired every 6 min, whereas for the long-term experiments, z-stacks of 30 × 20 × 3 µm were acquired every 12 min for 1 h, where two image stacks were acquired before the stimulus and four image stacks afterward.

We used NIH ImageJ and Matlab scripts for image processing and analysis. Analysis was done on raw data, whereas a one-pixel median filter was applied to the images shown in the figures. A maximum-intensity projection was applied on identified axon segments. To perform automated measurements, a reference image that consisted of a line of pixels running along the axon segment was produced by thresholding and binarizing the images before using a skeletonization plugin from ImageJ. The full width at half maximum (FWHM) was measured automatically along the axon guided by the reference image. For every position, intensity profiles from three parallel lines were averaged to reduce noise. Diameter profiles of full axon segments were generated and analyzed further (SI Appendix, Fig. S2). To determine bouton size changes over time, bouton area was calculated as the integral of the local diameter profile. Similarly, axon shaft diameter changes were calculated as the local average axon shaft diameter from each 1-μm segment. For each field stimulation experiment, approximately six axon segments were analyzed with a total axon shaft length of around 25 µm and 20 synaptic boutons. Instances of bouton turnover were excluded from this analysis. The gain and loss of boutons was analyzed separately (SI Appendix, Fig. S13).

The propagation of APs along unmyelinated axons was simulated using the simulation environment NEURON [version 7.3 (39)]. Two different kinds of compartmental models were used to analyze how axon morphology influences AP conduction velocity. The first model was a geometrically constructed using average characteristics of CA3 axons (axon shaft diameter = 200 nm; presynaptic boutons = 1 µm long and 0.5 µm wide, spaced every 3 µm along the axon), whereas the second one was based on a realistic morphological reconstruction (using STED data). The multicompartment model used standard channel kinetic equations (40), using published parameters for hippocampal pyramidal neurons (41, 42) (SI Appendix).