Abstract
The axon is a neuronal structure capable of processing, encoding, and transmitting information. This assessment contrasts with a limiting, but deeply rooted, perspective where the axon functions solely as a transmission cable of somatodendritic activity, sending signals in the form of stereotypical action potentials. This perspective arose, at least partially, because of the technical difficulties in probing axons: their extreme length-to-diameter ratio and intricate growth paths preclude the study of their dynamics through traditional techniques. Recent findings are challenging this view and revealing a much larger repertoire of axonal computations. Axons display complex signaling processes and structure–function relationships, which can be modulated via diverse activity-dependent mechanisms. Additionally, axons can exhibit patterns of activity that are dramatically different from those of their corresponding soma. Not surprisingly, many of these recent discoveries have been driven by novel technology developments, which allow for in vitro axon electrophysiology with unprecedented spatiotemporal resolution and signal-to-noise ratio. In this review, we outline the state-of-the-art in vitro toolset for axonal electrophysiology and summarize the recent discoveries in axon function it has enabled. We also review the increasing repertoire of microtechnologies for controlling axon guidance which, in combination with the available cutting-edge electrophysiology and imaging approaches, have the potential for more controlled and high-throughput in vitro studies. We anticipate that a larger adoption of these new technologies by the neuroscience community will drive a new era of experimental opportunities in the study of axon physiology and consequently, neuronal function.
Significance Statement
Recent experimental evidence is revealing the rich electrophysiology of the axon and the importance of this structure for setting important neuronal computations. Most of these recent discoveries have been achieved at the intersection of the fields of neurotechnology and neurobiology. In particular, key recent findings in axon function have been made possible by the development of specific cutting-edge techniques that allow for the targeted probing of axons with high spatiotemporal resolution and signal-to-noise ratio. Here we review the state-of-the-art electrophysiological and optical toolsets for in vitro axonal experimentation and summarize the most recent and relevant discoveries they enabled. Altogether, we give the reader an up-to-date understanding of the neurotechnologies available to dissect axonal function.
Introduction
Historically, axons have been viewed as simple transmission cables of upstream decisions. In the central nervous system (CNS), this function is reduced to the reliable transmission of information from the somatodendritic compartments to the presynaptic terminals in the form of stereotypical “all-or-none” action potentials (APs). This perspective is embedded in the origin of the word axon (from Greek “áxōn,” meaning “axis”) and has been driven, at least partially, by the technical difficulties in probing axonal function, which bias most technical approaches to the recording of somatic APs alone. However, seven decades after the pioneering studies on the active properties of signal conduction along the axon (Hodgkin and Huxley, 1952), an increasing body of knowledge is challenging this limiting view (reviewed in (Bucher and Goaillard, 2011; Debanne et al., 2011; Rama et al., 2018; Alcami and el Hady, 2019). Key findings such as distal AP initiation (Dugladze et al., 2012; Deemyad et al., 2018; Mateus et al., 2021; Liu et al., 2022a; Rózsa et al., 2023); activity-dependent axon plasticity (Grubb and Burrone, 2010a; Kuba et al., 2010; Sheffield et al., 2011; Chéreau et al., 2017; Lezmy et al., 2017); and analog–digital facilitation of synaptic transmission (Alle and Geiger, 2006; Zbili et al., 2020) have sparked new interest in the study of axon electrophysiology and are decisively shifting the field away from the view of axons as simple transmission cables. Still, the full implication of this computational repertoire is far from understood.
Most neurons encode and transmit information via the conduction of APs along the axon arbor. AP propagation controls the reliability and the timing with which neuronal networks communicate; thus the modulation of this process has repercussions for, at least, temporal coding. Remarkably, most (if not all) parameters that influence AP propagation along the axon (e.g., axon diameter, ion channel density) can be modified by, at least, neuronal activity levels (Chéreau et al., 2017). Important AP characteristics, such as AP shape, can also be modulated locally via, for example, voltage-gated channels’ heterogeneous distribution, to convey information beyond timing (reviewed in Rama et al., 2018). Intriguingly, APs in the CNS may even initiate in distal parts of the axon, independently of somatodendritic integration, and propagate (antidromically) toward the soma, as well as toward axon terminals (Dugladze et al., 2012; Liu et al., 2022a; Rózsa et al., 2023). On the other hand, sensory neurons, which typically initiate APs in the peripheral terminals, can assemble an axon initial segment (AIS)-like structure in the proximal stem axon that acts as a source of spontaneous activity in neuropathic pain (Nascimento et al., 2022). This bidirectional axonal conduction, with notable implications for neuronal communication, can also be observed in vitro in both CNS and sensory neurons (Mateus et al., 2021). Many of these recent and unexpected findings have been made possible by a novel and extending repertoire of tools that allow for the probing of mammalian axons with unprecedented spatiotemporal resolution and signal-to-noise ratio (SNR).
Originally, most of our knowledge about axonal electrophysiology are derived from in silico studies and experiments on invertebrate models, despite their relatively limited morphological and functional complexity (Bean, 2007; Bucher and Goaillard, 2011). While the seminal work of Hodgkin and Huxley made use of millimeter-sized invertebrate axons that enabled experimental access to direct electrophysiological recording (Hodgkin and Huxley, 1952), the mammalian axons’ morphology long precluded the application of the same technical approach. Mammalian axons are typically the longest neuronal subcompartment (measuring up to meters) but are also extremely thin (≈0.1–20 µm) due to demanding energy requirements and space constraints (Perge et al., 2012; Costa et al., 2018). Axons can emerge from dendritic processes (Kole and Brette, 2018; Hodapp et al., 2022) and branch profusely into collaterals that are even thinner than the main axon trunk (Sasaki et al., 2012a; Mateus et al., 2021). Altogether, these morphological characteristics have hindered the study of mammalian axon's electrophysiology but have also provided clues about structure–function relationships unique to mammals.
Recent technological developments in nano-microfabrication and optical sensing have led to the emergence of new tools that allow for axon probing in vitro with high fidelity. Compared with the in vivo setting, in vitro models have the advantages of easier accessibility and manipulation without many of the confounding factors (e.g., surrounding tissue) and ethical challenges of animal experimentation. Moreover, they are compatible with emerging human-induced pluripotent stem cell (hiPSC) technology, which allows for the engineering of multiple cell types directly from patients (Girardin et al., 2022; Revah et al., 2022). Taken together, these developments have greatly increased the number of experimental possibilities in vitro.
Here, we aim to review the state-of-the-art electrophysiological and optical toolsets for in vitro experimentation and summarize the most recent discoveries in axon function that they have enabled. We will also review several microtechnologies that have been developed for controlling axon guidance in vitro and that can be readily combined with these toolsets. Our aim is not to provide an overview of the various biological processes involved in axon electrophysiology or guidance but instead to highlight fundamental methodological contributions that have increased our understanding of the axon. Finally, we provide an outlook of future research directions and outline our views on the prospects of combining technologies that promote axon guidance and enable the simultaneous recording of axon electrophysiology.
Measuring Axonal Electrophysiology
Classically, the study of axon function relied heavily on electrophysiological techniques which were not adapted to the study of the thin and complex arborizations of mammalian neurons. However, recent technological developments are driving a new era of experimental opportunities. Here we will address the most relevant: subcellular patch clamp, microelectrode arrays (MEAs), and functional imaging techniques. A schematic overview of these different techniques is shown in Figure 1.
Methods for probing axonal activity. Schematic representation of the main experimental approaches to probe axonal activity and the respective stimulation and/or recording capabilities. The labels are merely illustrative and may not represent the signals/waveforms characteristic of each technique.
These approaches enable the probing of axonal activity on a variety of scales. Subcellular patch clamp allows for stimulation and extracellular-like (axon-attached configuration) or intracellular recordings (whole-cell configuration) from a particular point of a single axon. Standard MEAs can be adapted for the extracellular recording and stimulation of multiple points of multiple axons, while high-density MEAs can be used for the electrical mapping of axonal arbors. Functional imaging techniques (e.g., voltage imaging) can read out axonal arbors’ activity with the highest spatial resolution but, typically, at the cost of temporal resolution. Adding to this diversity of techniques, optogenetics is rapidly emerging as an optical tool for the manipulation of presynaptic terminals. A comparison of the key characteristics, advantages, and limitations of these different techniques is summarized in Table 1.
Comparison of key characteristics, advantages, and limitations of different techniques for axon electrophysiology
Subcellular patch clamp
For a long time, patch clamp has been the gold standard for studying neurophysiology at single-cell or even single-ion channel resolution (Hamill et al., 1981). Typically, a recording micropipette is pressed against a patch of the membrane to form a strong seal resistance, which allows high-fidelity recordings in the cell-attached or whole-cell (if the patch is ruptured via suction, gaining intracellular access) configurations. However, most studies are limited to large compartments of the neuron (usually the soma) due to their easier accessibility.
Until recently, direct patch-clamp recordings from intact mammalian axons were limited to specific “giant” axonal structures (3–5 µm; e.g., calyx of Held, mossy fiber terminals; Bischofberger et al., 2006), which comprise <1% of all CNS synapses. This approach showed that hippocampal mossy fiber boutons act as sites of presynaptic AP amplification (by ≈40 mV; Engel and Jonas, 2005). However, recordings from smaller structures could only be obtained upon severing the axon, since the resulting swollen ends (i.e., blebs) were much larger (3–6 µm) than the intact axon (Shu et al., 2006; Kole et al., 2008). For the study of axon physiology, this may be problematic, as axonal lesions lead to an alteration of the passive properties of signal conduction (e.g., membrane capacitance), aberrant reorganization of the cytoskeleton, and, potentially, ion channel redistribution. Still, dual soma-axonal bleb recordings in the whole-cell configuration were the basis for the identification of the AP initiation zone (Debanne et al., 2011) and provide a useful method for recording signal propagation along the axon (Kramer et al., 2020; Zbili et al., 2020; Rózsa et al., 2023).
Beyond the aforementioned cases, new methodological approaches have now been developed to implement subcellular patch clamp in both brain slices (Sasaki et al., 2012b; Vandael et al., 2021) and cell culture preparations (Novak et al., 2013), as detailed below.
Axon-attached
Axon-attached (or loose-seal patch) recordings from intact unmyelinated axons (<1 µm diameter) have been made feasible by fluorescence-guided subcellular patch clamp (Sasaki et al., 2012b). The key advantage of this technique is that both the axon and the recording glass pipette are fluorescently labeled, which greatly facilitates live optical targeting. In combination with conventional whole-cell somatic recordings (i.e., dual recordings), this method allows for assessing the fidelity of AP propagation. Still, this technique is limited to single-site short-term recordings (<1 h) of extracellular-like AP waveforms (µV range), which are around three orders of magnitude lower than those recorded intracellularly (mV range), and as with other extracellular-based techniques (e.g., MEAs), subthreshold events are not detected (Sasaki et al., 2012a,b).
Sasaki et al. pioneered the application of fluorescence-guided subcellular patch clamp and demonstrated that the local activation of AMPA receptors by glutamate released from periaxonal astrocytes caused axonal AP broadening in hippocampal CA3 neurons. Broadened APs increased synaptic transmission to postsynaptic neurons by inducing greater calcium influx in presynaptic boutons (Sasaki et al., 2011). Hamada et al. performed recordings from axon collaterals’ boutons to characterize the temporal fidelity of presynaptic APs in a cuprizone-induced demyelination model. This approach revealed reduced conduction velocity, presynaptic AP broadening, and frequency-dependent AP failure in demyelinated L5 pyramidal neurons (Hamada et al., 2017). Recently, Lang-Ouellette et al. performed dual loose-seal patch recordings of young Purkinje cell somata and axons, which led to the surprising finding that the formation of axonal swellings (normally associated with axon degeneration) increases the fidelity of high-frequency AP propagation (Lang-Ouellette et al., 2021). This discovery suggests that axonal swellings in healthy states can represent a homeostatic form of axonal plasticity.
Dual somatic and axon-attached recordings have also revealed that the distal axons of CA3 pyramidal neurons (>600 µm from the soma) can fire at much higher rates (≈4–5 times) than the corresponding somata during gamma oscillations (Dugladze et al., 2012). These distally generated APs (i.e., ectopic APs) do not depolarize the soma due to the action of inhibitory axo-axonic cells that target the AIS. This implies a functional separation of axonal and somatic activity during this network behavior. Even though somatic recordings usually allow ectopic APs to be distinguished from APs initiated in the AIS (due to their characteristic waveform devoid of a somatodendritic component; Bähner et al., 2011), Dugladze et al. provide a remarkable example of how dual recordings can reveal insights unattainable with somatic recordings alone.
Whole-cell
Whole-cell recordings from varicosities (putative en passant boutons; Vivekananda et al., 2017; Ritzau-Jost et al., 2021; Vandael et al., 2021) or terminals (Vasylyev and Waxman, 2012; Kawaguchi and Sakaba, 2015; Zorrilla de San Martin et al., 2017; Vyleta and Snyder, 2023), as well as outside-out patch from the axon shaft (Hu and Jonas, 2014; Hu et al., 2018), have also been performed for several neuronal types. Novak et al. achieved a key technical breakthrough by combining subcellular patch clamp and super-resolution scanning ion conductance microscopy (<150 nm 3D resolution). This semiautomated approach uses a nano-pipette for identification, topographical imaging, and recording of presynaptic boutons and is compatible with all subcellular patch-clamp modalities in neuronal cultures (Novak et al., 2013).
Unlike in the axon-attached configuration, gaining intracellular access allows measurement of presynaptic AP shape, control of the cytoplasmic composition, and direct examination of biophysical properties of synaptic transmission. Several studies have now reported the facilitation of synaptic transmission via subthreshold depolarization of the presynaptic terminals (“analog–digital facilitation”; reviewed in Alpizar et al., 2019; Zbili and Debanne, 2019; Debanne et al., 2013), highlighting the importance of this type of axon signaling in neuronal function. Whole-cell electrophysiology is currently the only method capable of recording such subthreshold depolarizations. Paired recordings of presynaptic terminals and postsynaptic neurons are particularly well suited to measure synaptic strength and short-term plasticity. Still, obtaining the precise axonal AP waveform (one of the main advantages of conventional intracellular recordings) is compromised by technical difficulties. The necessary amplifier configurations and small pipette tip sizes (with increased resistance), adapted for very thin structures, inevitably distort axonal APs and require post hoc reconstitution (Oláh et al., 2021). Quartz glass pipettes (instead of conventional borosilicate) can be used to minimize this problem since they have lower resistance and capacitance at comparable sizes (Ritzau-Jost et al., 2021).
Recording an accurate AP shape is particularly important in the presynaptic terminals, where it modulates voltage-activated calcium currents and neurotransmitter-containing vesicle release. Direct recordings from boutons have allowed the dissection of Kv channels’ importance in presynaptic AP shape (Begum et al., 2016; Rowan et al., 2016; Vivekananda et al., 2017); the characterization of the presynaptic AP duration and amplitude in different cell types (Kawaguchi and Sakaba, 2015; Kramer et al., 2020; Ritzau-Jost et al., 2021); and the finding that presynaptic GABAA receptors mediate depolarization in Purkinje (Zorrilla de San Martin et al., 2017), dopaminergic (Kramer et al., 2020), and hippocampal mossy fiber neurons (Ruiz et al., 2010). Interestingly, Ritzau-Jost et al. demonstrated that presynaptic APs broaden during high-frequency firing in excitatory L5 pyramidal neurons, but not in inhibitory fast-spiking interneurons. In both cellular models, APs propagated reliably and at constant large amplitude (larger than at the soma) into axon collaterals, even during high-frequency firing (Ritzau-Jost et al., 2021).
Recently, two studies reported a nonconventional form of AP initiation in striatal dopamine axons which is independent of somatodendritic activity. Liu et al. and Kramer et al. performed direct recordings (whole-cell and perforated configurations) to reveal that these axons can initiate EPSPs and full-blown APs distally (near the synaptic terminals) in response to local acetylcholine release (Kramer et al., 2022; Liu et al., 2022a). Since presynaptic acetylcholine receptors are abundant throughout the brain, this finding raises the question of whether distal AP initiation is an important mechanism in physiological and pathological contexts beyond the dopaminergic system.
Overall, subcellular patch-clamp techniques, specifically the intracellular configurations, presently allow the highest possible fidelity in axonal recordings. Unfortunately, the technical complexity and invasiveness of these techniques limit the throughput and compatibility with long-term experiments. Furthermore, currently available intracellular methods are limited to relatively large structures of the axon (e.g., boutons). Critically, the inability to record from multiple sites along the axon simultaneously precludes subcellular patch clamp from tracking AP propagation throughout the axon arbor. Still, it is worth mentioning that several research groups have made efforts to automate patch clamping, which dramatically increased the throughput and accessibility of the technique in recent years (Suk et al., 2019; Koos et al., 2021; Ghovanloo et al., 2023). While these automated approaches are currently only applicable to somata, future endeavors may develop to study axons.
Microelectrode arrays (MEA)
MEA technology is at the forefront of technologies for recording electrical activity from large neuronal ensembles (reviewed in Obien et al., 2015; Xu et al., 2021). Typically, standard MEAs are composed of a fixed grid of multiple planar microelectrodes (most often 60–256 electrodes, interspaced by ≈30–500 µm), which are embedded in a transparent substrate (e.g., glass wafer) that functions as a cell culture vessel. After cell plating on MEAs, the activity of self-organized neuronal ensembles can be monitored and modulated (e.g., via stimulation) for over a year (Potter and DeMarse, 2001).
Currently, MEAs provide a versatile, noninvasive, and high-throughput functional assay at the network level. However, standard planar microelectrodes only allow for the recording of strongly attenuated extracellular signals (µV range) and are limited to the detection of suprathreshold activity (i.e., APs). Moreover, the recorded potentials are highly dependent on the positioning of the neuron relative to the electrode, and, typically, the microelectrodes only probe a very small fraction (<1%) of the whole substrate (Spira and Hai, 2013). These limitations make apparent the main reasons why conventional MEA technology alone is not adapted to the study of axon electrophysiology: (1) axons navigate the substrate freely; (2) the neuropil complexity hinders source–target identification (optically or electrophysiologically); (3) the SNR of axonal signals is so low that propagating extracellular APs can only be identified after averaging noise several times (Bakkum et al., 2013; Tovar et al., 2018; Abbott et al., 2020).
These limitations have recently been circumvented by either guiding axon growth over the electrodes (i.e., MEA and microfluidics combination, 3D-structured MEA) or dramatically increasing the spatial resolution (and the number of electrodes) of MEAs (i.e., high-density MEA), as detailed below.
MEA and microfluidics (μEF devices)
Several research groups, including ours, have merged the benefits of MEAs and microfluidics to attain simultaneous axon and network recordings (Pan et al., 2014; Lewandowska et al., 2015; Jang et al., 2016; Gladkov et al., 2017; Habibey et al., 2017; Hong et al., 2017; Gribi et al., 2018; Lopes et al., 2018; Moutaux et al., 2018; Shimba et al., 2021). We named this combination of microelectrodes and microfluidics μEF devices (Lopes et al., 2018).
Generally, microfluidics make use of subcellular microchannels to guide the axon to a separate culture compartment from the soma. μEF devices comprise tailor-made microfluidic devices (with matching microchannel-microelectrode spacings) to guide axons over multiple electrodes, allowing the study of signal propagation along multiple axons with high spatiotemporal resolution. To this end, we have developed specialized algorithms that enable, for example, the characterization of conduction direction and velocity within a user-friendly computational tool (Heiney et al., 2019). This combination creates advantages beyond axon guidance over electrodes. Conveniently, the microchannels’ small cross section creates an electrophysiological environment that increases the electrical coupling between axons and electrodes that greatly amplifies (by 1–2 orders of magnitude) the axonal signals (Dworak and Wheeler, 2009; Wang et al., 2012). It has been speculated that the amplitudes of axonal APs recorded in these confined settings (up to units of mV) may be large enough to modulate adjacent axons’ activity via ephaptic coupling (Pan et al., 2014; Narula et al., 2017), though no study has observed this phenomenon conclusively (Lewandowska et al., 2015). Importantly, the axon-specific compartmentalization allows selective treatment (e.g., of chemical blockers) and/or manipulations (e.g., axotomy), while recording network and axonal responses simultaneously (Habibey et al., 2015; Moutaux et al., 2018; Mateus et al., 2021; Shimba et al., 2021).
Probably due to the distinctive advantage of allowing for multisite, noninvasive (hence repeated and long-term) recordings with very high temporal resolution (up to 20 µs), most insights obtained via this combination so far have been related to conduction velocity. In particular, we have used μEF devices to reveal an increase in conduction velocity after the pharmacological inactivation of non-muscle myosin-II, an important protein in the structure of the axonal cytoskeleton. Super-resolution microscopy experiments showed that this inactivation leads to axon enlargement, providing one of the first demonstrations that axon cytoskeleton dynamics may act as mechanisms for regulating conduction velocity (Costa et al., 2020). Other research groups have used μEF devices to characterize changes in conduction velocity during network maturation (Habibey et al., 2017; Hong et al., 2017) or upon Nav1.2 blockade (Shimba et al., 2021).
Recently, μEF recordings revealed the unforeseen presence of ectopic/antidromic APs (Sheffield et al., 2011; Liu et al., 2022a) in hippocampal and sensory neurons. This activity was observed in basal conditions but could also be modulated by lesioning or chemical stimulation of distal axon portions (Mateus et al., 2021). Since these axons extended in empty microfluidic compartments (i.e., without synaptic targets), one may ask whether this form of subcellular initiation of electrical activity plays a role during development, when axons have not yet established synapses.
MEA and microfluidics can also be adapted to the study of explanted nerves. Gribi and colleagues used this “nerve on-a-chip” platform to characterize myelinated peripheral nerve fibers’ electrophysiology (Gribi et al., 2018) and test biomimetic stimulation strategies (high-frequency amplitude-modulated bursts) ex vivo (Formento et al., 2020).
While this combination of microtechnologies certainly provides unique benefits for the study of axon physiology, most studies have not moved beyond the proof-of-concept stage. Given its unique advantages, a wider adoption of μEF devices by the neuroscience community may help dissect the relevance of axonal electrophysiology in network function.
High-density MEA (HD-MEA)
A different type of MEA technology has also gained momentum in the study of axons (reviewed in more detail in Emmenegger et al., 2019). Complementary metal oxide semiconductor (CMOS)-based planar MEAs with higher spatial resolution than conventional MEAs have been used to probe axonal arbors at single-cell resolution in low-density (Bakkum et al., 2013, 2018; Radivojevic et al., 2017; Yuan et al., 2020) and high-density (Abbott et al., 2020) random cultures. These high-density MEAs (HD-MEAs) can pack up ≈7,000 electrodes/mm2 (up to ≈236.000 total) that can be read out simultaneously with high noise levels [active-pixel sensor (APS) architectures] or in subsets with low noise (switch-matrix architectures; Yuan et al., 2020; Suzuki et al., 2023). Until recently, due to SNR constraints, mapping axonal arbors was only possible following electrical stimulation of the axon and noise averaging of several stimulation trials (“stimulus-triggered averaging”; Bakkum et al., 2013). New-generation HD-MEAs, with improved SNR, now enable the mapping of axonal arbors based on their spontaneous activity (“spike-triggered averaging”; Abbott et al., 2020; Yuan et al., 2020).
HD-MEAs’ relatively high spatial resolution (≈0.25–20 µm) allows pure electrical mapping (“electrical imaging”) of the axon morphology (Abbott et al., 2020; Yuan et al., 2020; Radivojevic and Punga, 2023; Suzuki et al., 2023). This has allowed the discovery that the AIS, instead of the soma, is the main contributor to the extracellular potential (Bakkum et al., 2018). This electrical mapping is not limited to unmyelinated axons. Recently, Shimba et al. demonstrated the HD-MEAs’ capability to detect saltatory conduction along myelinated axons of sensory neurons (Shimba et al., 2022). In addition, Radivojevic et al. used HD-MEA technology to reveal that mammalian cortical axons conduct with high temporal precision (low jitter) and reliability (no conduction or branch-point failures) (Radivojevic et al., 2017). This finding demonstrates that, despite their intricate arborizations, mammalian axons seem to follow Rall's 3/2 power law describing the ideal geometrical ratio between mother and daughter branches to maintain propagation fidelity (Rall, 1959; Swadlow et al., 1980; Radivojevic et al., 2017).
A key disadvantage of HD-MEAs is their substrate opacity, which limits simultaneous probing with optical methods. Moreover, their recordings’ temporal resolution is, typically, lower than conventional MEA setups, thus prone to less precise measures of conduction velocity. Finally, while HD-MEAs can be combined with microfluidics for the control of axon guidance (Lewandowska et al., 2015; Duru et al., 2022), their nonplanar surface complicates the proper attachment of microfluidic devices. Still, these exciting new technological developments should accelerate the study of the conduction properties of single-cell axonal arbors.
3D-structured MEA
In recent years, many research groups have made efforts to improve the SNR of MEA recordings by fabricating 3D electrodes at the micro- or nanoscale (reviewed in Acarón Ledesma et al., 2019; Cho et al., 2021). In general, 3D-structured electrodes increase the electrical coupling between the electrode and the electrogenic compartment, which leads to an increase in the seal resistance at the electrode–membrane interface. The field was fueled by the discovery that large invertebrate Aplysia neurons spontaneously engulfed mushroom-like microelectrodes (Hai et al., 2010), which enabled the intracellular-like recording of APs and synaptic potentials (reviewed in Teixeira et al., 2020). Since then, intracellular-like recordings (units of mV) from the soma of mammalian neurons after spontaneous partial membrane engulfment (Shmoel et al., 2016) or after poration (Robinson et al., 2012; Dipalo et al., 2017; Liu et al., 2017; Abbott et al., 2019) have been demonstrated. However, due to their relatively small size, intracellular-like recordings from axons have yet to be achieved. Even the smallest reported 3D nanoelectrodes are in the same size order as mammalian axons (Cho et al., 2021); thus it is unlikely that this approach will soon allow for stable poration and probing of axonal structures. Although smaller 3D nanostructures (10–100 nm) can be fabricated, these are limited in recording capability due to the increased electrode impedance (Wu et al., 2019).
Even though the effects of topography on axon contact guidance are well known (Simitzi et al., 2017), surprisingly, the effects of 3D electrodes on network organization and axonal recordings have been largely neglected. A few studies have demonstrated the potential of incorporating discontinuous 3D features (e.g., nanopillars) for axonal guidance (Santoro et al., 2014; Amin et al., 2018; Mateus et al., 2019), but ultimately, these were either not integrated into MEAs or not capable of selective axon recordings. Theoretically, 3D MEAs can also be combined with microfluidic chambers, though the assembly procedure would need to be automated in a way that prevents damage to the fragile microstructures. Given their higher SNR, 3D-structured electrodes should allow for the reliable recording of axonal signals and contribute to the dissection of their importance in network function.
Interestingly, CMOS-based MEAs that integrate 4,096 3D-nanostructured electrodes have shown intracellular-like recording capabilities from thousands of somata simultaneously (in “pseudo-current-clamp” mode, although for limited time recordings), allowing for a partial connectome mapping (Abbott et al., 2019, 2020). In the future, the combination of higher spatial resolution HD-MEAs with 3D electrodes should allow for the extracellular probing of full axonal arbors, while recording somata intracellularly, hence gaining access to the subthreshold fluctuations (e.g., synaptic potentials) that precede and follow axonal conduction.
Functional imaging
The use of light to investigate axonal activity avoids the need for direct axon-electrode contact (i.e., subcellular patch clamp) or the close-apposition of extracellular electrodes (i.e., MEA). Functional imaging methods offer the advantage of probing axons at the highest possible spatial resolution, albeit at the cost of temporal resolution (essentially limited by the field of view/acquisition rate relationship; reviewed in Wang et al., 2019). The main optical methods for monitoring axon activity are calcium (Ca2+), sodium (Na+), and voltage imaging. These techniques can be considered passive, as they rely on sensors or indicators to measure ionic concentrations with no intended influence on the underlying physiological processes, although they may interfere in them (e.g., via calcium buffering, changes in membrane capacitance). Axonal optogenetic actuators, however, use light stimulation to manipulate ion fluxes and control neurotransmitter release.
Synaptic reporters are a different class of optical tools routinely used to measure the exocytosis of synaptic vesicles (e.g., pH sensors→pHluorin variants) and neurotransmitter release (e.g., glutamate via iGluSnFR)—events that are linked to the arrival of APs at the presynaptic terminals. These have been reviewed in depth elsewhere (Gobbo and Cattaneo, 2020; Jang et al., 2021) and are beyond the scope of this review.
One general limitation of functional imaging techniques is that the activity measured in each pixel may result from the superposition of multiple neurites or other electrogenic compartments. It is particularly difficult to resolve the axonal signal due to the limited surface area (hence few emitted photons). Moreover, the rapid kinetics of APs force very high acquisition rates (hence few contributing photons), which require ever more sensitive (i.e., larger dynamic range), bright (i.e., higher SNR), and photostable fluorophores.
Calcium imaging
APs reliably lead to Ca2+ influx (≈10-fold rise in intracellular concentration, which reverts to baseline within ≈100 ms; Koester and Sakmann, 2000); thus, Ca2+ imaging has long been used as an indirect proxy for neuronal activity (reviewed in Grienberger and Konnerth, 2012). Ca2+ signaling is particularly important in the presynaptic terminals, where it controls vesicle release probability. In fact, presynaptic Ca2+ entry and neurotransmitter release exhibit a nonlinear dependence, with small changes in the level of Ca2+ having a profound impact on quantal release (reviewed in Dolphin and Lee, 2020). Classical studies loaded neurons with Ca2+ indicators (e.g., BAPTA-1 dye) to image the Ca2+ dynamics at presynaptic terminals (Koester and Sakmann, 2000); however, the rapid progress in viral vector tools [particularly adeno-associated viruses (AAVs); Haggerty et al., 2020] has established genetically encoded Ca2+ indicators (GECIs) as the most prominent tool for imaging Ca2+ in neurons (Broussard et al., 2014). Most GECIs are engineered by fusing a fluorescent protein (e.g., GFP) to a Ca2+-sensing domain (e.g., calmodulin). When Ca2+ binds, the sensing domain undergoes a conformational change, resulting in a change in fluorescence.
Currently, GECIs are routinely used to assess neuronal activity at the network, cellular, and subcellular scales with high SNR (Dana et al., 2019). State-of-the-art GECIs allow detection of single axonal APs without averaging, if imaged with sufficiently high spatiotemporal resolution (Broussard and Petreanu, 2021; Huang et al., 2021). Thus, Ca2+ imaging can be used to monitor activity at levels ranging from multiple axons to single presynaptic terminals (or boutons; reviewed in Broussard and Petreanu, 2021). Since Cav channels are particularly concentrated at presynaptic structures, these regions allow higher SNR measurements than the axon shaft. Interestingly, the amplitude of Ca2+ transients can vary significantly across boutons of the same axonal branch in response to a single AP (Koester and Sakmann, 2000). Ca2+ imaging has helped demonstrate that presynaptic Ca2+ signaling can be modulated by several mechanisms beyond AP arrival, such as AP amplitude (Zbili et al., 2020), subthreshold signaling (Christie et al., 2010), and neuromodulation of Cav channels (Burke et al., 2018).
Recent advances in multicolor Ca2+ imaging facilitate simultaneous imaging of pre- and postsynaptic activity from differentially labeled neurons (Inoue et al., 2019). For the detection of fast Ca2+ transients, the sensors with faster responses are XcaMP-Gf (Inoue et al., 2019), jGCaMP7f (Dana et al., 2019), and the newest jGCaMP8 family (Zhang et al., 2023). The fastest sensor to date (jGCaMP8f) has reported half rise and decay times of ≈7 and ≈92 ms, respectively, which has greatly reduced the mismatch between Ca2+ sensor kinetics and actual Ca2+ transients in the axon. Still, it is important to note that untargeted GECIs preferentially label the somatodendritic compartment and diffuse poorly along the axonal arbor and thus are biased toward the detection of somatic APs (Knöpfel and Song, 2019; Broussard and Petreanu, 2021). To circumvent this limitation, several variants of GcaMP have been fused to presynaptic scaffolding proteins (e.g., synaptophysin) to enable presynaptic Ca2+ imaging (Dreosti et al., 2009; Zhang et al., 2018). These probes are known collectively as syGCaMPs and have allowed the dissection of different Cav channels’ contribution to presynaptic Ca2+ influx (Brockhaus et al., 2019). A red-shifted variant combined with pHluorin (SypHy-RGECO) has allowed researchers to directly relate presynaptic Ca2+ entry to vesicular release (Jackson and Burrone, 2016). For high SNR Ca2+ imaging of axonal arbors, an axon-targeted GECI (axon-GcaMP6) has been developed by fusing GcaMP6 to the growth-associated protein 43 (GAP43; Broussard et al., 2018).
Despite their ubiquitous use in today's research, some limitations in Ca2+ imaging approaches need to be considered. First, Ca2+ is a critical intracellular messenger; thus, Ca2+ concentration can change via cellular processes independent of firing activity. Moreover, the origin of axoplasmic Ca2+ transients in different portions of the axon is a matter of ongoing debate. While Ca2+ signaling at the presynapse is thought to be mediated by Cav channels (Dolphin and Lee, 2020), recent findings using high-speed Ca2+ and Na+ imaging have diverged regarding the origin of Ca2+ elevations in the AIS and nodes of Ranvier. Different studies have suggested that these AP-associated transients are mediated by Nav channels (Hanemaaijer et al., 2020; Filipis et al., 2023) or by Cav channels (Lipkin et al., 2021). Furthermore, Ca2+ sensors are (inherently) not sensitive to hyperpolarizing activity, and debate continues regarding whether they can faithfully report the whole extent of suprathreshold activity (especially in high-firing-rate conditions; Huang et al., 2021). Finally, it is important to consider that Ca2+ sensors need to sequester free Ca2+ to produce a signal, thus acting as Ca2+ buffers that reduce the transients’ magnitude and rate of change. In hippocampal axons filled with fura-2 AM, such buffering action has been estimated to produce 20-fold changes relative to no-sensor conditions (McMahon and Jackson, 2018). The effect of these perturbations in intracellular Ca2+ signaling is largely unknown and rarely acknowledged. Still, proper calibration of the setup against test samples of known concentration can lead to experiments that minimally interfere with endogenous signaling (Cherkas et al., 2018).
Sodium imaging
Sodium currents underlie key processes such as AP generation and conduction. High-speed fluorescence Na+ imaging allows precise recording of Na+ currents; thus, in the context of axon electrophysiology, it has primarily been used to characterize the spatiotemporal pattern of AP initiation in the AIS. Sodium imaging relies on dye indicators that exhibit an increase in fluorescence emission (and a slight shift in wavelength) upon binding Na+. In Na+ imaging, only a small fraction of Na+ binds to the indicator; thus, a linear relationship exists between the fluorescent change and the total Na+ transient for a wide range of indicator concentrations (Filipis and Canepari, 2021). This contrasts with Ca2+ imaging, where calibration of Ca2+ influx needs to consider binding to indicators and endogenous buffers.
In most central neurons, the AIS functions as a lower-threshold AP initiation zone, presumably due to the local high density of specialized Nav channel isoforms (Bean, 2007; Kole and Stuart, 2012), especially the Nav1.6 (in myelinated axons) and Nav1.2 (in unmyelinated axons; Kole et al., 2008; Hu et al., 2009; Lorincz and Nusser, 2010). These specialized channels mediate the rapid Na+ influx underlying AP initiation, and their voltage dependence for both activation and inactivation are shifted by ≈10 mV compared with their counterparts in the soma (Kole et al., 2008; Hu et al., 2009). Na+ imaging has been used to characterize the fast-activating and fast-inactivating transient Na+ currents associated with the AP, as well as subthreshold and persistent Na+ currents that originate from the AIS (Filipis and Canepari, 2021; Shvartsman et al., 2021).
Despite a consensus that Nav density is higher in the AIS than elsewhere in the neuron, the order of magnitude of this difference is yet to be determined and its importance in AP initiation is not clear (Colbert and Johnston, 1996; Kole et al., 2008; Baranauskas et al., 2013; Katz et al., 2018; Lazarov et al., 2018). Findings obtained via Na+ imaging of L5 pyramidal neurons have shown that the AP-associated Na+ flux is ≈3-fold and ≈8-fold higher in the AIS than in the soma and basal dendrites, respectively [suggesting a ≈3 (axon):1 (soma):0.3 (dendrite) Nav density ratio; Fleidervish et al., 2010]. However, thought-provoking experimental and theoretical evidence has suggested that high Nav density in the AIS is not required for AP initiation in the axon but instead is crucial for precise AP timing (Lazarov et al., 2018). In the study, even when axonal Nav density of hippocampal neurons decreased below somatic levels (through a mutation of the ankyrin G anchor βIV-spectrin), APs were still initiated in the axon. Future studies using Na+ imaging may help reveal where and how exactly APs initiate in analogous channelopathies.
Sodium imaging studies have also helped reveal a spatial mismatch between the highest Na+ flux, which occurs in the middle of the AIS, and the actual AP trigger zone (distal AIS; Baranauskas et al., 2013). This finding highlights the importance of axon morphology (i.e., lower capacitive load) and electric isolation from the soma in AP initiation, beyond Nav density. In another study, Katz et al. used Na+ imaging to demonstrate that Nav1.6 deficiency (Nav1.6-cKO) did not keep APs from being initiated in the distal AIS of L5 pyramidal neurons, as this loss was compensated by Nav1.2 channel expression (Katz et al., 2018), putting into question the role of particular Nav subtypes in AP initiation.
Since Na+ kinetics are particularly fast, an acquisition rate of ≈10 kHz is estimated to be necessary to characterize fast transients (Popovic et al., 2011). This acquisition rate is technically demanding; thus, researchers have interfered with Na+ channels’ kinetic properties (by lowering the experimental temperature to ≈21°C) to image the whole AIS at 0.5–2 kHz (Baranauskas et al., 2013; Katz et al., 2018). Recently, Filipis and Canepari reported 10 kHz recordings near physiological temperature (32–34°C) along ≈40-μm-long AISs (Filipis and Canepari, 2021; Filipis et al., 2023). Besides recording equipment difficulties, presently available indicators have limitations in sensitivity and SNR that prevent higher acquisition rates. Typically, these Na+-binding dyes (e.g., SBFI, ING-2) are loaded intracellularly in a ≈0.5–2 mM concentration range that allows a 0.5–5% change of fluorescence per AP after averaging several trials. Higher dye concentrations (that could lead to better SNR) are not well tolerated and alter electrophysiological properties (Filipis and Canepari, 2021). Thus future improvements of the Na+ imaging technique depend critically on the development of novel Na+ indicators.
Voltage imaging
Voltage-sensitive dyes (VSDs) and genetically encoded voltage indicators (GEVIs) have the potential to overcome some of the limitations of Ca2+ and Na+ imaging, because the changes in fluorescence are a direct proxy for changes in voltage (reviewed in Peterka et al., 2011). These indicators’ responses to voltage change can be linear (e.g., opsin-based indicators) or nonlinear (e.g., ASAP-based indicators), with linear indicators having generally faster on/off kinetics (time constant <1 ms), but less sensitivity (low %/mV). Consequently, linear and nonlinear indicators are more suitable for the recording of AP (fast kinetics but large voltage change) and subthreshold (slow kinetics but small voltage change) activity, respectively. Still, the conversion of fluorescence into absolute membrane potential (Vm) requires calibration via simultaneous current clamp (Cohen et al., 2020).
Early studies using VSDs revealed multiple AP initiation zones in invertebrate neurons (Zečević, 1996). Later studies with higher-sensitivity dyes (e.g., JPW3028, membrane potential imaging with a ≈11 µs delay) allowed determination of the location (20–40 µm from the soma) and length (≈20 µm) of the AP trigger zone in L5 pyramidal neurons, as well as the reliability of propagation along the axonal arbor, even at very high firing frequencies (≈400 Hz; Popovic et al., 2011). However, concerns regarding VSDs’ effects on the physiological properties of the cellular membrane (Peterka et al., 2011), as well as their lack of neuronal specificity, have stalled their use.
GEVIs circumvent the need for intracellular dye loading to achieve neuronal specificity via targeted expression and thus are currently the preferred tool for probing voltage changes in the axon. Still, due to their comparatively lower brightness, most studies need to average several stimulation trials to distinguish axonal Aps (Hoppa et al., 2014; Cho et al., 2017, 2020; Ma et al., 2017) and subthreshold activity (Rowan and Christie, 2017) from noise levels. Nevertheless, Cho et al. used QuasAr (an opsin-based GEVI) to reveal a novel role for Navβ2 subunits in the prevention of conduction failures at branching points (Cho et al., 2017). And recently, new-generation GEVIs (e.g., opsin-based Archon1 and 2, FRET-opsin Ace variants) have displayed the necessary kinetics and brightness to resolve the axonal AP in single trials (reviewed in Panzera and Hoppa, 2019), although fast photobleaching remains problematic (Bando et al., 2019). Furthermore, voltage imaging with Ace2N-mNeon at ≈3 kHz has been used to assess the direction and velocity of AP conduction, as well as to compare the AP waveform between different axonal segments (proximal vs distal) during Kv blockade (Gonzalez Sabater et al., 2021).
Due to the very high spatial resolution, voltage imaging is particularly useful for the study of voltage changes in structures difficult to probe with electrophysiological techniques, such as the axon shaft and presynaptic terminals. Several studies have used voltage imaging to help unveil the contribution of Kv channels to the presynaptic AP width and amplitude (Hoppa et al., 2014; Cho et al., 2020; Gonzalez Sabater et al., 2021). For example, Kv3 channels were demonstrated to cluster at boutons of stellate interneurons and influence presynaptic AP width with synapse specificity, even within the same axonal branch; this finding was validated with both voltage imaging and multisite bouton-attached patch clamp (Rowan et al., 2016).
Still, the adoption of GEVIs has so far been restricted by several limitations such as phototoxicity, photostability, low SNR, and poor performance under 2-photon microscopy. Moreover, depending on the sensor expression levels and intrinsic properties, GEVIs have been shown to increase membrane capacitance, reducing EPSP amplitude and increasing AP initiation thresholds (Akemann et al., 2009). Thus, the major challenge of voltage imaging is to develop sensors and methodologies that improve sensitivity, photostability, kinetics, and brightness, while minimally interfering with axon physiology. Recently, Liu et al. identified JEDI-2P as a 2-photon-compatible GEVI capable of reporting voltage transients of Drosophila interneurons’ axon terminals (Liu et al., 2022b). This and further developments have the potential to revolutionize how axonal activity is measured.
Optogenetics
Optogenetic actuators are gaining considerable momentum in the all-optical study of neuronal function (Hochbaum et al., 2014). These tools comprise genetically encoded photosensitive proteins (i.e., opsins) which can be used to control neuronal activity in a light-dependent manner via ion influx or efflux. Traditionally, their application has focused on activity control at the somatodendritic level, even though most opsins are not targeted to specific subcellular compartments. A notable exception is that of light-gated Cl− channels (e.g., GtACR2), intended for inhibition of neuronal activity, which require somatic targeting due to the antagonistic effect of Cl− influx in the soma (inhibition) and the axon (excitation; Mahn et al., 2018; Messier et al., 2018). Subcellular targeting of opsins to the axon is less common, but it is rapidly emerging as a way to directly control neuronal output and manipulate neurotransmitter release within the target areas of projection neurons (reviewed in Rost et al., 2022).
Understandably, the AIS is of particular interest for activity control by optogenetic actuators. However, attempts at subcellular targeting of opsins to the AIS have failed to meet expectations. For example, targeting channelrhodopsin-2 (ChR2) to the AIS, via an ankyrin G-binding domain of Nav channels, disrupts the local endogenous Nav clusters or fails to elicit APs (Grubb and Burrone, 2010b; Zhang et al., 2015). The presynaptic optogenetics toolbox, however, has greatly expanded and has proven useful in eliciting and inhibiting neurotransmitter release for different applications (Liu et al., 2019; Oldani et al., 2021).
Most axon-targeting actuators rely on protein fusion to synaptophysin, a protein abundant on both GABA-containing and glutamate-containing vesicles, with no perceived negative effect on synaptic function. Actuators that target specifically the presynaptic axonal plasma membrane are also emerging. For instance, Hamada et al. recently combined ChR2 with the mGluR2 C-terminal domain for presynaptic enrichment (S. Hamada et al., 2021). Optogenetic manipulation of presynaptic function represents an excellent tool for studying the role of individual synapses in information processing.
Still, it is important to note that presynaptic targeting of opsins is not a requirement for revealing new roles of axons via optogenetics. Recently, Hari et al. used GAD2CreER mice to express ChR2 (excitation) in GAD2+ neurons (axoaxonic GABAergic neurons that project onto myelinated proprioceptive sensory axons). This tour de force work demonstrated that, contrary to what was previously believed, activation of GAD2+ neurons does not result in presynaptic inhibition but rather prevents AP propagation failure at branch points by activating GABAA receptors at nodes of Ranvier (Hari et al., 2022). This study challenges, once again, our basic understanding of axonal GABA and shows a, potentially generalizable, way (i.e., nodal facilitation) of gating mediated by axo-axonic neurons.
Technique combinations
Virtually all the described techniques can be combined to extract more information from a single experiment. The combination of electrophysiological and optical techniques, in particular, allows probing of axon subcompartments’ morphological and functional traits with superior spatiotemporal resolution. All-electric (e.g., MEA and patch clamp) and all-optical (e.g., voltage and Ca2+ imaging) approaches can also be used complementarily, even though they are less common.
Probably the most used combination of techniques in the study of axonal activity is that of patch clamp and functional imaging dyes. Typically, patch pipettes are first used for filling neurons with the sensor of interest. Then, intracellular recording and stimulation of the soma are used as a “ground truth” for spiking activity, while spatiotemporal information (e.g., varicosities’ depolarization duration) is extracted from functional imaging. Sasaki et al. used a Ca2+ dye (BAPTA-1) to measure the dependence of AP-evoked Ca2+ elevation, in varicosities, on the axonal length and number of crossed branch points (Sasaki et al., 2012a). Recently, Zbili et al. used the same principle to demonstrate that the amplitude of the varicosities’ Ca2+ transients was increased when APs were evoked by a synchronous-like input (i.e., a depolarizing pulse directly from the resting Vm). This suggests that synaptic transmission may be facilitated when the presynaptic AP originates from synchronous inputs (Zbili et al., 2020). Dual soma–axon patch-clamp recordings combined with Na+ imaging revealed homeostatic changes (e.g., distal relocation of Na+ flux) in the AP trigger zone by acute blockade of M-type K+ channels (Lezmy et al., 2017). Cohen et al. performed voltage imaging of myelinated axons at 20 kHz to characterize saltatory conduction in the spatiotemporal domains. These optical recordings were voltage-calibrated via somatic patch clamp and revealed the existence of a periaxonal space that supports AP “saltation”—high-amplitude potentials at nodes of Ranvier precede attenuated potentials traveling in the internode (Cohen et al., 2020).
MEA electrophysiology has the benefit of allowing concurrent, noninvasive recording/stimulation and imaging of the axon. Even though simultaneous MEA electrophysiology and Ca2+ imaging of axon dynamics have been shown to be feasible (Moutaux et al., 2018; Mateus et al., 2021; Xue et al., 2022), new insights from this combination remain to be obtained. The recent development of fully transparent MEAs, compatible with high-resolution microscopy (Middya et al., 2021), should expedite the adoption of this technique combination.
All-optical approaches have the potential to allow monitoring and modulation of axonal subcompartments at will, with minimal invasiveness, but so far their use has been limited. The advent of newly engineered opsins that target the axon may enable new strategies where electrode-free setups can stimulate/inhibit (via optogenetics) and measure activity (via functional imaging) of targeted regions of the axon. Cho et al. coexpressed QuasAr and GCaMP6f to measure the shaft depolarizations and presynaptic Ca2 transients caused by single APs within the same axonal arbor (Cho et al., 2020). Lipkin et al. used Na+ and Ca2+ dyes to characterize AP-evoked Ca2+ transients and the distinct roles of Cav channels in the AIS (Lipkin et al., 2021). Theoretically, nonlinear and linear voltage indicators may also be combined to assess, respectively, sub- and suprathreshold activity (Panzera and Hoppa, 2019), although no study has so far reported such a combination.
All-electric combinations may allow precise control of Vm (via whole-cell patch clamp) and mapping of axonal arbor conduction (via HD-MEAs) concurrently. However, HD-MEAs are considerably harder to combine with other techniques due to their substrate opacity, which precludes experimental setups comprising inverted microscopes and hinders optical microscopy with upright microscopes. Still, HD-MEA recordings have been combined with somatic whole-cell patch-clamp recordings (Jäckel et al., 2017). In that study, somata could be patched by imaging the cultures with an upright microscope equipped with differential interference contrast. By taking advantage of the concurrent recording capabilities, this all-electric approach allowed the mapping of EPSPs evoked by stimulating the patched neuron. As this combination has the highest temporal resolution in axonal recordings, spatial clamping can be used to study propagation jitter and how it influences efficient information transmission. It is important to note that the introduction of propagation jitter may impact the coincidence detection of multiple EPSPs in the postsynaptic neuron, impacting synaptic transmission strength.
Controlling Axonal Guidance
The first step for analyzing axonal function is to identify/isolate its structure and growth path. In complex networks, such a task becomes daunting, because axons, dendrites, and glial processes form very intricate arbors within the neuropil. Still, this task can be simplified by patterning axon growth in vitro. Researchers have long taken advantage of the axon's growth-cone motility and cue-sensing capacity to control axon guidance in vitro (for historical perspectives, see Dupin et al., 2013; Roy et al., 2013; Raj et al., 2021). This control streamlines the study of axonal function by isolating axons and allowing the entire axon length (from the soma to the terminals) to be tracked.
A wide range of patterning techniques has been employed to control axon guidance (Raj et al., 2021). The most used can be divided into physical (e.g., microfluidics) and chemical (e.g., surface functionalization) patterning methods. Physical-based guidance strategies do not distinguish between axon, soma, and dendrites per se but may use topographical features and principles of size restriction to isolate the axon. Chemical-based strategies can use axon-specific permissive or repelling cues to control axon guidance more precisely. Both strategies take advantage of the growth-cone sensitivity to direct axon growth (i.e., taxis) and can be merged for enhanced effect. For example, growth-permissive chemical gradients (i.e., chemical patterning) can be established along microfluidic channels (i.e., physical patterning) to specify axon growth to target regions.
The most used methods for in vitro patterning are engineered substrates, microfluidics, and microcontact printing, which typically define fixed patterns for subsequent axon growth. These patterns can be regarded as static because they are created before cell seeding and cannot be easily changed thereafter. Recent developments, however, have used photo-reactive molecules and substrates that allow manipulation of axon guidance by creating and/or modifying paths in an in situ manner (Broguiere et al., 2020). Schematic representations of these methods are displayed in Figure 2.
Methods for controlling axon guidance. Schematic representations of approaches to control axon guidance via physical (top) and chemical (bottom) patterning. Detailed explanations of each method can be found in the respective subsection [engineered substrates; microfluidics; microcontact printing (µCP); in situ techniques]. For each method, the nature of the guiding feature (continuous; discontinuous, or custom) and possible direction for axon growth (unidirectional; bidirectional; multidirectional, or custom) are described.
The next sections detail the most recent developments in axon patterning in vitro. These methods are compatible with conventional neuronal culture methods, concurrent live imaging, and electrophysiology techniques and thus are amenable to high-throughput probing of axons.
Physical patterning
Engineered substrates
During development, axon pathfinding partially depends on the physical features of the environment (“topotaxis”), as well as the existence of guidepost cells where axons make turning decisions. The axon growth cone, via filopodial interactions, senses features (down to the order of tens of nanometers) to guide axon growth. Taking advantage of the recapitulation of this contact-mediated guidance in vitro and the advances in micro- and nanofabrication, researchers have engineered substrates that promote axon guidance. The outcome of these axon–substrate interactions can vary across neuron origin or age, but also across the topographical feature dimension, geometry, material, stiffness, and chemical properties (reviewed in greater detail elsewhere, Marcus et al., 2017; Simitzi et al., 2017; Leclech and Villard, 2020; Raj et al., 2021).
Different guiding features can be broadly separated into continuous (e.g., grooves, gratings) or discontinuous (e.g., an array of pillars, pits). Most of the structured substrates presented in the literature can also be considered anisotropic since they impose two (unidirectional) or more (multidirectional) symmetrical orientations (Leclech and Villard, 2020). Pure isotropic structures (e.g., nanowires) do not impose a direction on axon outgrowth but can influence aspects such as axon branching (Gautam et al., 2017; Seo et al., 2018). Continuous unidirectional features tend to increase the outgrowth rate and decrease the complexity of the branching pattern of the axon (Chua et al., 2014; Li et al., 2015), while discontinuous multidirectional features tend to increase the number of branches, which themselves are guided by the topography (Gautam et al., 2017; Seo et al., 2018; Milos et al., 2021).
Numerous studies have demonstrated that axons from different neuronal types and origins elongate along discontinuous or continuous topographical features in the direction of the pattern. The most pronounced effects are usually achieved via continuous structures, such as grooves with widths and depths in the µm scale (Chua et al., 2014). Recently, human-derived motor neurons’ axons were shown to grow over 1 cm along linear microgrooves (Hagemann et al., 2022). Discontinuous topographies, such as arrays of micropillars, can also promote axon guidance (Micholt et al., 2013; Park et al., 2016; Milos et al., 2021) but in this case, the critical parameter for strong guidance is the spacing between features, with ≈0.5–3 µm being optimal (Simitzi et al., 2017). These guidepost-like structures are of particular interest, as they can be adapted for the fabrication of 3D devices capable of recording neuronal activity with high-fidelity (e.g., MEA; Abbott et al., 2019; Mateus et al., 2019).
Microfluidics
Microfluidic devices follow a different rationale, which was pioneered in the 1970s with Campenot chambers (Campenot, 1977). Then, and now, the objective was to isolate axons in a separate culture compartment. Campenot's concept together with advances in microfabrication allowed a key breakthrough ∼20 years ago (Taylor et al., 2003, 2005). This microfluidic device comprised two compartments where neurons could be plated, interconnected by straight, high-aspect-ratio microchannels, into which only neurites could grow. If the microchannels were longer than 450 µm, only axons were able to cross the entire microchannel (Taylor et al., 2005). Thus, it allowed the spatial and fluidic compartmentalization of the somatodendritic and axon subcompartments. This simple, yet effective, design is widely used to date (reviewed in Neto et al., 2016; Holloway et al., 2021).
Several asymmetric microchannel geometries (e.g., arrowheads) that promote unidirectional axonal outgrowth have since been proposed in the literature to engineer neuronal circuits in vitro with defined connectivity (Peyrin et al., 2011; Forró et al., 2018; Holloway et al., 2019). These can be used to recreate specific circuits (e.g., hippocampus’ trisynaptic pathway), network motifs (e.g., feedforward excitation), or logical operators (e.g., “AND gates”; Feinerman et al., 2008), which rely on unidirectional connectivity. Microfluidic devices take advantage of soft lithography for replica molding the desired pattern using biocompatible silicones, typically polydimethylsiloxane (PDMS; Duffy et al., 1998). Besides neuronal compartmentalization, microfluidics’ advantages include the usage of fewer reagents and sample amounts, low cost, and increased high-throughput potential when compared with traditional in vitro systems.
Although most neuroscience applications make use of passive microfluidic systems, flow can be driven by capillarity, hydrostatic pressure differences, or external pumps. Hydrostatic pressure differences are routinely used to establish chemical gradients along the microchannels (via diffusion) or to treat target compartments (Park et al., 2006). Precise control of the fluid flow rate and angle can also be used to, for example, induce mechanical stress on the axon, leading to axonal injury (Pan et al., 2022).
Due to their versatility, microfluidic devices are currently the main tool for the structuring of “brain-on-a-chip” platforms (Holloway et al., 2021; Nikolakopoulou et al., 2021). Importantly, they allow isolation of large numbers of axons’ distal ends, which are easily accessible for targeted chemical or mechanical manipulations or coculturing with different cell types (e.g., glia). This characteristic has been used extensively in studies of axon development (reviewed in Millet and Gillette, 2012; Fantuzzo et al., 2019), regeneration (Hur et al., 2011), and degeneration (reviewed in Kim et al., 2012), among others. The study of axonal transport, in particular, has benefited greatly from microfluidic devices’ application. For example, Akin et al. used microfluidics, in combination with advanced live imaging modalities and new fluorescent tags, to demonstrate how different ion channels are produced in the soma, trafficked to the axon, and, finally, inserted into the axon membrane of sensory neurons (Akin et al., 2019, 2021).
We have recently shown that the limits of subcellular patterning using microfluidic compartmentalization can be pushed down to the nanoscale. By using a combination of electron beam lithography and photolithography, we were able to fabricate features ranging from 150 nm (nanochannels) to a few millimeters (microchannels). This allowed control of both axon and dendritic spine growth and thus the establishment of synapses in user-defined regions (Mateus et al., 2022). Such a degree of control over synapse formation should facilitate the study of synapse maintenance and plasticity in long-term experiments, as well as the consequences in information transmission.
Chemical patterning
Microcontact printing (µCP)
Chemical patterning takes advantage of the growth cone's sensitivity (“chemotaxis”) to substrate-bound (e.g., proteins) and/or soluble chemical cues [e.g., nerve growth factor (NGF)] for controlling axon guidance. Classical approaches (e.g., Bonhoeffer stripe assay) of chemical patterning helped to identify several attractive and repulsive molecules for axon guidance. Currently, microcontact printing (µCP) is the most used technique for substrate patterning, while soluble cues are often used to establish chemical gradients within microfluidic devices (reviewed in Aebersold et al., 2016; Raj et al., 2021).
Most often, in µCP, replica-molded PDMS stamps with microscale features are initially inked in a coating solution. Then, the stamps are used to imprint the desired patterns on a cell culture substrate. Upon plating, neurons tend to adhere to or migrate toward the areas that promote cell adhesion and then extend within the patterned surface (Kleinfeld et al., 1988). Like engineered substrates, µCP can be used to create continuous (e.g., lines) or discontinuous (e.g., spots) patterns that either create paths for axon growth (Vogt et al., 2005) or promote axon collateral branching (Kim et al., 2014). µCP can be used to control axon guidance more precisely, via the use of both promoting and repelling cues (Oliva et al., 2003; Weydert et al., 2019) or by the application of geometrical constraints (Roth et al., 2012).
Like microfluidics, µCP allows the isolation of large numbers of axons, with the advantage of easier axon accessibility due to the absence of a physical constraint (i.e., microchannel). Still, it is important to note that µCP alone does not establish fluidic compartmentalization, nor is it compatible with long-term control of axon guidance, as neurons typically start extending beyond the micropatterned surface after a few weeks in culture (Aebersold et al., 2016). For more reproducible and long-term control, physical barriers need to be considered.
In situ techniques
In recent years, several techniques have enabled the in situ patterning of mammalian axon outgrowth with great spatiotemporal resolution. These methods provide localized sources of guidance cues or provide paths for axon outgrowth on-the-fly, and thus they are not dependent on nano-/microfabrication of predefined patterns.
The classical micropipette assay initiated the in situ studies of axon guidance and has been critical in the discovery of chemical cues and associated intracellular pathways. In this method, a micropipette is placed close to a growth cone to create a chemical gradient. This allowed the pioneering discovery that sensory neurons’ axons are attracted to NGF (Gundersen and Barrett, 1979). Photorelease techniques, such as optical uncaging of caged molecules (Ellis-Davies, 2007) and photolysis of encapsulated guidance factors (Pinato et al., 2012), follow the same principle and provide interesting alternatives for creating local sources of guidance with no physical perturbation. However, these techniques share a limitation: they are not amenable to high-throughput experimentation. Typically, a single (and isolated) growth cone is exposed to the local cue each time (Dupin et al., 2013).
Emerging light-based techniques such as stereolithography or two-photon lithography are capable of additive manufacturing, material polymerization, or material ablation/etching in 2D and 3D. These techniques make use of photosensitive biomaterials (e.g., photoresins) to laser pattern with high precision (limited by the point spread function of the system) paths for axon growth. Broguiere et al. used two-photon patterning to dynamically control the growth of axons over 3D patterns of NGF (Broguiere et al., 2020). A different rationale was followed by Hong and Nam, who used photothermal etching to melt an axon-repulsive agarose hydrogel and allow axon outgrowth and manipulation of network connectivity (Hong and Nam, 2020). These proof-of-concept works demonstrate new avenues for in situ manipulations of axon guidance with high throughput. The maturation of these techniques and combination with functional readouts may lead to new insights into axon function. For example, given the degree of control over axon trajectories, these techniques can be used to study forms of direct axo-axonal coupling, such as electrical synapses (i.e., axonal gap junctions) or chemical synapses (e.g., GABAergic terminals onto the AIS of principal neurons), that are difficult to study in vivo.
Other in situ methods such as the application of electric fields have long been employed to guide axon growth in avian (Pan and Borgens, 2012) and amphibian (Patel and Poo, 1982) models, but their efficiency on mammalian axons remains controversial (Kim et al., 2016). Moreover, due to their incompatibility with concurrent direct axon recordings, they are outside the scope of this review.
Discussion and Future Perspectives
As wisely stated by Edgar Adrian in 1932, “The history of electrophysiology has been decided by the history of electrical recording instruments.” Almost a century later, it is still experimentally challenging to record from axons and, particularly, from multiple axons simultaneously. Consequently, most studies are still based on data derived from recordings of somatic APs. Unfortunately, this emphasis on somatic APs introduces several biases: it assumes that suprathreshold activity recorded in the soma serves as a direct proxy for neuronal output; it presumes that signaling is not modulated along the axon; and it disregards the fact that the axon is a plastic compartment. From the previous sections, it should be evident that somatic recordings do not provide a complete picture of neuronal output. In fact, in certain network behaviors (e.g., gamma oscillations), distal axons can generate APs at much higher rates than recorded in their corresponding somata (Dugladze et al., 2012). Another striking case is that of substantia nigra dopaminergic neurons, in which dopamine release is often defined by dopamine and acetylcholine receptors in the axon terminals and independent from somatic APs (Berke, 2018; Kramer et al., 2022; Liu et al., 2022a). Ultimately, in some cases, somatic APs may represent only a fraction of a given neuron output. Besides unforeseen AP initiation mechanisms, axons are also highly dynamic and can change structural (e.g., AIS length or position; Grubb and Burrone, 2010a), morphological (e.g., axon diameter; Chéreau et al., 2017), or functional (e.g., conduction velocity, presynaptic AP shape) properties (Zbili et al., 2020) depending on the excitability levels. The significance of this extended computational repertoire has yet to be fully characterized, but it is increasingly clear that both single-neuron and circuit functions cannot be understood based on somatic recordings alone.
As discussed here, cutting-edge technologies have provided means to unveil and characterize AP initiation mechanisms (Liu et al., 2022a; Rózsa et al., 2023), structure–function relationships (Hodapp et al., 2022; Nascimento et al., 2022), voltage-gated channels’ implications on AP shape and release probability (Rowan et al., 2016), signal conduction properties (Radivojevic et al., 2017; Mateus et al., 2021), synaptic transmission biophysics (Vyleta and Jonas, 2014), or presynaptic plasticity (Chéreau et al., 2017; Vandael et al., 2020). Although these findings likely impact neural computations, it has been difficult to determine how they affect network function since the majority were obtained through recordings of single axons. In systems neuroscience, the number of neurons recorded by microelectrodes has followed a trend of exponential growth, doubling approximately every 7 years (Stevenson and Kording, 2011). Unfortunately, this trend does not directly translate to the number of recorded axons due to several intrinsic limitations (e.g., electrode–axon size mismatch). The integration of new microtechnologies into in vitro axon research, however, offers exciting prospects and can help dramatically increase the experimental throughput. Moreover, patterning methods and in situ techniques, in particular, can enhance the degree of control over the desired axon path to dissect axon physiology. HD-MEAs can be used to study multiple axons per culture with a spatiotemporal resolution compatible with inquiries on axonal conduction (Yuan et al., 2020; Radivojevic and Punga, 2023). μEF devices currently allow for the extracellular recording of hundreds of axons simultaneously (Moutaux et al., 2018; Mateus et al., 2021) and can be further improved to achieve single-axon resolution recordings via clever nano-microfluidic design. Nano-microfluidics can control neuronal growth down to the dendritic spine scale, paving the way for inquiries with single-synapse resolution (Mateus et al., 2022). These technologies can also be used to create testbeds amenable to the study of elusive forms of axo-axonal communication such as gap junctional (i.e., electrical synapses; Alcamí and Pereda, 2019) or ephaptic coupling (Han et al., 2020). Importantly, microfluidics’ compartmentalization can be leveraged to study axon interactions with other cell types (e.g., glia) in coculture configurations or the short- and long-term effects of insults (e.g., axon lesioning) and treatments (Hyung et al., 2021). Despite several studies highlighting the influence of gliotransmission in axon electrophysiology (Yamazaki et al., 2007; Sasaki et al., 2011; Deemyad et al., 2018), glia–axon interactions are particularly understudied and can benefit from this technique combination.
Ultimately, “on-chip” technologies provide a highly controllable environment for experiments that are time-efficient, highly informative, and in accordance with the 3R's principle (replacement, reduction, and refinement of animals used) given the high number of axons studied per experiment. Still, this approach has not permeated the neuroscience community as expected (Nikolakopoulou et al., 2021). This is probably because such commitment requires interdisciplinary research between researchers coming from various backgrounds, ranging from materials science to neuroscience. Hopefully, the establishment of standardized protocols (Lopes et al., 2018) and specialized companies, that offer “key-in-hand” solutions, will help fill this gap.
Despite their crucial contributions to neuroscience, conventional microelectrode technologies are not fundamentally different from the first electrodes used by Galvani (circa 1770). These technologies suffer from intrinsic limitations such as the need for a nonbiological interface with the axon; lack of cell-type specificity; a limited number of electrodes; adverse scaling of reduced electrode size and SNR; and limited capability to simultaneously record and stimulate neural activity. Optical probes can circumvent these limitations but not without shortcomings. Consequently, although the replacement of the “electron” by the “photon” has been anticipated for over a decade (Scanziani et al., 2009), it has not yet taken place. In axon physiology studies, optical sensing is particularly challenging due to, mainly, the very limited volume/membrane available to pack in reporters (consequently limiting the number of contributing photons). The development of fluorescent sensors tagged with axon-specific protein tags will enable more precise labeling of structures of interest compared with untargeted sensors or indicators. Still, it is important to note that this drawback may already be circumvented in vitro via axon guidance techniques that isolate axons.
Future endeavors will also benefit from bright red or far-red VSDs, GEVIs, or GECIs that can be spectrally separated from green markers or sensors (e.g., GCaMP family) to monitor properties associated with voltage and Ca2+ transients. These may be coupled with genetically encoded sensors of neurotransmitter release (e.g., iGluSnFR) for the dissection of synaptic transmission and facilitation. Importantly, axon-specific imaging allows for multiscale probing of axon transmission—from entire projection bundles to particular synapses. This strategically positions imaging techniques as the go-to tools for many research questions. Nevertheless, even though the future of the field will increasingly depend on optical approaches, a synergism between imaging and electrophysiology will continue to be crucial. For instance, the combination of electrophysiology and structural imaging techniques, such as super-resolution microscopy, is only beginning to reveal how nanoscale structural changes impact axon function (Chéreau et al., 2017; Costa et al., 2020).
Despite the early examples that invertebrate axons do not serve as simple transmission lines (reviewed in Swadlow et al., 1980), the inaccessibility of the mammalian axon long hindered attempts of dissecting its electrophysiology. Recently, cutting-edge techniques reignited the interest of the neuroscience community in axon physiology and have made possible several unexpected findings that carry significant functional implications. While a steady stream of technological developments is to be expected, we cannot anticipate what the future will teach us concerning axon physiology. It will ultimately take several key breakthroughs to understand how the computational repertoire of the axon influences network-level function and, possibly, behavior. To this end, future studies will need to go beyond in vitro experimentation; but current research can undoubtedly gain from such highly controlled environments. In particular, we suggest that the combination of state-of-the-art technologies to guide and probe multiple axons can shed light on how axon signaling coordinates network activity. Hopefully, these approaches will help us better understand not only axon physiology but also, ultimately, pathologies that arise from axonal dysfunction.
Footnotes
This work was partially funded by national funds through Foundation for Science and Technology (FCT), under the project PTDC/NAN-MAT/4093/2021. J.C.M. was supported by FCT (PD/BD/135491/2018) in the scope of the BiotechHealth PhD Program (Doctoral Program on Cellular and Molecular Biotechnology Applied to Health Sciences).
The authors declare no competing financial interests.
- Correspondence should be addressed to P. Aguiar at pauloaguiar{at}i3s.up.pt.
