9. Neuromodulation Through Axonal Stimulations in Brain

Narrow pulses can powerfully depolarize neuronal membranes (Lowet et al. 2022). Under normal physiological conditions, this depolarization can trigger action potentials in neurons. The typical refractory period of neurons is approximately 1 ms. DBS pulses at frequencies within 50–200 Hz have intervals exceeding the refractory period. Thus, most neurons—especially fast-responding axons for neural signal conductions—should be able to fire in response to every pulse in this frequency range. Given that peripheral nerve block requires kilohertz stimulations, debate has continued about whether DBS pulse at around 100 Hz can block axon firing and conduction. Although sustained axonal HFS can indeed decrease its excitatory effect on downstream neurons compared to initial period, the question is where this decrease stems from, the presynaptic axons or the synapses? Some researchers have attributed it to synaptic failures, considering that intense activations may deplete neurotransmitters through continuous releases, causing synaptic fatigues and failures (Anderson et al. 2006; Iremonger et al. 2006; Rosenbaum et al. 2014; Farokhniaee and McIntyre 2019). However, experiments on in-vitro brain slices from the thalamus and basal ganglia have shown that axons fail to produce action potentials following each pulse at frequencies above 50 Hz, resulting in axonal block (Shen and Johnson 2008; Zheng et al. 2011; Rosenbaum et al. 2014). Similar axonal block has been observed in brain slice experiments from other regions, including the cerebral cortex and hippocampus (Chomiak and Hu 2007; Jensen and Durand 2009). Our in-vivo experiments in intact rat brains further confirmed the HFS-induced axonal block. Nevertheless, the structural differences in axons across brain regions and peripheral nerves—such as thickness and myelin sheath presence—can influence the extent and progression of HFS-induced axonal block.

As presented in Sect. 5.2, we studied axonal HFS by utilizing the clear layered structures in the hippocampal region—especially the alveus axon fibers covering the dorsal surface of hippocampus beneath ventricle. The alveus, formed by the axons of CA1 pyramidal neurons, allowed us to use small bipolar stimulation electrodes to focus the stimulated area. We verified that HFS at around 100 Hz can produce intermittent axonal block by combining both antidromic and orthodromic stimulations in experiments. When the antidromically-evoked APSs significantly decreased during steady A-HFS period—indicating that the stimulation of pyramidal neurons’ axons failed to generate soma action potentials—the orthodromic test pulses (OTS) were still able to activate these somata. This result clearly showed that the failures occurred in the axons rather than the somata, providing strong evidence of axonal block. Furthermore, our series HFS experiments have provided additional supports. For instance, as shown in Chap. 7, the distinct soma responses to the opposite-polarity pulses during A-HFS_┬┴ also indicated the response attenuation originated in the axons. Otherwise, if the axons had reliably generated action potentials following each pulse, the somata would have received identical antidromic signals, making it impossible to distinguish between and respond differently to pulse polarities. Although these experiments have confirmed the axonal block, they cannot exclude other mechanisms, such as synaptic neurotransmitter depletion in the orthodromic pathway. Presumably, both mechanisms may coexist or dominate at different stages during HFS. Neurotransmitter depletion may precede axonal block in early HFS period. Once axonal block occurs, the resulting decrease in the excitatory inputs on synapses can then allow neurotransmitter to replenish.

The mechanism of HFS-induced axonal block may involve increased extracellular potassium ion concentration ([K⁺]_o) (Bellinger et al. 2008; Liu et al. 2009). The HFS-induced intense firing at onset can rapidly elevate [K⁺]_o, preventing the axonal membrane from repolarizing promptly after generating an action potential. The membrane can thus maintain a prolonged depolarized state and become unresponsive to continuous HFS pulses (Guo et al. 2018; Zheng et al. 2020)—as shown in the simulation results of axon model in Sect. 7.2. Although [K⁺]_o can play a key role, other ionic changes across the axonal membrane may also contribute to axonal depolarization block, including the accumulation of sodium ions ([Na⁺]_i) inside the membrane (Zang and Marder 2021). Additionally, structural factors like axonal branching and HFS-induced changes in axonal morphology and biophysical properties can promote the development of axon block (Chéreau et al. 2017; Chomiak and Hu 2007; Rama et al. 2018). Therefore, axonal block may result from multiple interacting mechanisms.

HFS has be used to block signal conduction in the spinal cord and peripheral nerves, achieving therapeutic efficacies such as pain relief (Mekhail et al. 2020). However, a complete block in axonal fibers requires HFS above kilohertz frequencies (Arle et al. 2016). Typical DBS using pulse frequencies below 200 Hz can only produce a “partial” or “intermittent” block. This means that the HFS can still trigger action potentials in axons but at a frequency lower than the stimulation pulses, thereby exciting post-synaptic neurons and establishing new firing activity to replace original activity in the axon projection areas. This effect resembles “ablation” or “information loss” (Lowet et al. 2022). The newly produced firing has some randomization (see Sect. 5.3), which can eliminate synchronous firing between neurons, modify their original burst firing patterns, and suppress rhythmic activity in neural networks (Feng et al. 2017; Yu et al. 2016; Barow et al. 2014; Lee et al. 2011). In addition, during this partial block situation, individual axons may fire at different rates. Some axons in the stimulated fiber may even completely stop firing until HFS ends.

Our experimental results also show that during HFS-induced axonal block, although pulses can intermittently trigger axons to fire, signals from upstream neurons should have been completely blocked. During O-HFS, the stimulation increased the firing of downstream neurons. However, once the O-HFS ended, these neurons immediately stopped firing and remained silent for seconds before gradually resuming firing (see Fig. 5.19). This indicates that during O-HFS, the downstream neurons were driven by the stimulation, not by the upstream neurons—had upstream control remained, the downstream neurons should have maintained at least partial firing after the stimulation ceased. The mechanism can involve O-HFS-induced intermittent firing propagating antidromically along axons to somata, suppressing the original activity at somata and altering their excitability (refer to Sects. 5.2.3 and 5.2.4 for details). The silent period of neuronal firing that occurred immediately after the end of A-HFS supports this mechanism (see Fig. 5.13). Consequently, when the O-HFS control ceased, the upstream somata needed time to recover their original activity before resuming control over downstream neurons. This produced the “silent period” in downstream neurons after O-HFS ended (Fig. 5.19).

These results verify that the intermittent depolarization block caused by axonal HFS can interrupt signal transmissions between upstream and downstream neurons, thereby disrupting pathological neural activity, as proposed by previous reports (Chiken and Nambu 2014). Besides signal blocking, the axonal HFS can bilaterally control both upstream somata and downstream post-synaptic neurons. Its action on downstream neurons and their networks is crucial for brain stimulation therapy. Our experiment results show that the intermittent block (or intermittent recovery) of stimulated axons can create randomness in the HFS-induced firing, weakening the phase-locking between induced firing and stimulation pulses, thereby reducing firing synchronization between individual axons. This leads to asynchronous activity in downstream neuronal populations (see Sect. 5.3). In addition, HFS with constant pulse intervals can produce non-uniform “cluster” firing (Sect. 5.2.5). Such HFS-induced activity may eliminate pathological synchronous activity—for example, suppressing excessive synchronous activity in epilepsy (Chap. 8)—thereby achieving therapeutic efficacies.

During intermittent depolarization block, small changes in axonal membrane potential can substantially change the state of membrane ion channels (refer to the potential-dependent changes in HH model parameters shown in Fig. 1.11). Besides the nonlinear dynamics of membrane ion channels, the random characteristics (“noise”) in various structural levels of the nervous system (Mino and Grill 2002; Faisal et al. 2008) may promote the changes in HFS-induced neuronal firing. As a result, even slight variations in HFS parameters can trigger diverse neuronal responses. Chapter 6 presents our findings with time-varying stimulation patterns, including HFS with gradually varying frequencies and intensities, alternating dual-parameters, single-pulse insertion and deletion, and randomly varying frequencies (IPIs).

Among these findings, the neuronal responses to randomly varying frequencies were surprising. Our results of constant HFS at 50–200 Hz showed that at lower frequencies (such as 50 Hz), neuronal firing strongly phase-locked with the stimulation pulses. As the HFS frequency increased, this phase-locking weakened and neuronal firing tended to random. Following these results, I originally expected that with IPIs varying randomly within a small range, the HFS would further randomize neuronal firing and reduce firing synchronization among neurons. However, our experiments of HFS trains with 5–10 ms randomly varying IPIs showed an opposite effect—the randomized pulse timings actually produced large PSs occasionally (see Sect. 6.4). This means that during HFS-induces axonal block, the stimulation can still activate downstream neuron populations strongly since PS events represent strong synchronized firing. Additionally, within a same small varying IPI range of 5–10 ms, different neuronal responses can be produced by designing various IPI sequences or adjusting IPI orders (Zheng et al. 2021). Notably, stimulation patterns with time-varying parameters offer broad options for neural modulation therapies beyond constant patterns. Like pharmaceutical treatments, these time-varying stimulation patterns can provide various “dosages” and “efficacies” for clinical applications, helping to expand the potential uses of DBS technology.

In addition, under intermittent block situation, axonal responses to opposite-polarity pulses can differ significantly from those under normal conditions. During the stimulations with alternating opposite-polarity pulses, the two pulse types can respectively activate two sub-groups of neurons. For axons that can response to both types of pulses at baseline, they only response to positive pulses rather than negative pulses during HFS of alternating pulses. This finding also provides valuable clues for developing new stimulation paradigms (Chap. 7).

Our studies presented in Chap. 8 show that excitatory axonal stimulations—delivered through both open-loop sustained HFS or closed-loop brief HFS—can suppress epileptiform spikes in downstream neurons. These suppressions may involve mechanisms of desynchronization and depolarization block. The effects of HFS patterns with time-varying parameters on abnormal neural activity require further study. Under intermittent depolarization block, for example, randomly-varying frequency stimulation can excite downstream neurons more strongly than constant frequency stimulation. Such stimulations might suppress certain types of epileptiform activity more effectively through a “fight fire with fire” effect (Gwinn and Spencer 2004)—though this hypothesis requires experimental verification.

Conventional DBS primarily targets the thalamus and globus pallidus to treat movement disorders. Theoretically, external electrical stimulations can potentially treat various neurological diseases through modulating neural activity in specific brain regions. Therefore, new DBS targets for different conditions are continuously being explored. For instance, studies have shown that cerebellar stimulations can enhance cognitive recovery in rodents with prefrontal traumatic injury, offering potentials for rehabilitating patients with trauma-induced cognitive impairments (Chan et al. 2022).

The hippocampus is a common pathogenic region for brain diseases such as epilepsy and Alzheimer's disease. It has served as a potential target for DBS treatment of refractory epilepsy in clinical practice (Vetkas et al. 2022; Geller et al. 2017; Wu and Sharan 2013). The hippocampal region has distinct structural layers—respectively containing axon bundles, somata, and dendrites. This feature enables precise positioning of stimulation and recording electrodes in different neuronal structures. Additionally, hippocampal neurons are densely arranged, facilitating the detection and measurement of biomarkers as quantitative indicators. For instance, the dense cell bodies in the soma layer enables the formation of large population spikes (PS) in the extracellular space during synchronous firing. The PS amplitude and area can indicate the amount of firing neurons and serve as feedback indicators for closed-loop control, as shown in Sect. 8.4. Although a PS cannot determine the absolute number of firing neurons, it can be used to compare relative quantities. Other extracellular signals can also be detected, including action potentials of individual neurons (unit spike), field excitatory postsynaptic potentials (fEPSP), and local field potentials (LFP). Recordings of these electrical signals from population and individual neurons simultaneously can both provide feedback indicators for DBS controls and reveal valuable insights into DBS mechanisms.

Beyond brain stimulations, neuromodulation technology encompasses various types of nerve stimulations widely used in clinical treatment. These include vagus nerve stimulation (VNS) for refractory epilepsy and migraines, spinal cord stimulation (SCS) for chronic pain, and sacral nerve stimulation (SNS) for bladder voiding dysfunction and fecal incontinence (Gurbani et al. 2016; Silberstein et al. 2020; Lam et al. 2023; Kaaki and Gupta 2020; Hull et al. 2013). Additional applications include cochlear implants (CI) for deafness, hypoglossal nerve stimulation (HNS) for sleep apnea, and functional electrical stimulation (FES) for restoring disabled limb function (Gay et al. 2022; Mashaqi et al. 2021; Pellot-Cestero et al. 2023). These techniques all work by applying electrical stimulations to nerve fibers—neuronal axons—either for blocking signal transmission or for producing axonal activity to modulate neuronal or muscle activity. Although our experiments in this book focused on the hippocampal region, our findings on axonal HFS (particularly from A-HFS experiments without involving synaptic transmissions) have general implications, providing valuable insights for stimulation applications in other brain regions and peripheral nerves.

In addition, we utilized electrophysiological methods in our studies, using microelectrode array recordings to directly measure neuronal activity with high spatiotemporal resolution. Research on DBS mechanisms has primarily relied on neuronal signal recordings from both animal and human being studies. These recordings fall into two categories: near-field and far-field. This book focuses on near-field recordings, which includes invasive recording of extracellular action potentials from surrounding neurons, as well as local field potentials. In contrast, far-field recordings include electroencephalogram (EEG) from the scalp and electrocorticogram (ECoG) from beneath the skull or dura mater. They have lower spatiotemporal resolution than near-field recordings due to signal attenuations by brain tissue, dura mater, skull, and skin. They mainly capture low-frequency synaptic potential signals. Although normal neuronal action potentials cannot be detected through far-field recording, integrated action potentials can become visible when large groups of neurons fire together abnormally—for instance, the spike waveforms appeared in the EEG signals of epilepsy patients. Only near-field recordings can detect action potentials from individual neurons, though EEG offers the advantage of being non-invasive. Beyond these direct electrical measurements, imaging techniques such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET) can indirectly track neural activity through blood flow changes. In addition, biochemical methods can examine neurotransmitter changes to provide more insights into the DBS mechanisms (Udupa and Chen 2015).

Conventional clinical applications of neuromodulation technologies, including DBS, primarily rely on real-time stimulation actions to control disease symptoms—once stimulation stops, symptoms return (Temperli et al. 2003). Our research on brain stimulations discussed in Part II also mainly focuses on neuronal responses during HFS. However, studies have shown that neurons can experience a period of inhibition after the end of HFS at a frequency around 100 Hz—commonly used by DBS. For example, significant reductions in excitatory postsynaptic currents were observed in rat primary motor cortex and subthalamic nucleus (STN) after HFS ended (Iremonger et al. 2006; Shen and Johnson 2008). Similarly, HFS in the globus pallidus internal segment (GPi) produced reduced post-stimulation neuronal firing in its thalamic projection area (Muralidharan et al. 2017). In both human GPi and our rat hippocampal experiments (Chap. 5), neuronal firing ceased immediately for several seconds after HFS ended (Lafreniere-Roula et al. 2010; Feng et al. 2017; Wang et al. 2018). These findings suggest that post-stimulation neuronal inhibition can occur in many stimulated brain regions. Importantly, clinical DBS studies have shown that Parkinson's patients experienced better outcome when they had longer periods of neuronal firing silence after 100 Hz STN stimulations (Milosevic et al. 2018), highlighting the clinical relevance of post-stimulation inhibitions.

Multiple mechanisms may cause reduced or absent neuronal firing after the end of HFS. In downstream projection neurons, increased inhibitions from GABAergic synapses may suppress neuronal firing when STN and GPi stimulations end (Chiken and Nambu 2013; Milosevic et al. 2018). For directly stimulated neurons, decreased excitability and elevated firing threshold can result in a silent period after HFS (Beurrier et al. 2001)—a finding supported by our experimental results described in Chap. 5. The antidromic conduction of HFS-induced axonal activity can affect the upstream somata (Sects. 5.2.3 and 5.2.4). This, combined with reduced excitatory inputs due to axonal block and synaptic failures, can produce an inactivity period in downstream neurons after the end of stimulation (Figs. 5.13 and 5.19). Furthermore, our experiments with both A-HFS and O-HFS in the rat hippocampal region showed clear differences in neuronal firing during and after HFS. While the stimulation enhanced neuronal firing during HFS, the firing decreased or completely stopped immediately after the HFS ended. This reflects a transition period of neuronal firing back to spontaneous activity after the removal of HFS modulation. Nevertheless, these post-stimulation inhibitory periods are very brief compared to the stimulation periods.

Conventional DBS, using continuous HFS lasting minutes to days or longer, can safely treat brain diseases. However, brief HFS lasting seconds or less, despite using identical pulses, can create animal epilepsy models through a “kindling” effect (Musto et al. 2009; Lothman and Williamson 1993). These brief HFS trains can cause irreversible long-term effects, including synaptic changes through plasticity mechanisms (Malenka 1994; Martin et al. 2000). In contrast, prolonged continuous HFS makes DBS reversible but restricts its therapeutic effects to the stimulation period. To achieve lasting effects after stimulation ends, or to have therapeutic benefits maintain between brief stimulations rather than only during them, alterative stimulation patterns are necessary. Animal studies have shown that certain patterns, such as burst stimulations, can produce therapeutic effects lasting several hours after stimulation (Spix et al. 2021). Our findings of different neuronal responses during initial and steady HFS periods have also revealed the distinctions between brief and prolonged HFS. Additionally, our studies on time-varying HFS and closed-loop HFS have provided new insights for developing stimulation patterns (detailed in Chapts. 6–8). Research into post-stimulation effects is one of important directions in DBS development—it can reduce stimulation duration, conserve electric energy, minimize treatment risks, and promote designs of sophisticated closed-loop stimulation paradigms.

Closed-loop DBS (also known as adaptive brain stimulation) has advantages including improved efficiency and efficacy, reduced side effects, and lower energy consumption. Although it has been used to treat epilepsy, open-loop stimulation remains the standard for other conditions like Parkinson's disease. The implementation of closed-loop DBS requires disease-specific biomarkers for feedback signals. Currently, two main types of neuro-electrical signals serve this purpose: rhythmic waves in local field potential (LFP) and action potentials from individual neurons. LFP, which contains lower frequency signals, represents the integrated electrical activity of neuronal populations measured locally. For example, β rhythmic waves (12–30 Hz) and γ rhythmic waves (60–90 Hz) in subthalamic nucleus LFP have been shown to correlate with motor dysfunction and bradykinesia in Parkinson's disease, showing promise as biomarkers for closed-loop DBS treatment (Bouthour et al. 2019; Little and Brown 2020). LFP biomarkers have also shown significant potential for adaptive stimulation in treating psychiatric disorders, such as major depression and obsessive–compulsive disorder, thereby advancing DBS applications in psychiatric treatment (Provenza et al. 2021; Sullivan et al. 2021). In refractory epilepsy treatment with adaptive stimulations, determining reliable feedback biomarkers remains challenging and requires more research to establish accurate epilepsy prediction signals (Ryvlin et al. 2021). The population spike (PS) recorded from the hippocampal soma layer, as introduced in this book, can serve as a biomarker for epileptiform activity to control closed-loop stimulations (refer to Sect. 8.4).

Additionally, when studying axonal stimulation—such as applying HFS to the CA1 alveus formed by pyramidal neuron axons—the soma ensemble can serve as an effective detector for recording APSs to reflect the axonal responses. The somatic membrane has a much larger surface area than thin axons, creating a stronger extracellular electric field during action potentials for accurate measurements from both population and single neurons (see Sect. 5.2). In the stratum oriens (basal dendritic layer) of hippocampal CA1 region, a composite action potential (CAP) can also be measured at the alveus axons near the soma layer to study axonal activity (Jensen and Durand 2009). However, CAPs have only sub-millivolt amplitudes—much smaller than somatic APSs—and can be influenced by somatic signals. Nevertheless, in spinal cord electrical stimulation, real-time recording of CAPs, also known as evoked compound action potential (ECAP), can achieve closed-loop spinal nerve stimulation (Mekhail et al. 2020; 2022). ECAP represents the extracellular potential created by the current leaking from the Ranvier nodes near the recording site when stimulation-evoked action potentials travel along axonal bundles.

In addition to electrophysiological biomarkers, biochemical substances can also serve as biomarkers for closed-loop DBS. When DBS alters neuronal firing and corrects pathological discharges, the resulting neural signals can travel to axon terminals and modify neurotransmitter releases. Real-time measurement of neurotransmitter concentrations, through electrochemical detection methods like voltammetry, can provide feedback signals to regulate stimulators, enabling closed-loop stimulations (Rojas Cabrera et al. 2020).

DBS uses electrodes implanted in the brain to connect stimulation devices with brain tissue, forming a neural interface that belongs to the broad category of brain-computer interface (BCI; also known as brain-machine interface, BMI). Open-loop DBS is a unidirectional interface, while closed-loop DBS is bidirectional, involving both the delivery of stimulation signals to the brain and the collection of neural feedback signals (Bouthour et al. 2019). DBS and BCI developed for different purposes. Medical experts and neurologists pioneered DBS as a treatment for brain diseases by applying electrical signals to modulate neural activity. Meanwhile, scientists and engineers with expertise in computational algorithms created BCI for robotics and neural rehabilitation, primarily aiming to read brain signals for controlling machines and computers. However, as both fields progress, they increasingly converge.

In its narrow sense, BCI conventionally refers to controlling machines, equipment, prostheses, and robots by decoding neuronal signals collected through intracranial electrodes (Chaudhary et al. 2016; Ajiboye et al. 2017; Willett et al. 2023). Alternatively, non-invasive BCI has also been used by collecting scalp EEG signals (Orban et al. 2022). While BCI technology primarily aims to assist disabled people, it also has a broader goal for advancing human intelligence through brain reading and writing (Yuste et al. 2017). In brain reading, a significant milestone occurred in 2006 when researchers first achieved mind-controlled machines by recording and decoding neural signals from the human brain (Hochberg et al. 2006). For brain writing, DBS represents a prototype by delivering electrical stimulations through implanted electrodes. Recent studies have shown that real-time closed-loop DBS applying to the human prefrontal cortex during sleep—synchronizing with endogenous slow-wave activity—can enhance memory consolidation (Geva-Sagiv et al. 2023). Beyond treating neurological diseases, DBS also aims to enhance brain function and human intelligence, aligning with BCI broader goals. Although BCI initially focused on “brain reading” and DBS on “brain writing”, both technologies are now evolving toward bidirectional capabilities and to a same destination through different paths.