Chronic pain poses one of the greatest public health challenges currently facing the United States, with more than 14 million Americans reporting they live with "a lot of pain" on a daily basis (1). In an effort to combat the pain epidemic, the prescription rate of opioids, a common pharmacological pain management strategy, quadrupled over the past 20 years (2). Over the same timeframe, there was a 200% increase in the rate of overdose deaths involving opioids (3). As the death rates from opioid-related drug overdoses have continued to climb in recent years (4), the need for non-addictive pain therapies has become even more pressing.
Neurostimulation therapies are effective, non-addictive treatment strategies for chronic pain that is refractory to conventional medical management. Spinal cord stimulation (SCS) is a neurostimulation therapy primarily used to treat intractable neuropathic pain in the lower limbs (e.g. failed back surgery syndrome) (5).
SCS is achieved by implanting an electrode lead in the spinal epidural space, and applying brief electrical impulses to the dorsal columns (6). However, due to the complex anatomy of the spinal column, SCS struggles to treat certain pain etiologies, particularly pain that is highly focalized to specific dermatomes (i.e. regions of the body), such as the groin and foot (7).
Administration in early 2016 for the treatment of intractable complex regional pain syndrome in the lower limbs (8,9), and has shown success at treating several other focal pain indications (e.g. phantom limb pain, painful diabetic neuropathy, groin pain) (10)(11)(12). In contrast to SCS, DRGS electrodes are placed in the intraforaminal space, where they apply electrical stimulation to a single dorsal root ganglion (DRG). There are bilateral pairs of DRG at each level of the spinal cord, with each DRG containing the cell bodies, and a portion of the axons, of all the primary sensory neurons innervating a single dermatome (13). DRG neurons are pseudounipolar: a single axon process extends from the soma, bifurcates at a large node of Ranvier called the T-junction, and forms an axon that projects to the spinal cord and an axon that extends to the periphery (14). Due to the precise targeting of a single dermatome's primary afferents, DRGS is rationalized to provide patients with focal, dermatome-specific pain relief.
Although preliminary clinical results indicate that DRGS provides adequate pain relief for many patients, approximately 20-30% of patients do not receive sufficient pain relief from DRGS (9). Furthermore, long-term studies showed that DRGS may lose efficacy over time (12,15), a trend also found in long-term clinical studies of SCS (16,17). These two shortcomings of DRGS can partially be attributed to the fact that we do not have a clear understanding of the physiologic mechanisms of action of DRGS. Uncovering the mechanisms by which DRGS provides pain relief will allow scientists and engineers to innovate the technology to specifically target these mechanisms to ultimately improve clinical outcomes.
A recent computational modeling study suggested that DRGS may provide pain relief by augmenting a low-pass filtering mechanism at the T-junction, preventing nociceptive impulses from propagating from the periphery to the spinal cord (18). However, the stimulation amplitudes necessary to augment T-junction filtering (> 9 mA) were far greater than the amplitudes used clinically ( 1 mA on average). Using a similar computational modeling approach, we recently showed that for stimulation parameters used clinically, DRGS may directly activate large-diameter myelinated A low threshold mechanoreceptors (A-LTMRs) (putative innocuous touch-sensing neurons), but not small-diameter nonmyelinated C-nociceptors (putative nociceptors) (19). However, this study did not examine the effect of DRGS on A-neurons, large myelinated afferents that innervate muscles (20), or A-neurons, a diverse class of medium-diameter, thinly myelinated afferents that can convey noxious or innocuous sensations (21). Furthermore, there are several clinical factors which affect the delivery of electrical stimulation to neural tissue, such as the placement of the electrode lead relative to the neural target (19,22,23) and the choice of stimulation parameters (e.g. pulse width, pulse frequency) (24,25).
Currently, it remains largely unclear how these clinically-controllable parameters affect the recruitment of different types of primary afferents for pain relief during DRGS. It is vital to understand how factors under clinical control influence neural activation, as these factors likely directly influence the efficacy of DRGS.
Therefore, these factors are key in ensuring positive patient outcomes.
In this work, we employed a computer model of DRGS to study how clinical factors, such as electrode position and stimulation parameter selection, affect primary afferent recruitment. We coupled a finite element model (FEM) of a human L5 DRG to multi-compartment models of primary sensory neurons to study how these clinically-controllable factors affected neural activation during DRGS. We considered our results in context with the recommendations of the Neuromodulation Appropriateness Consensus Committee (NACC) on best practices for DRGS (26).
Building upon our prior work (19), we developed a computer model of DRGS to investigate how clinically-controllable factors (e.g. electrode position, stimulus parameter selection) affect neural activation in the DRG. We coupled a finite element model (FEM) of a human L5 DRG to multi-compartment models of sensory neurons. We used the FEM to calculate the voltage distribution generated by DRGS, and applied these voltages to the multi-compartment models. We examined which types of neurons generated action potentials in response to DRGS, and how neural activation patterns changed as we varied electrode position and stimulation parameters (i.e. pulse width, pulse frequency, stimulus configuration).
We constructed an FEM of a human L5 DRG and its surrounding anatomy (e.g. intraforaminal tissue, foraminal bone) based on experimentally measured values (Table 1, Figure 1). We based the geometry of the model on imaging and cadaver studies of human DRG and foraminal tissues (27)(28)(29)(30). We wanted to examine the effect of electrode lead position relative to the ganglia, and the effect of lead distance from the ganglia on primary afferent recruitment. Therefore, we used a larger measured value for foraminal height (17.1 mm) in this model than in our previous model to allow a greater range of distances to be tested (27). We set the electrical conductivity of each tissue (Table 2) to the values used in our previous study of DRGS (19,(31)(32)(33). We modeled all conductivities as isotropic, with the exception of the nerve root, which we modeled as twodimensionally (2D) anisotropic white matter (19). We built the FEM in the commercially-available software 3matic Module within the Mimics Innovation Suite (Materialise, Belgium). We included an explicit representation of a four-contact DRGS electrode array in the FEM, with the second electrode contact centered above the middle of the DRG. In some simulations, we shifted the electrode 3.125 mm laterally along the nerve root axis, such that the midpoint between the second and third contacts was centered above the middle of the DRG (i.e. so the second and third contacts straddled the ganglion). We surrounded the electrode with a 300 m encapsulation layer to represent the foreign body response to implanted materials (33).
We imported the FEM into COMSOL Multiphysics (COMSOL, Inc., USA). To simulate bipolar DRGS (26), we applied a unit current stimulation boundary condition (i.e. 1 A) to the active electrode contact and grounded (i.e. 0 V) the return contact. To improve stimulation selectivity, clinical SCS sometimes utilizes a guarded cathode stimulation configuration: two electrode contacts adjacent to the active contact are used as return contacts (34)(35)(36). To our knowledge, the use of a guarded cathode stimulation configuration in DRGS has not been reported. To explore the utility of the guarded cathode configuration in DRGS, we applied current stimulation to either the second or third contact, and grounded the contacts immediately adjacent to the active contact (e.g. grounding the first and third contacts for an active second contact). In all simulations, we modeled the electrode lead shaft as a perfect insulator, and inactive contacts as equipotential with zero net current across their surface. To calculate the voltage distribution generated by DRGS, we used the conjugate gradient method to solve Laplace's equation: where 𝜎 is the tissue stiffness matrix, and Φ is the calculated voltage distribution. Figure 2 shows voltage distributions generated by DRGS using example stimulation configurations. We validated the FEM by its ability to produce bipolar impedances similar to impedances reported clinically (26). To calculate model impedance, we divided the average voltage across the active contact's surface by the applied stimulus current. From the ACCURATE clinical trial, average bipolar DRGS electrode impedances one year post-implant were 1458.9 714.5 (26). Our models produced a minimum bipolar impedance of 1357.55 , when the active contact was centered immediately above the DRG and the return contact was adjacent to the active electrode and more proximal to the spinal cord. Our models produced a maximum bipolar impedance of 1551.54 when the most distal contact and most proximal contact formed a longitudinal bipole pair. All model bipolar impedances fell within clinical range.
We implemented multi-compartment models of primary sensory neurons found in human DRG using the NEURON simulation environment (v7.4) (37). We implemented previously-published models of an A-LTMR and C-nociceptor (Figure 3a,d) (19). We developed a model of an A-afferent by extending the previously-published model of an Aβ-LTMR to include large-diameter (i.e. 15-16 µm) axons. Because A-and A-neurons share electrophysiological characteristics (38), we implemented the membrane dynamics of our previously-published Aβ-LTMR in our A-neuron model (Figure 3a). The A-neuron reproduced many action potential (AP) and conduction velocity (CV) data seen in experimental literature (e.g. AP height, duration) (Table 3). We developed two distinct multi-compartment models of A-neurons: an LTMR (Figure 3b) and a high-threshold mechanoreceptor (HTMR) (Figure 3c) (21). A-LTMRs and A-HTMRs express distinct voltagegated sodium channel profiles (39). A-LTMRs mainly express Nav1.6, similar to other non-nociceptive myelinated mechanoreceptors (40), while A-HTMRs express Nav1.7 and Nav1.8, similar to C-nociceptors (41,42). Therefore, we developed two A-neuron models, an LTMR and an HTMR model, distinguished by their active voltage-gated sodium channels. Both models had the same morphology. Each model had a soma 29 m long and 34 m wide, connected to a 3.0 m stem axon (43). The A-neuron axon morphologies (Table 4) were based on the MRG model of a mammalian peripheral axon (44). The stem axon extended 840 m to match the total stem axon and soma length of the A-neuron, A-LTMR, and C-nociceptor models (i.e. 869 m), before splitting into two axons. One axon projected towards the spinal cord, with a diameter of 2.0 m (43,45), while the other projected to the periphery and had the same diameter as the stem axon (i.e. 3.0 m) (43,46). The original MRG axon model was parametrized for axons of specific diameters (e.g. 2.0, 5.7, 7.3 m). To implement an axon with a diameter not included in the original model, such as the 3.0 m stem and Figure 3: Multi-compartment models of DRG sensory neurons. We implemented models of five types of sensory neurons found in DRG: (a) a large-diameter, myelinated Aα-and A-neurons; (b) a smalldiameter, thinly-myelinated A-LTMR; (c) a small-diameter, thinly-myelinated A-HTMR; and (d) a smalldiameter, nonmyelinated C-nociceptor. The Aα-neuron, A-LTMR, and A-LTMR putatively convey innocuous sensory information, while the A-HTMR and C-nociceptor are putatively nociceptive. For each cell model, the equivalent circuit diagrams show the active voltage-gated ion channels included in each cell type and a linear leak conductance. Inset action potentials represent the somatic membrane response to a brief intracellular current pulse applied to the peripheral axon. peripheral axons used in the A-neuron models, we performed a linear regression to calculate the values of each parameter (e.g. number of myelin lamellae, internodal length) for a given axon diameter (47).
The nodes of Ranvier in the myelinated axon models were separated by three distinct finite impedance myelin segments: two myelin attachment sections, two paranode main segments, and six internodal segments (48). To reduce computational demand of the A-neuron models (i.e. medium diameter myelinated axons with short internode distances), we modeled the internodal segments farther than 20 mm from the active electrode with only a single compartment. This simplification did not produce significant differences in the activation thresholds. The nodes of Ranvier contained active sodium and potassium conductances, and a linear leak conductance. Both models contained an A-type and delayed rectifier potassium conductance (49). The A-LTMR model nodes contained a Nav1.6 conductance (50), while the A-HTMR model nodes contained a Nav1.7 and Nav1.8 conductance (49). We set all nodal sodium conductances to 3.0 S/cm 2 (44). We set the LTMR model's somatic Nav1.6 conductance to 1.0 S/cm 2 , to best reproduce electrophysiological data reported by literature. To best reproduce electrophysiological data, we set the HTMR model's somatic Nav1.8 conductance to 0.3 S/cm 2 , A-type potassium conductance to 0.28 S/cm 2 , delayed rectifier potassium conductance to 6 mS/cm 2 , and all compartments' leak conductance to 2 mS/cm 2 . We validated the models based on their ability to reproduce AP and CV data reported in literature (Table 5) (38,51,52).
Next, we distributed our multi-compartment models of primary afferent neurons throughout the DRG FEM as described previously (19). Previous histological studies of mammalian and human DRG showed that cell bodies preferentially organize around the dorsal edge of the ganglion (53,54). Therefore, we generated two 2D regular grids -one intersecting the midpoint of the ganglion in the sagittal plane, and the other in the transverse plane -with 100 m spacing in all directions, resulting in 2,304 points. We used each point on the grid as a seed point for the somata of the cell models described in Step 2. From each seed point, the stem axon projected towards the midline of the ganglion, then bifurcated into central and peripheral axon processes that curved ventrally to enter the nerve root (Figure 4a).
We interpolated the extracellular potentials calculated in equation ( 1) onto the middle of each compartment of the cell models generated in Step 2. We applied the extracellular potentials to the multicompartment models using NEURON's extracellular mechanism within the Python programming language (55). We calculated each compartment's time-varying membrane voltage in response to DRGS by using a backward Euler implicit integration method with a time step of 5 s (Figure 4b). The tissue conductivities of the FEM were linear. Therefore, the voltage distribution generated by a specific DRGS amplitude was a scalar multiple of the voltage distribution generated by a unit stimulus (i.e. a 1 A stimulus) (56).
Our goal was to study how clinically-controllable factors (e.g. electrode lead position, stimulus parameters) affected neural activation in the DRG. Therefore, for each simulation, we calculated the minimum stimulus amplitude necessary to elicit one or more action potentials in each neuron type (i.e. the activation threshold). Each stimulus pulse was a charge-balanced, biphasic pulse with an active stimulus phase and a passive discharge phase with an interphase interval of 20 s (18,57). Unless otherwise noted, we calculated activation thresholds in response to a single stimulus pulse with a 300 s active phase (the approximate average pulse width reported by the ACCURATE clinical trial (9)), with the electrode lead positioned directly above the ganglion (i.e. with a 0 mm electrode shift). We calculated activation thresholds for both anodic-and cathodic-first pulses (i.e. stimulus pulses with a positive active phase and negative active phase, respectively), using a binary search algorithm with a resolution of 0.1 A.
We sought to identify which types of primary afferents (i.e. A-neurons, A-LTMRs, A-neurons, Cnociceptors) are likely activated by DRGS within parameter ranges used clinically. Therefore, we defined a maximum clinical amplitude of 2.0 mA, which is approximately the mean DRGS amplitude plus two standard deviations at one-year post implant reported by the ACCURATE clinical study (26). We considered any neuron with an activation threshold less than or equal to this maximum clinical amplitude (i.e. 2 mA) as activated within clinical ranges of stimulation parameters (i.e. clinical DRGS). When examining the effect of stimulus pulse frequency, we simulated 200 ms of DRGS with a pulse amplitude of 1 mA, and a pulse width of 300 s. Then, we calculated the response frequency, i.e. the frequency of action potentials generated in response to DRGS of different pulse frequencies, of different types of neural compartments (e.g. the soma, the spinally-projecting axon, etc.). Recent clinical studies have highlighted the importance of the electrode location relative to the ganglion to DRGS-induced pain relief (58,59). To examine the effect of electrode position on DRGS-induced neural activation, we calculated activation thresholds for each cell type in response to DRGS applied with several electrode locations relative to the ganglion. Figure 5 shows the percentage of modeled neurons with activation thresholds within clinical range (i.e. ≤ 2 mA) as distance between the lead and the DRG increased, both when the active electrode contact was centered above the ganglion and when the active and return contacts straddled the ganglion.
In general, for all electrode lead positions, clinical DRGS only activated myelinated afferents (i.e. A-, A-and A-neurons) and did not activate nonmyelinated C-nociceptors (Figure 5). As the distance between the active contact and the ganglion increased, the percentage of myelinated afferents activated by clinical DRGS decreased. Generally, straddling the active and return contacts across the ganglion activated a larger percentage of A-LTMRs than centering the active contact above the ganglion.
When the active contact was centered directly above the ganglion (i.e. with a 0 mm electrode shift), anodic-first DRGS (Figure 5a) activated 12% of A-LTMRs, while cathodic-first DRGS (Figure 5c) activated less than 1% of modeled A-LTMRs. When the active and return contacts straddled the ganglion, anodic-first (Figure 5b) and cathodic-first (Figure 5d) DRGS activated 2% and 3% of modeled A-LTMRs, respectively. However, shifting the electrode lead 1 mm dorsally abolished all A-LTMRs. We observed minimal activation in A-HTMRs. Anodic-first DRGS activated 3% of modeled A-HTMRs only during anodic-first DRGS when the active contact was centered directly above the ganglion (Figure 5a). Shifting the electrode lead 1 mm dorsally abolished all A-HTMR activation. We did not observe A-HTMR activation during cathodic-first DRGS (Figure 5c,d), nor during anodic-first DRGS when the active and return contacts straddled the ganglion (Figure 5b). We always observed a larger percentage of A-LTMR activation than A-neuron activation.
We observed A-neuron and A-LTMR activation for all electrode positions. When the active contact was centered directly above the ganglion, anodic-first DRGS activated 100% and 92% of modeled A-and Aneurons, respectively (Figure 5a), while cathodic-first DRGS activated 99% and 46% of modeled A-and Aneurons, respectively (Figure 5c). When the active and return contacts straddled the DRG, anodic-and cathodic-first DRGS both activated 88% of modeled A-LTMRs and more than 98% of modeled A-neurons (Figure 5b,d). Increasing the distance between the active contact and the ganglion decreased the percentage of both A-and A-neurons activated by clinical DRGS. When the electrode lead was shifted dorsally 1 mm or greater from the ganglion, cathodic-first DRGS applied with the active and return contacts straddling the ganglion produced the greatest A-LTMR activation (Figure 5d). When the electrode lead was shifted dorsally 3 mm from the ganglion, we only observed A-LTMR activation when cathodic-first DRGS was applied with the active and return contacts straddled the ganglion (Figure 5d). We observed A-neuron activation regardless of the distance from the electrode lead distance to the ganglion.
Stimulus pulse width is a critical parameter when programming a patient's DRGS system, and has been shown to affect neural activation (24) and paresthesia distribution (60) during SCS. Interestingly, increasing SCS pulse width lowered the activation threshold of small-diameter myelinated dorsal column axons (24). However, the effect of stimulus pulse width on neural activation during DRGS has not been rigorously studied.
Therefore, we calculated primary afferent activation thresholds for several pulse widths (i.e. 100, 200, 300, 500, and 1000 µs) both when the active contact was centered above the ganglion, and when the active and return contacts straddled the DRG. Figure 6 shows the percentage of modeled neurons with activation thresholds in clinical range (i.e. ≤ 2 mA) as the stimulus pulse width increased, both when the active contact was centered above the ganglion, and when the active and return contacts straddled the ganglion. For amplitudes within the clinical range, we did not observe activation of nonmyelinated C-nociceptors for any pulse width. However, we did observe activation of small-diameter thinly-myelinated A-neurons. When the active contact was centered above the ganglion, the minimum pulse widths for anodic-first DRGS (Figure 6a) to activate one or more A-LTMRs or A-HTMRs were 100 s and 200 s, respectively, while the minimum pulse widths for cathodic-first DRGS (Figure 6c) to activate one or more A-LTMRs or A-HTMRs were 300 s and 1000 s, respectively. When the active and return contacts straddled the ganglion, the minimum pulse widths for anodic-first DRGS to activate one or more A-LTMRs or A-HTMRs were 300 s and 1000 s, respectively (Figure 6b), while the minimum pulse widths for cathodic-first DRGS to activate one or more A-LTMRs or A-HTMRs were 200 s and 1000 s, respectively (Figure 6d). DRGS always recruited more A-LTMRs than A-HTMRs, regardless of pulse width or stimulus polarity.
We observed A-and A-neuron activation by clinical DRGS for all pulse widths. Increasing the stimulus pulse width increased the percentage of A-and A-neurons activated, regardless of stimulus pulse polarity or position of the active and return contact relative to the ganglion. For stimulus pulse widths below 300 s, cathodic-first DRGS applied with the active and return contacts straddling the ganglion activated the largest percentage of A-LTMRs (59-87%) (Figure 6d). For stimulus pulse widths greater than or equal to 300 s, anodic-first DRGS applied with the active contact centered above the ganglion activated the largest percentage of A-LTMRs (92-100%) (Figure 6a). Anodic-first DRGs applied with the active contact centered above the ganglion always activated the largest percentage of A-neurons (99-100%) (Figure 6a) For all pulse widths and stimulus polarities, there was always a larger percentage of A-and A-neurons activated by clinical DRGS than any other neuron type.
Stimulus pulse frequency is an important parameter during neurostimulator programming, and recent innovations in neurostimulation for pain have focused chiefly on this parameter (e.g. 10 kHz SCS (25)). DRGS is, on average, applied at 20 Hz, but can be applied as low as 4 Hz or as high as 80 Hz (26). Presently, we do not understand the physiological effect of varying stimulus pulse frequency, nor how those effects translate to clinical outcomes. Therefore, we applied anodic-and cathodic-first DRGS with a pulse amplitude of 1 mA, a pulse width of 300 s, and examined the time-varying membrane potential of different neural compartments (e.g. soma, axons) in response to DRGS applied at different pulse frequencies.
Regardless of pulse frequency, we did not observe activation of nonmyelinated C-nociceptors for any stimulus pulse frequency, for DRGS applied within standard clinical parameter ranges. We observed minimal activation of both thinly myelinated A-LTMRs and A-HTMRs (a maximum of 12% and 3%, respectively). As the majority of our data suggests that DRGS predominantly activates large myelinated neurons, and A-LTMRs are believed to play an important role in both DRGS-induced pain relief and physiologic pain inhibition in the spinal cord, (19,61), we focused the rest of our analyses on A-LTMRs.
Figure 7 shows the average response frequency of different A-LTMR compartments (i.e. soma, stem axon, centrally-projecting axon, peripherally-projecting axon) to DRGS applied at different frequencies, for A-LTMRs that responded to DRGS applied with a 1 mA stimulus pulse. For most stimulus polarities and positions of active and return contacts, all A-LTMR compartments responded in a one-to-one fashion with the DRGS pulse (i.e. one action potential for each stimulus pulse). However, in response to anodic-first DRGS with the active contact centered above the ganglion, the stem axon's response frequency was slightly larger than unity, while all other compartments' response frequencies were slightly below unity (Figure 7a). The increase in stem axon response frequency was caused by a rebound action potential propagating down the stem axon following a somatic action potential, which fails to propagate passed the T-junction. The decrease in the average response frequencies in other compartment was attributed to neurons with activation thresholds near 1 mA.
For those neurons, some DRGS pulses elicited an action potential, while some pulses failed to induce an action potential.
Clinical DRGS utilizes a bipolar stimulation configuration in which the cathode is typically placed directly beneath the pedicle (26). However, the position of the DRG within the foramen varies across patients and spinal level (27,62). X-ray fluoroscopy, the imaging modality used to visualize electrode position relative to bony structures during DRGS electrode implantation, is unable to resolve neural tissue, implying that clinicians are unable to precisely determine the position of the ganglia relative to the active contact. Therefore, the position of the stimulating contacts relative to the DRG is likely variable across patients, warranting investigation into how bipole placement relative to the ganglion affects neural activation during DRGS.
Figure 8 shows the percentage of modeled A-LTMRs activated by various bipolar DRGS configurations. For bipolar configurations where the active and return contacts were adjacent to each other (Figure 8a), cathodic-first DRGS (light gray bars) typically activated more A-LTMRs than anodic-first DRGS (dark gray bars), except when the active contact was centered above the ganglion. Cathodic-first DRGS applied with the return contact centered above the ganglion, and the active contact above the peripheral nerve root, activated the largest percentage of modeled A-LTMRs (99%). Anodic-first DRGS applied with the active electrode above the spinal nerve root activated the smallest percentage of modeled A-LTMRs (40%). In general, placing the cathode near the peripheral nerve root, or placing the anode near the ganglion, maximized A-LTMR activation.
For bipolar configurations with one or more inactive contacts separating the active and return contacts (i.e. separated bipoles; Figure 8b), cathodic-first DRGS typically activated more A-LTMRs than anodic-first DRGS. For separated bipole configurations, anodic-first DRGS only activated more A-LTMRs than cathodicfirst DRGS when the active contact was centered above the ganglion, similar to bipolar DRGS with adjacent active and return contacts. Three separated bipole configurations activated more than 98% of modeled A-LTMRs, each of which applied cathodic-first DRGS with the active contact above the peripheral nerve root, and the return contact as either the first or second most proximal contact. Anodic-first DRGS applied with a distal active contact and proximal return contact (i.e. placing the active contact above the peripheral nerve root and the return contact above the spinal root) activated the smallest percentage of modeled A-LTMRs (37%).
Similar to adjacent bipole configurations, placing the cathode near the peripheral nerve root, or placing the anode near the ganglion, maximized A-LTMR activation. However, except for the case when the active electrode was centered above the ganglion, cathodic-first DRGS using a separated bipole configuration always activated more than 80% of modeled A-LTMRs, while only half of the adjacent bipole configurations activated more than 80% of modeled A-LTMRs. Guarded cathode configurations are commonly used in clinical SCS (34), but to our knowledge have not been documented in DRGS studies. Therefore, we examined neural activation patterns resultant from DRGS applied with guarded cathode stimulation configurations. We again observed no activation of nonmyelinated C-nociceptors, and minimal activation of A-LTMRs and A-HTMRs (a maximum of 12% and 3%, respectively), and therefore we focused our analysis on A-LTMRs.
Figure 9 shows the percentage of modeled A-LTMRs activated by various guarded cathode DRGS configurations. Similar to conventional and longitudinal bipole configurations, anodic-first DRGS applied with a guarded active contact activated more A-LTMRs than cathodic-first DRGS only when the active contact was centered above the ganglion. When the active contact was close to a nerve root, or when the active contact and one of the return contacts straddled the ganglion, cathodic-first DRGS activated more A-LTMRs than anodic-first DRGS. For guarded active contact configurations, applying anodic-first DRGS with the active contact centered above the ganglion maximized A-LTMR activation (89%). For cathodic-first DRGS applied with a guarded active contact configuration, placing the active contact near the peripheral nerve root maximized A-LTMR activation (87%).
DRGS is a safe and effective therapy for chronic pain that is refractory to conventional medical management. To ensure accurate delivery of electrical stimulation to the DRG, several clinically-controllable parameters, such as the position of the active and return contacts and the shape of the stimulus pulse, must be carefully tuned to maximize pain relief without producing uncomfortable sensations. It is presently unclear how varying each of these parameters affects which types of sensory neurons are directly activated by DRGS. Our data suggest that DRGS applied with stimulation amplitudes within a clinical range (i.e. 2 mA) predominantly activates large-diameter myelinated afferents (e.g. Aα-and A-neurons). Furthermore, our results indicate that the position of the active and return contacts relative to the ganglion has the greatest effect on DRGS-induced A-LTMR activation.
Our data suggest that clinical DRGS directly activates myelinated neurons, regardless of electrode position, stimulation configuration, and stimulus pulse parameters (i.e. frequency, pulse width). This corroborates our previous study's findings that clinical DRGS is likely driving the activity of large-diameter myelinated A-LTMRs, without directly activating small-diameter nonmyelinated C-nociceptors (19), and is supported by recent experimental findings that DRGS applied with non-penetrating electrode arrays activates neurons with conduction velocities in the A-to A-axon range (63). Furthermore, our data also suggest that clinical DRGS activates Aα-neurons, and may activate A-neurons, though in a considerably smaller proportion than Aα-and A-neurons. Based on these findings, DRGS may provide pain relief by driving pain-gating mechanisms in the dorsal horn, via postsynaptic activation of inhibitory interneurons which receive input from large myelinated afferents. This hypothesis suggests similarities between the mechanisms of DRGS and SCS.
Previous animal studies of SCS demonstrated increased levels of -aminobutyric acid (GABA), an inhibitory neurotransmitter, in the spinal cord (64), and increased paw withdrawal thresholds in animals that responded to SCS (65) predominantly driven through the GABAB receptor (66,67). However, a recent study in rats concluded that DRGS does not cause GABA release in the dorsal horn (68), indicating DRGS may provide pain relief through other mechanisms.
Koetsier and colleagues suggested that instead of driving GABAergic inhibition in the dorsal horn, DRGS may provide pain relief by inducing GABAergic inhibition through GABA signaling within the DRG (68).
Du and colleagues concluded that in rats, activating small-, medium-, or large-diameter DRG neurons can induce GABA release in the DRG, and that nearly all small-diameter DRG neurons (putative nociceptors) can respond to GABA (69). Furthermore, they demonstrated that VGAT, the primary transporter for inhibitory neurotransmitter reuptake, was commonly found in DRG neurons that co-expressed 200 kDa neurofilament, a marker of myelinated afferents in rats. Optogenetically stimulating VGAT expressing DRG neurons produced a marked reduction in nocifensive behavior, and GABAA receptor antagonists increased nocifensive behavior even when no noxious stimuli were present, suggesting that a GABAergic pain-gating system exists at the level of the DRG. Taking these findings in context with our model predictions, it is possible that clinical DRGS provides pain relief by directly activating medium-and large-diameter myelinated afferents, causing local GABA release in the DRG to inhibit nociceptive afferents, thereby preventing pain signals from reaching the central nervous system. However, GABA is not the only inhibitory neurotransmitter in the spinal dorsal horn. Many dorsal horn neurons which release GABA co-release glycine, with some inhibitory postsynaptic currents in superficial dorsal horn laminae mediated exclusively by glycine (70). Recent studies have identified a glycinergic feedforward dorsal horn circuit that gates mechanical allodynia, mediated by paravalbumin (PV) expressing interneurons in lamina IIi and III of the dorsal horn, which silence lamina II PKC+ interneurons that form excitatory synapses on projection neurons in lamina I (71)(72)(73). These PV+ interneurons receive afferent input from both A-and A-axons (73), and gate mechanical pain via pre-and post-synaptic inhibition of primary afferents and lamina II excitatory interneurons, respectively (72,73). Subsequent studies uncovered novel neural circuits involved in, and further somatosensory functions of, glycinergic inhibition. Foster and colleagues demonstrated that selective ablation of dorsal horn interneurons which express GlyT2, the glycine transporter expressed in the spinal cord and brainstem, facilitates mechanical, and cold hyperalgesia, and can induce spontaneous pain behaviors (74). Cui and colleagues identified a separate subpopulation of deep laminae (III-V), predominantly glycinergic interneurons which receive mono-and polysynaptic input from both A-and C-axons, and form a feed-forward gate to silence pain transmission from PKC+ and somatostatin+ superficial dorsal horn neurons (75). The results of these studies emphasize the complexity of sensory processing in the dorsal horn, and taken together with the data presented in this work, suggest that DRGS may provide pain relief through a combination of glycinergic inhibition in the dorsal horn and GABAergic inhibition within the DRG itself. To fully elucidate the mechanisms of action of DRGS, we must uncover how innocuous and noxious stimuli are processed in dorsal horn and supraspinal structures in healthy and pathological states, and how the pattern of DRGS-induced afferent activity augments or abrogates neural activity throughout the neuraxis.
The location of the DRGS electrode lead relative to the DRG likely varies across patients, depending on the size of the patients' neuroforamina, the position of their DRG within the foramen (27,62), and the implanting physician's placement of the electrode lead relative to the patient's DRG. The NACC suggests that straddling the second and third contacts across the medial and lateral borders of the pedicle is the optimal position of the lead in the foramen (26). This positioning likely resembles our models where the second and third contacts are straddling the ganglion, as lumbar DRG are typically located in the 'foraminal zone' (i.e. beneath the pedicle) (27). Martin and colleagues found that power consumption by the implanted pulse generator (IPG) was minimized by electrodes placed superodorsally in the foramen, a similar position to what the NACC suggests to be optimal (58). However, Martin and colleagues found that clinical outcomes were not dependent on the position of the electrode in the foramen. That study suggests that DRGS is able to activate analgesic mechanisms with the electrode placed anywhere in the foramen, at the cost of additional power consumption, though the optimal electrode positioning would result in straddling the ganglion with the active and return contacts.
Our modeling results, and our hypothesis that DRGS provides pain relief by driving the activity of largediameter myelinated afferents, corroborate these notions. We demonstrated that DRGS consistently activates Aα-and A-neurons regardless of the positioning of active and return contacts relative to the ganglion, selection of stimulus parameters, and distance between the electrode lead body and the ganglion (Figure 5). Furthermore, we showed that straddling the active and return contacts across the ganglion maximizes A-LTMR activation (Figure 5d). Straddling the DRG with the active and return contacts, and placing the electrode lead in close proximity to the ganglion, would therefore enable consistent DRGS-induced analgesia while minimizing IPG power consumption, thereby reducing the need for battery replacement surgeries.
Programming a patient's DRG stimulator is a crucial, and often time-consuming process, as the stimulus parameter space is large and cannot be fully explored in a single programming session. This process is further complicated by the fact that we do not fully understand how varying each parameter, such as pulse width, pulse frequency, and the placement of cathodes and anodes, affects neural recruitment during DRGS.
The median DRGS pulse width and pulse frequency reported by the ACCURATE trial were 300 s and 20 Hz respectively, and the NACC highlights that pulse widths utilized by patients typically decrease over time, and that shorter pulse widths maximize the therapeutic window (26). Our results again corroborate the NACC recommendations; we demonstrated that regardless of pulse width, DRGS is predominantly activating A-LTMRs, especially with shorter pulse widths (Figure 6). However, when using 500 s or 1000 s pulse widths, our models predicted increased activation of both A-LTMRs and A-HTMRs (Figure 6a,d). Activation of smaller diameter axons in response to longer pulse widths has also been reported in SCS studies, where increased pulse widths increased activation of small diameter myelinated axons in the medial dorsal columns (24,36). Increased activation of A-neurons, particularly A-HTMRs, could explain why some patients report uncomfortable or painful sensations in response to DRGS of longer pulse widths. Therefore, shorter pulse widths (i.e. 300 s) may maximize activation of target neurons (i.e. A-LTMRs), while minimizing activation of nociceptive neurons, increasing the therapeutic window of DRGS.
From the ACCURATE clinical trial, the median DRGS pulse frequency was 20 Hz, and the maximum frequency across the DRGS patient cohort did not exceed 50 Hz at any time point (9). Our data demonstrate that in general, DRGS elicits one-to-one action potential generation in most A-LTMRs (Figure 7), suggesting DRGS consistently sends propagating action potentials to the dorsal horn via A-LTMRs. However, our data do not explain why DRGS patients typically utilize lower frequencies (around 20 Hz on average) compared to SCS patients (around 50 Hz on average (76)), especially considering the two therapies may operate through similar mechanisms (i.e. driving the activity of A-LTMRs in the DRG vs. in the dorsal columns). Future experimental and clinical studies are necessary to understand the effect of DRGS pulse frequency both on the mechanisms of the therapy and on clinical outcomes.
As described above, the NACC suggests straddling the second and third contacts across the target pedicle and applying bipolar, cathodic-first DRGS. However, it is currently unknown how selecting contacts as anodes and cathodes affects neural activation during DRGS. We showed that regardless of which contacts were set to be active, DRGS activated a substantial portion (~40%) of modeled A-LTMRs (Figure 8).
Generally, placing an anode near the body of the ganglion, or placing a cathode near a nerve root, particularly near the peripheral nerve root which contains larger-diameter axons, maximized A-LTMR activation. This agrees with our previous findings (19), and follows conventional neurostimulation theory that anodic stimulation results in lower activation thresholds when the electrode is near a cell body, and cathodic stimulation results in lower activation thresholds when the electrode is near an axon of passage (22). Guarded cathode configurations, placing an anode on either side of a cathode, are commonly used in traditional SCS. Because traditional SCS putatively targets the dorsal columns -axons of passage running parallel to the implanted electrode -guarded cathodes are thought to maximize the therapeutic window of SCS by maximizing the activating function along dorsal column axons (77). We explored if DRGS applied with a guarded active contact configuration would provide greater A-LTMR activation than other stimulus configurations. However, we did not find any added benefit of the guarded cathode configuration compared to bipolar configurations, regardless of stimulus polarity (Figures 8,9). As DRG neurons are not axons of passage, but instead have pseudounipolar morphologies (14), the guarded cathode configuration is unlikely to maximize neural activation during DRGS. Our data again add support to the NACC recommendation that 'complex programming arrays are not necessary, as simple bipolar arrays can achieve optimal activation of the DRG,' (26).
Although we built our models using previously-published clinical and experimental data, there are several limitations to our approach. The FEM of a human L5 DRG used in this study represented several anatomical compartments (e.g. foraminal bone, intraforaminal tissue) as largely concentric cylinders. Although the method of representing anatomical compartments as simplified concentric shapes has been commonly used to study other clinical neurostimulation therapies (25,78), recent work demonstrated that the complex anatomy of bony structures in the spine can affect model predictions of SCS-induced neural activation (23).
Future studies could employ a patient-specific modeling approach, similar to previous studies of SCS (79), which could elucidate how the complex anatomy of the spinal column affects DRGS model predictions.
Human lumbar DRG somata are typically located around the dorsal edge of the ganglion (54). To study how DRGS would affect a specific cell type at any location in the DRG, we homogeneously distributed each cell type throughout the DRG, with their cell bodies placed around the dorsal edge of the ganglion. However, the actual distribution of functional subpopulations of DRG neurons (e.g. A-neurons) is likely not homogeneous. Our data suggest DRGS causes widespread activation of large diameter A-neurons, which carry proprioceptive and stretch-receptor information from the muscles. However, a recent study in mice lumbar DRG showed that only 0.6% of DRG neurons were PV+, a marker of proprioceptive primary afferents (80), and our recent histological data suggests that only 3.6% of axons in human lumbar DRG have diameters greater than or equal to 12 m, an approximate lower limit of A-axon diameters (54,81,82). Furthermore, human lower leg muscles contain on the order of hundreds of muscle spindles (83), while there are tens of thousands of neurons in human DRG (54). These data suggest that A-neurons may sparsely populate the DRG, and therefore our model may be overestimating the amount of Aα-neuron activation during clinical DRGS. However, prolonged activation of muscle afferents could be another source of DRGS-induced discomfort, in addition to the activation of A-neurons with longer stimulus pulse widths described above.
Future studies should examine the functional organization of cells within human DRG and the corresponding implications for DRGS.
We modeled several classes of neurons important to the transmission of painful and non-painful stimuli: Aα-, A-, A-, and C-neurons. However, we ignored the potential effects of DRGS on non-neuronal DRG cells, such as satellite glial cells (SGCs), the glial cell type found in DRG. Glial cells in the central nervous system are known to play an important role in regulating both normal nociceptive pain processing and pathological chronic pain states (84,85). Furthermore, recent work has highlighted the contributions of SGCs to chronic pain, including visceral pain (86). To date, there are few studies examining the influence of clinical neurostimulation therapies on glial activity, though the effect of SCS-induced electric fields on glial cell function is a growing area of study (87)(88)(89). Glial cells express voltage-gated ion channels, the molecular targets by which neurostimulation therapies influence neural activity (90). Though SGCs do not contain voltage-gated sodium channels, the channel chiefly responsible for the generation of propagating action potentials, SGCs do express voltage-gated potassium channels (91). The extent to which DRGS affects voltage-gated channels in SGCs is unclear, though it is possible that DRGS could indirectly induce a myriad of potassium-mediated intracellular signaling cascades within SGCs. This notion is supported by recent studies highlighting the importance of glial cells in the development and maintenance of chronic pain at multiple levels of the nervous system, such as Schwann cells in the periphery (92), SGCs themselves in the DRG (93), and microglia in the spinal dorsal horn (94).
Finally, we examined the direct neural response to DRGS, i.e. which neurons are directly activated by one DRGS pulse or a short train of DRGS pulses. Clinically, DRGS is applied tonically, ideally over the period of years. In our multi-compartment neuron models, we included models of voltage-gated sodium and potassium channels, but ignored the contributions of other types of ion channels, such as calcium channels.
We believe that this simplification is justified for examining the direct neural response to DRGS, as voltage gated sodium and potassium channels are predominantly responsible for the generation of action potentials (95). However, due to this simplification, we were unable to study the long-term effects of tonic DRGS, as these channels activate and inactivate on time scales ranging from a few to tens of milliseconds. Furthermore, computational demands are prohibitive in simulating more than a second of neural activity. To complete the picture of the effect of DRGS on primary afferents, future experimental studies should study the effect of longterm DRGS on physiological processes mediated by ion channels that operate on long timescales, and on how tonic DRGS may modulate cause up-or down-regulation of different genes (96).
DRGS is a valuable clinical tool for managing intractable focal pain. Currently, we do not understand the physiological mechanisms of action of DRGS, nor how the clinical implementation of DRGS (e.g. lead placement, stimulator programming) affects the utilization of these mechanisms. In this work, we studied how clinically-controllable parameters affect neural activation during DRGS. Firstly, our data support the hypothesis that DRGS provides pain relief by directly activating A-LTMRs, leading to postsynaptic activation of paingating mechanisms in the dorsal horn and possibly pain-gating mechanisms within the DRG itself. Based on this hypothesis, our data corroborate several NACC recommendations: 1) straddling the active and return contacts across the pedicle (and presumably, the ganglion) may be the optimal electrode positioning in the foramen, 2) shorter pulse widths are preferred, based on maximizing activation of innocuous neurons while minimizing activation of potentially nociceptive neurons, and 3) conventional bipolar stimulation is sufficient to achieve analgesia, without the need for more complex programming configurations. Although the data presented here are pivotal to understanding the direct neural response to DRGS, future experimental and clinical studies are necessary to understand the downstream mechanisms of DRGS and how such effects influence long-term success with the therapy. Tables Table 1: Dimensions of the finite element model of a human L5 DRG.