Four food-restricted, male Lewis rats (Rats A, B, C, D; Charles River Laboratories, Wilmington, MA, United States) were studied on a behavioral task to measure current intensity detection thresholds for various MSE sector-activations of the sciatic nerve. The rats were trained with auditory stimuli prior to MSE implantation. They next proceeded to surgical implantation of an MSE in the right sciatic nerve and construction of a head cap with an embedded connector (Omnetics Connector Corporation, Minneapolis, MN, United States) for external interfacing. After 8–10 weeks of healing, the rats resumed training and transitioned to electrical stimuli under the method of constant stimuli (MCS), which applied stimulus intensities in random order from a pre-defined list without replacement. MCS was applied for nine different monopolar stimulus configurations in which the MSE stimulated the nerve and titanium screws embedded in the skull provided a return path for current stimuli. The first configuration passed equal currents through all eight channels simultaneously (i.e., multi-channel experiment), whereas the remaining eight configurations passed current through each channel individually (i.e., single-channel experiments). A flowchart representing the described protocol is presented in Figure 2. The data generated by this protocol enabled the derivation of psychometric curves depicting the probability of correct stimulus detections as a function of current intensity, and subsequent derivation of detection thresholds. All experimental procedures were conducted in accordance with regulations specified by the Institutional Animal Care and Use Committee at Washington University School of Medicine in St. Louis.

The experimental apparatus had four parts–a behavioral module, an electrophysiological module, a custom-built voltage converter for intermodular communications, and a commutator assembly (Figure 3A).

A latched male connector (Omnetics Connector Corporation, Minneapolis, MN, United States; #A76854-001) was modified as follows. Wires for channels 1–6 were removed. Wires for channels 7–10 were cut to 2.5″ and stripped of all insulation. Wires for channels 11–18 were cut to 8.75″ and soldered at 290°C to the MSE’s eight through-holes using a lead-free alloy (96.5% Sn, 3% Ag, 0.5% Cu) and a no-clean flux (Chip Quick, Inc., Ancaster, ON, Canada; #SMD291). A pair of 4-mm silicone guidance conduits with inner diameter 2 mm (A-M Systems, Sequim, WA, United States; #808500) were affixed on either side of the MSE’s middle polyimide ring using a small quantity of medical grade silicone adhesive (Factor II, Lakeside, AZ, United States; #A-564). After curing for 24 h, additional Factor II adhesive was used to insulate the solder joints and impart mechanical support to the conduit/polyimide interface. After curing, this last step was repeated as needed until complete insulation was achieved. Finally, channel impedances were measured in vitro at 1 and 5 kHz using an Autolab PGSTAT128N potentiostat (Metrohm Autolab, Utrecht, Netherlands).

Rats underwent surgery to implant the MSE in the right sciatic nerve and mount the attached connector atop the skull in a dental acrylic head cap. All surgical instruments and implants were sterilized by ethylene oxide or autoclave prior to surgery. Rats were anesthetized with isoflurane (administered IH, 4% induction, 2% maintenance) and injected with an analgesic for post-operative pain mediation (buprenorphine SR, administered SC, 1.2 mg/kg). The right hind leg, back, and scalp were shaved and sterilized using isopropyl alcohol and Betadine solution. Artificial tears applied with a cotton-tipped swab protected the eyes during surgery. The rats were head-fixed in a stereotaxic frame with a fitted nose cone to maintain anesthesia. The scalp was incised sagittally along the midline and cleaned of blood and soft tissue using a cotton-tipped swab dipped in hydrogen peroxide. Persistent bleeds were cauterized. Four to six holes were drilled into the skull using a #56-microdrill set in a hand-driven pin vise. Drilling stopped upon encountering resistance from the bone’s underlying cancellous layer. A 0–80 titanium hex screw was inserted into each hole to ensure that the head cap would remain firmly anchored to the skull.

Next, the sciatic nerve was exposed via a dorsolateral gluteal muscle-splitting incision, and blunt dissection. Nerve transection 5 mm proximal to the sciatic trifurcation preceded suturing of nerve stumps into the MSE assembly’s silicone guidance conduits using #8 microsuture (Figure 5). Separation of skin from fascia with a pair of blunt-tipped forceps created a subcutaneous dorsal tunnel between the leg and scalp incisions, through which the connector (with trailing microwires) was passed up to the skull. The connector was wrapped in a protective layer of Parafilm (Bemis Company, Inc., Neenah, WI, United States) prior to sterilization to prevent contamination of its electrical contacts during this step. This was removed once the connector was positioned over the skull. Exposed wires for channels 7–10 were wrapped around nearby skull screws; these would provide a return path for applied current stimuli. The connector was placed within a custom cylindrical titanium chamber whose threaded outer walls allowed the placement of a protective Delrin plastic (DuPont, Wilmington, DE, United States) screw cap when the connector was not in use. Care was taken to ensure that the connector’s latching mechanism rested well above the chamber’s rim. The connector, titanium chamber, and wrapped wires were fixed in place by building a head cap with two UV-cured dental acrylics (this technique was based on Park et al., 2016). A base layer of self-bonding acrylic (Fusio; Henry Schein Inc., Melville, NY, United States) was first applied to all areas of exposed skull. After setting for 30 s, this was cured with a dental UV lamp for a further 30 s. The body of the head cap was then built up little by little using a second UV-cured dental acrylic that was not self-adhering and therefore less expensive (Flow-it; Henry Schein Inc., Melville, NY, United States). Each small application of Flow-it was cured separately for 30 s using the dental UV lamp.

With MSE implantation and head cap construction completed, all muscle and skin incisions were closed with 5-0 vicryl and 4-0 nylon suture, respectively. Rats were released from the stereotaxic frame and returned to their cages for post-operative monitoring. Cages were placed partially on top of a heating pad to create a temperature gradient for the rats’ comfort. Post-operative monitoring continued until suture removal at 7–10 days post-surgery.

Rats resumed training with auditory stimuli 8–10 weeks post-surgery, progressing from interval training (2–5 days, 1–2 h each day) to withdrawal training (1–2 days, 1–2 h each day) and then reestablishment of criterion performance on the main auditory detection task (3–5 days, 1–2 h each day). They were next trained with combined auditory and electrical stimuli presented synchronously. Commencement of each experimental day with electrical stimulation was preceded by in vivo impedance measurements using the Synapse software suite (Tucker-Davis Technologies, Alachua, FL, United States). The electrical stimuli comprised equal currents passed through all eight MSE channels (i.e., multi-channel configuration). The IZ2H stimulus isolator was programmed to deliver 500 ms, 50 Hz biphasic pulse trains in which the leading and lagging phases each lasted 200 μs with no interphase gap. The stimulus intensity was of a fixed amplitude, chosen to elicit a visible twitch response in the leg and foot without any evident distress. The auditory stimuli were terminated mid-session to train responding to electrical stimuli alone. Following 2–4 days of these combined stimulus sessions in which the auditory stimuli were terminated mid-session and only electrical stimuli were presented in the latter part of the session, the rats progressed to sessions in which they were trained solely with fixed-intensity current stimuli, without any auditory stimuli, until they achieved criterion performance. Finally, the rats were trained with electrical stimuli whose amplitudes varied under MCS. Rats A, B, C, and D all performed MCS with the multi-channel stimulus configuration for up to 9 days of data collection, lasting up to 3 h per day. Only Rats C and D performed MCS with the single-channel stimulus configurations. For Rat C, data collection for each channel lasted 1–2 days with an average daily duration of 1¾ h (maximum duration: 3 h); one round of data collection was performed. For Rat D, data collection for each channel lasted 1–4 days. The average duration of data collection per day was roughly 3½ h (maximum duration: 6½ h). Two rounds of data collection were performed.

Psychometric curves for the multi-channel stimulus configuration were generated for Rats A, B, C, and D (see Figure 6 and Table 2). Rat A was stimulated at 21 unique amplitudes over nine sessions spanning 25 days (114–138 days post-implantation). The number of trials per amplitude varied from 10 to 106 trials. For Rat B, 14 unique stimulus amplitudes were used (10–59 trials). Data was gathered in five sessions spread over 8 days (119–126 days post-implantation). Rat C’s data collection occurred over eight sessions spanning 21 days (82–102 days post-implantation). The rat was stimulated at 13 unique amplitudes (18–94 trials). Rat D was stimulated at just eight unique amplitudes, with little variation in the number of trials per amplitude (98–103 trials). Data collection for Rat D took place over six sessions spanning 5 days (76–80 days post-implantation). For this stimulus configuration, the average threshold current per channel necessary to elicit a behavioral response was 19.37 μA (3.87 nC), and the average slope was 9.30 μA^–1 (46.50 nC^–1). Bootstrapped estimates of the thresholds’ CI₉₅ widths ranged from 0.87 to 4.34 μA (0.17–0.87 nC). For slopes, CI₉₅ widths for Rats A, B, and C were 3.60, 2.43, and 7.16 μA^–1 (18.03, 12.13, and 35.79 nC^–1), respectively. Rat D’s CI₉₅ for slope was much larger, spanning 88.57 μA^–1 (442.84 nC^–1).

Psychometric curves for single-channel stimulus configurations were generated for Rats C and D only. Single-channel stimulation of Rat C yielded seven psychometric curves, one each for channels C1, C2, C4, P1, P2, P3, and P4 (see Figure 7 and Table 3). Stimulation by C3 up to 200 μA produced no visible leg movement or behavioral response, precluding the generation of a psychometric curve for this channel. The remaining channels each stimulated the rat at six unique stimulus amplitudes, with the exception of P3, which used eight amplitudes. The distribution of trials among amplitudes was roughly uniform for each channel (approximately 30–60 trials per amplitude), although the dataset for C2 included 281 maintenance trials at 30 μA. Detection thresholds calculated for the three functioning core channels (C1, C2, and C4) ranged from 22.32 to 96.76 μA (4.64 to 19.35 nC; average value 56.17 μA or 11.21 nC), with CI₉₅ widths between 3.95 and 23.53 μA (0.79 and 4.71 nC). Corresponding slopes ranged from 1.35 to 7.27 μA^–1 (6.73–36.37 nC^–1; average value 4.88 μA^–1 or 24.39 nC^–1), with CI₉₅ widths from 3.26 to 72.17 μA^–1 (16.28–431.40 nC^–1). For the peripheral channels (P1, P2, P3, and P4), detection thresholds ranged from 90.18 to 144.20 μA (18.04–28.84 nC; average value 107.35 μA or 21.47 nC), with CI₉₅ widths between 11.26 and 25.09 μA (2.25 and 5.02 nC). The corresponding slopes varied from 1.94 to 5.07 μA^–1 (9.70–25.37 nC^–1; average value 3.46 μA^–1 or 17.29 nC^–1), with CI₉₅ widths between 2.31 and 29.15 μA^–1 (11.56–145.75 nC^–1).

Rat D underwent two rounds of data collection for single-channel stimuli. The first round resulted in the generation of seven psychometric curves for channels C1, C2, C3, C4, P1, P2, and P3 (see Figure 8 and Table 4). Stimulation by P4 produced no muscle twitch or behavioral response up to 200 μA, preventing psychometric curve generation for this channel. In the first round, the core channels (C1, C2, C3, and C4) had detection thresholds ranging from 49.05 to 57.75 μA (9.81–11.55 nC; average value 54.00 μA or 10.80 nC), with CI₉₅ widths between 1.58 and 3.28 μA (0.32 and 0.66 nC). Corresponding slopes were between 3.29 and 7.82 μA^–1 (16.44 and 39.10 nC^–1; average value 6.04 μA^–1 or 30.18 nC^–1), with CI₉₅ widths from 1.26 to 4.35 μA^–1 (6.29–21.77 nC^–1). The 3 functioning peripheral channels (P1, P2, and P3) had detection thresholds between 81.47 and 138.15 μA (16.29 and 27.63 nC; average value 115.20 μA or 23.04 nC), with CI₉₅ widths from 3.72 to 14.53 μA (0.74–2.91 nC). The corresponding slopes ranged from 0.92 to 4.23 μA^–1 (4.59–21.14 nC^–1; average value 2.61 μA^–1 or 13.07 nC^–1), with CI₉₅ widths between 0.29 and 2.40 μA^–1 (1.43 and 11.99 nC^–1).

Data collection for the second round occurred approximately 3 weeks after the first and yielded a second set of psychometric curves for channels C1, C2, C3, C4, and P1. Channels P2 and P3 were excluded because the rat’s waning performance lowered the SR (see Eq. 1) and prevented the accumulation of sufficient trials for analysis. Detection thresholds for C1 and C2 decreased by 20 and 36%, respectively. While C3’s threshold increased marginally by 7%, thresholds for C4 and P1 increased markedly by 70 and 75%, respectively. The longitudinal development of thresholds and slopes is presented in Figure 9.

Mann–Whitney U tests comparing core and peripheral channels’ detection thresholds and slopes yielded p-values of 0.0008 and 0.02, respectively. This indicated that the two groups differed significantly on both parameters.

Under multi-channel stimulation, Rats A, B, and D had similar detection thresholds of 21–22 μA per channel (corresponding to 4.2–4.4 nC of charge injection per channel, or a charge density of 157–163 nC/mm²). Rat C had a considerably lower threshold of 12 μA (2.4 nC, 91 nC/mm²). Notably, the threshold currents injected per channel for Rat C (12 μA) and Rat D (22 μA) fell well below their corresponding lowest single-channel thresholds of 22 μA (channel C2) and 49 μA (channel C3, first round), respectively. This can be explained by assuming that the axons activated by these single channels must have been the first to be activated under multi-channel stimulation. Since axon activation depends on local current density, the addition of equal currents from seven other channels likely lowered the per-channel current requirement to achieve the current density required for activation.

For single-channel stimulation, the average detection threshold for the core channels was approximately half that of the peripheral channels (Mann–Whitney U test, p < 0.0008); this relationship held true for both Rat C (56 μA vs. 107 μA, or 11.2 nC vs. 21.5 nC) and Rat D (56 μA vs. 115 μA, or 11.2 nC vs. 23.0 nC). In terms of current density, the discrepancy was even more pronounced: 351 nC/mm² vs. 954 nC/mm² for Rat C; 348 nC/mm² vs. 1,085 nC/mm² for Rat D. The divergence of core and peripheral channel thresholds may have had multiple causes.

Following transection, peripheral nerve regeneration commences with the formation of a bridge of dense extracellular matrix and inflammatory cells between the proximal and distal stumps. Vascularization of this intervening tissue creates a pathway for migrating Schwann cells to tow proximal axons toward distal targets (Cattin and Lloyd, 2016). Suzuki et al. (1998) has shown that regeneration of sensory axons precedes that of motor axons immediately after axotomy. Accordingly, there is an increased proliferation of sensory axons toward the nerve’s center and of motor axons toward the periphery (Negredo et al., 2004; Lago et al., 2005). This differential proliferation implies that the distance from a core channel to an average regenerated sensory axon should be less than from a peripheral channel. Previous simulations by our group (Zellmer et al., 2018) have predicted that regenerated axons’ thresholds for activation are not inherently higher or lower than those of undisrupted axons, but rather depend on proximity to the stimulating lead. Effectively, thresholds for nearby regenerated axons should be lower than for naïve axons of the same caliber, while thresholds for regenerated axons that are farther away should be higher than for their naïve counterparts. Thus, the higher density of regenerated axons at the nerve’s center, coupled with the pronounced dependence of regenerated axons’ recruitment on lead proximity, may have contributed to the observed discrepancy between core and peripheral channels’ detection thresholds.

Lead geometry may also have played a role. The core channels (area: 32,000 μm² each) are curved, while the peripheral channels (area: 22,500 μm² each) are straight. The core channels’ greater areas means that their current density for a given current level is lower than the corresponding current density for the peripheral channels. Generally, higher current density is associated with increased axon recruitment. Concurrently, the core channels’ curved geometry means that their “centers of mass” lie closer to centrally located axons than if their geometry was straight. Thus, the core channels may have had an outsized effect on central axons despite their larger areas, pushing their thresholds down. This effect would be magnified by the concentration of sensory axons toward the nerve’s center.

The peripheral channels’ radial placement is also of concern. These channels extend to a radius of 850 μm from the active region’s center, far exceeding the 1-mm diameter of regenerated rat sciatic nerve (MacEwan et al., 2016), and even the 1.4-mm diameter of undisrupted nerve (Tyler and Durand, 2003). This suggests that some fraction of the peripheral channels’ currents was injected into the extraneural space. However, the enclosure of the active region within a pair of 4-mm silicone guidance conduits would have ensured that any extraneural current must have flowed along the nerve’s periphery, contributing to axon recruitment. It remains unclear to what extent the peripheral channels extruded from the regenerated nerve, and how much their extrusion affected the detection threshold.

Longitudinal examination of Rat D’s detection thresholds using single-channel stimulation revealed disparate trends. Five channels (C1, C2, C3, C4, and P1) were measured at two timepoints 3 weeks apart. The decline in C1’s and C2’s thresholds may have been caused by the ongoing maturation of nearby regenerated axons (increased calibers, thicker myelin sheaths, and fewer unmyelinated axons), which is known to continue for up to 7 months post-implantation (Ceballos et al., 2002). Channel C3’s threshold was stable, showing a minor increase between the two rounds. The dramatic rise in C4 and P1’s thresholds coincided with increased channel impedances. For C4, the first round of data was gathered 100–101 days post-implantation. Corresponding impedances at 1 kHz ranged between 62.28 and 94.45 kΩ. From the 119th day onward, C4’s impedance became erratic. Measured values on this day ranged between 26.74 and 397.08 kΩ. The second round of data was gathered on the 121st day. Although the impedances measured on this day were lower than 2 days prior, on subsequent days (up to 127 days post-implantation) some measurements exceeded 1 MΩ. For P1, the first and second rounds of data were gathered 102–103 and 126–127 days post-implantation, respectively. The corresponding impedances for these two rounds ranged between 86.16 and 166.66 kΩ, and between 161.34 and 771.69 kΩ, respectively. Impedances for P2 and P3 also rose drastically after their respective first rounds of data collection, which may explain why the rat’s performance degraded during the attempted second round of data collection for these channels. Such sudden degradation of impedance suggests failure of the lead, the solder joint, the associated microwire, or damage to the skull-mounted connector. It remains unclear where the failure occurred.

The psychometric slope ψ(xT;θ)′ signifies the rapidity with which the percentagewise probability of stimulus detection rises with stimulus intensity. Measured slopes for multi-channel stimulation ranged from 3.76 to 15.71 μA^–1 per channel, and for single-channel stimulation from 0.92 to 12.29 μA^–1. A steep slope implies that activation of nearby axons occurs with sufficient reliability that the transition from low to high detection probability occurs over a short span of increasing current; a shallow slope implies the opposite. For both Rats C and D, single-channel stimulation yielded steeper average slopes for the core channels than the peripheral channels (Mann–Whitney U test, p < 0.02). This may reflect the differential proliferation of sensory axons described by Suzuki et al. (1998). Since core channels reside in a region of high axon density, small current increments should significantly increase the number of axons recruited and hence the probability of a behavioral response. Conversely, the lower density of axons surrounding the peripheral channels means that small current increments should recruit fewer additional axons, producing little change in the behavioral response probability.

No clear longitudinal trends for Rat D’s slope values were discerned. Five channels (C1, C2, C3, C4, and P1) were measured across two timepoints. Two channels’ slopes became markedly steeper (C1 and C3 by 57 and 203%, respectively). One channel’s slope increased slightly (P1 by 7%). The remaining two channels’ slopes became shallower (C2 and C4 by 21 and 70%, respectively). There was no consistent relationship between changes in threshold and changes in slope.

Wichmann and Hill (2001b) stress that bootstrapped confidence intervals do not measure a parameter’s underlying variability, but rather the variability inherent in the sampling scheme (i.e., the selection of current amplitudes), the number of trials for each amplitude, and interactions between the sampling scheme and parameter calculation. In the present study, the relation between choice of sampling scheme and CI₉₅ widths was readily apparent, as datasets whose sampling schemes placed fewer amplitudes in the sloped domain of the psychometric curve showed greater variation in bootstrapped parameter values and hence wider confidence intervals. This effect was more pronounced for slopes than thresholds. Bootstrapped slope values for Rat D under multi-channel stimulation showed considerable variation due to the placement of only one amplitude (24 μA) in the psychometric curve’s sloped domain (see Figures 6D,F). Similarly, bootstrapped slopes for Rat C under single-channel stimulation also varied considerably with the exception of channel C2, for which multiple amplitudes lay in the sloped domain (see Figures 7B,I). Bootstrapped thresholds also showed a degree of variation for some channels. In contrast, Rat D’s threshold and slope confidence intervals for single-channel stimulation were more constrained (see Figure 8), reflecting the adoption of a pseudo-adaptive strategy that evaluated the psychometric curve at multiple mid-session timepoints to determine whether the sampling scheme’s placement of amplitudes located them optimally in the curve’s sloped domain, so that more amplitudes could be added as necessary.

Future experiments will generate high-resolution maps of detection thresholds across the nerve using multipolar stimulus configurations, in which the coordinated application of cathodic and anodic currents across multiple MSE channels will restrict recruitment to spatially distinct axon clusters (i.e., current steering). Likewise, given the likely non-uniform distribution of regenerated sensory axons across the implant cross-section (i.e., higher density in the central region), the MSE’s transit zones could be redesigned to be smaller and more numerous in the central region while remaining larger toward the periphery. This might enable more even recruitment of sensory axons by matching recruitment slopes across the cross-section. Additional simulation work is also needed to understand the interplay of non-uniform axon distribution, lead geometry, and peripheral channel extrusion on axon recruitment.

The use of adaptive algorithms in place of MCS will reduce the number of trials required for reliable threshold estimation, enabling a more robust longitudinal analysis of threshold stability. Additionally, the present go/no-go detection task can be adapted to a 2-alternative forced-choice paradigm to assess whether such selective axon cluster recruitment elicits sensory percepts whose perceived locations in the phantom limb are discriminable.

The methods presented in this paper are not restricted to the MSE but can be harnessed for the investigation of other electrodes as well. For example, the micro-channel sieve electrode (MCSE) is an RE design with extruded transit zones that form electrically isolated “micro-channels” (FitzGerald et al., 2008; Lacour et al., 2008). Recent simulations by our group have shown that the MCSE is well-suited for bidirectional interfacing of peripheral nerve tissue, as it allows for the simultaneous stimulation of sensory axons and recording of motor axons without stimulus artifact (Coker et al., 2019a,b). The combined rat sciatic nerve and behavioral model could provide an ideal platform for evaluating the MCSE’s bidirectional capabilities in an in vivo setting.