Coupling multielectrode array recordings with silver labeling of recording sites to study cervical spinal network connectivity

We describe a method that reliably identifies the locations of multielectrode array (MEA) recording sites while preserving the surrounding tissue for immunohistochemistry. To our knowledge, this is the first cost-effective method to identify the anatomic locations of neuronal ensembles recorded with a MEA during acute preparations without the requirement of specialized array electrodes. In addition, evaluation of activity recorded from silver-labeled sites revealed a previously unappreciated degree of connectivity between midcervical interneurons.

Abstract

Midcervical spinal interneurons form a complex and diffuse network and may be involved in modulating phrenic motor output. The intent of the current work was to enable a better understanding of midcervical “network-level” connectivity by pairing the neurophysiological multielectrode array (MEA) data with histological verification of the recording locations. We first developed a method to deliver 100-nA currents to electroplate silver onto and subsequently deposit silver from electrode tips after obtaining midcervical (C3–C5) recordings using an MEA in anesthetized and ventilated adult rats. Spinal tissue was then fixed, harvested, and histologically processed to “develop” the deposited silver. Histological studies verified that the silver deposition method discretely labeled (50-μm resolution) spinal recording locations between laminae IV and X in cervical segments C3-C5. Using correlative techniques, we next tested the hypothesis that midcervical neuronal discharge patterns are temporally linked. Cross-correlation histograms produced few positive peaks (5.3%) in the range of 0–0.4 ms, but 21.4% of neuronal pairs had correlogram peaks with a lag of ≥0.6 ms. These results are consistent with synchronous discharge involving mono- and polysynaptic connections among midcervical neurons. We conclude that there is a high degree of synaptic connectivity in the midcervical spinal cord and that the silver-labeling method can reliably mark metal electrode recording sites and “map” interneuron populations, thereby providing a low-cost and effective tool for use in MEA experiments. We suggest that this method will be useful for further exploration of midcervical network connectivity.

NEW & NOTEWORTHY We describe a method that reliably identifies the locations of multielectrode array (MEA) recording sites while preserving the surrounding tissue for immunohistochemistry. To our knowledge, this is the first cost-effective method to identify the anatomic locations of neuronal ensembles recorded with a MEA during acute preparations without the requirement of specialized array electrodes. In addition, evaluation of activity recorded from silver-labeled sites revealed a previously unappreciated degree of connectivity between midcervical interneurons.

midcervical spinal interneurons form a complex and diffuse network that is synaptically coupled to both respiratory and nonrespiratory motor pools (Gonzalez-Rothi et al. 2015; Lane 2011; Lane et al. 2008b). Several groups have advanced the hypothesis that midcervical spinal interneurons can modulate phrenic motoneuron excitability and thereby influence the neural control of the diaphragm (Bellingham and Lipski 1990; Douse and Duffin 1993; Lane et al. 2008a,b; Palisses et al. 1989). Although there is some direct evidence to support this hypothesis (Marchenko et al. 2015; Sandhu et al. 2015), other studies have concluded the opposite (Duffin and Iscoe 1996). A significant hurdle in testing that specific hypothesis, or related hypotheses regarding cervical interneuronal circuits, is the difficulty of studying the “functional connectivity” in diffuse spinal cord networks. One of the foremost challenges is simultaneously recording numerous cells, and this can be addressed through the use of multielectrode arrays (MEAs). The MEA approach enables simultaneous recordings of multiple sites, but histologically identifying each recording location (vs. the electrode track) while also preserving tissue integrity poses a further challenge (Borg et al. 2015; Li et al. 2015; Nuding et al. 2015). Thus the initial thrust of the current work was modification and validation of a silver-labeling technique (Spinelli 1975) to enable postrecording deposition of a small amount of silver (i.e., for histological marking) from the tip of each electrode in an MEA. Additionally, we developed an electrical circuit to enable the use of small currents (100 nA) for silver electroplating and deposition to prevent tissue and electrode damage associated with high levels of current (Fung et al. 1998). Using the electroplated MEA, we recorded discharge from ensembles of neurons in the midcervical (C3–C5) spinal cord in adult rats and demonstrated a practical application of this technique by “matching” the array electrodes to the corresponding anatomic locations marked by silver.

The overall intent was to enable a better understanding of midcervical spinal discharge and “network-level” connectivity by pairing the neurophysiological MEA data with histological verification of the recording locations. Thus, using correlative techniques (Moore et al. 1970), we tested the hypothesis that the discharge patterns of midcervical (C3–C4) spinal neurons are temporally linked in time domains consistent with mono- and polysynaptic connections. In addition, midcervical neuronal discharge patterns were assessed relative to the endogenous inspiratory pattern, measured via bilateral phrenic nerve recordings, to determine whether bursting was temporally linked to phrenic motoneuron activity. To our knowledge, no prior study has comprehensively evaluated the temporal characteristics across multiple midcervical neurons using MEA technology. The results presented herein demonstrate a previously unappreciated degree of connectivity and indicate a high prevalence of temporally related discharge patterns between midcervical interneurons with characteristics consistent with mono- and polysynaptic connections and provide a comprehensive description of a cost-effective histological approach for validating anatomic locations of MEA recording sites.

MATERIALS AND METHODS

Animals.

All experiments were conducted with adult Sprague-Dawley rats obtained from Envigo (formerly Harlan Laboratories). Most experiments (n = 12) were performed with untreated, spinal-intact rats. A subset of rats (n = 2) received a cervical spinal injury (C3/C4 lateralized contusion, force: 205 kDa, displacement: 1,225 μm; Infinite Horizon pneumatic impactor; Precision Systems and Instrumentation) using published methods from our group (Lane et al. 2012). Spinal-injured rats were allowed to recover for 12 wk before electrophysiology. Histological results from spinal-injured animals were used to determine effectiveness of silver labeling and to compare micromotor depths relative to silver-labeling depths. Since there is evidence suggesting neurophysiological properties such as connectivity may be altered following spinal cord injury (SCI; Lane et al. 2008b, 2009), analysis of neurophysiology was limited to spinal-intact animals (n = 4). All rats were housed in pairs in a controlled environment (12:12-h light-dark cycles) with food and water ad libitum. All experimental protocols were approved by the Institutional Animal Care and Use Committee at the University of Florida.

In vivo electrophysiology.

All rats were anesthetized with 3% isoflurane (in 100% O₂) and transferred to a heated station where established surgical methods (Lee and Fuller 2010; Lee et al. 2009; Mahamed et al. 2011; Sandhu et al. 2015; Streeter and Baker-Herman 2014a,b; Strey et al. 2012) were used to set up electrophysiological experiments. Core body temperature was maintained at 37 ± 0.5°C with a servo-controlled heating device (model 700 TC-1000; CWE). A tail vein catheter was placed for in vivo delivery of urethane anesthesia and fluids. The trachea was cannulated (PE 240 tubing), and rats were pump-ventilated (Rodent Ventilator 683; Harvard Apparatus; volume: approximately 3-2.5 ml; frequency: 70/min). Once ventilated, CO₂ was added [fraction of inspired carbon dioxide (Fi_CO2): <3%] to maintain end-tidal CO₂ (EtCO₂) between approximately 40 and 50 mmHg throughout the protocol (Capnogard; Respironics). Tracheal pressure was monitored and lungs were periodically hyperinflated (2–3 breaths) to prevent atelectasis (∼1/h). A bilateral vagotomy was performed. Rats were slowly converted (6 ml/h; Harvard Apparatus syringe pump) to urethane anesthesia (1.7 g/kg in vivo in distilled water), and isoflurane was withdrawn. A femoral arterial catheter (PE 50 tubing) was placed to monitor blood pressure (TA-100; CWE) and sample blood gases (i-STAT 1 Analyzer; Abbott) throughout the protocol. Using a dorsal approach, the left and right phrenic nerves were isolated, cut distally, and partially desheathed (∼1/2 the length of the exposed nerve). A midline incision extending from the base of the skull to midthoracic region was made, and spinal vertebrae C3-T2 were exposed. Using a nose clamp and T2 spinous process, the rat was slightly elevated off the table to reduce ventilator-induced motion artifact and level the spinal cord. A laminectomy was performed from C3 to C6, the dura was cut and reflected back, and the pia was gently removed at the MEA insertion point. A unilateral pneumothorax was performed to decrease chest wall movement, and positive end-expiratory pressure (PEEP) of approximately 1–2 cm of water was applied to prevent atelectasis. Animals received the neuromuscular paralytic pancuronium bromide (2.5 mg/kg in vivo; Hospira) to eliminate spontaneous breathing efforts. Adequate depth of anesthesia was monitored by assessing blood pressure responses to toe pinch, and urethane supplements were given as necessary. Blood pressure and fluid homeostasis were maintained by a slow infusion of a 1:3 solution (8.4% sodium bicarbonate/Ringer lactate in vivo).

Bilateral phrenic nerve output was recorded using custom-made bipolar suction electrodes filled with 0.9% saline. Compound action potentials were amplified (20,000×; P511; Grass Instrument), band-pass filtered (3 Hz to 3 kHz), digitized [16-bit, 25,000 samples per second per channel; Power1401; Cambridge Electronic Design (CED), Cambridge, UK], and integrated (time constant: 20 ms) in Spike2 v8 software (CED). A custom-made MEA similar to that originally designed at the University of South Florida by Morris and colleagues (1996) and used in our previous report (Sandhu et al. 2015) was used to record bilateral midcervical (C3–C5) spinal activity (Fig. 1A). The array contained 16 tungsten electrodes coated with Epoxylite insulation (impedance: 10 ± 1 MΩ; shank diameter: 125 μm; tip diameter: ≤1 μm; cat. no. UEWLEGSE0N1E; FHC; Fig. 1B). A key feature of this array is the ability to control the depth of each of the electrodes independently using micromotors. This greatly improves the recording yield by allowing the experimenter to “hold” a recording on 1 electrode while continuing to search for neurons with other electrodes. The array was mounted on a stereotaxic frame, and 8 electrodes arranged in 2 staggered rows of 4 were placed into each hemicord at the dorsal root entry zone (Fig. 1C). The inner distance between the 2 sets of 8 electrodes was ∼1 mm, whereas the distance between electrodes within each row was ∼300 μm (Fig. 1D). Electrodes tips were maintained in this “fixed matrix” by the array guide. One by one, electrodes were advanced into the spinal cord while audio was monitored until single units with ∼3:1 signal-to-noise ratio were discriminated. Neural signals from single electrodes were amplified (5,000×), band-pass filtered (3-3 kHz), digitized (16-bit, 25,000 samples per second per channel; Power1401; CED), and recorded with Spike2 v8 software (CED). Once phrenic nerve activity and spinal discharge was stable, “baseline activity” was recorded with fraction of inspired oxygen (Fi_O2) set to 0.50. Using experimental procedures similar to our previous studies (Sandhu et al. 2015; Strey et al. 2012), the inspiratory gas mixture was altered using adjustable flowmeters to expose rats to a 5-min period of hypoxia (Fi_O2: 0.11) followed by 15 min of baseline oxygen levels (Fi_O2: 0.50). Silver deposition was performed immediately after the neurophysiology protocol. An arterial blood gas was sampled during baseline and hypoxia to ensure blood gases were within physiological limits.

Fig. 1.

Multielectrode array. A: multielectrode recording array containing 16 tungsten fine wire electrodes each controlled by micromotors and held in place by the array guide. B: high-resolution image of the electrode tips maintained in a “fixed matrix” by the array guide. C: image of the electrodes positioned in the dorsal C4 spinal cord at the dorsal root entry zone (e.g., black arrows). D: schematic of the recording matrix consisting of 8 electrodes arranged in 2 staggered rows of 4. The inner distance between the 2 sets of 8 electrodes was ∼1 mm, and the distance between electrodes within each row was ∼300 μm.

Analysis of electrophysiological signals.

All electrophysiological data were collected and sorted using Spike2 v8 software (CED). Extracellular action potentials from individual neurons were extracted from continuous recordings and converted to waveforms using spike-sorting tools of the acquisition software. Briefly, spikes were clustered using at least 80% template matching and principal component analyses. Sorted spikes were exported and analyzed using custom MATLAB software (MathWorks R2015a). A standard set of analyses (spike interval histograms, cycle-triggered histograms, and spike-triggered averages) were performed to phenotype each neuron electrophysiologically. Interval histograms were created for each spike train to ensure only single-unit activity was represented in each waveform. Similar to previous studies (Galán et al. 2010; Sandhu et al. 2015), cycle-triggered histograms were used to assess the preferred respiratory modulation of each recorded neuron relative to the phase of the respiratory cycle defined by phrenic nerve activity (Cohen 1968). Using the integrated phrenic nerve output, the beginning and end of the inspiratory, and therefore expiratory, phases were calculated as a departure of phrenic nerve activity ≥15 standard deviations above the average activity during the expiratory phase. Cycle-triggered histograms were constructed for each neuron by dividing the respiratory period into 20 bins of equal size, and spikes were counted within each bin and summated over 50 consecutive breaths during baseline, hypoxia, and posthypoxia. To determine whether neurons were respiratory-modulated, the spikes occurring during bins of the inspiratory phase were separated from those occurring during the expiratory phase, and the Wilcoxon signed-rank test was used to test the null hypothesis (i.e., no difference between inspiration and expiration). Cervical spinal neurons were classified into four categories: 1) neurons that discharged primarily during the inspiratory period (i.e., inspiratory-modulated; waveform 1 in Fig. 7, C and D); 2) neurons that discharged primarily during the expiratory period (i.e., expiratory-modulated; waveform 1 in Fig. 7, H and I); 3) neurons that discharged without respiratory modulation (i.e., tonic; waveform 2 in Fig. 7, C, D, H, and I); and 4) neurons that ceased firing at time points after baseline and were labeled as inactive at that time point.

Fig. 7.

Silver labeling of midcervical spinal interneurons. A: representative C4 section containing a silver-labeled site counterstained with cresyl violet and high-resolution image containing the silver-labeled site (callout). B: corresponding neuronal output and integrated phrenic output during baseline, hypoxia, and posthypoxia depicting a single tonic firing neuron at baseline and recruitment of a phasic inspiratory neuron during hypoxia and posthypoxia. C: integrated phrenic output, raw neuronal discharge, and sorted spikes (waveforms) during hypoxia. D: cycle-triggered histograms of both neurons during 50 consecutive breaths overlaid with the average integrated phrenic waveform during hypoxia. E: spike-triggered average of the raw and rectified phrenic nerve depicting a lack of positive features. F: representative C4 section containing a silver-labeled site counterstained with cresyl violet and high-resolution image containing the silver-labeled site (callout). G: corresponding neuronal output and integrated phrenic output during baseline, hypoxia, and posthypoxia depicting an expiratory and tonic firing neuron at baseline and hypoxia and only the tonic neuron posthypoxia. H: integrated phrenic output, raw neuronal discharge, and sorted spikes (waveforms) during hypoxia. I: cycle-triggered histograms of both neurons during 50 consecutive breaths overlaid with the average integrated phrenic waveform. J: spike-triggered average of the raw and rectified phrenic nerve depicting a lack of positive features. K: stacked bar graphs depicting the number of each cell type at baseline, hypoxia, and posthypoxia each normalized to the total number of interneurons (n = 28). L: anatomic location of each silver-labeled interneuron projected onto 1 side of the cord for simplicity, identified by the bursting pattern during baseline. Exp, expiratory; Insp, inspiratory; CC, central canal; DH, dorsal horn; VH, ventral horn. Scale bars: A, F, and L: 0.5 mm and 50 μm (callouts); B, C, G, and H: 1.5 s.

Spike-triggered averaging of raw and rectified phrenic nerve activity was used to examine the temporal relationship of neuronal waveforms and phrenic motor output (Lipski et al. 1983). Short-latency, offset peaks in the raw and rectified phrenic nerve average provided evidence that the recorded neuron was a phrenic motoneuron (Christakos et al. 1994). If features were not detected using spike-triggered averaging, the recorded cell was classified as a spinal interneuron. Cross-correlation histograms were constructed for all possible pairs of simultaneously recorded neurons using a bin width of 0.2 ms to evaluate functional connectivity (Moore et al. 1970). Similar to published methods (Aertsen and Gerstein 1985), the detectability index (DI) was calculated for each cycle-triggered histogram as the peak relative to average background activity (calculated during the interval −15 to −3 ms before 0) divided by the standard deviation. Only significant features occurring with a positive lag (e.g., ≥0) were counted. Features were considered significant if the DI was ≥3 (Melssen and Epping 1987). Central peaks (0–0.4 ms) between a pair of neurons supported synchronous firing between neurons, whereas offset peaks (≥0.6 ms) indicated functional excitation between the trigger and target neurons.

Plating electrodes and depositing silver.

Methods for electroplating and depositing silver described previously (Spinelli 1975) were adapted for use with MEA recordings. For each electrode, silver was electroplated (Fig. 2A) and deposited (Fig. 2B) using tightly controlled and monitored direct current (DC) circuits. The circuits consisted of an electrode selection interface (i.e., breakout box), two 9-V batteries, a custom-built μCurrent control device (Bare Electronics, Gainesville, FL), a μCurrent Precision nA Current Measurement Assistant v3 (cat. no. 882; EEVblog Adafruit Industries), and a digital multimeter (model 61-340; Ideal Industries) all connected with copper patch cables. Specifications of the μCurrent control circuit are provided in Fig. 2C. Before each experiment, 15 of the 16 fine wire tungsten electrodes were electroplated with dissolved silver cations in an electrolyte solution (AgNO₃/KCN), and an uninsulated silver wire (100 mm, 0.025-in. diameter; cat. no. 783500; A-M Systems) was used as the anode in the solution (Fig. 2A). The electrolyte solution was prepared by mixing equal parts of 1% potassium cyanide and 1% silver nitrate. The circuit was tested before each use by electroplating silver onto copper wire to ensure proper connections and current settings. Immediately following the testing procedure, the tip of each individual electrode was introduced to the electrolyte solution (as the cathode) for 100 s at 100 nA. Electrodes used to deposit silver were replated before each experiment. In a subset of experiments (n = 3), 2 of the 16 electrodes were coated with the lipophilic fluorescent dye 1,1′-dioctadecyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate [‘DiI’; DiIC18(3); Dil; cat. no. D282; Thermo Fisher Scientific] after electrodes were electroplated with silver and before electrophysiology. Each electrode was repeatedly dipped (∼10 times) in DiI (50 mg/ml in ethanol) and allowed to dry for 5 s between each dip (DiCarlo et al. 1996).

Fig. 2.

Silver electroporation and deposition circuits. Both circuits consisted of an electrode selection interface (i.e., breakout box) connected to the array by a DB9 connector, 2 9-V batteries connected in series, a custom-built μCurrent control device (constant current board), a μCurrent Precision nA Current Measurement Assistant (uCurrent Box), and a digital multimeter connected via copper patch cables (thick black lines). A: for the silver electroplating circuit, the array was connected as the cathode to the electrode selection box (breakout box) while an uninsulated silver wire was placed in the electrolyte solution (KCN/AgNO₃) and used as the anode. The μCurrent box was connected to breakout box corresponding to the electrode to be plated (Ch1-15). B: for the silver deposition circuit, the polarity of the circuit was reversed with breakout box used to select the reference channel (cathode/Ch16) and the electrode to be deposited (anode/Ch1-15). C: the μCurrent control device uses an NPN transistor (Q1 PN2222A) to regulate current (I) from the power source (2 9-V batteries; B1 and B2) through a load between J1 (i.e., connector to µCurrent box) and J2 [i.e., connector to silver wire (A) or electrode (B)]. Two small-signal diodes, D3 and D4 (1N4148), provide a fixed voltage (∼1.25 V_DC) used in parallel with voltage divider (R4). This adjustable voltage divider controls the base current (I_b) delivered to the transistor so that the emitter current of the transistor produces a stable current of 100 nA. As load between J1 and J2 changes, the transistor actively alters the voltage, and therefore current, drop across collector (C) to emitter (E), maintaining a constant voltage (across R3). Current was monitored by an EEVblog μCurrent Precision nA Current Measurement Assistant v3 with a consumer-grade multimeter and adjusted to maintain 100 nA.

Subsequent to ensemble recordings of individual neurons, electrodes were left in place while the MEA was disconnected from the acquisition system and attached to the electrode selection interface (i.e., breakout box; Fig. 2B). Sites to be labeled were chosen based on the quality of the signals during the electrophysiological recordings. The circuitry was connected as in the plating procedure with the exception that electrode 16 was used as the cathode and the individual electrode chosen for labeling was treated as the anode. Electroplated silver was released from the electrode tip using a 100-nA current for 10–100 s. Tissue was fixed approximately 15–30 min after silver was deposited (described below).

Tissue preparation.

Animals were transcardially perfused with ice-cold saline followed by 4% paraformaldehyde (PFA; cat. no. 19210; Electron Microscopy Sciences) in 1× Dulbecco phosphate-buffered saline (DPBS; cat. no. 21-030-CV; Mediatech). Spinal tissue was harvested and postfixed in 4% PFA overnight at 4°C. The cervical spinal cord was subsequently blocked (C3-C6; approximately 8–9 mm of tissue) and cryoprotected in 30% sucrose in 1× DPBS overnight at 4°C. The spinal cord was embedded in optimal cutting temperature compound (OCT; cat. no. 23-730-571; Fisher Scientific), flash-frozen using 2-methylbutane and dry ice, and transversely cut at 40 μm on a cryostat (Microm HM 500; GMI). Sections for silver labeling were placed into 12-well trays (cat. no. 3737; Corning) containing Corning Netwell inserts (2 sections per well; 15-mm diameter, mesh size of 74 μm; cat. no. 3477; Corning) filled with glass-distilled water (∼4 ml per well). Sections for immunohistochemistry were placed in 96-well plates containing 1× DPBS.

Silver staining protocol.

Silver staining was performed immediately after sectioning. Because of the sensitivity of the reaction, all trays, glassware, and instruments used for staining protocol were thoroughly acid washed (1% HCl) before staining to eliminate contamination and remove nonspecific catalysts. Plastic 12-well trays containing the developer solution were discarded after used for a single tray of tissue, whereas all other trays were discarded after each animal. Netwell mesh inserts were soaked in bleach between each animal. The deposited silver was developed as previously described (Spinelli 1975). Briefly, stock solutions were prepared as outlined in Table 1. The hydrogen peroxide solution was covered in foil throughout the entire protocol due to its light sensitivity, and the hydroquinone and citric acid solution was made in the same bottle. The developer was made fresh for each 12-well tray of tissue by mixing the 3 stock solutions (acacia, silver nitrate, and hydroquinone/citric acid) 5 min before use (at the start of step 9). All steps were carried out in a fume hood, and light exposure was limited while mixing developer solution and during “developing” by turning off the fume hood light.

Table 1.

Silver staining protocol

Dish no.	Contents	Time	Volume/Tray
1	dH2O	5 min	150 ml/dish
2	dH2O	5 min	150 ml/dish
3*	1% (NH4)2S	5 min	48 ml/tray
4*,†	0.125% H2O2	5 min	48 ml/tray
5	dH2O	5 min	150 ml/dish
6	dH2O	5 min	150 ml/dish
7*	1% ascorbic acid	5 min	48 ml/tray
8	dH2O	5 min	150 ml/dish
9	dH2O	5 min	150 ml/dish
10*,†	Developer	8 min
	25% acacia		16 ml (48 ml/tray)
	2% hydroquinone, 5% citric acid		16 ml (48 ml/tray)
	1% AgNO2		16 ml (48 ml/tray)
11	dH2O	Quick rinse	150 ml/dish
12	dH2O	Quick rinse	150 ml/dish
13	dH2O	Quick rinse	150 ml/dish
14	1% Na2S2O4·2H2O	3–4 min	150 ml/dish
15	dH2O	Quick rinse	150 ml/dish

Sequence of solutions necessary to develop histologically the deposited silver adapted from Spinelli (1975). Solutions were prepared and arranged in the order of the staining sequence in either glass petri dishes or plastic trays. Volumes of solution were calculated for 1 12-well tray of tissue (24 sections; ∼1 mm of tissue). Boldface text indicates the steps in which solutions were changed after processing 1 tray of tissue, and remaining steps indicate solutions were changed after processing 3 trays of tissue. Store in distilled water (dH₂O) at 4°C until ready to mount on charged slides, counterstain, and then cover with mounting medium.

^*Place solution in a 12-well tray.

^†Light-sensitive solutions. Change solution every 3 times except for boldface, which indicates change solutions every time.

The solutions were poured into either 200-ml glass petri dishes (cat. no. 3160152BO; Corning) or 12-well plastic trays according to the step (see Table 1) and arranged in order of the staining sequence. To maintain precision during the staining protocol, staining was performed in “rounds,” where each round contained 3 trays of tissue (∼1-mm tissue per tray). Trays were staggered by 5 min, and all 3 trays of tissue were stained from start to finish. Rounds were repeated until all tissue was stained. The Netwell mesh inserts were transported through the staining sequence using a Netwell carrier kit with handles (cat. no. 3520; Corning) to allow simultaneous processing and ensure all wells were in each solution for the appropriate time. Following staining, tissue was stored at 4°C in Netwells filled with distilled water. Sections were mounted onto charged microscope slides (cat. no. 12-550-15; Fisher Scientific) using 1× PBS and allowed to dry for 48–72 h. Sections were counterstained with cresyl violet covered with mounting medium (cat. no. 4111; Thermo Scientific) and coverslipped. Images were captured with bright-field microscopy using a Microscope Axio Imager.A2 (Carl Zeiss Microscopy). Images were stitched and white balanced using Adobe Photoshop. Anatomic locations of the silver-labeled sites within the spinal laminae were identified using the Atlas of the Spinal Cord (Sengul et al. 2012). All representative spinal cord images reflect camera lucida-style drawings of our histological images or C4 spinal section in the same atlas. The depth of silver-labeled sites were determined using 10× photomicrographs and adjusted by 10% to account for tissue contraction during the fixation and histological processing based on estimates of prior investigations (Deutsch and Hillman 1977; Quester and Schröder 1997).

Immunohistochemistry.

Initial immunohistochemistry (IHC) experiments determined the order in which IHC should be performed to achieve optimal fluorescent staining using tissue that did not contain deposited silver. IHC was performed either before or immediately after silver staining using two common markers (e.g., GFAP and NeuN). Using the optimal staining order, a subsequent experiment was performed with tissue containing deposited silver. All IHC was performed with free-floating sections in 96-well plates. All tissue was washed with 1× PBS with 0.2% Triton X-100 (3 × 5 min). For NeuN staining, tissue was incubated for 40 min at room temperature in 10% normal goat serum (NGS) in 1× PBS with 0.2% Triton X-100. Spinal sections were then incubated for 1 h at room temperature followed by 72 h at 4°C in the primary antibody solution: 10% NGS in 1× PBS with 0.2% Triton X-100 and NeuN (mouse anti-NeuN, 1:1,400; cat. no. MAB377; Millipore). Tissue was washed with 1× PBS (3 × 5 min) and incubated in the secondary antibody solution for 1 h at room temperature: 10% NGS in 1× PBS and Alexa Fluor 488 (goat anti-mouse, 1:500; cat. no. A11029; Invitrogen). Following incubation, tissue was washed with 1× PBS (3 × 5 min). For GFAP staining, tissue was incubated for 60 min at room temperature in 5% normal goat serum (NGS) in 1× PBS with 0.2% Triton X-100. Spinal sections were then incubated for 1 h at room temperature and overnight at 4°C in the primary antibody solution: 5% NGS in 1× PBS with 0.2% Triton X-100 and GFAP (mouse anti-GFAP, 1:200; G8393; Sigma-Aldrich). Tissue was washed with 1× PBS with 0.2% Triton X-100 (3 × 5 min) and then incubated in the secondary antibody solution for 1 h at room temperature: 5% NGS in 1× PBS with 0.2% Triton X-100 and Alexa Fluor 594 (goat anti-mouse, 1:500; cat. no. A11005; Life Technologies). Following incubation, tissue was washed with cold 1× PBS (3 × 5 min). Sections were mounted on positively charged glass slides (cat. no. 12-550-15; Fisher Scientific), covered with VECTASHIELD Antifade Mounting Medium (cat. no. H1-1200; Vector Laboratories), and coverslipped. Slides were air-dried and stored at 4°C. All fluorescent images were captured at ×10, ×20, and ×40 magnification with a Microscope Axio Imager.A2 (Carl Zeiss Microscopy).

RESULTS

Optimization of silver deposition from MEAs and subsequent histological development to enable visualization of recording locations.

Initial experiments determined the parameters needed to deposit and identify silver in the spinal cord without emphasis on recording neuronal bursting from the electrodes. As the DC resistance of the commercially purchased recording electrodes was constant and the current used for electroplating silver onto the electrodes was standardized (as described in materials and methods), we assessed the impact of two critical variables on the histological appearance of silver in the spinal cord: 1) the duration of deposition current; and 2) the duration of histological development. Figure 3 shows representative images of silver labeling obtained at different durations of deposition and histological development that demonstrate the variability of silver staining achieved by altering these parameters. Consistent and discrete labeling at a resolution between approximately 25 and 50 μm was obtained when current used to deposit silver was applied for 50 s and tissue was developed for 7–10 min (Fig. 3, A and B). When deposition current of <50 s was applied, few apparent labels were detected, presumably due to an inadequate amount of silver ions necessary for visualization. In contrast, deposition of 80–100 s resulted in a “halo effect” that increased the silver-labeled area to approximately 100–150 μm (Fig. 3, C and D), most likely due to a broad distribution of deposited silver. Similarly, histological development of <7 min was not long enough to “develop” the deposited silver required to visualize the label, whereas prolonged exposure to the developing solution tended to darken large areas of the section thus confounding differentiation due to insufficient contrast (data not shown).

Fig. 3.

Representative images of silver labeling. A–D: representative histological sections from the cervical spinal cord containing silver-labeled sites counterstained with cresyl violet and high-resolution images of the boxed areas. These images depict the variability of silver labeling at different durations of silver deposition (50–100 s) and histological development (7–10 min). E: images of a cervical spinal section depicting an electrode track (black arrows) coursing through the spinal tissue and terminating at the silver-labeled site. CC, central canal; DH, dorsal horn; VH, ventral horn. Scale bars: A–D: 0.5 mm and 50 μm (callouts); E: 0.5 mm, 100 μm, and 50 μm, respectively.

We next verified that the sites of silver deposition represented the location of the tip (i.e., recording location) of a given electrode. In a subset of animals (n = 2 with silver processing, n = 1 unprocessed positive control), 2 of the 16 silver-plated electrodes were coated with the fluorescent lipophilic dye DiI to visualize the tracks associated with the electrode insertion (DiCarlo et al. 1996; Márton et al. 2016; Naselaris et al. 2005). Although DiI was readily detected in unprocessed tissue (data not shown), DiI could not be detected after the histological processing required for silver labeling. Therefore, we conclude these two techniques are not compatible. However, in several histological sections, an electrode track could be identified terminating at the silver-labeled site (Fig. 3E). This is consistent with the interpretation that the silver-labeled site is located at the tip of the electrode.

Another objective was to determine whether the silver-labeling method could be coupled with fluorescent immunohistochemistry (IHC) procedures. Accordingly, we performed IHC with two markers commonly used in the central nervous system (GFAP and NeuN) either before or immediately after histological processing for silver development. These markers were chosen due to their ability to label astrocytes (Bignami et al. 1972) and neurons (Wolf et al. 1996). Fluorescent staining was detected in both cases but was more robust when IHC was performed after the silver staining procedures (data not shown), indicating IHC (at least for these commonly used markers) is compatible with the histological processing steps necessary to develop the silver. Using this staining order, we determined the location of deposited silver relative to NeuN-positive cells. Representative photomicrographs from a C4 spinal section depicting positive silver labeling in the intermediate gray is shown in Fig. 4A. Fluorescent labeling of the same section stained with NeuN and 4′,6′-diamidino-2-phenylindole (DAPI) are shown in Fig. 4B. High-magnification images of the silver-labeled site suggest the recording electrode was in close proximity to a NeuN-positive neuron (Fig. 4B). These qualitative results demonstrate that the silver-labeling technique can be coupled with fluorescent IHC.

Fig. 4.

Silver labeling coupled with fluorescent immunohistochemistry. A: representative cervical spinal section of positive silver labeling (brown/black) and high-resolution image of the boxed area. B: fluorescent labeling of neurons stained with NeuN (green) and nuclei stained with DAPI (blue) of the same section presented in A, and high-resolution image of the boxed area illustrating the silver-labeled site was in close proximity of a NeuN-positive cell. CC, central canal; DH, dorsal horn; VH, ventral horn. Scale bars: 0.5 mm and 50 μm (callouts).

Using silver labeling to identify anatomic locations of metal array electrodes.

In 6 animals (n = 4 spinal-intact and n = 2 SCI), we used the optimized deposition and development parameters discussed previously to label select electrodes. Overall, using this approach to identify the anatomic locations of the recording electrodes, we obtained positive labeling in 35 of 39 attempts. This indicates that the silver-labeling method can be used to identify the anatomic locations of array electrodes with a success rate of ∼90%. Positive silver labeling was identified in the spinal gray matter between C3 and C5. The locations of each silver-labeled site in spinal-intact and SCI animals was plotted (unilaterally for simplicity) according to their anatomic locations (Fig. 5A). Silver-labeled sites were identified between laminae IV and X, with the greatest number found in lamina VII (Fig. 5A).

Fig. 5.

Electrode and silver-labeled depth. A: representative C4 section of silver-labeled sites in spinally intact (●) and spinally injured (○) animals projected onto 1 side of the cord for simplicity. Each lamina was shaded according to the number (#) of silver-labeled sites within that lamina. B: scatter plot of the micromotor depths vs. the depths of the corresponding silver-labeled sites in spinally intact (●) and spinally injured (○) animals. Linear regression analysis was applied to determine the line of best fit and linear equation for each group. The line of identity is displayed to indicate the location where micromotor and silver-labeling depths are equal. C: normalized depth calculated as the difference between the micromotor depth silver-labeled site for spinally intact (●) and spinally injured (○) animals. *P < 0.001, unpaired t-test.

The success of the discrete silver labeling afforded the chance to compare the intended recording location (i.e., the micromotor coordinates used during the neurophysiology experiment) with the actual location of the labeling (Fig. 5, B and C). Each silver-labeled site was measured from the dorsal surface of the cord. Based on estimates of prior investigations regarding tissue shrinkage during paraformaldehyde fixation and subsequent tissue processing (Deutsch and Hillman 1977; Quester and Schröder 1997), the measured histological depth of each silver-labeled site was adjusted by 10%. This analysis indicated that the micromotor depth is likely to overestimate the actual depth of the electrode and perhaps even more importantly that this relationship is altered by SCI. Linear regression analysis indicated a significant relationship between recording coordinates and histological staining in spinal-intact animals (motor: 1.6 ± 0.04 mm; histological: 1.2 ± 0.05 mm; P = 0.0003) but not after chronic SCI (motor: 1.4 ± 0.07 mm; histological: 0.6 ± 0.05 mm; P = 0.3014; Fig. 5B). The depth of the silver-labeled site differed from the micromotor coordinate in both groups (Fig. 5C) but was significantly increased following SCI (0.8 ± 0.07 mm) compared with spinal-intact animals (0.4 ± 0.04 mm). These data highlight the need to identify multielectrode recording sites histologically, especially following experimental conditions such as SCI when tissue fibrosis and scarring can be expected to alter electrode movement within the central nervous system (CNS). Future studies can apply the linear fit as a proxy to calculate the actual depth of the electrode tips more accurately.

Coupling spinal discharge with anatomic locations.

All silver-electroplated electrodes were capable of discriminating single units, indicating that the pretreatment had a minimal, if any, impact on the ability to record and discriminate extracellular signals. Three silver-labeled sites were identified within lamina IX of the ventral horn (Figs. 5A and 6A). Spike-triggered averaging (STA) of the raw and rectified phrenic nerve activity in relation to neuronal discharge produced a distinct peak with an average lag time of 0.45 ± 0.16 ms (Fig. 6E) and therefore indicated that the recorded discharge was from phrenic motoneurons (Christakos et al. 1994; Mitchell et al. 1992). All three of these cells were active primarily during the inspiratory phase (Fig. 6, B and C), which is the typical firing pattern of phrenic motoneurons in this preparation (Sandhu et al. 2015). The silver-labeling data verified that the recordings were from the region of the phrenic motor nucleus (Furicchia and Goshgarian 1987; Goshgarian and Rafols 1981; Kinkead et al. 1998; Mantilla et al. 2009; Prakash et al. 2000). This close matching between neurophysiological data and anatomic data illustrates the effectiveness of the silver-labeling method.

Fig. 6.

Silver labeling of a phrenic motoneuron. A: representative C4 section containing a silver-labeled site counterstained with cresyl violet and high-resolution image containing the silver-labeled site located in the medial aspect of the ventral horn in lamina IX (callout). B: corresponding raw spinal discharge and integrated phrenic output during baseline, hypoxia, and posthypoxia depicting discharge during the inspiratory phase. C: integrated phrenic output, raw neuronal discharge, and sorted spikes (waveform) during hypoxia. D: cycle-triggered histogram during 50 consecutive breaths overlaid with the average integrated phrenic waveform during hypoxia. E: spike-triggered average of the raw and rectified (Rec) phrenic nerve depicting a delay of 0.76 ms. CC, central canal; DH, dorsal horn; VH, ventral horn. Scale bars: A: 0.5 mm and 50 μm (callout); B and C: 1.5 s.

The remaining silver-labeled sites were located between laminae IV and X (Fig. 7). Spike-triggered averaging provided no evidence of discharge synchrony in relation to phrenic motor output; thus these 28 recordings were considered to represent interneurons (Fig. 7, E and J). At baseline, most recorded interneurons (18/28 or 64%; Fig. 7K) fired tonically throughout the respiratory cycle (e.g., waveform 2 in Fig. 7, C, D, H, and I). These tonically discharging cells were not restricted to a particular lamina but rather were recorded throughout the cervical gray matter (Fig. 7L). A smaller proportion of interneurons (5/28 or 18%; Fig. 7K) primarily fired during the inspiratory phase (waveform 1 in Fig. 7, C and D), and these cells were found in laminae VI (n = 1), VII (n = 3), and X (n = 1). During hypoxia, 2 of these neurons (in laminae VII and X) switched to a tonic firing pattern but then resumed an inspiratory pattern posthypoxia. In addition, 1 inspiratory neuron in lamina VII was inspiratory-modulated during baseline and hypoxia but then assumed a tonic firing pattern posthypoxia. Similarly, during baseline conditions, 5 cells (5/28 or 18%; Fig. 7K) discharged primarily during the expiratory period (e.g., waveform 1 in Fig. 7, H and I) and were located in laminae VIII (n = 3), IX (n = 1), and X (n = 1). During hypoxia, 3 of these neurons [in laminae VIII (n = 2) and IX (n = 1)] switched to a tonic firing pattern that was then maintained posthypoxia. In addition, 1 expiratory-modulated neuron in lamina VIII ceased bursting immediately posthypoxia. Therefore, due to phase switching, the proportion of tonic firing interneurons increased to 82% (23/28) during hypoxia and remained elevated posthypoxia (20/28 or 71%). Interneuron bursting patterns during baseline, hypoxia, and posthypoxia are summarized in Fig. 7K, and the corresponding anatomic locations are provided in Fig. 7L. In this sample of neurons, respiratory-related discharge was observed only in the ventral gray matter (i.e., laminae VI, VII, VII, IX, and X), and tonic discharge was recorded throughout the midcervical gray matter.

Mapping anatomic locations and quantitation of interneuronal discharge across midcervical spinal laminae are shown in Fig. 8. A representative example from 1 recording in which silver labeling was used to “map” the anatomic locations of the array electrodes and corresponding electrophysiological data is shown in Fig. 8, A–C. A camera lucida-style drawing of the anatomic locations of 8 silver-labeled electrodes was constructed (Fig. 8B). Numerical identification of the anatomic positions of each electrode corresponds to the schematic presented in Fig. 1D and defines the rostral-caudal and medial-lateral positions of the array electrodes. Corresponding midcervical spinal discharge and integrated phrenic nerve activity during baseline and hypoxia on the left and right hemicord is shown in Fig. 8C. The majority of these recordings were from tonic firing interneurons (electrodes 3, 8, 9, 11, 12, and 15), and one represented a phase-switching (e.g., from tonic at baseline to expiratory during hypoxia) interneuron (electrode 6; Fig. 8C). Using this technique, a summary of the anatomic locations of all silver-labeled interneurons was constructed (Fig. 8D). In addition, the average discharge frequency of all silver-labeled interneurons was presented in Fig. 8E. The results indicate a dorsal-ventral discharge gradient, with higher discharge rates in laminae IV and X and lower values in laminae VIII and IX. This discharge map of the midcervical spinal network demonstrates the utility of using the silver-labeling technique to identify the anatomic locations of ensembles of interneurons recorded with a MEA. These results show the usefulness of using this technique to standardize the sampling distribution of recording locations across experimental groups.

Fig. 8.

Mapping anatomic locations of midcervical spinal neurons. A: representative photomicrograph of a C4 section containing a silver-labeled site (from electrode 9) counterstained with cresyl violet and high-resolution image of the silver-labeled site (callout). B: camera lucida-style drawing of the cervical spinal cord summarizing the anatomic locations of 8 silver-labeled sites obtained in 1 animal. C: integrated phrenic motor output and midcervical spinal discharge on the left and right hemicord corresponding to identified silver-labeled sites in B during baseline and hypoxia. Electrode 1 recorded phrenic motoneuron discharge, and the remaining electrodes recorded interneuron discharge. D: representative C4 section summarizing the total number of silver-labeled interneurons within each lamina in spinal-intact animals. E: average discharge frequency of spinal interneurons within each lamina represented in D. CC, central canal; DH, dorsal horn; VH, ventral horn; NaN, no cells recorded. Scale bars: A and B: 0.5 mm and 50 μm (callout); C: 1.5 s.

Temporal relationships between cervical interneuron discharge.

Anatomic data indicate that cervical interneurons are part of a diffuse and synaptically coupled propriospinal network (Lane et al. 2008b), but relatively little is known about functional connectivity between cervical interneurons. Therefore, initial analyses focused on all recorded neurons in spinal-intact rats (i.e., silver- and nonsilver-labeled neurons) to screen for short time scale (i.e., 0–10 ms) features (Aertsen and Gerstein 1985; Aertsen et al. 1989; Melssen and Epping 1987). Significant features were identified as departures in the cross-correlation histogram ≥3 standard deviations from the background noise (Melssen and Epping 1987). Using this approach, significant central (i.e., no lag time from 0) and offset correlogram peaks were detected (Fig. 9A). A summary of the latency of significant features relative to the trigger (i.e., time 0) are shown in Fig. 9B.

Fig. 9.

Cross-correlation analysis of midcervical spinal interneurons. A: representative cross-correlations obtained from 3 pairs of neurons depicting a significant central peak (left), 1.8-ms offset peak (middle), and 7.8-ms offset peak (right). The number of trigger spikes for each correlation: 43,891; 39,208; and 3,928; and the detectability index of each correlation: 45.2; 4.3; and 5.3, respectively. Black lines plotted at 0.5 and 2.9 ms to indicate how data were grouped by latency (see B). B: a histogram of the latency relative to the trigger for all correlations with a significant peak. The 1st bar in the histogram is a count of the central peaks (0-ms latency), and each successive bar in the plot represents counts obtained in an increment of 0.2. Yellow bars plotted at 0.5 and at 2.9 ms indicate how data were grouped based on latency. Black lines are 2nd-order polynomial fit of the data. C: the sum of significant cross-correlations with latencies ≤0.4 ms, between 0.6 and 2.8 ms, and ≥3.0 ms. Percentages reflect the proportion of positive correlations out of the total possible (n = 704). D: the average number of positive correlations per animal when both recordings were on the same side of the cord (20 ± 3.9 positive correlations per animal) or on opposite sides of the cord (25 ± 7.5 positive correlations per animal). E: the number of significant correlations expressed as a percentage of total possible connections obtained when both recordings were on the same side of the cord (81/332 total possible) or on opposite sides of the cord (89/372 total possible).

Inspection of the data suggested that latencies for correlogram peaks were distributed in 3 ranges: 1) ≤0.4 ms; 2) 0.6–2.8 ms; and 3) ≥3.0 ms (discrete integers reflect 0.2-ms bins used to construct histograms; Fig. 9, B and C). Few correlations (37/704 or 5.3% of possible correlations) had a peak with latency ≤0.4 ms, which is consistent with excitation from a common synaptic input (Aertsen and Gerstein 1985; Kirkwood 1979; Kirkwood et al. 1991; Melssen and Epping 1987). A greater number of positive correlations had latencies between 0.6 and 2.8 ms and ≥3.0 ms (54/704 or 7.7% and 97/704 or 13.8%, respectively). Correlogram peaks with latencies >0.6 ms are consistent with the interpretation of functional excitation involving mono- or polysynaptic connections (Aertsen and Gerstein 1985; Kirkwood 1979; Moore et al. 1970). We next evaluated the number of positive correlations relative to the recording locations. For this, neuronal pairs were classified as unilateral (i.e., recordings on the same side of the cord) or bilateral (i.e., recordings on opposite sides of the cord). Evaluation of unilateral pairs showed the average number of significant correlogram peaks was 20 ± 3.9 correlations per animal. A similar value was obtained for bilateral neuronal pairs (25 ± 7.5 correlations per animal; Fig. 9D). When normalized to the total number of possible unilateral (332) or bilateral (372) correlations, similar values were observed (24 ± 0.5 and 26 ± 5.5%, respectively; Fig. 9E). No significant differences in latencies were observed when data were separated into unilateral and bilateral neuronal pairs (data not shown).

Neuronal pairs with significant correlogram peaks that were also silver-labeled were used to construct correlation summary maps to illustrate the anatomic locations of each neuron (Fig. 10). The number of positive correlations within each lamina was normalized to the total number of neurons present in the corresponding lamina. Unilateral neuronal pairs are shown in Fig. 10A, and bilateral pairs are shown in Fig. 10B. Of the positive unilateral correlations, interneurons in dorsal lamina (i.e., IV, V, and VI) made and received the greatest number of excitatory connections (Fig. 10A). In contrast, when the trigger and target neuron were on opposite sides of the spinal cord, interneurons in laminae V, VIII, and IX made and received the most connections (Fig. 10B).

Fig. 10.

Anatomic locations of functionally connected interneurons. Summary maps of silver-labeled interneurons with significant peaks in cross-correlograms. The number of positive correlations in each lamina was normalized to the number of neurons present in the lamina and shaded accordingly. A: when both trigger and target neurons were on the same side of the spinal cord (e.g., unilateral recordings), interneurons in dorsal lamina (i.e., IV, V, and VI) made and received the greatest number of excitatory connections. B: when the trigger and target neurons were on opposite sides of the spinal cord (e.g., bilateral recordings), interneurons in laminae V, VIII, and IX made and received the most connections.

We also found 33 troughs in cross-correlograms that were significantly different from background activity using the same criteria used to detect peaks (data not shown). For comparison purposes, the number of troughs occurring ≤0.4 was 20/704 or 2.8%, whereas those between 0.6 and 2.8 ms was 1/704 or 0.1% and those ≥3.0 ms was 12/704 or 1.7%. Correlogram troughs with these latencies are generally thought to reflect functional inhibition (Aertsen and Gerstein 1985; Kirkwood 1979).

DISCUSSION

Here, we describe a silver-labeling technique that reliably identifies the position of MEA tips following in vivo neurophysiological recordings. The results indicate that silver labeling can be histologically identified with a resolution of ∼50 μm while also preserving the structural integrity of the tissue with a 90% success rate. A practical application of this technique was demonstrated by matching the location of midcervical MEA recording sites with neuronal discharge to create maps of bursting patterns (e.g., Figs. 7 and 8). Moreover, using correlative techniques, we also provide to our knowledge the first neurophysiological evidence for the presence of extensive connections between interneurons in midcervical gray matter (e.g., Figs. 9 and 10). Coupling silver labeling with MEA neurophysiology provides a powerful tool to investigate network-level properties of the spinal cord.

Commentary regarding the silver-labeling method.

The first challenge associated with adapting the silver-labeling method (Spinelli 1975) for MEAs was accurately measuring the low currents (in nanoamperes) needed for electroplating and depositing silver from tungsten array electrodes. Although devices that measure low currents are commercially available (e.g., for patch-clamp/iontophoresis), they are expensive and not easily adapted to electroplating or electrodeposition. Additionally, multimeters cannot resolve currents in the nanoampere range and have limited accuracy due to voltage drop caused by current flow through the current-measuring device (i.e., “burden voltage”; Jones 2010). To overcome these limitations, we used an EEVblog μCurrent Precision Assistant that permits nanoampere currents to be measured on a standard multimeter and also improves the accuracy to 0.2% (Jones 2010). We also designed a μCurrent control device capable of generating 100 nA while also allowing current adjustments of <1 nA (circuit details provided in Fig. 2C). The ability to produce and control this current precisely was critical since passing higher currents can alter electrode impedance and recording characteristics (Fung et al. 1998). The current-generating system was specifically designed to be compatible with both single electrodes and MEAs.

Published methods for marking the location of MEAs include iontophoresis of neural tracers (Fekete et al. 2015; Haidarliu et al. 1999; Kovács et al. 2005), topical application of fluorescent dyes (DiCarlo et al. 1996; Naselaris et al. 2005), lesioning (Brozoski et al. 2006; Townsend et al. 2002), electrical imaging (Li et al. 2015), and imaging-based approaches (Borg et al. 2015; Fung et al. 1998; Koyano et al. 2011; Matsui et al. 2007). The silver-labeling method offers several advantages over these previous techniques. First, this technique can be used with the standard metal electrodes typically used with MEAs (Borg et al. 2015; Cogan 2008). This is in contrast to the specialized electrodes necessary for iontophoresis (Fekete et al. 2015; Kovács et al. 2005; Naselaris et al. 2005). Second, the silver-labeling method does not require implanted electrodes or expensive and often unavailable equipment such as MRI to visualize electrode placement (Borg et al. 2015; Fung et al. 1998; Koyano et al. 2011; Matsui et al. 2007). Third, the histologically identified silver deposition spanned an area of ∼50 μm, and this provides an ∼3× greater resolution compared to previously published techniques compatible with metal MEAs. For example, MRI-based approaches enable visualization of microelectrode sites at an in-plane resolution of 150–200 μm (Fung et al. 1998; Koyano et al. 2011; Matsui et al. 2007). Fluorescent dyes achieve resolutions between 50 and 400 μm but identify electrode track trajectories rather than discrete locations of electrode tips (DiCarlo et al. 1996; Naselaris et al. 2005). Finally, whereas lesioning techniques can damage surrounding tissues (Townsend et al. 2002), low current levels required for silver deposition preserve the integrity of the surrounding tissue. This is a particularly important point since it permits neurochemical phenotyping of neurons at or near the electrode tip.

One of the features enabling matching of electrophysiological signals with anatomic locations is that the relative position of each electrode was known since the electrodes exist in a fixed matrix (Fig. 1D). Nevertheless, with multiple silver-labeled sites in a relatively discrete area, matching each silver-labeled site with its corresponding recording electrode can pose a challenge. To overcome this hurdle, several safeguards were implemented. First, since recordings were performed with a bilateral recording array, a small longitudinal cut in the dorsal spinal cord was made before sectioning to demarcate the left vs. right hemicord. Second, the rostral-caudal orientation of spinal sections was maintained throughout the histological process by sectioning sequentially into 12-well plates. Most importantly, the histologically identified silver deposition was linked to the corresponding recording data by using the lateral-medial aspect of the electrode matrix. The phrenic motoneuron data provided further verification of the ability to match silver labeling with the neurophysiology data. The anatomic location of phrenic motoneurons is well-defined in rat (Goshgarian and Rafols 1981; Mantilla et al. 2009; Prakash et al. 2000; Zhan et al. 1989), and the electrophysiological analyses (e.g., STA) confirmed an appropriate match between the anatomic location of silver labeling (i.e., ventral gray matter, lamina IX) and the recording electrode. Moreover, cells with histological features consistent with phrenic motoneurons (Furicchia and Goshgarian 1987; Prakash et al. 2000) were clearly identified near the silver label (approximately 50–100 μm).

To ensure that our recordings (and silver-labeled sites) were from neurons (vs. fibers of passage) in the immediate vicinity of electrode, we used high-impedance tungsten wires coated with Epoxylite insulation with an exposed tip of ≤1 μm. We suggest that using these insulated high-impedance electrodes, the following assumptions are reasonable: 1) the recorded neural discharge is of somatic origin (vs. axonal); and 2) the amplitude of the recorded action potentials decreases with distance between the soma and the recording electrode tip. The reasons for these assumptions are as follows. Extracellular action potentials recorded from axons are very brief (i.e., 0.1–0.3 ms), small, and extremely sensitive to electrode movement (Bellingham and Lipski 1990; Kirkwood et al. 1988). In contrast, action potentials recorded in the current study were broad (i.e., ≥1 ms) and biphasic, and the recordings were stable as electrodes were moved up to 100 μm. These features are consistent with extracellularly recorded somatic action potentials (Bellingham and Lipski 1990; Nelson 1959). The literature indicates that the amplitude of extracellularly recorded action potentials is inversely proportion to the distance between the soma and the recording electrode (Kirkwood et al. 1988; Nelson 1959). Therefore, by achieving a 3:1 signal-to-noise ratio when placing each electrode, we can be confident that the electrode tip is close to the cell soma generating the largest action potential. Last, the electrode design described above reduces the surface area available for silver plating, and thus silver is deposited from only the ∼1-μm plated length at each electrode tip.

The silver-labeling method was not 100% successful, and we suspect the small number of failures (n = 4 of 39 possible sites) occurred due to a lack of deposited silver or tissue damage during histological processing (e.g., tearing of sections or error during the placement of longitudinal notch in the spinal cord). A final methodological commentary relates to the difference between the histologically verified depth of the electrode and the coordinates used in the microdrive while placing each electrode (e.g., Fig. 5). The microdrive coordinates overestimated the actual depth of the recording and probably reflect some degree of “pillowing” of the tissue as the electrodes were advanced. The most salient point, however, is that the difference between histological and micromotor coordinates was increased in rats with chronic SCI (e.g., Fig. 5C). These data highlight the need to identify multielectrode recording sites histologically following SCI when tissue fibrosis and scarring can be expected to impede electrode movement within the spinal cord.

Midcervical spinal interneurons and respiratory motor output.

Several laboratories have suggested that phrenic motoneuron discharge can be modulated by synaptic input from spinal interneurons (Bellingham and Lipski 1990; Davies et al. 1985; Sandhu et al. 2015). Anatomic studies show spinal interneurons are uniquely situated to modulate phrenic motor output since they are in close opposition with medullary projections (Davies et al. 1985; Fedorko et al. 1983; Hayashi et al. 2003; Lane et al. 2008b) and have synaptic connections to phrenic motoneurons (Dobbins and Feldman 1994; Lane et al. 2008b, 2009; Lois et al. 2009; Yates et al. 1999). Electrophysiological recordings in multiple species have demonstrated that cervical interneurons have respiratory-related discharge patterns (Bellingham and Lipski 1990; Duffin and Iscoe 1996; Marchenko et al. 2015; Palisses et al. 1989; Sandhu et al. 2015) and respond to respiratory stimuli such as phrenic afferent stimulation (Iscoe and Duffin 1996; Speck and Revelette 1987) and hypoxia (Sandhu et al. 2015). Our results add to this literature in two primary ways. First, to our knowledge, only one prior study (Marchenko et al. 2015) has combined neurophysiology with the histological identification of the anatomic location of respiratory-related spinal interneurons associated with the phrenic motor pool. Our data show that interneurons with respiratory-related discharge are located throughout laminae VI, VII, VII, IX, and X. Second, the current results provide new information regarding the impact of respiratory stimulation with hypoxia on midcervical interneuronal discharge patterns. Indeed, during/following hypoxia, ∼29% of spinal interneurons alter their discharge in relation to the respiratory cycle. We noted that both inspiratory- and expiratory-modulated neurons had a tendency to adopt a tonic firing pattern during hypoxia. However, posthypoxia interneurons with baseline inspiratory discharge tended to resume an inspiratory firing pattern, whereas the expiratory interneurons maintained the tonic pattern. In addition, we found one example of a tonic firing interneuron during baseline and hypoxia that became inspiratory-modulated posthypoxia. Neurons exhibiting phase switching were located in intermediate and ventral laminae (i.e., VII, VIII, and X) rather than dorsal laminae. To our knowledge, this is the first report to describe hypoxia-induced phase switching of midcervical interneuron bursting patterns. When neuronal discharge frequency was evaluated relative to anatomic locations, higher frequencies were noted in laminae IV and X (e.g., Fig. 8E). These results again highlight the importance of matching anatomic location with discharge properties and show that midcervical interneuron discharge varies across cervical lamina, which is consistent with previous descriptions of lumbosacral neural activity (Borowska et al. 2013; Ruscheweyh and Sandkühler 2002).

Discharge synchrony between midcervical spinal interneurons.

Despite both historical (Duffin and Iscoe 1996; Palisses et al. 1989) and recent publications (Lane 2011; Marchenko et al. 2015; Sandhu et al. 2015) related to cervical respiratory interneurons, the functional connectivity of midcervical spinal circuit remains largely unexplored. This can be attributed to the fact that studies investigating respiratory-related interneurons in the spinal cord have primarily used a single-unit recording approach (Bellingham and Lipski 1990; Lipski and Duffin 1986; Lipski et al. 1993) and have instead focused on connections between motor- and interneurons (Davies et al. 1985; Duffin and Iscoe 1996; Lipski et al. 1993). In the current study, we utilized cross-correlation analysis to characterize the functional connectivity of pairs of spinal interneurons. Similar to previous studies investigating spinal interneurons in nonrespiratory-related networks (Brown et al. 1979; Prut and Perlmutter 2003), we found relatively few examples of synchronous discharge consistent with a shared excitatory presynaptic input (Aertsen et al. 1989; Kirkwood 1979). That is, when interneuronal pairs were examined, correlogram peaks with latencies between 0 and 0.4 ms were only observed in 5.3% of recordings. In contrast, 21.4% of the total possible positive correlations had latencies ≥0.6 ms. These longer latency peaks are consistent with mono- and polysynaptic connections between cervical interneuron pairs. Of the positive correlations involving neurons on the same side of the spinal cord, interneurons in dorsal lamina (i.e., IV, V, and VI) made and received the greatest number of excitatory connections. In contrast, when the trigger and target neuron were on opposite sides of the spinal cord, interneurons in laminae V, VII, and IX made and received the most connections. Taken together, our results provide support for the hypothesis that cervical interneurons form a dynamic network that is capable of rapid reconfiguration and modulation of cervical motor outputs. We suggest that the MEA silver-labeling method will help advance our understanding of the functional and anatomic correlates of ensembles of spinal neurons.

Application to spinal cord injury.

One potential application of MEA technology is to examine how SCI alters the cervical spinal networks. Spinal networks undergo substantial remodeling following SCI (Bareyre et al. 2004; Sperry and Goshgarian 1993), and interneurons have been implicated in SCI-induced plasticity and motor recovery (Alilain et al. 2008; Bareyre et al. 2004; Harkema 2008). However, relatively little is known regarding their contribution to the spontaneous recovery of phrenic output following SCI (Lane et al. 2008b). In the current study, we performed recordings on n = 2 rats with chronic cervical SCI. The intent of these experiments was not to map changes in the spinal network postinjury but rather to confirm that the method developed herein could be effectively utilized in rats with SCI. We successfully deposited and detected silver labeling following SCI using the same parameters optimized in spinal-intact animals. However, we noted a greater discrepancy between micromotor and silver-labeling depth in spinal-injured animals (e.g., Fig. 5C). Although multiple factors may contribute to this disparity, structural changes resulting from fibrosis and scarring induced following injury (Cregg et al. 2014) may increase electrode drag, thereby reducing the effectiveness of utilizing the electrode micromotor to predict the actual depth of the electrode tips. Regardless of the specific cause, our results highlight the importance of histologically verifying the locations of recorded neurons, especially following experimental conditions such as SCI that may alter electrode movement within the CNS. We suggest that future studies utilizing the methods described here will shed light on the contribution of the propriospinal network to the recovery of function following SCI.

Conclusions.

The methodology described here provides a solution to the fundamental limitation of using MEAs containing standard metal electrodes by identifying the discrete anatomic locations of the recording sites. Although prior studies have described methods to determine the approximate locations of MEA electrodes (Borg et al. 2015; Brozoski et al. 2006; DiCarlo et al. 1996; Fekete et al. 2015; Haidarliu et al. 1999; Kovács et al. 2005; Koyano et al. 2011; Naselaris et al. 2005; Townsend et al. 2002), to date labeling the discrete recording locations of nonspecialized metal array electrodes in an acute preparation while preserving the surrounding tissue has only been achieved with single-cell recordings. With MEA recordings, higher-order analyses can be used to investigate the functional connectivity of the neural network during physiological stimuli and following neurological injury (e.g., Figs. 9 and 10; also see: Aertsen and Gerstein 1985; Kirkwood 1979; Melssen and Epping 1987). In this regard, the current results are consistent with a high degree of synaptic connections between midcervical neurons. Collectively, our experiments show that: 1) MEA silver-labeling method enables the electrophysiological output of neural networks to be coupled with histological verification of electrode locations; 2) midcervical interneurons are capable of rapid phase switching of burst patterns relative to the respiratory cycle during and after hypoxia; and 3) a high percentage of midcervical interneuronal pairs have temporally related discharge patterns.

GRANTS

This work was supported by the National Institutes of Health Grants 1-R01-NS-080180-01A1 (D. D. Fuller), 1-F32-NS-095620-01 (K. A. Streeter), and T32-ND-043730 (M. D. Sunshine) and the Department of Defense Grant W81XWH-14-1-0625 (P. J. Reier).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

K.A.S., P.J.R., D.D.F., and D.M.B. conceived and designed research; K.A.S., S.R.P., S.S.L., and L.E.D. performed experiments; K.A.S., M.D.S., S.R.P., S.S.L., L.E.D., and D.M.B. analyzed data; K.A.S., M.D.S., D.D.F., and D.M.B. interpreted results of experiments; K.A.S. and M.D.S. prepared figures; K.A.S. and D.M.B. drafted manuscript; K.A.S., D.D.F., and D.M.B. edited and revised manuscript; K.A.S., M.D.S., S.R.P., S.S.L., L.E.D., P.J.R., D.D.F., and D.M.B. approved final version of manuscript.