Development of an integrated microperfusion-EEG electrode for unbiased multimodal sampling of brain interstitial fluid and concurrent neural activity

Abstract Objective. To modify off-the-shelf components to build a device for collecting electroencephalography (EEG) from macroelectrodes surrounded by large fluid access ports sampled by an integrated microperfusion system in order to establish a method for sampling brain interstitial fluid (ISF) at the site of stimulation or seizure activity with no bias for molecular size. Approach. Twenty-four 560 µm diameter holes were ablated through the sheath surrounding one platinum–iridium macroelectrode of a standard Spencer depth electrode using a femtosecond UV laser. A syringe pump was converted to push–pull configuration and connected to the fluidics catheter of a commercially available microdialysis system. The fluidics were inserted into the lumen of the modified Spencer electrode with the microdialysis membrane removed, converting the system to open flow microperfusion. Electrical performance and analyte recovery were measured and parameters were systematically altered to improve performance. An optimized device was tested in the pig brain and unbiased quantitative mass spectrometry was used to characterize the perfusate collected from the peri-electrode brain in response to stimulation. Main results. Optimized parameters resulted in >70% recovery of 70 kDa dextran from a tissue analog. The optimized device was implanted in the cortex of a pig and perfusate was collected during four 60 min epochs. Following a baseline epoch, the macroelectrode surrounded by microperfusion ports was stimulated at 2 Hz (0.7 mA, 200 µs pulse width). Following a post-stimulation epoch, the cortex near the electrode was stimulated with benzylpenicillin to induce epileptiform activity. Proteomic analysis of the perfusates revealed a unique inflammatory signature induced by electrical stimulation. This signature was not detected in bulk tissue ISF. Significance. A modified dual-sensing electrode that permits coincident detection of EEG and ISF at the site of epileptiform neural activity may reveal novel pathogenic mechanisms and therapeutic targets that are otherwise undetectable at the bulk tissue level.


Introduction
Seizures result in the fluctuation of both neurochemical and electrical biomarkers within the central nervous system (CNS), making simultaneous dual sampling from within the diseased microenvironment valuable for understanding disease mechanisms and treatment modalities [1].Historic collection techniques including microdialysis [2] and intracranial electroencephalography (EEG) [3,4] have provided insights into the relationship between neurochemical and electrical fluctuations in the CNS but have been limited with regard to spatial concurrency of the sampling modalities due to the physical separation of each probe type.Most previous designs have also been limited to collection of small molecular weight compounds due to the use of size-restricted semipermeable microdialysis membranes [5], precluding the unbiased collection of large macromolecular complexes and cell-derived membrane-bounded signaling entities such as extracellular vesicles [6][7][8].
Previous efforts to collect spatially coincident fluid and functional data within the brain include adhesion of microwire electrodes or Teflon-coated stainless steel recording electrodes to the outside of commercially available microdialysis catheters [9,10] and insertion of wire electrodes into the microdialysis probe lumen [11][12][13].The first method is associated with the potential separation of the electrode from the catheter during insertion, resulting in spatial decoherence of the signals, as well as the potential for tissue damage due to the large effective diameter of the electrode-catheter pair.The second method exhibits poor electrophysiological sensitivity due to interference from the microdialysis membrane and signal averaging across the fluid volume inside the probe.A probe design used by Fried and colleagues in which platinum iridium microwires splayed out along the tip of a microdialysis membrane provided unprecedented resolution of changes in neurotransmitter levels associated with single-unit recordings collected during seizures detected by platinum iridium macroelectrodes located along the microdialysis catheter body [14].Likewise, a modified Spencer electrode with perforations was used by Spencer and colleagues to perform zero-flow microdialysis to measure glutamate release in the hippocampus in response to 50 Hz stimulation [15].Building on this concept, we sought to use off-the-shelf components to develop a reproducible device for collecting EEG from macroelectrodes surrounded by large fluid access ports sampled by an integrated microperfusion system [1], thereby establishing a method with no bias for molecular or macromolecular size.

Electrode modification
The electrode subassembly was a Spencer depth electrode (SD04R-SP10X; AD-TECH) rated for 30 d of implantation, with a maximum charge density of 30 µC cm −2 and a safe charge injection capacity of 2.1 µC based on the 0.07 cm 2 surface area of each macroelectrode.The electrode is constructed as a multi-walled catheter bearing four distal platinum-iridium macroelectrodes connected to proximal contacts by polymer-coated wires running the length of the shaft (figure 1(A)).An internal polyimide sheath separates the lumen of the electrode shaft from the outer polyurethane annulus and serves to protect the wires from the stainless-steel stylus used to stabilize the electrode during insertion (figure 1(C)).In order to permit free exchange of fluid from the surrounding brain tissue into the lumen of the electrode, a triple wavelength femtosecond laser (PhotoMachining Inc) was used to ablate holes through the outer wall and the polyimide sheath.Exposure to femtosecond pulses of light at 343 nm in a cross-hatch pattern permitted ablation without significant heat dissipation [16,17].Wires were avoided using transillumination to visualize location and a non-reflective black plastic stylus was inserted into the lumen to prevent scattering and to protect the opposing wall.Ablated material was removed by vacuum during the process.We ablated three rows of four holes per row above and below the second most distal macroelectrode (figure 1(B)).Within a row each hole was positioned 90 • apart; the central row of holes was rotated 45 • from the top and bottom row.Each row was separated by 1 mm and the row closest to the macroelectrode was positioned 0.5 mm away.Three different focusing strategies were tested during ablation of the outer wall; for the low energy tests, after ablation of the outer wall the laser was focused on the underlying polyimide sheath and ten passes at 35% power was used to remove the material.

Pump and fluidics modification
The microperfusion pump subassembly was a battery-powered 107 syringe pump (M Dialysis AB) with adjustable flow rate (0.1, 0.2, 0.3, 0.5, 1.0, 2.0, 5.0 µl min −1 ).The pump was converted to push-pull operation by incorporating aluminum couplings that concurrently displace the plunger of the push syringe while moving the pull syringe plunger at the identical rate.The lead screw of the original pump was fitted with a threaded coupler that transferred shaft rotation to linear displacement moving each plunger in opposite directions (figure 1(D)).Our modified push-pull pump system used generic 1 ml syringes (figure 1(E)) instead of the proprietary 2.5 ml syringe that comes with the pump, reducing the standard 107 pump flow rates to 0.016, 0.032, 0.048, 0.08, 0.16, 0.32, and 0.80 µl min −1 .
The fluidics subassembly was from an off-theshelf M Dialysis-71 high molecular weight (100 kDa) microdialysis catheter with a 20 mm membrane.This membrane was removed, resulting in an open flow inlet/outlet within the lumen of the electrode after insertion.Luer-lock syringe tips (JG28-0.5HPX;Jensen Global) were used to connect the syringes to the microdialysis fluidics.

Electrical impedance testing
Modified electrodes were tested for ablation-induced changes in impedance using an external neurostimulator system delivering a single charge balanced current pulse (0.4 mA, 80 µs pulse width) to determine impedance.Electrodes were tested in vitro before and after laser ablation.Each macroelectrode was measured relative to a reference electrode using a saline bath (0.9% NaCl); electrode numbering proceeded proximal-to-distal, with electrode four being the most distal.The modified electrodes were also tested in vivo for impedance using the same pulse parameters employed in the in vitro tests.In addition to the single frequency test, modified and unmodified electrodes were assessed using electrochemical impedance spectroscopy and a frequency sweep from 1 Hz to 100 kHz, as previously described [18].Electrodes were tested on a Reference 600 (Gamry Instruments) potentiostat immediately upon insertion into saline containing 0.2 mg ml −1 bovine serum albumin.After incubation for 6 h in the solution the electrodes were tested again.Impedance and phase measurements were collected on the third most distal macroelectrode on the probe using a coiled platinum wire counter electrode (Gamry Instruments) referenced to an Ag/AgCl electrode (BASi).

Signal-to-noise ratio (SNR) measurement
The in vitro SNR exhibited by modified and unmodified electrodes were measured using artificially generated signal in a saline solution (0.9% NaCl) at 22 • C. Electrodes that had been previously implanted in a pig were cleaned and fixed in parallel in a custom-made plastic reservoir containing 100 ml saline.Background noise on the electrodes was measured in bipolar configuration between the first two macroelectrodes using a Neuralynx-Cube acquisition system.Modified and unmodified electrodes were measured simultaneously for 60 s at 30 kHz using 0.0305 µV quantization steps and bandpass filtering between 0.1 and 7500 Hz.Signal response on the electrodes was tested using an artificial signal mixture comprised of sinewaves of equivalent amplitude at frequencies corresponding to prime numbers between 1 and 41 generated with a National Instruments USB-6251 signal generator and an NPI Electronics ISO-isolation unit.The signal mixture was scaled and injected in current-supply mode with peak-to-peak magnitude of 2 µA into the saline bath using stainless steel electrodes positioned at each end of the reservoir.Signal response was measured for 60 s using the same configuration employed for noise assessment.Voltages detected on the electrodes ranged between 20 and 100 µV, consistent with in vivo measurements.The power spectrum density (PSD) in 10 s non-overlapping windows was calculated using MATLAB throughout the noise and stimulated response recordings.The average noise response at each frequency was derived from the mean of all background noise PSD windows.The SNR at each stimulus frequency was calculated using the ratio of the signal response power in each window divided by the average noise power, followed by conversion to dB.
The in vivo signal response profile was assessed by calculating the PSD in 4096 frequency bins across 15 kHz in 1 s time windows from 100 s of recording collected shortly after electrode implantation and from 100 s recorded during the end of the 3rd hour of implantation.The mean and standard deviation of the power in the early and late recordings was calculated in each frequency bin.Receiver-operating characteristic curves were calculated for the early and late recordings in Prism and area under the curve (AUC) and standard error were derived.The z-score for the difference between curves was calculated by dividing the absolute value of the difference between AUC late and AUC early by the square root of the sum of the squared standard errors for both curves [19].The z-score was used to calculate the two-tailed P value against the normal distribution.

Analyte recovery analysis
The no-net-flux method [1,20,21] was used to measure recovery of trypan blue, a small molecular weight molecule (870 Da), and fluorescein-conjugated dextran, a large molecular weight molecule (70 kDa).In this method the steady state concentration of analyte in the perfusate (C out ) relative to the input concentration (C in ) is plotted against various input concentrations and the slope of the fitted regression line is used to calculate the percent recovery.Target analytes were either dissolved in saline (0.9% NaCl) or distributed in 0.6% agarose, which served as a tissue surrogate.Analyte recovery was performed at 37 • C and endpoint samples collected after 30 min were measured by absorbance (trypan blue: 580 nm) or fluorescence (dextran: Ex490 nm, Em525 nm).

Pig neurosurgery
All study procedures were performed in accordance with the National Institutes of Health Guidelines for Animal Research (Guide for the Care and Use of Laboratory Animals) and were approved by the Mayo Clinic Institutional Animal Care and Use Committee.Pigs were socially housed in a controlled environment with humidity at 45%, temperature at 21 • C, with once daily feeding and ad libitum access to water.The experimental animal was fasted before the day of surgery.The subject was initially sedated with a mixture of Telazol (tiletamine and zolazepam) (5 mg kg −1 ) and xylazine (2 mg kg −1 ) via intramuscular injection.An endotracheal tube was placed, and an intravenous catheter was inserted and secured in an ear vein.Fentanyl was continuously infused at 10 µg kg −1 hr −1 through the IV catheter.Anesthesia was maintained with 1.5%-3% isoflurane throughout the surgery.Once the surgical preparation was completed, the pig's head was fixed in the stereotactic frame.Heart rate, indirect blood pressure, body temperature, and O 2 -saturation were monitored throughout the procedure and recorded every 15 min.When the heart rate stabilized under anesthesia, a midline cutaneous incision was performed in the scalp, burr holes were drilled in the skull, and the dura was opened.The modified microperfusion probe was secured in the Kopf micromanipulator and slowly inserted into the estimated location of the primary sensory cortex through a burr hole.During the 4th collection epoch 2-3 µl of an aqueous solution of benzylpenicillin (1100 U µl −1 , Penna-G, Sigma) was injected next to the microperfusion probe to induce epileptiform activity [22].

EEG
Electrical signals were collected from the macroelectrodes at 30 kHz using a wireless data acquisition system (Cube-64, NeuraLynx).For analysis, a 400sample median high-pass filter was applied to remove baseline drift.Line noise was suppressed with a notch filter at 60 Hz and relevant higher harmonics, including 120 and 180 Hz.All data analysis and review were performed in MATLAB R2021 A (MathWorks).

Proteomic analysis
Perfusates were collected into the pull pump fluid line at 0.8 µl min −1 .At the end of the epoch the perfusate was recovered into a microfuge tube and held on ice prior to freezing at −80 • C. cerebrospinal fluid (CSF) was collected at the electrode implantation site using a blunt needle and syringe.Bulk tissue interstitial fluid (ISF) was prepared using our established method [23], with slight modification.In brief, 10 mm 3 of cortical tissue was dissected from around the electrode site and placed in ice-cold Hibernate-E media for transport.The tissue was collected by centrifugation at 600 g for 5 min at 4 • C, then resuspended in 300 µl phosphate-buffered saline (PBS) containing 10 µg ml −1 aprotinin, 1 µg ml −1 leupeptin, and 1 mM phenylmethylsulfonyl fluoride.Following homogenization using ten strokes in a cold glass Tenbroeck homogenizer, the cell suspension was centrifuged at 1000 g for 5 min at 4 • C. The supernatant was collected and clarified by centrifugation at 16 000 g for 5 min at 4 • C.
Total protein levels in the microperfusate, CSF, and tissue ISF were measured and the samples were normalized based on concentration.Samples were processed by drying in a vacuum centrifuge concentrator and digested with 200 ng trypsin/Lys-C (Promega) in 20 mM Tris pH 8.2 containing 0.0005% zwittergent 3-16.Equivalent amounts of the acidified peptide digests were analyzed using a Thermo Scientific Exploris 480 mass spectrometer coupled to a Thermo Ultimate 3000 RSLCnano high performance liquid chromatography (HPLC) system.The digest mixtures were loaded onto a Halo C18 2.7 µm EXP stem trap (Optimize Technologies) and chromatography was performed using 0.1% formic acid in both the A solvent (98% water/2% acetonitrile) and B solvent (80% acetonitrile/10% isopropanol/10% water), with a 3% B to 35% B gradient over 120 min at 350 nl min −1 through a PepSep C18 2.4 µm, 100 µm × 40 cm column.The Exploris 480 mass spectrometer was set for data dependent acquisition with a 3 s cycle time between the MS1 survey scan from 340-1600 m z −1 at resolution 120 000 (at 200 m z −1 ), followed by HCD MS/MS scans at resolution 15 000 with a NCE setting of 30 and the isolation width set to 1.2 m z −1 .Ions with charge states of 2-4 were allowed for MSMS and if selected, were placed on an exclusion list for 30 s.The normalized AGC target settings were 100% for the MS1 and 80% for MS2 scans with max ion inject times of 50 ms and 120 ms respectively.The raw data were analyzed with MaxQuant (ver 1.6.17.0) using the Andromeda search engine [24] against the NCBI RefSeq pig database downloaded June 2022 and allowing for oxidized methionine and protein Nterminal acetylation as variable modifications.The protein identifications were filtered at a false discovery rate (FDR) [25] of 1% at the peptide and protein level with a two peptide minimum and reported with accompanying intensity-based absolute quantitation (iBAQ) [26] values.

Data analysis and presentation
Except for the in vivo microperfusion, experiments were repeated at least twice and at least three replicates were used per condition.Findings are reported as mean ±95% confidence interval where appropriate.Data were analyzed and graphed in Prism (GraphPad).Images were processed in Photoshop (Adobe) and schematics were drawn in Illustrator (Adobe) or Solidworks (Dassault Systèmes).FDRs and enrichment scores in figure 5(B) were calculated using WebGestalt [27].The heatmap and Z-scores in figure 5(C) were calculated using Heatmapper [28].

Electrode ablation parameters
Three approaches were tested to evaluate the collective impact of laser power (100% = 3 W average at 343 nm), focal depth, and number of passes on the quality and efficiency of ablation through both the polyurethane electrode wall and the inner polyimide sheath.In the first, 50% laser power (1.5 W) was focused on the exterior surface of the catheter and 20 passes were made to ablate holes 560 µm in diameter (figure 2(A)).In this approach an internal black plastic stylus was not used and the opposing wall of the internal lumen was partially ablated.In the second approach, the laser power was reduced to 35% (1.1 W) and the focus was stepped radially inward by 5 µm per pass for 40 passes (200 µm of (A) A clean hole was ablated using 50% laser power (1.5 W) and 20 beam passes focused on the exterior surface.However, these parameters damaged the opposing wall.(B) A black plastic stylus was introduced into the lumen to prevent collateral damage and the laser power was reduced to 35% (1.1 W).Using 15 µm pitch with 40 passes and a 5 µm step in focal depth for each pass resulted in incomplete ablation at the perimeter of the hole.(C) Reducing the beam pitch to 10 µm and laser power to 30% while increasing the number of passes to 120 with a 5 µm step in focal depth for each pass resulted in an acceptably clean perimeter without collateral damage or heat dissipation artifacts.(A) Fluid flow rate, diameter of the perfusion holes in the depth electrode, and the presence of microdialysis membrane on the fluidics were systematically modified to establish the optimal conditions for analyte recovery (optimal parameters shown in yellow).(B) No-net-flux (NNF) analysis was used to measure recovery of trypan blue dissolved at 60 µM in saline solution (yellow circles) or in a 0.6% agar tissue analog (blue squares).Flow rate was set to 0.8 µl min −1 for this assay.The NNF method introduces different concentrations of analyte into the perfusate (C in ) and measures differential recovery in the outflow (Cout-C in ).When the concentration in the inflow equals the actual concentration in the substrate the differential recovery is zero, equivalent to no-net-flux.The recovery when the inflow concentration is zero represents the maximal recovery potential of the system.Fitting a line to the data yields a slope that is equivalent to the effective percent recovery.Trypan blue recovery from saline was 91%; recovery from agar was 54%.Graphs show a representative experiment.(C) Flow rate was adjusted from 0.8 µl min −1 down to 0.08 µl min −1 and NNF recovery of trypan blue from agar was measured.Recovery plateaued at 97% at 0.16 µl min−1.
the 600 µm wall thickness).While this protocol successfully ablated both the outer wall and the inner polyimide sheath, a large amount of honeycomblike material remained around the perimeter of the hole (figure 2(B)).In the third approach, the laser power was reduced to 30% (0.9 W) and the focus was stepped radially inward by 5 µm per pass for 120 passes (through the full 600 µm wall thickness) (figure 2(C)).In addition, the pitch, defined as the spacing between beam paths during the passes, was reduced from 15 µm in the second approach to 10 µm in the third approach, resulting in complete overlap of beam paths.This produced a clean hole without significant collateral damage.

Molecular recovery properties of the modified microperfusion electrode
As outlined in figure 3(A), a systematic approach was taken to identify the optimal parameters for recovery of analytes using the modified device.In addition to testing the matrix from which the analyte was recovered (saline vs agar), we tested flow rate, perfusion hole size, and presence or absence of perforated microdialysis membrane inside the electrode.We established that recovery of trypan blue using just the modified fluidics was linear from both saline and agar at a fixed flow rate of 0.8 µl min −1 in the absence of the electrode (figure 3(B)).Recovery of the small molecular weight analyte from the tissue surrogate was ∼54%.Using this benchmark, a modified calculation was performed using the ratio of analyte in the perfusate (C out ) to the concentration of analyte in the agar (C in ) at various flow rates (figure 3(C)).We found that recovery plateaued at 97% at 0.16 µl min −1 and this flow rate was used for the remaining tests.Next, trypan blue recovery from agar was assessed using the complete device setup and the effect of ablated port size was determined.We found that increasing the perfusion hole diameter from 400 µm to 560 µm improved trypan blue recovery from 52.5 ± 4.9% to 61.6 ± 8.6% in the presence of a perforated microdialysis membrane and from 62.8 ± 8.4% to 77.5 ± 12.8% in the absence of membrane (i.e.open microperfusion).Finally, we recovered 72.2 ± 19.8% of 70 kDa dextran from agar using 560 µm ports, no dialysis membrane, and a flow rate of 0.16 µl min −1 .

Impact of ablation on electrical and mechanical integrity
The integrity of electrical connections in the ablated electrode was assessed by measuring impedance.A Spencer depth electrode as provided by the manufacturer was measured several times before ablation and immediately before and after laser exposure (table 1).Of note, simply testing the electrode for two days after removal from the sterile packaging resulted in a drop in impedance at all four macroelectrodes, despite cleaning and drying the device between measurements.The introduction of laser ablated pores in the electrode resulted in an impedance drop at all four electrodes, not just the 3rd electrode around which the holes were generated (table 1).This suggested that ablating the electrode induced a pervasive change in the electrical nature of the device.We hypothesized that rather than a structural effect, per se, the ablation changed impedance by allowing the free entry of saline from the testing bath into the lumen of the electrode, thereby changing conduction properties of the wire connectors.To test this, we injected saline into an unablated Spencer electrode.This resulted in a ∼29% drop in impedance, which is comparable to the ∼31% drop across all electrodes in the ablated electrode.
In parallel, impedance was measured in vivo across the two most distal macroelectrodes on modified and unmodified probes using the same parameters.
Immediately following implantation, the impedance on the unmodified electrode was 1950 Ω; after 6 h the impedance was 2098 Ω.The modified electrode impedance was 2043 Ω at implantation and 2015 Ω 6 h later.
Unmodified and modified electrodes were also assessed by electrochemical impedance spectroscopy for changes across a frequency spectrum (1 Hz-100 kHz) (figure 4).Impedance and phase were measured at the time of insertion into a saline bath containing protein (bovine serum albumin) and again after incubation for 6 h in the solution, to model the effects of implantation.The Bode plot (figure 4(A)) indicates that the modified electrode impedance and phase across the frequency sweep at outset was different than the unmodified electrode characteristics.However, across the 10-1000 Hz range that is most relevant to EEG, the modified electrode was comparable to the unmodified device (figure 4(B)).Two-way ANOVA detected no significant differences in impedance at any frequencies (F (150 200) = 13.12,P < 0.0001; Tukey's multiple comparison test between electrodes at each frequency) and the electrode modification only accounted for 4% of the total variance in impedance.Critically, there was no difference at any frequency in the impedance measured on the modified electrode at the end of the incubation relative to the start.
Signal-to-noise performance of unmodified and modified electrodes was assessed in vitro following explantation from the pig brain (figure 4(C)).Discrete responses at stimulation frequencies between 1 and 41 Hz were measured in 10 s epochs and the SNR was determined relative to the background noise.The overall SNR was not impacted by introduction of macroscopic perfusion ports in the modified electrode.Likewise, analysis of the power spectrum of the signal recorded from the modified electrode immediately after implantation into the pig brain versus the power of the signal 3 h later revealed no significant difference in the electrode response (P = 0.6360 between frequency response curves), indicating that the performance of the electrode was not degraded by active and ongoing microperfusion across the macropores (figure 4(D)).
Mechanical integrity was measured using a force gauge positioned coaxially against the probe tip.Buckling strength of the modified electrode was 1.0 N, compared to 3.8 N for the unmodified probe.The reported maximum penetration transient for 0.6% agarose or pig brain is <0.1 N with a continuing penetration force less than 0.03 N [29].In addition, the electrode is implanted during in vivo experiments with an internal stylet inserted in the lumen.Once implanted, the stylet is removed and the microperfusion subassembly is inserted.Overall, these findings indicate that ablation does not compromise the mechanical integrity of the probe to an extent that impacts surgical implantation.

Unbiased proteomic analysis of stimulation-induced changes in cortical microperfusate
The parameters established above were used to collect perfusate from porcine cortex during four ∼60 min epochs (figure 5(A)).Epoch 1 captured baseline proteins and macromolecular complexes (289 ng µl −1 protein; 40 µl).Epoch 2 collected perfusate while the macroelectrode surrounded by ablated perfusion holes was electrically stimulated at 2 Hz (0.7 mA, 200 µs pulse width) (300 ng µl −1 protein; 40 µl).Epoch 3 collected post-stimulation perfusate (287 ng µl −1 protein; 45 µl).Epoch 4 collected fluid while the neighboring cortex was stimulated by injection of benzylpenicillin (236 ng µl −1 protein; 70 µl).After the final perfusate epoch, CSF was collected (101 ng µl −1 protein; 150 µl) and a tissue block from the site of electrode implantation was dissected and processed for ISF isolation (1.18 µg µl −1 protein; 1.75 ml).Mass spectrometry with intensitybased absolute quantification [26] was performed using 3 µg protein from each perfusate sample and from CSF and ISF.Notably, in contrast to the loss of large molecular weight proteins normally associated with microdialysis, using microperfusion we readily recovered the same distribution of proteins by size in the perfusate as observed in CSF and ISF, including 248 proteins larger than 100 kDa (figure 5

(B)).
Intensity-based absolute quantification using a FDR of 1% and a minimum of at least two unique peptides from the expected tryptic library to qualify as 'detected' resulted in identification of 784 proteins in epoch 1, 756 proteins in epoch 2, 850 proteins in epoch 3, 689 proteins in epoch 4, 956 proteins in CSF, and 2855 proteins in the tissue ISF.In addition, 219 proteins were found only in the perfusates and not in the CSF, while 173 proteins were found only in the CSF and not in the perfusates.Likewise, 550 proteins were detected in the perfusates that were not found in the ISF.Finally, 154 proteins were unique to the perfusates and were not detected in either CSF or ISF.Within the perfusate samples, four response profiles were assessed: (a) proteins that were undetected at baseline and became detectable after stimulation; (b) proteins that were detectable at baseline but were increased >1.5-fold after stimulation; (c) proteins that were detected at baseline but become undetectable after stimulation; (d) proteins that were decreased >1.5-fold after stimulation but were still detectable.As shown in table 2 and figure 5(C), we found 180 category 1 proteins, 296 category 2 proteins, 99 category 3 proteins, and 85 category 4 proteins.Gene ontology profiling of category 1 and 2 proteins using over-representation analysis in WebGestalt [27] revealed a robust enrichment in extracellular proteins, matrix proteins, and lipoproteins (figure 5(D)), consistent with the expectation that most factors collected in the perfusate would be soluble proteins.Indeed, of the category 1/2 proteins, 141 are characterized as secreted in the UniProtKB cellular component keyword hierarchy [30].
Several proteins and biological responses stand out as stimulation-induced factors.The inflammatory cytokine IL6 was below detection limits in whole tissue ISF, CSF, and the first two perfusate epochs but was detected at 1.8 × 10 6 units in epoch 3 and 5.5 × 10 6 units in epoch 4. Likewise, the inflammatory protease MMP9 was below detection in ISF, Table 2. Proteins increased or decreased in perfusate after stimulation.Category 1 = proteins that were undetected at baseline and became detectable after stimulation.Category 2 = proteins that were detectable at baseline but were increased >150% after stimulation.Category 3 = proteins that were detected at baseline but become undetectable after stimulation.Category 4 = proteins that were decreased >150% after stimulation but were still detectable.

Category
GenInfo            CSF, and the first two epochs but was then detected at 2.5 × 10 7 units in epoch 3 and 6.5 × 10 7 units in epoch 4. The inflammatory stress proteins S100A8 and S100A9, which together form the calprotectin complex, were both below detection in ISF, CSF, and the first two epochs but were then detected at 1.4 × 10 7 and 2.9 × 10 6 units in epoch 3 and 4.9 × 10 7 and 3.8 × 10 6 units in epoch 4, respectively.Another notable response was observed in the family of complement factor proteins (figure 5(E)).
In the post-stimulation epoch (epoch 3) there was a clear increase in levels of nearly every component of the complement cascade, suggesting that stimulation drove the active release and processing of these factors.

Discussion
We describe the development of a modified brain microperfusion-EEG electrode device that uses offthe-shelf subassemblies.The Ad-Tech Spencer depth electrode, the M Dialysis AB fluidics assembly, and the M Dialysis AB adjustable flow rate pump are each approved for human use and are provided in sterile surgery-ready packaging.We are currently exploring the biocompatibility and sterility validation necessary for human use of the integrated system presented herein.Our findings support the use of this device for performing powerful unbiased proteomic analyses of perfusate collected from the brain during seizures and our near-term goal will be to gather such data from patients undergoing resection surgery for drugresistant epilepsy.The use of existing components allowed us to bypass the lengthy and costly process of developing new devices.We realized that the open lumen present in the Ad-Tech Spencer electrode was appropriately sized to accept the M Dialysis 71 microdialysis membrane and fluidics assembly.We further realized that precision ablation of fluid entry ports into the Spencer depth electrode would permit direct exchange of materials from the ISF surrounding the electrode into the lumen and across the dialysis membrane.In our initial tests we focused on cleanly ablating holes into the electrode body around the second-most distal macroelectrode.We identified the optimal laser power, focusing, and beam pitch parameters yielding clean, reproducible pores and we validated the maintenance of electrical connections through the electrode following ablation.While impedance was altered following ablation, we note that the decay was identical across all four macroelectrodes, an outcome that suggests a systemic change to the function of the electrode rather than a specific process of damage to the wires carrying current.Indeed, the wires emanating from the two macroelectrodes located above the modified macroelectrode do not pass through the polymer sheath at the site of ablation and therefore could not have been damaged by the laser.Importantly, despite the impedance changes associated with fluid access to the lumen of the electrode, electrochemical impedance spectroscopy across a physiologically relevant frequency band revealed no appreciable differences in the modified electrode response profile.Moreover, the modified probe did not exhibit degradation in signal-to-noise performance.
During the course of developing the device we recognized the potential value in removing the microdialysis membrane.As we have discussed elsewhere [1], microperfusion offers significant advantages with regard to an unbiased sampling of the ISF.In addition to removing the confounding and unpredictably stochastic nature of the molecular weight cutoff inherent to microdialysis membranes, running the device in open microperfusion mode offers the ability to capture large molecular weight macromolecules and large membrane-bound structures such as extracellular vesicles.Therefore, in addition to modifying the electrode to permit fluid flow, we also modified the M Dialysis 107 pump to operate in a pushpull manner.In order to prevent net fluid loss from the perfusate into the surrounding tissue, the flow rate through the perfusion space must be maintained and controlled by both a push pump and a pull pump acting simultaneously.The dual pump modification resulted in reproducible recovery of test analytes from fluid and agar substrates, including recovery of more than 70% of large molecular weight dextran from agar during a short collection window.In comparison, recovery of a 66 kDa protein by microdialysis using a 100 kDa membrane was less than 1% in the first hour of collection and recovery rapidly decreased thereafter due to membrane fouling [31].
The unbiased nature of collection in our system is highlighted by the recovery of numerous very large molecular weight proteins and by the unique detection of multiple inflammatory proteins following electrical stimulation at the macroelectrode coincident with microperfusion pores.We used intensitybased absolute quantification of proteins [26] to measure tryptic peptide concentrations in perfusates from each collection epoch and in CSF and ISF collected from the same animal.Albumin is a known protein constituent of CSF that is present at approximately 0.1 mg ml −1 under healthy conditions [32].We measured 2 × 10 10 iBAQ units of albumin in CSF and a remarkably consistent level of albumin in perfusate from all four epochs (1.5 × 10 10 , 1.7 × 10 10 , 1.6 × 10 10 , 1.2 × 10 10 respectively), providing confidence that the perfusates represented ISF.Using this metric as a quantitative unit to calculate estimated molar concentrations of other proteins measured in the samples, we found that IL6, which was undetected in CSF and in the initial baseline epoch, was detec-ted at approximately 0.5 nM in perfusate from the poststimulation epoch.This concentration is con-sistent with the amount of IL6 released by stimulated microglia [33] and suggests that electrical pulsing at 2 Hz during the previous epoch induced released of IL6 that was detectable and quantifiable using our microperfusion electrode device.Moreover, given that IL6 was not detected in bulk ISF collected from the tissue surrounding the probe implantation site, the IL6 detected in epoch 3 was likely restricted to a constrained spatial domain around the stimulated electrode.Our device may therefore provide unprecedented insight into the proteomic response induced within tissue localized to a seizure focus.

Conclusions
Laser ablation of an off-the-shelf Spencer clinical depth electrode coupled with push-pull pump modification of an off-the-shelf perfusion pump system and fluidics resulted in a device that permits coincident detection of electrophysiological changes and unbiased proteomic changes within a highly localized spatial domain in the brain of a living organism.This device recovered high molecular weight proteins with excellent efficiency and revealed stimulationinduced changes in the extracellular proteome that were not detectable in bulk tissue ISF from the stimulation site.Ongoing experiments in large non-human animal models of epilepsy include tuning the flow rate to increase detection of low-abundance proteins (by reducing total volume of the perfusate) or to increase temporal resolution (by shortening the collection epoch), employing additional enrichment steps upstream from the proteomics analysis, and collecting perfusates from non-stimulated control electrodes to discriminate stimulation-induced responses from innate immune responses to the implanted device.Based on the promising findings described herein, our future goal will be to stereotactically implant the microperfusion electrode (figure 6) into the ictogenic zone of patients with drug-resistant epilepsy undergoing assessment for resection surgery and correlate localized EEG changes measured on the macroelectrode at the site of the microperfusion pores with proteomic changes measured by unbiased mass spectrometry of the perfusate.Our current findings strongly support the use of coincident EEG and microperfusion as a powerful tool for uncovering pathogenic mechanisms operating at spatial scales that are inaccessible to current bulk tissue methods.

Figure 1 .
Figure 1.Electrode schematic and modified push-pull pump design.(A) Dimensions of the unmodified Ad-Tech Spencer depth electrode.(B) Location, dimensions, and spatial arrangement of the laser ablated microperfusion access ports in the depth electrode.An example of a modified electrode is shown above the schematic.The boxed region is magnified to reveal port layout.(C) Cross-sectional view through the probe showing the location of the wires, the internal polyimide sheath, and the polyurethane annulus.(D) Schematic of the modification used to convert the rotating lead screw motion of the M Dialysis 107 syringe pump (1) to linear displacement (2) to realize a dual syringe push-pull configuration.(E) Photograph of the modified pump setup with 1 ml syringes attached to the microperfusion fluidics.

Figure 2 .
Figure 2. Impact of laser ablation parameters on microperfusion port quality.Femtosecond pulses of 343 nm laser light were used to ablate the polyurethane outer sheath and polyimide inner sheath of the Spencer depth electrode.The ∼20 µm diameter beam (at focus) was fired under different laser powers, beam passes, and focal steps to ablate a 560 µm diameter hole in the electrode.(A) A clean hole was ablated using 50% laser power (1.5 W) and 20 beam passes focused on the exterior surface.However, these parameters damaged the opposing wall.(B) A black plastic stylus was introduced into the lumen to prevent collateral damage and the laser power was reduced to 35% (1.1 W).Using 15 µm pitch with 40 passes and a 5 µm step in focal depth for each pass resulted in incomplete ablation at the perimeter of the hole.(C) Reducing the beam pitch to 10 µm and laser power to 30% while increasing the number of passes to 120 with a 5 µm step in focal depth for each pass resulted in an acceptably clean perimeter without collateral damage or heat dissipation artifacts.

Figure 3 .
Figure 3. Optimization of device parameters and recovery rate assessment.(A)Fluid flow rate, diameter of the perfusion holes in the depth electrode, and the presence of microdialysis membrane on the fluidics were systematically modified to establish the optimal conditions for analyte recovery (optimal parameters shown in yellow).(B) No-net-flux (NNF) analysis was used to measure recovery of trypan blue dissolved at 60 µM in saline solution (yellow circles) or in a 0.6% agar tissue analog (blue squares).Flow rate was set to 0.8 µl min −1 for this assay.The NNF method introduces different concentrations of analyte into the perfusate (C in ) and measures differential recovery in the outflow (Cout-C in ).When the concentration in the inflow equals the actual concentration in the substrate the differential recovery is zero, equivalent to no-net-flux.The recovery when the inflow concentration is zero represents the maximal recovery potential of the system.Fitting a line to the data yields a slope that is equivalent to the effective percent recovery.Trypan blue recovery from saline was 91%; recovery from agar was 54%.Graphs show a representative experiment.(C) Flow rate was adjusted from 0.8 µl min −1 down to 0.08 µl min −1 and NNF recovery of trypan blue from agar was measured.Recovery plateaued at 97% at 0.16 µl min−1.

Figure 4 .
Figure 4. Electrochemical impedance spectroscopy assessment.(A) Bode plot showing the change in impedance (left y-axis) and phase (right y-axis) across a frequency range of 1 Hz to 100 kHz.The unmodified electrode is represented by a circle; the modified electrode is represented by a triangle.Color coding shows the impedance measured at the start (light red) and end (light blue) of the experiment and the phase measured at the start (light green) and end (light yellow) of the experiment.(B) Impedance and phase data highlighted between 10 and 1000 Hz.The symbol scheme is the same used in A. (C) Signal-to-noise response profile of explanted modified (red circle) and unmodified (blue triangle) electrodes measured in response to mixed signal 2 µA peak-to-peak current injection for 60 s at 30 kHz in a saline bath.Stimulus responses were assessed at prime number frequencies between 1 and 41 Hz.Graph shows mean ± standard deviation of six 10 s epochs at each frequency.(D) In vivo signal power from 100 s of recording collected immediately after implantation ('early') and from 100 s recorded at the end of the 3rd hour of recording ('late').Traces on the left show the power from 0.1 to 1000 Hz in 100 one-second epochs per recording time.Graph on the right shows the average power from 0.1 to 200 Hz across the 100 s recording epochs, with standard deviation shown for each of the frequency bins.Early signal is shown in blue; late signal is shown in red.

Figure 5 .
Figure 5.In vivo microperfusate collection and proteomic analysis.(A) A modified microperfusion-EEG electrode was placed in the cortex of a pig and perfusate and coincident EEG were collected during 4 one-hour epochs.Ten representative 10 s EEG traces collected on the macroelectrode bordered by perfusion holes are shown for each epoch; a single representative 500 msec trace is also shown.Note that the large repetitive signals in epoch 2 correspond with the 2 Hz electrical stimulation driven on the recording macroelectrode (red arrow heads); the 500 msec trace shows the EEG immediately after a 0.7 mA, 200 µs pulse.(B) Perfusate from each epoch (combined analysis of all four samples; red), total tissue ISF (gray), and CSF (yellow) were analyzed using quantitative mass spectrometry and the relative frequency of detectable proteins in 50 kDa bins is shown.(C) Of 1002 proteins detected in either epoch 2 or epoch 3, 476 were upregulated more than 1.5-fold following stimulation (red lines) and 184were downregulated more than 1.5-fold (blue lines).The graph shows the normalized percent expression between epochs, with many of the differentially regulated proteins either changing from undetectable to detectable (bottom left to upper right) or from detectable to undetectable (top left to bottom right).Overall, electrical stimulation led to a robust change in over 65% of the proteins detected in the perfusate.(D) Over-representation analysis of the proteins increased more than 1.5-fold following electrical stimulation (i.e. the red lines in panel C) indicated that secreted extracellular proteins were significantly enriched.The enrichment score was calculated based on the number of upregulated proteins overlapping the membership of each gene ontology cluster (color-coded in purple).The false discovery rate (FDR) was calculated based on expected stochastic overlap.(E) Twenty one complement pathway proteins were detected in the perfusate from at least one epoch.The heat map shows z-scores calculated for each complement factor across the 4 epochs.Nearly all detectable complement proteins were increased in epoch 3 or 4 or both relative to the first two epochs, suggesting profound upregulation of the complement pathway by electrical stimulation.

Figure 6 .
Figure 6.Schematic overview of the microperfusion-EEG dual-sensing probe.(A) Representative image of a modified depth electrode with associated fluidics.(B) Hypothetical trajectory for electrode implantation into the human hippocampus.(C) Hypothetical localization of the microperfusion ports and macroelectrode in the human hippocampus.(D) Schematic of the modified dual-sensing electrode showing the scale of the microperfusion ports to the surrounding tissue.Pyramidal neurons in black, astrocytes in red, microglia in green.(E) Expanded schematic view showing the overall concept and proposed microperfusion sampling of the extracellular fluid compartment around neurons and glia proximal to the macroelectrode.

Table 1 .
The impact of ablation on electrical properties of the Spencer electrode was tested by measuring impedance over time.Each of the four macroelectrodes was measured relative to a reference electrode in a saline bath.