A Framework for Benchmarking Emerging FSCV Neurochemical Sensors

Fast‐scan cyclic voltammetry (FSCV) in combination with carbon fiber (CF) sensors has been a longstanding tool for studying neurochemical signaling in the brain. However, further progress toward brain mapping requires improvements in sensitivity, chemical selectivity, and channel count. Addressing these critical needs has been an impetus behind the search for alternative carbon materials that could outperform CF. Given the significant research activities, an objective assessment of key sensor performance metrics is crucial for guiding the material discovery. Here, a framework for assessing the sensitivity and selectivity performance of FSCV sensors is put forth and visualizing the comparison in the form of a performance plot. An example of this analysis by using nanographitic carbon as a model material is provided. It is hoped that the proposed framework can help accelerate the research progress by providing an improved guideline for evaluating emerging FSCV sensors.


Introduction
Fast-scan cyclic voltammetry (FSCV) using carbon fiber (CF) electrodes is a well-established tool for monitoring the fast dynamics of neurochemical signaling in the brain.It has been extensively utilized for studying dopamine (DA) release in the striatum (e.g., [1][2][3][4][5][6] ).However, applying FSCV to other brain regions, which generally have lower neurotransmitter concentrations and a higher number of potential interferents, presents a challenge because of the sensitivity and selectivity limitations of CF electrodes. [7]Hence, significant research is ongoing to achieve new FSCV sensors which could enhance these performance metrics.
A critical need for advancing the FSCV capability is to discover a carbon material that outperforms conventional CF.Major research directions include modifying CF using nanomaterials [8][9][10][11] DOI: 10.1002/apxr.[14][15][16][17][18][19][20][21][22] The first step in this material discovery process is to produce sensors from a candidate carbon material and evaluate the sensor performance in vitro.However, before sensors from alternative carbon materials can be further considered for experiments in the brain, a quantitative comparison of their key performance metrics against those of CF will be insightful.Hence, consistent reporting of key performance metrics of FSCV sensors will be an important step in achieving an objective assessment of different forms of carbon.
Furthermore, visualizing the performance comparison of emerging FSCV sensors in the form of a summary plot will be an effective approach for benchmarking a candidate carbon material against other alternatives.The development of performance plots has been instrumental in guiding the materials discovery efforts for different technologies such as emerging field-effect transistors [23] and supercapacitors. [24]However, a similar framework for visualizing the performance comparison of FSCV sensors using a performance plot is still lacking.
In this work, we report a framework for reporting the key characteristics of FSCV carbon sensors, which allows an objective assessment of a candidate sensor material from an application standpoint, i.e., neurochemical sensing.Specifically, we use nanographitic (NG) carbon, a state-of-the-art carbon material, as a model material, and DA as an example neurochemical to illustrate our proposed guidelines for reporting and assessing the performance of the resulting sensor in terms of sensitivity and selectivity.

FSCV Sensor Characteristics
Figure 1a shows the schematic illustration of a planar NG sensor in an FSCV experiment configuration.Our NG sensor had a geometric area of 600 μm 2 .The details of NG sensors can be found in our previous studies. [20,21]An FSCV voltage waveform was applied to the sensor and the resulting current was recorded and subsequently analyzed for assessing the sensor performance in terms of sensitivity, selectivity, and limit of detection (LoD).
Figure 1b shows the FSCV waveform which we used for evaluating the performance metrics of the NG sensor.As we have shown in our previous studies, [21,25] our NG sensors using this waveform simultaneously provide high sensitivity and selectivity.We remark that this waveform has different parameters (scan rate , anodic and cathodic potential limits) than the optimal waveform for operating sensors made of CF (e.g., see [26] ).Since the goal of the ongoing research efforts is to discover an alternative carbon material that outperforms CF, it is reasonable that the characteristics of a candidate sensor to be reported under its optimal operating condition.While the potential range and  of the waveform can be chosen to achieve optimal sensitivity and selectivity, all experiments must be performed at the standard repetition frequency of 10 Hz.This consistent reporting allows the performance comparison of sensors at the desired temporal resolution of 100 ms.
When no DA is present, a background current is generated (i bg ), as shown in Figure 1c.This current consists of two main components.The first component is an electrochemical current (i quinone ) due to the redox of quinone-like species.The other component arises from the charge and discharge of the double-layer capacitance (C dl ) of the carbon material and is given by i dl = C dl .Since C dl and  are constant, the flat region of i bg represents the steady-state i dl , (marked in Figure 1c).To estimate C dl of the sensor accurately, care must be taken to remove the effect of parasitic capacitance on the measured background current.With the knowledge of  and i dl , this measurement gives an estimated C dl of 495 pF for this NG sensor.
Reporting i bg characteristics also provide valuable insights into the capacitive nature of the carbon sensors and their charging dynamics.The closely rectangular shape of the i bg in Figure 1c indicates the nearly ideal capacitive behavior of the NG carbon. [27]It also indicates a low equivalent series resistance (R s ) of the sensor, as we will quantify it later.If the i bg characteristic of a candidate carbon material deviates from a rectangular shape, it warrants further examination of the material to reveal the origin of this behavior.For example, past research attributes the i bg shape characteristics of CF sensors to the resistive-capacitive nature of the material. [28,29]fter adding the DA solution, the molecules undergo oxidation and reduction (redox) reactions on the sensor surface that involve exchanging electrons with the sensor.The inset in Figure 1a shows the redox reaction of DA.The electron transfer between the sensor and the target molecules adsorbed on the sensor surface results in a redox current (i redox ) that is superimposed on i bg (see Figure 1c).Figure 1d,e show the background-subtracted current (i redox ) against the FSCV potential at different DA concentrations.The curves in these plots are called the cyclic voltammogram (CV) and serve as the fingerprint for identifying a target analyte.

Absolute Sensitivity
The magnitude of i redox depends on the number of molecules that undergo the redox reaction, and hence it can be the predictor of the molecule concentration.From the redox reaction in the inset in Figure 1a, every DA molecule that gets oxidized generates two electrons, which contribute to i redox .Hence, i redox is expected to be zero in the absence of DA and increases linearly with the number of DA molecules (i.e., concentration).However, this linear relationship cannot go on indefinitely.As the concentration exceeds some critical value, a sensor will no longer have enough electrochemically active sites to support the redox reaction of every molecule.As a result, the sensor response begins to deviate from the linear relationship at those high concentrations.
For applications in brain studies, the detection and quantification of neurochemical molecules at low concentrations (i.e., nM range) are of interest. [30,31]From a practical standpoint, it is desired to describe the sensor response in this operating range using a linear model because it simplifies the quantification of an unknown molecule concentration from the measured i redox .In this region, a calibration curve (i.e., a univariant model) can be obtained by a linear fit with a zero intercept to the measured i redox data plotted against the analyte concentration.Figure 2a shows the calibration curve of the NG sensor for DA.The slope of this linear curve represents the absolute sensitivity (S) of the NG sensor in units of nA per unit concentration (e.g., μM), i.e., S = 28.3nA μM −1 .A higher sensitivity indicates the ability of a sensor to generate a stronger signal (i redox ) at a given molecule concentration.Hence, sensors with a higher sensitivity are desirable for detecting a target analyte at low concentrations.
The standard calibration curves are also the basis for quantifying a target analyte through univariant or multivariant analysis. [7,32]Therefore, the quality of the linear fit (measured by R 2 ) should be also reported as an indicator for assessing a sensor's ability to yield a reliable quantification of a target analyte.The calibration curve of the NG sensor gives an R 2 of 0.999, indicating the good quality of the calibration curve for quantification analysis.

Background-Normalized Sensitivity
Having established the best practices for analyzing and reporting the sensitivity and calibration curve of an FSCV sensor, the next step is to compare the sensitivity against other sensors, including CF.This requires the implementation of an objective assessment procedure that allows the identification of a promising carbon material or production method that can yield sensors with higher sensitivity than CF.This assessment procedure should ideally reveal the sensitivity shaped by the inherent properties of the material that control electron transfer kinetics and molecule adsorption on the sensor.
The FSCV sensitivity increases linearly with the sensor area and .For example, the NG sensor in Figure 2a would provide two times higher S at 400 V s −1 (i.e., ≈56 nA μM −1 ).Therefore, a desired analysis should remove the effect of these extrinsic factors (i.e., the sensor area and ).Recall that the magnitude of i bg also scales linearly with the sensor area and .Hence, normalizing the sensitivity to the background (SBR) can remove the effects of these two factors that are extrinsic to the sensor material.By removing the effects of the extrinsic factors, SBR provides insight into how the inherent properties of a sensor material shape the sensitivity.Furthermore, as we will show later, SBR could also be an excellent proxy for LOD of an FSCV sensor.
The SBR has units of nA.μM −1 .nA−1 which can be interpreted as the sensitivity of a sensor when the sensor produces i bg of 1 nA, for example by downscaling its area and while preserving its material properties.Following the above-described procedure, the NG sensor in this study provides SBR of 0.28 nA.μM −1 .nA−1 , which is up to 7 times higher than the typical SBR of CF (≈0.04 nA.μM −1 .nA−1 [15] ).

CV Shape Characteristics for Sensitivity Assessment
The sensitivity and selectivity of electrochemical sensors can be predicted qualitatively by the shape and separation between CV peaks. [33]The separation (ΔE p ) and width of the CV peaks are qualitative predictors of the apparent electron transfer kinetics.A smaller ΔE p and sharper CV peaks generally suggest energetically favorable electron transfer kinetics, indicating the efficacy of carbon material in shaping the sensitivity.Therefore, analyzing and reporting ΔE p provides an additional measurable metric for benchmarking different carbon sensors.We measured a ΔE p of 310 mV from the CV of the NG sensor in Figure 1.
A given FSCV sensor will produce a larger ΔE p with increasing .For example, our NG sensors typically produce a ΔE p of 390±10 mV at 400 V s −1 and 310±10 mV at 200 V s −1 .Therefore, the effect of  on ΔE p must be considered when using ΔE p as a qualitative metric for comparing the relative electron transfer kinetics of different sensor materials.

Sensitivity Performance Benchmarking Plot
In the previous sections, we discussed the merits of the SBR and ΔE p as both the quantitative and qualitative metrics for assessing the inherent properties of carbon material in shaping sensitivity.Hence by plotting the SBR against ΔE p , we introduce the performance plot that can provide the visual comparison of the FSCV sensitivity among different carbon sensors.10][11][12][13][14][15][16][17][18][19][20][21] Note that the list of chosen carbon materials in this plot is not exhaustive.We used the data from those studies that provided sufficient details for computing the SBR and ΔE p .
Based on our previous discussions on the merits of SBR ratio and ΔE p , a desirable carbon sensor will provide a higher SBR and a small ΔE p , i.e., the upper left corner of the performance plot.Therefore, our proposed performance plot clearly illustrates the prospects of carbon nanotube yarn [15][16][17] and graphitic carbon materials [18][19][20][21] for enhancing the sensitivity and selectivity of FSCV sensors beyond what conventional CF materials can provide.[36][37][38] This plot also illustrates the improved performance of our NG carbon materials compared to existing carbon material candidates.This enhanced performance is the result of our approach to engineering the structure of NG materials [20] and optimizing the FSCV waveform. [21]

SBR as a Proxy for LoD
The naturally occurring concentration of neurotransmitters is in the low nM range.Hence, the absolute sensitivity of a sensor is used for estimating the sensor's LoD to illustrate the prospects of detecting low nM concentrations of an analyte.The LoD is calculated using the following expression: where the pre-factor 3 is the signal-to-noise ratio (SNR) criterion for resolving the signal of interest, i noise is the total rms current noise, and S is the absolute sensitivity.Following the procedures of the previous FSCV studies, [39,40] we evaluated i noise of our measurements to be ≈70 pA rms .With this information, we estimated the LOD of our NG sensor to be ≈7 nM.Indeed, this method of calculating LoD gives valuable insight about the prospects of a given sensor for detecting neurotransmitters in low nM range.However, the applicability of LoD as a metric for guiding the sensor materials discovery calls for further examination.By re-writing Equation (1) in terms of SBR and expanding the total current noise, we obtain: where i sen , i e , and i en are the rms current noise contributions by the sensor, electronic detection circuitry, and environment, re-spectively.The latter two sources of noise are independent of the materials properties, and yet they could influence the estimation of LoD.i en can be made negligible by implementing proper noise mitigation strategies. [41]However, recent studies [42,43] illustrate that i e is more dominant than i sen , and thus it shapes the total (i.e., measured) current noise i noise .With this knowledge of i noise , Equation (2) predicts that, for sensors with similar SBR, LoD would improve with increasing i bg .
We examined this prediction of Equation ( 2) by plotting (see Figure 2c) the reported LoD of the previous sensor studies listed in Figure 2b.In this plot, we have also included two LoD curves as guides.We calculated these curves by using Equation (2), assuming i noise of 40 pA rms (solid line) and 200 pA rms (dashed line), and applying a fixed SBR of 0.05 nA.μM −1 .nA−1 .This value of SBR corresponds to the median value of SBR for the previously reported sensors in Figure 2b.The general trend of previously reported LoD data in Figure 2c is consistent with the above prediction of Equation ( 2).This observation is not surprising because for sensors with similar SBR, increasing i bg results in a proportional increase of the absolute sensitivity, and thereby the improvement of LoD.
Let us re-examine Equation (2) for sensors with equal i bg that are measured under identical experimental conditions (i.e., identical i e and negligible i en ).Given the dominance of i e in shaping the total noise of the existing FSCV sensors, it is therefore expected that sensors with higher SBR and identical i bg would provide an improved LoD.Therefore, SBR is an excellent predictor for LoD.

Inherent LoD
The main objective of our study is to provide guidelines for benchmarking the key sensor characteristics that are shaped by the inherent material properties of the sensor.Therefore, we introduce inherent LoD for evaluating the limit of detection of a sensor when the noise contributions due to the external sources of noise (i.e., those of the electronic readout circuit and environment) are negligible.Hence, the inherent LoD metric could be insightful in providing information about the best achievable limit of detection of a sensor.
The estimation of inherent LoD requires the extraction of the sensor noise.However, experimental extraction of the sensor noise is not easy because the noise measurements always include the contributions from the readout electronics and the environment.Removing the contributions from these extrinsic sources of noise in experiments could be highly involved, e.g., implementation of low-noise grounding and shielding procedures. [41]Alternatively, the inherent noise of an electrochemical sensor can be estimated through calculations by using the measurable electrical properties of a sensor. [41,44]By eliminating the effect of external sources of noise on estimating a sensor's noise performance, this approach could provide a preferable way of comparing inherent LoD for the emerging FSCV sensors produced at different research laboratories.
For FSCV sensors, the simplified electrical model of the noise in Figure 3a would be sufficient for obtaining reasonable estimates of the sensor's noise. [42,43]This model predicts the sensor's noise from the thermal noise arising from R s and the shot noise arising from i bg produced by C dl (i.e., i dl = C dl ).By fitting the i bg characteristic shaped by C dl (see Figure 3b), we extracted R s and C dl for our NG sensor to be 170 kΩ and 495 pF.With the knowledge of these circuit parameters, the inherent current noise of an FSCV sensor can be computed, as we explain below.
The current noise spectrum (S i ) illustrates the inherent thermal and shot noise power of an FSCV sensor at different frequencies.From the electrical model in Figure 3a, the corresponding S i of the thermal noise (S i,th ) and shot noise (S i,sh ) can be evaluated at the output node of the sensor (i.e., node "X" in Figure 3a) and written as follows: In Equation ( 3) and ( 4), k B , T, q, and f are the Boltzmann constant, temperature, elementary charge, and frequency.Figure 3c illustrates S i,th and S i,sh of the NG sensor plotted against the frequency.The yellow panel marks the typical measurement bandwidth of FSCV (BW = 5 kHz).
The noise spectra in Figure 3c illustrate the contributions of the thermal and shot noise to the total current noise at different frequencies.Integrating S i,th and S i,sh (Equation 3 and 4) over the FSCV measurement BW gives the root-mean-square (rms) of the thermal and shot current noise (i th and i sh ), respectively.In Figure 3d,e, we plotted the calculated i th and i sh as a function of different FSCV sensor parameters (i.e., C dl , R s , and i dl ).These calculations gave i th and i sh of 16 and 8.5 pA rms , respectively, for the NG sensor used in this study (marked with the symbols in Figure 3d,e).Assuming that these two sources of noise are uncorrelated, the total noise current of the NG sensor is 18.1 pA rms .Combined with its sensitivity in Figure 2a, this NG sensor (i bg = 99 nA) provides a calculated inherent LoD of 1.9 nM for the SNR criterion of 3.This inherent LoD level is encouraging for pursuing the application of NG sensors in future studies in the brain requiring the detection and quantification of neurochemicals at low nM concentrations.

Selectivity Performance Benchmarking Plot
So far, we introduced a performance plot for benchmarking sensitivity, which is a crucial metric for enabling reliable detection of neurotransmitters at low concentrations.Another equally important sensor performance metric is selectivity since brain is a complex chemical environment where multiple neurotransmitters could be present simultaneously.In this section, we introduce a selectivity performance plot by employing a two-step procedure.
The first step is to compute the extent of overlap between the normalized CVs of individual target chemical species.The oxidation and reduction peak potentials in a CV of an analyte provide information about the identity of the analyte.In experiments where different chemical species are present, significant overlap between their oxidation and reduction peaks could limit the ability of a sensor to distinguish those chemical species.We propose the use of accumulative mutual overlap (MO) to quantify the extent of overlap between the normalized CVs of target molecules.The computation of MO involves applying Gaussian fits to the oxidation and reduction peaks of the two target analytes and then evaluating the accumulative mutual overlap between the fitted curves.Figure 4a shows the representative normalized CVs of DA, norepinephrine (NE), and serotonin (5-HT).With this analysis, we computed an accumulative mutual overlap of ≈73.5% between DA and NE, and an accumulative mutual overlap of ≈8% between DA and 5-HT.The large accumulative mutual overlap between DA and NE in Figure 4a indicates the low selectivity of the sensor to reliably distinguish between these two analytes in a complex chemical environment.We also generate a complementary metric (as the second step) for assessing the selectivity.Our approach uses principal component analysis (PCA) method, which has been employed routinely in previous studies for analyzing in vivo FSCV signals. [32,45,46]pecifically, we aim at constructing the score plot for the target analytes (e.g., here DA and 5-HT) by employing the method to the in vitro standards constructed from known concentrations of those analytes.In this plot, we will then extract the angle between the score vectors of the two analytes as the new metric of selectivity.Nearly orthogonal score vectors is a measure of high selectivity, whereas small angles indicate the difficulty of the PCA model to reliably distinguish between the two signals.
Figure 4b shows the score plot for DA and 5-HT obtained from a representative NG sensor.The angle between the DA and 5-HT score vectors is ≈73.5°,indicating the prospects of the NG sensors for the selective detection and quantification of DA and 5-HT in mixtures.A detailed study to analyze and verify the DA and 5-HT selectivity using NG carbon sensors is beyond the scope of this work and will be reported elsewhere.However, we remark that the angle between the score vectors () in the score plot is an excellent complementary metric to the calculated mutual overlap for assessing the selectivity of different FSCV carbon sensors.
Lastly, we generated a new performance plot for benchmarking the selectivity of FSCV carbon sensors, as shown in Figure 4c.The plot presented here will clearly illustrate how well different FSCV carbon sensors can distinguish between two target analytes (here, DA and 5-HT).A similar plot can be constructed for different pairs of target analytes.

Conclusion
In summary, we described a framework for reporting and benchmarking the emerging FSCV carbon sensors.We discussed the merits of using the SBR as a fair metric for quantitative comparison of FSCV sensitivity among emerging sensors.Specifically, SBR represents the expected sensitivity of different sensors if they would produce identical levels of i bg (e.g., by adjusting the sensor area or ).By plotting the SBR against their corresponding ΔE p , we introduced a performance plot for sensitivity.We demonstrated the effectiveness of this performance plot in revealing the potentials of sp 2 -hybridized carbon materials for boosting the sensitivity of future FSCV sensors.We also demonstrated that SBR can be an excellent proxy for LoD when comparing sensors with identical i bg and assuming measurements under identical conditions.Lastly, we proposed a two-step procedure for generating a selectivity performance plot.The selectivity performance plot is constructed using two quantitative metrics: 1) accumulative mutual overlap between the CVs for a pair of target analytes, and 2) the angle between the score vectors of the target analytes obtained from the principal component analysis method.Given the significant research on FSCV sensors, we hope that our proposed benchmarking approach for sensitivity and selectivity can benefit the progress of the materials discovery process by improving the material assessment protocols.

Figure 1 .
Figure 1.FSCV sensor characteristics.a) Schematic illustration of an NG sensor in an FSCV measurement configuration.DA oxidation during the voltage ramp up produces dopamine-o-quinone (DOQ), which is reduced back to DA during the voltage ramp down.b) The N-shape waveform which we use with our NG sensors for optimal sensitivity and selectivity.c) Output current signals generated by the NG sensor without (black curve) and with DA (blue curve) in the 1X phosphate-buffered saline solution.d,e) The CVs of DA at different concentrations from 1 μM down to 50 nM.

Figure 2 .
Figure 2. FSCV sensitivity analysis.a) The calibration curve of the NG carbon of Figure 1.The dashed line illustrates the linear fit with a zero intercept.b) Our proposed in vitro performance plot for assessing the background-normalized FSCV sensitivity carbon sensors.FSCV sensors with higher SBR and smaller ΔEp are favorable.c) The reported LoD of existing carbon sensors plotted against their corresponding ibg.The desired corner is marked with a blue box in panel c).The symbols in panels b) and c) are consistent and the [Numbers] denote citations.The two curves in panel c) are plotted as guides and were calculated using Equation (2) by assuming SBR of 0.05 nA.μM −1 .nA−1 and a total noise of 40 pA rms (solid curve) and 200 pA rms (dashed curve).

Figure 3 .
Figure 3. Calculated inherent LoD of FSCV sensors.a) The equivalent electrical model of an FSCV sensor, showing its primary sources of noise.b) The procedure for extracting Rs and Cdl by fitting the ibg characteristics.c) The current noise spectra of the shot noise (top) and thermal noise (bottom) of the NG sensor.d) i th and e) i sh calculated by integrating Equation (1) and (2).The symbol shows the corresponding rms current noise of the NG sensor.

Figure 4 .
Figure 4. FSCV selectivity analysis.a) The representative CVs of DA, NE, and 5-HT measured with an NG sensor.The accumulative mutual overlap was used as a metric to quantify the extent of overlap between two individual CVs.b) The score plot of DA and 5-HT, obtained from PCA analysis of in vitro DA and 5-HT CVs.c) The proposed selectivity performance plot, obtained by plotting the angle between the score vectors of the target analytes (here, DA and 5-HT) against their corresponding accumulative mutual overlap (MO).