Optimising adsorptive stripping voltammetry; strategies and limitations

The widely employed electroanalytical technique (AdsSV) is critically assessed and evaluated at a wide and diverse range of unmodified and nano-particle modified carbon electrodes using the analyte fipronil as a paradigmatic case. The generic electroanalytical performances of the nano-particle modified electrodes are investigated and compared with the unmodified electrodes revealing similar LOD values and pointing to intrinsic limitations of AdsSV arising from the non-independence of the Faradaic and capacitive signals during the stripping step. Methods for facilitating the adsorption or using different waveforms that may offer a more favourable limit of detection (LOD) at the nanoparticle modified electrodes are suggested and assessed, specifically the use of adsorption onto particles prior to their use for modifying electrodes and also the recently introduced method of semi-circular voltammetry.


Introduction
Stripping voltammetry is a widely adopted strategy in electro-analysis since it offers the opportunity to enhance the sensitivity of the electrochemical method and to reach lower limits of detection than those direct forms of voltammetry where the size of the analytical signal is inevitably limited by the rate of diffusion to the electrode [2] . The constraint created by diffusional rate limitation imposes rather severe generic limitations on direct voltammetry such as linear sweep and cyclic voltammetry as well as, ultimately, the various forms involving pulses and waves [3] . This can be partly moderated by the use of microelectrodes [4] and their arrays [5] but invariably diffusion provides a fundamental rate determining limitation constraining sensitivity.
The attraction of stripping voltammetry is that the analyte is pre-concentrated, including from ultra-dilute solutions below the detection range for direct voltammetry, onto the electrode 3 surface prior to electrochemical analysis in the form of a potential sweep (linear or pulsed) in which a Faradaic current is passed as a result of the oxidation or reduction of the accumulated analyte. By prolonging the period of accumulation the limit of detection (LOD) can in favourable cases be suitably shrunk to meet the stringent demands of certain analytical challenges including the detection of manganese in tea [6] , lead in petrol [7] , and cadmium in blood [8] .
Stripping voltammetry comes in four forms of which anodic and cathodic stripping voltammetry (ASV and CSV) are the most familiar and where a Faradaic reaction is used to realise the pre-concentration step, for example by the reduction of Pb (II) to metallic Pb in the case of ASV [7,9] or its oxidation to PbO2 in CSV [10] . These chemical processes are reversed during a linear potential sweep which either oxidises the Pb or reduces the PbO2. A third type of SV is the recently introduced insertive stripping voltammetry [11] , whilst the topic of this paper is to explore the limitations of adsorptive stripping voltammetry (AdsSV).
In AdsSV, as the name suggests, pre-concentration occurs via adsorption onto the electrode surface. This may take place directly onto an electrode, for example, a mercury surface as in the AdsSV analysis of vanadium [12] or onto a modified electrode in which a convenient and attractive electrode, frequently carbon of some form, is supplemented by the addition of a surface layer designed to enhance adsorption so ideally promoting analysis via AdsSV. The surface modification may take the form of a chemisorbed monolayer, a polymeric layer, or as considered in this paper a layer of bespoke particles. In the case of particle modified electrodes (PMEs) a huge diversity of materials has been used including carbon nanotubes [13] , carbon black [14] , TiO2 nanoparticles [15] and metal (Au, Ag, Pt etc.) nanoparticles [16] . We have shown elsewhere [17] that the requirements for the modification are rather stringent unless the particles form a conducting layer since if the binding of the analyte with the modifying particles is too weak then little signal enhancement is seen, but if the interaction is too strong then the 4 analyte is not released sufficiently quickly to usefully benefit the analysis. Even with optimised absorption the merits of the approach can be limited to less than an order of magnitude improvement in analytical signals 26 .
In the present paper we consider the limitations of AdsSV with electrically conducting particle layers from a different perspective, namely to address the question as to what extent the quantity of material absorbed influences the analytical signal. This is an easily controllable variable in the design of PMEs via simply changing the number (or mass) of particles adsorbed on the electrode surface. On first thoughts it is obvious that the Faradaic signal seen on the stripping potential scan will increase with the amount of material adsorbed and that the latter will enhance the voltammetric signal. However ultimately the limit of detection is controlled by the ratio of the Faradaic analytical signal as compared to the background currents flowing during the potential scan in which the adsorbed analyte is oxidised or reduced. These Faradaic currents scale both with the voltage scan rate and with the amount adsorbed. However the amount adsorbed will reflect the surface area of the particles used for the modification. At the same time as the background currents are essentially capacitive they too will scale with both voltage scan rate and with the total particle surface area. Hence it is not obvious that increasing the number of particles modifying the surface is analytically helpful. It is the purpose of this paper to explore if this prediction reflects analytical reality and also to compare different adsorbing surfaces to see the extent to which the inferences are general.  5 consider surface modified with particles of carbon nanotubes (CNTs), graphene nano-platelets (GNPs) and of carbon black (CB) as well as unmodified surfaces of glassy carbon (GC), basal place graphite (BPPG) and edge plane graphite (EPPG). In the case of particles such as CNTs we complement the AdsSV data with single entity electrochemical measurements [18] to confirm adsorption and the amount of this on the individual particles used in the PME.
As a model target analyte we consider the molecule fipronil (Scheme 1), a phenylpyrazole insecticide used for pest control on crops, household pets and domestic animals [19] .
Fipronil and its metabolites display neurotoxicity [20] and potential genotoxicity [21] towards nontarget animals including humans [22] . The wide use of fipronil has raised concerns because of the residual fipronil present in soil or aquatic ecosystems [20,23] given its threat to human health [24] . The methods reported for the analysis of fipronil are mostly chromatographic [25] , while recently some cheap and rapid techniques have emerged [26] including a few electrochemical methods [26c-h] . The method we evaluate in the present paper employing AdsSV on the aforementioned carbon electrodes, particularly on a CNT modified surface can be seen in principle as an electroanalytical method alternative to the existing analytical methods for fipronil determination. However the intrinsic limitations of the AdsSV method as evaluated in then following indicates a clear lower limit for electroanalytical detection via this approach; clarifying the reasons for this are the aims of this paper.
Graphene nano-platelets (GNPs, 15 µm in width, 6-8 nm in thickness) with a known average area of 297 ± 152 µm 2 (from scanning electron microscopy [1]  All electrochemical experiments were performed at 298K with a standard threeelectrode configuration inside a Faraday cage by using an in-lab built potentiostat. As previously described [27] , the potentiostat was software controlled by Python 3.5 to generate the required potential waveforms and measurements with low noise and a high sampling rate (100 kHz maximum).

Voltammetry on Carbon Macroelectrodes
For voltammetric experiments, several carbon-based macroelectrodes were used variously as the working electrode. A saturated calomel electrode (SCE, BASi Inc., Japan) was used as the reference electrode and a graphite rod as the counter electrode. The macroelectrodes employed as working electrodes were a glassy carbon (GC) electrode with a diameter of 3.0 mm, a basal plane pyrolytic graphite (BPPG) electrode with a diameter of 4.9 mm, an edge plane pyrolytic graphite (EPPG) with a diameter of 4.3 mm and the same glassy carbon 7 electrode modified with carbon black (CB), multiwalled carbon nanotubes (MWCNTs) or graphene nanoplatelets (GNPs).
The GC and EPPG electrodes were polished on the soft lapping pads with alumina of decreasing particle sizes (1.0, 0.3 and 0.05 m) (Buehler, IL, UK), followed by sonication for 10 s in water. The BPPG electrode surface was prepared by polishing on the sand paper of grade P4000 and then repeatedly for several times pressing on and then removing from the sticky tape, 'cellotape' (Henkel AG, Dusseldorf, Germany), before finally washing with acetone [28] . The CB, MWCNTs and GNPs modified GC electrodes were prepared by a dropcast

Nano-impact Experiments
Nano-impact experiments were conducted by measuring the chronoamperometry at a carbon fibre wire microelectrode system using the same reference and counter electrode as above. Note the potentiostat used in this work ensures an accurate conservation of the charge transferred during a particle-impact event [18b] . The carbon fibre wire microelectrode was fabricated from a carbon fibre (diameter 7.0 m, Goodfellow, Cambridge, UK) following the procedure introduced previously [30] . The length of the microelectrode was 1mm. 10 L of MWCNT suspension prepared as described in 2.2 was added to 10 mL of the blank solution or the 0.1 mM FIP solution, followed immediately by a chronoamperometric measurement. The chronoamperograms were recorded for 20 s at varying potentials from 0.8-1.7 V vs SCE.

Results and Discussion
In the following sections, first, adsorptive stripping voltammetry (AdsSV) or, novelly, prior-adsorptive stripping voltammetry (pAdsSV) is applied to examine the electrochemical signals of fipronil (FIP) at six types of carbon electrode surfaces. The adsorption of FIP on glassy carbon (GC), basal plane pyrolytic graphite (BPPG), edge plane pyrolytic graphite (EPPG), multiwalled carbon nanotubes (MWCNTs) and graphene nanoplatelets (GNPs) surfaces are investigated and characterised. Next, the adsorption of FIP on single MWCNTs is investigated via nano-impacts. Finally, in the last section, the electroanalytical performances of AdsSV on these five carbon electrode surfaces are compared. The effects of the types and the amounts of carbon materials on the detection limit are demonstrated, then, in combination with the results obtained in the previous two sections, explanations are offered for the presented 9 merits and limitations of employing AdsSV at the unmodified and nano-particle modified surfaces. Also in this section, electroanalysis of FIP using adsorptive square wave (AdsSWV) and adsorptive semi-circular sweep voltammetry (AdsSCV) is evaluated at optimised MWCNT electrode surfaces.

Electrochemical Behaviour of Fipronil
The voltammetric responses of the FIP in a mixture of 60% pH 8.0 B-R buffer and 40% methanol were first explored at all carbon surfaces of interest. Figure  for EPPG and GNP and 0.89 V vs SCE for MWCNT surface. No FIP related signal was seen on CB modified electrodes. This likely reflects structural differences between CB and the graphitic forms of carbon [31] leading to negligible adsorption on CB where graphitic planes are absent. CB was not investigated further.
Next to further study the adsorptions of FIP onto the electrode surfaces, the effect of scan rate was investigated at the unmodified GC, BPPG and EPPG electrode, as well as the MWCNT and GNP modified GC electrodes using the stripping voltammetry analytical procedures as described above. The results are shown in Figure S2-S6 in the SI. The peak currents are linear with scan rate and the slopes of the corresponding log-log plots were close to 1 at the five surfaces in the low scan rate ranges, suggesting the oxidation processes of FIP are surface controlled. The oxidation is suggested to correspond to a two-electron transfer process generating fipronil-sulfone [26c] (structure shown in Figure S7), which is a wellrecognised oxidative degradation product of FIP [32] .
Having identified the surface-bound nature of the FIP oxidation processes, the adsorptions of FIP on the five carbon surfaces were next characterised by considering the possible approximate applicability of the Langmuir isotherm [33] : where θ is the fractional occupancy of the adsorption sites, K is the equilibrium constant for adsorption and [FIP] is the concentration of FIP. Eq. (1) can be re-written as: where Q is the voltammetrically measured charge transferred during oxidation of the species at the studied concentrations. Qmax is maximum charge transfer at the surface saturated with FIP. Voltammetric responses at the GC, BPPG, EPPG, MWCNT and GNP surfaces were 12 recorded in varying concentrations of FIP solutions using the same procedure as above. Figure   2 shows  The value was estimated using SEM [1] , see Section 5 of SI for the calculation.
14 surface areas of the two modified electrodes characterised by the supplier and an independent SEM study [1] . As can be seen in Table 1, the electro-active surface areas of the unmodified electrodes were approximately 2.3-5.5 times of the geometric areas. Under the aforementioned assumption of monolayer adsorption this suggests a considerable contribution from the roughness of theses surfaces. In fact the roughness factor for GC electrodes is reported as lying in the range 1.7-4.1 [36] High roughness levels of pyrolytic graphite electrode surfaces have also been noted [37] . However for the nano-entity modified electrode surfaces, the electro-active area values indicates likely sub-monolayer adsorption at both GNP and MWCNT surfaces. The low levels of adsorptions are possibly caused by desorption [34a] of FIP and/or the aggregation/agglomeration of the dropcasted MWCNTs leading to the reduced surface accessibility. To further investigate this and to assess the potential of improving the detection sensitivity the single entity electrochemistry of MWCNT was studied as an example in the next section.

Adsorption on Single Carbon Nanotubes: A Nano-impact Study
The oxidation of FIP on individual MWCNTs was investigated using the nano-impact  Figure S9). The average charge passed per spike were obtained from averaging the integration of each singlepeaked, sharp shaped spike for each potential (Figure 3(B)). The 'quasi-step' like impacts and the multi-peaked spikes were excluded since they may likely be the results of simultaneous MWCNTs as has been noted in other solvents [38] . As revealed by Figure 3(B), the average charges measured for the spikes detected in the MWCNTs-FIP suspension are very significantly larger than those detected in the blank MWNCT suspension and increase to a limiting value with the increase of potential particularly in the range of 1.1-1.5 V SCE consistent with the Faradaic oxidation of FIP. As established previously the spikes may result from the Faradaic and/or the capacitive charge transfer process occurring during the collision between the nano-particles and microelectrode [1, 18b] . The Faradaic charge transferred during the impact timescale increases substantially when an over-potential enabling the redox reaction to proceed is applied, which is consistent with the case observed for MWCNTs-FIP suspension.
In contrast the small spikes detected in the blank MWCNT suspension show a small and potential insensitive magnitude over the whole potential range, as shown in Figure 3(B), indicating the sole contribution is from the capacitive charge transfer at electrode-electrolyte interface taking place when the applied potential deviates from the potential of zero charge [1,39] .
Chronoamperograms at 1.6 V vs SCE were analysed where the average charge per Faradaic spike reaches a plateau (see Figure 3(B)) and the Faradaic oxidation of FIP was thought complete. A total number of 106 single-peaked, sharp-shaped spikes were collected from eight independent chronoamperograms. As the capacitative charge transferred is negligible, the measured charge transferred per spike can be approximated to the charge transferred in the electro-oxidation of the FIP at a single MWCNT. Figure 3

Analytical Performance of Different Carbon Electrode Surfaces
First the analytical responses of FIP at the particle modified electrodes (PME) and the unmodified GC, BPPG and EPPG electrodes were compared.  area by [40] : where ∆ is the difference between the measured forward and backward current, A is the effective surface area, C is the capacitance per unit area and v is the scan rate. Using Eq. (3) the capacitances can be determined from the cyclic voltammograms measured in the blank solution of 60% B-R buffer and 40% methanol at GC, BPPG, EPPG, MWCNTs and GNPs modified GC electrode. For each electrode, varying scan rates were applied and linear plots of ∆ (read at the potential of 0.5 V vs SCE) versus v were obtained, from the slope of the plots the capacitances at 0.5 V vs SCE can be calculated (see Figure S10). Table 2 lists the overall capacitance and the specific capacitance of each electrode estimated using the electro-active surface areas and characterised surface areas shown in Table 1 for unmodified electrodes and PMEs, respectively. It can be seen in Table 2 that the overall capacitances which are controlled by the surface areas are much smaller for unmodified electrodes in comparison to the PMEs.
The specific capacitances of all studied carbon surfaces lie in the similar ranges of 5.3-64.3 F/cm 2 , close to the typical double-layer capacitance values [40][41] . Note that unlike true capacitors, the double layer capacitance is often dependent on the potentials applied [40] .    Table 2. Summary of capacitances, specific capacitances, the lowest detectable concentrations and the signal to background ratios in the voltammograms of the lowest detectable concentration of the studied electrode surfaces. The ranges shown for the specific capacitance reflect both the ranges of the surface areas and the standard deviations of the estimated capacitances. 21 unmodified electrodes and the MWCNTs modified GC electrode are similar and are larger than the GNPs modified GC electrode. This pattern was also observed for LOD and hence it can be concluded that the LOD is dependent on the signal-background ratio rather than solely the magnitude of signal. Thus despite the fact that larger signals were seen at the PMEs with larger surface areas, the LODs of AdsSV are also affected by proportionally larger capacitances presented and are intrinsically limited by an upper limit of monolayer adsorption, as discussed in section 3.2.
Having compared the different types of carbon materials, we next focus on the PMEs and investigate the effect of the amounts of nano-particles used for modifying the electrode. Finally the possibility of further improving the limit and sensitivity of detection by employing the voltammetric techniques with different potential waveforms was examined at the optimised MWCNTs modified GC electrode surface where a relatively favourable analytical performance using AdsSV was observed. The commonly used adsorptive square wave voltammetry, AdsSWV, was first attempted however, the signal was lost in the voltammograms (see Figure S12 for the example voltammogram) since the backward and forward signals were found to cancel out. The forward scans was recorded in varying concentrations of FIP solutions and the LOD was found to be 4 M ( Figure S13).
Next we turned to the recently advocated semi-circular potential sweep voltammetry (SCV) but, novelly, with a pre-concentration step applied, AdsSCV. The SCV technique was reported in our previous work [42] . Briefly, the scan rate varies over the potential window generated by a semi-circular potential wave function. The midpoint of the potential window (Emid) can be adjusted to approximate the redox potential where the scan rate reaches a near infinite value and the present signals are greatly amplified. Here the Emid was optimised as 0.75 V vs SCE ( Figure S14) and the resulting adsorptive semi-circular voltammetric response of 0.1 mM FIP was compared with the adsorptive linear sweep voltammetric response in Figure 6 (A). The voltammograms of varying concentrations of FIP solutions ranging from 0 to 100 M obtained using AdsSCV are shown in Figure 6 (B) and the linear detection range was found to be 1-32 M. The LOD obtained using AdsSCV was 1 M, which is somewhat but not substantially lower in comparison to using the other two conventional electroanalytical techniques (4 M for the forward scan of AdsSWV and 2.5 M for the linear sweep AdsSV). 24

Conclusions
This work demonstrates the limitations and optimising strategies of the electroanalytical detection using adsorptive stripping voltammetry for the model molecule FIP so as to give generic insights. The adsorption of FIP at the unmodified electrodes (GC, BPPG EPPG electrode) and the PMEs (MWCNTs and GNPs modified GC electrode) was evidenced and characterised using the voltammetric studies and via nano-impact experiments at the single MWCNT surface. The adsorptions were identified to be no more than mono-layer for all the studied electrode surfaces. This intrinsically limits the LODs reflected by the signal to background ratios. The PMEs shows no obvious advantage in comparison to the unmodified electrodes in terms of improving the LOD despite larger signals are detected since an increased area for adsorption is offset by an increased capacitive charging signal. Nonetheless two strategies were developed for PMEs to enhance the analytical performance, specifically the use of pAdsSV technique facilitating the adsorption of FIP at GNPs modified GC electrode and the use of AdsSCV technique at MWCNTs modified GC electrode which selectively amplifies the signal and results in the low LOD.

Table of Contents Graphics
The adsorptions of FIP at the studied carbon electrodes are identified to be no more than monolayer, which intrinsically limits the electroanalytical performances of the nano-particles modified electrode using adsorptive stripping voltammetry despite the significantly larger signals are detected.