Validation of electrochemical sensor for determination of manganese in drinking water

Abstract

Manganese (Mn) is an essential nutrient for metabolic functions, yet excessive exposure can lead to neurological disease in adults and development deficits in children. Drinking water represents one of the routes of excessive Mn exposure. Both natural enrichment from rocks and soil, and man-made contamination can pollute groundwater that supplies drinking water for a substantial fraction of the U.S. population. Conventional methods for Mn monitoring in drinking water are costly and involve long turn-around time. Recent advancements in electrochemical sensing, however, have led to development of miniature sensors for Mn determination. These sensors rely on a cathodic stripping voltammetry electroanalytical technique on a miniaturized platinum working electrode. In this study, we validate these electrochemical sensors for the determination of Mn concentrations in drinking water against the standard method using inductively coupled plasma mass spectrometry (ICP-MS). Drinking water samples (n = 78) in the 0.03 ppb to 5.3 ppm range were analyzed. Comparisons with ICP-MS yielded 100% agreement, ~70% accuracy and ~91% precision. We envision the use of our system for rapid and inexpensive point-of-use identification of Mn levels in drinking water, which is especially valuable for frequent monitoring where contamination is present.

Graphical Abstract

1. Introduction

Manganese (Mn) is essential to human health at trace levels, playing a critical role in metabolism, growth, and development, yet causes adverse health effects in excess.^1,2 Health concerns are related to neural impairments, including neurodegenerative diseases in adults (e.g., ‘manganism’) and neurodevelopmental effects in children^3–5. Mn is introduced to the environment by human activities, such as from disposed lithium-ion batteries ⁶, agricultural fungicides ⁷, steel production⁸ and gasoline additives⁹, as well as by erosion of naturally-occurring Mn enriched rocks and soils. This directly contaminates both surface and groundwater.^10,11 Through public and private wells, groundwater supplies drinking water for nearly half of the total U.S. population, and most of the rural U.S. population.¹² Moreover, groundwater is heavily used in agriculture to irrigate crops.¹² Ingestion through food and drinking water are the main routes of Mn exposure for the general population. Although there are currently no enforceable regulations for Mn in drinking water, the US EPA has set the Secondary Maximum Contaminant Level (SMCL), a guideline, at 50 ppb.¹⁰ Thus, regular monitoring of Mn in water is important,¹³ but is typically only done at the municipal water sources or processing plants rather than at the point of use (i.e., in residential tap water).

The analytical laboratory approaches used to measure trace levels of Mn are based on either spectrometry (inductively-coupled plasma optical emission spectrometry (ICP-OES)¹⁴ and inductively-coupled plasma mass spectrometry (ICP-MS)^15,16) or spectroscopy (atomic absorption spectroscopy (AAS)^17,18). These techniques are widely used for environmental and water quality applications and are characterized by high accuracy, sensitivity, and the low limit of quantification of ~ 0.1 ppb.^19,20 For example, Al-Qutob et al.²¹ demonstrated detection of Mn using ICP-MS in ground water of South-West Bank/Palestine, where ground water is the main source of drinking water for the population. Further, Verma et al.²² applied AAS to assess Mn contamination of ground water from a thermal power plant in the Jahhnsi region of India. Despite strengths, these methods require bulky and expensive instrumentation as well as highly trained personnel. Ultimately, only centralized laboratories are able to house and maintain these instruments. Furthermore, turn-around time for analysis can stretch to months or longer if samples are batched to save on shipping and analysis costs. Thus, these conventional approaches are not suitable for rapid measurements and possible interventions, especially in low-income regions or developing countries.

Since classical analytical methods cannot easily or rapidly assess Mn levels at point-of-use (POU), alternative methods using electroanalytical techniques have been proposed. These techniques offer relatively low limits of detection (~g/L), coupled with miniaturization for POU measurements.^23–30 Stripping voltammetry is one of the most popular electrochemical techniques for determination of trace levels of heavy metals, including Mn.^23,24,31,32 It consists of two steps, as shown in Fig.1a. First, metal is deposited at the electrode by controlled potential electrolysis, which serves to pre-concentrate the metal analyte on the electrode surface. It is this pre-concentration feature that results in low detection limits. In the second step, the deposited metal is removed or “stripped” from the electrode by an appropriate potential scan that either oxidizes or reduces it back to ionic form in the solution, and the resulting current signal is used to quantify the metal. Although stripping voltammetry can be done at either anode (reduction) or cathode (oxidation), Mn cathodic stripping voltammetry (CSV) has proven to offer superior performance and has been reported using carbon,^33–35 boron doped diamond,³⁶ and indium tin oxide^23,31,37 electrodes. Our group has also reported on determination of Mn using miniature palladium (Pd)²⁴ and platinum (Pt)³² electrodes.

Figure 1.

Sensor with thin film Pt electrodes for the determination of Mn. (a) Schematic of the detection principle of CSV, illustrating the deposition phase to oxidize Mn²⁺ into Mn⁴⁺ which forms a layer of MnO₂ on the surface of the Pt WE, and the stripping phase to reduce Mn⁴⁺ back to Mn²⁺ off the electrode. (b) Photograph of the three-electrode microfabricated sensor with a 1 mm diameter circular Pt WE, a Pt CE, and a Ag/AgCl RE (WE = working electrode, RE = reference electrode, CE = counter electrode). (c) 3D printed interface housing the sensor, micro vibration motor, and the mini-USB connection to the potentiostat.

In this work, we report on validation of our electrochemical sensor system for determination of Mn in drinking water. We used our miniature sensor with Pt thin film electrodes for CSV analysis of Mn, and improved its performance using a new user-friendly interface with integrated agitation capabilities. We demonstrated a calculated limit of detection (LOD) of 10.1 nM (0.56 ppb), which is a nearly 2× improvement over our previous results with Pt³² and a 20× improvement over our previous results with Pd.²⁴ The sensor was validated with drinking water samples (n = 78) from a community in Holliston, MA, showing100% agreement, ~70% accuracy, and 91% precision when compared with measurements from ICP-MS. These results suggest that an electrochemical sensor based on Pt can be a useful and convenient component of a sensing system for rapid determination of Mn in drinking water at the point-of-use.

2. Experimental methods

2.1. Chemicals and reagents

Manganese solutions of desired concentrations were prepared from atomic absorption standard solution of 1000 mg/L Mn²⁺in 2–5% HNO₃ (Acros Organics). Sodium acetate buffers of two concentrations (0.1 M and 0.2 M) were prepared by dilution from a stock solution of 3.0 M, pH 5.2 ± 0.1 sodium acetate (Sigma Aldrich, St. Louis, MO). For electroplating the Ag/AgCl reference electrodes, Silver Cyless II RTU (Technic Inc., Cranston, RI) solution was used. For acidification of water samples, trace metal grade nitric acid (~70%) was purchased from Fisher Scientific. Unless otherwise noted, all other chemicals were purchased from Fisher Scientific.

2.2. Sensor and interface

The sensors in this work were purchased from Micrux Technologies (Oviedo, Asturias, Spain 33006). These sensors (ED-SE1-Pt) feature a 3-electrode configuration on a glass substrate (Fig. 1b). A 1 mm diameter circular Pt working electrode (WE) provided the active sensing area. A Pt counter electrode (CE) supported the current during electrochemical measurements. Both of these electrodes consisted of a 150 nm thick Pt film on top of a 50 nm Ti seed layer. A Pt reference electrode (RE) was transformed into an Ag/AgCl RE by electroplating in a two-step process. First, Ag was plated onto the Pt RE with a cathodic current of 12.55 μA for 240 s using Silver Cyless solution, then it was chloridized at 12.55 μA for 120 s in 1 M KCl. All three electrodes were confined by protective layer of SU-8 resist, with a 2-mm circular opening to ensure the same electrode surface area in all experiments. Before electrode modification or experiments, the sensors were electrochemically cleaned by scanning 10 cycles of cyclic voltammetry (CV) in the ±1.5 V range at 100 mV/s in 0.1 M KCl solution. The advantage of using sensors built on top of a glass substrate is that this prevents nonspecific absorption of metal ions by polymer components. To confirm that no absorption occurs, flat background scans in 0.1 M acetate buffer were obtained before and after 10 measurements of 25 ppb Mn.

The interface housing for the sensors was 3-D printed with Form2 SLA printer using Clear resin (FormLabs, Somerville, MA). It consisted of the top piece (Fig. 1c), which housed a two-layer printed circuit board (PCB) for electrical connection, and the bottom piece, which housed a coin/disc micro vibration motor (Amazon Inc.) for sample agitation. Once printed, the interface components were washed in isopropanol (IPA) to remove all the uncured resin and baked in the oven at 65°C for 20 min improve mechanical strength. Wet sandpapers of 120, 320 and 800 grit were used progressively to remove support marks and to achieve smooth surfaces in printed structures. Spring loaded pins were soldered into the PCB to ensure electrical connection to the sensor, whereas a mini-USB port was included to connect the PCB to the commercial potentiostat. The micro vibration motor was 10 mm in diameter and 2.7 mm in thickness, and was operated via 3.0 V DC and 100 mA to vibrate at 12,000 rpm. It was powered and controlled via an Arduino Uno microcontroller (Arduino, Somerville, MA) to ensure uniform vibration only during the electrochemical deposition phase. We designed a simple PCB which assembled directly onto the ARDUINO board to drive the circuit. It comprised of a switch to start the vibration and a potentiometer to precisely control voltage applied to the vibration motor.

2.3. Sample preparation

Residential tap water samples were collected from a subset of participants of the ACHIEVE (Assessing Children’s Environmental Exposures) study, a community-initiated pilot research study in the town of Holliston, MA. The ACHIEVE study was developed in response to community concerns about drinking water quality and their children’s health. Convenience samples were collected by study staff from participant homes between September 2018 and December 2019. The research study protocol and all study materials for ACHIEVE were approved by the Institutional Review Board at Boston University School of Public Health.

Samples were received in 9 different batches. Samples (n = 78) were collected in 50 mL conical tubes, with volumes ranging from 30 mL to 45 mL, in compliance with the EPA Lead and Copper Rule sampling protocol.³⁸ Locations for sample collection included kitchen sink (n = 34), bath tub (n = 41), and outdoor faucets (n = 3). Tap water sampling was performed from cold water faucets, free of water softener or point-of-use filters, with a 3 min flushing before collection in the provided tubes. Collected samples were shipped on ice to our laboratory in Chicago, IL for analysis. Upon arrival, samples were acidified with an average of 36 μL of ~70% nitric acid to pH < 2.0, per EPA protocol,³⁹ and to prevent precipitation, microbial activity or absorption to the container walls. Acidified samples were stored at 4 °C until analysis. For electrochemical measurements, the samples were diluted 2× with 0.2 M, pH 4.8 sodium acetate buffer, which yielded the adjusted pH ~ 4.7 for the experiments.

2.4. Analytical procedure

The procedure for the experiments was similar to our previous work.³² A miniature potentiostat WaveNow (Pine Research Instrumentation, Durham, NC) with AfterMath Data Organizer software was used in electrochemical experiments. A sensor was inserted into the interface and connected to the potentiostat. For each experiment, a 12 μL sample droplet was placed on top of the sensor, wetting all three electrodes. We performed cyclic voltammetry (CV) and Cathodic Stripping Voltammetry (CSV) in acetate buffer to confirm position of the Mn reduction peak. The CSV parameters were derived from previous work³² and consisted of deposition at 0.7 V for 900 s, with square wave parameters of 70 ms period, 25 mV amplitude, and 4 mV step potential. To enhance the analytical signal and improve diffusion limited mass transport, vibration was applied during the deposition phase. Mn concentrations from 5 ppb (91 nM) to 100 ppb (1.82 μM) were used to construct a calibration curve and to calculate the LOD using 3σ/slope. We used the same stripping parameters for the determination of Mn concentration in water samples, while applying the standard addition method to calculate concentration of Mn in the original samples. For this, spikes of 10 ppb, 30 ppb, and 50 ppb were used. Independent measurements of the water samples were performed using Thermo iCAP Q ICP-MS instrument (ThermoFisher Scientific, Waltham, MA).

3. Results and Discussion

3.1. Confirmation of CSV conditions

Determination of Mn²⁺ by CSV is based on the oxidation of Mn²⁺ to Mn⁴⁺, its deposition as MnO₂, and its subsequent cathodic stripping. Although we have used both Pd and Pt working electrodes in the past, and found them to behave similarly in electrochemical experiments, Pt is more difficult to oxidize than Pd (0.94V vs. 0.6V electrochemical potential, respectively). Thus, Pt working electrodes provide a wider positive potential window with reduced interference from electrode oxide formed during deposition. The experimental conditions for CSV determination of Mn²⁺ using Pt electrode were thoroughly investigated and optimized in our previous work,³² yielding deposition potential of 0.7 V and deposition time of 900 s in 0.1 M sodium acetate buffer at pH 5.5. Thus, these were the parameters used in this work. We did however use a slightly more acidic acetate buffer (pH 4.7) as supporting electrolyte, based on work from collaborators²³ that showed buffers with more acidic pH facilitate stripping of MnO₂ and thus improve CSV peak signal. We used cyclic voltammetry in this buffer with 10 ppm and 20 ppm Mn²⁺ to confirm the potential window and location of the reduction peaks of deposited MnO₂.

The background scan in Fig. 2a shows a large anodic current above +1 V, which is due to oxygen evolution on the Pt electrode, and a larger cathodic current below ~0 V, associated with hydrogen evolution on the Pt electrode. This ~1 V potential window is similar to what we observed previously.³² Addition of Mn²⁺ caused formation of insoluble MnO₂ at the initial positive potential. During the reduction scan towards more negative potentials, double peaks were observed. This behavior is attributed to the complex mechanism behind the MnO₂ reduction back to ionic Mn²⁺ as reported in literature.^40–43 Similar to our earlier work,³² the voltammogram with Mn peaks (solid line) overlaps well with the background voltammogram (dashed line), making evaluation of the peak potential and amplitude straightforward.

Figure 2.

Electrochemical performance of the sensors in standard solutions of 0.1 M sodium acetate buffer at pH 4.7. (a) Cyclic voltammetry on the Pt sensor in buffer and with the addition of 10 ppm and 20 ppm of Mn. (b) CSV in acetate buffer with and without addition of 100 ppb Mn, illustrating a stripping peak and a flat background signal in the relevant potential region for MnO₂ reduction.

Voltammogram for CSV over a more restricted potential range is shown in Fig. 2b. We scanned the background and 100 ppb Mn²⁺ using 0.7 V pre-concentration potential in acetate buffer, based on CV. We observed that the background signal in acetate buffer agrees with the voltammogram of 100 ppb Mn. The double peak of the stripping voltammogram can still be identified but appears to be less resolved than in our previous work.³² Nevertheless, due to the complexity of the peak shape, we used peak area in waveform analysis rather than amplitude, as it yielded better linearity and sensitivity.

3.2. Sample agitation

The reaction kinetics of electrochemical assays are often substantially influenced by mass transport. The role of mass transport in these assays is to efficiently bring analyte to the reactive electrode surface, ensuring that the reaction rate does not decay with time. Mass transport in liquid samples consists of both diffusion, driven by concentration gradients, and advection, in response to flow motion. Because diffusion alone is not sufficient to achieve this, advection must be generated. In macroscale assays, where sample volumes are often >1 mL, advection is predominantly generated using a magnetic stirrer. Magnetic stirrers, however, cannot be used in miniaturized assays aimed at point-of-care applications where sample volumes are small, often <100 μL. Thus, the microscale electrochemical assays are carried out as acquiescent, with diffusion often limiting the reaction rate.^44,45 Indeed, this can be seen in our earlier work, where increase in deposition time always resulted in current (or charge) saturation.^23,33 In biotechnology and medical diagnostics, where surface assays such as ELISA and immunofluorescence have become ubiquitous, sample volumes have decreased to ~200 μL in 96-well plates and the prevalent approach to generating advection is using an orbital shaker.⁴⁷ While shakers improve mass transport in miniature samples, and are widely accepted due to their ease of use, they are not suited for point-of-care applications due to size and power requirements.

In this work, we agitated the droplet-size liquid samples by integrating a small, coin-size vibration motor under the electrochemical cell. Vibrations from the motor yielded localized stirring of the sample droplet in a compact fashion. The compact nature of these motors, and the ability to operate from a portable power source (AA batteries or directly from USB) make them well suited for point-of-use applications. To quantitatively evaluate impact of agitation on sensor performance, we compared three experimental conditions, including no vibration (acquiescent), vibration provided by a stirring stage, and vibration provided by the coin disks.

Introduction of agitation led to an increase in signal and improvement in level of quantification (LOQ) from 5 to 3 ppb. Representative voltammograms for each of the conditions are shown in Fig. 3. In all cases, the waveforms appeared as double peaks for Mn concentrations above 25 ppb, becoming more pronounced with increasing Mn²⁺concentration. It is common for double stripping peaks to occur on solid electrode due to stripping of the bulk portion of the deposited analyte occurring at a different potential than stripping of the first monolayer on the electrode.³² For all concentrations, the reduction process starts at approximately 600 mV. However, introducing vibration resulted in a shift of the major peak toward less positive potentials, from ~450 mV to ~ 350 mV. At higher concentrations, above 25 ppb, it is possible that the larger amount of oxide deposited requires longer time to reduce. This leads to a longer stripping process and thus less positive Mn peaks. The Mn oxides from the lower concentration samples are easier and faster to reduce, so they are mostly stripped before 350 mV. Regardless of the agitation, the background waveforms are clean and do not show any minor reduction peaks of Pt oxide at ~300 mV as is the case for the Pd electrode.²⁴

Figure 3.

Effect of vibration during deposition phase in 0.1 M sodium acetate buffer at pH 4.7. (a) No vibration applied with Mn concentrations in the 5–100ppb range. Vibration provided by (b) the embedded small vibration disc and (c) the commercial stirrer stage, with Mn concentration in the 3–100 ppb range. (d) Comparison of the calibration curves obtained considering the peak area from the depicted voltammograms.

Measuring the area (charge) of the Mn reduction peak, calibration curves for each of the three conditions can be constructed (Fig. 3d). These calibration curves demonstrate 98–99% linear relationship between the stripping peak area and the concentration. The correlation equation of peak area measurements with no agitation is Q(μC) = 0.009 [Mn(ppb)] – 0.034 (R² = 0.998). Introducing agitation yields the correlation equations of Q(μC) = 0.021 [Mn(ppb)] – 0.035 (R² = 0.996). Introduction of agitation led to an increase in sensitivity by 2.3×, and improvement in level of quantification (LOQ) from 5 to 3 ppb (Table 1). Compared with our earlier work using a Cu-Pd sensor,²⁴ this is a 32× improvement in sensitivity and 20× improvement in detection limits.

Table 1.

Comparison of sensor performance for different vibration modes.

Approach	Linearity	Sensitivity (μC/ppb)	LOQ (ppb)	LOD (ppb)
Acquiescent	0.998	0.009	5	1.10
Thermix stirrer	0.982	0.034	3	0.39
Vibration disc	0.996	0.021	3	0.31

To further characterize the influence of vibration, we varied potential applied to the vibration disc in the range 0 V (no vibration) to 4 V. As expected, the higher driving potential generated higher agitation and thus stronger signal. This is shown in Fig. 4, which compares the peak area for 10 ppb Mn sample as a function of the vibration disk potential. Despite the significant increase in signal, biasing the vibration disk at 4 V also resulted in a significant increase in signal variability, with coefficient of variation approaching 26%. Such a high level in variability is undesirable, as it adversely impacts precision of the sensor. Therefore, the optimal voltage of 3V was selected for the following experiments, consistent with the rated voltage in the specifications of the vibration discs.

Figure 4.

Effect of potential applied to the vibration motor on the peak area of 10 ppb Mn in 0.1 M sodium acetate buffer at pH 4.7.

3.3. Calibration in water matrix

Calibration curve in water sample matrix was developed by spiking 5–100 ppb of Mn into tap water samples. The range was selected to bracket the EPA aesthetic guideline for Mn, which is 50 ppb.¹⁰ The representative voltammograms in Fig. 5a show complex peaks for the reduction of MnO₂ back to Mn²⁺ at concentrations above 25 ppb. No significant differences from stripping peaks in acetate buffer are observed, although there is a shift of ~50mV towards more positive potentials. The correlation equation of peak area measurements is Q(μC) = 0.026 [Mn(ppb)] – 0.181 (R² = 0.995, Fig. 5b). A sensitivity of 0.026 μC/ppb for the laboratory tap water was obtained as compared to 0.021 μC/ppb in the acetate buffer solutions. The increase in sensitivity is attributed to a more facile reduction of MnO₂ to Mn²⁺ in the water matrix, as corroborated also by the presence of a third reduction peak and the ~50mV shift towards more positive potentials. In terms of analytical considerations, the advantage of considering the overall area under the voltammetric curve instead of the peak amplitude values allowed to overcome this difference and fully exploit the effectiveness of the electrochemistry of MnO₂ reduction in water matrix. The LOD is calculated as 0.56 ppb based on 3σ/slope (n =10), which is an improvement over the previously reported detection limit of 0.89 ppb for Pt sensors³², and a 20× improvement over the Cu-Pd sensor in our earlier work²⁴.

Figure 5.

Performance of the developed system in tap water samples. (a) Voltammograms obtained in laboratory tap water spiked with known Mn concentrations. Samples diluted 2X with AB 0.2M, final pH 4.7. (b) Calibration curve Mn concentration 5 ppb - 100 ppb. (c) Sensor variability tested at Mn concentration of 5 ppb with 10 different sensors and n = 3 repetitions for each sensor.

To assess reproducibility of the sensor and determine precision of the measurements, we performed CSV at the limits of quantification, 5 ppb Mn, on 10 different sensors, repeating 3× (n = 3) for each sensor. The goal here was to assess reproducibility at the limits of quantification, since measurements become less challenging with higher concentrations. The measurements of peak area (Fig. 5c) show values from ~30 nC to ~40 nC, with a mean of ~35 nC. The variability of these measurements exhibited a mean of 8.05%, with 7 out of 10 sensors generating coefficient of variation (CV%) values under 10%. Our prior work using 6 sensors to repeatedly measure 10 ppb Mn concentration showed CV% to be approximately 9%.³² Moreover, as Fig. 5c illustrates, only 2 out of 10 sensors exhibited CV% values greater than 15%. Therefore, we have confirmed that these thin film sensors are capable of providing reproducible results in tap water with a precision of ≥92%.

Metals in water may potentially interfere with Mn CSV, which may impact accuracy and reliability of the sensor. Although CSV is less susceptible to interference than ASV (anodic striping voltammetry), we investigated potential interferers in our earlier work.³² We chose Zn²⁺, Mg²⁺, Cu²⁺, Pb²⁺, and Fe²⁺ metal ions since they can leach from water pipes and soldered joints that deliver water to tap, and also strip at potentials similar to Mn or can possibly interfere with deposition of Mn. Our results revealed that spiking these metals up to 100x in excess of the 10 ppb Mn in the solution yielded no significant alteration to the voltammetric response.³² Interference from Fe²⁺ was observed for concentrations approaching ppm levels. Fortunately, in water samples the concentration of Fe metal would normally be at sub-ppm levels.⁴⁹ Further, Fe(II) tends to be unstable and precipitate as insoluble Fe(III) hydroxide,⁵⁰ avoiding possible influence on Mn determination. Another metal ion, Ni²⁺ may be present in water samples, but is not considered an interferant since electrochemical detection requires chemical modification of the electrode surface to achieve acceptable sensitivity.^51,52

3.4. Determination of Mn in drinking water

We used the Pt sensors for the determination of Mn in drinking water from homes in Holliston, MA. Holliston is a Boston suburb where nearly 100% of the drinking water is drawn from very shallow aquifers (<20 feet to water table, <50 feet to bedrock). It has been shown that heavy metal contamination of water – both from natural and anthropogenic sources – is particularly significant when drinking water is drawn from shallow aquifers (<100 feet to water table).⁴⁶ Tap water samples (n = 78) collected from ACHIEVE study participant homes were acidified with 70% nitric acid per EPA protocol³⁹ to pH < 2. We analyzed samples using ICP-MS for confirmation and to compare the accuracy of our measurements. For electrochemical measurements, pH adjustment of acidified samples was necessary. However, titration with additional NaOH(s) to adjust sample pH was not preferred as this method can randomly reduce the amount of free Mn²⁺ by creating a temporarily-highly-basic region in the bulk sample, which may cause formation Mn oxide precipitates. Thus, samples were diluted 2× with 0.2 M, pH 4.8 sodium acetate buffer, which yielded the adjusted pH ~ 4.7 for the experiments.

All samples were analyzed using CSV, following dilution and addition of three spikes (10 ppb, 30 ppb, and 50 ppb Mn) in a standard addition approach. The representative voltammograms are shown in Fig. 6a for sample #W37. Compared to the results from the reference tap water, the voltammograms span the same potential range from 0.35 V to 0.65 V, and exhibit the similarly-complex shape. The resulting standard addition plot of peak area (Fig. 6b) yielded a correlation equation of Q (μC) = 0.054 [Mn(ppb)] + 0.348 (R² = 0.999). The resulting sensitivity of 0.054 μC/ppb is in the same order of magnitude as the sensitivity of calibration in tap water (0.026 μC/ppb). All samples were also analyzed using ICP-MS, revealing Mn concentrations in the range 0.03 ppb to 5.3 ppm, with a geometric mean of 4.74 ppb and a median of 2.29 ppb. Comparing CSV measurements with ICP-MS data yields an aggregate accuracy of ~70% and precision of ~91%. Table 2 summarizes sensitivity, specificity and concordance values for the comparison between our electrochemical method and ICP-MS. The EPA set the SMCL for Mn to 5 0ppb¹⁰ and therefore this value was selected as threshold to determine the True Positive (TP), False Negative (FN) , True Negative (TN) and False Positive (FP) values, as defined in conventional statistical analysis.⁴⁸ In terms of device testing, the most valuable result to be highlighted was the absence of False Positives in all 78 samples, which indicates that the devised approach properly identified all the 11 samples above the safety limit for Mn in drinking water among the tested samples. This further corroborates the applicability and reliability of our electrochemical platform for real, point-of-use analysis of environmental samples.

Figure 6.

Analysis of drinking water samples. Sample diluted 2X with AB 0.2M, final pH 4.7. (a) Cathodic stripping voltammograms (I-V) of unknown sample and known solutions for three-points standard addition and (b) corresponding linear regression model to extrapolate unknown Mn concentration (sample #W37).

Table 2.

Comparison between the electrochemical sensors and the ICP-MS analyses of water samples.

Statistical Index	Formula	Value
Statistical Sensitivity	TP/(TP+FN)=TP/nD	(11/11) =1
Statistical Specificity	TN/(FP+TN)=TN/nC	(67/67) =1
Concordance/Agreement	TP+TN/n	(78/78) =1
Mean accuracy detectable samples (%)	(1-Erel)*100	69.9%
Mean Precision (%)	(1-CV)*100	91.2%

In studies benchmarking a new sensor system with an existing gold standard approach, the Pearson correlation coefficient is often used to measure the strength of the correlation between the two approaches. This statistic varies between −1, indicating a total negative linear correlation, and +1, indicating a total positive linear correlation. In this work, the Pearson correlation coefficient was calculated to be ρ = 0.937 indicating a strong correlation between our electrochemical measurements and the ICP-MS. These results on a large set of drinking water samples expand prior work in the analysis of surface water samples which yielded a ~90% accuracy and ~96% precision when compared to ICP-MS³² for the two surface water samples tested in the concentration range 50–100 ppb.

Ultimately, in this work we validated the performance of Pt sensors for the electrochemical detection of Mn in drinking water on a large number of samples (n = 78) in the 0.03 ppb to 5.3 ppm range. Moreover, we devised a compact, small-sized interface to integrate agitation function and improve the LOQ to 6ppb. We achieved a 100% agreement and acceptable accuracy (~70%) and precision (91%) when benchmarked with ICP-MS. These results of Mn determination in drinking water samples expand prior work in the analysis of surface water and serve to validate our electrochemical sensor approach. In recent years, similar electrochemical sensors have been developed for the quantitation of a wide range of analytes, including uric acid⁵³ and blood β-ketone/glucose⁵⁴, white blood cells (WBCs)⁵⁵, air pollutants⁵⁶, and pesticides⁵⁷. Many of these can be integrated with smartphone-based analyzers for POU applications. We envision development of a similar POU system for our sensor, for more frequent Mn monitoring in drinking water in rural communities or settings at greater risk of exposure, where regular testing is of paramount importance and the conventional laboratory-based methods are time-consuming, cost-prohibitive, or not accessible.