Monitoring the lead-and-copper rule with a water-gated field effect transistor

Two common low-level toxic pollutants in drinking water are the heavy metal cations lead (Pb²⁺) and copper (Cu²⁺), e.g., lead leaches from historic water pipes, copper from ‘low tech’ water sterilisation (Sudha et al. 2012; Masindi & Muedi 2018). Lead and copper are subject to regulation, e.g., the US Environmental Protection Agency's (EPA) ‘lead-and-copper rule’ (EPA 2008) sets ‘action levels’ of 0.015 mg/L = 72 nM for lead and 1.3 mg/L = 20.5 μM for copper in the domestic water supply.

The additional parameter β < 1 describes inhomogeneity in the analyte/ionophore binding sites (Turiel et al. 2003). c_1/2=1/K remains true regardless of the value of β. The ratio of K's for a target analyte vs an interferant (or the inverse ratio of c_1/2's) quantifies the selectivity, S, of a sensitiser.

Transistor contact substrates were prepared by thermal evaporation of Au (100 nm) with Cr (10 nm) as adhesion layer onto clean glass substrates by a shadow mask. Each substrate contains five pairs of electrodes separated by a channel with a length L = 30 μm and width W = 1,000 μm (W/L = 33.3). SnO₂ films were prepared onto contact substrates by spray pyrolysis. We used a commercial airbrush at 20 cm distance to spray four short ‘puffs’ of 0.05 M SnCl₄.5H₂O in isopropanol in 1 min intervals (Vasu & Subrahmanyan 1991; Sankar et al. 2015; Anaraki et al. 2016) onto contact substrates preheated to 400 °C. Afterwards, substrates were left on the hot plate for 30 min for the full decomposition of the chloride precursor into SnO₂. We have shown previously (Alghamdi et al. 2019) that SnO₂ leads to WGTFTs with very low threshold, sufficient carrier mobility, and stability under water.

Poly(vinyl chloride) (PVC), 2-nitrophenyl octyl ether (2NPOE) and tetrahydrofuran (THF) were purchased from Sigma Aldrich. The zeolite clinoptilolite was sourced from DC Minerals' eBay shop (https://www.ebay.co.uk/str/DC-Minerals?_trksid=p2047675.l2568) as a fine powder, grain size <40 μm. It is a natural product mined in bulk from mineral deposits and may be a mixture of different but similar compounds, which the supplier does not fully characterise. An approximate overall composition is given as [(Ca, Fe, K, Mg, Na)_{(3 to 6)}Si₃₀Al₆O₇₂].24H₂O. PVC membranes were prepared based on the procedure described in Sohrabnejad et al. (2004). We dissolved 30 mg of PVC, 65 mg of plasticiser 2NPOE and 42 mg of clinoptilolite in 3 mL of THF. 500 μL of the solution was poured into a small vial and left overnight at room temperature to allow evaporation of THF. The resulting membranes were then conditioned for 4 hours in tap water which did not contain any deliberately added ions. Finally, the membrane was glued in between two plastic pools with epoxy (see Figure 1).

To simulate realistic conditions for practical use of our sensor, we did not work with deionised water but drew water samples from drinking water taps at Sheffield University in March/April 2019. The water supplier's umbrella organisation, Water UK, gives the most common cations in tap water as calcium, magnesium, sodium, selenium and potassium (Water UK 2016). For the assessment of water quality in the UK, the Drinking Water Inspectorate (DWI) releases an annual summary report (Dwi.defra.gov.uk. 2018), and our local supplier, Yorkshire Water, provides a list of ion concentrations in Sheffield tap water (Yorkshire Water n.d.). A 1 mM Cu²⁺ stock solution was prepared by dissolving copper nitrate, Cu(NO₃)₂, in tap water; we then get the desired (low) concentrations used in experiments by diluting with more tap water to (300, 200, 100, 50, 10, 1, 0.5 μM) Cu²⁺. For Pb²⁺, we prepared 1 μM stock solution of lead nitrate, Pb(NO₃)₂, dissolved in tap water and then diluted to low concentrations (0.5, 1, 5, 10, 25, 50, 100, 250 nM).

To test the response of membrane-sensitised WGTFTs to Cu²⁺/Pb²⁺, we used a two-chamber design, similar to some previous researchers (Arvand-Barmchi et al. 2003; Schmoltner et al. 2013), which is derived from the design of traditional potentiometric ion sensors (Menon et al. 2011). The SnO₂ transistor substrate was in contact with tap water held in an inner (reference) pool that is separated from an outer (sample) pool by the sensitised PVC membrane. The water in the reference pool was tap water as drawn, with no deliberately added ions. For sensor calibration, the outer pool is initially also filled with tap water, but this is then subsequently replaced with solutions of known and increasing concentrations of lead or copper, prepared as described in the section ‘Preparation of test solutions’, while the inner pool remains filled with tap water as a reference. For practical use of the WGTFT as lead and copper sensor, rather than calibration, the sample pool will be filled with the potentially contaminated water. The transistor is gated by a tungsten (W) contact needle that is submerged in the outer pool. As with all electrolyte-gated transistors, the potential applied to the gate contact will be communicated to the semiconductor surface via interfacial electric double layers (EDLs). However, the potential at the semiconductor surface will be different from the potential applied to the gate needle by any membrane potential, V_M(c) in response to different ion concentrations, c, in the outer (sample) vs inner (reference) pool. The setup is illustrated in Figure 1.

As V_M(c) adds to the applied gate potential, it can be measured as a shift in the WGTFT's threshold voltage, ΔV_th. We, therefore, recorded linear transfer characteristics by connecting all tungsten contact needles/probeheads shown in Figure 1 to two Keithley 2,400 source/measure units using coaxial cables. Both Keithley units are controlled by bespoke LabView software to record transistor characteristics, in this case, linear transfer characteristics. Each time a new sample was poured into the outer pool, we allowed 2 minutes to equilibrate. This may be longer than necessary, e.g., in the Alghamdi et al. (2019) study, 30 second equilibration was sufficient, but we do not regard fast(er) characterisation as our key objective. We then scanned V_G from −0.4 V to 0.7 V in steps of 20 mV at constant drain voltage V_D = 0.1 V. Gate voltage is stepped in time intervals of 0.5 seconds, i.e., recording each transfer characteristic takes 56 seconds. To determine membrane potential V_M= ΔV_th, we compensate for it by shifting recorded linear transfer characteristics along the gate voltage (V_G) axis to achieve the best overlap with the characteristic recorded under pure tap water in the sample pool. We identify the gate voltage shift required for the best overlap with the pure tap water characteristic as threshold shift ΔV_th. This method does not rely on any particular mathematical model of the linear transfer characteristics. It gives V_M= ΔV_th even when transistors do not exactly follow theoretical TFT equations, and independent of channel geometry and the semiconductor carrier mobility. The same analysis has been used previously in other WGTFT sensor work (e.g., Casalini et al. 2013; Althagafi et al. 2016a; Al Baroot & Grell 2019). Results were fitted against a quantitative model (Equation (3)) using the nonlinear fitting routine in Origin 2,018 software.

In Figure 2(a), we show the linear transfer characteristics of SnO₂ WGTFT transistors sensitised with a clinoptilolite membrane. The reference (inner) pool was filled with tap water, and the sample (outer) pool with tap water with increasing concentrations of Pb²⁺ up to 250 nM. Transfer characteristics clearly shift to more negative threshold voltages with increasing lead concentration, which indicates a lead concentration-dependent membrane potential. For quantitative analysis, we have shifted all transfer characteristics under leaded water (c_Pb> 0) to match the c = 0 characteristic, as described in the section ‘WGTFT characterisation and analysis’. The resulting ‘master’ transfer characteristic is shown in Figure 2(b).

Figure 2(b) shows an excellent overlap of all transfer characteristics into a single ‘master’ characteristic. This typical master curve confirms that increasing lead concentration in the sample pool impacts threshold voltage only, not any other WGTFT performance parameter. We identify the gate voltage shift required for best overlap as the WGTFT's threshold voltage shift under increasing lead concentration, ΔV_th(c_Pb). These are shown and analysed in the section below.

We then repeated the above experiment using nominally identical transistors, but adding increasing concentrations of copper (Cu²⁺) up to 300 μM rather than lead to the outer pool. Note the ∼1,000 times higher concentrations of Cu²⁺ vs Pb²⁺. Corresponding results are shown in Figure 3.

Figure 3(b) again shows the excellent overlap of all transfer characteristics into a single ‘master’ transfer characteristic. Threshold shifts ΔV_th(c_Cu) are shown and analysed in the section below.

Figure 4 shows ΔV_th(c_Pb) and ΔV_th(c_Cu) as evaluated from the shift of transfer characteristics along the V_G axis to construct ‘master’ characteristics.

We find that the threshold shifts observed in WGTFTs with increasing concentration of Pb²⁺ and Cu²⁺ increase rapidly for low concentrations and approach saturation ΔV_th(sat) of several 100 mV at high concentration. This response is different from the Nikolsky–Eisenman law (Equation (1)) but similar to our previous results with zeolite mordenite (Alghamdi et al. 2019), albeit we required the LF isotherm (Equation (3)) rather than the simpler Equation (2), for the fits shown in Figure 4. We find a satisfactory match for Pb²⁺ and excellent match for Cu²⁺. The values for the fit parameters K, β and ΔV_th(sat) from Equation (3) for both Pb²⁺ and Cu²⁺ sensing are summarised in Table 1.

The three-orders-of-magnitude larger K for lead vs copper indicates the stronger extraction of lead rather than copper by clinoptilolite, which is already evident from the concentration scales used in Figure 2 (nM) vs Figure 3 (μM). To determine values for the LoD, we re-plot the data in Figure 4 in linearised form, ΔV_th(c)((Kc)^β + 1) vs (Kc)^β (Figure 5), using K and β for Pb²⁺ and Cu²⁺, respectively, from Table 1. We then fit straight lines of the form y=mx+b; resulting parameters m (slope) and b (intercept) with their respective errors are listed in Table 2.

We here find LoD(Pb²⁺) = 0.9 nM and LoD(Cu²⁺) = 14 nM, which are already included in Table 2. To make sure Cu²⁺LoD is realistic rather than an artefact of mathematical analysis, we have repeated the experiment shown in Figure 3(b) with very small Cu²⁺ concentrations (15 and 50 nM), resulting threshold shifts are shown as triangle symbols in Figure 4(b). Note triangles agree well with the fit (solid line, and 15 nM is very close to the evaluated LoD and does lead to a recognisable threshold shift (≈30 mV); hence, the calculated LoDs for Cu²⁺ are realistic. LoD for lead is relatively larger than for copper when compared to 1/K, which reflects the larger scatter (poorer fit to the model, Equation (3)) in the original data, particularly at higher concentrations. Visually, the lead LoD formally evaluated by Equation (4) seems an overestimate when inspecting Figure 2(a), which shows a clear threshold shift under LoD = 0.9 nM lead. Nevertheless, formally evaluated LoDs for both lead and copper are significantly smaller than the action levels of the lead-and-copper rule, which qualifies our sensors for its monitoring.

While the tap water drawn in our laboratory has near-neutral pH (pH = 7.2, measured with pH meter (CyberScan PH 300)), drinking water generally may vary in pH, with the permitted range for drinking water (in the EU) being pH 6.5–9.5 (Council of the European Union 1998). Practically, water samples can be tested for pH with a pH meter and adjusted to pH 7 by adding small amounts of a strong base (or acid) before lead and copper testing. Contamination with, e.g., Na⁺ from NaOH will in itself not lead to significant threshold shift, as we show below in the section ‘Interference from common co-cations’. However, we here show that the impact of pH on sensing of lead and copper is small. We added a drop of acetic acid to our tap water to deliberately make it mildly acidic, pH 5.2 as measured with the same pH meter. We then tested clinoptilolite-based WGTFTs to sense lead and copper in acidified tap water. Threshold shifts at one representative heavy metal concentration for as-drawn (pH 7.2) vs acidified (pH 5.2) tap water are compared in Table 3. Concentrations were chosen to lead to near-saturated threshold shift according to Figure 4.

Heavy metal-induced threshold shifts under acidic conditions are slightly smaller than under near-neutral pH. However, shifts are still significant at pH 5.2, which is more than one pH unit below the permitted pH range for drinking water. Clinoptilolite membranes are therefore suitable to detect lead and copper within the permitted pH range of drinking water. For accurate quantitative determination at significantly non-neutral pH, we advise calibration (as in Figures 2 and 3) at several pHs, or prior neutralisation of acidic samples with small amounts of a strong base, e.g., NaOH.

As the usual application of clinoptilolite is to extract lead and copper pollution from the drinking water supply (Erdem et al. 2004; Perić et al. 2004), we have here used clinoptilolite membrane-sensitised WGTFTs to test extraction performance. We ‘spiked’ 15 mL of tap water with 1 μM lead and copper, respectively (same concentration to allow direct comparison of extraction), and then attempted to extract the heavy metal again. For this, we added 100 mg of clinoptilolite to spiked water, agitated, and left to settle for 2 hr. We then tested water samples resulting from this spiking/extraction procedure in a WGTFT transistor sensitised with clinoptilolite membrane in the same way as in the section ‘Lead and copper sensing results’. The resulting transfer characteristics are shown in Figure 6, which for comparison also includes the transfer characteristic for as-drawn tap water that has not been spiked/extracted.

We find that characteristics for both spiked/extracted samples do display a small threshold shift compared to a tap water sample that has never been spiked. However, the shift is significantly lower than what we found in the section ‘Lead and copper sensing results’. This response suggests that the extraction procedure has significantly reduced the initial 1 μM heavy metal concentration, albeit a small amount of pollution remains. Results are summarised in Table 4, which also shows the heavy metal concentration remaining after extraction. These are calculated with Equation (3) from the measured threshold shifts after extraction, using the parameters listed in Table 1.

Table 4 shows that clinoptilolite is, indeed, effective in extracting lead and copper from drinking water. The remaining heavy metal pollution after extraction is far below the action level. The larger K for lead vs copper established previously is reflected again in the lower residual concentration after extraction.

Drinking water naturally contains common cations of alkaline and alkaline earth metals (e.g., Na⁺, Ca²⁺, Mg²⁺) in concentrations typically ranging in the order (100 μM–1 mM). For example, our laboratory's water supplier, Yorkshire Water, quotes a typical ‘cocktail’ of 200 μM Ca²⁺, 99 μM Mg²⁺ and 783 μM Na⁺ (Yorkshire Water n.d.). These concentrations are significantly greater than the ‘action levels’ for heavy metals under the lead-and-copper rule, but alkaline and alkaline earth metal co-cations at these levels are not harmful and should not lead to ‘false positives’. As described in ‘Preparation of test solutions’, we account for the common tap water interference ‘cocktail’ by preparing calibration solutions and testing our WGTFTs, using tap water rather than DI water. We have, nevertheless, studied the interference from co-cations on our WGTFT heavy metal sensor. Figure 7 shows the transfer characteristics of a SnO₂ WGTFT transistor sensitised with a clinoptilolite membrane when using tap water with deliberately added sodium (Na⁺) ions (from NaCl) or calcium (Ca²⁺) ions (from CaCl₂) in the sample pool vs tap water as drawn in the reference pool.

There are measurable threshold shifts under co-cations, as summarised in Table 5.

We find that the highest threshold shifts due to Na⁺ and Ca²⁺ are significantly smaller than ΔV_th(sat) under Pb²⁺ or Cu²⁺. At 100 μM, we find a shift of 85 mV for Na⁺ and 120 mV for Ca²⁺ while ΔV_th(100 μM) >400 mV for Cu²⁺. According to Equation (3) with the parameters listed in Table 1, the action levels of 72 nM for lead and 20.5 μM for copper would lead to threshold shifts of 289 mV (lead) or 356 mV (copper), both significantly larger than 100 mV. Hence, at least qualitatively, we can still decide potability with respect to lead and copper despite interference. To quantify selectivity, we observe from Figure 7(a) that c_1/2 ≈ 30 μM for Na⁺, hence, selectivity S for lead over sodium is S(Pb²⁺ vs Na⁺) = K(Pb)/K(Na) = c_1/2(Na)/c_1/2(Pb) ≈ 13,000; log S ≈ 4.5.

For the sensor calibration in ‘Lead and copper sensing results’, co-cation (i.e., interferant) concentration in sample and reference were matched by calibrating sensors with sample solutions we prepared from the same tap water as we use for reference (cf. ‘Preparation of test solutions’). Clinoptilolite selects nanomolar lead from a ∼millimolar interferant cocktail when the reference pool carries a matched cocktail. In ‘Interference from common co-cations’, we study the practically unrealistic scenario of adding interferants to the sample solution without matching in the reference and find that such a mismatch would still allow lead and copper sensing, at least qualitatively. A more realistic interference ‘loophole’ arises not because we would deliberately add interferants to a sample, but because test samples taken in the environment would carry an a priori unknown interferant cocktail that will be different from our tap water. If we nevertheless use our tap water in the reference pool, interferants in reference and sample would not be matched.

To address this ‘interference loophole’, we propose a procedure based on extraction as described in the section ‘Lead and copper extraction with clinoptilolite’ to generate interferant (and pH) matched reference solutions. When obtaining a sample of unknown lead and/or copper content, we would first split it in two, and then generate an interferant-matched reference solution from one of the two by extracting lead and copper with clinoptilolite. We then first fill both reference and sample pool with extracted (i.e., self-generated reference) solution and record a reference transfer characteristic, corresponding to the ‘0 nM/μM’ curves in Figures 2 and 3. Then, we replace the extracted solution in the sample pool with non-extracted (i.e., actual) sample and test for threshold shift. We tested this procedure by applying it to a control, and a sample contaminated with lead at potability limit (72 nM) (Figure 8).

Figure 8(a) provides a control experiment, applying the above extraction procedure to Sheffield tap water (drawn on a different day as previously) without (deliberately) added lead. The two transfers in Figure 8(a) both were taken with extracted ‘reference’ in the reference pool, but compare extracted ‘reference’ and non-extracted ‘sample’ in the sample pool. The two curves are virtually identical, reflecting that the ‘sample’ was, in fact, tap water with no added lead, like the reference. This control experiment shows that the extraction procedure itself does not introduce false positives, e.g., by unintended extraction of interferants. In Figure 8(b), we then applied the same procedure to a sample which we prepared by adding 72 nM lead (potability limit) from lead nitrate to tap water drawn on a different day, so not necessarily identical to the water used for Figures 2 and 3. We then split this ‘spiked’ sample and generated an interferant matched reference by extraction. The reference pool was filled with extracted sample (i.e., interferant matched reference). The sample pool was first filled with the same extracted sample to record a reference transfer characteristic and was then replaced by untreated 72 nM lead-spiked sample to record the transfer characteristic under sample exposure.

Now we observe a significant threshold shift ΔV_th ≈ 110 mV between ‘reference’ (extracted sample) and actual sample in the sample pool. With the extraction procedure, we unambiguously detect lead at potability limit, relying only on clinoptilolite and the sample itself. The unknown interferant cocktail in the sample is accounted for by reference to an extracted sample with matching (albeit unknown) interferants.

The cheap and naturally abundant zeolite clinoptilolite is not only useful for the extraction of the toxic heavy metals copper and lead from contaminated water but also their sensing and monitoring of the lead-and-copper rule. When we embed powdered clinoptilolite into a plasticised PVC membrane that we use to separate a sample pool and a reference pool in water-gated SnO₂ thin-film transistor, we find a membrane potential that leads to transistor threshold shift in response to the presence of either Pb²⁺ or Cu²⁺ in the sample pool. Threshold shift follows a Langmuir–Freundlich (LF) characteristic (Equation (3). This is in contrast to Nikolsky–Eisenman (NE) characteristics (Equation (1)), which are usually found for potentiometric membranes sensitised with organic ionophores (e.g., Cadogan et al. 1992; Menon et al. 2011; Schmoltner et al. 2013; Melzer et al. 2014; Al Baroot & Grell 2019), but also in prior reports on zeolite-sensitised membranes (e.g., Arvand-Barmchi et al. 2003). The NE characteristic flat lines at concentrations c<c_st, hence, LoD ≈ c_st, which is typically in the order (100 nM to 1 μM). The LF characteristic lacks such a lower cut-off and, in fact, shows the steepest slope of membrane potential with c in the limit c → 0, opening a window to much lower LoDs. We here determine LoDs which for both Pb²⁺ and Cu²⁺ are significantly smaller than the ‘action levels’ stipulated by the lead-and-copper rule (EPA 2008), Pb²⁺: LoD 0.9 nM vs 72 nM action level, Cu²⁺: LoD 14 nM vs 20.5 μM action level. Threshold shift saturates for high ion concentrations, namely, at 341 mV (Pb²⁺) and 542 mV (Cu²⁺), which is large within the 1.23 V electrochemical window of water. Sensors work even in mildly acidic conditions. This qualifies clinoptilolite-sensitised WGTFTs as a low footprint sensor technology for monitoring the lead-and-copper rule, and to confirm the effectiveness of attempts to extract lead and copper from water. For the practical use of such sensors, potential interference from common co-cations such as Na⁺, Ca²⁺ and Mg²⁺ is a more serious challenge than LoD. However, we provide and verify a routine for generating interferant-matched reference solutions by using clinoptilolite as extractant as well as a sensitiser, closing the interference ‘loophole’.

The reason for the unusual but useful LF response characteristic warrants further study. We note an important difference between conventional macrocycle-sensitised potentiometric sensors and zeolite-based sensors; namely, macrocycles capture the target ion and hence charge the membrane. Zeolites are ion exchangers, so acquire no net charge under target ion exposure, but may well build up superficial dipoles.

Zahrah Alqahtani thanks the Cultural Attaché of Saudi Arabia to the UK and Taif University, Saudi Arabia, and Nawal Alghamdi thanks the Cultural Attaché of Saudi Arabia to the UK and Tabuk University, Saudi Arabia, for providing them with fellowships for their Ph.D. studies.