A Critical Review on the Development of Optical Sensors for the Determination of Heavy Metals in Water Samples. The Case of Mercury(II) Ion

Abstract

Recent publications are reviewed concerning the development of sensors for the determination of mercury in drinking water, based on spectroscopic methodologies. A critical analysis is made of the specific details and figures of merit of the developed protocols. Special emphasis is directed to the validation and applicability to real samples in the usual concentration range of mercury, considering the maximum allowed limits in drinking water established by international regulations. It was found that while most publications describe in detail the synthesis, structure, and physicochemical properties of the sensing phases, they do not follow the state of the art in the analytical developments. Recommendations are provided regarding the proper method development and validation, including the setting of the calibration concentration range, the correct estimation of the limits of detection and quantitation, the concentration levels to be set for producing spiked water samples, the number of real samples for adequate validation, the comparison of the developed method with a reference technique, and other analytical features which should be followed.

1. Introduction

Safe and readily available water is important for public health.¹ Drinking water may contain small amounts of heavy metals, some of which are known to be essential for human health. However, an excess of essential metal ions, or even small quantities of other ions may have a serious negative impact.² The presence of heavy metals in water is due to several reasons: (1) the release into the environment by natural causes; (2) anthropogenic activities, such as industrial activities generating wastes (electroplating, metal smelting and chemical industries); or (3) poorly treated wastewaters.³

There is a growing number of publications devoted to the development of new spectroscopic platforms for the determination of inorganic ions in samples such as drinking and natural waters, foodstuff, etc., which are of high importance for consumer safety. The methods are based on either absorptimetric or emissive probes, which are developed in such a way that they are highly selective and proposed to be sufficiently sensitive to the presence of specific metal ions in the studied samples. The advantages of the proposed sensors include simplicity, low cost, and speed.⁴ Typically, the concentrations to be measured in the latter samples are small, requiring limits of detection which should be compatible with the maximum levels set by international regulating agencies. Therefore, efforts are directed to maximize the sensitivity of the analysis, by incorporating a series of clever approaches in the new developments.

Due to the large number of papers published in recent years on the subject, the present report is concerned with a specific case: the determination of mercury in drinking water. This analyte was selected because of the challenges in measuring very low concentrations due to its toxicity. However, most of the considerations regarding chemical analysis at low concentrations also apply to other heavy metal ions, anions and organic pollutants in environmental samples, foodstuff, etc.

The sources of mercury in water include the erosion of natural deposits, the discharge from refineries and factories, and the runoff from landfills and croplands. For metal ions and other contaminants, the US Environmental Protection Agency (EPA) sets two limits:⁵ (1) the Maximum Contaminant Level Goal (MCLG), which is the level of a contaminant in drinking water below which there is no known or expected risk to health; and (2) the Maximum Contaminant Level (MCL), which is the highest level of a contaminant that is allowed in drinking water and is an enforceable standard. MCLs are set as close to MCLGs as feasible using the best available treatment technology and taking cost into consideration. In the case of mercury, the MCLG and MCL are both 2 μg L^–1 (2 ppb, 10 nM).⁵ Below this level of mercury in drinking water, there is no known or expected risk to health. Otherwise, serious problems may arise, the main of which is kidney damage.⁶

The analytical method recommended by the EPA for the determination of mercury in drinking water is cold vapor atomic absorption spectrometry.⁷ The concentration range of the method is 0.2–10 μg L^–1 (i.e., the lowest calibration concentration is well below the MCL for Hg in this type of samples). It is recommended to estimate the method detection limit (MDL) as 3 times the standard deviation of 7 replicates of a sample prepared to contain twice the concentration corresponding to a rough estimation of the true LOD. However, modern IUPAC recommendations use a different concept for the LOD estimation, which will be discussed below in detail.^8,9

It is apparent that both the LOD and LOQ must be low enough to detect and measure mercury concentrations at the required levels according to official regulations. If the LOQ of a new method is larger than the MCL, then it should probably be called a method to detect Hg in highly contaminated water samples and may require a substantial preconcentration step for analyzing drinking water. Nevertheless, even if the detection capability of a method is adequate, it should be appropriately validated to be considered reliable. In the present report, we review a number of selected publications on the development of spectroscopic methods for determining Hg in water samples, with emphasis on the requirements to be met for achieving the intended purpose.

2. RECOMMENDATIONS ON FIGURES OF MERIT

Analytical reports should not only describe with particular detail the synthesis, structure, and physicochemical properties of a new chemosensor or probe for specific analytes. The analytical section should be as complete as possible, following the standard guidelines which include the figures of merit, the statistical analysis of the results, and the comparison with those provided by previously published methods. Below we summarize the main figures of merit that should be reported and recommendations for the correct statistical analysis and method comparison.

2.1. Sensitivity

The sensitivity and analytical sensitivity should be reported, preferably including a comparison with the same parameters in previous analytical reports on the same analyte. The sensitivity is defined as the slope of the calibration graph; however, the analytical sensitivity is preferred for method comparison because it is independent of the measured signal.^10,11 The latter is defined as the ratio of the sensitivity and the instrumental noise level and has inverse concentration units.

2.2. Linearity

Most authors still employ the correlation coefficient (R²) of the calibration line as an indicator of the linear relationship between signal and concentration. IUPAC discourages this practice, recommending the estimation of an F value as the ratio of the residual variance of the calibration line to the variance of the instrumental noise. Comparison should be made with the critical F value, as detailed in a IUPAC report.¹² Other relevant discussions can be found in the literature.¹³

2.3. Limits of Detection and Quantification

The definition of the LOD has been evolving with time, from the old concept based on three times the standard deviation for the blank, to the modern view considering both Type I and Type II errors, also called α and β, or false positives and false negatives.^7,14,15 Furthermore, error propagation should consider the uncertainties in the measurement of the test sample and also those coming from the calibration phase. Specifically, the modern IUPAC recommendation first requires to define the limit of decision (LD), which only considers a risk of false positives (Figure 1, green-shaded area). The limit of detection is then defined as the level for which the risk of false negatives has a probability β, corresponding to the red-shaded area in Figure 1. According to this latter figure, the expression for the LOD is

1

where t_α,ν and t_β,ν are student coefficients with ν = N – 2 degrees of freedom and probabilities α and β respectively, N is the number of calibration samples, σ_c,0 and σ_c,LOD are the concentration standard errors at the blank and LOD levels, m is the slope of the calibration line, and s_y/x is the residual standard deviation. Assuming σ_c,0 = σ_c,LOD, the 95% confidence level (α = β = 0.05) and a large number of degrees of freedom, the right-hand side of eq 1 is obtained, where h₀ is the blank leverage value given by

2

where c̅_cal is the mean calibration concentration and c_n represents the concentration of the analyte in the n^th calibration sample.

Figure 1.

Scheme illustrating the modern IUPAC definition of the limits of decision (LD) and detection (LOD). Gaussian lines are shown centered at the blank and at the LOD level. The LD decides whether the analyte is detected or not with a rate α of false positives (green shaded area), whereas the LOD implies detection with a rate α of false positives and a rate β of false negatives (red shaded area). Adapted with permission from ref (8). Copyright 2015 Elsevier.

In some cases, authors refer to the IUPAC definition of the LOD but still use the old definition. The following important remarks by IUPAC should be considered: “It is accordingly recommended that...the approximate detection limit (is) calculated as 3S0. Note that with the recommended minimum number of degrees of freedom, this value is quite uncertain, and may easily be in error by a factor of 2. Where more rigorous estimates are required (e.g., to support decisions on based on detection or otherwise of a material), reference should be made to appropriate guidance (see, e.g., refs 22, 23).” In these latter references the new definitions of the LOD and LOQ are provided.^16,17

Another relevant issue is the use of the log transformation for coping with nonlinear signal-concentration relationships. However, it is preferable to employ the nonlinear calibration line, estimating the LOD according to a recent procedure, adapted from the IUPAC recommendations for linear calibration lines.¹⁸

Due to the presence of additional terms in the square root of eq 1 beyond the classical value of 1, the modern LOD definition is stricter (i.e., larger) than the old one. It is important to notice that all the reviewed publications quote the incorrect detection capability based on the abandoned LOD definition. They are consequently lower (perhaps significantly) than the true LOD values.

Even more important than the LOD is perhaps the limit of quantitation (LOQ) defined as

3

where the factor 10 sets a relative prediction uncertainty of 10%. The relevance of the LOQ is that the linear range of a developed method for the determination of mercury ranges from the LOQ to the upper concentration limit where the linear signal-concentration relationship holds. To be able to quantitate Hg below the levels set by official agencies, a sufficiently low LOQ is required.

2.4. Significant Figures

It is not uncommon to find reported analytical results with an excessive number of significant figures, which is unreasonable. In general, all results should be reported with a number of significant figures compatible with the associated standard error. In the case of uncertainty measures (standard errors, root-mean-square errors, relative prediction errors) and also in the case of parameters derived from uncertainties (detection limit, quantitation limit), they should be reported with one or at most two significant figures. A good rule of thumb is to use two significant figures when the first one is 1, or when the first is 2 and the second is smaller than 5; in the remaining cases, a single significant figure should be reported.⁸

2.5. Statistical Analysis of the Results

The number of studied samples is usually too small to gather reliable information on the applicability of the method. A new analytical development should include the validation of the protocol with a number of test samples (preferably certified reference materials or real samples, not artificially laboratory-prepared or spiked samples), comparing the determined concentrations of the analyte with those provided by a reference standard technique. The estimated recovered values should be compared with reference values. Statistical tests should be applied to assess whether a recovery is not statistically different than 100%, reporting both the experimental and critical t values. It should be noticed, however, that these tests assume certain conditions that the data should fulfill, such as constant variance.^19,20

2.6. Comparison of Average Errors

When comparing the average analyte errors estimated by different methods, the conclusions are often drawn on a visual basis. This impression may be false, and an objective statistical technique is required to compare error values, to be able to conclude that one error is significantly smaller than a second one. One interesting alternative is the randomization test described by van der Voet.²¹ The latter paper provides an easily implementable computer code in its Appendix.

3. CHEMICAL SENSORS FOR MERCURY IONS

The conventional analytical approaches for Hg(II) analysis include atomic absorption spectroscopy, inductively coupled plasma mass spectrometry, and atomic fluorescence spectrometry. Because these methods are rather complex, expensive, nonportable, and require specialized laboratories, the current scientific efforts are directed to developing efficient chemical sensors as alternatives for the determination of mercury in environmental samples. In general, these sensors display analytical features that make them attractive and useful to be implemented both in the analytical laboratory and in the field. For the present discussion, we have selected and analyzed some reported optical mercury sensors where absorption, fluorescence, and/or surface enhanced Raman scattering are the measured signals.

Most optical sensors for the detection of Hg(II) in aqueous media involve the use of noble metal nanostructures due to their unique physicochemical properties. For example, they have the ability to support surface plasmons which are generated by the coupling between the incident electromagnetic waves and the conduction electrons.²² In other words, when the electromagnetic radiation of a certain frequency interacts with the metal nanoparticles, their conduction electrons collectively oscillate on a metal/dielectric interface. This resonance is called surface plasmon resonance (SPR) or, more precisely, local SPR (LSPR) because nanoparticles are involved. Gold- and silver nanoparticles, either bare or with functionalized surfaces, are the two most widely used plasmonic nanomaterials, which have demonstrated sensitive LSPR signals used for the selective detection of Hg(II) ion.^22,23 The LSPR spectra can be modified by tailoring the size, the shape (rod, cube, disc, etc.), the structures (shell, cage, tube, etc.), and the environment of the metal nanostructures. The LSPR sensing mechanism can be produced through the aggregation of the nanoparticles or by changes in their refractive index.²⁴ The aggregation of nanoparticles, generally functionalized with oligonucleotides, oligopeptides, or chemical functional molecules, is promoted by the Hg(II) ion through its affinity with specific functional groups of the surface, producing a shift of the LSPR band as a result of the strong coupling between the localized SPRs of the nanostructures. On the other hand, refractive index sensors are based on the LSPR peak shift due to the variation in the local refractive index induced by a mercury amalgamation on the nanoparticle surface. In this case, the Hg(II) ion is reduced with an adequate reagent (NaBH₄, citrate, etc.) and the generated Hg(0) is strongly bonded onto the surface of either Au- or Ag- based nanomaterials to form an amalgam, shifting the LSPR spectrum of the nanoparticles to shorter wavelengths.²⁵

An alternative colorimetric assay is based on the inhibition of the peroxidase-like activity of noble metal nanostructures by the Hg(II) ion. The reaction between the chromogenic substrate 3,3′,5,5′-tetramethylbenzidine (TMB) and H₂O₂ is generally chosen for these probes. When the catalytic activity of the nanostructure decreases due to the presence of the Hg(II) ion, the concentration of the oxidation product decreases, leading to a color change of the system.^26,27

In other cases, the presence of mercuric ion produces the opposite effect to that described above. For example, the formation of either the stable thymine complex (T-Hg²⁺-T) or amalgams with noble metals, can reverse the inhibitory effect that certain molecules or nanoparticles have on the oxidase mimetic activity of specific nanostructures. In the presence of Hg(II) ions, the oxidase-like activity of the indicated nanostructures is restored and the concentration of the oxidation product increases.^28,29

Fluorescent sensors are constituted by nanomaterial-bound fluorescent units (the fluorophores). When the cation binds to the sensor, the fluorophore photophysical properties change due to different processes (photoinduced electron transfer or charge transfer, energy transfer, excimer formation, etc.), leading to either a fluorescence intensity enhancement or to a quenching effect.²³

Another class of optical sensors for Hg(II) ion resort to surface enhanced Raman spectroscopy (SERS), combining laser spectroscopy with the optical properties of nanosized noble metal structures, resulting in significantly increased Raman signals.³⁰ The SERS methods for the detection of Hg²⁺ are based on either the specific interaction between the ion and Raman active molecules exhibiting a strong and specific SERS spectrum (reporters) or the formation of the T-Hg²⁺-T complex. The detection of the Hg(II) ion is achieved by enhancing or turning off the SERS signals of the Raman reporters. The DNA-based T-Hg²⁺-T system relies on the coordination chemistry of the metal ion with the nucleotide and the formation of nanoparticles aggregates with the concomitant changes in the SERS intensity.

Some research groups have proposed the mercury determination by simultaneously measuring two types of signals with the same sensor, one detection mode generally being superior to the other. However, in some of these dual methods one of the two types of signal results in an unfeasible LOD for the determination of low levels of Hg(II) ion. Thus, it would be more correct to characterize the method with the signal that allows achieving the analytical purpose rather than as a dual one.

4. REVIEW ON MERCURY SENSOR REPORTS

As indicated above, numerous scientific articles report the determination of the mercury(II) ion in water samples. For the sake of brevity, we have randomly selected a number of publications on optical mercury sensors. The information from the literature search was compiled in Tables 1–4 following an increasing order of publication year and were organized according to the measured spectroscopic signals, as detailed in Section 3. Thus, whereas Tables 1, 2 and 3 describe sensors based on absorption, fluorescence, and SERS signals, respectively, Table 4 includes sensors displaying two different types of signals.

Table 1. Hg(II) Ion Determination in Water Samples Using Absorptimetric Signals.

probe/sensor fundamentals	calibration grapha	LR (μg L–1)	LOD (μg L–1)	samples/evaluated CHg(II)	ref
Hg interacts with AuNRs in a NaBH4 solution. CHg(II) is directly related with the blue shift of the maximum absorption wavelength of the longitudinal mode band of AuNRs. LOD after amalgam formation is ca. parts-per-trillions.	(Δλfd)b vs CHg(II)	1.98 × 10–6–0.031	(6.6 × 10–7)c	One tap water sample (multiple standard addition method). Original Hg(II) ion concentration: 1.67 × 10–3 μg L–1. No sample separation or preconcentration required.	(33)
NPs suspension of Hg(II) ion-dithizone complex is filtered and quantified on the surface of a WX mixed cellulose membrane through UV–visible DRS. The colored Hg–dithizone complex is directly measured on the surface of membrane	A485 vs CHg(II)	0.2–4.0	0.12c	Tap and river waters filtered through 0.22 μm membranes were treated with KMnO4 and K2S2O8. Three samples of each water were spiked with 0.5; 1.5 and 3 μg L–1 Hg(II) ion. AFS was applied.	(34)
Hg(II) ion induces both deprotection and morphology transition of 1-dodecanethiol-capped AgNPRs in the presence of iodide, producing a color change of the system. The Δλ of the SPR band is used for quantifying Hg(II) ion.	Δλ vs CHg(II)	2–100	0.66c	Tap, drinking, and lake waters were filtered through 0.22 μm membranes. Three samples of each water were spiked with 0; 40 and 80 μg L–1 Hg(II) ion.	(35)
Hg(II) ions are reduced by Ag-AuNPs electrodeposited on indium tin oxide film coated glass. The deposition/amalgamation leads to a blue shift of the SPR peak.	Δλ vs CHg(II)	0.05–500	0.02c	Hg(II) levels found in tap (0.61 μg L–1) and lake (0.87 μg L–1) waters were close to CVAAS (0.67 and 0.83 μg L–1, respectively). No sample separation or preconcentration required.	(24)
Aggregation of citrate-stabilized AuNPs in the presence of Hg(II) ion and lysine (positively charged) induces a color change of the system. Lysine is used as an aggregation promoter of Hg(II) ion-covered AuNPs.	A725/A525 vs CHg(II)	0.2–200	0.58c (dw),0.78c (tw)	Two tap water samples were spiked with 1 μg L–1 Hg(II) ion.	(36)
Hg(II) interacts with an aqueous dispersion of cysteamine stabilized AgNPs, forming a Ag–Hg nanoalloy, and the system changes from yellow to colorless	A402 vs CHg(II)	0.055–2.7	0.055	Lake, tap and packaged drinking waters were filtered through Whatman-1 paper. One sample of each water was spiked with 1 μg L–1 Hg(II) ion.	(37)
AuNPs/Hg amalgam catalyzes the reduction of 4-NP by NaBH4. As CHg(II) increases, the reduction time decreases. When Hg deposition reaches saturation, AuNPs aggregation occurs, the reduction time increases, and the color changes.	CCT400 vs log CHg(II)	0.2–20 (CCT ↓)	0.29c	One tap water spiked with 0.2; 1; 2; 10; 20; 60; 100; 160 and 200 μg L–1 Hg(II) (standard addition method). ICP-MS was also applied.	(38)
		20–160 (CCT ↑)
Hg(II) ion increases the absorbance of self-doped copper-deficient Cu2-xSe NPs system at 400–600 nm, while the NIR localized SPR band decreases and a red shift is detected. Changes are produced due to Hg(II) ions replace Cu+/Cu2+ ions forming a HgSe layer around the NPs.	A450/A920 vs CHg(II);	0–160	1.6c	Tap, pond, and river waters were filtered through a 0.22 μm filters. Three samples of each water were spiked with 0; 2 and 6 μg L–1 Hg(II) ion, respectively.	(39)
	Δλ vs CHg(II);
	A920 vs CHg(II) (for quantification)
Hg(II) inhibits the peroxidase-like activity of DNA-Ag/PtNCs on the oxidation of TMB by H2O2 due to the nanocluster aggregation and the decrease of Pt2+ surface content, resulting in the decrease of the absorption band at 652 nm.	A652 vs CHg(II)	2–40	1c	Samples of tap water, filtered through 0.22 μm membranes, were spiked with 4; 10; 20 and 40 μg L–1 Hg(II) ion.	(26)
AuNCs catalyze the reduction of HAuCl4 by H2O2 leading to AuNPs formation and partial aggregation. Hg(II) inhibits the catalytic activity and the AuNPs formed by residual H2O2 remain disperse, producing a color change. TSC accelerates the AuNPs formation.	A537/A650 vs CHg(II)	0.02–2000	1.78 × 10–3	Lake and river waters were filtered through 0.45 μm membranes. Two samples of each water were spiked with 10 and 20 μg L–1 Hg(II).	(40)
Hg(II) ion produces a color change in the thiamine modified AgNPs aggregate that is spectrophotometrically measured	A550/A395 vs CHg(II)	2–1000	1c	One river water spiked with 0; 4; 460 and 1000 μg L–1 Hg(II) ion (standard addition method).	(41)
Hg(II) ion interacts with bimetallic Ag-CuNPs producing changes in the SPR absorption. As CHg(II) increases, the band at 406 nm of the Ag-CuNPs absorption spectra decreases, and a second absorption peak at 504 nm is detected	A504/A406 vs CHg(II)(Milli-Q, Mengxi lake);A534/A404 vs CHg(II)(Chagan lake)	0.2–2	0.10c; 0.16c; 0.11c for Milli-Q, Mengxi, and Chagan waters	Lake waters were filtered through 0.22 μm membranes. Five samples of each lake were spiked with Hg(II) ion in the range 0.2 to 2 μg L–1	(42)
Hg(II) ion interacts with AuNPs encapsulated in porous wood membrane forming a gold amalgam. This system catalyzes the reduction of MB by NaBH4 and a decrease of the MB absorption peak is observed as CHg(II) increases.	A662 vs CHg(II)	0–2	6.4 × 10–3	Tap, deionized, and river waters. Three samples of each water spiked with 0.6; 1.2 and 1.8 μg L–1 Hg(II) ion, respectively.	(43)
Hg(II) ion inhibits the peroxidase-like activity of CuO/Pt nanoflowers on the oxidation of TMB by H2O2 due to the CuO/Pt–Hg trimetallic amalgam formation, resulting in a decrease of the absorption at 650 nm.	A650 vs CHg(II)	0.02–140	(9.8 × 10–3)c LOQ 0.024 μg L–1 (gw)	Groundwater samples (filtered through 0.45 μm syringe filters) were spiked with 20; 30; 40; 50; 60; 70; 80; 90; 100; 110 and 120 μg L–1 Hg(II) ion. CVASS was also applied.	(27)
Hg(II) ion interacts with Pzl through sulfur and nitrogen groups avoiding the aggregation of AuNPs, leading to both a decrease and an increase of absorbances at 680 and 520 nm, respectively.	ΔA680/A520 vs CHg(II)	2–300	1.54c	Five tap and five river waters (filtered with paper) spiked with 10; 80; 160; 200 and 240 μg L–1 Hg(II) ion. ICP-MS was also applied.	(44)
The T-Hg2+–T complex formed by interaction of Hg with ssDNA oligonucleotide avoids the adsorption of free oligonucleotide on graphene oxide/AuNPs nanocomposite, preserving catalytic activity in the oxidation of TMB by H2O2.	A655 vs CHg(II)	1.04–24	0.076	Three tap and three lake waters (previously dechlorinated by boiling) were spiked with 1.8; 12 and 60 μg L–1 Hg(II) ion. Three river waters were previously digested. Found values (verified by CVAAS) were: 0; 1.27 and 1.61 μg L–1.	(28)
SnTeNBs were grafted with AuNPs to form a SnTe/Au hybrid, resulting in depression of the oxidase-like activity, which is then recovered in the presence of Hg(II) due to a gold amalgam formation. Chloride ions are also determined, but [Cl–] > 3 mM (e.g., seawater) interferes in the Hg(II) determination.	A652 vs CHg(II)	40–11600	30	One tap water was spiked with 0; 1200; 2400; 5000; 7000 and 9000 μg L–1 Hg(II) ion.	(29)
DNAmb is immobilized on an optical fiber surface forming a stem-loop structure. In the presence of Hg(II) ion and DNAa, the T-Hg2+–T complex formation opens the stem helix in DNAmb allowing the binding of its terminal end with DNAr conjugated to AuNPs (AuNP-DNAr), producing nanoplasmonic absorption	(I0–IS)/I0 vs log CHg(II)d	1 × 10–5-0.2	(8.76 × 10–7)c	Two drinking and two tap waters were spiked with 0.02 and 0.2 μg L–1 Hg(II) ion.	(45)
Hg(II) ion induces aggregation of AuNPs modified with two peptides, generating SPR absorption and color change	A720/A530 vs CHg(II)	20–160	5.6	Lake waters, filtered through 0.45 μm membranes, were spiked with 60; 80 and 100 μg L–1 Hg(II) ion	(46)
Hg(II) ion produces a shift in the SPR wavelength of a sensing Au film due to the formation of a sandwich structure (Au/T-Hg2+-T/Au) between T-modified AuNPs and T immobilized on the Au film surface of a fiber-optic	SPR shift vs CHg(II)	16–40	2c	Tap waters were spiked with 16, 20, and 40 μg L–1 Hg(II) ion	(47)

^aSubscripts for A and CCT are the wavelengths in nm. Δλ is the wavelength shift.

^bWavelength difference of the first derivative of the absorption spectrum.

^cLOD = 3S_b/m, where S_b is the standard deviation of the blank signal and m is the slope of calibration line.

^dI₀ and I_S are the transmitted light intensity of the sensor in a blank and in a sample containing Hg(II) ion, respectively. Abbreviations: A, absorption; AFS, atomic fluorescence spectrometry; AgNPs, silver nanoparticles; AgNPRs, silver nanoprisms; AuNCs, gold nanoclusters; AuNPs, gold nanoparticles; AuNRs, gold nanorods; C_Hg(II), concentration of Hg(II) ion; CCT, color-change time; CVAAS, cold vapor atomic absorption spectrometry; dw, distilled water; DNA_a, free assisting oligodeoxyribonucleotide probe; DNA_mb, oligodeoxyribonucleotide probe as a molecular beacon; DNA_r, reporting oligodeoxyribonucleotide probe; DRS diffuse reflectance spectroscopy; gw; groundwater; ICP-MS, inductively coupled plasma mass spectrometry; LOD, limit of detection; LOQ, limit of quantitation; MB, methylene blue; NaBH₄, sodium borohydride; NCs, nanoclusters; NIR, near-infrared; 4-NP, 4-nitrophenol; NP, nanoparticle; SnTeNBs, tin telluride nanobelts; ssDNA, single-stranded DNA; Pzl, 3,5-dimethyl-1-thiocarboxamidepyrazole; SPR, surface plasmon resonance; T, thymine; tw, tap water; TMB, 3,3′,5,5′-tetramethylbenzidine; TSC, trisodium citrate.

Table 4. Hg(II) Ion Determination in Water Samples by Using Dual Spectroscopic Signals.

probe/sensor fundamentals	calibration grapha	LR (μg L–1)	LOD (μg L–1)	samples/evaluated CHg(II)	ref
Fluorimetric/Absorptimetric Modes of Detection
Hg(II) ion inhibits the PtNPs catalytic activity on OPD oxidation with H2O2 due to a Pt–Hg amalgam formation in the NP surface. Both the fluorescence intensity and the yellow color of the system decrease as the CHg(II) increase.	F555 vs CHg(II)	2–400 (Flu)	0.028b (Flu)	Two river water samples were centrifuged and filtered through 0.45 μm membranes, and then they were spiked with 10 and 20 μg L–1 Hg(II) ion.	(63)
	A420 vs CHg(II)	2–400(Abs)	0.16b(Abs)
Hg(II) ion produces both fluorescence enhancement and peroxidase-like activity decrease of FAM-labeled Hg2+ aptamer adsorbed on P-CeO2NRs. The T-Hg2+–T complex formation induces the separation of FAM-apt from the NRs surface, resulting in the recovery of FAM-apt fluorescence. The peroxidase-like activity of P-CeO2 NRs decreases and the color of TMB-H2O2 solution turned to light blue.	F520 vs log CHg(II)	0.016–2.5 (Flu)	0.0158b (Flu)	Canal waters were filtered through 0.22 μm membranes. Two samples were spiked with 0.6 and 2 μg L–1 Hg(II) ion (Flu and Abs); one sample was spiked with 0.06 μg L–1 Hg(II) ion (Flu); and one sample was spiked with 6 μg L–1 Hg(II) ion (Abs).	(64)
	A650 vs log CHg(II)	0.1–20(Abs)	0.062b(Abs)
Absorptimetric/SERS Modes of Detection
Hg(II) ion reverses the Bismuthiol II-induced aggregation of AuNPs, resulting in both UV–vis and SERS spectral changes.	A780/A530 vs CHg(II)	0–30(Abs)	0.4b(Abs)	One tap and one spring water samples were each spiked with 18 μg L–1 Hg(II) ion (Abs). CVAFS was also applied.	(65)
	I1364 vs CHg(II)	0–30 (SERS)	6b (SERS)
Hg(II) ion inhibits the oxidation of TMB by H2O2 catalyzed by Au core AgPt shell NPs modified with mercaptohexanol. As CHg(II) increases, the generation of the oxidized form of TMB (oxTMB) decreases, resulting in both a color change of the system and a decrease in the intensities of the SERS bands of oxTMB.	A652 vs log CHg(II)	200–20,000(Abs)	104b(Abs)	Six ultrapure waters were spiked with 0.4; 4 and 40 μg L–1 Hg(II) ion (SERS), and 1600; 4000 and 16 000 μg L–1 Hg(II) ion (Abs). Five filtered river waters were spiked with 0.4; 4 and 40 μg L–1 Hg(II) ion (SERS), and 1,600 and 16,000 μg L–1 Hg(II) ion (Abs).	(66)
	I1608 vs log CHg(II)	0.2–2000 (SERS)	0.056b (SERS)
Hg(II) ion suppress the etching effect of S2O32– on silver shells of Ag-coated AuNPs due to the HgS formation. Both the absorption at 394 nm and the Raman intensity of R6G increases as the CHg(II) increase.	ΔA394 vs CHg(II)	200–1000(Abs)	40 (Abs)	One tap water sample was spiked with 400 μg L–1 Hg(II) ion (Abs), and with 2 μg L–1 Hg(II) ion (SERS). One lake water sample, filtered through a 0.22 μm membrane, was spiked with 400 μg L–1 Hg(II) ion (Abs), and with 2 μg L–1 Hg(II) ion (SERS).	(67)
	ΔI1362 vs –log CHg(II)	0.02–200 (SERS)	0.02 (SERS)

^aSubscripts for F and A indicate the wavelength in nm, those for I the wavenumber in cm^–1.

^bLOD = 3S_b/m, where S_b is the standard deviation of the blank signal and m is the slope of calibration line. Abbreviations: A, absorbance; Abs, absorptimetric approach; AuNPs, gold nanoparticles; Bismuthiol II, potassium 5-mercapto-3-phenyl-1,3,4-thiadiazole-2-thione; C_Hg(II), concentration of Hg(II) ion; CVAFS, cold vapor atomic fluorescence spectrometry; F, fluorescence intensity; FAM, carboxyfluorescein; Flu, fluorimetric approach; LOD, limit of detection; LR, linear range; NPs, nanoparticles; OPD, o-phenylenediamine; P-CeO₂NRs, porous cerium oxide nanorods; PtNPs, platinum nanoparticles; R6G, rhodamine 6G; SERS, surface-enhanced Raman scattering; TMB, 3,3′,5,5′-tetramethylbenzidine.

Table 2. Hg(II) Ion Determination in Water Samples by Using Fluorimetric Signals.

probe/sensor fundamentals	calibration grapha	LR (μg L–1)	LOD (μg L–1)	samples/evaluated CHg(II)	ref
AuNPs modified with thioglycolic acid quench the rhodamine B fluorescence, and the signal is restored in the presence of Hg(II) ion.	F581 vs CHg(II)	0.2–6.2	0.08b	One tap, one pond, and two river samples were filtered through paper. Tap and pond waters were spiked with 1 μg L–1 Hg(II), and river samples with 2 μg L–1 Hg(II). Original found CHg(II) (verified by AAS) were 0.44 and 1.1 μg L–1 (tap and pond waters, respectively), and 12.2 and 10.9 μg L–1 (river samples).	(31)
Hg(II) ion displaces RBITC from the surface of RBITC-PEG-modified AuNPs resulting in an enhancement of RBITC fluorescence initially quenched by AuNPs.	F/Fo vs CHg(II)	2–100	0.46 (pure water)b	River water samples, filtered through 0.22 μm membranes, were spiked with 0; 2; 4; 8; 16; 30 and 60 μg L–1 Hg(II) ion.	(48)
			0.76 (river water)b
Nonfluorescent system formed by FAM-labeled ssDNA probe bound to another ssDNA probe adsorbed on WS2 nanosheets shows fluorescence in the presence of Hg(II) ion. FAM-labeled ssDNA probe and the other oligonucleotide probe hybridize through the T-Hg2+-T complex and is digested by T7 exonuclease, forming short FAM-oligonucleotide fluorescent fragments that are not adsorbed by the nanosheets.	F vs CHg(II)	0.1–4	0.02b	Tap and lake waters. Two samples of each water type were spiked with 0.4 and 2 μg L–1 Hg(II) ion (standard addition method).	(49)
Hg(II) ion replaces Mn(II) ion in the surface of nonemissive Mn-doped ZnSe QDs with small size (2.1 nm), and fluorescence is detected around 600 nm.	F600 vs CHg(II)	0–900	1.4b	Tap and lake waters were filtered with a 0.22 μm syringe filters. Four samples of each water type were spiked with 20; 100; 400 and 800 μg L–1 Hg(II) ion. ICP-AES was also applied.	(50)
Hg(II) ion quenches the fluorescence of l-cysteine-modified AuNPs. The fluorescence peak at 625 nm decreases as CHg(II) increases.	F625 vs log CHg(II)	0.4–6000	0.26b	One tap and one river water samples, filtered through a membrane, were spiked with 1; 2 and 4 μg L–1 Hg(II) ion (standard addition method). Original found CHg(II) (verified by AFS) were 0.99 and 2.40 μg L–1.	(32)
Nanocomposites formed by AgNP and reduced graphene oxide quench the rhodamine B fluorescence, and the signal is restored in the presence of Hg(II) ion.	F/F0 vs CHg(II)	2.5–40	0.4c	Mineral, tap, and pond waters. Two samples of each water type were spiked with 10 and 20 μg L–1 of Hg(II) ion.	(51)
The fluorescence of CDs@Eu-MOFs suspension at 430 nm decreases with the increasing of Hg(II) ion concentration, while at 614 nm the signal is almost constant.	F430/F614 vs CHg(II)	0–60,000	0.024c Lowest CHg(II) ≅ 2000 μg L–1	Three tap and three river water samples, filtered through a 0.22 μm filters, were spiked 1000; 2000 and 3000 μg L–1 Hg(II) ion.	(52)
Hg(II) ion quenches the fluorescence of nanocrystals formed by cesium, lead, bromide perovskite and methoxypolyethylene glycol/silica composite (CsPbBr3-mPEG@SiO2). Then, the fluorescence is restored through the interaction between Hg(II) ion and GSH. Fluorescent probe can also be used for GSH detection.	F0–F520/F0 vs CHg(II)	0.02–10	0.016	Tap water samples spiked with 2, 4, and 8 μg L–1 of Hg(II) ion.	(53)
Hg(II) ion interacts with a biosensor integrated with a selected FNA with T-rich DNA sequence and, due to the T-Hg2+–T complex formation and through FRET mechanism, the fluorescence signal is increased. Pb(II) ion is simultaneously detected using RNA-cleaving Pb2+-specific DNAzyme as FNA.	F vs CHg(II)	0–30	0.844b	No pretreatment needed for tap, bottled, lake, and underground waters. Three samples of each type of water were spiked with 30; 52 and 90 μg L–1 Hg(II) ion. ICP-AES was also applied.	(54)
	The first part of sigmoidal plot is used for calibration.
LYSO-QF probe forms tetrameric, trimeric, and dimeric complexes at different CHg(II) ranges giving three fluorescent emission ranges depending on the Hg(II) concentration.	F495 vs CHg(II) (signal ↓);F512 vs CHg(II) (signal ↑);F694 vs CHg(II) (signal ↑)	0–600; 1400–2300; 2800–4100	2.2b (F495 vs CHg(II) was used). Lowest CHg(II) ≅ 100 μg L–1	Two samples of each water (tap, pond, and river waters) were spiked with 2 and 5 μg L–1 Hg(II) ion. AAS was also applied.	(55)

^aSubscript for F is the emission wavelength in nm.

^bLOD = 3S_b/m, where S_b is the standard deviation of the blank signal and m is the slope of calibration line.

^cLOD = 3S_I/m, where S_I is the standard deviation of the intercept and m is the slope of calibration line. Abbreviations: AAS, atomic absorption spectroscopy; AFS, atomic fluorescence spectrometry; AgNPs, silver nanoparticles; AuNPs, gold nanoparticles; C_Hg(II), concentration of Hg(II) ion; CDs@Eu-MOFs, carbon dots-embedded europium metal–organic frameworks; F, fluorescence intensity; FAM, fluorescein amidite; FNA, functional nucleic acid; FRET, fluorescence resonance energy transfer; GSH, glutathione; ICP-AES, inductively coupled plasma-atomic emission spectrometry; LOD, limit of detection; LR, linear range; LYSO-QF, (Z)-N-((Z)-2-(1,3,3-trimethylindolin-2ylidene)ethylidene)quinoline-8-amine; NP, nanoparticle; PEG, poly(ethylene glycol); QDs, quantum dots; ssDNA, single-stranded DNA; RBITC, rhodamine B isothiocyanate; T, thymine; WS₂, tungsten disulfide.

Table 3. Hg(II) Ion Determination in Water Samples by Using Surface-Enhanced Raman Scattering Signal.

probe/sensor fundamentals	calibration grapha	LR (μg L–1)	LOD (μg L–1)	samples/evaluated CHg(II)	ref
Through the strong mercury and sulfur bond, Hg(II) ion prevents the adsorption of MESNA on the surface of AgNPs resulting in a decrease of SERS of the MESNA modified AgNPs.	ΔI795 vs CHg(II)	2–400	0.48b	Water samples of two different lakes and groundwater samples were spiked with 40 and 240 μg L–1 Hg(II) ion.	(56)
Hg(II) ion interacts with IP6 stabilized AuNPs modified by CV and TC, decreasing the intensity of SERS band of the probe at 1173 cm–1 due to the CV molecules detachment from the surface.	I1173/I0 vs CHg(II)	0.1–1	1 × 10–4	Five tap and five river water samples were spiked with 0.4 μg L–1 Hg(II) ion.	(57)
Hg(II) ion changes the SERS signal of DNA modified AuNPs and 4-nitrothiophenol system, through the transformation of ssDNA into double helical DNA by T–Hg2+–T interaction. Au nanochains length grew longer with increasing the CHg(II) ion, and the increase of SERS signal is detected.	I1344 vs CHg(II)	0.001–0.5	(4.5 × 10–4)b	Three drinking water samples were spiked with 0.04; 0.08 and 0.15 μg L–1 Hg(II) ion.	(58)
Nanostructured gold substrate functionalized with a crown ether formed by ADB18C6 and MPA as SERS reporter is employed. The binding of Hg(II) ion into the crown ether layer cavity increases the intensity of Raman band at 1501 cm–1.	I1501 vs log CHg(II)	0.002–200	(1.02 × 10–4)b	One tap water sample was spiked with 5 μg L–1 Hg(II) ion.	(59)
Surface modified AuNRAs with 4-MPBA as SERS reporter are employed. In the presence of Hg(II) ion spectral changes of reporter including the disappearance, emergence, and shift of Raman peaks are detected.	log I vs –log CHg(II)	0.02–2000	ca. 0.02 by visual inspection	One lake water sample and one groundwater sample, each spiked with 2 μg L–1 Hg(II) ion. AFS was applied for results comparison.	(60)
	I measured at 467, 1048, 1085, and 1536 cm–1
Hg(II) ion interacts with surface-stabilized HEPES-AuNSs on a paper substrate forming a AuHg amalgam. The substrate is incubated in a HAuCl4, HCl and H2O2 solution with 4-MBA (reporter). SERS bands decrease as the CHg(II) increases.	peak area (1590)vs −log CHg(II)	0.02–200	0.006b	Three seawater samples and three pond water samples, each spiked with 0.2; 2 and 20 μg L–1 Hg(II) ion.	(61)
			0.066 (seawater)b
			0.046 (pond water)b
A chip formed by AuNPs immobilized on aminopropyl trimethoxysilane modified ITO surface and assembled GSH molecules was constructed, and 4-Mpy was used as SERS reporter. The interaction of Hg(II) ion with the chip GSH molecules and 4-Mpy increases the intensity of SERS peaks of 4-Mpy. CHg(II) is measured through the 1093 cm–1 band intensity.	ΔI1093 cm–1 vs –log CHg(II)	0.002–200	1.76 × 10–3	Five seawater samples were filtered through filter paper and 0.22 μm microporous filters. Samples were spiked with 0; 0.002; 0.02; 0.2 and 2 μg L–1 Hg(II) ion. Samples with 0.2 and 2 μg L–1 Hg(II) also were investigated with ICP-MS.	(62)

^aSubscript for I is the wavelength in cm^–1.

^bLOD = 3S_b/m, where S_b is the standard deviation of the blank signal and m is the slope of calibration line. Abbreviations: A, absorbance; ADB18C6, aminodibenzo-18-crown-6 derivative; AFS, atomic fluorescence spectrometry; AgNPs, silver nanoparticles; AuNPs, gold nanoparticles; AuNRAs, Au nanorod arrays; AuNSs, gold nanostars; C_Hg(II), concentration of Hg(II) ion; CV, violet crystal; GSH, glutathione; HEPES, 2-[4-(2-hydroxyethel) piperazine-1-yl]ethanesulfonic acid; I, intensity; ICP–MS, inductively coupled plasma mass spectrometry; IP₆, inositol hexaphosphate; ITO, indium tin oxide; LOD, limit of detection; LR, linear range; MPA, mercaptopropionic acid; 4-MBA, 4-mercaptobenzonic acid; MESNA, sodium 2-mercaptoethanesulfonate; 4-MPBA, 4-mercaptophenylboronic acid; 4-Mpy, 4-mercaptopyridine; NIR, near-infrared; NPs, nanoparticles; SERS, surface-enhanced Raman scattering; ssDNA, single-stranded DNA; T, thymine; TC, trisodium citrate; TMB, 3,3′,5,5′-tetramethylbenzidine.

The first and second columns of these tables show the operational basis of the sensors and the relationship between the measured signal and the mercury concentration. In the next column, the analytical efficiency of each of the proposed methods is briefly mentioned, with an emphasis on both the linear range and the attained limit of detection. Regarding the values of these latter parameters in the tables, it is important to point out that none of them was calculated using the modern IUPAC definition (see above). Nevertheless, to assess and compare the detection capabilities of the sensors, the reported LODs were used in our discussion. To facilitate the comparison among the different works, the concentration units were unified to micrograms per liter (parts-per-billion).

The tables also include the type of studied water sample, the applied pretreatment (if any), the added mercury concentrations to the spiked samples and/or the original values found in real samples without addition of the analyte. Since in these latter samples the Hg(II) ion content is unknown, it is mandatory to compare the obtained results with those provided by a reference method.

From the 42 selected methods shown in the tables, only two of them meet the correct analytical standards for the validation protocol in real samples.^28,31 Qi et al.²⁸ developed a colorimetric mercury sensor blocking with the Hg(II) ion the inhibition produced by oligonucleotides on the peroxidase-mimicking catalytic activity of graphene oxide/AuNPs in the oxidation of 3,3′,5,5′-tetramethylbenzidine (TMB) by H₂O₂. Under the optimized conditions, a calibration line was built with the absorbance at a proper wavelength as a function of the mercury ion concentration. Seven replicate calibration samples containing Hg(II) concentrations from 1.04 μg L^–1 to 24 μg L^–1 were employed. Although the limit of detection (LOD = 0.076 μg L^–1) was calculated with the old LOD concept, the value was below the US EPA maximum level in drinking water of 2 μg L^–1. The method was successfully applied in both spiked and real samples without externally added analyte, and the estimated concentrations were compared with those provided by a reference method.

On the other hand, Zheng et al. quantified mercury in tap, pond, and river waters using a fluorescence sensor.³¹ The latter was based on the capability of the Hg (II) ion to remove the quenching effect AuNPs modified with thioglycolic acid on the fluorescence of rhodamine B by disrupting the absorption of the dye molecules on the surface of the nanoparticles. The calibration was performed under optimal experimental conditions, measuring the fluorescence intensity at the emission maximum as a function of analyte concentration, ranging from 0.2 to 6.2 μg L^–1. The LOD calculated as the ratio between three times the standard deviation of the blank signal, and the slope of the calibration line was 0.08 μg L^–1. This approach allowed determining the original Hg(II) ion levels in real water samples (verified by atomic absorption spectroscopy) and obtaining good recoveries at the low concentrations added to the studied matrices (1 and 2 μg L^–1).

Two additional reports fulfilled the requirements of achieving an appropriate LOD (<2 μg L^–1), demonstrating a successful application to both spiked and real samples without externally added analyte and comparing the results with those provided by a reference method, but the methods were applied to too few samples (i.e., three or less).^24,32

Another group includes the development of 15 sensors with adequate LODs but which were only applied to spiked samples, with five of them tested on very few samples. Although real matrices are preferable to synthetic ones for recovery assays, the validation of a new method should be completed by demonstrating its usefulness in real samples with no external addition of the analyte. Depending on the analytes and/or the investigated matrices, it may not be easy to obtain real samples. Despite this drawback, the authors should make an effort to include them in their working protocols and statistically compare the results with those given by a reference method. In any case, if only spiked samples are included in the analysis, at least 10 different water samples with the addition of various analyte levels are required. This would increase the analytical quality of the newly proposed methods and show that the required standards have been fully achieved.

More worrying are the 17 papers where the sensor was employed to determine mercury in samples spiked with unreasonably high analyte concentrations, even when they display LODs which apparently allowed to measure mercury levels below the official maximum. In addition, four of these latter reports studied either one or two spiked samples.

Finally, Tables 1–4 show six methods with limited applicability for the determination of mercury ion in drinking water, since they present LODs close to or higher than the target concentration of the metal ion in this type of samples. In addition, these sensors were tested on spiked samples with high concentrations of Hg(II).

An issue not considered in Tables 1–4 but of analytical relevance is the evaluation of the method selectivity. Although the chemical sensors are described as having high selectivity toward the analyte under investigation, their behavior in the presence of potential interferents must be reliably demonstrated. The most common interferents for metal ion sensors are other cations that may respond similarly to the analyte by interacting with the sensing unit. Nevertheless, it is also important to analyze the effect of certain anions that may react with the cation producing stable species, seriously affecting the analyte detection. In the specific case of the Hg(II) ion, the presence of chloride should be taken into account. In this regard, all reports shown in Tables 1–4 studied the selectivity by analyzing the effect of potentially interfering cations on the sensor signal. However, not all of them analyzed the influence of the anions. In one study, chloride was removed dechlorinating the water sample by previous boiling.²⁸ In other cases, the fact that this anion does not interfere was indirectly demonstrated either by using chloride salts to prepare the solutions of the investigated cations or by employing HgCl₂ as mercuric reagent.

5. FINAL RECOMMENDATIONS

In this section we provide some general recommendations for the development of methods for the determination of inorganic species in drinking water, using Hg(II) as an example. Most scientific reports on the development of sensing platforms for Hg(II) allocate ca. 90% of the space to the detailed description of the synthesis, preparation, structural study, physicochemical characterization, etc., while the analytical sections are usually very brief. We suggest to expand the latter section following some standard rules for analytical calibration and validation.

A first phase of any protocol attempting to develop a methodology for detecting Hg(II) in drinking water is to roughly estimate the LOD. If this latter value is above 2 μg L^–1, this means that a preconcentration procedure is required.

Once the detectability of the method has been found to be feasible, the following steps should be followed:

• 1.Prepare and measure a set of calibration standards from 0.1 to 0.2 μg L^–1 to ca. 50–100 μg L^–1.
• 2.If the calibration graph is linear, use the IUPAC test for setting the maximum concentration at which linearity is complied. Do not rely on the correlation coefficient.
• 3.Estimate the LOD using the modern IUPAC definition.
• 4.Analyze at least 10 spiked water samples, with final concentrations of Hg in the range 0.5 to 5 μg L^–1. Compare the estimated value of the added concentrations with the nominal ones using a statistical test.
• 5.Analyze at least 10 real, unspiked drinking water samples and compare the results with a reference method, i.e., atomic spectrometry. Use suitable statistical tests for the comparison.

Conclusion

In view of the above discussion and findings, the conclusions of the present report are rather disappointing. The proposal of new sensors for measuring the concentrations of heavy metal ions in drinking water, particularly in the case of the highly toxic mercury ions, demands authors to comply with a series of requirements for proper method development, validation, and applicability study. These are not met in the vast majority of the reviewed publications, with only a very small number of reports employing the correct analytical standards.

The authors of the present review would like to call the attention of reviewers and editors of international journals on the importance of verifying that submitted manuscripts describing spectroscopic sensors for the determination of low concentration species, from heavy metal ions to all potential organic contaminants, meet modern analytical standards. Otherwise, the usefulness of the proposed sensors would be highly limited.

The authors declare no competing financial interest.