Direct colorimetric detection of water pollutants with non-water soluble dyes Open Access

The detection or sensing of airborne or waterborne pollutants with optical methods is a well-established branch of sensor technology. Sensing can be achieved by the impact of a target analyte on either the fluorescence or optical absorbance of a sensitive dye to which the analyte selectively associates. A classic example is the detection of explosive nitroaromatic vapours by their fluorescence quenching of conjugated polymers, as reviewed by Thomas et al. (2007). The quenching efficiency of nitroaromatic/conjugated segment associates is effectively amplified by the exciton diffusion along the conjugated polymer chain, leading to very low limits-of-detection, LoDs. However, the technical demands on fluorimetric transducers are higher than for the detection of optical absorption, and fluorescent dyes are prone to degradation as the excited state is vulnerable to oxidation, as shown by Alshammari et al. (2019). ‘Colorimetric’, i.e. optical absorption-based, sensors often allow for simpler, sometimes ‘naked eye’, transduction and a longer shelf-and operational life. The development of bespoke colorimetric dyes for waterborne targets therefore remains a dynamic area of contemporary research, e.g.Noh et al. (2019); Noh et al. (2020); Levakov et al. (2023); Beard et al. (2024); Hladun et al. (2024).

As many colorimetric dyes are conjugated organic molecules, these typically are soluble in tetrahydrofuran, THF, a relatively benign organic solvent, but often are insoluble in water, e.g.Santos et al. (2013); Noh et al. (2019); Noh et al. (2020). This poses the challenge to mediate a contact between waterborne analyte and insoluble sensitive dye. A number of strategies have been explored to meet this challenge, all with their relative merits and drawbacks. Sometimes, organic dyes can be formulated as water-soluble organic salts, typically sulfonates, e.g.Alshammari et al. (2020); Beard et al. (2024). However, the sulfonation of organic dyes requires aggressive chemical treatment (Kopacz 2003) and is not universally applicable to all dyes. Alternatively, some water-insoluble organic dyes may dissolve in a mixture of water with an auxiliary solvent, often dimethyl sulfoxide, DMSO, e.g.Noh et al. (2019); Noh et al. (2020). However, this requires mixing the aqueous sample under test with DMSO. Hence, the use of both sulfonated dyes and auxiliary solvent creates polluted water even when the test clears the sample of the actual target analyte. Another strategy is to process the sensitive dye into a plasticised PVC membrane. THF acts as solvent for many dyes and the membrane components (PVC, plasticiser, surfactant), as shown by Santos et al. (2013). Casting from a common solution and subsequent evaporation of THF embeds the dye into a soft but solid plastic membrane that is permeable for water. The membrane thus allows diffusion of waterborne analyte to the embedded dye. This approach is also widely used to mediate contact between analyte and potentiometric sensitisers in electrochemical sensors, as shown by several authors (Al Baroot & Grell 2019; Alghamdi et al. 2019; Alqahtani et al. 2020a, 2020b). However, membrane preparation requires several processing steps, auxiliary chemicals and solvents, and patience as the complete evaporation of THF from PVC is slow.

The most direct application of water-insoluble dyes in colorimetric sensors is to immerse a dye film into the aqueous analyte and work with the change of optical absorbance at the immediate interface between dye and aqueous medium. As analyte/dye contact is limited to a thin layer, this demands light intensity measurement with a high signal-to-noise (SNR) ratio. Tuwei et al. (2016) and Alshammari et al. 2019 reported a fibre optic instrument with a Lock-In driven detection scheme for that purpose, with neat dye coated onto the core of a ‘stripped’ optical fibre. This approach simplifies dye processing, spray coating from the THF solution onto the stripped fibre, but the cleaving, stripping and handling of the fibre within the sensor setup is demanding. Also, Tuwei et al. (2016) relied on evanescent wave absorption of light propagating in the fibre, which is weak and demands a very thin coating.

We here present an alternative highly sensitive absorbance meter using Lock-In detection. Our instrument allows the direct deployment of insoluble dyes for the detection of waterborne pollutants, without the need for auxiliary solvents, sulfonation, or phase transfer membranes. Also, we do not require optical fibres, but coat sensitive dyes directly onto LEDs with emission wavelengths matching the dyes' sensitive absorption bands. Our approach significantly simplifies sensor manufacture.

We demonstrate our instrument on two waterborne pollutants as target analytes, namely hydrogen sulfide, H₂S, and ethyl amine, Et-NH₂. H₂S acts as a weak acid, pK_a = 7 (ChemBook 2025), and Et-NH₂ acts as a weak base, pK_a = 10.8 (Hall 1957), and both are soluble in water, and are hazardous in aqueous solution as well as in the gas phase. Water pollution by H₂S, H₂S_aq, is harmful by itself, but also is the ‘telltale’ of faecal pollution in water, indicating the risk of the presence of pathogens associated with faeces. A detailed discussion is shown by Alqahtani & Grell (2024), who also reviewed sensor technologies for H₂S_aq detection, and gave a quasi- ‘potability’ or ‘limiting value’ of 1.5 μM H₂S_aq. For Et-NH₂ vapours, the US National Institute for Occupational Safety and Health, NIOSH, sets an exposure limit of 10 ppm for Et-NH₂ vapours (NIOSH 2025), the European Chemicals Agency, ECHA gives a long-term exposure limit of 5 ppm (ECHA 2025), but neither define a potability for Et-NH_2aq. However, the US National Oceanographic and Atmospheric Administration, NOAA declare Et-NH_2aq as ‘harmful to aquatic life in very low concentrations’ and define an ‘aquatic toxicity’ of 40 ppm (NOAA 2025), corresponding to a molarity of 2.22 mM, which we will here use as limiting value for Et-NH_2aq.

As water-insoluble colorimetric dyes, we use two organic dyes, first synthesised and characterised by the group of scholars led by Prof. J. H. Kim from Ajou University, Suwon, South Korea (Noh et al. 2019, 2020). We demonstrate sensing of target analytes with LoDs, far below the ‘limiting value’ of 1.5 μM for H₂S_aq, and Et-NH_2aq ‘aquatic toxicity’ of 2.22 mM.

Prof. J. H. Kim from Ajou University, Suwon, South Korea, supplied us with two organic dyes: The squaryllium dye ‘SQ1’ (Noh et al. 2020) and the isoindolene dye ‘ID1’ (Noh et al. 2019). Neither SQ1 nor ID1 are soluble in (pure) water, but both dissolve in 4:1 mixtures of water with DMSO. Under titration of the respective target analyte, SQ1 responds to H₂S at a narrow band peaked at 649 nm, and ID1 responds to Et-NH₂ at a somewhat broader band around 726 nm. Response is based on the selective association of analyte, H₂S or Et-NH₂, and dye. Detailed density functional theory, DFT, quantum chemical calculations of the dye/analyte association and consequential absorption bleaching for both dyes are reported by Noh et al. (2019, 2020). For sensor calibration, cf. Section 2.3, we acquired saturated 1.647 M H₂S_aq from GELKY Lab & Equipments (GELKY 2025), and 2 M Et-NH_2aq from Sigma Aldrich. To probe the dyes' responses to waterborne analytes, we selected stock LEDs with emission wavelength closely matched to the analyte-sensitive SQ1/ID1 absorption band. For SQ1, we sourced a deep red plant growth light with peak emission at ∼660 nm (Led660), and for ID1, we sourced a far red/near infrared LED with 730 nm peak emission (Led730). After LED light passes through the dye film, we detect light intensity with an OPT101 light detector board (Opt101). To achieve very high sensitivity for small changes in optical absorbance, we drive LEDs and detect OPT101 output with an Anfatec 250 USB Lock-In amplifier (Anfa), as described in Section 2.2, “Instrument design”.

LEDs project light across the dye film coated onto their case and the water held in the cuvette onto the OPT101 photodetector board placed below the bottom of the cuvette. The LED is driven, and light intensity detected, with the electronic instrumentation described below. The sensor is then calibrated by titration of defined aliquots of analyte stock solutions, and detection of the increasing intensity of light projected onto the OPT101 as a result of dye bleaching at the water/dye film interface, as described in Section 2.3.

To calibrate the H₂S optical sensor, we drive the SQ1-coated 660 nm LED with a modulation amplitude V_P = 2 V, DC bias V_DC = 9 V, and a serial R of 47Ω, in a cuvette filled with 5 mL of pure water, and record V_OUT over time. We first allow for the LED and drive circuit to ‘warm up’ to a steady-state V_OUT which is achieved after ∼25 min. We then titrate aliquots of increasing volume of H₂S stock solutions into the cuvette. We prepare a ‘stock 1’ with 5 μM concentration H₂S, and a ‘stock 2’ with 100 μM H₂S by dilution from saturated H₂S_aq. We titrate aliquots of (10/20/40/80/200/400) μL stock 1 into the cuvette with a micropipette, raising the volume in the cuvette from 5 to 5.75 mL. After every titration step, we calculate the resulting concentration in the cuvette, allow V_OUT to settle to a new (higher) value as a result of dye bleaching by increased analyte concentration, and tabulate as V_OUT(c). We then switch to stock 2 and titrate the same series of aliquot volumes as for stock 1, eventually reaching a total volume of 6.5 mL of 11.663 μM H₂S_aq in the cuvette. The stock concentrations and aliquot volumes are chosen to scan a concentration range that overlaps with H₂S potability 1.5 μM. Titration range also overlaps 1/k = 0.9 μM with the association constant k = 1.11 × 10⁶ M⁻¹ reported by Noh et al. (2020) for the association of SQ1 and H₂S when both are dissolved in 4 H₂O:1 DMSO, but with an emphasis on exploring lower concentrations. As a final aliquot we add 10 μL saturated, 1.647M, H₂S solution to raise c in the cuvette to a very high value and ‘saturate’ the sensor response.

To calibrate the Et-NH_2aq optical sensor, we drive the ID1-coated 730 nm LED with a modulation amplitude V_P = 1.5 V, DC bias V_DC = 9 V, and a serial R of 100Ω in a cuvette filled with 4 mL of pure water. We first allow for the LED and drive circuit to ‘warm up’ to a steady-state V_OUT which is achieved after 27 min. For calibration, we use a 2 mM Et-NH_2aq stock solution 1 that we dilute from a 40 mM Et-NH_2aq stock solution 2 prepared from 2-M Et-NH_2aq. With a micropipette, we titrate aliquots of (10/20/40/80/160/300/600/790) μL Et-NH_2aq stock 1 into the cuvette, raising the volume in the cuvette from 4 to 6 mL and the Et-NH_2aq concentration to 667 μM. The stock concentration and aliquot volumes are chosen to scan a concentration range that overlaps with 1/k = 173 μM with the association constant k = 5.8 × 10³ M⁻¹ reported by Noh et al. (2019) between ID1 and Et-NH₂ when both are dissolved in 4 H₂O:1 DMSO, but with an emphasis on exploring lower concentrations. Finally, we added aliquots of (10/20/30/40/50/60) μL of 40 mM Et-NH_2aq stock 2 to achieve response saturation, and to raise c close to the ‘aquatic toxicity’ level of 2.22 mM. After every titration step, we wait for V_OUT to stabilise to a new level, record it, and titrate the next aliquot.

Following the warm-up and titration procedure described in Section 2.3, we find V_OUT(t) for SQ1-coated 660-nm LED under titration with H₂S stock solutions as shown in Figure 2.

V_OUT(t) in pure water stabilises within ∼20 min at 669.5 mV following an approximately exponential approach. After each titration step of analyte aliquots, V_OUT(t) increases rapidly (<1 min) to a new plateau, which we read as V_OUT(c) at the respective analyte concentration, c. We calculate analyte concentration in the cuvette resulting from addition of titration aliquots and tabulate c, V_OUT(c), and resulting ΔA(c) as calculated using Equation (1) in Table 1.

Following the warm-up and titration procedure described in Section 2.3, we find V_OUT(t) for ID1-coated 730 nm LED under titration with 2 mM Et-NH_2aq stock solution that is visually very similar to the previous Figure 2 for SQ1 dye/H₂S titration. We therefore do not show it here, but only list the stabilised V_OUT(c) values after every titration step in Table 2, together with the calculated Et-NH_2aq concentration, and resulting ΔA(c) as calculated using Equation (1), in Table 2.

The optical response ΔA(c) of solid SQ1 to H₂S_aq shown in Table 1 is negative due to the bleaching of the SQ1 absorption band at 649 nm after association of SQ1 dye with H₂S. This agrees with the prior report by Noh et al. (2020). However, the ΔA(c) we find here is of rather smaller magnitude. We note dye and analyte come in contact only at a thin interface layer between solid dye, and waterborne analyte. Also, the 649 nm SQ1 absorption peak is narrower than the 660 nm LED emission band, so only a part of the LED emission overlaps with dye absorption. Nevertheless ΔA is clearly resolved for every titration step. For example, the increment of ΔA is only 1.1 × 10⁻³ between 146 and 327 nM, still the corresponding titration step is well resolved in Figure 1. Our instrument can detect ΔA < 10⁻³. High resolution of small ΔA is achieved thanks to the excellent SNR of the Lock-In detection method.

The optical response ΔA(c) plotted in Figure 3 shows a steep, near-linear increase for small H₂S_aq concentration, c ≪ c_1/2, but saturation to ΔA_MAX at c≫c_1/2, with ΔA_MAX and c_1/2 as defined in Equation (2). Saturation to ΔA_MAX = −0.0197 is reached at c ∼2 μM indicating that all available dye molecules at the film/water interface now have associated with H₂S, achieving maximum bleaching. The final addition of an excess, > 11 μM, of H₂S_aq confirms prior saturation as ΔA does not increase further.

While the characteristic pattern in Figure 3 qualitatively matches the prediction of Equation (2) at small and large c, it does not fully follow the curve predicted by Equation (2) for intermediate concentrations, c ≈ c_1/2. This is most evident from the corresponding Hildebrand–Benesi plot (not shown here) which is not well described as a straight line, while Noh et al. (2020) found a good Hildebrand–Benesi plot for H₂S/SQ1 association in solution. We conclude it as an oversimplification to describe association at the solid dye/liquid analyte solution interface by a single association constant k, as it is assumed for Equation (2). We speculate there are different k values for dye molecules immediately accessible at the surface, and dye molecules buried slightly deeper, leading to the two-step response characteristics seen in Figure 3. Such a distinction between ‘surface’ and ‘buried’ dyes is absent in solution. We can nevertheless quantitatively determine H₂S_aq from measured ΔA by using Figure 3 as a calibration chart. We read c_1/2 from Figure 3 via ΔA (c_1/2) = ½ΔA_MAX = −0.0197/2 = −9.85 × 10⁻³ as c_1/2 = 105 nM. This is smaller, i.e. favourable for low level detection, than c_1/2 = 1/k = 900 nM calculated with the association constant k = 1.11 × 10⁶ M⁻¹ reported by Noh et al. (2020). We note that Noh et al. (2020) report k for SQ1/H₂S association in mixed H₂O/DMSO solvent, which is not expected to be the same as k for association between H₂S in pure water to a solid SQ1 dye surface. As there is no simple analytic expression for the sensor calibration in Figure 3, we cannot give a statistical evaluation of the sensor's LoD. However, it is obvious that the sensor response even at the lowest tested analyte concentration, 10 nM, already is clearly above the noise floor, hence we can safely claim LoD < 10 nM. This conveniently places potability, 1.5 μM, at the upper end of our sensor's dynamic range, given by LoD < 10 nM, saturation at 2 μM, covering ∼3 orders of magnitude. Our LoD is more than an order of magnitude lower than the LoD of 211 nM for H₂S in water/DMSO mix with the same SQ1 dye reported by Noh et al. (2020), and similar to LoD = 14.9 nM reported by Alqahtani & Grell (2024) for a simple potentiometric sensor for H₂S in water.

Commercial instruments for monitoring H₂S_aq typically first drive H₂S out of the sample by acidification, and then quantify by gas chromatography, e.g. ECHalle (2025). Our detection in the aqueous sample directly has a smaller procedural and equipment footprint.

The optical response ΔA(c) of solid ID1 to Et-NH_2aq shown in Table 2 is again negative due to the bleaching of the ID1 absorption band at 726 nm after association of ID1 dye with Et-NH₂. ΔA(c) saturates at ∼1.3 mM with ΔA_MAX > 16 times larger than for SQ1/H₂S. We assign this to the broader absorption band, peak at 726 nm, of ID1 dye, overlapping better with the 730 nm LED emission band than for SQ1/660 nm LED.

The optical response ΔA(c) plotted in Figure 4 again shows a steep, near-linear increase for small c, c ≪ c_1/2, but saturation to ΔA_MAX at c≫c_1/2. Saturation to ΔA_MAX = −0.277 is reached at c ∼1.3 mM indicating that all available dye molecules at the film/water interface now have associated with Et-NH₂, achieving maximum bleaching of absorption.

As observed previously for for SQ1/H₂S response, cf. Section 4.1, we again find the response characteristic does not fully follow the curve predicted by Equation (2), but deviates for concentrations c ∼ c_1/2, presumably for the same reasons as speculated in Section 4.1. We can nevertheless quantitatively determine Et-NH_2aq from measured ΔA by using Figure 4 as a calibration chart.

From Figure 4, we read c_1/2 ≈ 700 μM. This is somewhat larger than c_1/2 = 1/k = 172 μM calculated with the association constant k = 5.8 × 10³ M⁻¹ for ID1/Et-NH₂ in H₂O/DMSO solution reported by Noh et al. (2020). As there is no simple analytic expression for the sensor calibration in Figure 4, we again cannot give a statistical evaluation of the sensor's LoD. However, it is obvious that response is clearly above the resolution of our instrument ΔA < 10⁻³ even for the lowest tested concentration, 5 μM, we therefore estimate LoD < 5 μM. This is somewhat below the LoD of 167 ppb = 9.3 μM reported by Noh et al. (2020) for Et-NH₂ with the same ID1 dye when dissolved in H₂O/DMSO solution, and ∼440 times lower than the aquatic toxicity of Et-NH₂. This gives a dynamic range for Et-NH₂ detection of LoD < 5 μM..., saturation ∼1.3 mM, spanning almost three orders of magnitude. With an Et-NH₂ aquatic toxicity of 2.22 mM, this allows assessing toxicity with respect to Et-NH₂ even when samples are somewhat diluted.

The colorimetric detection of waterborne analytes with selective dyes is a powerful and convenient sensor technique. However, often the selective dyes are themselves insoluble in the aqueous sample. Researchers therefore traditionally mediate contact between waterborne analyte and water-insoluble dye by adding an auxiliary solvent, or embedding the dye in a phase transfer membrane, or chemical sulfonation of the dye.

We here present an optical absorption instrument with high resolution that allows direct deployment of colorimetric dyes without the above complications. Our instrument is based on widely available electronic components, and can easily be replicated elsewhere. We employ LEDs with bright, narrow emission targeting the ‘sensitive’ band in the dye's absorption spectrum, coat LEDs with a film of the dye using an organic solvent, and immerse coated LEDs into the aqueous sample. A bespoke Lock-In based drive-and-detection circuit lights the LED, and detects the intensity transmitted through the dye with low noise even in the presence of ambient light. The instrument is capable of detecting changes in absorbance <10⁻³. This is sufficient to ‘see’ the bleaching of a squaryllium dye, SQ1 from Noh et al. (2020), at the dye film's surface when films are in contact with hydrogen sulfide solutions (H₂S_aq) with a limit-of-detection <10 nM. This is at least 150 times lower than the potability of H₂S_aq, 1.5 μM. Similarly, we detect bleaching of an isoindolene dye, ID1 from Noh et al. (2019), in contact with ethyl amine solutions, Et-NH_2aq, with an LoD of <5 μM. This is ∼440 times lower than Et-NH_2aq ‘aquatic toxicity’ of 2.22 mM. The large ‘dynamic reserve’ of >2 orders of magnitude between LoDs and analyte limiting values enables practical application to environmental samples. A typical procedure would be to operate the LED in a sampling cuvette pre-filled with ∼3 mL of clean water, then adding ∼3 mL of environmental sample. As this dilutes the sample two-fold, we double the detected concentration to find the sample concentration. Even after the dilution of analyte concentration implied by this procedure, samples polluted above the respective potability will lead to a clearly detectable sensor response.

We have shown we can readily adapt our instrument from one target analyte to another simply by applying a different dye to a different LED. As narrow-band LEDs are available across the visible spectrum, the limit for general use of our instrument and method is solely in the availability of suitable dyes. We have not explored repeated use of the same coated LED after responding to analyte once. As dye/analyte associates are strong, they do not easily dissociate to prepare the sensor for repeated use. This is a common problem with optical sensors. However, we believe recovering the dye for repeated use may be somewhat easier for a neat film than for dye in a membrane or solution. We propose soxhletting or ‘steaming’ used coated LEDs, i.e. repeatedly rinsing in freshly distilled hot water, to wash out analyte from dye/analyte associates.

We are grateful to Prof. Jong H. Kim from the Department of Molecular Science and Technology, Ajou University, Suwon, Republic of Korea for the provision of SQ1 and ID1 dyes, and to Anfatec Instruments AG of Oelsnitz, Germany, for technical support with the Anfatec 250 Lock-In.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.