Corrections for matrix effects on fluorescence measurement of a multi-platform optical sensor

The application of optical techniques to in situ characterization of constituents in natural waters through the use of fluorescence, absorbance, or scattering has proven useful to the understanding of water chemistry. However, in situ fluorescence sensors suffer from the same instrumental limitations as their laboratory counterparts, in addition to some other challenges, all of which can affect the accuracy and precision of measurement. The principal, potential interference factors are inner filtering, temperature, intramolecular deactivation (quenching), turbidity, and pH. Each of these is addressed here with empirical calibration or by applying equations governed by the laws of chemistry. Finally, by addressing sensor-specific effects, the results obtained from different instruments can be compared more readily.

The fluorescence intensity of a water sample is only linearly proportional to the concentration of a fluorophore over a limited range. Water that has high absorbance, due either to the fluorophore itself or the absorbance of other compounds in the sample, can absorb the excitation light leading to primary inner filtering. Secondary inner filtering occurs when a fraction of the photons emitted by the target analyte is absorbed in the water. Inner filtering becomes increasingly important as sample absorbances increase. The effect of inner filtering also depends on the excitation-emission geometrical arrangement, including the path lengths within an instrument. In general, if absorbance measured over the relevant path length at the relevant wavelengths exceeds 0.1, correction of the fluorescence signal may be needed.

The effect of inner filtering on fluorescence measurement is discussed by several workers, including Kubista et al. (1994), Puchalski et al. (1991), Guibault (1990), Yappert & Ingle (1989), and Wiechelman (1986). The effect does not just produce errors in total concentration determination; for example, Ohno (2002) found that inner filtering affects the humification index of colored dissolved organic matter (CDOM) proportionally with increasing concentration. In waters containing high concentrations of CDOM (tens of mg-C/l) – e.g., some peatland drainage – this can be particularly problematic and can even result in fluorescence decreasing with increasing CDOM concentration. It is sometimes possible to deal with this issue by dilution in laboratory analysis but that is not practical with in situ instruments. An attenuation model based on the absorption of excitation and emission optical densities may be needed for mathematical correction of the real-time fluorescence measurement of in situ instruments.

In situ instruments are susceptible to both short-term diurnal and long-term seasonal temperature changes, and correction may be needed. A typical example would be the variation of water chemistry across a thermocline in a stratified lake. The fluorescence signal is generally reduced by higher temperatures, although the temperature and fluorescence relationship varies among analytes. Watras et al. (2011) investigated the effect of temperature on CDOM fluorescence measurement, and concluded that a temperature correction is necessary for field measurements. They propose a linear temperature compensation equation. Ng et al. (2014) demonstrated a linear temperature compensation equation for chlorophyll a, but found that an exponential equation gave better correction in the case of rhodamine b, showing that the functional relationship of temperature compensation and fluorescence measurement is analyte specific.

Natural waters are complex chemical solutions including many constituents, and the chemical background can influence the fluorescence signal. Quenching of fluorescence measurement is described by Lakowicz (2006) and literature cited therein. Various mechanisms can reduce fluorescence intensity including excited state reactions, molecular rearrangements, energy transfer, ground-state complex formation, and collisional quenching. The primary compensation for chemical quenching must deal with both dynamic and static quenching. Dynamic quenching results from collisions between the fluorophore and the quenching species in the absence of photochemical reaction – the quencher must diffuse to the fluorophore during the lifetime of the excited state. Excited state is any quantum state that has higher energy than the ground state. Static quenching results from binding of the fluorescent and quenching species to produce a non-fluorescent complex. The Stern-Volmer equation is linear when the quenching is either static or dynamic. When both occur, the equation is polynomial. The association of complex and/or Stern-Volmer coefficient of chemical quenching can be determined from the linear or polynomial Stern-Volmer equation, to correct for the effect of major quencher(s) on a fluorescence signal. The effect of quenching of humic material on the fluorescence of chlorophyll a was noted by Ng et al. (2014).

Natural waters also contain suspended particles, which cause turbidity and can interfere with fluorescence measurement by causing light reflection and/or absorption. Müller et al. (2001) found that scattering in turbid media can also alter spectral features, while Teale (2008) found that light scattering in turbid solutions caused depolarization of fluorescence. By recording turbidity simultaneously with fluorescence, the accuracy of measurement can be improved, typically using empirical calibrations.

The pH of water bodies is governed by many factors. The daily interplay of photosynthesis and respiration can result in cyclical changes of pH, especially in high-density aquaculture ponds and recirculation aquaculture systems. PH effects can be exploited constructively, for example, in fluorometric measurements of solvent extracts from disrupted algae cells, when acidification is used to estimate the relative amounts of chlorophyll and pheophytin [the excitation wavelength is varied according to the chlorophyll and pheophytin absorbance maxima (Saijo & Nishizawa 1969), and the ratio of the pre- and post- acidification fluorescence signal – the acidic ratio – is calculated (Arar & Collins 1992)]. The fluorescence of molecules containing acidic or basic functional groups is often pH dependent, as pH changes in the medium can alter the degree of ionization of the relevant functional group, thus affecting the extent of conjugation of the molecule and altering fluorescence emission. Effects can include changes in both intensity and the spectral properties of fluorescence. Yi et al. (2011), for instance, studied the effect of pH on the characteristics of oily waste water. For this material, they observed a spectral peak shift and an increase in fluorescence signal, when the medium was acidic. In a basic medium, the fluorescence signal decreased.

Finally, fluorescence spectra observed with different fluorometers are influenced uniquely by instrument-specific responses. DeRose et al. (2007) qualified (– i.e., characterized) a fluorescence spectrometer for measuring the fluorescence spectra of reference materials and generated excitation and emission correction factors. Cory et al. (2010) evaluated the removal of instrument-specific responses from dissolved organic matter fluorescence spectra collected using three different fluorometers, on the basis of manufacturer-provided emission and excitation correction files. They concluded that fluorescence spectra collected on different fluorometers differed significantly before correction, but in the cases examined, could be made sufficiently independent of the instruments concerned.

The work described here was carried out on a multi-platform optical sensor – the LEDIF (LED-induced fluorescence) – for the in situ determination of water chemistry, as described by Ng et al. (2012a, 2012b). The LEDIF uses fluorescence, absorbance, and scattering to quantify substances in water. In this paper, the use of auxiliary sensors (e.g. temperature, pH) is examined along with the LEDIF's ability to measure fluorescence, absorbance, and turbidity simultaneously, to enable development of corrections to the matrix effects. Laboratory-prepared and commercial standards with chemical characteristics resembling substances commonly found in natural water were used to demonstrate the corrections. As a test of ability to compensate for multiple, simultaneous matrix effects, the determination results from a liquid sample, subjected to known interferences and physical conditions, were corrected and compared with a reference sample. The spectrometer and associated optical systems in the LEDIF were qualified using the atomic lines of mercury (Hg) and argon (Ar), and National Institute of Standards and Technology (NIST)-traceable calibration light sources.

Light from the flow cell is observed with an Ocean Optics USB4000 spectrometer, and the data are recorded with a single-board computer manufactured by Technologic Systems (Model TS-7260-64-128F) running custom software (iLEDLIF). For land-based use in continuous monitoring, a Gotec (Model: EMX-08) piston pump feeds samples into the LEDIF flow cell manifold. When used in the autonomous underwater vehicle, ram pressure drives samples through the flow cell.

LEDIF spectrometer wavelengths were calibrated at multiple positions between 253 and 922 nm – see Table 1. The spectrometer was connected to the calibration source with a 50 μm diameter ultraviolet (UV)-visible (VIS) optical fiber. Wavelength calibration coefficients were obtained via a third-order linear regression, using true wavelength (atomic lines) as a function of pixel number, pixel number squared, and pixel number cubed (recorded with the spectrometer). The wavelength calibration coefficients are stored in the spectrometer's electrically erasable programmable read-only memory (EEPROM). The operating wavelength of the spectrometer is calibrated (Ocean Optics 2008) every 12 months.

Finally, the central wavelength and full width at half maximum (FWHM) bandwidth of each LED in the excitation system were measured using the LEDIF's NIST-calibrated spectrometer. The LED excitation system was connected to the spectrometer with a 1,000 μm diameter UV-VIS optical fiber.

All experiments were performed using laboratory-prepared and commercial standards in deionized (DI) water in amber glass bottles. Chlorophyll a was obtained from Tokyo Chemical Industry (CAS 1406-65-1), humic acid from Acros Organics (Sodium Salt, tech., CAS 68131-04-4, Lot A0268981), and turbidity standard solutions from AMCO CLEAR^®TURBIDITY STANDARD, GFS Chemicals, Inc. (NIST-traceable). For chlorophyll a, the concentrations reported in this paper are based on the manufacturer's reported assay of 0.5% plant-derived chlorophyll. For humic acid, the concentration reported is based on 55% carbon content; the manufacturer's assay reports carbon content ranging from 50 to 60%.

For all experiments, the baseline and background signals were corrected with the DI water spectrum used to prepare the laboratory standards. All stock solutions (except the turbidity standard) were filtered with Grade 42 filter paper (GE Healthcare Life Sciences), having a nominal pore size of 2.5 μm, to minimize possible effects of large particle interference (if any) on the fluorescence signal. Fresh solutions were prepared, typically within 2 to 3 hours of analysis, and all spectra were collected at room temperature (22 to 25 °C), except during temperature compensation experiments. The integration times for all fluorescence measurements were 10 seconds, except for those for turbidity and multiple matrix effects compensation to chlorophyll a.

For tests of temperature correction, the LEDIF inlet and outlet were dipped into a reservoir containing an analyte of known concentration. A 12 VDC Gotec (Model: EMX-08) piston pump was used to circulate the analyte continuously between the LEDIF and the reservoir, while a sensor recorded the temperature of the analyte in the reservoir. The liquid sample was first refrigerated for experiments below the reference temperature of 25 °C, and periodic measurements were made as the liquid rose to room temperature. For experiments at temperatures above reference, the reservoir was put on a temperature-controlled hot plate.

Humic acid (sample analyte) and chlorophyll a (quencher) were chosen for these experiments because both are important analytes relevant to natural water bodies. The observed fluorescence intensity was corrected for inner filtering effects before quenching assessment.

Fluorescence measurement is generally considered less susceptible to particle interference than absorbance measurement. The effect of turbidity on sensor response was assessed by mixing chlorophyll a standard with styrene divinyl benzene copolymer beads in water at concentrations up to 500 NTU. Monodisperse particles were thus responsible for the turbidity observed. Monodisperse particles have a one-to-one relationship between scattering intensity and scatter angle (Xu 2002). Natural waters contain polydisperse particles of different sizes and compositions (Fondriest.com 2015).

Laboratory-prepared chlorophyll a and humic acid samples of known concentration and pH were measured using the LEDIF. Sample pH was altered by adding potassium hydroxide (1,360, 3,400 and 6,800 mg/l) or hydrochloric acid (0.074, 0.44, and 40.4 mg/l) in a laboratory glass beaker in a dark room. The modified samples were mixed with a magnetic stirrer, and fluorescence was measured after the pH reached a constant value. These experiments were carried out on fresh samples derived from the same source. Sample dilution with acid or base additions never exceeded 0.3% by volume. Acid addition converts chlorophyll a to pheophytin a, and is used, typically, in the extractive measurement of disrupted algae cells. The observed fluorescence intensity of humic acid was corrected for inner filtering before pH assessment.

To assess whether humic material sorption to the flow cell or sample container walls occurs when the pH is lowered, potassium hydroxide was added to DI water to make it basic. This basic water was used to prepare the humic samples and the fluorescence signal measured. Hydrochloric acid was then added to lower the pH to neutral and the fluorescence signal was measured again.

Equal volumes of 1.5 mg/l chlorophyll a (analyte), 30 mg/l humic acid (quencher), and 600 NTU turbidity standard were mixed to form a solution containing 0.5 mg/l chlorophyll a, 10 mg/l humic acid, and 200 NTU. Hydrochloric acid was added to lower the pH to 4.3 and the solution was cooled to 10 °C. The LEDIF was used for fluorescence measurement, and matrix effects were compensated by applying the chemical law and empirical calibrations presented above. The results were compared with the fluorescence signal of 0.5 mg/l of chlorophyll a in DI water.

Figure 2(b) and 2(c) show the correction factor and the measured intensity of the NIST-traceable light source recorded by the spectrometer in the LEDIF. Figure 2(c) shows that the spectral irradiance of the NIST-calibrated spectrometer in the LEDIF matches very well with the NIST-traceable light source after the correction. Figure 2(d) shows the measured center wavelength and FWHM bandwidth of the six excitation sources in the LEDIF. The values reported by the LED manufacturer (Thorlabs) are shown in the legend of Figure 2(d). In Table 2, the LED manufacturer's reported values are compared with those measured by the LEDIF after calibration. The FWHM bandwidths of 375 and 405 nm as measured were somewhat larger than those reported by the manufacturer.

In vivo chlorophyll a internal to algae in water bodies in Singapore rarely exceeds 0.2 mg/l, and typically falls within the linear region for fluorescence (Ng et al. 2014, 2015). Because of this, inner filtering correction was not considered for measurement of in vivo chlorophyll a in this study. Some of the same compensation procedures could be applied should inner filtering effects become important.

Figure 5(b) shows the observed peak intensities and inner-filtering-corrected fluorescence measurements of humic acid quenched by chlorophyll a, for 500 and 520 nm emission wavelengths. Figure 5(c) shows the inner filtering and quenching-corrected fluorescence measurements of humic acid quenched by chlorophyll a, illustrating that the corrected value is a constant independent of chlorophyll a concentration. In other words, quenching compensation can improve the accuracy of humic fluorescence measurement in the presence of chlorophyll a. It is noted, however, that these experiments were carried out using extracted chlorophyll a as a model for a generic quencher, and do not necessarily represent possible effects of in vivo chlorophyll a internal to algae in natural waters.

Figure 7(b) shows the fluorescence spectra of humic acid at different values of pH, when excited at 285 and 375 nm. By contrast to chlorophyll a, fluorescence intensity decreases with decreasing pH; no noticeable fluorescence peak wavelength shift was observed with changes in pH.

The fluorescence signal of humic samples prepared at pH = 9.8 and excited at 375 and 405 nm was measured before and after neutralization (pH = 7). The maximum percentage difference in all signals measured was below 2.7, and readily accounted for by small dilution and/or experimental errors. Mass loss by sorption to vessel walls (if any) was negligible and the humic materials appeared to be entirely dissolved in samples used for the pH adjustment experiments.

The experiments described illustrate the variety of conditions under which matrix effects to fluorescence measurement of natural waters occur, and suggest that, in general, compensation for them is both necessary and possible in many in situ applications. Corrections involving both inner filtering and turbidity are simplified by the LEDIF's ability to measure absorbance and turbidity almost simultaneously with fluorescence in situ. Sensing of pH and temperature requires that an in situ optical instrument be accompanied by auxiliary pH and temperature sensors. Although the work discussed here was carried out with the LEDIF, the compensation techniques described should be equally adaptable to other in situ optical sensors.

Other potentially common matrix effects were not investigated. They include, for example:

• o quenching by dissolved oxygen, so that the inclusion of a dissolved oxygen sensor could be helpful.

• o turbidity effects resulting from inhomogeneous mixtures of particles, which could also prove more complex than those investigated – in some cases in-line filtration to remove high concentrations of larger particles could prove helpful.

Corrections for ionic strength were not investigated, either, but variations of this type could be important, for example, in estuarine studies. Fortunately, it is possible to estimate this parameter using standard auxiliary sensors.

Funding for this work was provided by the National Research Foundation Singapore under its Campus for Research Excellence and Technological Enterprise programme through the Singapore MIT Alliance for Research and Technology's Center for Environmental Sensing and Modeling interdisciplinary research group. We thank S. Senft-Grupp for advice and helpful discussions, and Andre Teo, Fitri Jianna and Chow Shan Bin for assistance with data collection.