- Email: [email protected]
Correction of inner-filter effect in fluorescence excitation-emission matrix spectrometry using Raman scatter
Analytica Chimica Acta 583 (2007) 357–363
Correction of inner-filter effect in fluorescence excitation-emission matrix spectrometry using Raman scatter Tobias Larsson ∗ , Margareta Wedborg, David Turner Department of Chemistry, G¨oteborg University, Kemiv¨agen 10, SE-41296 G¨oteborg, Sweden Received 19 May 2006; received in revised form 5 September 2006; accepted 18 September 2006 Available online 14 October 2006
Abstract Fluorescence excitation-emission matrix (EEM) spectroscopy is a useful tool for interpretation of fluorescence information from natural water samples. One of the major problems with this technique is the inner-filter effect (IFE), i.e. absorption of light at both the excitation and emission wavelengths. The common solutions are to either dilute the sample or apply some form of mathematical correction, most often based on the measured absorbance of the sample. Since dilution is not always possible, e.g. in on-line or in situ EEM recordings, and corrections based on absorbance are hampered primarily by the use of a separate absorbance instrument, neither of these solutions is optimal. In this work, we propose a mathematical correction procedure based on the intensity of Raman scatter from water. This procedure was found to reduce the error after correction by up to 50% in comparison with two absorbance correction procedures. Furthermore, it does not require the use of a separate absorbance measurement, and it is applicable to on-line and in situ EEM recordings, where the IFE would otherwise cause problems. © 2006 Elsevier B.V. All rights reserved. Keywords: Fluorescence; Excitation-emission matrix; Inner-filter effect; Absorbance correction
1. Introduction The fluorescence of humic substances, or gelbstoff, in seawater has been recognised for a long time [1,2]. The low limit of detection has made it easy to measure and useful as a proxy for organic material. However, there is more information available from the fluorescence of humic substances, or, more generally, fluorescent organic matter (FOM), than can be acquired from a single or a few excitation/emission wavelength combinations. A wide variety of wavelength combinations has been used to measure humic fluorescence [3–7], which makes overview difficult. For the last 15–20 years, it has become increasingly common to measure excitation-emission fluorescence matrices (EEMs) [8–11], sometimes also described as total luminescence spectra. This technique involves recording a series of fluorescence emission spectra at different excitation wavelengths and concatenating them into a matrix. With this technique, more information about the FOM is collected, and it also becomes more straightforward to compare results
∗
Corresponding author. Tel.: +46 31 7722284; fax: +46 31 7722785. E-mail address: [email protected] (T. Larsson).
0003-2670/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.aca.2006.09.067
from different researchers. The technique also enables characterisation of FOM from different locations, which opens the possibility to use FOM as a water tracer. With modern-day technology, including diode-array and CCD detectors, it may also become possible to do real-time, on-line or in situ EEM recordings. One of the major problems associated with measurement of FOM in natural samples is the inner-filter effect (IFE), sometimes referred to as self-absorption. This is caused by the absorption of the exciting as well as the fluorescent light (primary and secondary IFE, respectively) by the fluorophore itself, alternatively by another component of the sample. The effect is important already at low absorbances; an absorption coefficient (a) of 11.5 m−1 (corresponding to 0.05 absorbance in a 1 cm cuvette [12]) in the sample reduces the fluorescence by 5% [13]. The mechanism of IFE is different from quenching, although the effect is similar [14]. Quenching is the process in which the fluorescence is reduced because the excited molecules lose their energy via other pathways than emitting light, such as interaction with other molecules. In general there are two ways of reducing the IFE: dilution or mathematical correction of the intensities. The simplest procedure for laboratory measurements is to dilute the sample to
358
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
an acceptable absorbance level (i.e. below a = 11.5 m−1 ) [15]. However, this may lead to problems, such as contamination (from the diluting water) and, perhaps most important, perturbation of the chemistry of the sample (affecting e.g. colloidal stability). Another way is to apply some form of mathematical correction to the fluorescence data. A number of such corrections has been suggested [16–20], but most, if not all, involve the need for a separate absorbance measurement. This usually requires the employment of a separate absorbance instrument, with instrument characteristics different from the fluorescence instrument used for recording the EEM. Consequently, another error is introduced into the analysis. The rather short linear range of the absorbance measurement procedure is another limiting factor. Rayleigh and Raman scatter occur when a molecule has been excited to a virtual energy state by a photon with insufficient energy to completely excite the molecule. The virtual state is unstable, and the molecule quickly relaxes, emitting light. Depending on the original and final vibrational energy level of the molecule, the emitted light may have higher, lower or the same energy as the exciting light. In Rayleigh scatter, the emitted light is of the same energy as the exciting light (Fig. 1). This process, known as elastic scattering, is the most probable one, and, consequently, it has the highest intensity. Rayleigh scatter in the EEM is an exactly diagonal structure occurring at λem = λexc . Raman scatter is an inelastic process, in which the molecule relaxes to a different vibrational energy level of the ground state than the original one. When the molecule relaxes to a higher vibrational level, the emitted light has a lower frequency than the exciting light, and vice versa for the case where the molecule relaxes to a lower vibrational level. The difference in energy is called the Stokes’ shift, a term borrowed from fluorescence spectroscopy, although the underlying phenomena are different. Since the vibrational energy levels depend on the molecule, the Stokes’ shift is independent of the wavelength (frequency) of the exciting light, i.e. the Stokes’ shift of a certain molecule will be constant regardless of the frequency of the exciting light. For example, the Stokes’ shift of water is ∼3400 cm−1 ,
Fig. 1. Jablonski diagram showing different excitation and emission processes. (a) Fluorescence; (b) Rayleigh scatter; (c) Stokes’ shifted Raman scatter; (d) anti-Stokes’ shifted Raman scatter.
regardless of the excitation frequency. The wavelength is inversely proportional to the frequency, which causes the Raman scatter to appear as a slightly curved structure in the EEM (Fig. 2). Both Rayleigh and Raman scatter are dependent on the number of scatterers, i.e. the concentration of the scattering molecule. Raman scattering is a fairly weak phenomenon, requiring either high concentrations of the scatterer or a high-energy light source (such as a laser). As a result, the only molecule of sufficient concentration to produce visible Raman scattering in fluorescence measurements of aqueous samples is water. Consequently, differences in the Raman scatter intensity in EEMs of natural water samples will only depend on the IFE. Raman scattering therefore provides the more appropriate basis for a correction method since it is much less dependent on sample composition. 2. Experimental An EEM consists of a number of emission spectra, each of which contains a Raman scatter peak (Fig. 2a and b). The area
Fig. 2. (a) Typical fluorescence emission spectrum with the different peaks marked. The upper abscissa scale is wavelength in nm and the lower scale is frequency in cm−1 . (b) Contour plot of an EEM with the different structures marked.
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
359
of the Raman scatter peak of each emission spectrum was integrated in Matlab (Version 6.5, The MathWorks, Inc.) using an in-house script, which approximates the area as a series of trapezoids. The areas were stored as a vector, which was subsequently smoothed using a mean filter with window size 3. The smoothing is necessary to reduce integration related errors, which would otherwise impede further data processing. Each value in the smoothed vector thus corresponds to a specific emission spectrum of the sample (or blank sample). For correction of an EEM of a sample for the inner-filter effect, the emission spectra of the EEM were multiplied by the quotient between the values from the vector of the blank (ultrapure water, Milli-Q Plus 185, Millipore Inc.) and the values from the vector of the sample. In mathematical form, the corrected emission spectrum scorr can be expressed as (bold type denotes a mathematical vector): scorr (exc ) = s0 (exc )
rMQ (λexc ) r0 (λexc )
Here, s0 (λexc ) is the uncorrected emission spectrum obtained at excitation wavelength λexc , rMQ (λexc ) is the value for the blank at λexc , and r0 (λexc ) is the value for the uncorrected emission spectrum. The RC procedure was compared to two other procedures (CL and CG) suggested by, respectively, Lakowicz [13] (CL) and Gauthier et al. [16] (CG). In CL, the EEMs are corrected for absorbance by multiplication of each value in the EEM with a correction factor based on the idea that the average path length of the absorption of the excitation and emission light is 1/2 of the cuvette length. Mathematically: Fcorr = Fobs × 10(Aexc +Aem /2) where Fcorr and Fobs are the corrected and uncorrected fluorescence intensities and Aexc and Aem are the absorbance values at the current excitation and emission wavelengths. In CG, an attempt is made to account for the geometry of the cell, i.e. the actual widths of the excitation and emission light paths: Fcorr 2.3dAex 2.3sAem = × 10gAem × −dA ex Fobs 1 − 10 1 − 10−sAem Fcorr , Fobs , Aex , and Aem are as defined above and d, g, and s are defined in Fig. 3. Measurements were made with a slide caliper. The comparison was carried out in two parts: the dichromate and the humic lake (HL) experiment. In the dichromate experiment, the effect of potassium dichromate on the fluorescence of quinine sulphate dihydrate (QS) was investigated: two different QS concentrations (15 and 38 g/L in 0.05 M sulphuric acid) were used for fluorescence. To each of the QS solutions, four different concentrations of potassium dichromate (1, 5, 10, and 25 mg/L) were added to produce IFE. For the HL experiment, the samples were collected from three different humic lakes in western Sweden: Lake Lille Vektor (LV), Lake Kringlevattnet (KV) and Lake Trollvattnet (TV) in winter conditions.
Fig. 3. Cuvette dimensions for CG procedure (adapted from Gauthier et al. [16]). s represents the width of the excitation light beam, g is the distance from the edge of the excitation light beam to the edge of the cuvette, and d is the width of the cuvette seen by the detector. In this instrument setting, s was estimated to 0.9 cm, g to 0.1 cm, and d to 0.6 cm.
2.1. Chemicals All reagents were of analytical grade. QS was obtained from Carl ROTH, potassium dichromate was obtained from Scharlau and sulphuric acid was obtained from Merck. All solutions and dilutions were prepared with ultrapure water, which was also used for blank samples. From each of the HL samples, four subsamples were prepared by dilution with ultrapure water. 2.2. Data recording The EEMs for the dichromate experiment were recorded by scanning emission wavelengths from 230 to 600 nm with 1 nm increments. Excitation wavelengths ranged from 230 to 450 with 5 nm intervals. For the HL experiment, emission was scanned from 230 to 600 nm with 1 nm increments. Excitation wavelengths ranged from 230 to 420 nm with 10 nm intervals. All fluorescence measurements were performed on a Fluoromax-2 spectrofluorometer (Jobin-Yvon Inc.). The bandwidth used for both excitation and emission was 5 nm. The absorbance spectra were recorded from 230 to 600 nm with 1 nm increments on a Hewlett–Packard 8453 UV–vis diodearray absorbance spectrophotometer with 1 nm bandwidth and a path length of 1 cm. Absorbance spectra are reported as absorption coefficients for the above range as recommended by Hu et al. [12]. 2.3. Data processing The raw data from both instruments was imported into Matlab (Version 6.5, The Mathworks, Inc.), where the different corrections were subsequently applied.
360
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
As a measure of the accuracy for each of the three procedures, the deviation of the corrected from the true EEM was used. This deviation was defined as: true corr − Ii,j Ii,j i
j
In the literature, absorbances of both 0.02 [21] and 0.05 [13] have been judged sufficiently low to avoid IFE. This corresponds to an absorption coefficient that does not exceed 4.6 m−1 or 11.5 m−1 , respectively, within the wavelength range used. In this work the lower value was used. For the dichromate experiment, the QS solutions with no added dichromate were used as the true values, since the QS did not contribute to the absorbance. For the HL experiment, the lower criterium above was met with subsamples diluted to 1/64 of the original concentrations. The Raman and Rayleigh scatter were excluded from the comparison. 3. Results and discussion The absorbance spectra from the two experiments are shown in Fig. 4. The impact of the absorbance, i.e. the inner-filter effect, is clearly visible in Fig. 5 (top row) and Fig. 6 (top row), which show the true and distorted EEM for 15 g L−1 QS (Fig. 5, top row) and for Lake Kringlevattnet (Fig. 6, top row). In both cases, the inner-filter effect has reduced the fluorescence intensities by up to 50% of the total fluorescence. An important practical consideration concerning the absorbance correction procedures is that the bandwidths of the two instruments were not the same. This is one of the more serious objections to the absorbance correction procedures. The spectrofluorometer had adjustable bandwidth settings for the excitation as well as the emission monochromator. These settings must be optimised for each sample set, and it is not uncommon that it is preferable to have different bandwidths for excitation and emission. In this work, the optimal bandwidth setting was found to be 5 nm for both excitation and emission. The spectrophotometer used in this work, on the other hand, had a fixed bandwidth of 1 nm. The bandwidth affects both peak height and peak width, and consequently the use of two different instruments will likely be a source for errors, although the magnitude is as yet unknown.
Fig. 4. (a) Absorbance spectra for the dichromate experiment; (b) absorbance spectra for the humic lake experiment. Absorption coefficient as defined by Hu et al. [12].
The results from the correction procedures are shown in Fig. 5 and Table 1 (the dichromate experiment) and in Fig. 6 and Table 2 (the HL experiment). In general, the deviation of the correction increases as the IFE increases, regardless of correction procedure (Tables 1 and 2). Figs. 5 and 6 show some of the correction results; the residuals are of particular interest. Note that the fluorescence scales of the residuals from the corrections are not the same as shown for the original EEMs. The effect of the dual absorbance peaks of dichromate is clearly visible in the residuals of the dichromate experiment (Fig. 5, right column). For both the CG and the CL procedures (Fig 5 CG and CL procedure residuals), there is an obvious correspondence between the two residual peaks and the dichromate absorbance (Fig 4a). The residual for the RC procedure also has some correspondence to the dichromate absorbance, but it is not as pronounced as for the other procedures. A similar pattern can be observed in the results from the HL experiment. The residuals for the CG and
Table 1 Results of the dichromate experiment QS (g L−1 )
15
38
K2 Cr2 O7 (mg L−1 )
Total intensity of true spectrum (108 cps)
0 1 5 10 25
31.6
0 1 5 10 25
81.5
Deviation (108 cps) Uncorrected EEM
CG
CL
RC
3.1 8.4 9.4 14.6
2.9 7.3 7.3 10.5
2.6 6.6 5.9 7.3
1.5 4.9 3.3 4.4
9.5 22.9 25.5 38.5
8.9 20.3 20.4 28.3
8.2 18.5 16.8 20.4
5.5 12.9 12.9 17.2
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
361
Fig. 5. Sample EEM results from the dichromate experiment. Note that the residuals of the three correction procedures have a different fluorescence scale. Rayleigh and Raman scatter peaks have been removed manually by setting their value to Not-a-Number (NaN) [23].
CL procedures are highest at the shorter wavelengths, where the IFE also is the largest. The same is equally true for the RC procedure, but, again, not as pronounced. The CG procedure is associated with the largest residuals for both experiments (Figs. 5 and 6). This is somewhat contradictory to the work by Puchalski et al. [22], who also compared the CG and CL procedures. However, Puchalski et al. used only one excitation spectrum and one emission spectrum for each
experiment, as opposed to using a major part of the EEM, as was the case in the present study. This might possibly account for the discrepancy. Furthermore, the work of Puchalski et al. was done using protocols from studies of fluorescence quenching; it is possible that the different approaches may also cause discrepancies. The residuals from the CL procedure are generally on the same level as the residuals from the CG procedure or slightly
362
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
Fig. 6. Sample EEM results from the humic lake experiment. Note that the residuals of the three correction procedures have a different fluorescence scale. Rayleigh and Raman scatter peaks have been removed manually by setting their value to Not-a-Number (NaN) [23].
smaller (Figs. 5 and 6). It is somewhat surprising that the simple correction equation of the CL procedure produces a better correction than the more elaborate equation, derived from Parker [21] and Lloyd [15], of the CG procedure. However, it is quite difficult to estimate the different measures of the light paths (Fig. 3), and it was further noted that the light beam of the instrument used in this work was not quite parallel. Consequently, the CG procedure is hampered by errors in the estimation and cal-
culation of correction factors. Unfortunately, neither Gauthier et al. [16] nor Puchalski et al. [22] give any recommendations concerning minimisation of these errors. The RC procedure has overall produced the best corrections, i.e. the corrections that are closest to the true EEM. At best, the residuals from the RC procedure are almost 50% lower than the corresponding residuals from the other two correction procedures.
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
363
Table 2 Results of the humic lake experiment Sample namea
Total intensity of true spectrum (108 cps)
Deviation (108 cps) Uncorrected EEM
CG
CL
RC
KV 0 KV 1 KV 2 KV 3 KV 4
163.3 81.4 40.5 20.0 9.8
73.8 27.7 11.1 3.8 1.4
56.1 22.3 9.6 3.5 1.3
43.3 18.9 8.7 3.2 1.2
22.3 9.7 5.1 2.7 0.6
LV 0 LV 1 LV 2 LV 3 LV 4
91.8 45.7 22.6 11.1 5.3
47.7 21.9 10.0 4.6 1.7
44.8 21.1 9.9 4.6 1.7
43.0 20.6 9.7 4.6 1.7
36.6 16.0 7.8 4.1 1.5
TV 0 TV 1 TV 2 TV 3 TV 4
194.4 97.0 48.3 23.9 11.8
97.6 36.0 13.2 4.7 1.6
67.7 26.4 10.1 4.0 1.4
45.7 20.5 8.4 3.6 1.2
35.9 10.5 4.8 1.9 0.7
a
Samples are named as follows: KV = Lake Kringlevattnet, LV = Lake Lille Vektor, TV = Lake Trollvattnet, number = log2 (dilution factor).
4. Conclusions One of the conclusions from the experiments is that none of the corrections managed to completely reproduce a true EEM. For the CG procedure, it should also be noted that it is not straightforward to acquire the cuvette dimensions required to calculate the correction factors. This makes the results of the procedure highly instrument and laboratory dependent. For all correction procedures, the deviations increased with increasing IFE, most likely due to deteriorating S/N ratio in the fluorescence measurement. The experiments showed that the Raman correction procedure presented here is superior to the two absorbance correction procedures. The deviation for the Raman procedure was found to be lower by up to 50%. The Raman correction procedure is experimentally simple and does not require the use of a second instrument. Thus, it could be of special interest for on-line and in situ measurements of fluorescent material. Acknowledgments The authors wish to thank Niklas Str¨omberg for fruitful discussions and valuable comments on the manuscript. Thanks also to G¨osta Larsson for assistance with sampling for the HL experiment. Two anonymous reviewers are also gratefully acknowledged for comments, which considerably improved the manuscript. References [1] K. Kalle, Deutsche Hydrografische Zeitung 2 (1949) 117. [2] K. Kalle, Oceanogr. Mar. Biol.: Annu. Rev. 4 (1966) 91.
[3] T. Almgren, B. Josefsson, G. Nyquist, Anal. Chim. Acta 78 (1975) 411. [4] M. Ewald, C. Belin, P. Berger, J.H. Weber, Environ. Sci. Technol. 17 (1983) 501. [5] G.R. Harvey, D.A. Boron, L.A. Chesal, J.M. Tokar, Mar. Chem. 12 (1983) 119. [6] O.F.X. Donard, M. Lamotte, C. Belin, M. Ewald, Mar. Chem. 27 (1989) 117. [7] E.K. Duursma, in: N.G. Jerlov, E. Steemann Nielsen (Eds.), Optical Aspects of Oceanography, Academic, New York, 1974, p. 237. [8] P.G. Coble, S.A. Green, N.V. Blough, R.B. Gagosian, Nature 348 (1990) 432. [9] K. Mopper, C.A. Schultz, Mar. Chem. 41 (1993) 229. [10] M.M. de Souza Sierra, O.F.X. Donard, M. Lamotte, Mar. Chem. 58 (1997) 51. [11] J. Callahan, M. Dai, R.F. Chen, X. Li, Z. Lu, W. Huang, Mar. Chem. 89 (2004) 211. [12] C.M. Hu, F.E. Muller-Karger, R.G. Zepp, Limnol. Oceanogr. 47 (2002) 1261. [13] J.R. Lakowicz, Principles of Fluorescence Spectroscopy, Kluwer Academic/Plenum, New York, 1999. [14] C.A. Parker, W.T. Rees, Analyst 87 (1962) 83. [15] J.B.F. Lloyd, in: J.N. Miller (Ed.), Standards in Fluorescence Spectrometry, Chapman and Hall, London, 1981, p. 27. [16] T.D. Gauthier, E.C. Shane, W.F. Guerin, W.R. Seitz, C.L. Grant, Environ. Sci. Technol. 20 (1986) 1162. [17] S.A. Tucker, V.L. Amszi, W.E. Acree, J. Chem. Educ. 69 (1992) A8. [18] D.M. McKnight, E.W. Boyer, P.K. Westerhoff, P.T. Doran, T. Kulbe, D.T. Andersen, Limnol. Oceanogr. 46 (2001) 38. [19] R. van Slageren, G. den Boef, W.E. van der Linden, Talanta 20 (1973) 501. [20] S. Determann, R. Reuter, P. Wagner, R. Willkomm, Deep-Sea Res. Part I—Oceanogr. Res. Papers 41 (1994) 659. [21] C.A. Parker, Photoluminescence of Solutions. With Applications to Photochemistry and Analytical Chemistry, Elsevier Pub. Co, Amsterdam, New York [etc.], 1968. [22] M.M. Puchalski, M.J. Morra, R. von Wandruszka, Fresenius J. Anal. Chem. 340 (1991) 341. [23] ANSI/IEEE Std 754-1985, 1985.
Correction of inner-filter effect in fluorescence excitation-emission matrix spectrometry using Raman scatter Tobias Larsson ∗ , Margareta Wedborg, David Turner Department of Chemistry, G¨oteborg University, Kemiv¨agen 10, SE-41296 G¨oteborg, Sweden Received 19 May 2006; received in revised form 5 September 2006; accepted 18 September 2006 Available online 14 October 2006
Abstract Fluorescence excitation-emission matrix (EEM) spectroscopy is a useful tool for interpretation of fluorescence information from natural water samples. One of the major problems with this technique is the inner-filter effect (IFE), i.e. absorption of light at both the excitation and emission wavelengths. The common solutions are to either dilute the sample or apply some form of mathematical correction, most often based on the measured absorbance of the sample. Since dilution is not always possible, e.g. in on-line or in situ EEM recordings, and corrections based on absorbance are hampered primarily by the use of a separate absorbance instrument, neither of these solutions is optimal. In this work, we propose a mathematical correction procedure based on the intensity of Raman scatter from water. This procedure was found to reduce the error after correction by up to 50% in comparison with two absorbance correction procedures. Furthermore, it does not require the use of a separate absorbance measurement, and it is applicable to on-line and in situ EEM recordings, where the IFE would otherwise cause problems. © 2006 Elsevier B.V. All rights reserved. Keywords: Fluorescence; Excitation-emission matrix; Inner-filter effect; Absorbance correction
1. Introduction The fluorescence of humic substances, or gelbstoff, in seawater has been recognised for a long time [1,2]. The low limit of detection has made it easy to measure and useful as a proxy for organic material. However, there is more information available from the fluorescence of humic substances, or, more generally, fluorescent organic matter (FOM), than can be acquired from a single or a few excitation/emission wavelength combinations. A wide variety of wavelength combinations has been used to measure humic fluorescence [3–7], which makes overview difficult. For the last 15–20 years, it has become increasingly common to measure excitation-emission fluorescence matrices (EEMs) [8–11], sometimes also described as total luminescence spectra. This technique involves recording a series of fluorescence emission spectra at different excitation wavelengths and concatenating them into a matrix. With this technique, more information about the FOM is collected, and it also becomes more straightforward to compare results
∗
Corresponding author. Tel.: +46 31 7722284; fax: +46 31 7722785. E-mail address: [email protected] (T. Larsson).
0003-2670/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.aca.2006.09.067
from different researchers. The technique also enables characterisation of FOM from different locations, which opens the possibility to use FOM as a water tracer. With modern-day technology, including diode-array and CCD detectors, it may also become possible to do real-time, on-line or in situ EEM recordings. One of the major problems associated with measurement of FOM in natural samples is the inner-filter effect (IFE), sometimes referred to as self-absorption. This is caused by the absorption of the exciting as well as the fluorescent light (primary and secondary IFE, respectively) by the fluorophore itself, alternatively by another component of the sample. The effect is important already at low absorbances; an absorption coefficient (a) of 11.5 m−1 (corresponding to 0.05 absorbance in a 1 cm cuvette [12]) in the sample reduces the fluorescence by 5% [13]. The mechanism of IFE is different from quenching, although the effect is similar [14]. Quenching is the process in which the fluorescence is reduced because the excited molecules lose their energy via other pathways than emitting light, such as interaction with other molecules. In general there are two ways of reducing the IFE: dilution or mathematical correction of the intensities. The simplest procedure for laboratory measurements is to dilute the sample to
358
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
an acceptable absorbance level (i.e. below a = 11.5 m−1 ) [15]. However, this may lead to problems, such as contamination (from the diluting water) and, perhaps most important, perturbation of the chemistry of the sample (affecting e.g. colloidal stability). Another way is to apply some form of mathematical correction to the fluorescence data. A number of such corrections has been suggested [16–20], but most, if not all, involve the need for a separate absorbance measurement. This usually requires the employment of a separate absorbance instrument, with instrument characteristics different from the fluorescence instrument used for recording the EEM. Consequently, another error is introduced into the analysis. The rather short linear range of the absorbance measurement procedure is another limiting factor. Rayleigh and Raman scatter occur when a molecule has been excited to a virtual energy state by a photon with insufficient energy to completely excite the molecule. The virtual state is unstable, and the molecule quickly relaxes, emitting light. Depending on the original and final vibrational energy level of the molecule, the emitted light may have higher, lower or the same energy as the exciting light. In Rayleigh scatter, the emitted light is of the same energy as the exciting light (Fig. 1). This process, known as elastic scattering, is the most probable one, and, consequently, it has the highest intensity. Rayleigh scatter in the EEM is an exactly diagonal structure occurring at λem = λexc . Raman scatter is an inelastic process, in which the molecule relaxes to a different vibrational energy level of the ground state than the original one. When the molecule relaxes to a higher vibrational level, the emitted light has a lower frequency than the exciting light, and vice versa for the case where the molecule relaxes to a lower vibrational level. The difference in energy is called the Stokes’ shift, a term borrowed from fluorescence spectroscopy, although the underlying phenomena are different. Since the vibrational energy levels depend on the molecule, the Stokes’ shift is independent of the wavelength (frequency) of the exciting light, i.e. the Stokes’ shift of a certain molecule will be constant regardless of the frequency of the exciting light. For example, the Stokes’ shift of water is ∼3400 cm−1 ,
Fig. 1. Jablonski diagram showing different excitation and emission processes. (a) Fluorescence; (b) Rayleigh scatter; (c) Stokes’ shifted Raman scatter; (d) anti-Stokes’ shifted Raman scatter.
regardless of the excitation frequency. The wavelength is inversely proportional to the frequency, which causes the Raman scatter to appear as a slightly curved structure in the EEM (Fig. 2). Both Rayleigh and Raman scatter are dependent on the number of scatterers, i.e. the concentration of the scattering molecule. Raman scattering is a fairly weak phenomenon, requiring either high concentrations of the scatterer or a high-energy light source (such as a laser). As a result, the only molecule of sufficient concentration to produce visible Raman scattering in fluorescence measurements of aqueous samples is water. Consequently, differences in the Raman scatter intensity in EEMs of natural water samples will only depend on the IFE. Raman scattering therefore provides the more appropriate basis for a correction method since it is much less dependent on sample composition. 2. Experimental An EEM consists of a number of emission spectra, each of which contains a Raman scatter peak (Fig. 2a and b). The area
Fig. 2. (a) Typical fluorescence emission spectrum with the different peaks marked. The upper abscissa scale is wavelength in nm and the lower scale is frequency in cm−1 . (b) Contour plot of an EEM with the different structures marked.
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
359
of the Raman scatter peak of each emission spectrum was integrated in Matlab (Version 6.5, The MathWorks, Inc.) using an in-house script, which approximates the area as a series of trapezoids. The areas were stored as a vector, which was subsequently smoothed using a mean filter with window size 3. The smoothing is necessary to reduce integration related errors, which would otherwise impede further data processing. Each value in the smoothed vector thus corresponds to a specific emission spectrum of the sample (or blank sample). For correction of an EEM of a sample for the inner-filter effect, the emission spectra of the EEM were multiplied by the quotient between the values from the vector of the blank (ultrapure water, Milli-Q Plus 185, Millipore Inc.) and the values from the vector of the sample. In mathematical form, the corrected emission spectrum scorr can be expressed as (bold type denotes a mathematical vector): scorr (exc ) = s0 (exc )
rMQ (λexc ) r0 (λexc )
Here, s0 (λexc ) is the uncorrected emission spectrum obtained at excitation wavelength λexc , rMQ (λexc ) is the value for the blank at λexc , and r0 (λexc ) is the value for the uncorrected emission spectrum. The RC procedure was compared to two other procedures (CL and CG) suggested by, respectively, Lakowicz [13] (CL) and Gauthier et al. [16] (CG). In CL, the EEMs are corrected for absorbance by multiplication of each value in the EEM with a correction factor based on the idea that the average path length of the absorption of the excitation and emission light is 1/2 of the cuvette length. Mathematically: Fcorr = Fobs × 10(Aexc +Aem /2) where Fcorr and Fobs are the corrected and uncorrected fluorescence intensities and Aexc and Aem are the absorbance values at the current excitation and emission wavelengths. In CG, an attempt is made to account for the geometry of the cell, i.e. the actual widths of the excitation and emission light paths: Fcorr 2.3dAex 2.3sAem = × 10gAem × −dA ex Fobs 1 − 10 1 − 10−sAem Fcorr , Fobs , Aex , and Aem are as defined above and d, g, and s are defined in Fig. 3. Measurements were made with a slide caliper. The comparison was carried out in two parts: the dichromate and the humic lake (HL) experiment. In the dichromate experiment, the effect of potassium dichromate on the fluorescence of quinine sulphate dihydrate (QS) was investigated: two different QS concentrations (15 and 38 g/L in 0.05 M sulphuric acid) were used for fluorescence. To each of the QS solutions, four different concentrations of potassium dichromate (1, 5, 10, and 25 mg/L) were added to produce IFE. For the HL experiment, the samples were collected from three different humic lakes in western Sweden: Lake Lille Vektor (LV), Lake Kringlevattnet (KV) and Lake Trollvattnet (TV) in winter conditions.
Fig. 3. Cuvette dimensions for CG procedure (adapted from Gauthier et al. [16]). s represents the width of the excitation light beam, g is the distance from the edge of the excitation light beam to the edge of the cuvette, and d is the width of the cuvette seen by the detector. In this instrument setting, s was estimated to 0.9 cm, g to 0.1 cm, and d to 0.6 cm.
2.1. Chemicals All reagents were of analytical grade. QS was obtained from Carl ROTH, potassium dichromate was obtained from Scharlau and sulphuric acid was obtained from Merck. All solutions and dilutions were prepared with ultrapure water, which was also used for blank samples. From each of the HL samples, four subsamples were prepared by dilution with ultrapure water. 2.2. Data recording The EEMs for the dichromate experiment were recorded by scanning emission wavelengths from 230 to 600 nm with 1 nm increments. Excitation wavelengths ranged from 230 to 450 with 5 nm intervals. For the HL experiment, emission was scanned from 230 to 600 nm with 1 nm increments. Excitation wavelengths ranged from 230 to 420 nm with 10 nm intervals. All fluorescence measurements were performed on a Fluoromax-2 spectrofluorometer (Jobin-Yvon Inc.). The bandwidth used for both excitation and emission was 5 nm. The absorbance spectra were recorded from 230 to 600 nm with 1 nm increments on a Hewlett–Packard 8453 UV–vis diodearray absorbance spectrophotometer with 1 nm bandwidth and a path length of 1 cm. Absorbance spectra are reported as absorption coefficients for the above range as recommended by Hu et al. [12]. 2.3. Data processing The raw data from both instruments was imported into Matlab (Version 6.5, The Mathworks, Inc.), where the different corrections were subsequently applied.
360
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
As a measure of the accuracy for each of the three procedures, the deviation of the corrected from the true EEM was used. This deviation was defined as: true corr − Ii,j Ii,j i
j
In the literature, absorbances of both 0.02 [21] and 0.05 [13] have been judged sufficiently low to avoid IFE. This corresponds to an absorption coefficient that does not exceed 4.6 m−1 or 11.5 m−1 , respectively, within the wavelength range used. In this work the lower value was used. For the dichromate experiment, the QS solutions with no added dichromate were used as the true values, since the QS did not contribute to the absorbance. For the HL experiment, the lower criterium above was met with subsamples diluted to 1/64 of the original concentrations. The Raman and Rayleigh scatter were excluded from the comparison. 3. Results and discussion The absorbance spectra from the two experiments are shown in Fig. 4. The impact of the absorbance, i.e. the inner-filter effect, is clearly visible in Fig. 5 (top row) and Fig. 6 (top row), which show the true and distorted EEM for 15 g L−1 QS (Fig. 5, top row) and for Lake Kringlevattnet (Fig. 6, top row). In both cases, the inner-filter effect has reduced the fluorescence intensities by up to 50% of the total fluorescence. An important practical consideration concerning the absorbance correction procedures is that the bandwidths of the two instruments were not the same. This is one of the more serious objections to the absorbance correction procedures. The spectrofluorometer had adjustable bandwidth settings for the excitation as well as the emission monochromator. These settings must be optimised for each sample set, and it is not uncommon that it is preferable to have different bandwidths for excitation and emission. In this work, the optimal bandwidth setting was found to be 5 nm for both excitation and emission. The spectrophotometer used in this work, on the other hand, had a fixed bandwidth of 1 nm. The bandwidth affects both peak height and peak width, and consequently the use of two different instruments will likely be a source for errors, although the magnitude is as yet unknown.
Fig. 4. (a) Absorbance spectra for the dichromate experiment; (b) absorbance spectra for the humic lake experiment. Absorption coefficient as defined by Hu et al. [12].
The results from the correction procedures are shown in Fig. 5 and Table 1 (the dichromate experiment) and in Fig. 6 and Table 2 (the HL experiment). In general, the deviation of the correction increases as the IFE increases, regardless of correction procedure (Tables 1 and 2). Figs. 5 and 6 show some of the correction results; the residuals are of particular interest. Note that the fluorescence scales of the residuals from the corrections are not the same as shown for the original EEMs. The effect of the dual absorbance peaks of dichromate is clearly visible in the residuals of the dichromate experiment (Fig. 5, right column). For both the CG and the CL procedures (Fig 5 CG and CL procedure residuals), there is an obvious correspondence between the two residual peaks and the dichromate absorbance (Fig 4a). The residual for the RC procedure also has some correspondence to the dichromate absorbance, but it is not as pronounced as for the other procedures. A similar pattern can be observed in the results from the HL experiment. The residuals for the CG and
Table 1 Results of the dichromate experiment QS (g L−1 )
15
38
K2 Cr2 O7 (mg L−1 )
Total intensity of true spectrum (108 cps)
0 1 5 10 25
31.6
0 1 5 10 25
81.5
Deviation (108 cps) Uncorrected EEM
CG
CL
RC
3.1 8.4 9.4 14.6
2.9 7.3 7.3 10.5
2.6 6.6 5.9 7.3
1.5 4.9 3.3 4.4
9.5 22.9 25.5 38.5
8.9 20.3 20.4 28.3
8.2 18.5 16.8 20.4
5.5 12.9 12.9 17.2
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
361
Fig. 5. Sample EEM results from the dichromate experiment. Note that the residuals of the three correction procedures have a different fluorescence scale. Rayleigh and Raman scatter peaks have been removed manually by setting their value to Not-a-Number (NaN) [23].
CL procedures are highest at the shorter wavelengths, where the IFE also is the largest. The same is equally true for the RC procedure, but, again, not as pronounced. The CG procedure is associated with the largest residuals for both experiments (Figs. 5 and 6). This is somewhat contradictory to the work by Puchalski et al. [22], who also compared the CG and CL procedures. However, Puchalski et al. used only one excitation spectrum and one emission spectrum for each
experiment, as opposed to using a major part of the EEM, as was the case in the present study. This might possibly account for the discrepancy. Furthermore, the work of Puchalski et al. was done using protocols from studies of fluorescence quenching; it is possible that the different approaches may also cause discrepancies. The residuals from the CL procedure are generally on the same level as the residuals from the CG procedure or slightly
362
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
Fig. 6. Sample EEM results from the humic lake experiment. Note that the residuals of the three correction procedures have a different fluorescence scale. Rayleigh and Raman scatter peaks have been removed manually by setting their value to Not-a-Number (NaN) [23].
smaller (Figs. 5 and 6). It is somewhat surprising that the simple correction equation of the CL procedure produces a better correction than the more elaborate equation, derived from Parker [21] and Lloyd [15], of the CG procedure. However, it is quite difficult to estimate the different measures of the light paths (Fig. 3), and it was further noted that the light beam of the instrument used in this work was not quite parallel. Consequently, the CG procedure is hampered by errors in the estimation and cal-
culation of correction factors. Unfortunately, neither Gauthier et al. [16] nor Puchalski et al. [22] give any recommendations concerning minimisation of these errors. The RC procedure has overall produced the best corrections, i.e. the corrections that are closest to the true EEM. At best, the residuals from the RC procedure are almost 50% lower than the corresponding residuals from the other two correction procedures.
T. Larsson et al. / Analytica Chimica Acta 583 (2007) 357–363
363
Table 2 Results of the humic lake experiment Sample namea
Total intensity of true spectrum (108 cps)
Deviation (108 cps) Uncorrected EEM
CG
CL
RC
KV 0 KV 1 KV 2 KV 3 KV 4
163.3 81.4 40.5 20.0 9.8
73.8 27.7 11.1 3.8 1.4
56.1 22.3 9.6 3.5 1.3
43.3 18.9 8.7 3.2 1.2
22.3 9.7 5.1 2.7 0.6
LV 0 LV 1 LV 2 LV 3 LV 4
91.8 45.7 22.6 11.1 5.3
47.7 21.9 10.0 4.6 1.7
44.8 21.1 9.9 4.6 1.7
43.0 20.6 9.7 4.6 1.7
36.6 16.0 7.8 4.1 1.5
TV 0 TV 1 TV 2 TV 3 TV 4
194.4 97.0 48.3 23.9 11.8
97.6 36.0 13.2 4.7 1.6
67.7 26.4 10.1 4.0 1.4
45.7 20.5 8.4 3.6 1.2
35.9 10.5 4.8 1.9 0.7
a
Samples are named as follows: KV = Lake Kringlevattnet, LV = Lake Lille Vektor, TV = Lake Trollvattnet, number = log2 (dilution factor).
4. Conclusions One of the conclusions from the experiments is that none of the corrections managed to completely reproduce a true EEM. For the CG procedure, it should also be noted that it is not straightforward to acquire the cuvette dimensions required to calculate the correction factors. This makes the results of the procedure highly instrument and laboratory dependent. For all correction procedures, the deviations increased with increasing IFE, most likely due to deteriorating S/N ratio in the fluorescence measurement. The experiments showed that the Raman correction procedure presented here is superior to the two absorbance correction procedures. The deviation for the Raman procedure was found to be lower by up to 50%. The Raman correction procedure is experimentally simple and does not require the use of a second instrument. Thus, it could be of special interest for on-line and in situ measurements of fluorescent material. Acknowledgments The authors wish to thank Niklas Str¨omberg for fruitful discussions and valuable comments on the manuscript. Thanks also to G¨osta Larsson for assistance with sampling for the HL experiment. Two anonymous reviewers are also gratefully acknowledged for comments, which considerably improved the manuscript. References [1] K. Kalle, Deutsche Hydrografische Zeitung 2 (1949) 117. [2] K. Kalle, Oceanogr. Mar. Biol.: Annu. Rev. 4 (1966) 91.
[3] T. Almgren, B. Josefsson, G. Nyquist, Anal. Chim. Acta 78 (1975) 411. [4] M. Ewald, C. Belin, P. Berger, J.H. Weber, Environ. Sci. Technol. 17 (1983) 501. [5] G.R. Harvey, D.A. Boron, L.A. Chesal, J.M. Tokar, Mar. Chem. 12 (1983) 119. [6] O.F.X. Donard, M. Lamotte, C. Belin, M. Ewald, Mar. Chem. 27 (1989) 117. [7] E.K. Duursma, in: N.G. Jerlov, E. Steemann Nielsen (Eds.), Optical Aspects of Oceanography, Academic, New York, 1974, p. 237. [8] P.G. Coble, S.A. Green, N.V. Blough, R.B. Gagosian, Nature 348 (1990) 432. [9] K. Mopper, C.A. Schultz, Mar. Chem. 41 (1993) 229. [10] M.M. de Souza Sierra, O.F.X. Donard, M. Lamotte, Mar. Chem. 58 (1997) 51. [11] J. Callahan, M. Dai, R.F. Chen, X. Li, Z. Lu, W. Huang, Mar. Chem. 89 (2004) 211. [12] C.M. Hu, F.E. Muller-Karger, R.G. Zepp, Limnol. Oceanogr. 47 (2002) 1261. [13] J.R. Lakowicz, Principles of Fluorescence Spectroscopy, Kluwer Academic/Plenum, New York, 1999. [14] C.A. Parker, W.T. Rees, Analyst 87 (1962) 83. [15] J.B.F. Lloyd, in: J.N. Miller (Ed.), Standards in Fluorescence Spectrometry, Chapman and Hall, London, 1981, p. 27. [16] T.D. Gauthier, E.C. Shane, W.F. Guerin, W.R. Seitz, C.L. Grant, Environ. Sci. Technol. 20 (1986) 1162. [17] S.A. Tucker, V.L. Amszi, W.E. Acree, J. Chem. Educ. 69 (1992) A8. [18] D.M. McKnight, E.W. Boyer, P.K. Westerhoff, P.T. Doran, T. Kulbe, D.T. Andersen, Limnol. Oceanogr. 46 (2001) 38. [19] R. van Slageren, G. den Boef, W.E. van der Linden, Talanta 20 (1973) 501. [20] S. Determann, R. Reuter, P. Wagner, R. Willkomm, Deep-Sea Res. Part I—Oceanogr. Res. Papers 41 (1994) 659. [21] C.A. Parker, Photoluminescence of Solutions. With Applications to Photochemistry and Analytical Chemistry, Elsevier Pub. Co, Amsterdam, New York [etc.], 1968. [22] M.M. Puchalski, M.J. Morra, R. von Wandruszka, Fresenius J. Anal. Chem. 340 (1991) 341. [23] ANSI/IEEE Std 754-1985, 1985.