2-Furaldehyde in Honey by a Modified Winkler Method

ANALYTICAL SCIENCES APRIL 2016, VOL. 32 413 2016 © The Japan Society for Analytical Chemistry

Flow Injection Analysis of 5-(Hydroxymethyl)-2-furaldehyde in Honey by a Modified Winkler Method

Karine CASTOLDI, Maria Izabel MILANI, Eduardo L. ROSSINI, Leonardo PEZZA, and Helena R. PEZZA†

Instituto de Química, Universidade Estadual Paulista “Julio de Mesquita Filho” (UNESP), R. Prof. Francisco Degni 55, PO Box 355, 14800-900, Araraquara, SP, Brazil

One of the quality indicators for honey is 5-(hydroxymethyl)-2-furaldehyde (HMF), which is formed during the heating or aging of honey. The International Honey Commission recommends three methods for the determination of HMF in honey: the Winkler method, the White method, and determination by HPLC. The Winkler method uses the carcinogenic substance p-toluidine, which is not in accordance with the principles of Green Chemistry. The present work describes the determination of HMF in honey by flow injection analysis (FIA) using a modified Winkler method, replacing p-toluidine with p-aminobenzoic acid. The linear range was 1.00 to 40.0 mg L–1, the limit of detection (LOD) was 0.43 mg L–1, and the limit of quantification (LOQ) was 1.32 mg L–1. The method is an efficient and environmentally friendly technique for the analysis of HMF in honey.

Keywords 5-(Hydroxymethyl)-2-furaldehyde (HMF), flow injection analysis, p-aminobenzoic acid, honey

(Received October 7, 2015; Accepted November 13, 2015; Published April 10, 2016)

The Winkler method should not be used when the other two Introduction methods are available, because of the need for p-toluidine, which is a carcinogenic compound.10 The White method is Honey is a natural substance produced by Apis mellifera bees, unable to provide accurate measurements of low concentrations with high nutritional value due to its complex composition. of HMF in honey.11 The HPLC method uses methanol as a It essentially consists of sugars and water, with important solvent,10 which is potentially harmful to the environment and secondary components such as minerals, organic acids, enzymes, human health due to its high toxicity.12 According to Bogdanov,10 proteins, and vitamins.1–3 The composition and nutritional samples for HPLC analysis must be prepared and analyzed properties of honey vary depending on the flora, climate, and within half an hour, after which the level of HMF present in the soil.4–6 sample decreases drastically.13 Honey is submitted to a heating and filtration process in order Some methodologies described in the literature for the to remove bacteria, yeasts, and spores, and to retard granulation determination of HMF in foods use flow injection analysis.14–16 and prevent fermentation.1,7,8 Although this heating process Disadvantages of these techniques are the use of p-toluidine14,15 facilitates the processing of honey, high temperatures can or the need for heating.16 destroy vitamins and enzymes, and lead to the production of The advantages of flow injection analysis (FIA) are that it hydroxymethylfurfural (HMF), reducing the quality of honey.1,4 enables increases in the speed and number of analyses, while HMF is practically absent in fresh honey, but concentrations also decreasing the risk of analyst-induced errors.17 The increase with the temperature and time of heating, as well as versatility of the FIA system enables it to be adapted to different due to improper storage and adulteration.4,8 At low heating detectors, using different settings, which amplifies the number temperatures, the composition and pH can also influence the of potential applications of the technique.18 formation of HMF in honey.8 HMF is formed in foods by Analytical methods often employ reagents (such as p-toluidine) heating monosaccharides under acid conditions,9 and it can be that are more harmful to the environment than the substances to used in the quality control of honey.6 be analyzed.17 There are currently increasing efforts to develop Numerous analytical methods have been developed for the analytical methods that are less harmful17 and that are coherent determination of HMF. According to the International Honey with the principles of Green Chemistry. The aim is to develop Commission,10 the recommended methods are two new analytical methodologies that reduce or eliminate the use spectrophotometric techniques (the Winkler and the White) and and generation of substances harmful to human health and the high performance liquid chromatography (HPLC). According environment.19,20 to the Codex Alimentarius2 and the European Union,3 only the In order to minimize analytical risks and follow the principles White and HPLC methods are acceptable. of Green Chemistry, we developed an alternative method for the analysis of HMF, modifying the Winkler method by replacing † To whom correspondence should be addressed. p-toluidine with a non-toxic compound and using flow injection E-mail: [email protected] analysis (FIA) in order to increase the sampling rate and reduce 414 ANALYTICAL SCIENCES APRIL 2016, VOL. 32

manual injector-commutator, loops were filled with volumes of the sample and reagent. In the injection position, the two solutions were loaded into the carrier stream (0.08 mol L–1 HCl) and merged at the confluence point. A yellow-colored product was formed inside the reactor coil (kept at room temperature) and was then carried to the detector flow cell, where the transient absorbance signal of the product was measured at 390 nm. This signal was proportional to the HMF content in the sample, and a second signal was measured at 700 nm. After measurement of the absorbance maximum, the manual injector-commutator was Fig. 1 Schematic illustration of the FIA system manifold merging returned to the sampling position and another cycle was started. zones. R, Reagent ([PABA] = 6000 mg L–1, [barbituric acid] = 1000 mg L–1, and volume = 603 μL); S, sample (volume = 427 μL); Preparation of samples CA, carrier stream ([HCl] = 0.08 mol L–1); P, peristaltic pump (flow Seven samples of honey originating from different flowers –1 rate = 1.9 mL min ); I, manual injector-commutator; C, confluence; were used to evaluate the performance of the method. The RC, reactor coil (volume = 1005 μL); D, detector; W, waste. samples were purchased in local markets (in the Brazilian States of Paraná, Rio Grande do Norte, and São Paulo), and from an apiary (in Araraquara, São Paulo State). The samples were prepared by dissolving 2.5 g portions of contact with reagents and waste for those performing the honey in approximately 25 mL of water and transferring analysis. quantitatively to 50 mL volumetric flasks. To these were added 0.5 mL of Carrez I solution (15% (m/v) of potassium hexacyanoferrate), followed by mixing and addition of 0.5 mL Experimental of Carrez II solution (30% (m/v) of zinc acetate dehydrate). The volumes were made up to 50 mL with water. These Apparatus solutions were filtered through filter paper (Whatman No. 1), The flow injection system (Fig. 1) comprised a peristaltic and the first 10 mL of the filtrates were discarded. pump (Gilson Model MINIPULS 3 with 8 channels), Tygon® tubing (1.02 mm i.d.) for propelling the reagent and sample Experimental design solutions, polytetrafluoroethylene (PTFE) tubes (0.8 mm i.d.), After identification of the significant parameters, the an acrylic “y” confluence piece, a manual acrylic injector- operational variables were optimized by fractional factorial commutator, and a quartz cuvette flow cell (10 mm optical path design (27–3) in order to obtain the best analytical conditions. length and 160 μL internal volume). Measurements of The variables involved in the reaction were the HCl, absorbance at 390 and 700 nm were carried out in kinetic mode p-aminobenzoic acid, and barbituric acid concentrations, using a photodiode array spectrophotometer (Model HP8453). together with flow rate, reactor coil size, and the volumes of the sample and reagent loops. The matrix was designed using Chemicals Minitab software and optimization graphs were constructed All reagents employed were of analytical grade and were used using Statistica 8.0 software. without prior purification. Ultrapure water (18 MΩ cm, Milli-Q After completion of the fractional factorial design, a univariate system, Millipore) was used to prepare the solutions. The procedure was conducted, varying the factor likely to have the carrier stream (0.08 mol L–1 HCl solution) was prepared by greatest effect in order to achieve higher sensitivity. This factor appropriate dilution of concentrated HCl (Merck, Darmstadt, was the reactor coil size, and the volumes used were 804, 904, Germany). An aqueous solution containing 6.00 × 103 mg L–1 1005, and 1105 μL. The HMF concentration used was p-aminobenzoic acid (PABA) (Henrifarma, Brazil) and 1.00 × 15 mg L–1. 103 mg L–1 barbituric acid (Merck) was prepared daily. The solution was left for 20 min in a thermostatic bath at a Study of interferences temperature of 50°C to ensure solubilization of the reagents. An evaluation was made of possible interferences from the Stock aqueous standard solutions of HMF (Sigma-Aldrich, major compounds (glucose, fructose, and sucrose) commonly China) were prepared daily at concentrations of 200 mg L–1 and present in honey. Solutions containing 2.0 mg L–1 of HMF and were diluted to the required working concentrations. each of the major compounds at concentrations equal to or 10 times greater than that of HMF were evaluated under the same Determination of product structure conditions described above. In order to determine the possible structure of the product, a solution was prepared containing barbituric acid, p-aminobenzoic Reference method acid, and HMF. This solution was analyzed by mass The official method used was Method 980.23 (hydroxy- spectrometry, using a Thermo Scientific LCQ Fleet Ion Trap methylfurfural in honey) described in the 18th edition of the LC/MSn instrument operated in full scan negative mode with AOAC International manual.21 In this spectrophotometric capillary voltage (ESI) of 5 kV, N2 flow rate 8 (arbitrary units), method, the measurements were carried out using a Model transfer capillary temperature of 275°C, transfer capillary HP8453 photodiode array spectrophotometer. voltage of 11 V, and sample solution flow rate of 5 μL min–1.

Procedure Results and Discussion Figure 1 shows the merging zones of the FIA system, where discrete aliquots of reagent and sample were used to avoid Based on the work of Winkler22 and Foley, Stanford and unnecessary wastage of reagent. In the sampling position of the McKennis,23 the reaction of HMF with barbituric acid and ANALYTICAL SCIENCES APRIL 2016, VOL. 32 415

Table 1 Description of the factors used in the fractional factorial design (27–3)

Lower level Upper level Factor (–1) (+1)

Volume of reagent loop/μL 151 603 Volume of sample loop/μL 151 427 Reactor coil size/μL 422 804 Flow rate/mL min–1 1.05 1.9 Fig. 2 Probable product of the reaction involving HMF, barbituric [HCl]/mol L–1 0.01 0.08 acid, and p-aminobenzoic acid. [PABA]/mg L–1 1000 6000 [Barbituric acid]/mg L–1 200 1000

.

Fig. 3 Mass spectrum of the solution (negative full scan mode) containing the barbituric acid and p-aminobenzoic acid reagents and the HMF analyte.

p-aminobenzoic acid probably leads to the product shown in Fig. 2. In the present work, p-aminobenzoic acid was chosen to Fig. 4 Pareto chart for optimization using a 27–3 fractional factorial replace p-toluidine in the Winkler method, which in its original design. form uses barbituric acid and p-toluidine as the reagents. Since p-toluidine is a toxic compound that is potentially carcinogenic, the International Honey Commission only recommends use of the original Winkler method if one of the other methods (White gradients formed during the transport of solutions in the flow or method or HPLC) is not available.10 Aniline analogues to in the sample zone, which disturb the transient signals. When p-toluidine could be adopted to modify the Winkler method for the sample solution has a different physico-chemical composition HMF. In this context, PABA has been selected because it is to that of the carrier solution, the form and height of the transient widely distributed in nature as a B complex factor, and has low signals can be modified. The dispersion of the sample in the acute oral toxicity. The low toxicity of p-aminobenzoic acid12 sample zone also influences the Schlieren effect. One way to makes its use preferable to p-toluidine and anilines analogues. eliminate this effect is to perform spectrophotometric measurements at two wavelengths. The first wavelength Determination of product structure corresponds to the analytical signal plus a non-specific The solution of reagents with the analyte was submitted to absorption, while the second wavelength is where the absorption analysis by mass spectrometry in negative full scan mode. The of the analyte is negligible, leaving only the signal corresponding spectrum obtained can be seen in Fig. 3. The proposed structure to the Schlieren effect. The difference between the two signals of the colored product (Fig. 2) is analogous to that proposed by is proportional to the analyte concentration and is used for Winkler,22 but with substitution of aniline by p-aminobenzoic quantification.28 acid. The suggested product would therefore have a molecular Experiments were carried out to establish the conditions that mass of approximately 373 g mol–1. provided a maximum absorbance response at 390 nm, as well to The spectrum acquired in negative mode showed a base peak select the significant variables. A second wavelength of 700 nm at m/z 372.08, corresponding to the molecular ion (M-1), in was employed, at which only the signal corresponding to the agreement with the proposed structure. Peaks at m/z 127 and Schlieren effect was present. The signal at 700 nm was 136 corresponded to the barbituric acid and p-aminobenzoic subtracted from the signal at 390 nm in order to generate the acid reagents, respectively. We suggest that peaks at m/z 373, calibration curve for quantification of HMF. 342, 329, 299, and 256 represented the protonated molecular ion A fractional factorial design (27–3) was used to select the values

(M) with losses of CH2O, CONH, (H2O+CONH), and of the variables that maximized the analytical signal, requiring (H2O+(CONH)2), respectively. 16 experiments. In order to minimize environmental effects, the experiments were performed randomly, in triplicate. Upper Optimization of variables (+1) and lower (–1) levels were selected for the seven factors Various studies have reported problems due to Schlieren (Table 1), based on the preliminary experiments, and the HMF effects in flow injection systems.24–27 This effect was also concentration was 20 mg L–1 in all the tests. observed in the present work. The Schlieren effect is a The Pareto chart (Fig. 4) shows the individual effects of the phenomenon caused by refractive index and concentration various parameters. The chart includes a vertical line showing 416 ANALYTICAL SCIENCES APRIL 2016, VOL. 32

Table 2 Analytical parameters and figures of merit for the Table 3 Recovery of HMF added to honey samples proposed method for determination of HMF in honey Added value/ Found value/ Sample Recovery, % Parameter Value mg L–1 mg L–1 a

Wavelength 390 and 700 nm Orange honey 2.00 2.08 ± 0.01 103.9 ± 0.7 Flow rate 1.9 mL min–1 4.00 3.70 ± 0.02 92.4 ± 0.4 Volume of reagent loop 603 μL 6.00 5.51 ± 0.08 91.9 ± 1.3 Volume of sample loop 427 μL 8.00 7.07 ± 0.06 88.4 ± 0.8 Reactor coil 804 μL Eucalyptus honey 2.00 2.08 ± 0.08 103.8 ± 3.8 HCl concentration 0.08 mol L–1 4.00 3.73 ± 0.02 93.3 ± 0.5 Sampling frequency 20 samples h–1 6.00 5.46 ± 0.03 91.1 ± 0.4 Linear range 1.00 – 40.0 mg L–1 8.00 7.42 ± 0.15 92.7 ± 1.8 Linear equation A = –0.00103 + 0.03069CHMF Correlation coefficient (R) 0.999 a. Average of three determinations. LOD 0.43 mg L–1 LOQ 1.32 mg L–1

3.5% for the 20 mg L–1 level, respectively. These values are within the acceptable ranges for such analyte concentrations.29 the 95% statistical significance limit. Any bar that exceeds this The calibration curve showed linearity in the concentration line indicates a significant influence on the response range from 1.00 to 40.0 mg L–1 (R = 0.999). The calibration (absorbance). curve and other figures of merit are shown in Table 2. It can be seen from Fig. 4 that the only variable that had a The criteria used for the LOD and LOQ calculations were significant effect was the PABA concentration, with a positive LOD = 3SD/S and LOQ = 10SD/S. The standard deviation impact. This variable therefore had the greatest influence on the (SD) was based on the residual SD of the regression line with response. However, it was not possible to increase the PABA slope S.30 concentration, as this caused precipitation of the reagents. The other variables (HCl and barbituric acid concentrations, flow Study of interferences rate, reactor coil size, and volumes of the sample and reagent The major compounds present in honey (glucose, fructose, loops) showed no significant influence on the reaction. The best and sucrose) were evaluated in terms of their potential result was obtained when all the factors were set at the upper interferences in the method. The tests were conducted using level (+1), so the corresponding values were used in the analyte:interferent ratios of 1:1 and 1:10, and analyses were also subsequent experiments. performed in the presence of the interfering substances alone. A univariate procedure was performed in order to increase the A value of ±5% in the absorbance signal was used as the sensitivity of the method. The factor that most influenced the criterion to indicate interference. The percentages of HMF reaction was the concentration of PABA, however it was not found in the fortified solutions were in the range from 101.3 to possible to use a higher concentration, due to the problem of 104.6%, showing that there was no interference from any of the precipitation, and for the same reason it was not possible to three compounds evaluated. increase the concentration of barbituric acid. The next factors that could be modified were the volumes of Accuracy and recovery the reagent and sample loops. However, the volumes of these Addition and recovery tests were carried out to evaluate loops were already sufficiently large, and further changes in accuracy and to detect possible matrix interferences. Samples them did not result in any increase in sensitivity. The volumes of two different types of honey were fortified with HMF and of the loops were therefore left at the maximum values, as each sample was analyzed three times. The amounts added and discussed above. the percentage recovery values are provided in Table 3. The For these reasons, the factor studied was the volume of the recoveries were between 88.4 and 103.9%, indicating good reaction coil. This was increased by extending its length, so that accuracy and an absence of matrix effects. the reaction mixture took longer to reach the detector and the The recovery percentages were within the expected range; reaction proceeded further towards completion, resulting in a recovery efficiency is influenced by the sample matrix, the gain in sensitivity. The reaction coil volumes tested were 804, sample processing procedure, and the analyte concentration, and 904, 1005, and 1105 μL. It was found that increasing the for this concentration range, the expected recovery is between volume of the coil improved the sensitivity, up to a maximum 80 and 110%.29,30 volume of 1005 μL. After this point, further increase of the volume had a negative influence on the absorbance, due to Application of the method to honey samples dispersion of the reactants and sample in the carrier stream. The method was applied to seven honey samples originating A reaction coil volume of 1005 μL was therefore selected for from flowers of different plants, such as: assa-peixe, cipó-uva, use in the subsequent tests. eucalyptus, orange, and wild flower, and the results showed that the method could be successfully used, irrespective of the origin Analytical characteristics of the proposed method of the honey. The following analytical parameters of the method were Statistical comparison, using the F-test and t-test at 95% determined: limit of detection (LOD), limit of quantification confidence levels, was made of the values obtained by the (LOQ), precision, accuracy, and linear range. The precision of proposed method and an official method21 (AOAC International, the method was evaluated in terms of the relative standard 2005). The results (Table 4) showed that the methods were in deviation (%RSD), using intra-day and inter-day measurements good agreement. The calculated t and F values did not exceed of two standard solutions (5 and 20 mg L–1). The %RSD values the critical values, indicating there was no significant difference obtained were 0.6 and 2.3% for the 5 mg L–1 level, and 1.8 and between the two methods (with the exception of sample 4). ANALYTICAL SCIENCES APRIL 2016, VOL. 32 417

Table 4 HMF concentrations obtained using the proposed Toxicol., 2013, 34, 733. method and the official method 2. Codex Alimentarius Commission, Alinorm 01/25, http:// Proposed Official www.codexalimentarius.org/input/download/report/277/ RSD, RSD, Sample method/ method/ t-valuec F-valued Al01_25e.pdf. %b %b mg kg–1 a mg kg–1 a 3. European Union, Official Journal of the European Communities, Council Directive 2001/110/EC of 20 1 34.62 ± 0.62 1.79 34.04 ± 1.38 4.05 0.64 4.91 December 2001 relating to honey, http://eur-lex.europa.eu/ 2 27.36 ± 0.92 3.36 27.09 ± 0.88 3.25 0.28 1.09 LexUriServ/LexUriServ.do?uri=OJ:L:2002:010:0047:0052: e — e — — — 3 nd nd EN:PDF. 4 nde — 2.66 ± 0.11 4.14 — — 4. S. Ajlouni and P. Sujirapinyokul, Food Chem., 2010, 119, 5 25.70 ± 0.64 2.49 22.44 ± 2.55 11.36 1.78 15.9 6 14.55 ± 0.92 6.32 10.73 ± 0.72 6.71 2.72 7.03 1000. 7 32.10 ± 0.75 2.34 32.55 ± 0.80 2.46 0.67 1.17 5. X. Feás, J. Pires, A. Iglesas, and M. L. Estevinho, Food Chem. Toxicol., 2010, 48, 3462. a. Average ± standard deviation (SD) of three determinations. 6. E. Crane, “O livro do mel”, Nobel, 1985, São Paulo, 226. b. Relative standard deviation (RSD) of three determinations. 7. J. W. Jr. White and J. Siciliano, J. Assoc. Off. Anal. Chem., c. Critical values of t at 95% confidence level (t = 4.303). d. Critical values of F at 95% confidence level (F = 19). 1980, 63, 7. e. nd = not detected. 8. B. Fallico, M. Zappalà, E. Arena, and A. Verzera, Food Sample: 1, assa-peixe; 2, cipó-uva; 3 and 5, eucalyptus; 4 and 6, Chem., 2004, 85, 305. orange; 7, wild flower. 9. H. D. Belitz, W. Grosch, and P. Schieberle, “Carbohydrates. In Food Chemistry”, 2009, Springer, New York, 1070. 10. S. Bogdanov, Harmonized Methods of the International Honey Commission, http://www.ihc-platform.net/ihcmethods In other work, it was found that the official method failed to 2009.pdf. provide an accurate measurement of HMF in honey samples 11. D. A. Wunderlin, S. F. Pesce, M. V. Amé, and P. F. Faye, J. containing low quantities of the analyte.11 Agric. Food Chem., 1998, 46, 1855. The proposed method is therefore suitable for the analysis of 12. J. O’Neil (ed.), The Merck Index, Whitehouse Station, honey samples. The amount of waste generated was low, with 2006, Cambridge, 1896. each analysis consuming 3.62 mg of PABA, 0.60 mg of 13. A. Käzig, D. Kaufmann, and S. Bogdanov, Stability of barbituric acid, and 21.35 mg of sample. The technique is also Hydroymethylfurfural during Determination by HPLC, more sensitive and has a lower detection limit, compared to http://www.ihc-platform.net/kaenzig2001.pdf. methods previously reported in the literature14–16 for the 14. F. Salinas, A. Espinosa-Mansilla, and J. J. B. Nevado, determination of HMF in food. Fresenius. J. Anal. Chem., 1991, 340, 250. 15. F. de la Iglesia, F. Lázaro, R. Puchades, and A. Maquieira, Food Chem., 1997, 60, 245. Conclusions 16. A. Espinosa-Mansilla, A. M. de la Peña, and F. Salinas, J. AOAC Int., 1993, 76, 1255. The proposed method for the determination of HMF is efficient 17. F. R. P. Rocha, J. A. Nóbrega, and O. Fatibello-Filho, Green and more environmentally friendly than the official Winkler Chem., 2001, 3, 216. method, since it uses a compound (p-aminobenzoic acid) that is 18. J. Ru˚žicˇ ka and E. H. Hansen, “Flow Injection Analysis”, less toxic to humans, compared to the compound used in the 2nd ed., John Wiley & Sons, 1988, Chap. 2, New York, official method (p-toluidine). Another important advantage is 498. the use of an FIA system, where the analyst has less contact 19. P. Anastas, Crit. Rev. Anal. Chem., 1999, 29, 167. with reagents and waste, less waste is generated, and analytical 20. P. Anastas and N. Eghbali, Chem. Soc. Rev., 2010, 39, 301. errors are reduced because the system is semi-automated. The 21. W. Horwitz (ed.), Official Methods of Analysis of AOAC analysis is faster, compared to the Winkler22 and official AOAC International, 2006, Chap. 44, AOAC International, methods,21 which operate in batch mode. In its application to Maryland, 53. real samples, the results obtained with the proposed method 22. O. Winkler, Z. Lebensm. Unters. For., 1955, 102, 161. showed no statistically significant differences, compared to 23. W. M. Jr. Foley, G. E. Sanford, and H. Jr. McKennis, J. Am. those obtained using the methods described in the literature. Chem. Soc., 1952, 74, 5489. This new flow injection procedure does not require any sample 24. A. C. B. Dias, E. P. Borges, E. A. G. Zagatto, and P. J. pretreatment steps. Worsfold, Talanta, 2006, 68, 1076. 25. I. D. McKelvie, D. M. W. Peat, G. P. Matthews, and P. J. Worsfold, Anal. Chim. Acta, 1997, 351, 265. Acknowledgements 26. F. R. P. Rocha and J. A. Nóbrega, J. Brazil. Chem. Soc., 1997, 8, 625. We would like to thank the Brazilian National Council for 27. E. A. G. Zagatto, M. A. Z. Arruda, A. O. Jacintho, and I. L. Scientific and Technological Development (CNPq) for financial Mattos, Anal. Chim. Acta, 1990, 234, 153. support and for a scholarship grant. 28. F. R. P. Rocha and J. A. Nóbrega, Quim. Nova, 1996, 19, 636. 29. L. Huber, LC/GC Magazine, http://geocities. References internetarchaeology.org/HotSprings/Spa/6896/methval.pdf. 30. M. Rambla-Alegre, J. Esteve-Romero, and S. Carda-Broch, 1. N. Islam, I. Khalil, A. Islam, and S. H. Gan, J. Appl. J. Chromatogr. A, 2012, 1232, 101.