Total Synchronous Fluorescence Spectroscopy Combined with

E-mail: [email protected]. Phone: (+39) 0223996010. ... Please contact your librarian to recommend that your institution subscribe to this publica...
0 downloads 0 Views 388KB Size
Download PDF
Anal. Chem. 2009, 81, 1784–1791

Total Synchronous Fluorescence Spectroscopy Combined with Multivariate Analysis: Method for the Classification of Selected Resins, Oils, and Protein-Based Media Used in Paintings Austin Nevin,*,† Daniela Comelli,‡ Gianluca Valentini,‡ and Rinaldo Cubeddu‡ Dipartimento di Fisica, Politecnico di Milano, Piazza Leonardo da Vinci 32, Milano, 20133, Italy, and IFN-CNR, Piazza di Leonardo da Vinci 32, Milano, 20133, Italy Recent interest in the fluorescence of binding media and varnishes (proteins, oils, and resins) commonly used in paintings is based on the potential for discriminating these organic materials. A useful way of studying the presence of the broad-band fluorescence emissions found in these complex organic materials is fluorescence excitation emission spectroscopy. However, due to the presence of Raman and Rayleigh scattering which may necessitate correction or preprocessing for statistical analysis and visualization, an alternative approach has been adopted for the analysis of different samples of artist materials based on total synchronous fluorescence spectroscopy. Films of selected drying oils, glue, egg, and casein and the resins mastic, dammar, copal, and shellac were analyzed using total synchronous fluorescence spectroscopy, and an interpretation of the differences between spectra is given. A data reduction method based on the transformation of fluorescence contours extracted from total synchronous fluorescence from Cartesian to polar coordinates is presented and is followed by the comparison of data using multivariate analysis and hierarchical cluster analysis. Results suggest that the new method can be used to classify samples on the basis of their fluorescence spectra, clearly differentiating oils, resins, and protein-based media into groups. The analysis of paintings is particularly challenging due to the vast range of inorganic and organic materials which often constitute a work of art: canvas fibers, wooden supports and inorganic fillers, mixtures of binding media and pigments, metal decorations, varnishes, modern polymer treatments, and protective layers. Protein-based adhesives and natural resins are often used in surface treatments and conservation interventions in the treatment of paintings. In addition to the inherent heterogeneity and complexity of multilayered paintings, the constraints of sampling, which is rarely possible, and the degradation of original materials are two significant complications for the determination of the composition and structure of a painting. The identification * To whom correspondence should be addressed. E-mail: austinnevin@ gmail.com. Phone: (+39) 0223996010. † Politecnico di Milano. ‡ IFN-CNR.

1784

Analytical Chemistry, Vol. 81, No. 5, March 1, 2009

of the varnishes, pigments, and binding media employed may be especially important for decision making regarding conservation and restoration. For example, knowledge of original materials will influence the choice of cleaning of paintings, their treatment and storage, and may also assist in the attribution of a painting to a particular workshop or artist. Although many techniques are available for the analysis of organic painting materials, the most reliable and quantitative analysis is provided by destructive analytical techniques,1 which include gas chromatography (GC),2,3 matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS),4-6 and direct temperature-resolved mass spectrometry (DTMS).7 Indeed, recently a combined GC/MS approach has been proposed for the determination of lipids, waxes, proteins, and resins from the same painting sample, which is particularly powerful in the analysis of organic materials and their degradation products.8 Protein-based binding media (which range from animal glues to eggs and milk) can be determined by analysis of different amino acids present in varying concentrations. Drying oils (including linseed, walnut, and poppy seed oil) are generally distinguished on the basis of the presence of different fatty acids (for example, azelaic, palmitic, and stearic acids as well as other markers).9 Resins, often used as varnishes when dispersed in a solvent, may also be identified by specific markers (for example, in triterpenoids, moronic acid, and nor-olean-17-en-3-one, present only in mastic, or the ursane skeleton, and nor-R-amyrone found only in dammar10). (1) Dome´nech-Carbo´, M. T. Anal. Chim. Acta 2008, 621 (2), 109–139. (2) Casoli, A.; Musini, P. C.; Palla, G. J. Chromatogr., A 1996, 731, 237–246. (3) van den Berg, J. D.; van den Berg, K. J.; Boon, J. J. J. Chromatogr., A 2002, 950, 195–211. (4) Kuckova, S.; Nemec, I.; Hynek, R.; Hradilova, J.; Grygar, T. Anal. Bioanal. Chem. 2005, 382, 275–282. (5) van den Brink, O. F.; Boon, J. J.; O’Connor, P. B.; Duursma, M. C.; Heeren, R. M. J. Mass Spectrom. 2001, 36, 479–492. (6) Dietemann, P.; Kalin, M.; Zumbuhl, S.; Knochenmuss, R.; Wulfert, S.; Zenobi, R. Anal. Chem. 2001, 73, 2087–2096. (7) Scalarone, D.; Lazzari, M.; Chiantore, O. J. Anal. Appl. Pyrolysis 2003, 6869, 115–136. (8) Andreotti, A.; Bonaduce, I.; Colombini, M. P.; Gautier, G.; Modugno, F.; Ribechini, E. Anal. Chem. 2006, 78, 4490–4500. (9) Mills, J. S.; White, R. Organic Chemistry of Museum Objects; ButterworthHeinemann: Oxford, U.K., 1999. (10) Colombini, M. P.; Modugno, F.; Giannarelli, S.; Fuoco, R.; Matteini, M. Microchem. J. 2000, 67, 385–396. 10.1021/ac8019152 CCC: $40.75  2009 American Chemical Society Published on Web 02/04/2009

Nondestructive means are available for the analysis of organic materials found in works of art. Fourier transform infrared (FTIR) spectroscopy is very useful and widely employed for the analysis of inorganic materials and has recently been proposed for in situ analysis.11 Alternatively, micro-Raman spectroscopy has been proposed for a range of different media,12 and multivariate analysis of data may be able to resolve apparently similar spectra.13 Both FT-IR and Raman spectroscopy have been extensively used on samples from paintings for the analysis of inorganic and, to a lesser extent, organic materials. Fluorescence spectroscopy and imaging are other nondestructive methods for the analysis of painting materials and have also been widely employed for the in situ examination of paintings and other works of art, providing specific advantages for the mapping of differences in varnish and retouching and repainting, for example. Complementary laboratory analysis is provided with spectrofluorimetry, which has recently received significant attention for the analysis of colorants14 as well as of protein-based binding media15 and resin-based varnishes.16 Excitation spectra of different media have been used to differentiate between the diterpenoid resin sandarac and triterpenoids mastic and dammar.16 Synchronous fluorescence spectroscopy (SFS) combined with multivariate analysis was proposed for the discrimination between different protein-based binding media (casein, egg white, egg yolk, and various collagen-based glues).15 Other research has ascribed the fluorescence of drying oils employed as binding media17,18 and other commonly employed varnishes.19 Clear analogies can be drawn between painting materials and food products (for example, collagen glues and gelatin, egg white, casein and cheese, drying oils and edible oils). In the analysis of foods, spectrofluorimetry has wide-ranging applications which include the determination of freshness in eggs20 and the identification of contaminants in oils,21 where a variety of fluorimetric methods have been employed, which include fluorescence excitation emission spectroscopy (EES). Recently, EES has been proposed for the analysis of different kinds of food samples, ranging from beer to yoghurt, olive oil, and fish,22 which has been strengthened with statistical treatment of data; because of the broad signals from overlapping fluorophores, the interpretation of excitation emission (EE) spectra can be greatly improved with chemometrics. One method which is particularly powerful is parallel factor analysis (PARAFAC) which has successfully been (11) Miliani, C.; Rosi, F.; Borgia, I.; Benedetti, P.; Brunetti, B. G.; Sgamellotti, A. Appl. Spectrosc. 2007, 61, 293–299. (12) Vandenabeele, P.; Wehling, B.; Moens, L.; Edwards, H.; De Reu, M.; Van Hooydonk, G. Anal. Chim. Acta 2000, 407, 261–274. (13) Nevin, A.; Osticioli, I.; Anglos, D.; Burnstock, A.; Cather, S.; Castellucci, E. Anal. Chem. 2007, 79, 6143–6151. (14) Clementi, C.; Doherty, B.; Gentili, P. L.; Miliani, C.; Romani, A.; Brunetti, B. G.; Sgamellotti, A. Appl. Phys. A: Mater. Sci. Process. 2008, 92, 25–33. (15) Nevin, A.; Cather, S.; Burnstock, A.; Anglos, D. Appl. Spectrosc. 2008, 62, 481–489. (16) Thoury, M.; Elias, M.; Frigerio, J. M.; Barthou, C. Appl. Spectrosc. 2007, 61, 1275–1282. (17) de la Rie, E. R. Stud. Conserv. 1982, 27, 102–108. (18) Miyoshi, T. Jpn. J. Appl. Phys. 1990, 29, 1727–1728. (19) Miyoshi, T. Jpn. J. Appl. Phys 1987, 26, 780–781. (20) Karoui, R.; Schoonheydt, R.; Decuypere, E.; Nicolaı¨, B.; De Baerdemaeker, J. Anal. Chim. Acta 2007, 582, 83–91. (21) Poulli, K. I.; Mousdis, G. A.; Georgiou, C. A. Anal. Chim. Acta 2005, 542, 151–156. (22) Christensen, J.; Nørgaard, L.; Bro, R.; Engelsen, S. B. Chem. Rev. 2006, 106, 1979–1994.

used for the assignment of concentrations of different known fluorophores in foods23 and marine water,24 and various methods for the deconvolution of three-dimensional spectra data have been proposed. EES has also been employed for the mapping of fluorophores present within the complex mixtures in protein-based binding media used in paint.15 As with samples from food, the interpretation of spectra is not straightforward, first because of the large number of fluorophores which are present in the impure mixtures used in binding media25,26 and second due to relatively limited knowledge regarding the fluorophores in solid and aged artists’ materials. Emissions from amino acids in protein-based media are known, Maillard reaction products and oxidation products which may develop with aging are numerous; therefore, rather than the identification of specific fluorophores, EE spectra can instead be used for the rapid differentiation between proteinaceous media. For oils and resins, the isolation of individual fluorophores has been limited, with most interest focused on the analysis of solutions of the former, rather than aged and solid samples.21,27 A similar method, analogous to EES for the mapping of fluorescencestotal synchronous fluorescence spectroscopy (TSFS)shas been proposed for the analysis of petroleum and natural oils.28 In TSFS, rather than scanning a range of excitations and the construction of a matrix of fluorescence emissions as in EES, a range of offsets between excitation and emission are scanned over a fixed range of excitation, and the emission at each offset is plotted. Although total synchronous fluorescence (TSF) spectra do not yield better resolved spectra than analogous EE spectra, no corrections or preprocessing for the elimination of Rayleigh or Raman scattering are required, as may be necessary for EE spectra.23 In this work, TSFS of a range of different artificially aged films of selected organic painting materials has been undertaken. The materials studied can be used both as binding media for pigments (proteins and oils) but may also be used as adhesives or surface treatments (varnishes).29 Materials were selected on the basis of their chemical composition, as well as their use and availability, and are not an exhaustive selection of resins, oils, or proteins but rather represent general classes of materials used as adhesives, and binding media. Materials studied include selected drying oils (walnut, poppy seed, uncooked linseed, and cooked linseed oils), resin-based paint varnishes (the triterpenoids dammar (from the Dipterocarpaceae family of trees) and mastic (from Pistacia lentiscus), a diterpenoid varnish prepared with semifossilized Manila copal, and shellac (in insect secretion)), and protein-based binding media (egg white, egg yolk, casein, isinglass (glue extracted from sturgeon bladders)). Artificial and natural aging may lead to the oxidation of many of the molecules found in binding media, which include specific amino acids,8 terpenoids,3 and fatty acids.5 It has been shown that light exposure also affects Andersen, C. M.; Bro, R. J. Chemom. 2003, 17, 200–215. Booksh, K. S.; Muroski, A. R.; Myrick, M. L. Analyst 1993, 118, 917–921. Deyl, Z.; Miksik, I.; Zicha, J. J. Chromatogr., A 1999, 836, 161–171. Nevin, A.; Cather, S.; Anglos, D.; Fotakis, C. Anal. Chim. Acta 2006, 573574, 341–346. (27) Sikorska, E.; Romaniuk, A.; Khmelinskii, I. V.; Herance, R.; Bourdelande, J. L.; Sikorski, M.; Koziol, J. J. Fluoresc. 2004, 14, 25–35. (28) Patra, D.; Mishra, A. K. Trends Anal. Chem. 2002, 21, 787–798. (29) Merrifield, M. P. Original Treatises on the Arts of Painting; Dover Publications: London, 1967. (23) (24) (25) (26)

Analytical Chemistry, Vol. 81, No. 5, March 1, 2009

1785

the fluorescence of binding media, likely after the oxidation of specific molecules and a change in the concentration of fluorophores, relevant for the spectrofluorimentric analysis of oils,17 proteins,26 and resins.16 Therefore, light aging of samples was selected to simulate cumulative museum exposure. For the discrimination between selected media on the basis of TSFS, an alternative and novel approach is proposed; instead of the decomposition of a matrix of data for PARAFAC,22,23 statistical analysis based on principal component analysis and hierarchical cluster analysis of polar coordinate plots of contours describing fluorescence extracted from TSF spectra is proposed. EXPERIMENTAL SECTION Samples. Protein-based binding media were prepared following recipes from treatises, as described elsewhere,15 and samples of films of egg white, egg yolk, casein (milk), and isinglass (collagen from sturgeon bladder) were analyzed. Drying oils without the addition of a siccative (cold-pressed and cooked linseed oil, cold-pressed walnut oil, and poppy seed oil) and selected varnishes (dammar (35%) and mastic (30%) dissolved in turpentine, Manila copal (30%) cooked in linseed oil, and shellac (10%) dissolved in ethanol) were purchased from Zecchi (Florence, Italy) and were used as supplied. Five samples of each material were prepared by evaporation (or drying for oils) on 2 cm fusedsilica disks to yield samples of between 10 and 20 µm in thickness. In order to simulate the cumulative effects of light exposure and natural aging, samples of materials were artificially aged using a custom-built aging chamber. Samples were irradiated with daylight fluorescent lamps (T ) 6500 K) at 40 °C and ambient RH for a total of 45 × 106 lx · h (lux · hour), approximately equivalent to 100 years of indoor lighting conditions.30 Samples were analyzed 6 months after their preparation. Instrumentation. Synchronous fluorescence (SF) spectra were collected using a Jobin Yvon Horiba Fluorolog spectrometer equipped with a Xenon arc lamp in a front face geometry (23° between excitation and emission monochromators.) Slits were fixed at 1 nm for the excitation and 5 nm for the emission monochromator; integration time was 0.1 s. TSF spectra were acquired with excitation between 300 and 500 nm at a resolution of 2 nm, with an offset between 10 and 150 nm (in steps of 5 nm) between excitation and emission monochromators, and spectra were corrected for detector efficiency and lamp emission; this range of excitation was chosen as all materials studied present at least one broad fluorescence emission. For each material, TSF spectra were collected on different days and from areas approximately 2 mm × 2 mm on the surfaces of different samples. Data Analysis. TSF spectra were recorded from five different samples of each material. Following acquisition, spectra were assembled into a matrix of SF spectra (excitation wavelength vs offset between excitation and emission). Spectra were then transformed into a matrix equivalent to a fluorescence EE spectrum using MatLAB 7.5 (Mathworks), and new matrices have been normalized by the maximum emission. For easy comparison in figures, matrices are represented with 20 different contours ranging from 10% of the maximum (in blue) to the maximum value (shown in red) given for each plot, and in the text maxima in (30) Feller, R. Accelerated Ageing: Photochemical and Thermal Aspects; Getty Conservation Institute: Los Angeles, CA, 1994.

1786

Analytical Chemistry, Vol. 81, No. 5, March 1, 2009

Figure 1. EE spectra of drying oils: (a) cooked linseed oil and coldpressed (b) linseed, (c) poppy seed, and (d) walnut oil.

fluorescence bands are described by the excitation and emission (ex/em) wavelengths. In our approach, we have chosen to reduce EE data to contours in order to reduce the number of variables and consequently the number of required observations for statistical analysis. Specifically, for data reduction, contours in normalized EE spectra were projected in polar coordinates with dimensions (F, θ) around a common origin, allowing a consideration of the shape of the contours at different values of fluorescence intensity. Polar coordinates where determined with a spacing of 5° between 0° and 360° (72 variables). To the authors’ knowledge this approach has not been applied for the treatment of EE of TSF spectra, although polar coordinates have been suggested for data reduction of other types of 2D spectra.31 Multivariate analysis of contours in polar coordinates was performed using principal component analysis (PCA) with the Matlab Statistics toolbox. Thereafter,

Table 1. Fluorescence Properties of Media (Average of Five Different Samples) medium

band max intensity/ 107 counts

excitation max/nm

linseed oil cooked linseed oil poppy oil walnut oil

17.9/18.3 18.7/18.9 21.3/19.5 19.7/19.5

Drying Oils 338/369 340/366 332/368 334/368

dammar mastic copal shellac

15.1 16.7 10.3/10.9 6.2

Resin-Based Varnishes 362 368 323/410 312

casein egg white egg yolk isinglass

8.9 14.5 16.3 8.4

Protein-Based Binding Media 325 330 328 320

hierarchical cluster analysis of the first two principal components based on their weighted Euclidian distance was carried out, and a dendrogram has been used to represent the spectral similarity between the fluorescence of different media. The data sets used for statistical analysis are shown as polar plots for each material in the Supporting Information. RESULTS AND DISCUSSION Transformed TSF Spectra of Artificially Aged Media. Drying Oils. From an inspection of the EE spectra recorded from different drying oils (Figure 1), it is clear that the samples share common features: all are characterized by a broad-band emission with full width at half-maximum (fwhm) of approximately 130 nm and two local maxima; one with a longer Stokes shift at ex/em of approximately 330-340/400 nm and the other with a shorter Stokes shift with ex/em maximum at approximately 370/425 nm. Small differences in the relative intensity of the two peaks and in their position may be characteristic of different oils and suggest greater similarity between poppy and walnut oil compared with the two differently prepared linseed oils. In contrast to the fluorescence of fresh oils which is ascribed to tocopherols, chlorophyll, and phenolic antioxidants, as has been suggested for various oils,21,32 the broad fluorescence which develops in solid drying oils likely follows the polymerization of fatty acids.17 The small difference between EE spectra of the cold-pressed and cooked linseed oils, which is accompanied by a slight hypsochromic shift in the broad emission, is ascribed to differences in oxidation products which result from the high-temperature treatment of cooked linseed oil compared to the less oxidized uncooked linseed oil.33 Indeed, the ratio of azelaic/palmitic acids measured in GC/MS is one means for the differentiation between cooked and uncooked oils,9 and although these specific acids are not responsible for fluorescence, fluorophores are likely to be oxidized during cooking and others may also develop with cooking which slightly shifts the fluorescence emission and excitation maxima, as shown in Table 1. (31) Bouveresse, D. J. R.; Malabat, C.; Rutledge, D. N. Trends Anal. Chem. 2005, 24, 839–842. (32) Sikorska, E.; Gorecki, T.; Khmelinskii, I. V.; Sikorski, M.; Koziol, J. Food Chem. 2005, 89, 217–225. (33) van den Berg, J. D.; Vermist, N. D.; Carlyle, L.; Holcapek, M.; Boon, J. J. J. Sep. Sci. 2004, 27, 181–199.

emission max/nm

Stokes shift/nm

406/427 400/420 398/422 398/422

68/58 60/54 66/54 64/54

425 423 440/454 390

63 55 117/44 78

398 400 396 390

79 70 68 70

Only subtle differences are found between the EE spectra of cold-pressed oils from walnut and poppy seeds, whereas linseed oil is characterized by longer Stokes shifts, especially at lower excitation wavelengths. Indeed, rather than the position of the maximum fluorescence emission, variation between samples is observed in the shape of the contour plots. Varnishes: Mastic, Dammar, Copal, and Shellac. In contrast to the drying oils, triterpenoid resins mastic and dammar are found with a narrower fluorescence band centered at approximately 365/ 425 nm, and significantly different EE spectra are associated with the different varnishes (Figure 2). The fluorescence emission in dammar and mastic (which are most similar to the oils in terms of the spectral shape) is less broad than that found in oils, with a fwhm of approximately 100 nm, and differences in the shape of the emission are appreciable from individual emission spectra, SF spectra, and subtle differences between EE spectra. Although differences between many constituent molecules isolated in aged mastic and dammar have been documented following high-performance liquid chromatography MS (HPLC/MS),34 GC/MS,10 and graphite-assisted laser desportion ionization (GALDI) MS,6 the differences in the fluorescence of mastic and dammar have not yet been ascribed to specific molecules. The fluorescence of diterpenoid Manila copal (cooked in oil), a solid semifossilized resin which hardens following cross-linking of polycommunic acids, is distinctive and dissimilar to that observed in triterpenoids, characterized by large Stokes shifts, as can be appreciated from the TSF spectrum. Both fluorescence emission spectra and SF spectra of copal can easily be used to distinguish copal from the other varnishes. The fluorescence from copal is ascribed to the presence of multiple aromatic fluorophores with different chemical composition which may arise during the degradation reactions after the formation of the semifossilized resin, and other contributions from the oil fraction are also likely; however, the broad peaks observed in the spectrum of copal do not correspond to those observed in linseed oil (cooked or coldpressed) but are found at longer emission wavelengths, suggesting that different fluorophores account for the many emissions in the resin, thus highlighting limitations of fluorescence spectroscopy (34) van der Doelen, G. A.; van den Berg, K. J.; Boon, J. J.; Shibayama, N.; Rene de la Rie, E.; Genuit, W. J. L. J. Chromatogr., A 1998, 809, 21–37.

Analytical Chemistry, Vol. 81, No. 5, March 1, 2009

1787

Figure 2. EE spectra of resin varnishes: (a) copal, (b) dammar, (c) mastic, and (d) shellac.

Figure 3. EE spectra of protein-based binding media: (a) casein, (b) egg white, (c) egg yolk, and (d) isinglass.

in the detection of mixtures. Among compounds identified using DT-MS and GC/MS,7 derivatives of agathic, communic, and sadaracopimaric acids may also fluoresce and could give rise to the peaks in the EE spectra which have two characteristic bands, found at ex/em of approximately 325/440 nm and another very broad emission centered at 410/455 nm. Shellac has an emission detected with a maximum around 310/ 390 nm, distinctly different from the other varnishes considered. Although this fluorophores has not been attributed, it is likely that the visible but weaker fluorescence in shellac is from laccaic acids, responsible for the red color of lac dyes, which have been identified using laser-induced fluorescence of analogous samples.18 Protein-Based Binding Media. Films of casein and egg white have EE spectra (Figure 3) of similar shape and position, reflecting the similarity in their chemical composition,35 both sharing a maximum excitation/emission at approximately 325-330/390-400

nm, ascribed in this case to the fluorescence from photooxidation of tryptophan and the accumulation of Maillard reaction products. Indeed, significant differences are found between collagen-based glues including isinglass which contains no tryptophan and the other protein-based binding media, but the differences in fluorescence of protein-based media are most apparent at excitation below 300 nm, not considered in this work. A discussion of the use of SF spectroscopy for the classification of protein-based media can be found elsewhere.15 In Table 1, the maximum excitation and emission of the materials considered are given, and despite subtle differences in the shape of the fluorescence emissions described in the EE spectra in Figures 1-3, the data in the table highlight one of the complications in the use of fluorescence spectroscopy of such

1788

Analytical Chemistry, Vol. 81, No. 5, March 1, 2009

(35) Colombini, M. P.; Modugno, F. J. Sep. Sci. 2004, 27, 147–160.

Figure 4. (a) Polar coordinate plots of the contour describing fluorescence (with origin at ex/em 330/400 nm) of dammar at various intensity levels with respect to the maximum fluorescence and (b) polar plots of other binding media representing the contour at 75% intensity (with origin at ex/em 330/400 nm).

Figure 5. (a) Loading plot of the first two principal components and (b) score plot following principal component analysis of polar plots of TSF spectra of media at 75% of the maximum intensity; five contour plots were considered for each material and are represented with different symbols on the score plot.

complex samples; even the small set of samples considered as part of this work cannot be differentiated with confidence on the basis of fluorescence intensity or emission or excitation maximum. Therefore, multivariate statistical analysis was explored as rapid alternative for differentiation between spectra. Statistical Analysis of Data. In order to better compare data from different media, rather than the position of the ex/em maxima, or specific SF spectra, which are often quite similar due to similar Stokes shifts in the media considered (as seen in Table 1), a new method for the evaluation of the fluorescence of materials has been used and is based on a consideration of the shape of the fluorescence profile or contour at different relative intensity levels from the maximum. Following normalization (based on the maximum fluorescence intensity), contours describing the shape of the fluorescence curve at any offset from the maximum or at a percentage of the maximum fluorescence can be determined. Contours were then converted from Cartesian to polar coordinates. When considering a large set of different contours, a common origin for the polar coordinate plot is chosen, thereby allowing the consideration of all spectra using the same coordinate system. Specifically in the following example, for the conversion to polar coordinates, a group of contours extracted from EE spectra were transformed with the average maximum excitation and emission (330/400 nm) as the origin for each plot.

This method has a variety of advantages, which include the possibility of comparing data from different samples which are associated with different excitation and emission maxima and differently shaped contour plots. As an example, in dammar, polar coordinate plots are used to represent the fluorescence contours at 90%, 70%, and 50% of the maximum intensity, as shown in Figure 4a. Clearly, closest to the maximum the contour describes a smaller area, but the distribution of the different contours around the origin is similar, reflecting, in this case, the uniformly oblong shape of the broad fluorescence emission seen in Figure 2b. When considering spectra from different materials at the same relative intensity level (70%, for example), clear variation is observed, as exemplified in dammar, isinglass, and linseed oil (Figure 4b), due to differences in the distribution of fluorescence about different emission and excitation maxima for the media shown. Spectra of the other organic materials at 75% are included in the Supporting Information. Because of the large variation in fluorescence spectra of the materials considered, and hence the large number of variables in TSF spectra, some of which do not have single bands of fluorophores and are recognizably different from the other spectra, an initial mathematical filtering of data whereby copal was excluded from the data set was used. This is based on two criteria. First, copal has EE spectra with more than one distinct fluoresAnalytical Chemistry, Vol. 81, No. 5, March 1, 2009

1789

Figure 6. Dendrogram representation of results of hierarchical cluster analysis of polar graphs of fluorescence spectra of 11 different media (excluding copal) following filtering of data. Clusters of four different colors are formed beneath a threshold distance of 100 based on the weighted Euclidean distance between spectra.

cence contour at 75% of the maximum and cannot thus be easily compared with the contours from other materials. Second, in copal the contour centered at 330/400 nm does not share a common point with the contours from the other more similar materials. A requirement of multivariate analysis in the consideration of contours in this work is that polar spectra have positive F; in other words, contours should describe an area containing the common origin. Due to the range of excitation emission maxima and the distribution of fluorescence around the maximum, this criterion is met by selecting a suitable origin for the spectra from all the materials, which may be impossible if contours are not broad enough or at a significantly different ex/em position to contain a common maximum. Principal component analysis (covariance) was then carried out on a matrix of fluorescence contours (five for each sample, provided in Supporting Information) at 75% of the maximum with fixed θ from 0° to 360° with steps of 5° and variable F following the application of the filtering described above, and results are shown in Figure 5. The loading plot of the first two principal components which account for 95% of the variance within the spectra considered indicates that polar plots of contours are separated well on the basis of the distribution of the fluorescence around the origin. Spectra with a greater F at 60° relative to that at 240° are found with negative PC1 (oils, dammar, and mastic), whereas spectra from shellac (which are centered at 312/390 nm (Figure 4b)) and protein-based binders are found with positive PC1. As appreciable from the score plot, contours from shellac are found with PC1 greater than 220 and PC2 around -25, distant from the other materials. Isinglass and other protein-based binders are found with PC1 between 0 and 150 and PC2 between +50 and -50. Oil-based media are clustered together with negative PC1 and are separated from dammar and mastic by PC2. Dammar and mastic are both found with PC1 around -100 and PC2 between 50 and 100. Within drying oils, the greatest variation is 1790

Analytical Chemistry, Vol. 81, No. 5, March 1, 2009

observed in walnut oil, but no significant differences are observed between cooked and cold-pressed linseed oil. Within the dairybased proteins, casein and egg white cluster together and can be separated from egg yolk, which has greater PC1. In order to quantitatively describe the distribution of data, hierarchical cluster analysis was applied to the first two PCs, and a dendrogram representing weighted average Euclidean distances indicates the similarities between binary clusters or pairs of contour plots. In the dendrogram, data from more than one sample may be represented in each line. A threshold Euclidean distance greater than 0.5 for different branches yields different colored groups of nodes. As shown in Figure 6, data is classified in clusters which reflect the origin and chemical composition of the studied materials, with shellac the most distant from the other materials. With a distance of less than 100 (arbitrary units), four different groups are found in separate nodes: (1) shellac, (2) casein, egg white, egg yolk, and isinglass, (3) drying oils, and (4) mastic and dammar. The protein-based binding media are represented in different clusters, and spectra from egg white are clustered with egg yolk and casein, as can be understood from the score plot. Dammar and mastic are found within a homogeneous cluster and are more similar to the cluster formed with drying oils than to the protein-based media. Drying oils are found within a separate cluster, in which poppy seed is grouped with walnut oil but never with linseed oils and similarities between the various drying oils compromise their differentiation. Likewise, within the proteinbased binding media, although casein and egg yolk are always found in different clusters, spectra from egg white are found distributed between the clusters formed by egg yolk and casein. In the example above, contours at 75% of the maximum have been demonstrated, and the method was successfully applied to other contours from the same EE spectra at other levels (below 78% where contours do not share a common origin), which yielded a similar clustering of PCs following multivariate analysis of data. CONCLUSIONS In this work multivariate analysis of polar plots of contours of fluorescence spectra demonstrates the potential of the method for the specific problem of the classification of artist materials used as binders and varnishes (as well as in conservation treatments) for paintings. With the advantage of classifying media into different clusters based on data reduction and a consideration of the contours (which describe spectral shape) of EE spectra, this method could also be used for the analysis of other fluorescent artist materials (colorants and dyes, for example). Further statistical methods including linear discriminant analysis (LDA)27 and factor analysis (including PARAFAC)22 could also be considered for the analysis of contours and TSF spectra, as has, for example, been successfully demonstrated for EE spectra of adulterations in olive oil.36 As shown, TSFS can allow the differentiation of media found in paintings, but similarities in spectra between oils, resins, and protein-based binders may be such that additional data analysis is required. With a larger database of spectra from a range of different naturally and artificially aged binding media of different origins it is possible that this method could be extended to the analysis of real samples. Analysis of other proteins could include (36) Guimet, F.; Ferre`, J.; Bouque´, R. Anal. Chim. Acta 2005, 544, 143–152.

ox and rabbit glue, and the expansion of the range of excitation and SF offsets would allow the consideration of other fluorophores. Other varnishes should be considered and could include copals of different origin (including samples from South America and Africa) as well as other diterpenoid resins (including colophony, sandarac, and Venice terpentine, for example) and shellac from a variety of sources. Indeed, this method could usefully be employed for an assessment of differences between similar materials. Additionally, mixtures of oils and varnishes should be investigated but may be particularly complicated by contributions from different mixtures of fluorophores from both media. As pigments have a significant effect on the chemistry and observation of the fluorescence of media, as has been shown for protein-based binding media,37 the application of the proposed method could assist in the assessment changes in fluorescence as a function of pigment concentration. Further, various mathematical models have been suggested to correct fluorescence spectra for the absorption of pigments which could also be extended to the consideration of TSF and EE spectra.14 (37) Nevin, A.; Anglos, D.; Cather, S.; Burnstock, A. Appl. Phys. A: Mater. Sci. Process. 2008, 92, 69–76.

Following the careful selection of suitable experimental conditions a fiber-optic coupled to a spectrofluorimeter could be used for the collection of spectra from painted surfaces, and provided access to a significant database of reference materials, the method proposed could be used to classify fluorescence. Interferences from pigments and colorants, as well as fluorescent conservation materials, are expected to be problematic. Therefore, future work should focus on the analysis of pigmented materials and a larger set of laboratory samples. A similar method could also be employed for the analysis of food, natural oils, and biological materials, where florescence EES is commonly employed and where subtle changes in fluorescence may be related to food freshness, material contamination, or pathology. SUPPORTING INFORMATION AVAILABLE Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review December 27, 2008.

September

9,

2008.

Accepted

AC8019152

Analytical Chemistry, Vol. 81, No. 5, March 1, 2009

1791