A Comparison of the Fading of Dyestuffs as Textile Colourants and Lake Pigments

Abstract

Dyed wool samples and lake pigments prepared from the same dyestuffs were exposed to light over the course of 14 months. Brazilwood or sappanwood, cochineal, madder, and weld were used for both wools and pigments, with the addition of dyer’s broom, indigo, and tannin-containing black dyes for the wools and eosin for the pigments. The wools were dyed within the MODHT European project on historic tapestries (2002–2005), using recipes derived from fifteenth- to seventeenth-century sources. The pigments were prepared according to European recipes of the same period, or using late nineteenth-century French or English recipes. Colour measurements made throughout the experiment allowed for overall colour difference (ΔE₀₀) to be tracked and half-lives to be calculated for some of the colour changes. Alterations in the samples’ hue and chroma were also monitored, and spectral information was collected. The results showed that, for both textiles and pigments, madder is the most stable red dye, followed by cochineal, and then brazilwood. Eosin was the most fugitive sample examined. Comparisons of textile and lake samples derived from the same dyestuff, whether red or yellow, indicate that the colourants are more stable when used as textile dyes than in analogous lake pigments.

1. Introduction

Previous studies have examined the behaviour of traditional red and yellow lake pigments on exposure to light. These pigments, prepared from natural dyestuffs by precipitation of the colouring matter onto a usually translucent substrate, were used in western European easel paintings until well into the nineteenth century, both to produce red and yellow passages of colour [1] and in mixtures, particularly with blue to produce purple or green shades [2]. Many of the dyestuffs that were used to make lake pigments were also employed in the dyeing of textiles; indeed, the latter industry was, historically, the principal user of the natural dyestuffs that were the subject of these earlier studies.

Textiles dyed with natural dyes had already been the subject of much of the work carried out on fading and colour change [3,4,5,6]; equivalent pigments were less well studied, apart from the work on alizarin crimson by Ruth Johnston-Feller and co-authors [7,8,9]. These studies were often carried out within a museum or conservation context, fading the colourants under daylight, artificial daylight, or other light sources, to provide information on lighting conditions appropriate for the textiles or painted artefacts that were displayed. By the early 2000s, the light sources most frequently found in museums and galleries were tungsten-halogen and fluorescent lamps, together with daylight itself. To assess the effect of these different white light sources on works of art that are vulnerable to damage by light, an experiment was conducted at the National Gallery, London, using sets of lake pigment and textile samples prepared using similar natural dyes, thus representing a wider range of light-sensitive artefacts [10]. As textile samples, the authors were given a set of dyed wools prepared as part of the MODHT (Monitoring of Damage in Historic Textiles) project, which was supported under the EU 5th Framework programme (contract no. EVK4-2001-00020) and ended in June 2005. For that project, a range of traditional colourants were dyed onto wool and silk substrates to produce red, yellow, blue, green, and black textile samples; only the wool samples were used in the National Gallery experiment. In parallel, many of the same colourants were prepared as lake pigments and painted out in linseed oil (Figure 1). This comparison found that, perhaps expectedly, filtering out ultraviolet radiation made the largest contribution to reducing damage to the colourants but also concluded that “providing the light levels are controlled to a similar level, neither of the two common museum illuminants (tungsten halogen and fluorescent) is more damaging” [10] (p. 766).

While the study published in 2008 focused on a comparison of the behaviour of the same samples when exposed to the different light sources under evaluation, the more than 6000 measurements made over a 14-month exposure also contained the information needed to assess colour change of different samples under the same light exposure. In particular, we wished to examine the relative stability of the various colourants, both when dyed on wool and prepared as lake pigments, and to investigate whether the same colourants were less or more stable when dyed on wool than when present as lake pigments. It is the outcome of these comparisons that is the focus of this paper.

2. Materials and Methods

2.1. Light Exposure

One of the light boxes used in the study of white light sources contained so-called ‘daylight’ fluorescent lamps (Figure 2). The output from these lamps contains rather more ultraviolet radiation than some of the other lamps used in the experiment certainly more than would be normal in a museum context where ultraviolet is generally excluded or greatly reduced. However, since the pattern of colour change under this illuminant was not significantly dissimilar from that in the other light boxes, differing mainly in the rate at which changes occurred, the data from this light box—designated ‘1’ in the 2008 publication—were used in the current study as the alterations to colours could be seen more rapidly.

The light box was equipped with 12 General Electric F65W AD (artificial daylight) fluorescent lamps that produced a light level on the sample surface of ca. 20,000 lux. The ultraviolet content of the lamps was about 310 μW per lumen. Over the course of the experiment, the light level in the box changed as lamps aged, failed, and were replaced. Periodic light level measurements allowed these fluctuations to be factored into the calculation of light exposure across the course of the experiment, which is expressed in megalux hours (Mlux·h). The cumulative exposure across the whole experiment was around 210 Mlux·h. This is equivalent to about 300 years of exposure in a museum at a standard illumination of approximately 200 lux for 8 hours a day, or 3000 h per year, with the caveat that, under those circumstances, ultraviolet radiation would normally have been filtered out.

2.2. Samples

2.2.1. Paint Samples

The paint samples used in the experiment are listed in

Table 1. List of paint samples—see Figure 1.

Code	Pigments (in Linseed Oil Medium) †
C3A	Cochineal C3 (Sn substrate) + lead white 1:20 w/w
C3B	Cochineal C3 (Sn substrate) + lead white 1:8 w/w
C6	Cochineal C6 (Sn substrate + starch extender)
CD1A	Cochineal CD1 (Al substrate) + lead white 1:40 w/w
CD1B	Cochineal CD1 (Al substrate) + lead white 1:20 w/w
CD1C	Cochineal CD1 (Al substrate) + lead white 1:8 w/w
M7A	Madder lake from plant M7i
MD6A	Madder lake from wool MD6ai
E1	Eosin lake E1
B10A	‘Brazilwood’ BBr 10 * + lead white 1:40 w/w
B10B	‘Brazilwood’ BBr 10 * + lead white 1:10 w/w
B10C	‘Brazilwood’ BBr 10 *
VE	Vermilion
LW	Lead white
ILW	Indigo + lead white
PBC1	Prussian blue SC + lead white 1:400 w/w
PBC2	Prussian blue SC + lead white 1:100 w/w
PBC3	Prussian blue SC + lead white 1:20 w/w
SG	Sap green lake
INY	Indigo + Naples yellow
BLUW	Buckthorn lake BuDS3 + ultramarine + lead white
GLW	Gamboge + lead white 1:3 v/v
W6LW	Weld lake W6 (Al substrate) + lead white 1:3 v/v
W7LW	Weld lake W7 (Al/Ca substrate) + lead white 1:3 v/v

† Sn substrate: tin(IV) oxide, SnO₂; Al substrate: amorphous hydrated alumina; Al/Ca substrate: calcium carbonate + sulfate + alumina. * This pigment was prepared from sappanwood, not brazilwood.

. Those that do not contain lakes or are not discussed further in this paper are indicated in italics. The list is the same as that in the 2008 paper [10], which also provides a brief summary of the recipes used to make the lake pigments, although these have also been cited in other papers, as many of the lakes were already used in earlier work [1,11]. Most of the lake pigments were prepared in the National Gallery laboratory following European recipes largely dating from the fifteenth to the seventeenth centuries, examples of which can be found in Kirby, Van Bommel, and Verhecken [12] (pp. 78–85). In this particular historical context, a lake pigment is one made by precipitating the dye onto a substrate, usually of amorphous hydrated alumina but also, in some cases, a calcium salt, largely calcium carbonate or tin(IV) oxide (stannic oxide). The cochineal lakes on tin(IV) oxide substrates and the eosin lake used in this experiment were prepared during a workshop on nineteenth-century red lake pigments held at the Instituut Collectie Nederland (ICN) laboratory (now the Cultural Heritage Laboratory of the Rijksdienst voor het Cultureel Erfgoed (RCE), Cultural Heritage Agency of the Netherlands), Amsterdam in April 2003 [13].

The colourants that were examined in detail in this experiment were cochineal (Dactylopius coccus Costa, 1829), sappanwood (Biancaea sappan (L.) Tod.)—which behaves in the same way as brazilwood (until recently both were classified in the genus Caesalpinia) and contains the same principal dye constituents—madder (Rubia tinctorum L.), weld (Reseda luteola L.), and also eosin, the only synthetic dye. The sample of eosin was manufactured by Meister, Lucius & Brüning, Höchst am Main, Germany, and supplied by the ICN laboratory. Some of the other organic colourants (indigo, buckthorn, gamboge, and sap green) and the inorganic pigments (vermilion, ultramarine, Prussian blue, and lead white) have not been considered in detail here except for the effect of adding increasing amounts of lead white to some of the other pigments, discussed below. These pigments are italicised in Table 1 above.

Lakes containing lac and kermes dyes were not studied during this experiment although they were important pigments; their use continued even after the introduction of the cochineal insect and its dye from New Spain during the sixteenth century [14,15,16]. These dyes were not included in the samples prepared for the MODHT project, partly because it is difficult to source kermes today and lac seems not to have been widely used in textile dyeing in Western Europe; therefore no comparison would have been possible. The lake pigments prepared from these dyes were, however, studied in 1994, when it was found that kermes lakes were more sensitive to light than madder lakes, but slightly more stable than lac and cochineal lakes. Cochineal and lac lakes were found to show very similar behaviours [1] (pp. 86–87).

For the three red lake pigments that were examined as mixtures with lead white, the letter ‘A’ is used to denote the tone with the highest proportion of lead white. The samples labelled ‘B’ have a lower lead white content and—where applicable—those designated ‘C’ contain even less, or no, lead white.

The two madder samples differ in that for M7A the dye was extracted directly from madder root using a late nineteenth- or early twentieth-century recipe, while for MD6A, the source of the dye was a gentle alkali extraction from dyed wool, a very typical fifteenth- or sixteenth-century method, although it should be noted that the latter type of recipe can result in the presence of very variable proportions of individual dye constituents depending on the proportions of constituents in the original textile and the strength of alkali used [17].

To prepare the paint outs, the pigments or pigment mixtures were ground in cold-pressed linseed oil and painted onto 3 mm thick Teflon plates, the surface of which had been roughened to aid wetting and adhesion.

2.2.2. Textile Samples

The textile samples used in the experiment are listed in

Table 2. List of dyed wool samples—see Figure 1.

Code	Textiles (MODHT)
RW1A	Madder (mordanted + semelwater * and alum; dyed + semelwater)
RW1B	Madder (mordanted + semelwater and alum; dyed + semelwater; treated + K2CO3)
RW2A	Madder (mordanted + oak galls then + alum; dyed)
RW2B	Madder (mordanted + oak galls then + alum; dyed; treated + K2CO3)
RW3A	Brazilwood (mordanted + alum; dyed)
RW3B	Brazilwood (mordanted + alum; dyed; treated + K2CO3)
RW4	Cochineal (mordanted + alum, D-tartaric acid, salt and sandalwood; dyed + starch, salt and turmeric)
RW5	Cochineal (mordanted + alum, D-tartaric acid, salt and sandalwood; dyed + gum arabic, alum, salt and turmeric)
BKW1	Black (mordanted + oak galls; dyed + Fe(II) sulfate; all hanks rinsed together)
BKW2	Black (mordanted + oak galls; dyed + Fe(II) sulfate; hanks rinsed separately)
BKW3	Black (mordanted + alder bark; dyed + Fe(II) sulfate)
BKW4	Black (mordanted + oak galls; dyed + Cu(II) + Fe(II) sulfates)
YW1	Weld (mordanted + alum; dyed + K2CO3)
YW2	Dyer’s broom (mordanted + alum; dyed + K2CO3)
GW1	Green (weld as in YW1, overdyed with woad indigo)
GW2	Green (woad indigo, overdyed with weld as in YW1)
IW1	Indigo (all woad indigo dyed by the late John Edmonds)

* Fermented bran solution.

. As with the pigment samples, the list is very similar to that in the 2008 paper [10]. As mentioned in the introduction, the wool textile samples were dyed and woven as part of the MODHT project and were supplied to the National Gallery by the National Museums of Scotland, Edinburgh, and the Royal Institute for Cultural Heritage, Brussels [18]. Samples of each, measuring about 1 cm², were mounted on conservation grade cardboard using double-sided adhesive tape. As part of the MODHT project itself, other sample sets had been submitted to accelerated light ageing in a Xenotest-150S chamber fitted with an Atlas xenon lamp and a window glass filter (UV cut-off 320 nm) [19]. However, the aim of that study was to investigate the mechanical properties of tapestries using strain testing of the aged samples; examination of the fading or other effects on the colour was not carried out [19,20]. Light ageing using Xenotest equipment was carried out on another sample set as part of later research connected with the digital reconstruction of the tapestry, The Oath and Departure of Eliezer, in the Historic Royal Palaces collection, Hampton Court, also carried out at the University of Manchester [21,22].

The dyeing recipes were derived from sixteenth- and seventeenth-century Netherlandish and Italian sources: T’bouck van wondre (Brussels, 1513); archival recipes from the Six and Kerspin families of Amsterdam and Rotterdam, quoted by Willem de Nie in sections XVI–LXI of the Appendix of his book; and Gioanventura Rosetti, Plictho de larte de tentori … (Venice, 1548) [23,24,25]. Three-ply wool was used for both warp and weft. The original recipes were modified in two ways. First, the original volumes and weights were converted to metric measures and scaled as appropriate for the equipment used. Second, certain ingredients were omitted: poisonous compounds, such as arsenic salts; materials which were not deemed essential for the procedure; and ingredients whose purpose was unclear [19] (pp. 35–43). The dyes used were madder, brazilwood (Paubrasilia echinata (Lam.) Gagnon, H.C.Lima & G.P.Lewis), cochineal, weld, dyer’s broom (Genista tinctoria L.), and woad indigo, the last of which was dyed by the late John Edmonds for the project. The black dyes were based on oak galls or alder bark, with the addition of iron(II) and, in one case, copper(II) sulfates.

‘Semelwater’ (from the Dutch zemelen, bran), used in the madder recipes RW1A and B, is a mildly acidic fermented bran solution that was widely used, particularly in madder dyeing, during the period from which the recipes date. It is also known as ‘sour water’ (zuurwater), ‘bran water’, and ‘strong water’ [12] (pp. 40–41); [24] (pp. 144–145, 154–155, Appendix XLVI–XLVIII); [25] (pp. 17–19, 107–109). When used during the extraction of the dye, it promotes some hydrolysis of the anthraquinone sugars, increasing the quantity of anthraquinones available for dyeing and producing a better result.

For samples RW2A and RW2B, the wool was mordanted twice, first with a solution of oak galls and then with the alum solution. The madder and brazilwood samples labelled ‘B’ (RW1B, RW2B and RW3B) had a final treatment with alkali (potassium carbonate, K₂CO₃), which slightly darkens the final colour. The recipes for samples RW4 and RW5 include the addition of turmeric to the cochineal dye bath, which results in a rather less purple red than cochineal alone. Turmeric was often used when dyeing with cochineal and is undoubtedly safer than the yellow arsenic sulfide orpiment, which was also used [24] (p. 157). In both yellow wool samples (YW1 and YW2), the dyestuff was extracted from the plant material in the dye bath using alkali, a very common process that intensifies the colour obtained. The recipe used is close to that in T’Bouck van wondre: ‘Om ghelu met wouwen [weld] te verwen—capit. XII’ (and also the following recipe), and the same procedure was employed for green using a blue-dyed wool [12] (pp. 41–42) [23] (pp. 18–19), as described for the green sample GW2 [19] (p. 39).

2.3. Colour Measurement

To follow the progress of light-induced colour changes to the colourants, the colour of each of the samples was measured prior to exposure to light and periodically throughout the course of the experiment. The frequency of measurement lessened as the study progressed, and the visible changes became smaller. Five sets of forty-one samples—shown in Figure 1—were measured; four sets were placed in the different light boxes, and a fifth set was stored at ambient temperature and relative humidity in the dark as a control. Only the results from the samples exposed to artificial daylight fluorescent lamps are described here. Each measurement was recorded once; however, as four sets of samples were being exposed to light and, as mentioned above, the principal difference between the sets of samples was in the rate of change, not in the pattern of change, it was possible to make a comparison between sets of measurements, to check on consistency and to confirm that unexpected features observed were indeed real.

There were some changes to the samples kept in the dark, but these were due to phenomena unrelated to the light-induced fading of the samples in the light boxes. A common change, for example, was the yellowing of the linseed oil medium (see Section 3.5).

All colour measurements were made using a Minolta CM2600d spectrophotometer/colorimeter. The sample area measured was 8 mm in diameter and the instrument was set to SCE (specular component excluded) mode, to ensure that the readings were not affected by surface gloss. For each sample, a spectrum from 400 to 700 nm, at an interval of 10 nm, was recorded, selecting the option to exclude ultraviolet radiation from the illuminating radiation in order to minimise fluorescence. The spectral data were converted by the instrument into co-ordinates in the CIE Lab colour space, calculated using the D65 standard illuminant and the CIE 2° observer [26]. In addition to the L*, a* and b* data for each measurement, the hue and chroma (C*, the colour saturation or strength) were calculated. For each sample, the CIE Lab data were used to calculate the colour difference between the sample before light exposure and at the time of measurement during the experiment. The colour differences were calculated using the CIE 2000 colour difference formula and are given in ΔE₀₀ units [27].

2.4. Exposure Half-Lives

Many of the plots of colour change in ΔE₀₀ units against cumulative light exposure in Mlux·h show exponential behaviour (see, for example, the first of the graphical figures below 4). In the conservation field, this pattern of fading behaviour is very common, certainly among the natural dyes. It is perhaps the most commonly encountered among the five types of curve described and plotted by Giles in 1963 [28] (Figure 14), and 1965 [29] (Figure 10), largely for synthetic dyes, and replotted by Crews for her study of natural dyes in 1987 [6]. Changes that proceed in an exponential manner are often quantified by calculating a half-life. This is normally the time taken for half the total change to occur, for example, half the radioactive nuclei in an element to decay. In this experiment, the half-life can be expressed as exposure (in Mlux·h) and will correspond to the exposure needed for half the total colour change to occur. The shorter the half-life, the more fugitive the material.

The half-lives were calculated by plotting log_e[ΔE_00(∞) − ΔE_00(Ex)] against Ex

ΔE_00(Ex) is the colour change after an exposure of Ex, and ΔE_00(∞) is the colour change at the end of the experiment.

The best fit straight line was calculated using the data for the first 50 Mlux·h of the experiment, and the slope of this line (s) was used to calculate the half-life (Ex_½) in Mlux·h using:

Ex_½ = log_e(½)/s    or Ex_½ = −0.69315/s

The method tends to underestimate the stability of the most stable materials as they are more likely to have retained some colour at the end of the experiment, so that ΔE_00(∞) does not quite reflect the final colour they might reach if exposure was continued.

3. Results

As might be expected after an exposure equivalent to around 300 years of unfiltered daylight, many of the samples showed dramatic colour changes, often fading completely (Figure 3). In general, the loss of colour in the textile samples is less severe than that in analogous pigment samples; the details are set out in the sections that follow.

The changes measured for a set of samples kept in the dark for the duration of the experiment were reported in the 2008 paper [10] (

Table 3. Half-lives for selected samples.

Sample	Sample Code	Half-Life (Mlux·h)
Cochineal C3 (Sn substrate) + lead white 1:20 w/w	C3A	27
Cochineal C3 (Sn substrate) + lead white 1:8 w/w	C3B	80
Cochineal CD1 (Al substrate) + lead white 1:40 w/w	CD1A	26
Cochineal CD1 (Al substrate) + lead white 1:20 w/w	CD1B	29
Cochineal CD1 (Al substrate) + lead white 1:8 w/w	CD1C	46
Madder lake from wool MD6ai	MD6A	114
Madder lake from plant M7i	M7A	120
Eosin lake E1	E1	8
‘Brazilwood’ BBr 10 + lead white 1:40 w/w	B10A	6
‘Brazilwood’ BBr 10 + lead white 1:10 w/w	B10B	12
‘Brazilwood’ BBr 10	B10C	36
Weld lake W6 (Al substrate) + lead white 1:3 v/v	W6LW	25
Weld lake W7 (Al/Ca substrate) + lead white 1:3 v/v	W7LW	29
Madder (mordanted + semelwater and alum; dyed + semelwater)	RW1A	59
Madder (mordanted + semelwater and alum; dyed + semelwater; treated + K2CO3)	RW1B	60
Madder (mordanted + oak galls then + alum; dyed)	RW2A	55
Madder (mordanted + oak galls then + alum; dyed; treated + K2CO3)	RW2B	52
Brazilwood (mordanted + alum; dyed)	RW3A	21
Brazilwood (mordanted + alum; dyed; treated + K2CO3)	RW3B	21
Cochineal (mordanted + alum, D-tartaric acid, salt and sandalwood; dyed + starch, salt and turmeric)	RW4	53
Cochineal (mordanted + alum, D-tartaric acid, salt and sandalwood; dyed + gum arabic, alum, salt and turmeric)	RW5	43
Weld (mordanted +alum; dyed + K2CO3)	YW1	30
Dyer’s broom (mordanted +alum; dyed + K2CO3)	YW2	32
Green (weld as in YW1, overdyed with woad indigo)	GW1	27
Green (woad indigo, overdyed with weld as in YW1)	GW2	24

). The only sample for which a colour change greater than one ΔE₀₀ unit was observed was GW2 (woad indigo overdyed with weld), the colour of which is rather less homogeneous than most of the other textile samples.

3.1. Half-Lives

The half-lives (Ex_½) calculated from plots of log_e[ΔE_00(∞) − ΔE_00(Ex)] against Ex for selected samples are tabulated in Table 3 and used in the discussion of the results below.

3.2. Red Pigments and Dyes

A wider range of red dyes is represented in this study than any other colour, not only in the number of dyes used but also in the way in which they were used. This can be seen in the different substrates in the case of the lake pigments, or in the different mordants and other ingredients used in the case of the dyed wools. The pattern of fading shown by cochineal is a good example of what happens in general and the different cochineal-containing samples of pigment studied provide a representative selection of the different parameters examined, including the different substrates and the addition of white pigment. Following its introduction into European dyeing and pigment-making technology in the sixteenth century, cochineal dye was used to prepare a lake pigment on a hydrated alumina substrate in the same way as the other commonly used red dyes, such as kermes. Some time after the seventeenth-century discovery that it could give a brilliant scarlet red on textiles using a tin-based mordant [30], cochineal pigments on tin-containing substrates came into use during the eighteenth century. The fading behaviour of these cochineal lakes on two different substrates can thus be compared. The colour change of the dye in pigment form can also be compared with that of the same dye used on wool, although as wool dyed with cochineal on a tin-containing mordant was not part of the MODHT project survey, this particular comparison cannot be made, only that with the alum-mordanted dye on wool. The colour change behaviour shown by the cochineal-containing samples is then compared with that exhibited by the other red dyes, first as lake pigments and then as dyed wools.

3.2.1. Cochineal-Containing Samples

Figure 4a shows a plot of colour change in ΔE₀₀ units against cumulative light exposure in Mlux·h. In Figure 4b, the colour changes over the first 60 Mlux·h can be seen in more detail. Comparison of the two figures allows differences between the overall extent of colour change in Figure 4a, which very much depends on the amount of colourant in the sample, to be distinguished from changes in the rate of colour change, seen more clearly in the initial stages of the experiment in Figure 4b and dependent on the nature of the colourant and other materials in the paint film. For example, we see that although the overall colour change for sample C3B is greater than that for C3A (because the former contains a higher proportion of the same red lake pigment and thus has more colour to lose), the initial rate of colour change is greater for C3A (Ex_½ = 27 Mlux·h) than C3B (Ex_½ = 80 Mlux·h), for reasons explained below that are connected with the presence of more lead white in the mixture.

In general, the cochineal-containing lake pigments are more fugitive—less stable to light—than the same dye on wool, exemplified by textile samples RW4 and RW5. Among the pigment samples, those on hydrated alumina substrates, labelled CD1A–C, tend to be somewhat more stable to light than those on tin oxide substrates.

A typical example of the fading of cochineal-containing pigment is shown in Figure 5. This is a portrait of Anne Dashwood by Sir Joshua Reynolds, an artist whose use of cochineal lakes was much criticised because they faded during his lifetime, notably in the paler tints of the sitter’s flesh, as here [1] (pp. 79–80), [31].

For the three samples of cochineal on hydrated alumina mixed with different proportions of lead white, there is an increase in the initial rate of colour change—seen in Figure 4b—as the proportion of lead white increases; i.e., CD1C (Ex_½ = 46 Mlux·h) < CD1B (Ex_½ = 29 Mlux·h) < CD1A (Ex_½ = 26 Mlux·h). In the same manner, the curve for C3A, the cochineal lake on tin oxide, mixed with quite a lot of lead white, rises more steeply than that for sample C3B, with less lead white added. The higher rate of fading seen in lighter shades, which has already been noted by the authors in mixtures of red lakes with lead white [1], has been attributed to increased internal reflectance of light within the paint film by the white pigment particles so that the lake particles are exposed to a greater proportion of incident light [8].

The consequence of this pattern of colour loss in mixtures with white over time is that the mid tones and highlights tend to become indistinguishable, with the colour preserved to any extent only in the deepest tones. This can be seen in Figure 6, a mid-fifteenth-century panel, Ecce Agnus Dei, by the Sienese painter Giovanni di Paolo, one of a group of 12 that perhaps formed the doors of a reliquary shrine to the saint. In this example, the lake pigment is not cochineal; it could contain either lac or kermes dye as the colouring matter, but the principle is the same. The red colour is preserved only in the shadowed folds, while the mid tones have been lost, giving the folds a rather flat appearance.

The behaviour of sample C6, the lake on a tin oxide substrate and a starch extender with no addition of lead white, differs from that of the other samples. The initial colour change is very slow, before accelerating midway through the experiment. There are a number of possible reasons for this, but it is difficult to be precise in the absence of analyses of the intermediate and end products. It is conceivable that the reaction is autocatalytic and as colourant molecules are destroyed, they generate further activated species that propagate the reaction. However, under these circumstances, one would not expect the rates to be so different for this particular cochineal sample. Another possibility is that one of the components in this paint system, perhaps associated with the starch extender, quenches the active species generated when the samples are exposed to light and decelerates the reaction. Once the supply of this quenching material is exhausted, the reaction rate would increase, as we see for sample C6 as the exposure proceeds.

Another scenario is that, initially, a new coloured intermediate species is produced that is quite stable and only as the experiment proceeds further does this species react, causing the sample to lose colour. Figure 7 shows a plot of chroma (C*) versus lightness (L*) for the cochineal-containing samples; the initial colour is indicated by the filled circle in each case. Most of the samples gradually lose chroma as their lightness increases, whereas sample C6 increases in chroma over the first six months of exposure (100 Mlux·h), and only later in the experiment does its chroma fall below the initial value. A plot of redness–greenness (a*) versus yellowness–blueness (b*) for C6 (not illustrated) shows that the colour initially becomes both more red and more yellow before the colour is lost in the final stages of the experiment. A comparable, if less extreme, pattern of colour change has been observed in the fading of brazilwood lakes [1] (Figure 5).

A similar initial increase in chroma was also noted by Johnston-Feller and Bailie. During the early stages of the fading of alizarin glazes over a white substrate, the chroma and lightness increased until a point, termed the ‘critical concentration’ by the authors, was reached. Thereafter, while the lightness continued to increase, the chroma decreased [7] (Figure 6). As sample C6 in the current experiment was one of the more transparent samples (due to the presence of a starch extender, which is transparent in the oil binding medium), it might be considered optically, if not chemically, analogous to the glazes studied by Johnston-Feller and Bailie, who also noted a shift towards a more yellow hue as the sample approached the critical concentration. Recent experimental work has shown that cleavage of the sugar side chain in carminic acid (the colouring matter present in cochineal) can take place during light ageing (in solution and under rather more aggressive conditions than in the present experiment), giving rise to a range of products [32,33]. In addition, cochineal pigment in oil appeared to behave rather similarly to alizarin crimson during light ageing [32].

3.2.2. Red Pigments

Figure 8 plots overall colour change in ΔE₀₀ units against exposure for all the red pigments included in this experiment. The initial steepness of the curves can again be used to gauge the relative stability of the pigments. Of the pure pigments, eosin lake E1 is least stable to light (Ex_½ = 8 Mlux·h), as will be discussed in more detail later. The mixture of sappanwood lake containing the most lead white (B10A) was the least stable sample studied (Ex_½ = 6 Mlux·h). The sappanwood sample with a lower proportion of lead white (B10B) is only slightly less fugitive (Ex_½ = 12 Mlux·h). For these pigments (E1, B10A, and B10B), the most rapid changes occur in the earlier stages of the experiment, indicated by the steeply rising curves that plateau quite quickly.

The initial colour change shown by the sappanwood lake without admixture of lead white (B10C) is slightly slower (Ex_½ = 36 Mlux·h) than for the cochineal samples containing a high proportion of lead white. This may be partly due to the very deep colour of the undiluted pigment, but a plot of a* versus b* for B10C (not illustrated) shows that the sample initially becomes more yellow before it begins to lose redness and lighten, leading to a total loss of colour in this sample (Figure 3). It is known that the dye derived from sappanwood or brazilwood, the most important component of which is brazilein, contains a yellowish-beige constituent, urolithin C, which is often the only indication left in the pigment or on the textile that the dye was ever present [34]. However, it seems likely that the marked increase in yellowness seen as the dyestuff deteriorates—which was also noted in earlier experiments [1]—is the result of the formation of a yellow intermediate compound that is lost on further light exposure. If the sappanwood lakes are compared with the cochineal lakes discussed above, it is clear from Figure 8 that the cochineal lakes are broadly more light-stable than the sappanwood lakes, with those on hydrated alumina substrates generally being more light-stable than those on tin oxide substrates. As only sappanwood lakes on a hydrated alumina substrate were used in this experiment, no equivalent comparison of the relative stability of the dye on different substrates was made, although earlier work has shown that sappanwood or brazilwood lakes on hydrated alumina substrates are more light-stable than those on substrates with a high calcium carbonate content [1,11].

The madder lake pigments are the most stable of those examined in this study. Analysis of these pigments by high-performance liquid chromatography with diode array detection (HPLC–PDA) in the National Gallery laboratory as part of a study of fifteenth- to seventeenth-century pigments showed that the dye present in M7A, extracted directly from madder root, had a high alizarin content. This pigment shows the least change of all, with a colour difference ΔE₀₀ = 10 over the entire experimental period. The dye used for MD6A, which was extracted from dyed wool using alkali following a fifteenth-century recipe, had a high pseudopurpurin/purpurin content and showed a greater degree of change (ΔE₀₀ = 20) [17]. The stability of these samples is also reflected in their half-lives (MD6A = 114 Mlux·h and M7A = 120 Mlux·h), which are greater than those of the other red pigment samples in Table 3 (the half-life was not calculated for cochineal sample C6 as it shows an anomalous pattern of fading, discussed above). Recent experimental work by Willemen and others, including examination of the light stability of the principal anthraquinones in madder and the flavonoids in weld, confirmed the overall pattern shown by madder and weld samples examined here, including the relative light stability of alizarin [35].

3.2.3. Red Dyes on Wool

The same pattern of stability emerges when the three red colourants included in this study are dyed on wool: the brazilwood dyes are the least stable to light, followed by cochineal, then madder (Figure 9).

Both brazilwood samples showed a more-or-less complete loss of colourant (Figure 3). Sample RW3B, which was treated with an alkaline solution after dyeing, shows a greater degree of colour change than RW3A (without alkali treatment), but this is probably because this sample was initially darker and had more colour to lose. Indeed, both samples show the same half-life (Ex_½ = 21 Mlux·h). Alkali treatments, which were used fairly commonly in both dyeing and pigment-making with brazilwood and sappanwood, increase the range of colours obtainable to include purples, although pigments produced in this way have notably poor permanence [36] (pp. 78–81).

The cochineal dyes on wool (RW4: Ex_½ = 53 Mlux·h and RW5: Ex_½ = 43 Mlux·h) have similar light stability to the madder-dyed samples in which the wool was mordanted twice, first with oak galls and second with alum, before being dyed (RW2A: Ex_½ = 55 Mlux·h and RW2B: Ex_½ = 52 Mlux·h). The most light-stable colourant on wool is madder dyed with the addition of semelwater, the mildly acidic fermented bran solution, at both the mordanting and dyeing stage (RW1A: Ex_½ = 59 Mlux·h and RW1B: Ex_½ = 60 Mlux·h). Samples RW1B and RW2B underwent an alkali treatment after dyeing but were otherwise identical to samples RW1A and RW2A, respectively. This does not seem to have made much difference to either the initial colour (Figure 1) or their behaviour when exposed to light (Figure 9).

3.3. Yellow Pigments, Yellow Wools, and Green Wools

A narrower range of yellow colourants was included compared to the reds. The two yellow lake samples examined and the yellow- or green-dyed wools all contain luteolin-based dyes (weld or dyer’s broom), which are more stable to light than quercetin-based dyes such as buckthorn [1,3]. The two weld lakes had different substrates, one hydrated alumina (W6LW) and the other with the addition of calcium salts (largely carbonate, W7LW), but as both were mixed with lead white, the effect of this cannot be judged. Both had lost most of their colour by the end of the experiment (Figure 3). The wool samples containing weld (YW1) or dyer’s broom (YW2) were quite heavily dyed and thus retained some colour even after prolonged exposure. Despite the different end colours, the rate of initial colour change did not differ greatly (YW1: Ex_½ = 30 Mlux·h; YW2: Ex_½ = 32 Mlux·h), although the pigment samples altered slightly more quickly than the dyed wools based on weld or dyer’s broom (Figure 10).

The rate at which the yellow weld lakes examined in this experiment fade can be compared with the red pigments considered earlier. They do not lose colour as rapidly as eosin lake and the pigments from sappanwood and cochineal, particularly C3A, on a tin oxide substrate. Placing the red and yellow pigments in order of increasing lightfastness, gives: Eosin lake < Sappanwood lakes < Cochineal lakes < Weld lakes < Madder lake (dye from wool) < Madder lake (dye from root).

The green wools dyed with the indigo and weld faded at much the same rate as the yellow lake pigments (GW1: Ex_½ = 27 Mlux·h; GW2: Ex_½ = 24 Mlux·h). Both green wools lost their yellow component, incompletely in the case of sample GW1 (woad indigo over weld), which presented an uneven pale green colour at the end of the experiment, and completely in GW2 (weld over woad indigo), which appeared entirely blue (Figure 3).

In Figure 11, the alteration in colour of the yellow lakes, yellow wools, and green wools is plotted as changes in the CIE a* and b* co-ordinates, with the initial colour indicated by the filled circle in each case. All the yellow samples show a straightforward loss of yellow (diminishing positive values of b*). Weld tends to give a very slightly lemon yellow rather than the warmer yellow of quercetin-containing dyes, and this is seen in the fact that the curves of the two lake pigments are just on the green (negative) side of zero on the a* axis. The bright yellow weld-dyed wool YW1 has its starting point far higher on the b* axis than the browner yellow of the dyer’s broom-dyed wool. The starting colours for the two green wool samples are much further into the green region (a* ≈ –12). The principal change is again a loss of yellowness as the weld component fades. However, as noted above, the end colour for GW1 is a pale green, reflected by the position of the final colour in the yellow–green region of the a* vs. b* plot, while GW2 fades to blue, with the final colour exhibiting a negative b* value indicating blueness.

It is often very hard to demonstrate colour changes in yellow lakes in paintings because the pigments are frequently used in very inconspicuous ways: to brighten other yellows or mixed with other pigments in landscape paint, although they are also found as final translucent glazes on yellow draperies. If used with blues (or other pigments) to make greens, by mixing or by glazing a layer of yellow over blue paint, fading can be more obvious [2]. Where yellow glazes have been used, a hazy whiteness is often seen on the surface, which is the remains of the white substrate of the lake pigment. In mixtures, the final colour will tend to be that of the more stable components, with the residual white substrate perhaps lightening the colour slightly or contributing to the translucency of the paint.

Where mixtures of a stable blue and a lake pigment have been used to depict green foliage, fruit, or vegetables, the context of the blue passages that remain can be a good indication that a fugitive yellow pigment was once present. The blue tubular leaves of the spring onions and other foliage in Figure 12 are a good example, as these would undoubtedly have been green when freshly painted.

3.4. Black Dyes on Wool

The black wool samples were all dyed using a source of tannins, extracts of oak galls, or—in the case of BKW3—alder bark as mordant, followed by either iron(II) sulfate alone or, in the case of BKW4, with the addition of copper(II) sulfate. The wools show little change over the period of the experiment, although the alder bark-mordanted wool (BKW3) is slightly more affected by light (Figure 13).

Although all three wool samples mordanted with oak galls (BKW1, BKW2, and BKW4) lighten slightly when exposed to light, plots of a* against b* (not illustrated) for these three samples show changes so small that it is difficult to ascribe a particular change in hue. The alder bark black (BKW3) also shows an increase in lightness, accompanied in this case by a more distinct increase in yellowness (and a slight increase in redness). Although the constituents present in this particular alder bark are unknown, its behaviour is of interest, as alder bark could also be used for dyeing red [37,38].

3.5. Admixtures with Lead White

Some of the pigments were studied as mixtures with greater or lesser amounts of lead white. When plots of a* vs. b* for these mixtures are examined, a notable feature is a large change during the first 24 h of exposure. The alteration is principally a reduction in b*—a loss of yellowness. Both drying oils and egg yolk, used in egg tempera painting, contain yellow compounds, and there are numerous reports in earlier literature of drying oils being left in the light in order to bleach them [39,40]. Many are mentioned by Sir Theodore de Mayerne in his accounts of conversations with painters at the court of King Charles I of England between 1620 and 1646 [41]. Linseed oil was used to bind the paint samples in this experiment, and these sharp losses of yellowness occur during the first 24 h as the oil is bleached. While this rapid loss of yellowness is seen for the cochineal samples mixed with lead white discussed above (CD1A, CD1B and CD1C: not illustrated), the phenomenon is seen most clearly in mixtures with blue or the sample of lead white alone (Figure 14). The initial colour shift is greatest for the pure lead white, but it is also very marked in the light-coloured mixture of indigo with lead white. The change in colour during the first 24 h can be seen to become more prominent for the three mixtures of Prussian blue as the proportion of lead white in the mixture increases from 1:20 to 1:100 and 1:400.

3.6. The Dyeing Process

For some samples, the colorimetric measurements also offered an opportunity to learn more about the dyeing processes. In the early stage of the experiment, the cochineal-dyed wools show an initial rapid loss of yellowness, followed by a more gradual loss of redness, seen on a plot of a* vs. b* (Figure 15). In the seventeenth century, dyers often used turmeric with cochineal to give a less blue-toned crimson result [24] (pp. 157, 167, 174; Appendix XXI–XXVI). Both recipes used to produce the cochineal-dyed wools examined in this experiment included turmeric in the dye bath. Turmeric is, however, a very fugitive dye, which accounts for the initial rapid loss of yellowness seen in Figure 15. The seventeenth-century purchaser of a textile dyed with cochineal and turmeric might appreciate its warm crimson red colour, akin to that given by dyeing with kermes, a resemblance that was apparently intentional [24] (p. 157), but the warmth of the red would have been fleeting, leaving a strong, but bluer, shade of crimson. Modern researchers hoping to find evidence of the practice of adding turmeric during the dyeing process are unlikely to find evidence from surviving textiles, unless they are fortunate enough to be able to examine an area—perhaps from inside the folds of a seam—that has escaped light exposure over the centuries.

Absorbance spectra obtained from the two yellow-dyed wool samples YW1 and YW2 (dyed using alkaline solutions of dye extracted from weld and dyer’s broom, respectively) before any fading had taken place showed a small feature with a maximum at about 660 nm. After 24 h of exposure to light, this feature was almost invisible. Results from the other three light boxes and the dark control samples were also checked, revealing the presence of this feature in the yellow wools in each sample set initially and its disappearance over slightly longer times, depending on the lighting conditions in each box; it remained in the dark control sample. This absorption has been attributed to the presence of chlorophyll A [42,43]. The effects of increasing the pH during the extraction of weld dye, the presence of aluminium ions (Al³⁺) on the fluorescence of its constituents luteolin and apigenin (also present in dyer’s broom), and the resulting variable presence of chlorophyll A have all been noted [42], as well as the effect of varying the amount of alum mordant used on the lightfastness of the dye [43,44]. In the context of the MODHT samples studied in the present experiment, it is probably more significant that the dye was extracted from the plant material using alkali, which has been shown in earlier work to give a strongly coloured result, more so than using the same batch of plant material extracted in water [12] (pp. 54–55, 64–65). These alkali-extracted samples, which were dyed as part of the CHARISMA European project (Cultural Heritage Advanced Research Infrastructures, Synergy for a Multidisciplinary Approach to Conservation/Restoration, funded by the European Union 7th Framework Programme (2009–2014), Grant Agreement no. 228330) exhibited the same absorption feature (unpublished results), whereas it was absent in the other samples in which the dye had been extracted using water. It has also been observed in certain yellow lake pigments made using weld dye extracted using alkali [45] (p. 31), although not in those used in the current study, where alkali was not used for the extraction. The extreme sensitivity to light demonstrated here suggests that, in a museum context, this feature might only be observed in an area of a textile that has always been protected from light [43].

3.7. Eosin

While most of the lake pigments studied in this experiment are derived from fifteenth- to seventeenth-century recipes, eosin lake dates to the late nineteenth century. The eosin lake used in this study is on an alumina substrate and was prepared as part of a workshop held in Amsterdam in 2003, dedicated to the pigments used by Vincent van Gogh [13]. Research into the photodegradation of dyes has progressed considerably in recent years, so there is a growing understanding of the mechanisms underlying some of the changes caused by light in the pigments and dyed textiles in which they were used [46]. This has included many studies of eosin and the pigment made from it, often known as geranium lake [33,47,48,49], not least because of its changed appearance in many paintings by Vincent van Gogh [50,51], including Roses from 1890 (Figure 16), studied in detail by Centeno et al. [52].

The plots in Figure 8 and the data in Table 3 indicate that eosin (Ex_½ = 8 Mlux·h) is the most fugitive of the pure pigments examined in this study and go some way to explaining why some paintings from the late-nineteenth century are so changed, while older paintings and textiles in which other red dyestuffs were used are in rather better condition.

The progress of the colour change in eosin can be followed in a plot of a* vs. b* (Figure 17). There is a rapid loss of yellowness (b*) in the earlier stages of light exposure. As seen in other samples, this is likely to result—at least in part—from the bleaching of the linseed oil in which the pigment was painted out. The sample then loses redness (a*) with a small additional loss of yellowness in the late stages of fading.

The same pattern of change can be seen in absorbance spectra recorded during the fading of the eosin lake sample (Figure 18). The absorbance at the blue end of the visible spectrum declines quickly, corresponding to the loss of yellowness seen in Figure 17. The sample then shows a decrease in the absorbance peaks centred in the 500–570 nm region as fading progresses, accompanied by small shifts in their maxima. It is worth noting that the maxima seen here differ slightly from those sometimes reported for pure eosin in the literature [51], presumably due to the nature of the interactions between the eosin molecule and the substrate to which it is laked.

4. Discussion and Conclusions

The red and yellow natural dyes used in both pigments and textile dyes well into the nineteenth century are not stable to light. Their tendency to fade was, to some extent, known from quite early times, particularly when used as textile dyes, as indicated by medieval regulations controlling the use of certain colourants, notably brazilwood, for example, the Venetian dyers’ statutes, 1243 [53], and those of Ypres, dating from the end of the thirteenth century [54]. The order of permanence of the traditional red colourants in this study, whether used as pigments or dyes, is: sappanwood/brazilwood < cochineal < madder (the most permanent). The nineteenth-century eosin lake is even less light-stable than brazilwood.

When a pigment is used at full strength, it appears more stable to light, principally because more colourant is present, so the loss of a certain proportion of the colourant molecules is less apparent visually. As found in earlier studies, mixing the pigments with white or using a great deal of extender during pigment preparation leads to an increase in the rate of change in colour due to greater internal reflection of light within the paint layer.

Of the yellow colourants, only the luteolin-containing weld and dyer’s broom were examined in this study. These are rather more stable than the redwood dyes or cochineal, although more fugitive than pigments or dyes based on madder. Quercetin-containing dyes such as buckthorn, which is less light-stable than weld, were not examined [1].

While indigo and the black dyes are the most stable to light, the damaging effects of acid hydrolysis, known to occur with black iron/tannin complexes, were not part of this study and may contribute more to the deterioration of textiles dyed in this manner than exposure to light [55].

The pattern of fading shown by the same MODHT dyed wools, examined at Manchester University as part of the work on the digital reconstruction of The Oath and Departure of Eliezer in 2009, was very similar to that observed in the work described here. Comparison of the ΔE results obtained from wools dyed with madder, brazilwood, cochineal, weld, dyer’s broom, and woad indigo after 500 h of Xenotest fading showed that the woad indigo-dyed blue wool was the most stable and brazilwood was the least stable, followed by weld, cochineal, dyer’s broom, and madder. The yellow colour of the faded brazilwood-dyed wool (also observed in the experiment described here: see the two samples on the extreme right of the top row of faded textile samples in Figure 3) was explained as being due to yellowing of the wool by the light ageing process. This may be a factor, but longer-lasting yellowish dye constituents such as urolithin C also contribute to the colour [21,34].

An indication of what can happen to the appearance of a tapestry in about 300 years, during which it is exposed to daylight for about 8 hours a day on average, is illustrated in Figure 19. Eight samples, one each of the wools dyed with madder, brazilwood, cochineal, weld, indigo, black, and both the green wools, shown in the key on the left of Figure 19, have been arranged in a simple pattern. Pink brazilwood and crimson cochineal samples are arranged around a brownish-red madder sample on the left. These are separated from the green, bright-yellow weld, pink brazilwood, brownish-red madder and crimson cochineal on the right by blue indigo and black samples.

This can be compared with the faded examples of the same set of eight samples, arranged in the same pattern, shown on the right of the figure. The blue and black are lighter, but still recognisable, while the madder has faded to orange-brown. The light green is still weakly green, but the dark green is now blue. The cochineal has faded to an unattractive brownish pink. The most serious change has occurred to the brazilwood and weld samples, which are very hard to distinguish from one another, so that this section of the pattern has, in effect, been lost.

Similar effects caused by fading can be seen in tapestries in museum collections and historic houses today. Figure 20a shows an early seventeenth-century tapestry cushion cover in the collection of the Metropolitan Museum, New York. Woven in England, it illustrates a scene from the story of the Prodigal Son, here seen tending the pigs and repenting of the wastrel lifestyle that has brought him to this sorry state. Although the tapestry has been reworked in several areas [56], comparison of the front and reverse sides of a detail of the bottom left corner (Figure 20b) shows the considerable loss of colour that has taken place (the image of the back of the tapestry has been reversed horizontally for ease of comparison). The pale-pink background to the border was originally a bright crimson. Light-blue areas on the variegated foliage were a bright green, together with the blue areas of landscape below the left-hand pig, formerly a dark green, and the landscape in the bottom border: the yellow dye has faded completely. Pale orange-yellow stripes on the vase at the bottom left corner were originally a bright orange, and the house in the bottom border, now a nondescript beige, was originally largely pink. Even the indigo blue shows some fading, although considerably less than the other colours described above.

The study described here is, of course, not without its limitations. First, the damaging effects of light observed in this experiment are considerably reduced in the real museum world by filtering out the ultraviolet component of the illuminating radiation. Nevertheless, many objects were exposed to unfiltered daylight prior to entering museums, and light from which ultraviolet has been filtered remains a threat to most of the colourants studied here.

Second, the experiments examined the behaviour of laboratory-prepared pigments and dyed wools; in the real world, both are more complicated. Apart from anything else, any individual historical pigment or dyed wool may be richer, or less rich, in dye content than the laboratory-made example, influencing its lightfastness. Paints are complex, both in their layer structure and, frequently, in the mixture of pigments present within a layer. The situation represented by Giovanni di Paolo’s pink drapery shown in Figure 6, and in the samples tested, is relatively simple compared to that which may be present in many paintings. This complexity may affect not only how fading proceeds but also the interaction over time of the lake pigment with other ingredients in the paint layer.

Dyers operated under a variety of conditions. Dye baths and indigo vats could be used over a period of weeks or months, being topped up with one ingredient or another, notably, for example, the baths for dyeing black; conditions within them could well change over time until a point came when they were exhausted. In addition, the mordant used, which was not necessarily an aluminium salt, could influence the light- or wash-fastness of the dye, as well as its colour. Perhaps the most important factors to be recognised are those of scale and time. A sixteenth- or seventeenth-century dyer would be dyeing many metres of wool or silk yarn or cloth. This required large vessels, a great deal of clean water, and long times to heat up or cool down, and to extract the colouring matter from the source material to the dyer’s satisfaction, to say nothing of physical strength. A lake pigment maker of the same period may also have worked on a large scale, depending on the type of pigment being made [57].

When this study was conducted, the practice of assessing the fugacity of objects themselves using microfade testing (MFT) had not been widely adopted. MFT can, of course, often provide information on the behaviour of the colourants in specific objects whose light stability is being considered. Nevertheless, the results reported here offer a guide to the relative stability of a number of traditional colourants that have been widely used as pigments and dyes, and give an indication of the changes in appearance that might be expected—or might already have taken place at some time in the past—when the objects in which they have been used are placed on display.