The development of a generic analysis method for natural and synthetic dyes by ultra-high-pressure liquid chromatography with photo-diode-array detection and triethylamine as an ion-pairing agent

In cultural heritage the characterization of organic colorants is a challenging task. Currently


Introduction
Dyes and pigments have fascinated humankind ever since their discovery.Colorants have been used to enhance the aesthetics of paintings, textiles, furniture, and later also plastics.Dyeing itself is one of the most ancient arts, with history dating at least 50 0 0 years back [3] .Until the mid-19th century all organic colorants were derived from natural sources, such as plants and animals.Hence, these are called 'natural dyes'.After the discovery of the first synthetic dye, mauveine, in 1856, brighter colors could be produced that were also cheaper to manufacture on a large scale.Within just a few decades natural dyes were replaced by their syn-thetic analogues, which led to a decline in the production of natural dyes [3] .Objects originating from this time period may contain natural or synthetic dyes or a mixture of both.Natural and synthetic dyes differ significantly in their chemical properties.Natural dyes are mostly neutral or slightly acidic compounds, while there is a huge range of neutral, acidic and basic synthetic dyes.Such compounds are typically analyzed using liquid chromatography (LC).Since the chromatographic behavior of these components varies significantly, different chromatographic methods are needed for an optimal separation of the dyestuffs in question.In order to select a suitable analysis method for cultural-heritage objects, information on the origin of the used dyestuffs is required.Unfortunately, in most cases information about the investigated object is insufficient to determine which method should be used.The typical solution to this problem is to apply multiple analytical methods on the sample to obtain sufficient chemical information, however, this is more time consuming.In addition, in cultural heritage sam-ple material is limited as one must avoid jeopardizing the integrity of the object.Therefore, it is paramount that LC methods yield as much information as possible.
To illustrate the vast diversity and specificity of LC methods for dyestuffs in cultural heritage, we discuss several of them here.In 1985, Wouters and colleagues introduced an LC method using photodiode-array (PDA) detection for red anthraquinone dyes [4] .The chromatographic method was based on a reversed-phase (RP) mechanism and employed gradient elution using water and methanol (MeOH) with 1% formic acid (FA).The sensitivity of this method allowed determining kermesic acid in both Polish and ordinary species of cochineal, insect species often used to dye red.In later research, the authors replaced FA by 0.5% phosphoric acid for the analysis of natural dyes, which could then also be applied to purple and blue indigoid dyes [5][6][7] .Another LC-PDA method for the analysis of natural dyes was presented in 1996 by Halpine et al., who used a water/acetonitrile gradient with 0.1% trifluoroacetic acid (TFA) to analyze fibers from furniture tapestries and lake pigments from a watercolor box [8] .The organic colorants that were analyzed ranged from indigo and Indian yellow to madder lakes.Subsequently, in 2013, Serrano et al. described the performance and method development of a UHPLC system for the characterization of natural dyestuffs, including flavonoids, carotenoids, anthraquinones and indigoids [9] .Seven UHPLC columns were tested with different gradient-elution programs, different eluents, flow rates, temperatures, and run times.In addition, the differences in performance between 5% phosphoric acid and 1% FA were also evaluated.The final UHPLC method used a water/methanol gradient with 1% FA instead of phosphoric acid.The use of FA as an additive yielded better resolution and a larger peak capacity compared to phosphoric acid and allowed for hyphenation with mass spectrometry (MS), thanks to its volatility.
The analysis of synthetic dyes proved to require a different approach.Van Bommel et al. performed LC analyses on early synthetic dyes that were used in the period 1850-1900 [10] .The aim of their study was to investigate the performance of several analytical methods on a selection of 65 synthetic dyes, covering all dye classes from that period.They applied two gradient methods.The first was similar to the phosphoric acid method developed by Wouters, except that a different C18 column was used.Many of the synthetic acid dyes could be detected with this method.However, for most acid dyes broad and tailing peaks were observed.Therefore, a second method was investigated, using tetrabutylammonium (TBA) as an ion-pairing agent, which was based on earlier research on modern synthetic dyes, where TBA was introduced as a positively charged counter ion to neutralize sulfonic-acid groups [11][12][13] .Using the TBA method, good chromatographic behavior was observed for all acid and direct dyes.Direct dyes are bound to the substrate directly and are generally defined as anionic.Neutral and basic dyes, however, showed better peak shapes and higher peak capacities with the phosphoric-acid method.Both methods were successfully tested in practice by analyzing samples taken from two faded embroideries from Emile Bernard (ca.1892 and 1904) [14] .As a result of this research, the phosphoric-acid and TBA methods were then accepted by the authors as the appropriate methods for natural and synthetic dyes, respectively.It should be stressed that each of the above mentioned methods either focuses on the analysis of natural or synthetic dyes, due to the vast molecular differences between the two dye classes.While each of the methods offered selective separation power, none provided a general approach.
Since then, there have been several reports that describe the successful analysis of both natural and synthetic dyes by reversed phase LC methods.However, most of these methods were limited to a small number of dyes or dye classes [15][16][17] and/or needed re-analysis for identification of synthetic dyes by a second system, mostly MS, using different gradient programs and solvents [ 18 , 19 ].Chen et al. [20] reported on the identification of 42 synthetic and 17 natural dyes in fourteen silk samples by a single LC method.They applied LC-DAD-ESI-MS using gradient elution with a mixture of water, acetonitrile and a constant amount of FA at 0.1%.Although successful, the authors stated that some synthetic dyes could be better analyzed using an ion-pair.They also shared their concern about the complexity of establishing a generic method for all dyes.
In an attempt to resolve this problem, Pirok et al. applied comprehensive two-dimensional LC (LC × LC) to analyze different types of synthetic dyes and their degradation products in one method [21] .In LC × LC fractions of the first-dimension eluents are subjected to a very different ('orthogonal') second-dimension separation to yield additional selectivity and separation power.One drawback of LC × LC is that method development is rather complex and time consuming.To reduce method-development time, the authors developed a computer-optimization tool called 'Program for Interpretive Optimization of Two-dimensional Resolution' (PIOTR) [1] .It was based on retention modeling and on efficient methods to obtain the unique retention parameters for each compound.The latter was used to derive method parameters that provided optimal separations considering various optimization criteria.Among the possible chromatographic parameters that must be selected in reversed-phase liquid chromatography (RPLC) experiments are the starting time of the gradient ( t init ), the duration of the gradient ( t G ), the flow rate ( F ), and the initial ( ϕ init ) and final ( ϕ final ) fractions of the organic modifier.Other conditions such as column dimensions, the type of solvents and type of instrument are considered by the program to improve the accuracy of the prediction.Adjusting chromatographic parameters by means of a software program, instead of by trial-and-error experiments in the laboratory, saves a significant amount of work, time, and solvents.More information about the mathematics behind PI-OTR is described in the Supporting Information (SI).The authors used the program to develop a method for a mixture of 54 synthetic dyestuffs.Compounds were separated in the first LC dimension by ion-exchange chromatography (IEC), followed by ion-pair RPLC (IP-RPLC) in the second dimension, using 10 mM tetramethylammonium (TMA) or TBA at a pH of 3 (water/acetonitrile) with an Agilent ZORBAX Eclipse Plus C18 RRHD column [21] .
Later, Pirok et al. optimized the LC × LC method for both natural and synthetic dyes, with TMA and TBA being replaced by 5 mM triethylamine (TEA) [22] .This method was successfully applied to extracts from cultural-heritage objects containing natural and synthetic dyes.While the method was widely applicable to all classes of dyes and allowed more unknown species to be detected, costs and detection sensitivity still favor a one-dimensional UHPLC method for routine analysis.
Therefore, we set out to develop a novel highly-optimized UH-PLC method for the simultaneous analysis of acidic, basic and neutral dye components in a single experiment.With the rather large number of dyes, baseline separation between all compounds within a reasonable analysis time is challenging.We therefore aim to the computer-optimization strategy (PIOTR) to optimize a gradient-elution method for a large number (nearly 130) natural and synthetic dye components.The final UHPLC method is applied on several artefacts to evaluate its performance and practicality for the analysis of historical objects, taking analysis time and resolution into consideration.The results will be compared with the results obtained with the previously discussed LC × LC method for natural and synthetic dyes [22] .Our final objective is to use the optimized method to create a compound library based on retention times and PDA spectra obtained from the most commonly applied dyestuffs in the field of cultural heritage.

Instruments
All analyses were carried out on a Waters Acquity H-class UHPLC system (Waters, Nilford, MA, USA) equipped with a quaternary solvent-delivery system, a column oven, an autosampler and a PDA detector.The flow rate was set at 0.2 mL •min −1 and the column oven was held at 40 °C.The two chromatographic columns that were used for separation of analytes were a BEH Shield RP C18 column (150 × 2.1 mm i.d., 1.7 μm) from Waters (part number 186,003,977) and a ZORBAX Eclipse RRHD C18 column (150 × 2.1 mm i.d., 1.8 μm) from Agilent (part number 821,725-901).Both columns were protected with guard columns (5 × 2.1 mm i.d.) containing the same C18 packing from the same manufacturer as the analytical column.PDA data were recorded from 200 to 800 nm with a resolution of 1.2 nm and a sampling frequency of 2 Hz.The equipment was controlled by Empower 2.0 Chromatography Data Software (Waters).

Preparation and optimization of mobile phases
For the preparation of the mobile phases, ultrapure water and methanol or acetonitrile were used.First, FA and NaOH were added to 1 L of ultrapure water at concentrations of 0.1 and 0.02 M, respectively, to obtain a pH of 3. The eluent was prepared by mixing buffered water and MeOH in ratios of 95/5 (by volume) for mobile phase A, and 5/95 for B, respectively.TEA was added as an ionpairing agent to both A and B at a concentration of 1 or 5 mM, or was left out of mobile phase B entirely.In addition to MeOH, analyses were also executed with ACN as the organic modifier to compare the performance of the two modifiers.

Reference dyestuffs
Different mixtures containing reference dyestuffs were used throughout the process of method optimization.An overview of the composition of the used dye mixtures is presented in the SI (Tables S1-S3).It should be noted that many of these reference samples comprise original dye powders, which may originate from several decades ago.As a consequence, some dye powders may contain degradation products and/or impurities or side products formed during the synthesis, which would otherwise not be found in pure analogues.However, samples from cultural-heritage objects may feature these same components, which renders the references very useful for method development.
Stock solutions of each dyestuff were prepared at 1 mg/mL in DMSO.Using the stock solutions of the dyestuffs (henceforth referred to as 'dyes'), ten mixtures (henceforth referred to as 'dye mixtures') were created.The first two, mixtures A and B, were created for selection of the chromatographic column and the opti-mization of chromatographic parameters, such as the composition of the mobile phase (SI, Tables S1 and S2).Both mixtures contained several natural and synthetic dyes covering different dye classes.Their molecular structures are presented in Figs.S1 and S2.The dyes were divided over mixtures A and B based on their retention times with the aim to prevent co-elution as much as possible for easy interpretation.Mix A was created by combining 100 μL of eight dyes into one vial with a final concentration of 125 μg/mL for each dye.The same procedure was followed for ten other dyes with a final concentration of 100 μg/mL for each dye to obtain mix B.
For the computer-aided optimization of the gradients, eight additional dye mixtures were created (SI, Table S3).The natural dyes were divided across mixtures 1-5 and the synthetic dyes across mixtures 6-8 by transferring 100 μL of each dye in a new vial to obtain final concentrations ranging from 66.7 to 100 μg/mL for each dye.
Lastly, a dye mixture containing all dyes, hereafter called mix X, was prepared by combining 100 μL each of dye mixtures 1 through 8 in one sample vial, which was used to test the performance of the developed LC gradient methods.As a result, mix X contained 98 dyes, representing close to 130 dye components (including side products, impurities and degradation products).

Sample preparation and analysis of historical objects
To examine the suitability of the optimized LC-PDA method described in this paper, hereafter called the TEA method, samples from several historical objects of different age were analyzed.The samples were selected such that a wide range of natural and synthetic dyes were represented.The first group of samples originated from a unique archeological find of textiles in a shipwreck near Texel [2] , an Island in the north of The Netherlands.Sport divers discovered the wreck in 2009 and in 2015 several garments were discovered and brought to the surface.Among those were a complete 17th century gown, two kaftans, a large cape and several interior textiles, such as used for cushions (Figs.S8-S11).The colors were surprisingly well preserved and, based on the dating, natural dyes were expected to be present.To complement the research, two late 19th century embroideries were investigated (Figs.S12  and S13), which were designed by Emile Bernard (1868-1941), who was known as a painter and a good friend of Vincent van Gogh.He designed a number of embroideries, several of which are now owned by the van-Gogh museum.Due to their date of origin, synthetic dyes were expected to be present.The same samples were also analyzed by the LC × LC method for natural and synthetic dyes recently developed by Pirok et al. [22] .Sample preparation was the same for both analytical methods, using the HCl method that was described in detail by Pirok et al. [22] .

Chromatographic column
The first step towards optimization of the method was the comparison of the two types of stationary phases used by Serrano et al. [9] and Pirok et al. [ 21 , 22 ], a BEH Shield C18 and a ZORBAX Eclipse Plus C18 RRHD column, respectively.The BEH column contains trifunctional silica particles, with an additional hydrophilic carbamate group bonded to C18 groups, and is endcapped.The ZORBAX column contains silica particles treated with a monolayer of C18 stationary phase and is doubly endcapped.
Figs. 1 and S3 show the results of the analyses of mixtures A and B, respectively, on the two RP columns.The mobile phase was prepared as described in Section 2.3 , containing MeOH and 5 mM TEA.A gradient was applied with mobile phase B increasing  from 5% to 95% from 1.50 till 20.0 min, isocratic between 20.0 and 25.0 min, then again from 95% to 5% B within 2 min, and finally isocratic for 5 more minutes.The chromatogram obtained with the ZORBAX column shows narrower peaks and less tailing when compared to the BEH Shield column.This is especially observed for amaranth (A2), crystal ponceau 6R (A6), martius yellow (A8) and congo red (A9) in mixture A, as well as for tartrazine (B1), ponceau G (B6) and silk scarlet N (B9, B10) in mixture B. The broader peaks observed for the BEH Shield column may be explained by the additional hydrophilic carbamate groups of the stationary phase, which increase the affinity of very acidic compounds.Natural dyestuffs, in contrast to synthetic dyestuffs, contain mostly neutral or slightly acidic compounds, which is why natural dyes show narrower peak shapes on the BEH column and also why this column has proven to work well for natural dyes before [9] .Most of the dyes that show peak broadening are indeed synthetic acid dyes, except for tartrazine which is a synthetic direct dye.The latter, however, shows chromatographic behavior more like that of an acid dye, due to the presence of two sulfonic acid groups.
Other compounds that showed poor peak shape on the BEH column contain either one or more sulfonic-acid groups that is charged at a pH of 3.These charged groups may undergo specific interaction with free amide groups on the surface of the stationary phase of the BEH column, resulting in increased band broadening [10] .This effect is not present with the ZORBAX C18 column as the stationary phase does not contain amide groups and free silanol groups that may cause additional band broadening are shielded off more effectively due to the double endcapping.The challenging collection of compounds included in this study asks for an analytical column that facilitates sufficient retention for a wide range of analytes.As the stationary phase of the ZORBAX column yielded a more homogenous performance for this wide set of acidic, neutral and basic compounds, further experiments were conducted using this column.

Effect of organic modifier and ion pair
The effect of the composition of the mobile phase on the chromatographic performance was studied by varying the concentration of the ion-pairing agent between 1 and 5 mM TEA in mobile phase A and B, and by omitting TEA from eluent B during analyses of mixtures A and B. In addition, the effects of MeOH or ACN as organic modifiers were studied.The same gradient as in Section 3.1 was applied.Increasing the concentration of TEA from 1 to 5 mM resulted in improved peak shapes and an increased peak capacity ( Figs. 2 and S4, for mixtures A and B, respectively).Improved peak symmetry was especially observed for compounds that possess multiple charged sites (crystal ponceau 6R, tartrazine, rhodamine B and silk scarlet N).More polar compounds (i.e.martius yellow and cochineal) eluted earlier using 5 mM compared to 1 mM TEA.
The decreased retention may be caused by competition between the dye and TEA for interaction with the stationary phase, or by ion pairing of TEA with remaining silanol groups on the stationary phase.Several compounds that elute later at a higher concentration of TEA are those that contain additional sulfonate and/or azo groups (amaranth, crystal ponceau 6R and tartrazine).Positively charged ion-pairing agents, such as TEA, are generally added to neutralize anionic fractions and to enhance retention of the ion pair based on hydrophobicity through alkyl chains on the ammonium ion [3] .The improved peak shapes and increased retention times for acid dyes with 5 mM TEA compared to 1 mM TEA can be assigned to an increased neutralizing effect of the ionpairing agent, as there are more counterions present.
When TEA was omitted from mobile phase B, a narrower retention window was observed, with poor peak shapes, especially for later eluting peaks (Fig. S5).This starts to be noticeable after approximately 12 min, when the ratio of A and B is roughly 50/50%.If both mobile phases contain an equal amount of TEA, the ion-pair concentration will remain constant throughout the analysis.Therefore, TEA should be added to both A and B, as is generally done for IP-RPLC [ 9 , 23 , 24 ].In some particular situations the ion-pair agent may be left out of the organic modifier, as Pirok et al. did to reduce the elution time in the second dimension in LC × LC [21] .
The differences between MeOH and ACN as the organic modifier were assessed by analyzing dye mixtures 1-8 and importing the obtained retention times in PIOTR.Analyses with ACN resulted in significantly shorter retention times for all analytes, as shown for mixture 1 in Fig. S6.Their respective peak widths, however, did not change, which led to more peak overlap and an overall decrease in peak capacity.Fig. 3 shows the retention plots for MeOH and ACN generated by PIOTR.The lines in these graphs represent all 127 dye components and show their retention characteristics ( ln k ) at any organic modifier fraction ( φ).For MeOH the retention lines are distributed more evenly than for ACN, which implies that many of the compounds show different elution behavior.This phenomenon increases the possibility of computing a gradient method that results in the separation of as many compounds as possible.Based on these graphs, and because ACN is less environmental friendly and more expensive, further experiments were carried out with MeOH.

Gradient optimization by PIOTR
The gradient-elution method was optimized using PIOTR and included the evaluation of both linear and step gradients.To optimize the LC gradient for a mixture of compounds based on a log-linear retention model, the retention parameters, i.e. the slope ( S ) and the intercept ( ln k 0 ), of the retention lines must be known ( Fig. 3 ) [25] .To obtain these, two or more experimental retention times are needed, which can most-conveniently be established from two gradient experiments, the effective slopes of which are substantially different.In general, more measurements lead to more accurate retention parameters.However, previous reports showed that robust findings could be obtained using just two scouting gradients [ 26 , 27 ].Conventional wisdom prescribes a difference in slopes by at least a factor of three, but recent research has shown that smaller differences may suffice [26] .Nevertheless, two gradient experiments of 10 and 30 min duration ranging from 5% to 95% of B were performed for mixes 1 through 8.The retention times of all dye components for both the short and the long gradient were entered in PIOTR.Next, the ln k 0 and S values for each analyte were calculated (SI, Table S4).These values were used for computational optimization of the gradient method.
Parameters that were optimized for both step gradients and linear gradients were the time before the gradient is programmed to start ( t init ), the duration of the gradient ( t G ), and the initial ( φ init ) and final ( φ final ) concentration of the organic modifier.The flow rate F was fixed at 0.2 mL •min −1 .The maximum t G for linear and step gradients was set at 70 and 90 min, respectively.The concentration of the organic modifier was varied from at least 5% as Fig. 5. LC-PDA chromatograms of dye mixture X obtained after analysis using two linear (L1 and L2) and two step gradients (S1 and S2), which closely matched the retention predictions of PIOTR.Indicated peaks are (1) quercitrin, (2) martius yellow, (3) xantho-purpurin and (4) emodin.

Table 1
Overview of the selected natural and synthetic dyes that were used to assess the within-day and inter-day repeatability.The RSDs were calculated based on the differences between triplicates on each day.initial condition up to 95% at the end of the gradient.For step gradients t init , t G (in this case the time another gradient starts), φ init and φ final were varied for each individual step.This was done by a 'funneling approach', with broad settings being applied first and narrowing these down after observing their effects in the Paretooptimization (PO) plots, the framework in which objectives such as the resolution, time of last eluting peak, or orthogonality are considered.In a PO plot every optimization criterion (e.g.analysis time, resolution score, detector sensitivity) is a dimension.POpoints are those for which no conditions exist that yield equal or better values for all optimization criteria simultaneously.Eventually, a final optimization experiment was conducted for which the settings can be found in the SI, Tables S5 and S6.Two linear (L1 and L2) and two step gradients (S1 and S2) were chosen from the PO plots.L1 and S1 were chosen based on the optimum theoretical resolution score ( Fig. 4 ).Gradients L2 and S2 are not PO-points, however, were chosen as sub-optimal references to compare to L1 and S1.The settings of the four gradients can be found in the SI, Table S7.
The four gradients were first applied on dye mixtures 1-8 to obtain retention times of each analyte.Next, the mixture of all 98 dyes (mix X, Fig. 5 ) was analyzed.The retention times and absorption spectra retrieved from the analyses of mixes 1-8 were used to interpret the results obtained from the analysis of mix X.For several compounds that could not be assigned, we noted that the concentration was very low, which caused those peaks to be overshadowed by overlapping peaks of more-concentrated or more-absorbing compounds.These components were re-injected individually, several at higher concentrations, and all eluted at the expected retention times.It is good to note that most overlapping peaks showed very distinguishable PDA spectra, which would make it easier to identify both peaks when present at higher concentrations.This also brings forward the importance of the use of MS for identification of dyes in general, but especially in such complex mixtures.Fortunately, the ion pair in the optimized method is compatible with MS, however, implementation thereof was out of the scope of this work.
The performance of PIOTR to predict retention models was evaluated based on the accuracy of the prediction of the retention times for each compound in mix X.The error between the predicted and experimentally obtained retention times was very small for all four gradients and are presented as two overlays (SI, Fig. S7 and Table S8).Prediction accuracies of 97.8% and 97.1% for gradients L1 and L2, and 96.4% and 97.2% for gradients S1 and S2 show that PIOTR can predict retention models accurately for both linear and step gradients.
The experimental results of the four gradients were evaluated in terms of chromatographic separation, the number of analytes that could be assigned, and the total analysis time.Due to the high complexity of the mixture, in contrast to what the PO plots may show, it is difficult to make a distinction between the four experimentally obtained chromatograms based on chromatographic separation.Furthermore, the percentage of compounds that could be assigned in mix X analyzed by each gradient method was very similar: 75% for gradients L1, S1 and S2, and 77% for gradient L2, the latter being the sub-optimal PO-point.This shows that the resolution score as a criterium for PO plots is not very meaningful for the optimization of such complex mixtures.The objective function of PIOTR was to maximize the product of corrected resolution scores, which means that two overlapping peaks will lower the resolution score towards zero.In PIOTR, the peak widths are estimated based on a standard plate number and a general Van Deemter model, meaning that peaks with higher retention times will be simulated broader than early eluting peaks.Nowadays, this may not be an accurate estimation anymore as UPLC methods with smaller particle columns are used more frequently, such as for the TEA method.With regards to the current work, the calculated resolution score was extremely low due to the use of nearly 130 compounds in the simulations.A better criterium for method optimization of complex samples may have been 'number of peaks', which unfortunately, was not an available option at the moment of optimization.
A choice for a method could not be made based solely on the number of identified peaks.The patterns and peak shapes obtained with gradients L1, L2 and S2 are quite similar.Method S1 results in slightly broader peaks due to the small gradient step from 60/40 to 50/50 A/B in 20 min, the relatively long isocratic step of 15 min, followed by the slow gradient (30 min) to 5/95 A/B (Table S7).Step gradient S1 shows good peak shapes early in the chromatogram.However, the space between 40 and 80 min is relatively empty and this space could be used more efficiently.The overall analysis time is one of the criteria in the PO optimization.A shorter analysis time is generally preferred.Since the better resolution score seen in Fig. 4 for method L1 is barely reflected in the chromatogram of the very complex mixture, both linear methods (L1 and L2) are thought to perform equally well.Therefore, the shortest method (L1) was selected as the decrease in analysis time weighed heavier than the marginal improvement in resolution score.Using the results obtained with this gradient on mixture X, a compound database was created, which included the retention times and absorption spectra of each analyte.

Repeatability
The final method, using gradient L1, was validated for withinday and inter-day repeatability.This was done by analyzing several natural and synthetic compounds (listed in Table 1 ) in triplicate during five successi ve days (a technical error on the 5th day caused the exclusion of data for a few compounds).The data were interpreted manually using the new compound database and the within-day and inter-day repeatability were calculated from the variation in retention times.
The repeatability is calculated as the RSD% of the retention times between triplicates within each day.The obtained RSDs ranged from 0.00% to 0.39%, indicating a good repeatability.An overview of the RSDs of each compound is shown in Table 1 in order of elution time.The inter-day repeatability of the method from day to day was calculated by applying a one way ANOVA on the complete data set.A P -value of 0.999 was obtained, indicating a good inter-day repeatability.

Analysis of historical objects
To assess the performance and applicability of the TEA method for the characterization of historical artefacts containing natural and synthetic dyes, several samples taken from the objects described in paragraph 2.5 were analyzed.The same samples were also analyzed by the LC × LC method described by Pirok et al. [22] since this method was successfully applied for both natural and synthetic dyes and can be used as a benchmark for comparison.Table 2 presents the results obtained by both methods.A more extensive overview of the obtained results are given in section S7 in the SI, together with the obtained LC-PDA data and measured masses in Section S8.
The results described in Table 2 and the SI show that, like the LC × LC method, the TEA method is perfectly capable of analyzing a wide range of natural and synthetic organic colorants used in historical objects from different periods.Therefore, even with little knowledge about the historical background of a sample, this method may be successfully applied, saving time and precious sample material.
In comparison with the LC × LC method, the TEA method was able to detect more compounds especially in the samples from the Texel shipwreck.This may be due to dilution of the sample in the second dimension in LC × LC, causing analytes that are already in low abundance to be non-detectable.However, the LC × LC method also showed non-identified analytes in the second dimension, which could be isomers, side-products or degradation products, which were not separated by the 1D TEA method.
Overall, both methods are comparatively strong, but may be used for different purposes.The LC × LC method provides more in-depth information about the presence of possible isomers, sideproducts and degradation products, but data processing requires more time.In situations where the additional information is not required, the TEA method may serve as a robust routine analysis method providing high throughput and good sensitivity.

Conclusions
A UHPLC-PDA analysis method for natural and synthetic dyes was developed and a compound library for nearly 130 dyestuff compounds was created.It is important to note that dyes are a very diverse group of small molecules with vastly ranging chemical properties (e.g.charge and hydrophobicity) which renders them representative of other chemical classes.Consequently, this work and the method-development strategy will be of use to the development of methods for other classes of small molecules.
The method included the use of a ZORBAX Eclipse Plus C18 RRHD column and a linear gradient elution with a mobile phase consisting of water and methanol with 5 mM TEA at a pH of 3. PI-OTR proved to be an extremely useful tool to build retention models and to predict the outcomes of linear and step gradients for RPLC.However, the resolution score used in this study proved of limited use for the separation of such a complex chromatogram.Thus, more research is required into the development of suitable chromatographic response functions (i.e.objective function or optimization criteria).Fortunately, a more advanced version of PIOTR, called MOREPEAKS [28] , is being developed which does include this feature.
The TEA method described in this paper was successfully applied for the characterization of historical artefacts containing synthetic and natural dyes.Comparison with the LC × LC method for natural and synthetic dyes proved that this method is equally capable of analyzing a broad range of acidic, basic and neutral dye components in a single run, which circumvents the need of having to perform different methods on the same sample, saving precious sample material and time in the lab.The TEA method requires less data processing and specialized knowledge than the LC × LC method and, therefore, provides a good alternative for high-throughput routine analysis.

Fig. 2 .
Fig. 2. UHPLC-PDA chromatograms of dye mixture A extracted at 300 nm analyzed with 1 mM (blue) or 5 mM (beige) TEA in the mobile phase.The same gradient was applied as in Fig. 1 .

Fig. 3 .
Fig.3.Retention plots for 127 analytes calculated by PIOTR after analyzing all components by UHPLC-PDA with 10 and 30 min linear gradients from 5% to 95% ACN or MeOH.

Fig. 4 .
Fig. 4. Pareto-plots presented as in PIOTR obtained after computing different experimental parameters for linear and step gradients.Two of both were chosen to be experimentally tested in the lab: L1, L2, S1, and S2.