Methodological Approach for Identification of Organic Residues Preserved in Roman Amphorae

Several studies have focused on local workshops in the Picton territory, the western center of Gaul. Picton potters notably reproduced the forms of amphorae imported from the north of Spain (Pascual 1, Dressel 2/4), or from the Narbonensis province (Gauloise 4 and 5). To ascertain to what end these amphorae were manufactured in the region, the chemical study of the organic residues left in the vessels was undertaken. An analytical protocol with a two-step analysis is proposed herein. Firstly, py-GCMS of sherds gives rise to a global print of the different compounds present in their pores, and permits to choose an adapted protocol of extraction for further GCMS analysis. In the present work, the methodology is tested using amphorae sherds originating from local workshops-hence never used-that we soaked in a wine of known composition, before analyzing them via both GCMS and py-GCMS showing the complementary of these approaches and their efficiency in the study of archaeological organic residues originating from amphorae found in consumption sites in Poitiers .

Sherd samples: Two different ceramics were used as model samples: a home-made one [1], and a sherd of a Dressel 2/4 amphora [2] from the archaeological site of Gourgé [11] known as a production workshop [12]. This late sample was chosen as a model since it probably never contained any wine or resin due to its provenance. A residue collected from an amphora [3] found in a well on the hill fort of the Ermitage in Agen [13] (Table 1).

Imbibition:
The imbibition protocol was implemented as previously described in [6]. Each ceramic sherd was split between an imbibed sample and its control (1-C and 2-C). Samples were then thoroughly cleaned with distilled water, dried in an oven at 100 °C for 24 hours and weighed (1-C and 2-C) or immersed in solution A (1-A and 2-A) or in wine (1-W and 2-W) for 3 days. Solution A was made of a concentration of 2g/L of syringic, tartaric, citric, p-coumaric and fumaric acid in an ethanol: water solution (10% ethanol) and the pH were adjusted for the dissolution of the different acids.

Extraction:
The extraction was performed as described by Pecci and her collaborators [6]. 1g of each sample reduced in powder was extracted with 3mL of a 1M KOH solution and sonicated for 90min. The supernatant collected after centrifugation (20min. at 2000rpm) was adjusted to pH 2 with a HCl solution before adding 3mL of ethyl acetate. After shaking, the organic phase was collected and dried with anhydrous sodium sulfate to remove any trace of water. The procedure was repeated 2 times and all the previously dried organic phases were combined and evaporated under a nitrogen flow until obtaining a dry extract.

GC-MS analysis:
Before the GC-MS analysis, the extracts were silylated as follow: 100μL of an N, O-Bis (trimethylsilyl) trifluoroacetamide/trimethylchlorosilane (BSTFA/TMCS; 99:1) mixture and 100μL of acetonitrile were added to the dry extract. The samples were then left for 1 hour at 70 °C for the silylation reaction to take place. 1μL of each sample was injected for analysis and the carri-er helium flow rate was set at 1.2mL/min. The injector was programmed to operate in splitless mode and the analysis of the compounds obtained after separation was carried out using the electron ionization mode (70eV) over a detection range of 36-800m/z.
Pyrolysis GC-MS analysis: 5μL of tetramethylammonium hydroxide (TMAH) were added to approx. 2mg of each sample, then left for a few minutes at 70°C so that the derivatization of the compounds took place and heated for a few seconds until complete evaporation of the solvent, using the thermally assisted hydrolysis and methylation protocol developed by Challinor [14]. The samples were then analyzed by Pyrolysis-GC-MS. As previously, the injector was programmed to operate in split mode at 500 °C and the analysis of the compounds obtained after separation was carried out using the electron ionization mode (70eV) over a detection range of 36-800m/z.

Identification of the compounds:
The compounds were identified by comparing their mass spectrum to the NIST libraries with respect to mass fingerprint fragmentation.
NMR Spectra: Two types of NMR spectra were acquired (CESA-MO facilities). A 1 H NMR spectrum of Mulsum wine (0.5mL) was recorded at 400.13MHz using a pulse sequence standard with water presaturation and the lock was ensured by the presence of D 2 O in a capillary. A 13 C CPMAS spectrum of the residue 3(150mg of powder) was recorded using a standard CPMAS pulse sequence at 100.6MHz, with a rotation speed of 8kHz.

Analytical strategy
The main concern was to distinguish organic contaminants from the specific rest giving information about the use or even the re-use of the containers. Since it is commonly accepted that the form induces the function and as the amphorae selected in the present work were intended to contain mainly wine, at least for a first use, tracking grape/wine biomarkers, such as tartaric, malic, syringic, citric, p-coumaric and fumaric acids was the main focus [3,4,6]. For this purpose, analyses were firstly carried out on the home-made sherd rinsed with distilled water (1-C), or imbibed with solution A (1-A) or with wine (1-W) for testing different approaches. Ceramic samples were weighed before and after soaking and its subsequent drying to have an idea of the quantity of organic matter embedded in the pores. 2612mg of dried ceramic was imbibed with solution A, and after an imbibition of 3 days and a drying of 1 day in an oven, the sherd reached a weigh of 2616mg, corresponding to 4mg of solution A absorbed. The 3529mg sherd reached a weight of 3626mg after the same treatment with wine, corresponding to a wine absorption of 97mg. 1g of each crushed ce-ramic was extracted using the protocol of Pecci et al optimized for wine organic acids [6], and the resultant pellet was taken back for silylation before the GC-MS analysis. The previous imbibition performed with the five reference acids permits both their assignment and their identification in more complex mixtures such as wine, showing that the protocol of extraction used was adapted to the defined objective. (Figure 1) displays the chromatograms obtained for the blank (bottom) compared to sample 1 imbibed with a model solution (middle) or with wine (up). Firstly, this comparison evidences contaminants: beyond the ubiquitous presence of palmitic and stearic acids often attributed to their presence on hands, the detection of syringic acid-commonly used as a biomarker of wine -in the blank was unexpected but can suggest its presence in the clay used for the preparation of the ceramic. Consequently, this acid has not been considered anymore as a wine biomarker in the present study. On the other hand, the different acids sought were identified, along with three others, fumaric, vanillic and lactic acids. If the latter is undoubtedly a biomarker of the fermentation process, fumaric acid is known to be a regulating agent in the malolactic fermentation. For vanillic acid, its presence may have various origins: Figure 1: Annotated gas chromatogram (total ion current display, relative abundance against retention times) for ceramic 1 rinsed with distillated water (blank), imbibed with a solution of reference organic acids (acids) or a wine «mulsum» of the «mas des tourelles».

a.
Grapes and wines contain at least seven benzoic acids including vanillic acid [10], whose concentrations vary depending on factors such as vine variety, winemaking process, maturity of the grapes [15].

b.
A degradation product of anthocyanins during winemaking due to the action of wine lactic acid bacteria, O. Oeni [16]. c.
The thermochemolysis of lignin, especially from guaicyl unit, for wine aged in barrels [17].
On the other hand, 2mg of crushed ceramics 1 and 2 imbibed with wine were directly analyzed by Py-GCMS after methylation with TMAH. The pyrogram profiles obtained were similar for every ceramic imbibed. (Table 2) reports on the relevant compounds with respect to wine detected using this strategy: two organic acids were detected, succinic and fumaric acid, and other characteristic markers of wine that were not observed using the extraction procedure, directed at extracting mainly organic acids. We notice the presence of glycerol markers (2 propanol 1,3 methoxy, and glyceraldehyde dimethyl ether), glycerol being the most abundant constituent of wine after ethanol and water, and also a marker of alcoholic fermentation [10]. Proline was the sole amino acid detected owing to its abundancy in wine, about 30 to 85% of the total amino acid content [18]. It is usually explained by its specific chemical struc-

SBB.000591. 4(4).2021
ture (secondary amine) which is not metabolized by yeast during the alcoholic fermentation. Sugars, such as glucose and its oxidized gluconolactone are also detected and probably reflect the addition of honey to wine, as mentioned on the label (5%). Lastly, 3,4,5-trimethoxy-benzoic acid, methyl ester was identified; it has been previously described as a degradation product of galloylated polyphenols under pyrolysis, and its presence is expected mainly in red wine [9]. In order to confirm the presence and/or source attribution of all the compounds detected by py-GCMS, a proton NMR spectrum of the wine used was recorded. (Figure 2) displays this spectrum. It was acquired under water presaturation to visualize the signals of relevant compounds (water represents about 87% of wine). As expected, the spectrum is dominated by the two resonances of ethanol (compound 7 in the spectrum) (13% alcoholic degree is indicated on the label of the mulsum). Other resonances of relative intensity are observed between 3 and 4ppm and should be assigned to several compounds related to wine such as glucose (1), galactose (2), fructose (8), glycerol (9), or even the proline amino acid (4). The anomeric protons of glucose 1 (d, 5.83ppm, 3 J 3.5Hz) and galactose 2 (d, 4.60ppm, 3 J 8.0Hz), standing out from those in the 3-4ppm range, are noteworthy by their chemical shifts and coupling constant values. The tartaric 3 (s, 4.82ppm), succinic 6 (s, 2,6ppm) and lactic 5 (d, 1,35ppm, 3 J 6.5ppm, coupled to those at 4.2ppm) acids are also observable in the spectrum. Finally, the presence of coupled resonances at 1.97 and 2.3ppm can be assigned to the protons β and γ of proline 4, and the signals around 7ppm to aromatic protons of polyphenols, mainly of the procyanidin family.

Application to an archaeological sample
The analysis of residue 3 collected in an intact Dressel 1A amphora (ST41.296) found in well 41 of the Ermitage hill fort in Agen [13] was performed using the two protocols: extraction/silylation on 1g and methylation/pyrolysis upon 2mg of the residue followed by GCMS analysis. The chromatogram (left) and pyrogram (right) profiles recorded for the two protocols are displayed in (Figure 3), and the different compounds found are reported in (Table 3) for GC-MS analysis of the extract and (Table 4) for Py-GCMS performed directly on the residue. The compounds observed whatever the protocol used appear in bold in the two tables (6 in total). Firstly, the predominance of four diterpenic acids of the abietic series, is apparent whatever the protocol used. This diterpene family is known to come from Pinus sp. which has traditionally been used for coating the inner side of amphorae since Antiquity [19,20]. The observation of succinic acid, known as a biomarker of the alcoholic fermentation, in both pyrograms and chromatograms of the residue supports the use of this type of amphora for wine transport and/ or conservation. The last common compound observed with both approaches is vanillic acid or its decarboxylated derivative trimethoxybenzene, and its possible origins were discussed previously (vide supra). However, hypotheses (iii) and (iv) cannot be retained as its presence in an amphora excludes, a priori, the use of barrels for Italian amphora since wine was made in dolia, and because the vanillic acid was found even in non-pyrolyzed extracts, excluding the degradation due to pyrolysis. This compound has already been observed in archaeological wine and can thus be considered as a wine and grape biomarker [21,22]. However, the absence of the other benzoic acids normally present in wine or grape raises questions. Some compounds were exclusively found in the extract treated by GC-MS: Oxalic acid, commonly found in wine, like benzoic acid and its derivatives, is attributed to degradation products of polyphenols [16]; the ubiquitous palmitic and stearic fatty acids and their derivative azelaic acid (produced by the oxidation of unsaturated palmitic C18:1 w9) were also observed [23].  to attribute to human activity (such as heating) or to natural decay in the sedimentary matrix [24]. These different resinic acids corresponded to the main resonances observed in the solid-state NMR spectrum of the residue (Figure 4). The resonances observed are as follow: the aliphatic (15-50ppm), aromatic (120-150ppm) and carboxylic regions (170-190ppm); oxidation effect are also perceptible through the resonances 7 DH-CO centered at 156ppm and 7,15 DH-OH at 80ppm [25]. All the chemical structures of these resinic derivatives are reported on ( Figure 5). This result confirms the presence of different chemical species observed using Py-GCMS and shows, as expected, that resin is the predominant component of the residue.   Contrarily to NMR, and fortunately, Py-GCMS also permits the detection of minor but decisive components. Sugars and derivatives (such as glucuronic acid) were identified. They should reflect the addition of honey, a common practice to obtain a Mulsum (20, NH XIV, 85). However, their presence can also reflect the use of sundried grapes (that permits the concentration of sugars) to obtain a Passum (20,NH XIV,81). The detection of di/trimethoxybenzenoids in the pyrogram was of interest: their presence can reflect thermochemolysis products of procyanidins [9] whose abundance provides information on the colour of the grape (red grapes are richer in tannins than white, 10). Finally, the presence of p-cumic aldehyde is noteworthy: it is responsible for the odor of cumin, and is known to be present in myrrh. Could its detection be explained by the addition of myrrh to this wine? Since herbs and spices were frequently added to wine, as explained by Pline the Oldest in his Naturalis Historia dedicated to wine and vine (20,NH,XIV,134), this attribution may be plausible.

Discussion
If adapting an analytic protocol enables to better elucidate specific compounds, it is fortuitous to find an unexpected compound. As example, in amphorae forms known to have contained wine, most researchers focused on finding wine biomarkers such as organic acids and adapted an extraction procedure according to this purpose. To date, there are three of them, the approach proposed by McGovern et al. [5,6] and Garnier [8] with the same objective,

SBB.000591. 4(4).2021
detecting L-tartaric acid or its salt specific to grapes and thus wine and, since the analytical procedure permits it, other organic acids responsible for wine acidity such as malic, fumaric, succinic and lactic acids. Herein, another analytical strategy is proposed : a technique developed in the 1960s, pyrolysis coupled with GCMS -Py-GCMS-was implemented. It offers the possibility to analyze organic macromolecules of low volatility and is currently utilized in the vast domain of archaeological sciences for characterizing organic paint binders [26] or natural and synthetic resins [27]. In the present work, the Py-GCMS is directly applied on sherds without any prior extraction before to obtain a molecular print of the sherd content. This approach has been previously used twice: Oudemans and Boon [28] applied Py-GCMS on a sherd and detected both aromatic and aliphatic compounds, and Kaal and collaborators [29] explored this method on a series of potsherds and provides a rapid molecular screening of the organic matters embedded in the pottery matrix. In those two examples, a methylation reaction was conducted using a protocol adapted from Challinor [14], as this step increases the compounds volatility for further GCMS detection.
As shown in the present work, this approach evidences the presence of compounds that were not detected in the extract, the chosen method permitting to extract mainly the organic acid fraction. If the use of Py-GCMS does not allow observation of all ten acids observed in the extract (only three of them were observed: fumaric, succinic and benzoic acids; Table 2), some other compounds of interest were detected. Sugars are identified in accordance with the addition of honey as indicated on the Mulsum bottle, as well as an amino acid known to be commonly found in wine. As proof of their presence in Mulsum, all these compounds were detected by 1 H-NMR. This approach was applied to an archaeological residue sampled from an amphora. Four acids were identified in the extract (oxalic, benzoic, succinic, vanillic acids, Table 3), as well as resinic acids, whose presence is in accordance with a pretreatment of the amphora with resin and content such as wine. In addition, other components of interest were also observed on the pyrogram, possibly providing information about the amphora content or even the winemaking process. In that regard, the presence of sugars is remarkable since it suggests the addition of honey or/and the use of sun-dried grapes. The presence of tannins (procyanidins) derivatives is equally noticeable, as it indicates the red color of wine -their presence being at least 10 to 50-fold larger than in white wine [10]. The combination of the two methods (extraction followed by GCMS, and Py-GCMS) allows the identification of a lot of different components. It will be applied in the future in amphorae found in the Picton territory (western center of Gaul) where local workshops were discovered [12]. Picton potters notably reproduced different forms of amphorae (Dressel 2/4 or Gauloise 4 and 5), and the chemical analysis of the residues found in those vessels discovered in consumption sites will ascertain to what end they were manufactured. Indeed, the direct use of Py-GCMS on a ceramic fragment or a solid residue (1-2mg) can give an idea of the nature of the compounds present and thus direct their extraction for a more detailed analysis.