New Non-Invasive Approaches for Proteomics and Metabolomics Analyses SSD CHIM/01

Home , Carminic acid

Università degli Studi del Piemonte Orientale “Amedeo Avogadro” Department of Sciences and Technological Innovation Ph.D. Program in “Chemistry and Biology” Cicle XXXII, a.a. 2018/2019

New non-invasive approaches for proteomics and metabolomics analyses SSD CHIM/01

Elettra Barberis

Supervised by: Prof. Emilio Marengo Ph.D. Program Co-Ordinator: Prof. Luigi Panza

DOTTORATO DI RICERCA IN CHEMISTRY & BIOLOGY

Via Duomo, 6

13100 – Vercelli (ITALY)

DECLARATION AND AUTHORISATION TO ANTIPLAGIARISM DETECTION

The undersigned ELETTRA BARBERIS student of the Chemistry & Biology Ph.D

course (XXXII Cycle).

declares:

- to be aware that the University has adopted a web-based service to detect

plagiarism through a software system called “Turnit.in”,

- his/her Ph.D. thesis was submitted to Turnit.in scan and reasonably it resulted

an original document, which correctly cites the literature;

acknowledges:

- his/her Ph.D. thesis can be verified by his/her Ph.D. tutor and/or Ph.D

Coordinator in order to confirm its originality.

- Date: 16/06/2020 Signature:

Contents Chapter 1 Non-invasive analysis of cultural heritage

1.1 Introduction……………………………………………………………………..…………….7

1.2 Development, validation and application of a new non-invasive method for the analysis of proteins and small molecules from cultural heritage objects……………………………………………………………………………………..9

1.2.1 Synthesis and characterization of the EVA film………………….…………11 1.2.2 Application of the EVA film for surface analysis…………………………….14 1.2.3 Proteomics sample preparation and analysis…………………….…………14

1.2.3.1 Protein extraction and digestion protocols…………………..…………..14

1.2.3.2 Proteins analysis and data processing……………………….………………15

1.2.4 Small molecules and lipids sample preparation and analysis…..……17

1.2.4.1 Colorant extraction and analysis……………………………………………….18

1.2.4.2 Small molecules and lipid extraction and analysis …………………….19

1.2.5 Validation of the EVA method………………………………………………………23

1.2.5.1 Preparation of simulated samples ……………………………………………23

1.2.5.2 Proteins and colorants calibration curves and recoveries…………31

1.2.5.3 Non-invasiveness of the method………………………………………………36

1.2.5.4 Method comparison: EVA vs micro-sampling analysis……………….38

1.2.5.5 Test with varnish samples……………………………………………….………..43

1.2.6 Application of the method to historical objects……………….…………47 1.2.6.1 Application of the EVA method to Historical Frescoes……………….47

1.2.6.2 Application of the EVA method to the Collection of Palazzo Madama of Turin…………………………………………………………………………………53 1.2.6.3 Application of the EVA method to a Leonardo da Vinci’s painting from Hermitage museum of St. Petersburg ……………….………62 1.2.6.4 Application of the EVA method to the investigation of the 3500 years old mummy of Ancient Egyptian dignitary. …………………………….….75

1.2.6.5 Non-invasive analysis of Stradivari Violins………………….……………89

1.3. Non-invasive multispectral imaging to uncover ancient Egyptian fresco……………………………………………………………………………………………….100

1.3.1 Imaging system and data processing ………………………………………100

1.3.2 Imaging results and discussion…………………………….…………………..102

1.4 Cultural heritage monitoring and analysis with Infrared Spectroscopy……………………………………………………………………………………109

1.4.1 Non-invasive characterization of colorants by portable diffuse reﬂectance infrared Fourier transform (DRIFT) spectroscopy and chemometrics……………………………………………………………………………………109 1.4.1.1Colourants preparation…………………………………………………………111 1.4.1.2 DRIFT, data processing and chemometrics analysis……….………..113 1.4.1.3 Infrared library result: binders and colourants……………………….114 1.4.1.4 Statistical analysis……………………………………………………………….…125 1.4.1.5 Real Case Study…………………………………….………………………………..128

1.4.2 Applicability of a portable ATR device to monitoring the conservation state of plastic materials in modern and contemporary art……………………………………………………………….…………………………………….131

1.4.2.1 Sample collection of the plastics: Polypropylene and Polycarbonate ………………………………………………..…………………………………135

1.4.2.2 Sample analysis, instrumentation and software……………………..136

1.4.2.3 Statistical analysis: PCA–LDA and PLS-DA…………….………………….136

1.4.2.4 Artificial light aging ………………………………………………………………..139

1.4.2.5 Characterization of plastic materials……………………………………….139

1.4.2.6 PCA–LDA……………………..…………………………………………………………142 1.4.2.7 PLS-DA……………………………………………………………………………………145 References………………………………………………………………..……………………….151 Chapter 2 Non-invasive analysis of biological materials

2.1 Introduction………………………………………………………………………………..173

2.1.1 Development of a new method for the non-invasive analysis of adenoma in CRC ……………………………………………………………………….………174

2.1.2 Method development………………………………………………….……………177

2.1.3 Validation of the method……………………………………………………….….181

2.1.4 Application of the method to colon rectal cancer……….……………..186

2.2 Identification of new biomarkers for the diagnosis of prostate cancer………………………………………………………………………………………………..191

2.2.1 Study population, materials and methods………………………………….194

2.2.2 Results…………………………………………………………………..………………….197

2.2.3 Relevance of our findings………………………………………..…………………212

References………………………………………………………………………..……………….215

List of publications………………………………………………………….………………….221

Patents………………………………………………………………………………………………222

Acknowledgements…………………………………………………………………………..223

Chaapter 1

NON-INVASIVE ANALYSIS OF CULTURAL HERITAGE

1.1 Introduction Recent technological developments in analytical chemistry spurred the analysis of historical, archaeological and paleontological objects. The Identification of proteins and small molecules from cultural heritage objects is crucial to characterize the materials used by the artists, to investigate the context of the object and it can provide invaluable information for designing restoration interventions1, 2.

The application of analytical chemistry in the field of archeology research represents today an important methodological area of study within the archaeological sciences. Although there are several techniques that allow the identification and quantification of minimum amount of molecules from small samples, the majority of the current protocols are at best not completely non-invasive, and usually a micro sampling is required, especially for mass spectrometry analysis.

In fact, based on the recent scientific literature in this field, all the developed analytical procedures require at least a micro sampling from the object3-5. However, non-invasive instruments and techniques are always preferred for the analysis of precious and unique objects6-9, especially in the cultural and archeological fields.

The aim of my PhD research was the development and application of new non-invasive methods for the analysis of cultural heritage.

The following chapters will deal with three main topics: 1) the development of a new method for the non-invasive analysis of proteins and small molecules with mass spectrometry from cultural heritage objects (this technique makes use of a special functionalized film); 2) the discussion of results obtained using a non-invasive imaging instrument; 3) the development and application of non-invasive methods that use portable infrared spectroscopy instrumentation.

1.2 Development, validation and application of a new non- invasive method for the analysis of proteins and small molecules from cultural heritage objects

Proteinaceous materials derived from animal products and tissues (eggs, milk, skin, bones, etc.) have been widely used in cultural heritage as binders and adhesives. Their accurate identification is crucial to study the context of objects and fundamental to improve the conservation practices. Proteomics approaches are often employed for the identification of protein binders in cultural heritage objects, in particular in historical paints10, frescoes11,12 and polychrome pottery.13 Reliable protein extraction and identification from old-age parchment documents had already been performed by sampling tiny fragments,3 while parchment proteins were non-invasively extracted by using an electrostatic charge generated by the gentle rubbing of a PVC eraser.14 The in situ analysis of proteins has been investigated using DESI-MS, obtaining significant results and good preservation of the protein material15. The identification of proteins allowed also a better understanding of human history: the recent uses of shotgun proteomics on a 500-year-old Andean mummy provided for the first time the evidence of active pathogenic infection in an ancient sample16. Moreover, the extraction of proteins and metabolites from the surface of a manuscript page of Bulgakov allowed the identification of morphine and three proteins biomarkers of the nephritic syndrome that affected the author.17

Conventional analytical techniques used to identify and analyze organic substances present in artworks include Enzyme Linked Immunosorbent Assay (ELISA). The method applied for the analysis of proteinaceous media still suffers the problem of the large amount of sample required and the sensitivity to the contamination with other proteins18, 19.

Over the centuries, artists used organic colorants on paper as inks, on wall for mural paintings, on wood panels and canvas. The characterization of these small molecules is very important in order to study the painting techniques or to dating an artwork.

Liquid chromatography mass spectrometry (LC-MS) plays an important role in the analysis of historical organic colorants but all the available protocols require a micro-sample for the analysis20. The non-invasive approach, like visible fiber optics reflectance spectra, can give preliminary results, but mass spectrometry is necessary for the confirmation21,22.

The following chapter describes the development, validation and application of a new non-invasive analytical method that allows the analysis of proteins and small molecules from an artworks surface, leaving unchanged the object23.

The method consists of a special ethyl-vinyl acetate (EVA) film, functionalized with strong cation/anion exchange and C8 resins, which can be used for the extraction of proteins and small molecules from the surface of different types of supports (i.e. canvas, mural painting, painting on wood, parchments...). The extracted molecules are then analyzed by mass spectrometry.

The following paragraphs will report: 1) the preparation and characterization of the functionalized film; 2) the validation of the analytical method, together with the tests performed to evaluate the invasiveness of the technique; 3) the application of the method to the characterization of ancient recipes from historical artworks, such as archeological findings and historical musical instruments.

1.2.1 Synthesis and characterization of the EVA film A special plastic-like film based on ethyl-vinyl acetate (EVA) as binder of ground AG 501 mix-bed cation/anion exchange and C8 hydrophobic resins (all from Bio Rad) was prepared. A mixture was made comprising 70% 1-10 μm size ground beads and 30% EVA (the melting temperature was 75°C). This mixture of melted EVA and Bio-Rad resins was then poured into a "Brabender" mixer W30 and successively extruded via a "Brabender" extruder KE19 (both from Brabender GmbH Duisburg, Germany) in the form of a thin film (figure 1). The thickness of the film was very uniform and ranged from 150 to 200 μm. The synthesis of the film can be performed also without using an extruder: the resins can be dispersed in an EVA solution (20% EVA / 80% cyclohexane (w/v)) and then dried on a cleaned surface until the complete evaporation of cyclohexane using a laboratory fume hood.

Figure 1. EVA functionalized film.

The morphologies and elemental mapping of the functionalized EVA film with embedded beads was then characterized by environmental scanning electron microscopy, ESEM (Quanta 200, FEI Company, Eindhoven, The Netherlands) equipped with EDX (EDAX Inc., Mahwah, USA), in order to characterize its surface. The distribution of the trapped resins was analyzed in order to investigate the chemical behavior and the ability of the material to adsorb proteins and small molecules. Figure 2 shows the ESEM images of the EVA strip: the surface roughness, the morphology of the material and the presence of 10 μm-size particles quite uniformly distributed can be appreciated (fig. 2a). The material of the strip is porous as it can be noted in the cross-section in figure 2b.

Given the fact that the film contains three different types of beads, it was of interest to investigate if they were uniformly distributed on its surface or if there were clusters of any type. In particular, we wanted to ascertain if aggregates of anion and cation exchange resins were present (possibly because of electrostatic interactions between the positively and negatively charged residues), as this would have hampered their ability to

12 capture proteins and metabolites from the specimens under analysis. Panels C to E in Figure 2 indeed show that such aggregates do not exist. Another aspect to be mentioned is that crushing the beads before the preparation of the film is an advantageous procedure, since it enhances the capturing surface area thus favouring a more efficient harvest of the analytes.

Figure 2. ESEM images of the EVA strip: the roughness of the EVA surface (a) is due to the 10-μm size SCX, SAX, and C8 resins as shown on the section of the strip (b). Elemental distribution of sulfur (c), nitrogen (d), and carbon (e) on the EVA strip. Image f shows the overlapping of the sulfur (orange pixels) and carbon (blue pixels) particles.

1.2.2 Application of the EVA film for surface analysis The non-invasive analysis of the surface of the object of interest consists of 3 phases: 1) The functionalized film is first humidified with ultrapure water in order to activate the exchange between the resins; 2) any water residues were then removed from the surface of the strip in order to avoid the release of water micro-drops on the object surface; 3) the film is then positioned with extreme caution on the surface of the object for 10-15 min, based on the type of surface, but also on the thickness of the paint layer, on the level of varnish oxidation and its state of conservation, always in agreement with conservators. After the adsorption of the molecules, the film is then placed in a tube and conserved at 4°C until the analysis that is performed within few hours. Proteins, lipids and small molecules can be then analyzed.

1.2.3 Proteomics sample preparation and analysis

The following paragraphs describe the methods used to prepare and analyze the samples for the proteomic analysis.

1.2.3.1 Protein extraction and digestion protocols

The proteins adsorbed on the EVA film were eluted from the strip with 500 µL of 1.0 M ammonium acetate in a tube for 30 minutes. Then, the strip was removed and the proteins present in the solution were denatured with TFE at 60 °C, reduced with DTT 200.0 mM, alkylated with IAM 200.0 mM and digested with trypsin overnight. The peptide digests were desalted on the Discovery® DSC-18 solid phase extraction (SPE) 96- well plate 25 mg/well (Sigma-Aldrich Inc., St. Louis, MO, USA).

1.2.3.2 Proteins analysis and data processing

The extracted proteins were then analyzed with a micro-LC Eksigent Technologies (Eksigent, Dublin, USA) system that included a micro LC200 Eksigent pump with flow module 5-50 µL, interfaced with a 5600+ TripleTOF system (AB Sciex, Concord, Canada) equipped with DuoSpray Ion Source and CDS (Calibrant Delivery System). The stationary phase was a Halo C18 column (0.5 x 100 mm, 2.7 µm; Eksigent Technologies Dublin, USA). The mobile phase was a mixture of 0.1% (v/v) formic acid in water (A) and 0.1% (v/v) formic acid in acetonitrile (B), eluting at a flow-rate of 15.0 µL min−1 at an increasing concentration of solvent B from 2% to 40% in 30 min. The injection volume was 4.0 μL and the oven temperature was set at 40 °C. For identification purposes the mass spectrometer analysis was performed using a mass range of 100–1500 Da (TOF scan with an accumulation time of 0.25 s), followed by a MS/MS product ion scan from 200 to 1250 Da (accumulation time of 5.0 ms) with the abundance threshold set at 30 cps (35 candidate ions can be monitored during every cycle). The ion source parameters in electrospray positive mode were set as follows: curtain gas (N2) at 25 psig, nebulizer gas GAS1 at 25 psig, and GAS2 at 20 psig, ionspray floating voltage (ISFV) at 5000 V, source temperature at 450 °C and declustering potential at 25V. Each LC-MS/MS instrumental analysis of a sample was preceded and followed by at least one injection of a blank solution in order to assess peptide carryover and a new column and freshly acquired consumable were used in order to avoid contaminations. Common protein contaminants were included in the database search and removed when identified.

The mass spectrometry data were searched using Mascot (Mascot v. 2.4, Matrix Science Inc., Boston, USA), the digestion enzyme being trypsin with 2 missed cleavages. The instrument was set to ESI-QUAD-TOF and the following modifications were specified for the search: carbamidomethyl cysteins as fixed modification, oxidized methionine as variable modification and hydroxylation of prolines and lysines when collagen was present. A search tolerance of 0.08 Da was specified for the peptide mass tolerance, and 10 ppm for the MS/MS tolerance. The charges of the peptides to search for were set to 2+, 3+ and 4+, and the search was set on monoisotopic mass. The databases employed were selected based on the type of support: Swissprot with taxonomy restriction to Chordata database, Uniprot Bos taurus and Uniprot Ovis aries were used (e.g. collagen from animal glue obtained from bovine bones and cartilages were search against Chordata and then Bos taurus, while collagen from the support of the most commonly used parchments obtained from calf or sheep skins, were searched against Chordata, Ovis aries and Bos taurus). Only proteins identified with at least two peptides were considered as significant. Peptides with individual ion scores > 10 were considered for identification purposes11.

For the quantification of peptides for the calibration curve the samples were subjected to cyclic data independent analysis (DIA) of the mass spectra, using a 25-Da window24,25: the mass spectrometer was operated such that a 50-ms survey scan (TOF-MS) was performed and subsequent MS/MS experiments were performed on all precursors. These MS/MS experiments were cyclically performed, using an accumulation time of 40 ms per cycle, per 25-Da swath (36 swaths in total), for a total cycle time

16 of 1.5408 s. The ions were fragmented for each MS/MS experiment in the collision cell using the rolling collision energy. The MS data were acquired with Analyst TF 1.7 (AB SCIEX, Concord, Canada). The quantification was carried out in Skyline 3.1, an open source software project (http://proteome.gs.washington.edu/software/skyline)26 on MS1 (i.e., precursor ion masses). Quantitative MS1 analysis was based on the extracted ion chromatograms (XICs) for the top three resulting precursor ion peak areas e.g. M, M + 1, and M + 2. The signal at m/z 740.4014 (M), m/z 740.9028 (M+1), and m/z 741.4042 (M+2), corresponding to the peptide sequence LGEYGFQNALIVR of BSA, were obtained and employed for the calibration curve. The signal at m/z 820.7771 (M), m/z 821.1114 (M+1), at m/z 821.4455 (M+2), corresponding to the peptide sequence NVLQPSSVDSQTAMVLVNAIVFK of Ovoalbumin, were obtained and employed for the calibration curve.

The identified proteins in the lung of the mummy Nebiri were also subjected to gene ontology classification based on biological processes with the open source software Cytoscape [https://cytoscape.org/] and the ClueGO plug-in. The conceptual framework obtained allowed to construct a global protein-protein interaction network integrated with functional attributes and expression profiles and to associate proteins to specific biological functions.

1.2.4 Small molecules and lipids sample preparation and analysis

The following paragraphs present the methods used to prepare and analyze the samples for the small molecules and lipids analysis.

1.2.4.1 Colorant extraction and analysis

After EVA extraction from the surface of the object, the colorant was eluted from the strip with 500.0 µL of a mixture of methanol and 0.1% of formic acid. The extracted colorants were analyzed with the same instrumental system adopted for the determination of the proteins, described in the previous paragraphs.

The mobile phase was a mixture of water/acetic acid 99.0/1.0 (v/v) solution (A) and methanol/isopropanol/acetic acid 98.0/1.0/1.0 (v/v/v) solution (B), eluting at a flow-rate of 20.0 μL min-1 according to the following gradient conditions: 0.0 min 5% B, 7.0 min 100% B, 8.0 100% B, 8.1-10.0 min 5% B. The injection volume was 300.0 nL. The oven temperature was set at 40 °C. The DuoSpray ion source worked in negative ion (NI) mode. The instrumental parameters were set as follows: curtain gas (N2) at 30 psig, nebulizer gas GS1 and GS2 at 25 and 35 psig, respectively, temperature (TEM) at 400 °C, collision activated dissociation gas (CAD) at 6 units of the arbitrary scale of the instrument and ionspray floating voltage (ISFV) at -4300 V. The mass spectrometer acquired two MS experiments, namely a full scan TOF-MS between m/z 100 and 1000 (accumulation time of 0.1 s), and a MS/MS of the deprotonated carminic acid ion at m/z 491.0835 (accumulation time of 0.15 s), to obtain a total cycle time of 0.3 s. The declustering potential (DP) was set at -50 V both in TOF scan and in MS/MS experiments. The collision energy (CE) was set at -10 V in the TOF scan experiment, whereas at -35 V with the addition of ±5 V due to the collision energy spread (CES).

For alizarin and indigotin analysis, the ionspray floating voltage (ISFV) was in positive ion mode at 5500 V and the mobile phase was water (A) and 0.1% (v/v) formic acid in acetonitrile (B), eluting at a flow-rate of 20.0 μL min-1 with the same gradient conditions described for carminic acid.

Via post-acquisition extraction of the ions (isolation window of 0.02 Da) at m/z 327.0516, m/z 447.0940, and m/z 357.0622. corresponding to the product ions of carminic acid, at m/z 185.0597, m/z 157.0647, and m/z 139.0542, corresponding to the product ions of alizarin, and at m/z 235.0866, m/z 219.0844, and m/z 77.0386, corresponding to the product ions of indigo, data were obtained from the MS/MS experiment at high- resolution multiple reaction monitoring-like (HR MRM-like).

1.2.4.2 Small molecules and lipid extraction and analysis

After the EVA application on the surface (the same reported in the paragraph n. 1.2.2), the molecules were eluted from the film and then prepared using three different protocols. The samples were not splitted in three parts, but each protocol was used for specific application, in example for fatty acids analysis or untargeted metabolomics or lipidomics analysis.

For metabolomics analysis by GC-MS, two different methods were performed in order to extract a wider information. In the first one the elution was performed with 1 mL of KOH in ethanol (10 wt%) for 30 min. After saponification (4 hours at 80 °C), the solution was diluted in water, and the unsaponifiable fraction extracted in 400 μL of n-hexane for three times. The residue of n-hexane extraction was acidified with HCl (6 M) and then extracted with diethyl ether (400 μL, three times). The neutral

19 and acidic fractions were admixed, and the resulting solution was subjected to derivatization with 40 μL of N,O-Bis(trimethylsilyl) trifluoroacetamide (BSTFA), 50 μL of isooctane and 5 μL of tridecanoic acid solution at 90 °C for 30 minutes. The solution was gently dried under nitrogen steam. Finally, 10 μL of isooctane and 1 μL of hexadecane solution were added and 1 μL of sample was analyzed by GC–MS.

In the second method, the metabolites captured by the EVA film were eluted using 1 mL of Ethanol for 30 min. Then, the strip was removed and the metabolites were subjected to derivatization. The derivatization protocol was performed by adding 20 μL of methoxamine hydrochloride in pyridine (20 mg/mL) and 30 μL of BSTFA. The samples were incubated at 80 °C for 20 minutes, and then centrifuged for 15 minutes at 14,500g. Tridecanoic acid (1 ppm) and hexadecane (0.1 ppm) standard solutions were added as internal standards before the derivatization and GC–MS analysis respectively. Nitrogen steam was used to gently dry the samples, before the gas-chromatographic analysis.

As regards lipidomic analysis, the elution of the analytes from the EVA film was performed using 200 μL of acetone for 10 minutes. The sample was then dried using nitrogen stream and then re-suspended in methanol for the LC-MS analysis.

Gas chromatography–time of flight mass spectrometry (GC-TOF/ MS) was performed using an Agilent 7890B GC (Agilent Technologies, USA) and Pegasus (BT) TOF-MS system (Leco Corporation, USA) equipped with an Rxi–5 ms column (30 m × 0.25 mm × 0.25 μm, RESTEK, USA), stationary phase 5% diphenyl-95% dimethyl poly-siloxane. High-purity helium

(99.999%) was used as the carrier gas at a flow rate of 1.00 mL/min−1. The samples were injected in splitless mode at 280 °C. The chromatographic conditions were: initial temperature 40 °C, 5 min isothermal, 8 °C/min up to 300 °C, 20 min isothermal. MS parameters: electron impact ionization source temperature (EI, 70 eV) was set at 250 °C; scan range 40/630 m/z, with an extraction frequency of 30 kHz. The chromatograms were acquired in TIC (total ion current) mode. The mass spectral assignment was performed by matching with NIST MS Search 2.2. libraries, implemented with the MoNa Fiehns libraries.

For the 2D analysis, a LECO Pegasus 4D GCXGC/TOFMS instrument (Leco Corp., St. Josef, MI, USA) equipped with a LECO dual stage quad jet thermal modulator was used. The GC part of the instrument was an Agilent 7890 gas chromatograph (Agilent Technologies, Palo Alto, CA), equipped with a split/splitless injector. The first dimension column was a 30 m Rxi-5Sil (Restek Corp., Bellefonte, PA) MS capillary column with an internal diameter of 0.25 mm and a stationary phase film thickness of 0.25 μm, and the second dimension chromatographic columns was a 2 m Rxi-17Sil MS (Restek Corp., Bellefonte, PA) with a diameter of 0.25 mm and a film thickness of 0.25 μm. The carrier gas (Helium) was used with a flow rate of 1.4 mL/min. After derivatization performed with the method previously described in this paragraph, 1 μL of sample was injected in splitless mode at the same programming rate of the GC-MS analysis. The secondary column was maintained at +5°C relative to the GC oven temperature of the first column. Also, the MS method was the same of the mono-dimensional analysis, while the extraction frequency was 32 kHz, the acquisition rates was 200 spectra/s and the modulation period

21 was 4s for the entire run. The modulator temperature offset was set at +15°C relative to the secondary oven temperature, while the transfer line was set at 280°C.

The extracted lipids were analyzed with a micro-LC Eksigent Technologies system (Eksigent, Dublin, USA) that included a micro LC200 Eksigent pump, with a flow module of 5–50 μL, interfaced with a 5600+ TripleTOF system (AB Sciex, Concord, Canada) equipped with DuoSpray Ion Source and CDS (Calibrant Delivery System). The stationary phase was a Halo C18 column (0.5 × 100 mm, 2.7 μm; Eksigent Technologies Dublin, USA). The column was maintained at 65 °C, at a flow-rate of 15 μL/min. For the LC- ESI(+)-MS analysis the mobile phase consisted of: (A) 60:40 (v/v) acetonitrile:water with ammonium formate (10 mM) and formic acid (0.1%) and (B) 90:10 (v/v) isopropanol:acetonitrile with ammonium formate (10 mM) and formic acid (0.1%). For the LC-ESI(−)-MS analysis the organic solvents for mobile phases were the same with the exception of using ammonium acetate (10 mM) as mobile-phase modifier. The separation was conducted under the following gradient for both ESI positive and negative: 0 min 10% (B); 0–2 min 30% (B); 2–2.5 min 48% (B); 2.5–11 min 82% (B); 11–11.5 min 97% (B); 11.5–12.5 min 97% (B); 12.5– 13.6 min 16% (B); 13.6–15 min 10% (B). A sample volume of 5 μL was used for the injection.

The mass spectrometric detection of the lipids was performed using data dependent acquisition (DDA). The mass spectrometry analysis was performed using a mass range of 150–1150 Da (TOF scan with an accumulation time of 0.25 s), followed by an MS/MS product ion scan from 150 to 1150 Da (accumulation time of 5.0 ms) with the abundance

22 threshold set at 30 cps (30 candidate ions could be monitored during every cycle). The ion source parameters in electrospray positive mode were set as follows: curtain gas (N2) at 25 psig, nebulizer gas GAS1 at 25 psig, and GAS2 at 20 psig, ions pray floating voltage (ISFV) at 5000 V, source temperature at 450 °C and declustering potential at 25 V. The ion source parameters in electrospray negative mode were set as follows: curtain gas (N2) at 25 psig, nebulizer gas GAS1 at 25 psig, and GAS2 at 20 psig, ions pray floating voltage (ISFV) at −5000 V, source temperature at 450 °C and declustering potential at −25 V. For the data processing MS- DIAL (v. 4) software was used (5). The software allows the automated mass spectral deconvolution of the high-resolution MS data, followed by MS/MS library search (LipidBlast).

1.2.5 Validation of the EVA method

1.2.5.1 Preparation of simulated samples

One of the most important issues in the analysis of historical and precious samples is the non-invasive identification of the material composition. A very efficient method of analysis should enable the extraction, the identification, and the quantitative estimation of major and trace components leaving unchanged the original sample.

In order to simulate the real cases, the panel of different supports, binders and colorants was prepared following the ancient recipes used by the most important artists27-29. Replicas of canvas (a), painting on wood (b), fresco mural painting (c) “secco” mural painting (d) and supports of bovine bone (e) and parchment (f) were prepared (figure 3).

a) The canvas support was prepared with linen, rabbit skin glue 1:14 in water and gypsum ground, while the binder was fresh whole egg. b) The tempera painting was prepared using a wood table spread with bovine glue, linen with rabbit skin glue 1:14 in water and gypsum ground, while the binder was fresh whole egg. c) The fresco mural painting was prepared with lime paste, large and fine granules of sand, red ochre, ultramarine blue and yellow ochre. The colors were spread in lime and protein binders on the wet plaster. d) The “secco” mural painting was also prepared by drying the intonaco before the application of the color. For the fresco rabbit skin glue (1:14 in water) and ammonium caseinate (15% milk

casein + 82% of water and 3% of NH4OH) were employed as binders. For the “secco” mural painting egg yolk and egg white were used as binders. e) The bovine bone was a scapula that was carefully washed and heated for 10 hours at 65°C in an oven. f) The parchment sample was from bovine and the linen was pure.

In order to evaluate the performance of the methods described in the previous sections for proteomic analysis, all these simulated objects were analyzed according to the protocols reported in the paragraph 1.2.4.2.

Table 1 lists the proteins that could be extracted and identiﬁed via MS/MS. It can be appreciated that a fairly large number of proteins could be recognized with high conﬁdence, those characterized by a larger abundance obviously with a larger number of peptides (for instance as

24 many as 42 peptides for ovotransferrin and 20 for ovalbumin for the binders containing egg), and a minimum of two peptides for the species present at low-abundances (e.g., cystatin, vitelline membrane outer layer protein 1, lysosomal-traﬃcking regulator, and interferon gamma for painting on wood), typically identified with 2 peptides.

Interestingly, even from an old and dry bovine bone as many as 26 different proteins could be extracted in sufficient amounts to permit their characterizations via mass spectrometry. Even more strikingly, one protein (collagen alpha-1(I) chain) with a good Mascot score of 644 could be extracted from a bovine parchment and authenticated with 11 peptides. This nondestructive analysis compares favorably with the data of Toniolo et al.3 where eight proteins (four types of collagen, type VI, alpha 3-like isoform 3, alpha-1(XIV) chain- like, alpha-2(I) chain precursor, alpha-1(I) chain precursor, mimecan, histone H2A type 3, protein disulfide isomerase, and tubulin alpha-1C chain) could be identified in a 800 year old parchment, however at the cost of destroying and digesting a fragment of this valuable item.

In the fresco mural painting, the method was able to identify both the collagen proteins (collagen alpha-1(I) chain and collagen alpha-2(I) chain) and the casein proteins (alpha-S1-casein and beta-casein). Moreover, as reported in Table S1, the LC−MS analysis on the “secco” mural painting samples was able to discriminate between the use of egg yolk (vitellogenin-2) and egg white (lysozyme C, ovalbumin, and ovotransferrin) as binders.

The suitability of the EVA film for the quantitative analysis was evaluated for both proteins and small molecules. The validation of the method was carried out on Bovine Serum Albumin (BSA) and Ovoalbumin from egg tempera (to simulate the real case) for the proteins, and on carminic acid, alizarin and indigotin for the small molecules. The solutions of the investigated binders and colorants were spread on different areas of 1x1 cm size on a painting on wood.

Figure 3. Preparation of the painting on linen canvas with fresh egg yolk and pigments, following ancient recipes.

Table 1. Identification of proteins in canvas, painting on wood, fresco mural painting, “secco” mural painting, bovine bone and parchment by strip extraction and trypsin digestion.

SAMPLES PIGMENTS BINDERS PROTEINS (Mascot score) Accession name No. of peptides

Canvas Ultramarine Egg white and Ovalbumin (4157) OVAL_CHICK 20 (linen canvas with blue yolk Ovomucoid (3546) IOVO_CHICK 15 rabbit skin glue Ovotransferrin (3531) TRFE_CHICK 42 1:14 in water and Vitellogenin-1 (2290) VIT1_CHICK 14 gypsum ground) Vitellogenin-2 (1998) VIT2_CHICK 18 Ovoinhibitor (1459) IOV7_CHICK 10 Collagen alpha-2(I) chain (1455) CO1A2_BOVIN 20 Collagen alpha-1(I) chain (1096) CO1A1_BOVIN 21 Ovalbumin-related protein Y (1004) OVALY_CHICK 13 Serum albumin (733) ALBU_CHICK 15 Lysozyme C (636) LYSC_CHICK 5 Ovalbumin-related protein X (Fragment) (128) OVALX_CHICK 2 Riboflavin-binding protein (103) RBP_CHICK 3 Painting on wood Yellow ochre Egg white and Collagen alpha-1(I) chain (10005) CO1A1_BOVIN 68 (wood table yolk Collagen alpha-2(I) chain (7621) CO1A2_BOVIN 59 spread bovine Collagen alpha-1(III) chain (4849) CO3A1_BOVIN 35 glue, linen with Collagen alpha-1(I) chain (3540) CO1A1_CHICK 26 rabbit skin glue Ovalbumin (4595) OVAL_CHICK 18 1:14 in water and Ovotransferrin (4021) TRFE_CHICK 44 gypsum ground) Vitellogenin-2 (2969) VIT2_CHICK 23 Ovomucoid (2742) IOVO_CHICK 12 Vitellogenin-1 (2717) VIT1_CHICK 16 Lysozyme C (1856) LYSC_CHICK 9 Ovalbumin-related protein Y (1203) OVALY_CHICK 14 Ovoinhibitor (1178) IOV7_CHICK 14

Serum albumin (672) ALBU_CHICK 14 Ig lambda chain C region (431) LAC_CHICK 2 Apolipoprotein A-I (179) APOA1_CHICK 2 Ovalbumin-related protein X (Fragment) (141) OVALX_CHICK 4 Riboflavin-binding protein (97) RBP_CHICK 2 Cystatin (78) CYT_CHICK 2 Fresco mural Red ochre Rabbit skin Collagen alpha-1(I) chain (394) CO1A1_BOVIN 11 painting glue 1:14 in Collagen alpha-2(I) chain (268) CO1A2_BOVIN 7 (lime paste, large water and fine granules Ultramarine Ammonium Alpha-S1-casein (61) CASA1_BOVIN 2 of sand, binders blue caseinate Beta-casein (34) CASB_BOVIN 2 and pigments on (15% milk wet) casein + 82% of water + 3% NH4OH) “secco” mural Red ochre Egg yolk Vitellogenin-2 (86) VIT2_CHICK 3 painting (lime paste, large Yellow ochre Egg white Lysozyme C (83) LYSC_CHICK 2 and fine granules Ovalbumin (42) OVAL_CHICK 1 of sand, binders Ovotransferrin (21) TRFE_CHICK 1 and pigments on dry) Bovine bone / / Serum albumin (7624) ALBU_BOVIN 53 (washed + 10 Creatine kinase M-type (4126) KCRM_BOVIN 29 hours at 65 °C in Triosephosphate isomerase (3537) TPIS_BOVIN 21 oven) Beta-enolase (2708) ENOB_BOVIN 26 Serotransferrin (2248) TRFE_BOVIN 28 Alpha-2-HS-glycoprotein (613) FETUA_BOVIN 8 Alpha-1-antiproteinase (563) A1AT_BOVIN 7 Phosphoglycerate kinase 1 (499) PGK1_BOVIN 9 Carbonic anhydrase 3 (466) CAH3_BOVIN 5

Alpha-1-acid glycoprotein (430) A1AG_BOVIN 2 Phosphoglycerate mutase 2 (425) PGAM2_BOVIN 8 Actin, cytoplasmic 1 (392) ACTB_BOVIN 6 Lumican (347) LUM_BOVIN 7 Adenylate kinase isoenzyme 1 (331) KAD1_BOVIN 11 Phosphoglucomutase-1 (280) PGM1_BOVIN 6 Alpha-1B-glycoprotein (278) A1BG_BOVIN 3 Pigment epithelium-derived factor (241) PEDF_BOVIN 3 Malate dehydrogenase, cytoplasmic (221) MDHC_BOVIN 6 Vitamin D-binding protein (206) VTDB_BOVIN 2 Protein DJ-1 (197) PARK7_BOVIN 4 Hemopexin (197) HEMO_BOVIN 6 Transthyretin (193) TTHY_BOVIN 3 Kininogen-1 (192) KNG1_BOVIN 3 Phosphatidylethanolamine-binding protein 1 PEBP1_BOVIN 5 (185) FIBA_BOVIN 2 Fibrinogen alpha chain (180) CO3_BOVIN 4 Complement C3 (180) A2MG_BOVIN 3 Alpha-2-macroglobulin (179) PRDX6_BOVIN 1 Peroxiredoxin-6 (174) COF1_BOVIN 2 Cofilin-1 (162) GELS_BOVIN 2 Gelsolin (156) HSP7C_BOVIN 3 Heat shock cognate 71 kDa protein (148) CFAB_BOVIN 3 Complement factor B (142) TYB4_BOVIN 2 Thymosin beta-4 (136) FABP4_BOVIN 3 Fatty acid-binding protein, adipocyte (135) CO1A2_BOVIN 3 Collagen alpha-2(I) chain (128) KCRB_BOVIN 1 Creatine kinase B-type (101) FETUB_BOVIN 3 Fetuin-B (95) CO1A1_BOVIN 5 Collagen alpha-1(I) chain (95) APOA1_BOVIN 2 Apolipoprotein A-I (90) PGCA_BOVIN 7

Aggrecan core protein (80) TAGL2_BOVIN 3 Transgelin-2 (72) THIO_BOVIN 4 Thioredoxin (25) Bovine / / Collagen alpha-1(I) chain (644) CO1A1_BOVIN 11 Parchment Collagen alpha-2(I) chain (573) CO1A2_BOVIN 10

1.2.5.2 Proteins and colorants calibration curves and recoveries

The calibration curve for Bovine Serum Albumin (BSA) was carried out by spreading 100 µL of BSA solution at 1.00, 0.50, 0.10, 0.05 and 0.01µg/mL, on five different areas of 1x1 cm size on a painting on wood. The calibration curve for egg tempera (fresh egg incorporated with the pigment Ultramarine Blue) was carried out by spreading 100 µL of the solution at 1.000, 0.250, 0.125, 0.025 and 0.003 ppm dilutions from fresh egg, on five different areas of 1x1 cm size on a painting on wood.

The calibration curve of three common used red dyes, namely carminic acid, alizarin and indigotin, was carried out by spreading 100.0 µL of each solution at 100.0, 50.0, 10.0, 5.0, 2.5 and 1.0 µg/mL on different areas of 1x1 cm size on a painting on wood. The protein and colorant extractions for the calibration curves and from the historical frescoes were carried out with 1x1 cm strips. For each analyte a calibration plot reporting the peak area of the “quantifier” transition signal (y) versus the standard concentration (x) was built. Five concentration levels in the range between 0.005-1.000 ug mL-1 for BSA and 0.003-1.000 dilution from fresh egg for Ovoalbumin, and six concentration levels in the range between 1.00-100.00 µg mL-1 for carminic acid, alizarin and indigotin were considered.

Moreover, to overcome possible memory effects, the standard solutions were injected in randomized order. For all the analytes a linear regression fit with a weighting factor 1/x was used, and a good linearity was obtained: the regression coefficients were 0.9866 for BSA, 0.9959 for Ovoalbumin, 0.9989 for carminic acid, 0.9983 for alizarin and 0.9927 for

31 indigotin. The limit of detection (LOD) was calculated as the concentration of the analyte that gives a signal (peak area) equal to the average background (Sblank) plus three times the standard deviation sblank of the blank (LOD = Sblank + 3sblank), while the limit of quantification LOQ was calculated as LOQ = Sblank + 10sblank. LOD and LOQ for the proteins were respectively 0.001 µg mL-1 and 0.004 µg mL- 1, for BSA and Ovoalbumin. As concerns the dyes, LOD and LOQ were 0.30 µg mL-1 and 1.01 µg mL-1 for carminic acid, 0.28 µg mL-1 and 0.93 µg mL-1 for alizarin and 0.46 µg mL-1 and 1.54 g µmL-1 for indigotin, as reported in Table 2.

The LODs and LOQs of the method are very good considering that the extraction of the analytes was performed on a painting on wood and that the egg tempera sample simulates a real case. The inter-day precisions on the concentration were evaluated by analyzing the analytes every day (five replicates) for six days. The results show that the inter-day precisions are 6.88% and 7.32% for BSA and Ovoalbumin and 8.37%, 6.32% and 9.71% for carminic acid, alizarin and indigotin respectively.

To evaluate the recovery R(%) of each analyte and to investigate its possible dependence on the concentration, the analyte standard solutions at different concentration levels used for the calibration curve were analyzed. The recovery values were calculated as Cobs/Cref where

Cobs is the difference between the concentration determined for the spiked sample and the native concentration in the same sample, and Cref is the spiked concentration. Table 3 reports the average percentage of recovery R(%) calculated for all analytes. For all the analytes an average percentage of recovery R(%) was calculated, which is reported in Table 3.

As it can be observed the R (%) values for BSA ranges from 3.97% to 9.36% while the R (%) values for Ovoalbumin ranges from 4.01 % to 11.31%. The recovery of the carminic acid ranges from 0.010% to 0.83%, for alizarin ranges from 0.15% to 2.23% and for indigotin ranges from 0.03% to 2.10%. It may be remarked that the low but reproducible recovery is particularly indicated for the heritage field, because the extraction method can be considered non-invasive, especially for the small molecules, which often represent the most interesting substances in cultural heritage materials.

Table 2. Calibration Curve Results for the Quantiﬁcation of Proteins and Colorants.

SAMPLES LINEARITY RANGE (μg/mL) R2 LOD (μg/mL) LOQ (μg/mL) RSD % conc. interday (n = 30) bovine serum albumin 0.005−1.000 0.9866 0.001 0.004 0.004 ovalbumin (fresh egg) 0.003−1.000 (dilution from fresh 0.9959 0.001 0.004 0.004 egg) carminic acid 1.01−100.00 0.9989 0.30 1.01 1.01 alizarin 0.93−100.00 0.9983 0.28 0.93 0.93 indigotin 1.54−100.00 0.9927 0.46 1.54 1.54

Table 3. Recovery % for the Carminic acid, Alizarin, Indigotin, Bovine Serum Albumin and Ovoalbumin.

Concentration Carminic acid Alizarin Indigotin (µg/mL) Recovery (%) Recovery (%) Recovery (%)

1.00 0.83 ± 0.07 2.23 ± 0.65 0.03 ± 0.01

2.50 0.39 ± 0.04 1.01 ± 0.04 0.12 ± 0.01

5.00 0.18 ± 0.01 0.20 ± 0.05 0.20 ± 0.03

10.00 0.13 ± 0.02 0.18 ± 0.03 0.37 ± 0.05

50.00 0.10 ± 0.01 0.16 ± 0.01 0.70 ± 0.05

100.00 0.010 ± 0.001 0.15 ± 0.01 2.10 ± 0.05

Bovine Serum Albumin Ovalbumin (Fresh egg)

Concentration Recovery (%) Concentration Recovery (%) (µg/mL) (dilution from fresh egg)

0.010 8.80 ± 0.60 0.003 4.01 ± 0.50

0.050 6.72 ± 0.41 0.025 8.55 ± 0.54

0.100 3.97 ± 0.25 0.125 5.63 ± 0.31

0.500 5.41 ± 0.39 0.250 7.98 ± 0.45

1.000 9.36 ± 0.78 1.000 11.31 ± 0.99

Moreover, the ability of detecting proteins and small molecules at the sub-nanomolar and micromolar concentrations (140 picomolar for BSA and 2.03 micromolar for carminic acid) on a surface is very encouraging to explore the methodology for detecting other biological and organic substances on the surface of cultural heritage objects.

1.2.5.3 Non-invasiveness of the method

Two fundamental aspects of this novel methodology were successively investigated, namely that the EVA film, once applied to a given object, would not leave resin residues on its surface or imbedded into its fibers and also, in the case of frescos or paintings, that it would not alter precious material, thus permanently damaged it. To that aim, the characterization of the strip surface before (Figure 4a) and after (Figure 4b) its use for the extraction of proteins from a fresco was performed by Environmental Scanning Electron Microscope (ESEM): as it can be noted, no visible residues from the fresco appear on the strip surface. Moreover, the parchment surface before (figure 4c) and after (Figure 4d) was also investigated with the ESEM microscope in order to detect possible residues that may have been deposited on it by the EVA film. Figure 4e shows the parchment sample with some chromatographic beads deposited on its surface. The ESEM images confirm that the strip did not leave any residues on the parchment support. A light-emitting diodes (LEDs) multispectral imaging system was employed for the acquisition of multispectral images in order to detect possible changes or damages of the parchment surface. The sample was illuminated by LEDs generating different spectral bands while a monochrome camera captured the reflected light. Two identical LED panels (Equipoise Imaging LLC, MD,

USA), each consisting of 8 different LEDs (370, 446, 466, 498, 521, 600, 640, and 923 nm), were employed. A CCD camera with an 8.3-megapixel Kodak CCD monochrome sensor array (ST-8300M, Santa Barbara Instrument Group, CA, USA), 17.96 × 13.52, with 3326 × 2504 pixels, 5.4 × 5.4 μm was employed for the images acquisition30,31. The system, which measures quantitatively the reflectance spectra of a surface, was performed on a parchment sample colored with carminic acid. The imaging analysis of the same area before and after the color extraction with the EVA film showed that the reflectance spectra did not change after the application of the method (Figure 5).

Figure 4. ESEM images of the strip before (a) and after (b) the extraction on fresco and ESEM images of the parchment sample before (c) and after (d) the extraction with the EVA strip. Figure (e) shows the parchment sample with some beads deposited on the surface. It can be noted that the extraction method is not invasive because on the strip are not visible residues from the fresco and the strip did not leave any residues on the parchment support.

Figure 5. Reflectance spectra of a parchment area colored with carminic acid before (black dot line) and after (orange line) the colorant extraction with the EVA film. The reflectance of the surface among the UV-Vis-NIR spectra did not change.

All these data, together with the low recovery measured for both proteins and small molecules, confirmed the non-destructivity of the analytical method developed here and that the analyzed surface was left unchanged, in a detectable way, by the sampling procedure.

1.2.5.4 Method comparison: EVA vs micro-sampling analysis The performances of the EVA film method were compared with a traditional procedure that requires a micro sampling from the object. Two micro samples of about 4 mg were collected from the model fresco mural painting and from the model canvas. Then proteins were denatured in AMBIC (200 L) with 15 L TFE at 60 °C, reduced with 100 L DTT 200.0 mM, alkylated with 100 L IAM 200.0 mM and digested with 10 L of trypsin overnight. The supernatant was then recovered for the LC-MS analysis

The protein extraction from fresco mural painting and canvas were compared: table 4 reports the proteins identified in the samples using the two methodologies. The invasive procedure allowed the identification of the main five milk proteins (Alpha-S1-casein, Beta-casein, Kappa-casein, Alpha-S2-casein, Beta-lactoglobulin) while the EVA method was able to identify only Alpha-S1-casein and Beta-casein. The most important aspect is that the invasive method left a 4x4 mm hole on the surface of the fresco, while using the EVA film the artwork was not damaged. Moreover, the identification of the two milk proteins extracted with the EVA is sufficient to detect the binder. The lower performance of the EVA method is due to the fact that in the mural painting the binders are embedded into the intonaco, therefore the extraction is more difficult. However, the protein extraction from canvas allowed the identification of all the egg (Ovalbumin, Ovomucoid, Ovotransferrin, Vitellogenin-1, Vitellogenin-2, Ovoinhibitor, Ovalbumin-related protein Y, Serum albumin, Lysozyme C, Ovalbumin-related protein X and Riboflavin-binding protein) and glue (Collagen alpha-2(I) chain and Collagen alpha-1(I) chain) proteins. Nevertheless, with the invasive analysis we were able also to identify other less abundant egg and glue proteins.

Table 5 shows that also small molecules can be extracted and quantified from different materials, such as parchment, linen, and painting on wood. In this particular case, carminic acid dye could be assessed even at rather low amounts, such as ng/μL.

Table 4. Methods comparison: identification of proteins in fresco mural painting and canvas by strip extraction and by traditional micro sampling.

PROTEINS (Accession name and mascot score) SAMPLE BINDER EVA film Micro sampling

Fresco mural Ammonium CASA1_BOVIN (44) CASA1_BOVIN (3197) painting caseinate (15% CASB_BOVIN (33) CASB_BOVIN (1859) milk casein + CASK_BOVIN (835) 82% of water + CASA2_BOVIN (753) 3% NH4OH) LACB_BOVIN (459) MFGM_BOVIN (86) GLCM1_BOVIN (82) TRFL_BOVIN (78) SG1D_BOVIN (54) CDC5L_BOVIN (39) XDH_BOVIN (33) PLCA_BOVIN (28) BT1A1_BOVIN (27) ARL11_BOVIN (23) CO3_BOVIN (20) LONM_BOVIN (19) PIGR_BOVIN (19) PNML1_BOVIN (18) CD36_BOVIN (18) DRS7C_BOVIN (17) Canvas Egg white and OVAL_CHICK (4157) VIT2_CHICK (5706) (linen canvas yolk IOVO_CHICK (3546) OVAL_CHICK (5061) with rabbit TRFE_CHICK (3531) CO1A1_BOVIN (3741) skin glue VIT1_CHICK (2290) CO1A1_CHICK (1577) 1:14 in water VIT2_CHICK (1998) TRFE_CHICK (3142) and gypsum IOV7_CHICK (1459) CO1A2_BOVIN (2842) ground) CO1A2_BOVIN (1455) VIT1_CHICK (2174) CO1A1_BOVIN (1096) ALBU_CHICK (1015) OVALY_CHICK (1004) OVALY_CHICK (860) ALBU_CHICK (733) LYSC_CHICK (763) LYSC_CHICK (636) APOV1_CHICK (671) OVALX_CHICK (128) TENP_CHICK (360) RBP_CHICK (103) CO3A1_BOVIN (215) IOVO_CHICK (313) LAC_CHICK (235) MUC5B_CHICK (202) OVALX_CHICK (197) APOB_CHICK (144) OVOS_CHICK (128) IOV7_CHICK (69) CYT_CHICK (62)

RBP_CHICK (53) VMO1_CHICK (43) RM01_BOVIN (37) EXFAB_CHICK (24)

Table 5. Identification and quantification of carminic acid in parchment, linen and painting on wood by strip extraction and LC-MS TRANSITIONS CE (V) Concentration SAMPLES DYE (Q1 > Q3) (ng/μL)

491.0835 > 327.0516 -40 Parchment Carminic acid 37.13 ± 2.50 491.0835 > 447.0940 -30 (Bovine parchment) 491.0835 > 357.0622 -40

491.0835 > 327.0516 -40 Linen Carminic acid 14.30 ± 1.15 491.0835 > 447.0940 -30 (Pure linen canvas) 491.0835 > 357.0622 -40

Painting on wood 491.0835 > 327.0516 -40 (wood table spread bovine glue, linen with Carminic acid 8.28 ± 0.68 491.0835 > 447.0940 -30 rabbit skin glue 1:14 in water and gypsum 491.0835 > 357.0622 -40 ground)

1.2.5.5 Test with varnish samples

Since the most of wood panels and canvases are varnished, during the validation experiments, it was also demonstrated the efﬁcacy of the ﬁlm- functionalized method on varnished paintings and on a wide range of different supports, in order to apply the EVA method to historical objects2,32.

Although the varnish is usually employed to protect the surface from dirt, dust and smoke, or for aesthetic purposes, it can be replaced along time, during restoration or conservation procedures, and its thickness can fluctuate depending on the substrate or its oxidation.33

The efficacy of the non-invasive method was tested on varnished canvas samples, made with collagen as binder. Animal glue, egg white and yolk were deposited on a canvas and, once dried, the following varnishes were spread on different areas, in order to simulate the most commonly used protective films adopted in old paints34. The following varnishes were tested: Shellac, natural Mastic of Chios, Copal and Dammar (all purchased from Zecchi, Firenze). Shellac is constituted by a resin of animal origin secreted by a female lac bug. In art it has been mainly used for finishing wooden artwork until XX century. The natural Mastic of Chios is a brilliant natural resin obtained from the plant Pistacia lentiscus, originally from a Greek island. It has been used in paintings since antiquity for the preparation of oil paints and protective varnishes. Copal resin is made by a subfossil vegetal resin insoluble in the most common solvents, while Dammar is constituted by a vegetal resin derived from dipterocarpaceae plant.35,36

These four resins were selected on the base of ancient recipes and uses: since the 16th century they were employed mixed with binding media or spread on the painting to protect and preserve the artworks. These varnishes are also widely employed by restorers for conservation purposes34,37.

The application of the method to a canvas support painted with several types of varnishes showed that the technique is able to adsorb low amounts of binders under the ﬁnish, without damaging the surface.

As shown in table 4, the method was able to identify the binders underneath all varnishes. Table 6 lists the proteins identified in the different samples. In all the samples prepared with animal glue, the most abundant protein of the binder Collagen alpha-1(I) was identified, with a large number of peptides and a high percentage of coverage. The type of animal glue (different animal, different preparation, etc.) may influence the distribution of the different types of collagens: i.e. in this application only Collagen alpha 1(I) chain was detected and identified. This also indicates that no artifacts or contaminations are present in the workflow.

The method was also able to identify egg-based binders: several proteins from egg white (i.e. Ovalbumin, Ovotransferrin, Ovomucoid, Ovoinhibitor, Ovalbumin-related protein Y, Lysozyme C, Apovitellenin- 1, etc.) and egg yolk (i.e. Vitellogenin-2, Vitellogenin-1, Ovalbumin, Ovotransferrin, Ovomucoid, Serum albumin, etc.) were identiﬁed.

The varnish layers were about 1 mm thick and no cracks were visually present. Although varnish layer is mainly used to protect paintings, the surfaces are not completely sealed because nanoscopic pores holes inside

44 the varnish may allow the exchange of protein material. Moreover, the brush application of the varnish can be irregular and this can lead to an increase in the oxidation state of the varnish over time, which could generate the formation of deep cracks, facilitating this exchange more easily. Thick varnishes, dust, contaminations and ageing may affect the protein extraction. On the other hand, the sampled area on a real artwork must be chosen together with an expert conservator and the regions with a clear and thick varnish layer should be avoided when possible.

Table 6. Binders identiﬁed under different varnishes: Shellac (SH 2), Mastic of Chios (CH 1), Copal (CO 3) and Dammar (DA 4). The table reports the protein name and its Mascot score, the accession name, the number of identiﬁed peptides and the protein coverage.

Binder Varnish PROTEINS (Mascot score) Accession No. of Protein coverage % name peptides Animal glue SH 2 Collagen alpha-1(I) chain (11335) CO1A1_BOVIN 83 64 Animal glue CH 1 Collagen alpha-1(I) chain (8658) CO1A1_BOVIN 72 64,5 Animal glue CO 3 Collagen alpha-1(I) chain (9291) CO1A1_BOVIN 55 58 Animal glue DA 4 Collagen alpha-1(I) chain (8833) CO1A1_BOVIN 65 64,3 Egg white SH 2 Ovotransferrin (2301) TRFE_CHICK 45 68,1 Ovalbumin (1935) OVAL_CHICK 18 60,1 Ovomucoid (1625) IOVO_CHICK 15 69,5 Ovoinhibitor (966) IOV7_CHICK 17 47,2 Ovalbumin-related protein Y (663) OVALY_CHICK 15 42,3 Lysozyme C (559) LYSC_CHICK 12 76,9 Egg white DA 4 Ovalbumin (1369) OVAL_CHICK 14 56,5 Ovotransferrin (1223) TRFE_CHICK 27 49,6 Vitellogenin-2 (769) VIT2_CHICK 20 14,6 Ovomucoid (603) IOVO_CHICK 8 53,8 Ovoinhibitor (438) IOV7_CHICK 9 27,1 Ovalbumin-related protein Y (390) OVALY_CHICK 10 28,1 Lysozyme C (379) LYSC_CHICK 10 74,8 Apovitellenin-1 (250) APOV1_CHICK 5 51,9 Egg white CO 3 Ovalbumin (1478) OVAL_CHICK 17 64,5 Ovomucoid (892) IOVO_CHICK 12 53,8 Ovotransferrin (860) TRFE_CHICK 19 38,7 Vitellogenin-1 (806) VIT1_CHICK 19 13,3 Lysozyme C (464) LYSC_CHICK 7 69,4 Ovalbumin-related protein Y (392) OVALY_CHICK 10 35,6 Ovoinhibitor (315) IOV7_CHICK 8 25,4 Apovitellenin-1 (143) APOV1_CHICK 3 33 Egg white CH 1 Ovalbumin (1614) OVAL_CHICK 15 56,5 Ovotransferrin (1050) TRFE_CHICK 22 40,7 Ovomucoid (800) IOVO_CHICK 10 57,1 Vitellogenin-2 (616) VIT2_CHICK 18 14 Ovalbumin-related protein Y (426) OVALY_CHICK 12 34 Lysozyme C (285) LYSC_CHICK 6 59,9 Ovoinhibitor (236) IOV7_CHICK 5 17,8 Egg yolk SH 2 Vitellogenin-2 (4688) VIT2_CHICK 77 47,2 Vitellogenin-1 (2617) VIT1_CHICK 54 33,4

Ovalbumin (1154) OVAL_CHICK 17 50,5 Ovotransferrin (725) TRFE_CHICK 17 27,8 Ovomucoid (496) IOVO_CHICK 9 52,4 Serum albumin (488) ALBU_CHICK 14 29,1 Lysozyme C (370) LYSC_CHICK 6 59,2 Ovalbumin-related protein Y (290) OVALY_CHICK 8 30,4 Ovoinhibitor (172) IOV7_CHICK 5 15,7 Egg yolk DA 4 Vitellogenin-2 (5888) VIT2_CHICK 86 50,3 Vitellogenin-1 (3392) VIT1_CHICK 64 37,9 Ovalbumin (1114) OVAL_CHICK 14 50,8 Ovotransferrin (691) TRFE_CHICK 16 34,8 Ovomucoid (627) IOVO_CHICK 9 50 Serum albumin (596) ALBU_CHICK 14 30,2 Ovoinhibitor (297) IOV7_CHICK 6 26,1 Apovitellenin-1 (291) APOV1_CHICK 6 59,4 Ovalbumin-related protein Y (286) OVALY_CHICK 9 24,2 Lysozyme C (268) LYSC_CHICK 6 59,9 Apolipoprotein B (Fragment) (231) APOB_CHICK 8 21,2 Egg yolk CO 3 Vitellogenin-2 (1670) VIT2_CHICK 37 23,7 Vitellogenin-1 (435) VIT1_CHICK 10 6,8 Ovomucoid (425) IOVO_CHICK 6 35,7 Ovalbumin (181) OVAL_CHICK 3 11,1 Serum albumin (141) ALBU_CHICK 5 7 Egg yolk CH 1 Vitellogenin-2 (9305) VIT2_CHICK 103 55,6 Vitellogenin-1 (5936) VIT1_CHICK 87 49,1 Serum albumin (1098) ALBU_CHICK 25 46 Apovitellenin-1 (896) APOV1_CHICK 9 65,1 Ovotransferrin (793) TRFE_CHICK 21 35,7 Ovomucoid (592) IOVO_CHICK 12 54,3 Ig lambda chain C region (545) LAC_CHICK 4 71,8 Ovoinhibitor (503) IOV7_CHICK 12 32,8 Apolipoprotein B (Fragment) (485) APOB_CHICK 12 27,5 Lysozyme C (469) LYSC_CHICK 10 68 Ovalbumin-related protein Y (297) OVALY_CHICK 8 25 Vitellogenin-3 (Fragments) (277) VIT3_CHICK 8 18,2 Apolipoprotein A-I (241) APOA1_CHICK 10 36,4

1.2.6 Application of the method to historical objects After the fully validation of the method performed in our laboratory, its effectiveness was tested on real samples, thanks to our collaboration with several National and International Institutions and Museums, like the Egyptian Museum of Turin, Soprintendenza dei Beni Culturali e Ambientali of Piemonte and Provincia di Parma, Museo Civico di Antichità of Turin (collection of Palazzo Madama, Turin), Parella Castle, St. Petersburg Hermitage Museum, Laboratorio Arvedi di Diagnostica non Invasiva CISRiC – from University of Pavia and Violin Museum of Cremona. Several historical samples were investigated in order to identify the binders used by the artist or the materials that constituted the artwork, but also to test the ability of the method on a wide range of real old supports, such as important historical objects, mummies and archeological finds, historical musical instruments, frescoes, canvas, parchments and wood panel.

1.2.6.1 Application of the EVA method to Historical Frescoes

An important aspect of the EVA film technology is its ability to extract proteins and other molecules from ancient frescoes without damaging their surface23. We selected two frescoes of the XVI century with a “secco” details, whose images are shown in Figure 6A (Flemish painter Paul Brill (1554-1626)) and 6B (a recently discovered fresco from Isidoro Bianchi (1581-1662)). The two frescoes are situated in the Parella castle in Italy (XIII century), which is decorated with precious allegorical frescoes recently attributed to the baroque painter Isidoro Bianchi and to the

Flemish artist Paul Brill, who painted several artworks conserved today in the Vatican Museums. In both cases, different types of caseins were identified, as reported in Table 7. In the insert of Fig. 6B, where the gloved finger is pointing, one can see, underneath it, the EVA film diskette applied to the surface of the fresco. Even if in the fresco the binder is embedded into the intonaco, and as a consequence the extraction from the support is likely to be more difficult than with other supports (like canvas), the beta-casein protein was identified with three peptides in both the samples. In the fresco from Isidoro Bianchi we also identified Alpha-S1-casein and Kappa-casein with one peptide each.

The peptide GPFPIIV from beta-casein was identified in the two historical frescoes but not in the model one using the non-invasive EVA extraction. But the same peptide was identified using the traditional invasive method. We have to consider that the two frescoes are five hundred years old and the proteins may have undergone degradation. The peptide could be linked to some sort of degradative process. Moreover, the simultaneous presence of several organic and inorganic materials, the environmental and ageing contamination and the low amount of milk used in the fresco preparation may complicate the proteins extraction.

Table 7. Identiﬁcation of Proteins in Frescos from the XVI Century and Identiﬁcation: The Analysis was carried out by the EVA Strip Extraction and LC−MS.

SAMPLE BINDER PROTEINS (MASCOT ACCESSION PEPTIDES (ION SCORE IDENTITY SCORE) NAME THRESHOLD) Fresco I (fresco CASEIN Beta-casein (91) CASB_BOVIN GPFPIIV (36-21) recently VLPVPQK (46-13) discovered under the Alpha-S1-casein (53) CASA1_BOVIN AVPYPQR (22-29) plaster) FFVAPFPEVFGK (53-30)

Beta-casein (71) CASB_BOVIN GPFPIIV (25-24) Fresco II (fresco CASEIN Kappa-casein (32) CASK_BOVIN VLPVPQK (33-18) from the AVPYPQR (26-26) Flemish painter YIPLQYVLSR (35-25) Paul Brill)

Figure 6. Color images of the XVI century fresco from the Flemish painter Paul Brill (a) and of a recently discovered fresco from Isidoro Bianchi (b). In the magniﬁcation at the bottom left of image b the extractive procedure on the fresco and the 1 cm × 1 cm size EVA ﬁlm are visible. Images taken during the analysis at the Parella castle, Ivrea, Turin.

The non-invasive method was also applied directly in situ to investigate a cycle of secco mural paintings from Reggia di Colorno (Parma, Italy)2. In the mural painting of Reggia di Colorno, we selected two points of an important cycle of secco mural paintings (beginning of XIX century) located in the dining room of the Napoleon's wife, Maria Luigia d'Austria: a mural painting with still life scenes (Fig. 7a, sample 1) and a mural painting with allegorical figures (Fig. 7c, sample 2), located in the central vault of the salon. As shown in Table 8, the binders identified in sample 1 are animal glue and casein. The first binder was characterized by Collagen alpha-1(I) with a score of 157, and the second one is represented by Beta- casein with a score of 27. In sample 2 we found a higher concentration of glue and milk proteins of the same binder: Collagen alpha-1(I) with a score value of 635 and 7 peptides, Collagen alpha-2(I), Collagen alpha-1(III), Collagen alpha-1(XVII), Kappa-casein, Alpha-S1-casein, Beta-casein and Alpha-S2-casein.

The analysis of the secco mural paintings from Reggia di Colorno enabled the identification and quantification of different types of binders. Depending on the scene represented in the painting, animal glue and milk proteins are present in different ratios. As shown in Table 8 the relative quantitation through the Exponential Modified Protein Abundance Index (emPAI) was used to compare the two samples analyzed from the same painting. The results evidenced that the concentration of the binders in the central vault is higher than in the decorative one. The emPAI values of the Collagen alpha-1(I) and Beta-casein in sample 2 are exactly twice than in sample 1. According to historians and restorers, this is probably due to the presence of more painted details such as in correspondence

50 of allegorical figures. In fact, in these regions the artist used more colors, and thus more binders, to paint details and figures with respect to areas where a simple brushstroke was sufficient to give color and shape the image.

Fig. 7. Color images of mural painting of Reggia di Colorno with the details during the extraction: “still life scenes” (a) and details (b); “allegorical ﬁgures” (c) and details (d).

Table 8. Relative quantitation of binders in the secco mural painting from the Reggia of Colorno. Sample 2, which was taken in correspondence of an allegorical ﬁgure, showed a higher concentration of binders with respect to sample 1, as shown by the emPAI values. The table reports the protein name and their mascot score, the accession name, the number of peptides, the protein coverage and the emPAI value for each identiﬁed protein

Sample Proteins (Mascot score) Accession No. of peptides Protein coverage % emPAI name Sample 1 “Still life scenes” Collagen alpha-1(I) chain (157) CO1A1_BOVIN 4 4,2 0,13 Beta-casein (27) CASB_BOVIN 2 6,3 0,18 Sample 2 “Allegorical Collagen alpha-1(I) chain (635) CO1A1_BOVIN 7 7,1 0,24 ﬁgures” Kappa-casein (335) CASK_BOVIN 2 14,7 0,49 Alpha-S1-casein (332) CASA1_BOVIN 3 17,3 0,69 Beta-casein (102) CASB_BOVIN 2 18,3 0,4 Collagen alpha-1(III) chain (61) CO3A1_BOVIN 2 4 0,1 Alpha-S2-casein (43) CASA2_BOVIN 1 3,6 0,18 Collagen alpha-1(XVII) chain (38) COHA1_BOVIN 1 0,6 0,03

1.2.6.2 Application of the EVA method to the Collection of Palazzo Madama of Turin The non-invasive method was applied directly in situ to several historical artworks of European provenance conserved at the Civic Museum of Ancient Art, Palazzo Madama (Turin, Italy)2.

In the collection of Palazzo Madama, the following artworks were analyzed:

• a polychrome wooden altarpiece with Episodes from the life of St. Mary Magdalene, by the Maestro of Oropa (half of XIV century, inv. 1053/L) shown in Fig. 8b; • a polychrome sandstone capital with Annunciation, Saints and Angels, by Maestro of San Domenico of Chieri (end of XIV century, inv. 353/PM) shown in Fig. 8e; • a rare polychrome alabaster with the Coronation of the Virgin, from the Novalesa abbey (XV century, inv. 281/PM) shown in Fig. 8f; • a medieval wooden panel representing the Stigmatization of St. Francis, Madonna and Child, the Trinity and Saints by Pietro Gallo from Alba (end of XIV century, inv. 774/D) in Fig. 9a; • a panel from the French artist Jean Bapteur (middle of XV century) in 9d, a Cruciﬁxion, inv. 432/D, a renaissance polyptich by Defendente Ferrari (St. Jerome, inv. 464/D) shown in Fig.9b; • a panel by Martino Spanzotti with Christ on the tomb, with angels (429/D) shown in Fig. 9e, the Madonna and Child with St. Anne by Gerolamo Giovenone (XVI century, 431/D) in Fig. 9c;

• a detached fresco by Antoine de Lonhy (XV century, inv. 428/D) shown in Fig. 8a; • a rare sample of painted leather casket from Parisian manufactory (beginning of XIV century, inv. 4590) shown in Fig. 8d; • a wooden reliquary attributed to the Tuscan painter Buonamico Buffalmacco (beginning of XIV century, inv. 411) shown in Fig. 8c; • a precious parchment manuscript of Francesco Marmitta, that belonged to the cardinal Domenico Della Rovere (about 1462- 1465), shown in figure n 10a; • a Greek vessel (kylix) with black figures (VI century B.C.) shown in figure n 10b.

Figure 8. Color images of: the detached fresco from Antoine de Lonhy, inv. 428/D (a), the wooden altarpiece with Episodes from the life of St. Mary Magdalene, by the Maestro of Oropa (half of XIV century, inv. 1053/L) (b), the wooden reliquary attributed to Buonamico Buffalmacco (beginning of XIV century, inv. 411) (c), leather painted casket from Parisian manufactory (inv. 4590) (d), the polychrome sandstone capital with Annunciation; Saints; Angels, by Maestro of San Domenico of Chieri (end of XIV century, inv. 353/PM (e) and the polychrome alabaster with the Coronation of the Virgin, from the Novalesa abbey (XV century, inv. 281/PM).

Figure. 9. Color images of the wood panels from Palazzo Madama, Turin: Stigmatization of St. Francis, Madonna and Child, the Trinity and Saints by Pietro Gallo from Alba (end of XIV century, inv. 774/D) a); St. Jerome Polyptich by Defendente Ferrari, inv. 464/D (b); the Madonna and Child with St. Anne by Gerolamo Giovenone (XVI century, 431/D) (c) Cruciﬁxion, by Jean Bapteur (middle of XV century) inv. 432/D (d); Martino Spanzotti, Christ on the tomb, with angels (429/D) (e).

Figure 10. Color images of the parchment manuscript of Domenico della Rovere (a), and of the Greek vessel with black figures (b).

The results showed that the method was able to identify one of the most common binders used for the painting on table in medieval and renaissance times: namely the animal glues. This is the case of the wood panel from Pietro Gallo, composed by collagen alpha-2(I), collagen alpha- 1(I) and collagen alpha-1(III). In the same panel, although the identification and extraction of colorants was extremely complicated, the method allowed the non-invasive identification of carminic acid in correspondence of the red bishop cassock, as reported in table 10. While the surface of the wood panel from the artist Defendente Ferrari, is characterized by collagen alpha-1(I) and collagen alpha-2(I), in the wood panel from Gerolamo Giovenone several collagen proteins were identified, as Collagen alpha-1(I), the most abundant protein with a Mascot score value of 6894 and coverage of 46%. On the other hand, in the polychrome wooden table of the miniaturist Jean Bapteur, Crucifixion with a female donor, we investigated two different areas of the same panel: a “figure of warrior under the cross” (Fig. 11a) and a “figure of child” (Fig. 11b). In both subjects, the animal glue was identified as the binder. The protein Collagen alpha-1(I) showed the highest score in both cases, 961 with 18 peptides in the first sample, and 82 with 2 peptides in the second. Additionally, milk was found as second binder but only in sample 11a, with the Beta-casein protein characterized by 2 peptides and a score value of 35. The results were confirmed in a second replicate analysis of the same areas. The identification of casein protein only in sample 11a is probably due to the presence of more details painted in light color in the decoration of the armor of warrior under the cross.

Figure 11. Color images of two particulars of the wood panel from Jean Bapteur, Crucifixion inv. 432/D: scene with “figure of warrior under the cross” (a); scene with “figure of child” (b).

In the case of the wood panel “Christ on the tomb with angels”, painted by Martino Spanzotti, we could appreciate the use of mixed media as revealed by the LC-MS analysis. In fact, we found several glue and milk proteins. The extraction allowed to identify Collagen alpha-1(I) and Col- lagen alpha-1(III) for the glue binder and Alpha-S1-casein, Kappa-casein, Beta-casein, Beta-lactoglobulin and Alpha-S2-casein for the milk one.

The analysis on the sample obtained from the polychrome wooden altarpiece by the Master of Oropa inv. 1053/L, allowed the identiﬁcation of protein Alpha-S1-casein with a score value of 76, so the binder used by the artist was milk.

In the detached fresco from the French artist Antoine de Lonhy, the proteins identiﬁed were Collagen alpha-1(I) and Collagen alpha-1(XVII), the ﬁrst one with a high score of 8968 and a coverage of 54%. The fresco was restored after the detachment and reposition in the frame, we

57 cannot exclude that the presence of collagen could be the result of the recent restoration intervention. In fact, we did not ﬁnd degradation markers of collagen.

Another example of stone painting is the case of a rare, well-conserved, painted sandstone capital from the Italian abbey of San Domenico, in the town of Chieri. Here we are in presence of a case of glue on stone: the binder identiﬁed was glue, characterized by collagen alpha-1(I) with a score of 31. Also the analysis performed on a precious polychrome alabaster from the Novalesa Abbey revealed the presence of animal glue (Collagen alpha-2(I)) as principal binder, in agreement with the historical literature35.

The leather-painted casket from a Parisian manufactory was characterized by the presence of Collagen alpha-1(I) with a score of 6309, and Collagen alpha-1(III), both suggesting the presence of glue binder. The identiﬁed collagen proteins could also be attributed both to the leather support and/or to the medium employed for painting.

Thanks to the application of the EVA method, it was possible to identify the proteins from a parchment page of a precious manuscript created by the illuminated artist Francesco Marmitta for Cardinal Domenico Della Rovere. In this case, a confident assignment of the alpha 1 (I) collagen protein to the animal origin (sheep/lamb) was possible. The bio informatic research in the NCBI database, allowed us to identify three peptides with unique sequence of different amino acids specific of the genus Ovis aries.

In the kylix Greek vessel, the method allowed to identify different egg proteins on the bottom of the jar and collagen proteins near the cracks,

probably due to gluing occurred with the last restoration. The morphology of the vessel and the presence of egg proteins as Ovomucoid and Lysozime C from Gallus gallus suggests that the object was used to prepare or consume food.

Table 10. Proteins identified in each artwork. The table reports the name of the artwork, the proteins identified and their accession name, and the binder used by the artist.

Artwork PROTEINS (Mascot score) Accession name Binder

Wood panel from Collagen alpha-2(I) chain (278) CO1A2_BOVIN Pietro Gallo (XIV century) Collagen alpha-1(I) chain (274) CO1A1_BOVIN Animal glue

Collagen alpha-1(III) chain (169) CO3A1_BOVIN

Wood panel from Collagen alpha-1(I) chain (226) CO1A1_BOVIN Defendente Ferrari Animal glue (XVI century) Collagen alpha-2(I) chain (95) CO1A2_BOVIN

“Crocifissione con Collagen alpha-1(I) chain (961) CO1A1_BOVIN donatrice”, Collagen alpha-1(III) chain (64) CO3A1_BOVIN Mixed sample with “figure of media: warrior under the Collagen alpha-1(XVII) chain (39) COHA1_BOVIN animal glue cross”, wood panel and milk from Jean Bapteur Beta-casein (35) CASB_BOVIN (XV century)

“Crocifissione con donatrice”, sample with “figure of child”, Collagen alpha-1(I) chain (82) CO1A1_BOVIN Animal glue wood panel from Jean Bapteur (XV century)

“Madonna con Collagen alpha-1(I) chain (6894) CO1A1_BOVIN bambino e sant'Anna”, wood Animal glue panel from Gerolamo Collagen alpha-1(XVII) chain (51) COHA1_BOVIN Giovenone (XVI century)

“Cristo nel sarcofago Collagen alpha-1(I) chain (5210) CO1A1_BOVIN e quattro angeli”, wood panel from Collagen alpha-1(III) chain (1046) CO3A1_BOVIN Martino Spanzotti (XV century) Alpha-S1-casein (448) CASA1_BOVIN Mixed media: Kappa-casein (377) CASK_BOVIN animal glue Beta-casein (349) CASB_BOVIN and milk

Beta-lactoglobulin (323) LACB_BOVIN

Alpha-S2-casein (166) CASA2_BOVIN

“Storie di Santa Maria Maddalena”, Polychrome wooden Alpha-S1-casein (76) CASA1_BOVIN Milk altarpiece from Maestro of Oropa (XIV century)

Detached fresco from Collagen alpha-1(I) chain (8968) CO1A1_BOVIN Antoine de Lonhy (XV Animal glue century) Collagen alpha-1(XVII) chain (45) COHA1_BOVIN

“Annunciazione; Santi; Angeli”, polychrome Collagen alpha-1(I) chain (31) CO1A1_BOVIN Animal glue sandstone capital, San Domenico di Chieri (XIV Century).

“Incoronazione della Vergine”, polychrome alabaster, abbey of Collagen alpha-2(I) chain (19) CO1A2_BOVIN Animal glue Novalesa (XIV Century).

Leather casket from Collagen alpha-1(I) chain (6309) CO1A1_BOVIN Leather as Parisian manufactory support and (beginning of XIV Collagen alpha-1(III) chain (223) CO3A1_BOVIN animal glue century)

Wooden Reliquary Collagen alpha-1(I) chain (1243) CO1A1_BOVIN from Buffalmacco Animal glue (beginning of XIV Collagen alpha-1(III) (147) CO3A1_BOVIN century)

Manuscript of Francesco Marmitta Collagen alpha-1(I) (…) CO1A1_OVIS Parchment (about 1462-1465)

Ovomucoid IOVO_CHICK Egg Greek vessel (VI Lysozime C LYSC_CHICK Egg century BC) Collagen alpha-1(I) (…) CO1A1_BOVIN Animal glue

Table 11. Identification of a colorant in wood panel of Pietro Gallo (XIV century). The analysis was carried out by the EVA strio Extraction and LC-MS.

SAMPLES DYE TRANSITIONS CE (V) CONC (μg/μL) (Q1>Q3) Wood panel Carminic 491.0835>327.0516 -40 8.43±1.40 (from Pietro acid 491.0835>447.0940 -30 Gallo) 491.0835>357.0622 -40

1.2.6.3 Application of the EVA method to a Leonardo da Vinci’s painting from Hermitage museum of St. Petersburg The EVA method allowed to investigate, for the first time, the artistic technique of one of the greatest Italian artists: Leonardo da Vinci32.

Although some researches had already been conducted, regarding the various media used for his master wall paintings, no scientific evidences and characterizations of the binding material in his painting on panels and canvas are reported.

While there is a strong interest in understanding the Leonardo's artistic techniques, on the other hand the need of safeguarding his masterpieces limits the possibility of scientific investigations. Analyses carried out on the “Last Supper” during a recent restoration have established that the technique used by Leonardo was a painting in tempera (perhaps mixed with oils) on two layers of preparation, the first being thicker of calcium carbonate, the second thinner in order to make the painting stick based on white lead38-41. The unsatisfactory results were due to several restorations that altered the original materials: from the analyses the authors hypothesized a mixed technique in which the pigments were dispersed in a soluble protein binder, perhaps egg, and in an insoluble binder, oil42.

Although some attempts permitted the identification of part of the materials used by Leonardo, to date no analytical investigations were able to fully characterize the organic constituents of the binding media and to decipher their recipes. This is mainly due to the fact that the most useful and performing analytical techniques adopted up to the present are invasive43, and they require at least a micro sampling,

62 while the information that can be obtained from the non-invasive ones, such as spectroscopy techniques44 or imaging methods45, cannot be complete and exhaustive.

As previously reported in the first part of the thesis (paragraph 1.2.5.3) we have demonstrated that the surface of the artworks does not change, and that no damages or contamination is present. Thus, we had the possibility to investigate in a non-invasive manner the surface of the painting “Donna Nuda” by Leonardo (Hermitage museum, St, Petersburg), in order to ascertain the techniques used in its drawing32. The painting (canvas transferred from wood panel, dimensions 86,5 × 66,5 cm, inventory number Э-110; collection: European Fine Art; Sub- collection: Italian Painting of the 13th–18th Centuries) was acquired from the R. Walpole collection, Houghton Hall, England, in 1779 by Catherine II for her Hermitage residence in St. Petersburg, as part of the paintings collection for the palace decoration. Until 1917 the Hermitage was Russian Tsars residence and palace. After the Great October Revolution in 1917, the Hermitage was transformed into a Government Museum.

To characterize and decipher the complex original recipes used by Leonardo, mainly composed by resins and oils, as reported by Vasari in Le Vite (1568)46, we decided to extract metabolites and lipids from the surface of the painting using our EVA functionalized film. For the first time, untargeted metabolomics and lipidomics were used to deeply investigate the chemical composition of the materials employed for painting the “Donna nuda”, as reported in Fig. 12

Figure 12. The painting “Donna Nuda” from the Hermitage museum in St. Petersburg. The red squares and white numbers refer to the five EVA diskettes applied in the different zones of the canvas.

After the EVA extraction from the surface of the painting, the molecules were eluted from the film and then prepared using the different methodologies described in the paragraph 1.2.4.2 for metabolomics and lipidomics analyses.

As regards the lipidomic approach, the following lipid standards were also used to identify the lipids classes: 1,3(d5)-diheptadecanoyl-2-(10Z- heptadecenoyl)-glycerol (TAG(17:0/17:1/17:0)); 1,2-dihepta-decanoyl-

64 sn-glycero-3-phosphoethanolamine (PE(17:0/17:0)); 1-dodecanoyl-2- tridecanoyl-sn-glycero-3-phosphocholine (PC(12:0/ 13:0)). All the standards were purchased from Sigma (Sigma-Aldrich, Inc., St. Louis, MO, USA).

The numbers one to five (see the painting in figure 12) refer to the five EVA films (about 2.5 × 1.5 cm in size) placed for 15 minutes in contact with the 5 different positions on the surface of the painting. The main components identified by GC–MS were fatty acids, which are originally derived from triglycerides. Their distribution indicates that the major compounds are myristic, palmitic, stearic, oleic and arachidic acids together with small amounts of dicarboxylic acids, as shown in Tables 12 and 13. Below are reported some examples of the GC–MS analyses of eluates from the various EVA films applied to the canvas (numbered 1 to 5 in Fig. 12).

Table 12. Artistic techniques, ingredients, identified molecular markers and analytical methods used for the analysis of each sample. The “TEMPERA GRASSA” is composed by linseed oil and egg yolk; the Venetian turpentine is mainly composed by conifer resin, while for the preparation of the essential oil a rosemary extract was used.

Table 13. Measured retention index, tabulated retention index, main m/z and software similarity of the main markers identified in the samples using the GC–MS analysis

The presence of “tempera grassa” was confirmed by the analysis of the eluates of EVA films 2, 3 and 5, while conifer resin was identified from EVA films 1, 4 and 5. Rosemary oil was clearly identified in EVA film 1 (Table 12) Fig. 13 shows the total ion chromatogram of EVA film 1 (refer to Fig. 12 for positioning of the film on the canvas) as analyzed by GC– MS. The main fatty acid molecular markers, namely myristic acid, pentadecanoic acid, palmitic acid, heptadecanoic acid and stearic acid are reported. Fig. 14 reports the total ion chromatogram of the same EVA film 1 eluate analyzed with GC–MS. The main rosemary oil markers, namely α-Pinene, D-Limonene, (+)-2-Bornanone, D-Pinitol and Borneol are identified.

Figure 13. Total ion chromatogram of the EVA film 1 analyzed with GC–MS. The main fatty acid markers, namely myristic acid, pentadecanoic acid, palmitic acid, heptadecanoic acid, oleic acid and stearic acid are reported. Some peaks were not identified using the database search.

Figure 14. Total ion chromatogram of the same EVA film 1 eluate analyzed by GC–MS. The main rosemary oil markers, namely α-Pinene, D-Limonene, (+)-2-Bornanone, D-Pinitol and Borneol are identified. All the compounds were derivatized. Some peaks were not identified or were contaminants from the stationary phase of the column.

Fig. 15 shows the total ion chromatogram of EVA film 2 analyzed with LC-MS. The main lipid classes corresponding to Phosphatidylethanola- mine (PE), Phosphatidylcholine (PC), Sphingomyelin (SM), Diacylgly- cerol (DAG) and Triacylglycerol (TAG) are reported.

Fig. 16 shows the total ion chromatogram of EVA film 3 analyzed with GC–MS. The main linoleic and egg yolk markers, namely pentadecanoic acid, palmitic acid, heptadecanoic acid, stearic acid, cholesterol and cholest-5-en-3-ol, (3α)- are identified. In fig. 17 is reported the total ion chromatogram of the EVA film 4 analyzed with GC–MS. The main conifer resin markers, namely longifolene, methyl dehydroabietate,

10,18-bisnorabieta-8,11,13-triene and 10,18- bisnorabieta- 5,7,9(10),11,13-pentaene are detected.

Figure 15. Total ion chromatogram of the EVA film 2 eluate as analyzed by LC-MS. The main lipid classes corresponding to Phosphatidylethanolamine (PE), Phosphatidylcholine (PC), Sphingomyelin (SM), Diacylglycerol (DAG) and Triacylglycerol (TAG) are identified.

Figure 16. Total ion chromatogram of EVA film 3 eluates analyzed by GC–MS. The main linoleic and egg yolk markers, namely pentadecanoic acid, palmitic acid, heptadecanoic acid, stearic acid, cholesterol and cholest-5-en-3-ol, (3α)- are marked. All the compounds were derivatized. Some peaks were not identified using the database search or were contaminants from the stationary phase of the column.

Figure 17. Total ion chromatogram of the EVA film 4 eluates as analyzed with GC–MS. The main conifer resin markers, namely longifolene, methyl dehydroabietate, 10,18-bisnorabieta-8,11,13- triene and 10,18-bisnorabieta-5,7,9(10),11,13-pentaene are clearly identified.

Linseed oil, walnut oil and poppy-seed oil were traditionally used in paintings. It is known that they show an increase of dicarboxylic acids during ageing, whereas saturated fatty acids are less involved in the oil polymerization process. It is well known that palmitic (P) and stearic (S) acid concentrations change over time but their ratio remains stable enough to be used as a marker of the type of oil47,48. The average P/S ratio in the Donna Nuda resulted 1.87, clearly indicating the presence of linseed oil. In fact, several authors found that a P/S value within the range 1.4–1.9 is typical of this type of oil. Linseed oil was also confirmed by the lipidomic analysis: the TAGs Oleic Stearic Palmitic (OSP), Palmitic Linoleic Palmitic (PLP), Palmitic Linolenic Palmitic (PLnP) and Oleic Oleic Stearic (OOS), which are characteristic markers of linseed oil, were

71 identified together with oleic acid. These TAGs are still present after a strong oxidation and reticulation caused by the ageing, as reported by Degano et al.49.

Most interestingly, the GC-MS analysis revealed also the presence of cholesterol and its derivatives cholest-5-en-3-ol, (3α)- that, combined with odd chain length fatty acids, clearly indicates the presence of animal fats, in particular egg yolk48. Egg yolk was confirmed by the lipidomic analysis that allowed the identification of the complex lipids phosphatidylethanolamines and phosphatidylcholines, which are the most abundant lipid classes in egg yolk, after TGs 50,51. The change from egg tempera painting to oil painting was a slow and gradual process and was mainly aimed at making the application of tempera more fluid and easy, the drying process slower, and in general, the colors more easily manipulated. Before oil paint was adopted as principal painting technique, Leonardo da Vinci was one of the first artists to use a combination of tempera and oil, called “tempera grassa”. He mixed one part of egg yolk with one part of linseed oil to increase the intensity of his colors and lengthen their drying time; this offered him the opportunity to create the dramatic ‘chiaroscuro’ and the subtle blends of ‘sfumato’ tone, that characterize his work. The presence of both oil and egg in the “Donna nuda”, revealed that the technique employed by Leonardo da Vinci was a “tempera grassa”: this is the first scientific evidence of the employment of this painting technique by the most important Renaissance master.

During the second half of the 15th century the use of oil and natural resins became widespread mainly for increasing flexibility in oil binders

72 and to protect the painting. In this case, the metabolomic analysis revealed the presence of natural resin: dehydroabietic acid, methyl dehydroabietate, dehydroabietine, 10,18-bisnorabieta-8,11,13-triene and 10,18-bisnorabieta-5,7,9(10),11,13-pentaene, which are specific marker of aged conifer resins, were identified. Moreover, the detection of the tricyclic sesquiterpene longifolene confirmed the use of conifer resins in the painting preparation52. All these diterpenic markers indicate that Leonardo used Venetian turpentine. Interestingly, the metabolomic analysis identified the following terpenoids: Borneol, D- Limonene, D-Pinitol, α-Pinene, (+)-2-Bornanone (camphora), Thymol and 2-Pinen-4-one. All these compounds derive from plant extracts/essential oils that can be attributed to the specific botanical specie of rosemary53. The use of plant extracts was also confirmed by the presence of vegetable tannins (pyrogallol, cathecol and hydroquinone) and aromatic acids. Resins were usually adopted to improve the spreading of the binding media on the support or as a protective film. Essential oils and turpentine were often added to better combine different colors, but also to slow down or accelerate, respectively, the drying process of oil painting54.

As reported in Table 12, the presence of almost all the markers was confirmed in two or more samples, even if it is clear that the choice of the sampled area can affect the final results. As an example, in sample four there is a strong presence of dehydroabietic acid degradation products, which are related to the presence of conifer resin and thus of Venetian turpentine, while the identification of the main markers of the “tempera grassa” was not possible. This in agreement with the

artistic technique, which indicates that resins were mainly used as protective layer. The presence of essential oil from rosemary was clearly identified in sample 1, which is the only one that comes from the landscape of the painting. In fact, essential oils were often used to slow down the drying process and to guarantee a more fluid and transparent layer of color, which is very suitable to perform the glazing technique. The use of diluents, such as rosemary oil, allows to enhance the sense of depth in correspondence of the landscape, and to create the typical blur effect called “aerial perspective”. Moreover, both Venetian turpentine and rosemary essence were very precious ingredients, as they were obtained through distillation processes, indicating that the materials used for this painting were of high quality. Interestingly, the markers of the “tempera grassa” were mainly identified in the samples collected from the face and the hair, which are rich of particulars.

The present data have been discussed with Pinin Brambilla Barcilon, who spent 20 years (between 1978 and 1998) restoring the Last Supper by Leonardo in the refectory of S. Maria delle Grazie in Milan55. She has critically evaluated our work and agreed that this deep exploration of a Leonardo's painting is quite unique in that, even in her case, she could not perform this chemical analysis also due to the fact that in her times advanced analytical instrumentation was not available. She could only explore the surface of the fresco via spectrophotometric techniques, in order to evaluate the presence of color changes and different hues of the pigments utilized. The conclusion that in this fresco “tempera grassa” had been used was simply based on the statements of Vasari46.

1.2.6.4 Application of the EVA method to the investigation of the 3500 years old mummy of an Ancient Egyptian dignitary

Naturally preserved and embalmed bodies from archeological contexts represent a powerful source of information in many fields. In this chapter it is reported, for the first time, the use of the non-invasive EVA method for the analysis of an Egyptian mummy and his embalming. Thanks to this method we were able to analyze proteins, metabolites and fatty acids, leaving unchanged the precious archeological remains (the manuscript is under submission: Barberis et. al, Uncovering Nebiri: non-invasive metabolomics and proteomics analyses of the 3,500 years old Ancient Egyptian dignitary).

The mummy of Nebiri, is conserved at the Egyptian Museum of Turin. Nebiri was an ancient Egyptian dignitary who lived 3500 years ago under the reign of Thutmoses III (1479– 1424 BC; 18th Dynasty). His tomb (QV30) was discovered by the first director of Egyptian Museum off Turin, Ernesto Schiaparelli, in 1904 in Luxor and only his head (S.5109) and the canopic jars containing the internal organs (lung, stomach, liver, and intestines) were preserved56. Figure 18 shows the photos of the subject of the investigation: the head of the mummy covered by the bandage (18 a) and the canopic jar containing the lung (18 b).

Figure 18. In the image are reported two photos of the head of the mummy (18 a) and of the canopic jar containing the lung (18 b) during the EVA extraction.

Proteomics analysis

Mass spectrometry was used to sequence ancient protein residues. We applied shotgun proteomics using LC-MS/MS for the analysis of proteins from the head and the lung of Nebiri. The proteins were extracted from the surface of the parietal bone, from the head skin and from the lung, using the methodology described in paragraph 1.2.2.

Parietal bone proteins

The analysis of the proteins extracted from the parietal bone allowed the identification of several collagens and keratins, as reported in Table 15. Human collagen alpha-1(I) chain (CO1A1_HUMAN), collagen alpha-2(I) chain (CO1A2_HUMAN) and collagen alpha-1(III) chain (CO3A1_HUMAN) were the most abundant proteins, but two collagens from fowl, namely Collagen alpha-1(I) chain (CO1A1_CHICK) and Collagen alpha-1(IX) chain (CO9A1_CHICK) were detected with unique and specific peptides. The presence of proteins with modifications that are associated to protein biological ageing confirmed the authenticity of the proteomic data and of

76 the head. In fact, proteins with deamidated Asn and Gln were detected in all the collagens, but also the presence of Aminoadipic acid from lysine (CO1A2_HUMAN), which is among the most important age-dependent forms of oxidative damage, indicated the presence of endogenous proteins. Acetylation and hydroxylation were also identified. Collagen alpha-2(I) chain (CO1A2_ONCMY) from fish species was identified with two unique peptides. The presence of human collagens with specific modifications confirmed that the head analyzed was certainly ancient and also showed the ability of the method to extract ancient proteins.

With regard to the presence of collagens from fowl, CO1A1_CHICK and CO9A1_CHICK from Gallus gallus were identified. Both proteins are characterized by ancient modifications, indicating that the collagen is part of the original material and it does not come from a more recent conservation treatment. Animal glues were widely used as binders and adhesives, especially in Egyptian cartonnage, but this is the first scientific evidence of proteinaceous material on a mummy, in particular of animal glue from fowl. During the 18th Dynasty domestic fowl were already present57.

Lung proteins The proteomic analysis of the lung sample revealed the presence of 60 human proteins: among them there are several specific biomarkers of lung tissue. The main proteins are listed in table 15. Lung is undoubtedly a major "immunological organ" since it contains a considerable amount of lymphoid tissue. Neutrophil defensins (DEF1_HUMAN) which is an antimicrobial peptides present in large amounts in the neutrophil58,

77 hemoglobins (HBA_HUMAN and HBB_HUMAN) which are involved in oxygen transport from the lung to the various peripheral tissues, the neutrophil serine proteases cathepsin G (CATG_HUMAN) and neutrophil elastase (ELNE_HUMAN) which are involved in immune-regulatory processes and exert antibacterial activity against various pathogens and Haptoglobin (HPT_HUMAN)59, which has been shown to be associated with the host-defense response to infection and inflammation and is expressed at a high level in lung cells60. The identified proteins were also subjected to gene ontology classification based on biological processes with the Cytoscape software and the ClueGO plug-in. The analysis showed several lung biological processes such as immune response, defense response to bacterium and oxygen transport (Figure 19). Proteins from animal origin were not detected.

Figure 19. Cytoscape based ClueGo pathway analysis and visualization. Enriched pathways were obtained from the Kyoto Encyclopaedia of Genes and Genome (KEGG) database. The figure reports the functionally grouped networks of identified proteins. Terms are linked based on к- score (≥0.4), edge thickness indicates the association strength while node size corresponds to the statistical significance for each term. Biological processes are also reported.

Head skin Proteins extracted from head skin are mainly human collagens (CO1A1_HUMAN, CO1A2_HUMAN and CO3A1_HUMAN) and human keratins (K1C9_HUMAN and K1C10_HUMAN). Besides these human proteins, an animal collagen, namely collagen alpha-2(I) chain (CO1A2_CHICK), was also identified. While all the proteins showed modifications specific for an ancient material, the presence of collagen

from fowl confirmed the use of an original conservation treatment with animal glue.

Table 15. Proteins found in the head and in the lung of Nebiri, with relative score and number of peptides.

N. of Sample PROTEINS (Mascot score) Accession name Score peptides Collagen alpha-1(I) chain CO1A1_HUMAN 1436 23 Collagen alpha-2(I) chain CO1A2_HUMAN 1127 22 Collagen alpha-1(I) chain CO1A1_CHICK 764 12 Collagen alpha-2(I) chain CO1A2_ONCMY 66 3 Keratin, type I cytoskeletal 9 K1C9_HUMAN 464 5 Parietal Keratin, type I cytoskeletal 10 K1C10_HUMAN 236 5 bone Keratin, type II cytoskeletal 1 K2C1_HUMAN 225 5 Keratin, type II cytoskeletal 2 epidermal K22E_HUMAN 40 3 Keratin, type II cytoskeletal 5 K2C5_HUMAN 39 3 Collagen alpha-1(IX) chain CO9A1_CHICK 207 3 Collagen alpha-1(III) chain CO3A1_HUMAN 82 2 Lung Serum albumin ALBU_HUMAN 942 12 tissue Protein S100-A9 S10A9_HUMAN 640 6 (main proteins) Ig alpha-1 chain C region IGHA1_HUMAN 469 5 Alpha-1-antitrypsin A1AT_HUMAN 376 7 Cathepsin G CATG_HUMAN 367 5 Hemoglobin subunit beta HBB_HUMAN 350 3 Histone H2A type 1 H2A1_HUMAN 340 3 Isoform H14 of Myeloperoxidase PERM_HUMAN 288 7 Alpha-1-antichymotrypsin AACT_HUMAN 236 6 Collagen alpha-1(III) chain CO3A1_HUMAN 210 5 Ig gamma-1 chain C region IGHG1_HUMAN 209 5 Hemoglobin subunit alpha HBA_HUMAN 190 3 Neutrophil defensin 1 DEF1_HUMAN 177 3 Lysozyme C LYSC_HUMAN 166 2 Histone H4 H4_HUMAN 158 2 Histone H2B type F-S H2BFS_HUMAN 137 4 Peroxiredoxin-2 PRDX2_HUMAN 125 2 Actin, cytoplasmic 1 ACTB_HUMAN 120 2 Tubulin beta-2B chain TBB2B_HUMAN 118 4 Fibrinogen beta chain FIBB_HUMAN 110 4

Neutrophil elastase ELNE_HUMAN 104 2 Isoform Gamma-A of Fibrinogen gamma chain FIBG_HUMAN 100 3 Band 3 anion transport protein B3AT_HUMAN 77 2 Isoform 2 of Haptoglobin HPT_HUMAN 77 2 Isoform 2 of Complement C4-A CO4A_HUMAN 76 3 Isoform 2 of Heat shock protein HSP 90-alpha HS90A_HUMAN 76 2 Myeloblastin PRTN3_HUMAN 75 2 Head skin Collagen alpha-1(I) chain CO1A1_HUMAN 1145 12 Collagen alpha-2(I) chain CO1A2_HUMAN 1010 13 Collagen alpha-2(I) chain CO1A2_CHICK 134 3 Collagen alpha-1(III) chain CO3A1_HUMAN 400 9 Keratin, type I cytoskeletal 9 K1C9_HUMAN 327 5 Keratin, type I cytoskeletal 10 K1C10_HUMAN 103 3

Ancient protein damage Although proteins survive longer than DNA, they still decay naturally over time. The identification of diagenetic protein modifications was used to distinguish ancient from modern proteins. Hydroxylation of proline, which is one of the main modifications of collagen, was identified in all the samples. Deamidation and aminoadipic acid from lysine are more specifically related to degradation. Deamidation is usually associated with protein biological aging, it plays an important role in protein degradation and has been correlated with time of aging11. Aminoadipic acid from lysine is the most important age-dependent forms of oxidative damage. Presence of 2-aminoadipic acid in the ancient sample can be associated to the decomposition that occurred immediately after Nebiri died, or to infiltration of biogenic taphonomic factors during the prolonged deposition of the sample in soil61. As reported in table 16, parietal bone, lung and head skin samples were all characterized by the presence of

81 deamidation and aminoadipic acid from lysine. The results confirmed the ancient origin of the samples.

Table 16. Ancient modifications identified in proteins extracted from Nebiri.

Skin Parietal Protein Modification Lung head bone Collagen alpha-1(I) chain Deamidated (NQ) 8 11 4 Collagen alpha-1(III) Deamidated (NQ) 3 2 2 chain Deamidated (NQ) 3 4 1 Collagen alpha-2(I) chain Lys-> AminoadipicAcid - 1 - (K) Collagen alpha-5(VI) Deamidated (NQ) - 1 - chain Collagen alpha-3(VI) Deamidated (NQ) - - 1 chain Collagen alpha-1(IX) Deamidated (NQ) 1 - - chain Deamidated (NQ) - - 1 Collagen alpha-1(V) Lys-> AminoadipicAcid chain - - 1 (K) Collagen alpha-1(XXVIII) Deamidated (NQ) - 1 - chain Collagen alpha-1(X) Lys-> AminoadipicAcid - 1 - chain (K)

Metabolomic analysis

For the first time this method was applied also for the non-invasive identification of the mixture used for embalming the ancient Egyptian dignitary, through GC-MS and GCxGC-MS.

Mono and two-dimensional GC-MS analysis allowed a robust identification of trace compounds present in the lung of Nebiri. The detailed experimental procedure and the GC-MS and GCxGC-MS methods are described in the previous paragraph n. 1.2.4.2.

The lipid component of the samples analyzed consists mainly of:

-Linear monocarboxylic saturated fatty acids ranging from 3 to 24 carbon atoms, which are the majority of the observed biomarkers. In the samples of lung, canopic jar, head, we found: palmitic, stearic and myristic acids as the most preminent, and at lower concentrations propionic, butyric, valeric, caproic, heptanoic, caprylic, pelargonic, decanoic, undecanoic, lauric, tridecanoic, pentadecanoic, margaric, nonadecanoic and arachidic acids. The palmitic/stearic ratio ranging from a minimum of 1.2 for the textile samples to a maximum of 2 for the lung tissue, suggests the use of vegetable oil as a major ingredient of the mixture.

-Dicarboxylic acids as azelaic acid, succinic acid, pimelic acid, adipic acid, which are formed by the degradative oxidation of originally unsaturated fatty acids.

-Hydroxycarboxylic acids derived from the oxidation of the acyl chain of triglycerides as: 2-Hydroxy-2-methylbutyric acid, 2-Hydroxybutyric acid, 2-Hydroxyisocaproic acid, 3-Hydroxybenzoic acid, 3-Hydroxyisovaleric acid, 9,18-Dihydroxyoctadecanoic acid, α-Hydroxyisobutyric acid, 7- Hydroxyheptanoic acid.

-Monounsaturated fatty acids as: oleic, gondoic acids and erucic acid, (quite unusual in ancient samples for its great tendency to oxidation during the time)62. The lipid profile thus seems to indicate that a plant oil or a mixture of plant oils was present in the embalming recipe.

-Cholesterol degradation products are very abundant in lung and lung bandage samples, and they probably derive from the organic tissue, as reported in figure 20.

-Signals of diterpenoids from conifer resins and their relative oxidations products as Dehydroabietic acid, 15-Hydroxy-7-oxodehydroabietic acid, 7-Oxodehydroabietic acid, are consistent with what was found in previous studies of the funerary treatments63,64.

-Low amount of aromatic acids characteristic of plant products are present in the mixture as constituent of the balm: 3-Hydroxybenzoic acid, 4-Hydroxybenzoic acid an Vanillic acid. They are present in the samples of the lung, lung bandage and canopic jar.

-The presence of monosaccharides, more abundant in the lung than in the textile wrapping, could suggest the possible trace of a plant gum or sugar as a component of the balm. However, the absence of sugars in the samples of head and canopic bandages does not allow to establish if these signals exclusively originate from the lung tissue.

-Triterpenoids acids from Pistacia resin as 28-Norolean-17-en-3-one, Olean-18-en-3-ol, O-TMS, (3β), Β-Amyrin, Dammaran-3-one,20,24- epoxy-25-hydroxy.

-Vegetable tannins and phenols with antiseptic activities, derived from plants as catechol and guaiacol, have been found in most samples. The presence of guaiacol in ancient recipes could be due to the use of wood smoke/cedar wood in the mummification process65.

The final recipe used for embalming Nebiri is summarized in table 17.

Figure 20. Extracted ions (contour 3D plot) of main cholesterol derivatives in lung GCXGC analysis.

Table 17. Final recipe used for embalming Nebiri. List of main class of compounds associated with the samples and relative assignment CLASS OF COMPOUND SAMPLE RECIPE PRODUCTS Linear monocarboxylic Lung, canopic jar, PLANT OILS AND RELATIVE saturated fatty acids; head OXIDATION PRODUCTS Dicarboxylic acids; Hydroxycarboxylic acids; Monounsaturated fatty acids Diterpenoids Lung, canopic jar, PINACEAE RESINS head Aromatic acids Lung, canopic jar, VEGETABLE BALMS head Monosaccharides Lung, canopic jar, HUMAN TISSUE OR GUMS head Triterpenoids Lung PISTACIA RESIN Tannins Lung, canopic jar, CEDAR OIL/WOOD SMOKE head Collagen proteins head ANIMAL GLUE (FISH AND FOWL)

Some consideration about Pistacia resin

Although the most specific triterpenoid of masticadienonate series (moronic, masticadienoic, isomasticadienoic and oleanonic acids) have not been detected in the monodimensional GC-MS analysis, in the bidimensional one we found other biomarker compounds, generally used to identify an antique Pistacia resin66: penta and tetra-cyclic triterpenes from resins of Pistacia species as oleanane-type molecules, in particular β-Amyrin, with most abundant fragment ion at m/z 218 and Olean-18-en- 3-ol, o-TMS with main fragment ions at m/z 189 and 204; a triterpenoid with dammarane skeleton, Dammaran-3-one, 20,24-epoxy-25-hydroxy with characteristic base peak at m/z 14367. In this case, as well described by Daifas et. al68, we confirmed that Pistacia resins were used for embalming treatments, probably for their antibacterial, antifungal and antiseptic properties68,69 or they might have been employed in the preparation of kyphi ointment, as well as been used for its religious capacity 70. Although various authors reported the presence of two species of Pistacia, lentiscus and atlantica71, we were not able to discriminate the different species within the sample, possibly for the very low abundance of characteristic markers. However, the presence of Pistacia and other triterpenoid compounds can lead us to strongly hypothesize that the employed mummification method used for Nebiri was expensive and typically reserved for royalty and very wealthy nobles65.

Table 18. The table shows the name of heated Pistacia markers obtained in the lung sample with the untargeted analysis by GCxGC-MS, the respective formula, the retention time in first and second dimension and the similarity > 700 (index of identification reliability).

GCxGC-MS

Name Formula R.T. (s) Similarity 163 100 st S28-Norolean-17-en-3-one C29H46O 2474,88 (1 ) 790 m/z 163.16 2,840 (2nd)

191

410

95 175 81 105 69 91 119 55 135 205 41 147 395 st 204 100 29 215 243Olean-18-en-3-ol, O-TMS, (3β) C H OSi 2484,88 (1 ) 877 0 33 58 20 30 40 50 60 70 80 90 100 110 120 130 140189150 160 170 180 190 200 210 220 230 240 250 260 270 280 290 300 310 320 330 340 350 360 370 380 390 400 410 420 nd (mainlib) 28-Norolean-17-en-3-one m/z 189.17 2,616 (2 )

H 177 73

95 109 H 121 Si 231 O 129 81 H 69 218 498 147 161 218 483 st 100 55 Β-Amyrin C H O 2479,88 (1 ) 779 279 30 50 41 393 257 369 293 nd 29 61 243 271 306 320 339 348 408 426 441 455 472 0 m/z 218.21 3,846 (2 ) 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 250 260 270 280 290 300 310 320 330 340 350 360 370 380 390 400 410 420 430 440 450 460 470 480 490 500 510 (mainlib) Olean-18-en-3-ol, O-TMS, (3β) OH

50 203

69 95 189 55 81 43 107 119 135 st 147 175 161 Dammaran-3-one,20,24- C30H50O3 2639,87 (1 ) 797 257 426 29 229 243 271 283 297 311 337 365 393 411 0 nd 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 250 260 270epoxy280 290 300-25310-320hydroxy330 340 350 -360 370 380 390 400 410 420 430 440 1,014 (2 ) (mainlib) β-Amyrin 100 143 m/z 143.11 OH

H 50

O 125 399 59 81 95 205 43 71 107 119 161 135 175 381 31 189 219 245 357 425 443 0 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 250 260 270 280 290 300 310 320 330 340 350 360 370 380 390 400 410 420 430 440 450 460 470 (mainlib) Dammaran-3-one, 20,24-epoxy-25-hydroxy-,

In conclusion, the non-invasive chromatographic analysis performed both on the bandages that originally enveloped Nebiri's head and on the lungs, allowed to identify some of the components of the "recipe" used for embalming: animal fat, balms, essential oils, aromatic plants, heated Pistacia and coniferous resins.

The use of Pistacia and coniferous resins, which are plant species not native of Egypt, and thus probably imported from the Levant, is indicative of the high social status of Nebiri. This statement is further corroborated by the title "Head of the Stables" attributed to him. Since the introduction of horses in Egypt, dates back to the Second Intermediate Period (1785- 1580 BC), the attribution of a title as "Head of the Stables" to an individual who lived at the beginning of the XVIII dynasty (1479-1425), lead to speculate that Nebiri was a high-ranking officer or a member of the royal family.

The cosmetic treatment applied to him, highlights the use of an extremely refined embalming technique: a similar treatment can be found only at the end of the XVIII Dynasty (more than 100 years later) in the mummies of Yuya and Thuju (KV46), parents of Queen Tiye, royal bride of the pharaoh Amenhotep III (1338-1348 BC), now conserved at Egyptian Museum in Cairo72,73. Finally, the metabolomics results contributed to obtain more information regarding the social status of Nebiri, while the proteomics results confirmed the authenticity of the analyzed tissues and highlighted the first use of animal glue in an Egyptian mummy.

1.2.6.5 Non-invasive analysis of Stradivari Violins

Historic stringed musical instruments are a unique class of cultural heritage objects. Crafted during the 16th-18th centuries, these instruments remain somehow mysterious. Due to the absence of written historical documents of traditional varnish recipes, their chemical characterization is the only way to recover the lost ‘secret’ of the Cremonese stringed instrument maker Antonio Stradivari and his contemporaries, whose traditions has been lost since 1800. The analytical characterization, both with non-invasive and minimally invasive techniques, provided information about the material composition, methods of manufacture, and past restorations of the finishes.

Wali et. al74 claim that the application of the varnish can significantly alter the mechanical properties of the wood support: this effect depends on the type of wood, on the control of its hardening conditions, on the environmental conditions as temperature and humidity and, above all, on the control of the penetration depth of the varnish into the wood. Stradivari and his contemporaries seem to have been able to make the best use of all three aspects of varnish control. But this process still remains a mystery.

Recent works reported that before the varnish, sometimes a layer of proteinaceous glue (with inorganic particles, like silicates of potassium and calcium) was applied as a sort of “ground” layer75,76. After the preparatory layers, the varnish was applied. The varnish exhibits multiple functions: it protects the wood but also improves optical characteristics as well as the sound of the instruments by hardening the wood.

The varnish applied on musical instruments can be divided into three different types: 1. Spirit varnishes (resins dissolved in alcohol); 2. Essential oil varnishes (resins dissolved in volatile oils); and 3. Oil varnishes (resins mixed with siccative oils)77.

Historic varnishes of musical instruments may contain inorganic particles (such as inorganic colorants and fillers), suspended in organic binders. For example, mercury (Hg) is an element in cinnabar and vermilion (HgS). Possible organic compounds include drying oils, essential oils, tree resins, tree gums, insect resins, dyes, and proteins or polysaccharides, used alone or mixed together. The compounds may be purified, pre-treated, or diluted in a volatile solvent77.

In previous studies, Echard et al. described some of the difficulties related to the study of varnish of Stradivari’s instruments77-79. The organic components could be derived from animal parts, such as albumin, casein, or collagen (derived from cartilage, bones, skin, etc.), or derived from plants, i.e. from oils rich in unsaturated fatty acids such as drying oils and terpene exudates such as gums, resins and fossil amber77,79.

During my PhD research, the EVA method was also applied to historic and famous instruments from Antonio Stradivari (1644-1737), conserved at The Violin Museum of Cremona (collaboration with Prof. Marco Malagodi, University of Pavia). This method was confirmed to be appropriate, fast and effective for the study of proteins, resinous and fatty organic in materials of complex cultural heritage composition, such as the historic musical instruments.

The EVA technology was applied to a group of three historical Stardivari's violins produced in three different period of his life, in order to compare the recipes used by the luthier along his life.

Ancient violins are made by wood, natural organic substances as resins, siccative oils, solvents, dyes and, in small amount of inorganic additives as driers, fillers and pigments. The search for identifying these natural substances, especially in aged or ancient work of arts, has been performed using various non-invasive techniques80,81, but obtaining information on the composition of the organic fraction is possible only using gas or liquid chromatography.

Thanks to the non-invasive EVA method, it was possible to extract several proteins and about 1800 molecules from the surface of the Stradivari violins. In this case the proteomics analyses were performed using Nano LC-MS/MS, as previously described in paragraph 1.2.3.2 and in other real case applications, while the metabolites were separated and identified by the use of comprehensive bi-dimensional gas chromatography (see paragraph 1.2.4.2). The analytical results were then processed and compared with the open source MetaboAnalyst software.

The ground layer of the violins was characterized by the presence of animal proteins: bovine glue and casein. While in the varnish, the analytical technique was able to identify the use of siccative and essential oils, aged conifer resins and their relative oxidation products, but also polycyclic aromatic hydrocarbons as markers of red organic dyes, as shown in table 20. The statistical analysis allowed us to discriminate the

91 recipes used by Stradivari in three different instruments built along the 50 years of his master career.

The sampling with EVA method is still in progress and received the approval to be extended to other important instruments conserved in the same museum collection.

Results

It is well known that glue was used as an impregnating agent before applying the paints in several artworks especially in wood panels and canvas. However, according to historians, museum conservators and luthiers, this system was also applied to Stradivari instruments most probably for better isolate the acoustic case of the violin and improve the quality of the sound82-85. The proteomics and metabolomics analyses were performed in 72 different areas from several parts of three different violins, on the top and on the back: Clesbee (1669), Cremonese (1715) and Vesuvius (1727) (figure n. 21).

Figure 21. a) Photo taken in the laboratory of the Violin Museum of Cremona during the EVA extraction on the top of the violin Vesuvius (1727); b) Sampling map of the violins, in orange the points for the metabolomics, in blue those for the proteomics.

The presence of proteins was confirmed both on the top and on the back of the three violins, as shown in table 19.

Table 19. Identification of the proteinaceous material used by Stradivari for the manufacturing of the three violins analyzed. The table reports the proteins, the accession name, the mascot score and the number of peptides identified in each analyzed sample.

Instrument Side Material Protein Accession name Score N. of peptides Front Animal glue Collagen alpha-2(I) chain CO1A2_BOVIN 378 4 Collagen alpha-1(I) chain CO1A1_BOVIN 170 2 Front Bovine casein Alpha-S1-casein CASA1_BOVIN 56 2

CLISBEE Front Bovine casein Alpha-S1-casein CASA1_BOVIN 51 2 Back Bovine casein Alpha-S1-casein CASA1_BOVIN 52 2 Front Animal glue Collagen alpha-2(I) chain CO1A2_BOVIN 334 4 Collagen alpha-1(I) chain CO1A1_BOVIN 233 3 Intern Animal glue Collagen alpha-2(I) chain CO1A2_BOVIN 411 5 Collagen alpha-2(I) chain CO1A2_BOVIN 221 2

VESUVIUS Back Bovine casein Alpha-S1-casein CASA1_BOVIN 60 2 Beta-Casein CASB_BOVIN 55 2

Front Animal glue Collagen alpha-2(I) chain CO1A2_BOVIN 255 3

Front Bovine casein Alpha-S1-casein CASA1_BOVIN 56 2 Beta-Casein CASB_BOVIN 48 2 Intern Animal glue Collagen alpha-2(I) chain CO1A2_BOVIN 387 4

Collagen alpha-1(I) chain CO1A1_BOVIN 154 2 CREMONESE Back Bovine casein Alpha-S1-casein CASA1_BOVIN 54 2

The results of the analysis showed that the top and the back of the violins were treated and prepared using different techniques. On the top, the proteins identified are from collagen and casein, while on the back only casein proteins were observed. This suggests that different ground preparations were employed for the two main parts of the violin.

Moreover, on the top of the instruments several signals of diterpene resins of vegetable origin were identified, probably employed as finishes (various derivatives and oxidation products of abietic acid, characteristic of the Pinaceae family) or as antiseptics (labdanes). Fatty acids of vegetable origin and their oxidation products (carboxylic acids) instead indicate the probable presence of a drying oil spread over the entire instruments surface. The palmitic/stearic ratio was from 1 to 2, indicating that linseed oil has been used. Several lignin oxidation products, strictly related to the substrate, were detected, while the presence of numerous amino acids confirmed the use of animal glue, as already reported by the proteomic analyses. Finally, the use of essential oils was suggested by the presence of sequiterpenes, while the identification of tannins could be related to the use of dyes, probably galls. The list of the main identified metabolites is reported in table 20.

Table 20. List of the identified small molecules related to the materials used by Stradivari for the manufacturing of the three analyzed violins. CLASS NAME OF METABOLITES TANNINS AND (1R,4aR,4bS,7S,10aR)-1,4a,7-Trimethyl-7-vinyl-1,2,3,4,4a,4b,5,6,7,8,10,10a-dodecahydrophenanthrene- 1-carbaldehyde; 1,6-Dihydroxy-8-methoxy-3-methylanthraquinone, O,O'-bis(trimethylsilyl)-; ANTHRAQUINONE DERIVATIVES 1-Phenanthrenecarboxaldehyde, 7-ethenyl-1,2,3,4,4a,4b,5,6,7,9,10,10a-dodecahydro-1,4a,7-trimethyl-, [1R-(1α,4aβ,4bα,7β,10aα)]- 4b,8-Dimethyl-2-isopropylphenanthrene, 4b,5,6,7,8,8a,9,10-octahydro-; 7-Isopropyl-1,1,4a-trimethyl-1,2,3,4,4a,9,10,10a-octahydrophenanthrene; 9,10-Anthracenedione, 1,4- diamino-2,3-dihydro-; Hydroquinone, 2TMS derivative; Parietin, 2TMS derivative; Pyrogallol, 3TMS derivative PHENOLS Phenol, TMS derivative; Thymol, TBDMS derivative ESSENTIAL OILS Hydroxycitronellal; Linolool oxide, TMS derivative; Menthol, TMS derivative TERPENIC RESINS FROM 10,18-Bisnorabieta-5,7,9(10),11,13-pentaene; 10,18-Bisnorabieta-8,11,13-triene; 1-Methyl-10,18- bisnorabieta-8,11,13-triene; Abietic acid, TMS derivative; Dehydroabietic acid, TMS derivative PINACEAE Isopimaric acid, TMS; Labda-8(20),14-dien-13-ol, (13S)-, O-TMS MONOUNSATURATED FATTY Oleic acid; cis-Vaccenic acid ACIDS CRBOXYLIC ACIDS 1-Monomyristin, 2TMS derivative; 1-Monopalmitin, 2TMS derivative; 2-Aminomalonic acid, N,O,O,-TMS; 3-Aminoisobutyric acid, 3TMS derivative; 4-Methylvaleric acid, TMS derivative; 6-Octadecenoic acid; 9(E),11(E)-Conjugated linoleic acid, ethyl ester; 9,12-Octadecadienoic acid (Z,Z)-, TMS derivative; 9- Hexadecenoic acid, hexadecyl ester, (Z)-; 9-Hexadecenoic acid, octadecyl ester, (Z)-; 9-Octadecenoic acid, (E)-, TMS derivative; Acetic acid, TMS derivative; Arachidic acid, TMS derivative; Behenic acid, TMS derivative; Butanedioic acid, 2TMS derivative; cis-13-Octadecenoic acid; Decanoic acid, TMS derivative; Dodecanedioic acid, 2TMS derivative; Dodecanoic acid, TMS derivative; Formic acid, TMS derivative;

Heneicosanoic acid, TMS derivative; Heptadecanoic acid, TMS derivative; Hexacosanoic acid, TMS derivative; Hexadecanedioic acid, 2TMS derivative; Hexanedioic acid, 2TMS derivative; Hexanoic acid, TMS derivative; Lignoceric acid, TMS derivative; Myristic acid, TMS derivative; Nonadecanoic acid, TMS derivative; Nonanoic acid, TMS derivative; Octanoic acid, TMS derivative; Oxalic acid, 2TMS derivative; Palmitelaidic acid, TMS derivative; Palmitic Acid, TMS derivative; Pentadecanoic acid, TMS derivative; Propanoic acid, TBDMS derivative; Propanoic acid, TMS derivative; Pyrrole-2-carboxylic acid, 2TMS derivative; Pyruvic acid, TMS derivative; Sebacic acid, 2TMS derivative; Stearic acid, TMS derivative; Tetradecanedioic Acid, 2TMS derivative; trans-13-Octadecenoic acid; Undecanoic acid, TMS derivative; Methyl trans-4-methylcinnamate; Vinyl trans-cinnamate HYDROXYCARBOXYLIC ACIDS α-Hydroxyisobutyric acid, 2TMS derivative; 3-Hydroxybutyric acid, 2TMS derivative AMINO ACIDS Alanylglycine, 2TMS derivative; Glycine, 3TMS derivative; l-Norleucine; Tyramine, 3TMS derivative; Valine; Valylvaline; d-Proline SUGARS D-Fructose, 5TMS derivative; D-Glucopyranose, 5TMS derivative; D-Glucose, 5TMS derivative; Galactopyranose, 5TMS derivative; Glucopyranose, 5TMS derivative; Ribitol, 5TMS derivative; Talose, 5TMS derivative; α-D-(+)-Talopyranose, 5TMS derivative; α-D-Glucopyranose, 5TMS derivative; α-D- Mannopyranose, 5TMS derivative; β-D-(+)-Talopyranose, 5TMS derivative

Results comparison

Metabolomics confirmed the presence of conifer resins markers, which were more abundant on top of the violins than on its back. As reported in the heat map represented in figure 22, the markers Deydroabietic, Abietic and Isopimaric acids are more abundant in the samples collected from the top of the violins, as well as the myristic and pyruvic acids, while oleic acid (cis and trans 13- Octadecenoic acid) is present in larger amount on the back of the instrument. These results indicate that Stradivari might have used more finishing steps on the top of the instruments. In fact, the dendrogram shown in figure 23 highlights that the two parts of the violins are quite different. The results obtained from this analysis confirmed that the choice of the materials was extremely important. In fact, as suggested by Wali et al.74, the sound quality of the violin depends greatly on its vibrations, but also on the composition of both the top and the bottom plate, especially on their finishing.

As suggested by the conservators, thanks to this analytical method, it has been possible to affirm that Stradivari used different recipes to enrich the finishing of his violins, composed both of animal proteins, ad of resins and of vegetable oils, that altogether, more than 300 years after the instrument creation, are able to make its sound of still unique and inimitable all over the world.

Figure 22. Heat map with the bottom vs the top of the three sampled violins. The 7 regulated metabolites (fold change > 1.5 and p value 0.05) confirm that different compositions have been used to finish the top rather than the bottom of the violins.

Figure 23. In the image it is reported the MetaboAnalyst partition dendrogram obtained from the samples of the top and of the back of the instruments.

1.3. Non-invasive multispectral imaging to uncover hidden features in ancient Egyptian frescoes

Multispectral imaging coupled to digital image processing and computing is an important tool for the non-invasive analysis of artworks. Multispectral imaging is a non-invasive technique that acquires the spectral reflectance of an object with accuracy comparable to a spectrometer. The method is widely used to study cultural heritage objects, to analyze their conservation treatments, for documentation and archiving.

This section reports the results of the application of LED multispectral imaging for discovering new insights related to the frescoes of the Tomb of Iti and Neferu (about 2000 B.C.) from the Egyptian Museum of Turin, Italy. These important frescoes were detached from the Tomb of Iti and Neferu, situated in Gebelein near Luxor in Egypt, in the 1911.

1.3.1 Imaging system and data processing

A portable customized imaging system was used to acquire the multispectral images86. The fresco was actively illuminated by light emitting diodes (LEDs) generating different spectral bands, while a monochrome camera synchronized with LED, was able to capture the reflected light. LEDs were used as light source because they are stable, highly monochromatic and non-damaging as there is no heat on the imaged objects. Two identical panels (Equipoise Imaging LLC, MD, USA) were used, each panel consisting of 10 different LEDs centered on the following wavelengths 370, 444, 466, 598, 521, 600, 640, 923, 1250 and

100

1450 nm. The LED bandwidth ranged from ∼10 nm in UV to 40 nm in the IR. A CCD camera with an 8.3-megapixel Kodak CCD monochrome sensor array (ST-8300M, Santa Barbara Instrument Group, CA, USA), 17.96 x 13.52, with 3326 x 2504 pixels, 5.4 x 5.4 μm was employed.

The multispectral images were then processed using digital image algorithms, principal component analysis and statistics, using the open source program ImageJ version 1.52t. In particular, we employed the infrared false color technique based on the overlapping of two pictures, taken through different set-ups with infrared and visible light (RGB). Integrating the green parts of the image as blue, the red parts as green, and the near infrared information as red, it is possible to generate a new image. The objective of this technique is to characterize and deeply investigate the various pigments that constitute the artwork, in order to obtain information on its origin, the period of its creation and the presence or absence of restorations87. Another imaging tool that was used is DStretch (https://www.dstretch.com/). DStretch, acronym of “Decorrelation stretch”, is an image enhancement technique that can be free downloaded as plugin of ImageJ. The technique, first developed in 1996 at JPL (NASA), consists of applying a Karhunen-Loeve transform to the colors of the image. This diagonalizes the covariance (or optionally the correlation) matrix of the colors. Then, the contrast for each color is stretched to equalize the color variances. At this point the colors are uncorrelated and filled to the colorspace. Finally, the inverse transform is used to map the colors back to an approximation of the original88.

101

1.3.2 Imaging results and discussion

The multispectral imaging analysis performed on different areas of the Egyptian fresco allowed the discovery of several scriptures and hidden paintings that were not visible at the human eyes, and that were not reported in the travel book of the archaeologist Ernesto Schiaparelli, who discovered the ancient frescoes at the beginning of the last century. The analysis revealed also new insights on the collocation of a Stele founded near the Tomb.

Figure 24 reports the color image in visible light of the detached fresco named “painting of Grus”. In yellow are reported the areas involved by the imaging analysis.

Figure 24. The image reports the photo in visible light of the part of detached fresco named “painting of Grus”. In yellow the areas involved by the analysis LED-MSI are shown.

The multispectral analysis performed on the area numbered as three in the center of the fresco, allowed to distinguish the different phases of painting execution. Interestingly, the results showed new details of the bird’s plumage, obtained with the so-called “rope painting technique”. This painting technique was usually performed only by masters with

102 great technical expertise. Moreover, as shown in figure 26, the analysis performed with the DStretch plugin revealed the presence of trilobed- shaped (on the top of the image), and traces of paint underlying. The image also showed the presence of black shoes, probably belonging to a person depicted near a chariot pulled by oxen, originally painted above the “painting of Grus”.

Figure 25 In the image some multispectral images of the central part of the “painting of the Grus” (point 3 in figure n.) processed with different algorithms are shown. With the LED-MSI technique, it was possible to distinguish the different phases of painting execution.

Figure 26. A particular of the multispectral photo processed in false color of the upper register of the “painting of the Grus” is shown.

103

In figure 27 it is reported the color image of the fresco “painting of boat”, while in figure 28 two multispectral details of the rope area of the same fresco are reported: on the left an IR image is presented, on the right the MSI image with post-processing in false color. During the discovery of the tomb, the Egyptologist Virginio Rosa described the presence of animal figures attributable to monkeys. The multispectral images confirmed these evidences: the profile of a monkey appeared on a rope of the boat while the head is lost. The false color processed images showed also the presence of numerous traces of color on the animal's abdomen and on the hind legs.

Figure 27. In the image it is reported the photo in visible light of the detached fresco named “painting of boat”.

104

Figure 28. Multispectral details of the rope area of the “painting of the boat: on the left the IR image at 1200 nm and on the right the MSI image with post processed in false color.

Figure 29 on the left reports a detail of the center of the painting of the boat. Close to the subject, positioned in the center of the boat, it emerged an inscription previously unknown and not indicated in the diary of Virginio Rosa (right). The resulted image (right) was obtained by stacking 9 different wavelengths (from 280 to 1250 nm). Figure 30 reports a particular of another inscription not observed after the discovery in 1911 from the “painting of grain”. From left to right it is shown the same area of the fresco acquired in visible light, after MSI stacking and after false-color processing. The inscription/caption reported: the name of a servant piling up grain in a granary and identifies him as belonging to Iti’s estate.

105

Figure 29. Unknown Inscription (left). Post processed image using by stacking 9 different wavelengths (right).

Figure 30. Particulars of a caption not observed after the discovery in 1911 from the “painting of grain”. From left to right it is shown the same image acquired in visible light, after MSI stacking and after processing in false colors.

106

Another important discovery was achieved thanks to the use of this non- invasive technique for the analysis of the fresco “painting of the tomb” (figure 31 left). In fact, an afterthought of the painter not visible at human eyes was discovered. The enlargements (right) of the area within the red box (left) shows the color image of the representation analyzed. The image at the bottom-right reported the same area after the MSI elaboration. The caption reports the words uttered by the servant, who is offering the foreleg of an ox to the deceased: "To your life spirit". The multispectral image has highlighted what appears to be the painter’s correction of an excessively slanted thumb in the hieroglyph meaning “life spirit” (ka).

Figure 31. The figure shows details of the “painting of the tomb”. From the left to the right, the enlargements of the area with the red box, the image in visible light (above) and after MSI processing (below) are shown.

In conclusion, the MSI analyses allowed the in-depth investigation of the layers of the paintings of Iti and Neferu Tomb, and to rediscover traces

107 of lost colors. Thanks to this information it was possible to proceed to the virtual digital restoration of the frescoes and to hypothesize the original design and arrangement of the paintings. The imaging analyses have been enriched and supported by the precious historical research of the Egyptologists and conservators of the museum. Thanks to the archive work, a reconstruction of the probable location of the scene depicting the boat on the Nile and the Grus was also carried out.

All the results are now part of a temporary exhibition “Archeologia Invisibile” (12 March 2019 – 7 June 2020) organized by the Egyptian Museum of Turin. Part of results are summarized in the exhibition’s catalogue89. The technique used will be soon extended to the analysis of all the frescoes of the Iti and Neferu tomb conserved in the Egyptian Museum.

108

1.4 Cultural heritage monitoring and analysis with Infrared Spectroscopy

Today, non-invasive and non-destructive devices are widely employed for the in-situ characterization of the materials used by the artists in cultural heritage investigations. The rapid growth and diffusion of the modern technology allowed the development of new portable and handle instruments for the analysis of cultural heritage objects.

In the following paragraphs it is reported the application of non-invasive infrared methods (IR) methods based on Attenauted Total Reflection spectroscopy (ATR) and Diffuse Reflectance Infrared Fourier Transform spectroscopy (DRIFT) portable instruments, to the monitoring of common plastics used in contemporary artworks, and to characterize the most used colorants and pigments used by artists.

1.4.1 Non-invasive characterization of colorants by portable diffuse reﬂectance infrared Fourier transform (DRIFT) spectroscopy and chemometrics In this paragraph it is reported the application of a portable diffuse reﬂectance infrared Fourier transform (DRIFT) method for the non- invasive characterization of colorants prepared according to ancient recipes, and using egg white and arabic gum as binders44. With this method there is no need to touch the sample and the analysis can be done directly on site. No sample needs to be removed to analyze the object and, in fact, a large number of areas of the object surface can be analyzed quickly and in a non-destructive manner.

109

The scientific study of the colorants used in the fields of cultural heritage is of great importance for the examination of paintings, textiles, illuminated manuscripts and other objects. During the last years the in- situ characterization of materials has been performed by using a wide variety of analytical techniques: infrared spectroscopy was used to identify pigments, binders and fillers in paintings90-92, ultraviolet and visible reflectance spectrophotometry with optical fibers (FORS) was employed for the non-invasive identification of several colorants used by ancient illuminators93, X-ray fluorescence spectrometry has been often applied to paintings analysis94-96, even if it is an elemental technique that does not provide unique pigments identification. Multi-techniques researches are becoming very popular because of their great diagnostic power7,9,97-99, but nonetheless, multi-analytical in-situ non-invasive approaches are always preferred for the analysis of paintings: i.e. Miliani et al. used five portable spectroscopic techniques, including X-ray fluorescence, mid-infrared reflectance spectroscopy, near infrared spectroscopy and UV–Vis spectroscopy for the analysis of paintings100. Diffuse reflectance infrared Fourier transform (DRIFT) spectroscopy is a rapid and powerful technique that can be applied to the analysis of works of art: DRIFT handled spectrometers have already been employed to perform in situ analysis of frescoes and canvas grounds. The use of external reflection FTIR spectroscopy for the identification of pigments and binders in illuminated manuscripts was also exploited, but only few inorganic pigments were recognized8,101,102. The diffuse reflectance spectra are very similar to those obtained in transmission mode, but overtones and combination bands increase their

110 signals because many reflections occur and the interpretation of the vibrational profiles is often more difficult because distortions as derivative-like spectral features, inverted bands and Reststrahlen effect may occur. The applicability of a DRIFT handled device to perform the in-situ analysis of colorants will be shown, creating and discussing a spectral database that can be very useful to conservators and historians. PCA–DA was also employed with the aim of generating a classification model based on the chemical/mineralogical classes of colorants and of identifying the characteristic bands of each group. Finally, in order to verify the suitability of the DRIFT method, the analysis of a Graham Sutherland's gouache was performed: he was an influential and imaginative British artist, renowned for his visionary landscapes and religious iconography; he also painted a full-size portrait of Winston Churchill in 1954. The colorants used by the painter were identified directly in situ and in a non-invasive manner.

1.4.1.1 Colorants preparation A total of fifty colorants widely used by the first artistic expressions like the Egyptian Blue (about five thousand years ago) to those used by the modern artists were selected and prepared by painting areas on parchment, and using egg white and arabic gum as binder, according to ancient recipes103. The colorants were purchased by Kremer Pigmente GmbH & Co (Aichstetten, Germany). The colorants are classified on the basis of their origin (seeTable 21): mineral (M), synthetic-artificial (S) and organic (O). The wide variety of the color table allowed the building of a

111 spectral library that can be used for the identiﬁcation of colorants from manuscripts to gouache, but also for other types of support. The complete list of colorants used is shown in Table 21, together with the bands positions and characteristics of each infrared spectra.

Table 21. Colorants analyzed with DRIFT spectroscopy and band assignments. The colorants are classiﬁed based on their origin: mineral (M), synthetic-artiﬁcial (S) and organic (O).

112

1.4.1.2 DRIFT, data processing and chemometrics analysis Diffuse reflectance infrared Fourier transform (DRIFT) spectroscopy was performed using an Agilent 4100 Exoscan FTIR portable spectrometer (Agilent Technologies, Santa Clara, CA, USA) equipped with a diffuse reflectance sampling interface. The sample spectra were obtained in absorption mode, using a diffuse reflectance device over a wavelength interval from 650 cm−1 to 4000 cm−1, at 32 scans per sample and a resolution of 4 cm−1. The system had a ZnSe beam splitter and a DTGS detector. The background signal was acquired with a diffuse gold reference cap, and the treatments of the spectral data were performed using OPUS 5.5 software (Bruker, Billerica, MA, USA). The spot size of the beam was estimated to be about 2 mm2 and the time required to perform a full scan was about 15 s. The measurements were performed holding the instrument by hand. Three DRIFT spectra of each sample were collected in different positions and the average spectra were used for identification purposes. In this handled device the optical head measured a higher proportion of the diffuse reflectance signal and a low proportion of the specular one, that guaranteed a less distorted spectrum. The Unscrambler 10 (Camo Inc., Norway) was used for data processing (smoothing moving average with step 7 followed by a Standard Normal Variate (SNV) transformation), and MarkerView software 1.2.1 (AB SCIEX) was used for Principal Component Analysis–Discriminant Analysis (PCA– DA).

113

1.4.1.3 Infrared library result: binders and colorants In the experiment, two binders were used: arabic gum and egg white. Gum Arabic is a natural gum extracted from various species of the acacia tree, and it is composed by a mixture of glycoproteins, mono- and di- saccharides. Gum Arabic is the most commonly used gum in preparation of paintings, especially employed in antiquity for manuscript, inks and textile sizing, but also for the preparation of different 19th century paintings using techniques like watercolor or gouache91,102,103. Egg white is among the most traditional tempera media used for painting. It is composed by water, proteins, fats, lecithin, salts and other substances. Egg white has been used as a medium for illuminated manuscripts or for attaching gold leaf, employed in panel painting and also in mural painting for integration and ﬁnishing works35,105. The absorption spectra of the two binders (reported in Fig. 32a) show similarities with the infrared spectra found in literature102-103: Gum Arabic shows its characteristics IR absorption bands at 2956 cm−1 (C=H stretching band), 1642 cm−1 (OH bending), 1450 cm−1 (C=H bending), 1140 and 976 cm−1 (C=O stretching bands), while egg white shows its typical signals at 2978 cm−1 (C=H stretching band), 1688 cm−1 (C-O- stretching band) and 1575 cm−1 (C=N=H bending band).

Yellows The non-invasive identiﬁcation of yellow dyes is often very difﬁcult: in particular, the discrimination of different types of yellow ochres with infrared spectroscopy has never been carried out in a conclusive way. Frequently, the IR spectra of ochres or earth pigments, found in both

114 ancient and modern paintings, correspond to the spectra of silica and silicates106. As the natural earth pigments were simply dug out from the ground, they contain a mixture of minerals, including clay and quartz. Clay and quartz are readily identiﬁed by their absorption bands, while the compounds responsible for the pigment color rarely present absorption bands in the mid-IR range. The color of red ochre is mainly due to anhydrous iron oxide (Hematite Fe2O3), while the color of yellow ochre is usually generated by hydrated iron oxides, most often goethite FeO·(OH)105-106. Fig. 32b shows the DRIFT spectra of four yellow iron-based pigments prepared with Gum Arabic: Burgundy Yellow Ochre (BYO), Iron Oxide Yellow (IOY), Jarosite from Cyprus (JC) and German Gold Ochre (GGO). All the spectra are very similar in intensity and band wavenumbers, especially the silicates ﬁngerprint area at 1100–1010 cm−1 (Si=O stretching mode of the kaolin) and near 3600 cm−1, assigned to the hydroxyl group8,90,107. Through the DRIFT spectroscopy, the discrimination of different ochres could be carried out evaluating the shape and position of the bands and their overtones: these can be attributed to the different chemical composition based on the region of provenience of the constitutive minerals. Burgundy Yellow Ochre (from France) and Iron Oxide Yellow show both similar peaks at 3690–3680 cm−1 and near 3650 cm−1, which can be assigned to the OH stretching from different yellow iron earth pigments108, while Jarosite and German Gold Ochre are characterized by the presence of silicates compounds at 1050 cm−1 (Si=O asymmetric stretching mode)109. The Iron Oxide Yellow is also characterized by two peaks centered at 2516 and 1792 cm−1

115 associated to carbonates110,111 and the Jarosite sample can be discriminated by the two peaks at 2045 and 1989 cm−1, corresponding to

3+ δ OH bands, characteristic of the Jarosite mineral KFe 3(SO4)2(OH)6. In Iron Oxide Yellow a shoulder can be observed at 1015 cm−1, due to the Si=O anti-symmetric stretching mode of kaolin, while at 922 and 883 cm−1 there are respectively two diagnostic peaks of kaolinite and goethite FeO·(OH). The band at about 3620 cm−1 is reported in literature as a typical signal of yellow ochre in Gum Arabic and egg white. The same band is also present in Burgundy Yellow Ochre, which is characterized by a shoulder at 982 cm−1, probably due to the Si=O asymmetric vibration. The peak at 916 cm−1 is characteristic of the German Gold Ochre and the Burgundy Yellow Ochre8. According to the above reported information, the DRIFT infrared spectra are suitable for the characterization and identification of the different yellow ochres that can be classified on the basis of their provenience and mineral composition. The Lead-tin Yellow is available in two varieties: the first (type I) is composed by lead stannate while the type II also contains silicon in the chemical formula112. The discrimination of Lead-tin Yellow pigments can be performed by Raman spectroscopy93 but it is nevertheless very complex because the two forms are very similar: type I is a Pb2SnO4, while type II is a PbSnO3 or PbSn1−xSixO3. According to literature, the only differences are melting temperature and time, and the possible presence of SiO2 in type II. This justifies the presence of the peaks at 3548 and 3508 cm−1 (O=H vibrations) only in type II and the signal of silicates compounds at 1056 cm−1, probably due to the presence of quartz in the chemical structure113. As shown in Fig. 32c, the several intense bands in the range

116

900–665 cm−1 (856, 818, 736, 713, 687, 665 cm−1) are typical of the metal-oxygen vibrations (Pb=O). These signals can be interpreted as a ﬁngerprint of type II. Instead, in type I the absence of peaks in the region near 900 cm−1 conﬁrms the similarity with the same pigment described in the work of Bagdzevičienė et al.114.

Fig. 32. DRIFT infrared spectra of Gum Arabic (GA) and egg white (EW) binders (a); of the four yellow iron-based pigments prepared with Gum Arabic Burgundy Yellow Ochre (BYO), Iron Oxide Yellow (IOY), Jarosite from Cyprus (JC) and German Gold Ochre (GGO) (b); of Lead-tin Yellow type I (LTY Type I) and Lead-tin Yellow type II (LTY Type II) in Gum Arabic (c); and of Naples Yellow (NY) and Saffron (SF) in Gum Arabic and of Litharge (LI) and Orpiment (OR) in egg white (d).

Fig. 32d shows the DRIFT spectra of Naples Yellow (NY) and Saffron (SF) in Gum Arabic and of Litharge (LI) and Orpiment (OR) in egg white. Naples

117

Yellow consists essentially of lead antimoniate106,112. The Litharge (or Massicot) is a yellow lead oxide pigment (PbO) obtained by the direct

112 oxidation of Pb or White Lead (PbCO3)2 Pb(OH)2 at about 300–400 °C . The Orpiment is a natural mineral, but it can also be artificially prepared. The natural Orpiment forms as a low temperature product in hydrothermal veins, as a volcanic sublimation product, as a hot spring deposit and in fire mines. It is often associated with stibnite, pyrite, Realgar, calcite and gypsum. The pigment is an arsenic trisulphide, which can be artificially obtained by sublimation of sulphur and arsenic oxide

115 As2O3 . Two particularly strong peaks at 699 cm−1 and 784 cm−1 characterize the DRIFT spectrum of Naples Yellow (NY)91 while the Litharge (LI) differs for its characteristic signal at 679 cm−1[116]. In the spectrum of LI other bands are present at 1736 cm−1, 1046 cm−1 and 844 cm−1, similar to the signals of lead white, due probably to the impurities coming from the synthesis of PbO using lead carbonate as starting material113,116. In the spectrum of Orpiment there is a band at 1233 cm−1, that was also found in the work of Kendix116. The spectrum of Saffron shows a peak at 3586 cm−1 assigned to the absorption of –OH, while according to literature117,118 the spectral region from ~ 1800 to 800 cm−1 can be considered as a saffron ﬁngerprint, in particular the bands in the region 1500–800 cm−1 are associated with the skeletal vibrations of the colorant components and have been attributed to –CH2-, CH3-, –OH, C=C, C=O, C=OC groups. As shown in the infrared spectrum of the Italian Saffron, reported by Anastasaki et al.119, the C=O stretching vibrations of the ester group lay at 1101 cm−1. Furthermore, the signals around ~ 980–920 and

118

~ 780–700 cm−1 result respectively from C=H (trans-) and C=H (cis-) out- of-plane vibrations.

Red and orange Fig. 33a shows the DRIFT spectra of the three main red ochres employed by the artists: Burgundy Red Ochre (BRO), Spanish Red Ochre (SRO) and Hematite (HE). The vibrational bands in the OH stretching region, respectively at 3689, 3619, 3536 cm−1, are the typical signals of ochres. The characteristic peak at 1867 cm−1 of the Burgundy Red Ochre is assigned to the presence of quartz (Si=O stretching): this signal can be found only in red ochres and not in the yellow ones, as already highlighted by Miliani et al.90. The peaks at around 927–928, 814–812 and 710–705 cm−1 are present exclusively in Burgundy and Spanish Red Ochre and not in Hematite, as they indicate the presence of different minerals within the pigment: the peaks at 927–928 are assigned to kaolinite or polymorphous Nacrite (O=H deformations), while the peaks at 814–812 and 710–705 are related respectively to the O=H out of plane bending and the Fe=O bending typical of Magnetite mineral120. Hematite is also characterized by a shoulder at about 1109 cm−1 assigned to the pure red iron oxide pigment121. Mineral Cinnabar (HgS) is a red mineral constituted by mercuric sulphide.

Red Cochineal is a natural organic dye based on carminic acid (C22H20O13), which was usually extracted from Cochineal insects. The lacquer investigated in this research was obtained in laboratory from the aqueous extract of the species Coccus cacti ﬁxed on calcium carbonate CaCO3. The extracted colored substance of the natural Madder dye (C26H28O14) is

119 obtained from the Rubia tinctorum plant and contains a mixture of hydroxyanthraquinone compounds, the most abundant ones being alizarin (1,2-dihydroxyanthraquinone), purpurin (1,2,4- triidrossiantrachinone) and a small part of the quinizarin (1,4- dihydroxyanthraquinone). Fig. 33b shows the spectra and relative assignments of Mineral Cinnabar (CI) in Gum Arabic, Cochineal lake (CO) in egg white, Madder (MD) in egg white. According to the unique available literature122, in the IR spectrum of natural Madder there are strong absorptions in the region of O=H stretching vibrations; in our case we found two maxima at 3553 and 3403 cm−1. In the spectrum of natural Cinnabar the strong band at 1129 cm−1 could be assigned to sulphate compounds. The spectrum of Cochineal shows many IR signals in the OH region at 3500–2500 cm−1, attributable to some strong intra- and inter- molecular hydrogen bonds. The principal constituent of Cochineal is carminic acid and FTIR absorption bands of the main component of carminic acid correspond to anthraquinone compounds. Another typical intense peak found in the spectrum of Cochineal123 is visible at 730 cm−1 (C=H bending), while other two signals (2515 and 1792 cm−1) are due to the presence of calcite employed to prepare the Cochineal lake91.

Green and blue

Malachite is a natural basic copper carbonate CuCO3 Cu(OH)2 and it is similar for composition to blue basic copper carbonate Azurite 2CuCO3

Cu(OH)2 except that it contains a greater amount of combined water. Both pigments present the absorption bands of carbonates in

120

2− − [90,102] correspondence of the vibration modes of CO3 and OH . Another important blue pigment is the precious Lapis Lazuli (Na,

Ca)8[(SO4,S,Cl)2(AlSiO4)6], sodium calcium aluminum silicate, which can be described rich in lazurite 2Na2Al2−Si2O6 NaS2, but also containing other blue colored minerals, such as haüyne, sodalite and nosean102. Natural indigo (C16H10N2O2) is among the oldest organic blue dyes to be used for textile dyeing and printing. The genuine dye is obtained from the processing of Indigofera plant's leaves. Phthalocyanine Blue, also called Monastral Blue, is a bright synthetic blue pigment from the group of phthalocyanine dyes (C32H16N8Cu). The pigment Egyptian Blue is a blue copper silicate, made by heating at 900 °C quartz with copper, calcium and a ﬂux (CaCuSi4O10). Another relatively unknown pigment investigated, but very similar for the basic structure and chemical stability

10 124 with the Egyptian Blue is the Han Blue (BaCuSi4O ) . Fig. 33c shows the DRIFT spectra of Malachite (MC), Azurite (AZ) and Lapis Lazuli (LL): the three infrared spectra are characterized by several bands that allowed the clear identiﬁcation of each pigment. Some interesting features in the spectrum of natural Azurite are the bands attributed to the carbonate ion situated at 1496 and 1448 cm−1 (ν3 antisymmetric

2− 90,102 −1 stretching of CO3 ) . The strong peak at 862 cm can be assigned to the out-of-plane bending (ν2)102; instead, the signals of OH vibrations fall from 3400 to 3550 cm−1 (OH stretching) and at 976 cm−1 (OH bending), while the signal at around 1833 cm−1 can be probably attributed to a

2− 90,102 combination band of the carbonate anion (ν1 + ν4, CO3 ) . As regards the spectral region at higher wavenumbers, the signals at 3878 and 3838 cm−1 can be attributed to the combination 2ν3 + δ(OH)102. In addition, as

121 reported by Vetter et al.91, the carbonate combination band ν1 + ν3 at about 2500 cm−1 (2505 cm−1 in Fig. 33c) can be used as a marker of Azurite125. According to literature90,91 the spectrum of natural ultramarine is characterized by the presence of residues such as calcite (CaCO3), an associated mineral in the natural pigment. In this case we observed its band at 2521 cm−1. The spectrum of natural ultramarine in Gum Arabic shows also a peak at 2952 cm−1, probably due to the C=H stretching bands of the binder104. In addition, the speciﬁc feature of natural ultramarine is visible at 2340 cm−1 as a strong and sharp band, reported in literature as absorption band of CO2 entrapped in the mineral lazurite, that occurs only in some natural ultramarines obtained from the Badakhshan mines in Afghanistan126. Instead, the presence of the mineral lazurite, responsible for the color in ultramarine pigment, can be associated to the band at around 681 cm−1[102]. The green carbonate mineral Malachite shows a band at 3440 cm−1 assigned to O=H stretching (ν3 of OH−) and two main peaks due to the OH bending at around 1071 and 908 cm−1[102,104]. Other signals, attributed to

2− −1 antisymmetric CO3 stretching modes, are visible at 1447 cm . Fig. 33d shows the DRIFT spectra of natural Indigo (IN), Phtalocyanine Blue (PB), Egyptian Blue (EB) and Han Blue (HB): the characteristic signals are associated to several bands that allow the clear identiﬁcation of each pigment. The DRIFT spectrum of Egyptian Blue spread in Gum Arabic shows signiﬁcant similarities with the IRUG infrared spectrum of the same pigment and, as reported in literature90,127, in the region of asymmetrical Si=O=Si stretching the pigment shows peculiar bands: a shoulder at 1250 cm−1, which can be attributed to a double silicate of Cuprorivaite

122

−1 (CaCuSi4O10), and two inverted bands at 1057 and 1004 cm , while two infrared bands appear in the region of symmetric Si=O=Si stretching at 818 and 698 cm−1. The spectrum of Han Blue shows strong similarity with the infrared spectrum of Egyptian Blue. Fig. 33d reports four bands at 1250, 1057, 1004 and 698 cm−1. According to literature, the infrared spectrum of Phthalocyanine Blue produces very distinctive sharp bands from 1700 to 700 cm−1 range, characteristics of the organic aromatic group112. The DRIFT spectrum of the synthetic organic pigment in Fig. 33d, shows several characteristic bands and has many similarities with the assignments performed for the FTIR spectrum of the CAMEO on-line database. In our case, the most representative signals found in the Phthalocyanine spectrum are: aromatics bands at 1616, 1515 and 1425 cm−1; 1346 cm−1 (C=H bending bands); 1296 and 1170 cm−1 (C=N stretching bands); 784 and 750 cm−1. The spectrum of natural indigo in Gum Arabic presents similarities with the FTIR assignments identiﬁed by Vetter et al.91 and, as reported in literature128, the signal at 1323 cm−1 is assigned to C=H in plane bending vibrations, and the two peaks at 759 and 719 cm−1 are assigned to C=H out of plane bending vibrations of trans-Indigo.

123

Figure 33. DIRFT spectra of red ochres: Burgundy Red Ochre (BRO), Spanish Red Ochre (SRO) and Hematite (HE) (a); of Mineral Cinnabar (CI) in Gum Arabic, Cochineal lake (CO) in egg white, Madder (MD) in egg white (b); of Malachite (MC), Azurite (AZ) and Lapis Lazuli (LL) (c); and of natural Indigo (IN), Phtalocyanine Blue (PB), Egyptian Blue (EB) and Han Blue (HB) (d).

White White Lead is a complex salt, containing both carbonate and hydroxide ions. It is composed by cerussite (PbCO3) and hydrocerussite 2PbCO3

Pb(OH)2. In Fig. 34a the spectra of White Lead in both binders can be clearly identiﬁed. Typical White Lead vibrations in Gum Arabic were found at 3540 and 1732 cm−1[91]. The peak at 3540 cm−1, shifted to 3538 cm−1 in egg white, it is a signal of hydrocerussite and it is representative of the OH stretching mode in basic carbonates90,93. The combination bands at lower wavenumbers are then assigned to the ν1+ ν4 mode of cerussite

124 and lay at 1733 and 1735 cm−1 for White Lead both in Gum Arabic and in egg white129. The band at 2400 cm−1 is frequently used for discriminating White Lead from the other carbonate pigments90. The band positioned at 2417 cm−1 for the pigment spread in Gum Arabic and at 2414 cm−1 for the

2− pigment in egg white, suggest the presence of cerussite: ν1 + ν3 (CO3 ) of the anhydrous and hydrous forms of lead white.

Fig. 34. DRIFT spectra of White Lead pigment (LW) in Gum Arabic (GA) and egg white (EW) (a); and of the yellow and blue colors identiﬁed in the Sutherland's artwork (b).

1.4.1.4 Statistical Analysis PCA–DA was performed using MarkerView. The dataset consisting of 50 samples (25 colorants prepared with both egg white and Gum Arabic) was divided into ﬁve groups based on the chemical class to which the colorants belong: carbonates (Malachite, Azurite, White Lead), sulphures (Orpiment, Realgar, Cinnabar), silicates/iron-oxides (Burgun- dy Yellow Ochre, Iron Oxide Yellow, Jarosite from Cyprus, German Gold Ochre, Burgundy Red Ochre, Spanish Red Ochre, Hematite, Lapis Lazuli, Egyptian Blue, Han Blue), lead-oxides (Lead-tin Yellow I and II, Naples Yellow, Litharge) and organics (Cochineal, Phthalocyanine Blue, Mad- der, Saffron, Indigo). PCA is a statistical method widely used to identify

125 clustering patterns in data, by reducing the dimensionality of the variables. To improve the separation of the five sample groups PCA-DA was performed and the discriminant scores D1 and D2, rather than principal components, were used. The results of the PCA-DA were displayed as the variable projection scores and loadings plots. The scores plot (Figure 35 left) represents the variance of the original variables while the loadings plot (Figure 35 right) describes the variable behavior and differences between the five groups. As shown in Figure 35 right, D1 and D2 scores are able to discriminate between the five different sample groups, representing the different chemical class of the colorants. On the score plot, five clusters can be distinguished: the silicates/iron-oxides (pink), the carbonates (red), the lead-oxides (yellow), the sulphures (green) and the organics (blue). At negative D1 scores there are the carbonates and the suplhides, while silicates are at positive D1 scores. At negative D2 scores there are the sulphides and the lead-oxides while at positive D2 there are the carbonates. The organic samples are more dispersed than the other samples, and are mainly characterized by positive scores along D1. The chemical markers, namely the bands that mostly contributed to the differences among the five groups, were explored by looking at the loadings plot in Fig. 35 right. The loading scores indicate the characteristic bands that are specific to a given group. Several signals were found significant for each given group: i.e. the silicates area characterized by the fingerprint band around 3600 cm−1, the sulphides by the band at 1129 cm−1 assigned to sulphate compounds, the lead oxides samples that present several intense bands in the range of 900–665 cm−1 (in particular

126

834, 780 cm−1), typical of the metal-oxygen vibrations, the carbonate samples characterized by the bands around 2400– 2500 cm−1. The organic group, which includes Cochineal (animal origin) Phthalocyanine Blue (synthetic origin), Madder, Indigo and Saffron (vegetal origin), is more dispersed, but the samples are distributed on the basis of the colorants origin: the two upper samples are from Cochineal and are characterized by the band at around 1700 cm−1, while the two lower samples are from Phthalocyanine Blue, characterize by some bands near 1200 cm−1 (1174 cm−1). The statistic was also able to discriminate the silicates/iron-oxides group from the lead-oxides group, which are two different types of oxides.

Fig. 35. Principal Component Analysis–Discriminant Analysis (PCA–DA) of the colorants divided into ﬁve groups based on the chemical/mineralogical class: carbonates (red points: Malachite, Azurite, White Lead), sulphides (green points: Orpiment, Realgar, Cinnabar), silicates/iron-oxides (pink point: Burgundy Yellow Ochre, Iron Oxide Yellow, Jarosite from Cyprus, German Gold Ochre, Burgundy Red Ochre, Spanish Red Ochre, Hematite, Lapis Lazuli, Egyptian Blue, Han Blue), lead- oxides (yellow points: Lead-tin Yellow I and II, Naples Yellow, Litharge) and organics (blue points: Cochineal, Phthalocyanine Blue, Madder, Saffron, Indigo). The scores plot (left) represents the variance of the original variables while the loadings plot (right) describes the variable behavior and differences between the ﬁve groups.

127

1.4.1.5 Real Case Study The method for the non-invasive identification of the colorants was then used to characterize the pigments employed by the painter Graham Sutherland (1903–1980) in his gouache from an Italian private collection shown in Fig. 36. The painting is a part of the famous collection of “stairs study” executed in 1950s. The analysis was carried out directly in situ, pointing the infrared beam on the gouache without touching the painting surface. Fig. 34b shows the infrared spectra of the blue and yellow colors of the Sutherland's gouache: by comparing the spectra with the DRIFT reference pigments database, it was possible to attribute the composition of the two pigments most widely used from the artist in this series of gouaches. The results were confirmed by art historians and by the palette used by the author, that was recorded in a manual on painting techniques125. The yellow identified in the Sutherland's gouache shows evident correspondences with the Iron Oxide Yellow present in the DRIFT spectra database. In the spectrum, it is present a doublet centered at 2516 and 1792 cm−1, associated to the carbonates of the ochres and other two diagnostic peaks at 920 and 883 cm−1, respectively of kaolinite and goethite FeO·(OH). In the region of OH stretching at about 3550-3500 cm−1, many bands typical of yellow ochres are visible. The blue characterized in the Sutherland's painting is similar to the ultramarine in Gum Arabic present in the DRIFT database. Although the pigment is in artificial form, as shown by the missing of the strong and sharp band at 2340 cm−1 typical of natural ultramarine, in the spectrum of the blue colorant the characteristic pigment absorptions are still

128 present at 2520 and 681 cm−1, as well as another signal at 3691 cm−1, in the region of OH stretching, due to the presence of reagent residues such as kaolin in the synthetic pigment; kaolin, in fact, it is used in the production of the artificial pigment.

Summary of results In this research the use of a handle portable DRIFT instrument for the non-invasive identification of colorants is presented. Approximately 50 colorants were analyzed with DRIFT spectroscopy. The colorants were prepared with Gum Arabic and egg white as binders. The method was able to identify and discriminate several yellow pigments like Orpiment, Litharge, lead–tin yellows, lead antimonite and all the yellow ochres. The characterization of many red and orange colorants like Realgar, Burgundy Red Ochre, Spanish Red Ochre, Hematite, Cinnabar, Cochineal and Madder was carried out. Blue and green colorants (Lapis Lazuli, Azurite, indigo, etc.) were also analyzed creating a complete and exhaustive spectral library. The library was then employed for the analysis of a Graham Sutherland's gouache, where Iron Oxide Yellow and ultramarine blue were identified with Gum Arabic as binder. Moreover, using DRIFT spectroscopy followed by PCA–DA, the separation of the colorants samples, which was not evident by visual analysis of the infrared spectra, was obtained, based on direct DRIFT analysis. Examination of the loading plots showed the characteristic bands of each given group. According to the chemical classification, DRIFT spectroscopy coupled to statistics may

129 be also employed for the classiﬁcation of colorants based on the chemical/mineralogical class or provenance. Among the advantages of this analytical method we can highlight that there is no need to touch the sample, there is no need to use the correction algorithms Kubelka–Munk (KM) and Kramers–Kronig (KK), because the specular contribution is low, the analysis is non-invasive, and it can be done directly in situ, which is very important in cultural heritage and art ﬁelds.

Fig. 36. Pictures of Sutherland's gouache, series of “stairs study”, private collection, 1950s

130

1.4.2 Applicability of a portable ATR device to monitoring the conservation state of plastic materials in modern and contemporary art.

In the following paragraph it is reported the development and application of an attenuated total reflection (ATR) spectroscopic method based on a portable infrared spectrometer, for the non-invasive characterization and monitoring of the degradation of plastics used in modern and contemporary art130.

Today, artworks partially or completely made of plastic materials can be found in almost all international museums and collections. The deterioration of these objects is now becoming evident, mainly because these synthetic materials are not designed for a long life and the characterization of their state of conservation can of help to curators and conservators.

During the last century, synthetic polymers have been widely employed by artists, mainly to explore new esthetic features and materials. The versatility (color, shape, strength) of plastics is one of the advantages that allowed the diffusion of these materials in the modern design and contemporary art: a recent example is the land art installation ‘‘The Floating Piers’’ of Christo and Jeanne-Claude, constituted by high-density polyethylene cubes (Lago d’Iseo, Italy, 2016).

The need to know more about the plastic deterioration mechanisms131- 133, and to develop new conservation strategies to preserve these artworks, have pushed up the interest of researchers in this field134,135.

131

Polymeric materials are present in almost all heritage collections across the world136,137, but at the same time these contemporary materials are less studied and their chemical decomposition mechanisms have not yet been fully elucidated. Several institutions around the world launched research programs on plastic conservation, like the famous research project ‘‘POPART’’ (preservation of plastic artefacts in museum collections), funded by the Europe Community and aimed at the preservation of plastic museum objects138,139. Synthetic and semisynthetic polymers degradation phenomena are mainly caused by environmental effects and by their chemical reactivity140-142. The cause of deterioration of plastic objects can be due to: (1) their composition, in particular when migration of plasticizers and other fillers can produce irreversible tacking and warping phenomena; (2) aging, i.e., many chemical additives used in the past are unstable along time; (3) environmental factors like pollutants, light, temperature and relative humidity, that can accelerate the degradation process; and (4), the improper handling and cleaning of the objects.

Many degradation phenomena on plastics used in design and contemporary art have already been observed and reported143-148.

Infrared (IR) spectroscopy is suitable for the qualitative and quantitative analysis of polymers degradation because it is reliable, fast and cost- effective and last but not least, it responds exactly to the modification involved in the degradations. The attenuated total reflection-Fourier transform infrared spectroscopy (ATR-FTIR) has been extensively used for the characterization of plastics objects and, in particular, coupled with principal component analysis (PCA), it allowed the discrimination of aged

132 foam samples149, and the identification of some synthetic resins used in works of art150. Portable infrared instruments have been already employed with success for the analysis of cultural heritage8,151 and new conservation materials98,152. The use of portable FTIR reflectance spectroscopy has already been investigated for the in situ analysis of historic plastics, to build a complete database of polymer spectra of a set of reference materials widely used in the cultural heritage community153 and employed as a robust method for infield prescreening of the chemical composition of plastics artworks and historical objects154. Nuclear magnetic resonance and in situ FTIR-ATR have been employed for the full characterization of the structure of plastic artifacts in museum environments155, while a multi-analytical approach (FTIR, optical and electronic microscopy, X-ray diffraction, and portable X-ray fluorescence) was used to study in detail the constituent materials, the manufacturing technique, and the state of conservation of a contemporary sculpture156.

Laser-induced fluorescence has been applied for the assessment of the damage, the biological growth, and the analysis of specific materials on various surfaces in cultural heritage, including modern synthetic materials of contemporary works of art, while the time-resolved photoluminescence spectroscopy and fluorescence lifetime imaging were used as sensors for the detection of the degradation of design objects made of plastic157.

The effect of multiple degradation agents and pollutants on polymers, has been studied through the use of a colorimeter with the aim to prioritize the most important environmental variables of interest for polymer

133 conservation158. But the degradation of these materials can often occur before it becomes detectable in the visible spectra, i.e., recognizable at a visual inspection. Cucci et al.159 investigated the possibility of extending the applications of fiber optic reflectance spectra to synthetic polymers, and proposed this technique as a new non-invasive analytical tool for making diagnosis on plastic artworks conservation in museum collections. Simulated photo-aging was employed to estimate the degradability and to investigate the main reasons of deterioration in four contemporary works of art realized in plastics160, and recently the development of non- invasive analytical methods coupled to statistics, for the monitoring and characterization of cultural heritage objects acquired interest, particularly for ancient and historical objects7,31.

In this study we investigated the applicability of a portable attenuated total reflection (ATR) infrared spectrometer for the non-invasive characterization and for monitoring the degradation of plastics used in modern and contemporary art. Several polypropylene and polycarbonate samples were artificially aged in solar box, simulating about 200 years of museum light exposure, and they were monitored with the portable ATR, creating an infrared library of the conservation state of plastics. Through the use of chemometric techniques like principal component analysis– linear discriminant analysis and partial least square—discriminant analysis, we built a robust degradation model of each material, that can be used to predict and classify the degradation state of artworks and to identify the priority of intervention in the museum collections. Portable ATR coupled to multivariate statistics can be employed for taking care of

134 plastic artworks as it is non-invasive, the analysis is very fast and it can be performed directly in situ.

1.4.2.1 Sample collection of the plastics: Polypropylene and Polycarbonate

Polypropylene is a thermoplastic polymer derived from petroleum which exhibits exceptional mechanical properties. The polymer can be accompanied by a wide variety of additives and plasticizers in order to enhance its stability, hardness and physical resistance161. The behavior of this material under the effect of strong accelerated degradation has been already studied162, while Llamas et al.163 investigated its attitude as contemporary art material using FTIR spectroscopy.

Polycarbonate is a thermoplastic polymer containing carbonate groups. It can be easily worked, molded, and thermoformed, because of these properties it found many applications in the fields of arts and design. This material has already been characterized after natural aging149.

For the present research, a total of five different polycarbonate samples and ten different polypropylene samples (all available from PROPLAST, Tortona, Italy) were employed. The samples were of different colors: black (carbon black), orange (perinone dye), brown (mixed blends of cellulose), blue (methylene blue), gray (carbon black–white metal oxides), transparent (natural) and white (inorganic pigments include titanium dioxide) for the polypropylenes and black (carbon black), orange (isoindolinone organic pigment), transparent (natural) and white (inorganic pigments include titanium oxide) for the polycarbonates.

135

1.4.2.2 Sample analysis, instrumentation and software

Fourier transform infrared spectral analysis of polycarbonate and polypropylene samples was conducted using ATR-FTIR. All the spectra were obtained using an Agilent 4100 Exoscan FTIR portable spectrometer (Agilent Technologies, Santa Clara, CA, USA). The spectra of the amples were obtained in absorption mode, over a wavelength window of 650– 4000 cm-1, at 32 scans per sample and a resolution of 4 cm-1. Background readings of air were established prior to data collection. The background was subsequently subtracted from each spectrum before any data treatment. Three infrared spectra of each sample were collected in different position, and the average spectra were used for identification purposes and for monitoring the samples conservation state over time.

The spectra were processed by OPUS v.2.3 (Bruker, Billerica, Massachusetts, USA), the statistical analysis and the data pre-treatment were performed by The Unscrambler 10 (Camo Inc., Norway) and by Statistica v7.1 (Statsoft, Tulsa, OK, USA).

1.4.2.3 Statistical analysis: PCA–LDA and PLS-DA

PCA is a multivariate pattern recognition method164 representing the objects, described by the original variables, into a new hierarchic reference system given by new variables, the Principal Components (PCs), linear combinations of the original ones, that are orthogonal one to each other. The experimental noise and random variations are collected in the last PCs, while the relevant information is present in the first PCs. PCA has already been applied to separate systematic information from

136 experimental noise and to describe the state of conservation of cultural heritage by using the most significant PCs6.

LDA is a Bayesian classification method providing the classification of the objects by taking into consideration the multivariate structure of the data165. In Bayesian methods, each class is usually described by a Gaussian multi-variate probability distribution and each object is assigned to a particular class if the so-called discriminant score is minimum. The discriminant score represents the position of the object along the discriminant direction. The variables used in the LDA model can be chosen by a stepwise algorithm that selects only the most discriminating ones. In this investigation, the PCs were used instead of the original variables, as PCA operates a very effective dimensionality reduction thanks to their intrinsic characteristic to sum up the systematic information contained in the original variables into a non-redundant, restricted set of PCs. Forward stepwise linear discriminant analysis was applied to select the most discriminant principal components. The classification performance of the LDA models was evaluated by the calculation of the non-error rate (NER%) that represents the percentage of overall correct assignments. The prediction performance of the models was evaluated using a leave- more-out cross-validation, based on a random sampling of the samples to be included in the training set: at each iteration 80% of the samples were included in the training set (used to build the classification model), while the remaining 20% of the samples was included in the test set (used to estimate the model classification performance), for a total of 1000 iterations9. This method constitutes a very severe validation of the classification performance and guarantees that no overfitting is present.

137

PLS-DA is a modification of the partial least squares (PLS) regression algorithm that makes it suitable for discriminant analysis. PLS establishes a relationship between one or more dependent variables (Y) and a group of descriptors (X)166-168. X and Y variables are modeled simultaneously, to find the latent variables (LVs) in X that will best predict the LVs in Y. These LVs (PLS components) are similar to the principal components calculated from PCA and are computed hierarchically169. A leave-one-out cross-validation is applied to evaluate the predictive ability and to select the optimal number of LVs on X and Y, to be used in the final model. In this work, it was applied a procedure for variable selection, based on the elimination of groups of not significant X variables, according to the minimum error in cross-validation. PLS was originally developed to model continuous responses, but it can be applied also for classification purposes (PLS-DA) by establishing an appropriate Y related to the association of each sample to a class.

In this case, we separated the samples into three groups on the basis of their degradation time; at this purpose, a Y variable was added to each dataset, coded so that 1 is attributed to the samples subjected to a short aging, 2 for the samples subjected to a medium aging, and 3 for the samples that were more degraded. The samples were arbitrary grouped, based on the progression of the degradation; the variable is semi- quantitative as the distance between 1, 2 and 3 does not follow a definite metric: here, PLS-DA is a hybrid as classification is performed by a regression method. The regression/classification is then carried out between the X-block variables and the Y just established.

138

1.4.2.4 Artificial light aging

Five polycarbonate samples and 10 polypropylene samples (1 x 1 cm of size) were subjected to artificial accelerated degradation for 800 hours in solar box, with light exposure of 700 Watt/m2, simulating about 200 years of museum light exposure. The maximum degradation temperature of the solar box was set at 35 °C, in order to avoid thermal effects. The samples were characterized before the solar box treatment and monitored during the accelerated degradation by using a portable ATR- FTIR.

The monitoring was performed at the following time intervals: 4, 28, 94, 123, 150, 320, 600 and 800 h of aging in the solar box (Fig. 37).

Fig. 37 Workflow of the research: the plastic samples were firstly characterized with the portable ATR-FTIR. The samples were then subjected to the artificial light aging in solar box that was monitored with ATR-FTIR. The data were used to build the statistical models in order to classify the samples on the basis of their degradation state

139

1.4.2.5 Characterization of plastic materials

Figure 38 shows the infrared spectrum of polypropylene and polycarbonate samples. The most significant vibrations identified in the infrared spectrum of polypropylene (Fig. 38-up) are: a band at 2953 cm-1

149,163,170 (asymmetric stretching of CH3) ; two strong sharp bands in the C- H stretch region at 2917 and 2851 cm-1, which indicate the presence of a long-chain hydrocarbon polymer104; of the two methylene absorption

149 bands (respectively, asymmetric and symmetric stretching of CH2) . A peak of the carbonyl group can be observed at 1735 cm-1, probably due to the presence of a stabilizer compound within the polymer163,171,172; two

-1 strong characteristic signals at 1460 cm (asymmetric bending of CH2) and 1377 cm-1 (symmetric bending of CH3)149,163,173 can be detected, while at 1179, 973 and 841 cm-1 there are three characteristic peaks of the polypropylene isotactic form149,174], together with a typical band at 720 cm-1 (methylene rocking vibration)163. The characteristic infrared bands observed in the spectrum of polycarbonate (Fig. 38-down) correspond to signals typical of this polymer, in particular there are two peaks at 2923 and 2858 cm-1, respectively, due to asymmetric and symmetric stretching of CH2 [149,153,175]. Another typical signal is in the region of C=O stretching at about 1772 cm-1 [149,153,175], while at 1601 and 1503 cm-1 there are two bands of aromatic carbon–carbon stretching149,175. Others intense signals appear in the region of the carbon–oxygen stretching from 1300 to 1000 cm-1 range, at 1228, 1188 and 1156 cm-1[153]. The spectrum shows the last three significant peaks at 1083, 1009 and 825 cm-1 attributable to the

140 presence of para-substituted phenol rings in the backbone of the polycarbonate polymer[149,175].

Figure 38. Spectroscopic characterization: infrared spectra of polycarbonate (down) and polypropylene (up). Samples unaged (black), aged for 94 h (blue) and aged for 800 h (red) are shown.

Each polypropylene and polycarbonate spectrum was smoothed (moving average with step 5), to reduce the noise, followed by a standard normal variate transformation.

141

In order to create a robust and representative aging model, able to explain the degradation of the selected plastics, for each of the two materials, we grouped together the monitored samples in three classes of aging as previously described: the first class included the samples aged for 0–28 h, the second class included the samples aged for 94–150 h, while the last class was constituted by the samples aged for 320–800 h. Moreover, the creation of three classes allowed to take into account the degradation phenomena in a better manner and with improved classification performance because, for example, in the range of 0–28 h the infrared spectra of the samples did not present significant changes. The number of samples, five different polycarbonate and ten different polypropylenes, allowed the building of a robust model because each sample was monitored 9X over the degradation. In fact, 45 and 90 infrared spectra, for polycarbonate and polypropylene, respectively, were employed to statistically assess their aging behavior.

1.4.2.6 PCA–LDA PCA, which allows to reduce the variable dimensionality, was carried out after range scaling [-1÷1], separately for polypropylene and polycarbonate samples. For both materials, the explained variance was mainly distributed along the first PCs as expected (polypropylene: PC1 = 66.21%, PC2 = 14.41%, PC3 = 6.67%, PC4 = 4.31%, PC5 = 3.28%; polycarbonate: PC1 = 64.70%, PC2 = 14.82%, PC3 = 6.26%, PC4 = 5.23%, PC5 = 2.51%). The development of a degradation model of the materials was then achieved by linear discriminant analysis applied to the first 20 PCs: the

142

use of the PCs instead of the original variables allows a dimensionality reduction and the elimination of the experimental noise. Moreover, the variable selection procedure was applied exploiting a stepwise algorithm

in forward search (Fto-enter=6) to select only the most effective PCs for classification purposes. The performance of the model in prediction was evaluated in cross-validation by a leave-more-out procedure (random sampling of 20% of the samples in the test set at each iteration; 1000 overall iterations). LDA provided good results for both models: for polycarbonate, the % of correct classification (% corr.) were 97, 100 and 100% respectively for the 0–28, 94–150 and 320–800 hours groups, while for polypropylene the results were slightly worse: 83, 60 and 85%, respectively, for the 0–28, 94–150 and 320–800 h groups, as shown in Table 22.

Table 22. Classification matrix of polypropylene (up) and polycarbonate (down) samples for PCA– LDA and PLS-DA both performed in cross- validation: rows correspond to the observed classification and columns corresponds the expected observations for each group. The last column corresponds to the corrected percentages of classification.

Polypropylene (h) PCA–LDA PLS-DA 0–28 h 94–150 h 320–800 h % corr. 0–28 h 94–150 h 320–800 h % corr.

0–28 35 6 1 83 37 5 0 88 94–150 13 21 1 60 2 32 1 91 320–800 0 5 29 85 0 4 30 88 Polycarbonate (h) PCA–LDA PLS-DA 0–28 h 94–150 h 320–800 h % corr. 0–28 h 94–150 h 320–800 h % corr.

0–28 32 1 0 97 32 1 0 97 94–150 0 10 0 100 1 9 0 90 320–800 0 0 9 100 0 1 8 89

These results are graphically represented in figure 39, which reports the samples along the two discriminant roots calculated by the LDA canonical analysis (left): for both materials the first root separates the first two

143 groups of samples (0–28 and 94–150 h) that were subjected to a shorter degradation (at negative values) from the 320–800 h group (at positive values), while the second root separates the 0–28 h group (at positive values) from the 94–150 and 320–800 h groups (at negative values). While for the polycarbonate, the three groups are well separated (Fig. 3- down), the 0–28 and 94–150 h groups of polypropylene (Fig. 39-up) are slightly overlapped, with several samples of the latter that are projected in the root space of the 0–28 h one. This is also confirmed by the percentage of correct classifications, which is lower than in the case of polycarbonate materials, where many samples are not assigned to the correct class. Since the principal components are linear combinations of the original variables, we calculated the weight of each original variable (absorbance) on the final model, in order to identify the variables characterized by the most significant contribution to the model, namely the spectral bands responsible accounting for the 3 different states of deterioration. Figure 39 (right) reports the original variables on the x axis and the weights of the corresponding root on the y axis: negative weights correspond to signals characterized by a large intensity in the 0–28 and 94–150 h groups and small signals in the other class, while positive weights correspond to signals with an opposite behavior. The very good prediction performances of the models and the severe validation guarantee, that no over-fitting is present: the method can be considered a good solution for the estimation of the degradation state of plastics.

144

Figure 39. Canonical scores of the PCA–LDA (left) and loading plot (right) of PC1 for polypropylene (up) and polycarbonate (down) samples. The first root separates in both cases the first two groups from the third, which contains the more degraded samples.

1.4.2.7 PLS-DA PLS-DA was applied with a backward elimination variable selection algorithm in leave one out cross- validation. The variables included in the final model represent the most discriminant variables according to PLS- DA. The final model for polypropylene contained the first two latent variables (LVs), that account approximately for 89% of the class belonging, while the final model for polycarbonate contains the first four latent variables, that explain approximately 90% of the class belonging.

145

Figure 40 shows the plot of the predicted versus measured values for the calibration models of PLS, for the three groups of polypropylene (up) and polycarbonate (down) samples, aged for different times. Not all the samples were correctly classified: the R2 values in fitting and cross-validation were, respectively, 0.879 and 0.826 for polypropylene and 0.900 and 0.869 for polycarbonate. Table 1 reports the % of correct classification in cross-validation (red samples in Fig. 40) for both materials, the results are very good for the two models: for polycarbonate, the % of correct classification (% corr.) were 97, 90 and 89%, respectively, for the 0–28, 94–150 and 320–800 h groups, while for polypropylene were 88, 91 and 88%, respectively, for the 0–28, 94–150 and 320–800 h groups.

Figure 40. Predicted versus measured values for the calibration models of PLS of the three groups of polypropylene (up) and polycarbonate (down) samples aged for different times.

Figure 41 reports the samples along the first two latent variables (left): for both materials the first latent variable separates the first two groups

146 of samples (0–28 and 94–150 h) that were subjected to a shorter degradation (at negative values) from the 320–800 h group (at positive values). Figure 41 (right) shows the loadings plot of the first two latent variables on the x axis and the weight on the corresponding latent variables on the y axis: negative weights for LV1 correspond to signals characterized by a large intensity in the 0–28 and 94–150 h groups, and a small signal in the other class, while positive weights for LV1 correspond to signals with an opposite behavior. The loading plot and the latent variables can be used to identify the variables that describe the model and that change along the degradation. The infrared spectra of polypropylene (Fig. 41-up) showed the appearance of new absorption bands in the range of 1800–1650 cm-1 that could indicate the formation of carbonyl groups in various oxidation products (aldehydes and ketones)171,176. Around 3300 cm-1, a new absorption band indicates the formation of the hydroxyl group – OH[176,177]. After the light degradation, a rising of the polypropylene spectrum background in the 1300–700 cm-1 region is generally observed. Other changes in relative absorbance intensity could be detected in the methyl group region, corresponding to 1377 cm-1, in the methylene group at 1460 cm-1, as well as the region of the polymer isotacticity, at 1179 and 973 cm-1[173,178]. Instead, for polycarbonate (Fig. 41-down), we observed some variations of the bands in the carbonyl stretching of the carbonate functional group at 1772 cm-1, which decreased with increasing time of exposition, while the bands in the 3300–2800 cm-1 region (carbon hydrogen stretching) showed quantitative changes and deformations as degradation

147 proceeded179-181, as well as in the peaks at 1601 and 1503 cm-1, which correspond to a ring stretching and skeletal vibration of phenyl groups. In the IR region of the carbon–oxygen stretch between 1300 and 1100 cm- 1, there were others relevant changes: the band centered at 1228 cm-1 increased during the degradation, while the ones at 1188 and 1156 cm-1 showed a significant decrease of the peaks intensity. The oxidation of polycarbonate aromatic rings was visible with the decrease of the signal at 825 cm-1 which indicated an increase of substituted phenolic species180. For what regards the other signals identified as significant in the discrimination between the classes, they all belong to regions were a relevant absorption was recorded for the samples; however, these signals corresponded in general to shoulders of the major bands. The overall prediction and classification performance of the models are very good for both materials.

Summary of results Although the degradation was monitored at more intervals, the samples were separated into three groups on the basis of their degradation time: samples subjected to a short, medium and strong aging. In general, the first two degradation times are less identifiable because the materials have a long induction period, during which the plastics are stable: for instance, if a sample is assigned to a border class, a wrong assignment is not completely wrong, while here no sample was classified to non- confining class. The information obtained by this approach could be used for the full and detailed characterization and diagnosis of plastic in museum

148

environments: this is very important because the deterioration of these objects is now becoming evident. Synthetic materials are not designed for a long life and the characterization of their state of conservation can help curators and conservators. In this example, through the use of a portable ATR, the rapid and in situ diagnosis of a selected artwork could be performed: the degradation models can be then used to predict and classify its degradation state and to identify the priority of intervention in the museum collections.

Figure 41. Score plot (left) and loading plot (right) of latent variables 1 and 2 of polypropylene (up) and polycarbonate (down) samples

The assessment of the level of degradation of a work of art made from plastic is very difficult, especially because there is no way to predict the

149 time of deterioration. Moreover, plastics have a relatively long induction period, during which the material is stable enough, which is followed by a rapid and irreversible degradation. In this work we showed that differences existing between the degradation of plastics could be detected by infrared spectroscopy and investigated by chemometric methods to build degradation models. The infrared spectra contain information that can be extracted using mathematical algorithms and it can be used to predict the degradation state of plastics and to establish the degradation state of an object. The non-invasive analysis and degradation monitoring of plastic surfaces was carried out with a portable ATR. Several polypropylene and polycarbonate samples, two types of plastics often used in modern and contemporary art, were artificially aged in solar box, simulating about 200 years of museum light exposure, and an infrared library of the conservation state of plastics was created. PCA–LDA and PLS-DA were employed to build robust degradation models that can be used to predict and classify the degradation state of artworks and to identify the priority of intervention in the museum collections. The models presented satisfactory classification performances so that we can conclude that this method could provide a useful support to the activity of conservators and curators. Portable ATR coupled to chemometrics can be employed for taking care of plastic artworks because it is not invasive, it is fast and the analysis can be performed directly in situ.

150

References 1. Giuffrida MG, Mazzoli R, Pessione E. Back to the past: deciphering cultural heritage secrets by protein identification. Appl Microbiol Biotechnol. 2018;102(13):5445‐5455. doi:10.1007/s00253-018-8963-z 2. Barberis E, Baiocco S, Conte E, Gosetti F, Rava A, Zilberstein G, Righetti PG, Marengo E, Manfredi M. Towards the non-invasive proteomic analysis of cultural heritage objects. Microchemical Journal Volume 139, June 2018, Pages 450-457 3. Toniolo, L.; D’Amato, A.; Saccenti, R.; Gulotta, D.; Righetti, P.G., J. J. Proteomics 2012, 75, 3365−3373. 4. Vinciguerra, R.; De Chiaro, A.; Pucci, P.; Marino, G.; Birolo, L., Microchem. J. 2016, 126, 341−348. 5. Bottari, F.; Oliveri, P.; Ugo, P. Biosens. Bioelectron. 2014, 52, 403−410. 6. Marengo, E.; Manfredi, M.; Zerbinati, O.; Robotti, E.; Mazzucco, E.; Gosetti, F.; Bearman, G.; France, F.; Shor, P. Development of a technique based on a multispectral imaging for the conservation of cultural heritage objects. Anal. Chem. 2011, 83, 6609−6618. 7. Manfredi, M.; Bearman, G.; France, F.; Shor, P.; Marengo, E., Quantitative multispectral imaging for the detection of parchment ageing caused by light: a comparison with ATR-FTIR, GC-MS and TGA analyses Int. J. Conserv. Sci. 2015, 6 (1), 3−14. 8. Manfredi, M.; Barberis, E.; Rava, A.; Robotti, E.; Gosetti, F.; Marengo, E. Portable diffuse reflectance infrared Fourier transform(DRIFT) technique for the non-invasive identification of canvas

151 ground: IR spectra reference collection Anal. Methods 2015, 7, 2313−2322. 9. Manfredi, M.; Robotti, E.; Bearman, G.; France, F.; Barberis, E.; Shor, P.; Marengo, E. Direct analysis in real time mass spectrometry for the non-destructive investigation of conservation treatments of cultural heritage J. Anal. Methods Chem. 2016, 2016, 6853591. 10. Fremout, W.; Dhaenens, M.; Saverwyns, S.; Sanyova, J.; Vandenabeele, P.; Deforce, D.; Moens, L. Anal. Chim. Acta 2010, 658, 156−162. 11. Leo G, Bonaduce I, Andreotti A, et al. Deamidation at asparagine and glutamine as a major modification upon deterioration/aging of proteinaceous binders in mural paintings. Anal Chem. 2011;83(6):2056‐ 2064. 12. Van der Werf, I. D.; Calvano, C. D.; Palmisano, F.; Sabbatini, L. Anal. Chim. Acta 2012, 718, 1−10. 13. Yan, H. T.; An, J. J.; Zhou, T.; Li, Y. H. Chin. Sci. Bull. 2013, 58, 2932−2937. 14. Fiddyment, S.; Holsinger, B.; Ruzzier, C.; Devine, A.; Binois, A.; Albarella, U.; Fischer, R.; Nichols, E.; Curtis, A.; Cheese, E.; Teasdale, M. D.; Checkley-Scott, C.; Milner, S. J.; Rudy, K. M.; Johnson, E. J.; Vnouček, J.; Garrison, M.; McGrory, S.; Bradley, D. G.; Collins, M. G. Proc. Natl. Acad. Sci. U. S. A. 2015, 112 (49), 15066−15071. 15. Heaton, K.; Solazzo, C.; Collins, M. J.; Thomas-Oates, J.; Bergström, E. T. J. Archaeol. Sci. 2009, 36, 2145−2154. 16. Corthals, A.; Koller, A.; Martin, D. W.; Rieger, R.; Chen, E. I.; Bernaski, M.; Recagno, G.; Dávalos, L. M. PLoS One 2012, 7 (7), e41244.

152

17. Zilberstein, G, Baskin E, Righetti, P.G., Maestro, Marguerite, morphine: The last years in the life of Mikhail Bulgakov, Journal of ProteomicsVolume 131, January 10, 2016, Pages 199-204. 18. Potenza, M.; Sabatino, G.; Giambi, F.; Rosi, L.; Papini, A. Anal. Bioanal. Chem. 2013, 405, 691−701. 19. Jurado-Loṕez, A.; De Castro, M. D. L. Anal. Bioanal. Chem. 2004, 380, 706−711. 20. Pauk, V.; Bartak, P.; Lemr, K. J. Sep. Sci. 2014, 37, 3393−3410. 21. Londero, P. S.; Lombardi, J. R.; Leona, M. Anal. Chem. 2013, 85, 5463−5467. 22. Gulmini, M.; Idone, A.; Diana, E.; Gastaldi, D.; Vaudan, D.; Aceto, M. Dyes Pigm. 2013, 98, 136−145. 23. Marcello Manfredi†‡$, Elettra Barberis†‡$, Fabio Gosetti†Orcid, Eleonora Conte‡, Giorgio Gatti†, Clara Mattu§, Elisa Robotti†, Gleb Zilberstein∥, Igor Koman⊥, Svetlana Zilberstein∥, Emilio Marengo*†Orcid, and Pier Giorgio Righetti*, Method for Noninvasive Analysis of Proteins and Small Molecules from Ancient Objects, Anal. Chem. 2017, 89, 6, 3310–3317. 24. Gillet, L.C.; Navarro, P.; Tate, S.; Röst, H.; Selevsek, N.; Reiter, L.; Bonner, R. Mol. Cell. Proteomics. 2012, 11, O111.016717. 25. Manfredi, M.; Martinotti, S.; Gosetti, F.; Ranzato, E.; Marengo, E. J. Proteomics 2016, 145, 3–10. 26. MacLean, B.; Tomazela, D.M., Shulman, N.; Chambers, M.; Finney, G.L.; Frewen, B.; Kern, R.; Tabb, D.L.; Liebler, D.C.; MacCoss, M.J. Bioinformatics 2010, 26, 966–968.

153

27. Cennini, C. Il libro dell’Arte; a cura di F. Brunello: Vicenza, Italy, 1971. 28. Salazaro, D. L’arte della miniatura nel sec. XIV, Napoli 1877 (and D.V. Thompson Jr.). An anonymous XIV Century treatise: De arte illuminandi; Yale University Press: New Haven, CT, 1933. 29. Armenini, G. B. Dei veri precetti della pittura; Appresso F. Tebaldini: Ravenna, Italy1587. 30. Manfredi, M.; Barberis, E.; Rava, A.; Poli, T.; Chiantore, O.; Marengo, E.; Anal. Bioanal. Chem. 2016, 408: 5711-22. 31. Manfredi, M.; Barberis, E.; Bearman, G.; Marengo, E. Digital Heritage International Congress, Digital Heritage 2015, 7419545, 449- 452. 32. Elettra Barberis, Marcello Manfredi, Emilio Marengo, Gleb Zilbersteinb, Svetlana Zilbersteinb, Alexander Kossolapov, Pier Giorgio Righetti, Leonardo's Donna Nuda unveiled, Journal of Proteomics 207 (2019) 103450. 33. P. Baglioni, D. Chelazzi, Nanoscience for the Conservation of Works of Art, RSC, Pages 307-308. 34. P. Vandenabeele, B. Wehling, L. Moens, H. Edwards, M. De Reu, G. Van Hooydonk, Analysis with micro-Raman spectroscopy of natural organic binding media and varnishes used in art, Anal. Chim. Acta 407 (2000) 261–274. 35. M. Matteini, A. Moles, La Chimica nel Restauro: I Materiali Dell'arte Pittorica, 1989. 36. [19] http://www.kremer-pigmente.com/it.

154

37. D. Scalarone, M.C. Duursma, J.J. Boon, O. Chiantore, MALDI-TOF mass spectrometry on cellulosic surfaces of fresh and photo-aged di- and triterpenoid varnish resins, J. Mass Spectrom. 40 (2005) 1527–1535, https://doi.org/10.1002/jms.893. 38. M. Matteini, A. Moles, A preliminary investigation of the unusual technique of Leonardo's mural "the last supper", Stud. Conserv. 24 (1979) 125–133. 39. D. Tal, A. Gordon, Intellectual leadership of Leonardo da Vinci: a Bibliometric study, Society 55 (2018) 549–554. 40. T.H. Newton, Leonardo da Vinci as mural painter: some observations on his materials and working methods, Arte Lombarda 66 (1983) 71–83. 41. H. Khun, A scientific examination of Leonardo's Last supper in Santa Maria delle Grazie, Metechnik Ed. Restauro 4 (1985) 24–51 Milano. 42. G. Basile, M. Marabelli, Nardini (Ed.), Leonardo: l'Ultima Cena: Indagini, Ricerche, Restauro, 2007 43. S. Dallongeville, N. Garnier, C. Rolando, C. Tokarski, Proteins in art, archaeology, and Paleontology: from detection to identification, Chem. Rev. 116 (2016) 2–79. 44. M. Manfredi, E. Barberis, M. Aceto, E. Marengo, Non-invasive characterization of colorants by portable diffuse reflectance infrared Fourier transform (DRIFT) spectroscopy and chemometrics, Spectrochim. Acta A Mol. Biomol. Spectrosc. 181 (2017) 171–179. 45. M. Manfredi, G. Williamson, D. Kronkright, E. Doehne, M. Jacobs, E. Marengo, G. Bearman, Measuring changes in cultural heritage objects

155 with reflectance transformation imaging, Proc. Dig. Herit. 1 (2013) 189– 192. 46. Giorgio Vasari, Vite de' più eccellenti pittori, scultori ed architettori, vol. 2, Firenze, Giunti, 1568. 47. M.P. Colombini, A. Andreotti, I. Bonaduce, F. Modugno, E. Ribechini, Analytical strategies for characterising organic paint media using GC–MS, Acc. Chem. Res. 395 (2010) 715–727. 48. A. Andreotti, I. Bonaduce, M.P. Colombini, G.I. Gautier, F. Modugno, E. Ribechini, Combined GC/MS analytical procedure for the characterization of glycerolipid, waxy, resinous, and proteinaceous materials in a unique paint microsample, Anal. Chem. 78 (2006) 4490– 4500. 49. I. Degano, J. La Nasa, E. Ghelardi, F. Modugno, M.P. Colombini, Model study of modern oil-based paint media by triacylglycerol profiling in positive and negative ionization modes, Talanta 161 (2016) 62–70. 50. A.M. Campos, F. Ricardo, E. Alves, A. Reis, D. Couto, P. Domingues, M.R.M. Domingues, Lipidomic investigation of eggs' yolk: changes in lipid profile of eggs from different conditions, Food Res. Int. 89 (2016) 177– 185. 51. F.D. Gunstone, J.L. Harwood, F.B. Padley, The Lipid Handbook, 2nd edition, Chapman & Hall Eds, London, 1994. 52. K. Van den Berg, J.J. Boon, I. Pastorova, L.F.M. Spetter, Mass spectrometric methodology for the analysis of highly oxidized diterpenoid acids in Old Master paintings, J. Mass Spectrom. 35 (2000) 512–533. 53. A. Sonia, M.A. Socaci, C. Socaciu, GC-MS analysis of rosemary essential oil, Bull. Univ. Agric. Sci. Vet. Med. 65 (2008) 405–409.

156

54. C. Maltese (Ed.), Le Tecniche Artistiche, Mursia Ed, Milano, 1973, pp. 343–358. 55. P. Brambilla Barcilon, La Mia Vita con Leonardo, Mondadori Electa Ed, Milano, 2015. 56. Raffaella Bianucci, Robert D. Loynes, M. Linda Sutherland, Rudy Lallo, Gemma L. Kay, Philippe Froesch, Mark J. Pallen, Philippe Charlier and Andreas G. Nerlich, Forensic Analysis Reveals AcuteDecompensation of Chronic Heart Failure in a3500-Year-Old Egyptian Dignitary, J Forensic Sci, September 2016, Vol. 61, No. 5. 57. J. B. Coltherd, The domestic fowl in Ancient Egypt, International Journal of Avian Science, Vol. 108, Issue2, April 1966, Pages 217-223. 58. Jamil Aarbiou, Marloes Ertmann, Sandra van Wetering, Peter van Noort, Denise Rook, Klaus F. Rabe, Sergey V. Litvinov, J. Han J. M. van Krieken, Willem I. de Boer and Pieter S. Hiemstra, Human neutrophil defensins induce lung epithelial cell proliferation in vitro, Journal of Leukocyte Biology Vol. 72, July 2002. 59. Steinwede K, Maus R, Bohling J, et al. Cathepsin G and neutrophil elastase contribute to lung-protective immunity against mycobacterial infections in mice. J Immunol. 2012;188(9):4476‐4487. 60. Yang F, Haile DJ, Coalson JJ, Ghio AJ. Haptoglobin in lung defence. Redox Rep. 2001;6(6):372‐374. 61. Cappellini E, Jensen LJ, Szklarczyk D, et al. Proteomic analysis of a pleistocene mammoth femur reveals more than one hundred ancient bone proteins. J Proteome Res. 2012;11(2):917‐926.

157

62. Maria Perla Colombini,Francesca Modugno,Flora Silvano, Massimo Onor, Characterization of the Balm of an Egyptian Mummy from the Seventh Century B.C., Studies in Conservation, Vol. 45, Issue 1, 2000. 63. Raffaella Bianucci, Michael E. Habicht, Stephen Buckley, Joann Fletcher, Roger Seiler, Lena M. Öhrström, Eleni Vassilika, Thomas Böni, and Frank J. Rühli, Shedding New light on the 18th dynasty mummies of the royal architect Kha and his spouse Merit, PLoS One. 2015; 10(7): e0131916. 64. Jones J, Mirzaei M, Ravishankar P, et al. Identification of proteins from 4200-year-old skin and muscle tissue biopsies from ancient Egyptian mummies of the first intermediate period shows evidence of acute inflammation and severe immune response. Philos Trans A Math Phys Eng Sci. 2016;374(2079):20150373. 65. Gomaa Abdel-Maksouda, Abdel-Rahman El-Aminb. A review on the materials used during the mummification processes in Ancient Egypt. Mediterranean Archaeology and Archaeometry, Vol. 11, No. 2, pp. 129- 150. 66. Charrié-Duhaut A, Connan J, Rouquette N, Adam P, Barbotin C, de Rozière M-F, Tchapla A, Albrecht P. The canopic jars of Rameses II: real use revealed by molecular study of organic residues. J Archaeol Sci. 2007; 34:957–967. 67. Assimopoulou AN, Papageorgiou VP. GC-MS analysis of penta- and tetra-cyclic triterpenes from resins of Pistacia species. Part I. Pistacia lentiscus var. Chia. Biomed Chromatogr. 2005;19(4):285‐311. 68. Daifas DP, Smith JP, Blanchfield B, Sanders G, Austin JW, Koukoutisis J. Effects of mastic resin and its essential oil on the growth of

158 proteolytic Clostridium botulinum. Int J Food Microbiol. 2004;94(3):313‐ 322. 69. Nicholson TM, Gradl M, Welte B, et al. Enlightening the past: analytical proof for the use of Pistacia exudates in ancient Egyptian embalming resins. Journal of Separation Science. 2011 Dec;34(23):3364- 3371. 70. Baumann B.B., 1960, The Botanical Aspects of Ancient Egyptian Embalming and Burial, Economic Botany 14 (1), 84-104. 71. Mills JS, White R, The identity of the resins from the late bronze age shipwreck at Uluburun (Kaş), Archeometry 31, 1 (1989), 37-44. 72. Loynes R.D., Charlier P., Froesch P., Houlton T.M.R., Lallo R., Di Vella G., Bianucci R. Virtopsy shows a high status funerary treatment in an early 18th Dynasty non-royal individual. Forensic Science, Medicine and Pathology, DOI 10.1007/s12024-017-9879-0. 73. Zahi A. Hawass, Sahar Saleem, Scanning the Pharaohs: CT Imaging of the New Kingdom Royal Mummies, AUC press, 2016, EAN: 9789774166730. 74. Wali, K. C. Cremona Violins: A Physicist’s Quest for the Secrets of Stradivari. World Scientific Publishing Co Pte Ltd, 2014. 75. Fiocco, G.; Rovetta, T.; Gulmini, M.; Piccirilo, A.; Licchelli, M.; Malagodi, M. Spectroscopic Analysis to Characterize Finishing Treatments of Ancient Bowed String Instruments. Appl. Spectrosc. 2017, 71, 2477- 2487. 76. Sacconi, S. F. The “Secrets” of Stradivari. Libreria del Convegno: Cremona, 1979.

159

77. Echard, J. P.; Lavédrine, B. Review on the characterisation of ancient stringed musical instruments varnishes and implementation of an analytical strategy. J. Cult. Herit. 2008, 9, 420-429. 78. Echard, J. P.; Benoit, C.; Peris-Vicente, J.; Malecki, V.; Gimeno- Adelantado, J. V.; Vaiedelich, S. Gas chromatography/mass spectrometry characterization of historical varnishes of ancient Italian lutes and violin. Anal. Chim. Acta. 2007, 584, 172-180. 79. Echard, J. P.; Lavédrine, B.; von Bohlen, A.; Le Ho, A. S.; Paris, C.; Bellot-Gurlet, L.; Soulier, B. ; Lattuati-Derieux, A.; Thao, S. ; Robinet, L.; Lavedrine, B.; Vaiedelich, S. The Nature of the Extraordinary Finish of Stradivari’s Instruments. Angew. Chem. Int. Ed. 2010, 49, 197-201. 80. Invernizi, C.; Daveri, A.; Rovetta, T.; Vagnini, M.; Licchelli, M.; Cacciatori, F.; Malagodi, M. A multi-analytical non-invasive approach to violin materials: The case of Antonio Stradivari “Hellier” (1679). Microchem. J. 2016, 124, 743-750. 81. Invernizzi, C.; Fichera, G. V.; Licchelli, M.; Malagodi, M. A non- invasive stratigraphic study by reflection FT-IR spectroscopy and UV- induced fluorescence technique: The case of historical violins. Microchem. J. 2018, 138, 273-281. 82. Tai, B.H. Stradivari’s Varnish: A Review of Scientific Findings – Part I. J. Violin Soc. Am.: VSA Papers, Summer 2007, pp 119-144. 83. Skrodzka, E.; Linde, B.; Krupa, A. Modal Parameters of Two Violins with Different Varnish Layers and Subjective Evaluation of Their Sound Quality. Arch. Acoust. 2013, 38, 75-81. 84. Kasprzok LM, Off-line pyrolysis solid-phase microextraction gas chromatography-mass spectrometry with on-fiber silylation for the

160 characterization of historic stringed instruments. Master’s thesis in analytical chemistry of organic materials. A.A. 2017/2018. Pages 12-20. 85. Bonizzoni, L.; Canevari, C.; Galli, A.; Gargano, M.; Ludwig, N.; Malagodi, M.; Rovetta, T. A multidisciplinary materials characterization of a Joannes Marcus viol (16th century). Heritage Sci. 2014. 86. Manfredi M, Barberis E, Bearman G, Marengo E. Portable non- invasive imaging method for monitoring the conservation of frescoes. 2015 Digital Heritage 2015, 2, 449–452. 87. Aldrovandi A, Picollo M, Metodi di documentazione e indagini non invasive sui dipinti, Il prato, 2007, ISBN 978-88-89566-62-6. 88. Ronald E. Alley, 1996, Algorithm Theoretical Basis Document for Decorrelation Stretch, NASA, JPL. 89. Archeologia invisibile. Catalogo della mostra (Torino, 13 marzo 2019-6 gennaio 2020), Franco Cosimo Panini Ed., 2019. 90. C. Miliani, F. Rosi, A. Daveri, B.G. Brunetti, Reflection infrared spectroscopy for the non-invasive in situ study of artists' pigments, Appl. Phys. A Mater. Sci. Process. 106 (2) (2012) 295–307. 91. W. Vetter, M. Schreiner, Characterization of pigment-binding media system – comparison of non-invasive in-situ reflection FTIR with transmission FTIR microscopy, Morana RTD e-PS. 8 (2011) 10–22. 92. C. Miliani, F. Rosi, B.G. Brunetti, A. Sgamellotti, In situ non-invasive study of artworks: the MOLAB multitechnique approach, Acc. Chem. Res. 43 (6) (2010) 728–738. 93. M. Aceto, A. Agostino, G. Fenoglio, A. Idone, M. Gulmini, M. Picollo, P. Ricciardi, J.K. Delaney, Characterisation of colourants on

161 illuminated manuscripts by portable fibre optic UV-visible-NIR reflectance spectrophotometry, Anal. Methods 6 (2014)1488–1500. 94. Z. Szökefalvi-Nagy, I. Demeter, A. Kocsonya, I. Kovács, Non- destructive XRF analysis of paintings, Nucl. Instrum. Methods Phys. Res., Sect. B 226 (1–2) (2004) 53–59. 95. L. De Viguerie, V.A. Sole, P.Walter, Multilayers quantitative X-ray fluorescence analysis applied to easel paintings, Anal. Bioanal. Chem. 395 (7) (2009) 2015–2020. 96. M. Alfeld, J.V. Pedroso, HommesM. Van Eikema, G. Van der Snickt, G. Tauber, J. Blaas, M. Haschke, K. Erler, J. Dik, K. Janssens, A mobile instrument for in situ scanning macro-XRF investigation of historical paintings, J. Anal. At. Spectrom. 28 (5) (2013) 760–767. 97. F.C. Izzo, V. Vitale, C. Fabbro, H. Van Keulen, Multi-analytical investigation on felt-tip pen inks: formulation and preliminary photo- degradation study, Microchem. J. 124 (2016) 919–928. 98. M. Manfredi, E. Barberis, A. Rava, T. Poli, O. Chiantore, E. Marengo, An analytical approach for the non-invasive selection of consolidants in rubber artworks, Anal. Bioanal. Chem. 408 (21) (2016) 5711–5722. 99. Manfredi M, Williamson G, Kronkright D, Doehne E, Jacobs M, Marengo E, Bearman G, Measuring changes in cultural heritage objects with Reflectance Transformation Imaging, Proceedings of the DigitalHeritage 2013, Vol. 1, 189-192. 100. C. Miliani, F. Rosi, A. Burnstock, B.G. Brunetti, A. Sgamellotti, Non- invasive in-situ investigations versus micro-sampling: a comparative study

162 on a Renoirs painting, Appl. Phys. A Mater. Sci. Process. 89 (4) (2007) 849– 856. 101. R.A. Eisenbeis, Avery RB (1972) Discriminant Analysis and Classification Procedures: Theory and Applications, Lexington books, 1972. 102. C. Zaffino, V. Guglielmi, S. Faraone, A. Vinaccia, S. Bruni, Exploiting external reflection FTIR spectroscopy for the in-situ identification of pigments and binders in illuminated manuscripts. Brochantite and posnjakite as a case study, Spectrochim. Acta A Mol. Biomol. Spectrosc. 136 (PB) (2015) 1076–1085. 103. F. Brunello, in: Neri Pozza (Ed.), De arte illuminandi (XIV sec.), 1992 (a cura di). 104. M.R. Derrick, D. Stulik, J.M. Landry, Infrared Spectroscopy in Conservation Science, Scientific Tools for Conservation. Los Angeles, The Getty Conservation Institute, 1999. 105. A. Karpowicz, Ageing and deterioration of proteinaceous media, Stud. Conserv. 26 (4) (1981) 153–160. 106. R. Newman, Some applications of infrared spectroscopy in the examination of painting materials, J. Am. Inst. Conserv. 19 (1) (1979) 42– 62. 107. J. Van derWeerd, A. van Loon, J.J. Boon, FTIR studies of the effects of pigments on the aging of oil, Stud. Conserv. 50 (1) (2005) 3–22. 108. M. JournAbdel-Ghani, M. Mahmoud, Spectroscopic investigation on paint layers of Sabil- Kuttab Umm 'Abbas ceiling,Mohammed Ali Era in Cairo, Egypt: identification of unusual pigment and medium, J. Archaeol. Restor. Stud. 3 (2) (2013) 95–105.

163

109. F. Rosi, C. Miliani, C. Clementi, K. Kahrim, F. Presciutti, M. Vagnini, V. Manuali, A. Daveri, L. Cartechini, B.G. Brunetti, An integrated spectroscopy approach for the non-invasive study of modern art materials and techniques, Appl. Phys. A Mater. Sci. Process. 100 (3) (2010) 613–624. 110. S.A. Sandford, M. Nuevo, G.J. Flynn, S. Wirick, Mid-infrared study of samples from several stones from the Sutter's mill meteorite, Meteorit. Planet. Sci. 49 (11) (2014) 2017–2026. 111. R. Ravisankar, G. Senthilkumar, S. Kiruba, A. Chandrasekaran, P.P. Jebakumar,Mineral analysis of coastal sediment samples of Tuna, Gujarat, India, Indian J. Sci. Technol. 3 (7) (2010) 775–781. 112. L. Campanella, Chimica per l'arte, Zanichelli, 2007. 113. R. Mazzeo, S. Prati, M. Quaranta, E. Joseph, E. Kendix, M. Galeotti, Attenuated total reflection micro FTIR characterization of pigment– binder interaction in reconstructed paint films Anal, Bioanal. Chem. 392 (1–2) (2008) 65–76. 114. J. Bagdzevičienė, G. Niaura, E. Garškaitė, J. Senvaitienė, J. Lukšėnienė, S. Tautkus, Spectroscopic analysis of lead tin yellow pigment in medieval necklace beads from Kernavė-Kriveikiškės cemetery in Lithuania, Chem. Aust. 22 (4) (2011) 216–222. 115. Fitzhugh, E.W., Orpiment and Realgar, in Artists’ Pigments, A Handbook of Their History and Characteristics, Vol 3: E.W. Fitzhugh (Ed.) Oxford University Press, 1997. 116. E.L. Kendix, Transmission & Reflection (Atr) Far-Infrared Spectroscopy Applied in the Analysis of Cultural Heritage

164

Materials(Doctoral thesis) 2009 (http://amsdottorato.unibo.it/2266/1/Kendix_Elsebeth_tesi.pdf). 117. P.A. Tarantilis, A. Beljebbar, M. Manfait, M. Polissiou, FT-IR, FT- Raman spectroscopic study of carotenoids from saffron (Crocus sativus L.) and some derivatives, Spectrochim. Acta A Mol. Biomol. Spectrosc. 54 (4) (1998) 651–657. 118. S.A. Ordoudi, M. De losMozos Pascual, M.Z. Tsimidou, On the quality control of traded saffron bymeans of transmission Fourier- transform mid-infrared (FT-MIR) spectroscopy and chemometrics, Food Chem. 150 (2014) 414–421. 119. C. Anastasaki, C. Kanakis, L. Pappas, C.P. Maggi, M. Del Campo, G.L. Carmona, M.G. Alonso, M. Polissiou, Differentiation of saffron from four countries by mid-infrared spectroscopy and multivariate analysis, Eur. Food Res. Technol. 230 (2010) 571–577. 120. R.S. Kumara, P. Rajkumarb, Characterization of minerals in air dust particles in the state of Tamilnadu, India through ftir spectroscopy, Infrared Phys. Technol. 67 (2014) 30–41. 121. E.M. El Afifi, M.F. Attallah, E.H. Borai, FTIR studies of the effects of pigments on the aging of oil, utilization of natural hematite as reactive barrier for immobilization of radionuclides from radioactive liquid waste, J. Environ. Radioact. 151 (1) (2016)156–165. 122. V.A. Stikhin, A.I. Ban'kovskii, M.E. Perel'son, Lucidin primeveroside from Rubia iberica. II, Chem. Nat. Compd. 4 (5) (1968) 231–233. 123. L. Fuster López, M.F. Mecklenburg, D.L. Yusá Marco, S. Vicente Palomino, A.F. Batista dos Santos, Effects of mordants on the mechanical

165 behaviour of dyed silk fabrics: preliminary tests on cochineal dyestuffs, Arché 2 (2007) 115–120http://hdl.handle.net/10251/32241. 124. B. Heinz, The invention of blue and purple pigments in ancient times, Chem. Soc. Rev. 36 (2007) 15–30. 125. J. Russell, A study of the materials and techniques of Francis Bacon (1909–1992), Doctoral thesis, Northumbria University, 2010, http://nrl.northumbria.ac.uk/3156/. 126. C. Miliani, F. Rosi, I. Borgia, P. Benedetti, B.G. Brunetti, A. Sgamellotti, Fiber-optic Fourier transformmid-infrared reflectance spectroscopy: a suitable technique for in situ studies of mural paintings, Appl. Spectrosc. 61 (3) (2007) 293–299. 127. Y. Xia, Q. Ma, Z. Zhang, Z. Liu, J. Feng, A. Shao,W.Wang, Q. Fu, Development of Chinese barium copper silicate pigments during the Qin Empire based on Raman and polarized light microscopy studies, J. Archaeol. Sci. 49 (1) (2014) 500–509. 128. P. García-Fernández, M. Moreno, J.A. Aramburu, Origin of the exotic blue color of copper-containing historical pigments, Inorg. Chem. 54 (1) (2015) 192–199. 129. F. Rosi, A. Daveri, C. Miliani, G. Verri, P. Benedetti, F. Piqué, B.G. Brunetti, A. Sgamellotti, Non-invasive identification of organic materials in wall paintings by fiber optic reflectance infrared spectroscopy: a statistical multivariate approach, Anal. Bioanal. Chem. 395 (7) (2009) 2097–2106. 130. M. Manfredi, E. Barberis, E. Marengo. Prediction and classification of the degradation state of plastic materials used in modern and contemporary art, Appl. Phys. A (2017) 123:35.

166

131. F. Cappitelli, P. Principi, C. Sorlini, Biodeterioration of modern materials in contemporary collections: can biotechnology help?, Trends Biotechnol. 24, 350–354 (2006). 132. D. Atek, N. Belhaneche-Bensemra, FTIR investigation of the specific migration of additives from rigid poly (vinyl chloride). Eur. Polym. J. 41, 707–714 (2005). 133. T.R. Newman, Plastics as an Art Form (Chilton Book, Radnor, 1969). 134. A. Werner, Synthetic materials in art conservation. J. Chem. Educ. 58, 321–324 (1981). 135. C.V. Horie, Materials for Conservation: Organic Consolidants, Adhesives and Coatings (Butterworths, London, 1994). 136. E. Then, V. Oakley, A survey of plastic objects at the Victoria and Albert Museum. V&A Conser. J. 06, 11–14 (1993) 137. Y. Shashoua, C. Ward, plastics: modern resins with ageing problems, in Proceedings of SSCR 2nd Resins Conference, Resins-Ancient and Modern, ed. by Wright M.M., J.H. (Scottish Society for Conservation and Restoration, Townsend Edinburgh; 1995), pp. 33–37. 138. http://popart.mnhn.fr/ 139. B. Lave´drine, R. Rivenc, M. Schilling, POPART. An International Research Project on the Conservation of Plastics (The GCI Newsletter, Getty Conservation Institute, Los Angeles; 2009), pp.10–12 140. N.S. Allen, M. Edge, Fundamentals of Polymer Degradation and Stabilization (Springer, New York, 1992). (Ed. by S. Halim Hamid). 141. S. Halim Hamid, Handbook of Polymer Degradation, 2nd edn. (CRC Press, Boca Raton, 2000).

167

142. H.W. Moeller (ed.), Progress in Polymer Degradation and Stability Research (Nova Science Publishers, Hauppauge, 2008). 143. Y. Shashoua, Conservation of Plastics. Material Science, Degradation and Preservation (Butterworth-Heinemann LTD, London, 2008) 144. H.C. Flemming, Relevance of biofilms for the biodeterioration of surfaces of polymeric materials. Polym. Degrad. Stab. 59, 309–315 (1998) 145. F. Cappitelli, Y. Shashoua, E. Vassallo, Macromolecules in Cultural Heritage. Macromol. Symp. 238, 1–104 (2006) 146. F. Cappitelli, Synthetic Polymers in Cultural Heritage, in Microbiology:Fundamental Studies in Conservation Science, ed. by 147. R. Mitchell, C.J. McNamara (ASM Press, Washington, 2012), pp. 153–166 148. F. Cappitelli, S. Vicini, P. Piaggio, P. Abbruscato, E. Princi, A. Casadevall, J.D. Nosanchuk, E. Zanardini, Investigation of fungal deterioration of synthetic paint binders using vibrational spectroscopic techniques. Macromol. Biosci. 5, 49–57 (2005). 149. G. Mitchell, F. France, A. Nordon, P.L. Tang, L.T. Gibson, Assessment of historical polymers using attenuated total reflectance- Fourier transform infra-red spectroscopy with principal component analysis. Herit. Sci. 1, 28 (2013). 150. M.T. Domenech-Carbo, A. Domenech Carbo, J.V. Gimeno- Adelantado, F. Bosch-Reig, Identification of synthetic resins used in works of art by Fourier transform infrared spectroscopy. Appl. Spectrosc. 5(12), 1590–1602 (2001).

168

151. I. Arrizabalaga, O. Gomez-Laserna, J. Aramendia, G. Arana, J.M. Madariaga, Spectrochim. Acta Part A 129, 259–267 (2014). 152. E. Barberis, E. Marengo, M. Manfredi, Quantitative imaging to study new conservation materials, in 2015 Digital Heritage International Congress, Digital Heritage 2015, 7419538, (2015), pp. 417–420. 153. D. Saviello, L. Toniolo, S. Goidanich, F. Casadio, Non-invasive identification of plastic materials in museum collections with portable FTIR reflectance spectroscopy: reference database and practical applications. Microchem. J. 124, 868–877 (2016) 154. C. Cucci, G. Bartolozzi, V. Marchiafava, M. Picollo, E. Richardson, Study of semi-synthetic plastic objects of historic interest using non- invasive total reflectance FT-IR. Microchem. J. 124, 889–897 (2016). 155. A. Pereira, A. Candeias, A. Cardoso, D. Rodrigues, P. Vandenabeele, A.T. Caldeira, Non-invasive methodology for the identification of plastic pieces in museum environment—a novel approach. Microchem. J. 124, 846–855 (2016). 156. B. Salvadori, E. Cantisani, M.P. Colombini, C.G.R. Tognon, Painted fiberglass-reinforced contemporary sculpture: investigating composite materials, techniques and conservation using a multi-analytical approach. Appl. Spectrosc. 70(1), 174–185 (2016). 157. A. Nevin, A. Cesaratto, S. Bellei, C. D’Andrea, L. Toniolo, G. Valentini, D. Comelli, Time-resolved photoluminescence spectroscopy and imaging: new approaches to the analysis of cultural heritage and its degradation. Sensors 14, 6338–6355 (2014). 158. G. Pastorelli, C. Cucci, O. Garcia, G. Piantanida, A. Elnaggar, M. Cassar, M. Strlic, Environmentally induced colour change during natural

169 degradation of selected polymers. Polym. Degrad. Stab. 107, 198–209 (2014). 159. C. Cucci, L. Bigazzi, M. Picollo, Fibre optic reflectance spectroscopy as a non-invasive tool for investigating plastics degradation in contemporary art collections: a methodological study on an expanded polystyrene artwork. J. Cult. Herit. 14, 290–296 (2013). 160. M. Lazzari, A. Ledo-Sua´rez, T. Lo´pez, D. Scalarone, M.A. Lo´pez- Quintela, Plastic matters: an analytical procedure to evaluate the degradability of contemporary works of art. Anal. Bioanal. Chem. 399, 2939–2948 (2011). 161. J.A. Brydson, Plastics Materials, 7th edn. Butterworth, London, 2000. 162. D. Song, J. Gao, X. Li, L. Lu, Evaluation of aging behavior of polypropylene in natural environment by principal component analysis. Polym. Test. 33, 131–137 (2014). 163. R. Llamas, M.C. Talamantes, An analytical study of polypropylene as a support for paint layers. From concept to material in contemporary art. J. Cult. Herit. 15, 136–143 (2014). 164. D.L. Massart, B.G.M. Vanderginste, S.N. Deming, Y. Michotte, L. Kaufman, Chemometrics: A Textbook, vol. 3 (Elsevier, Amsterdam,1988). 165. W.R. Klecka, Discriminant Analysis (Sage Publications Inc., Newbury Park, 1980). 166. D.L. Massart, B.G.M. Vanderginste, S.N. Deming, Y. Michotte, L. Kaufman, Chemometrics: A Textbook, vol. 3 (Elsevier, Amsterdam,1988). 167. D. Kleinbaum, L. Kupper, K. Muller, Applied Regression Analysis and Other Multivariate Methods, 4th edn. (PWS-Kent, Boston, 1988).

170

168. H. Martens, T. Naes, Multivariate Calibration, 1st edn. (Wiley, London, 1989). 169. E. Marengo, E. Mazzucco, E. Robotti, F. Gosetti, M. Manfredi, G. Calabrese, Characterization study of tomato sauces stored in different packaging materials. Curr. Anal. Chem. 12, 1–15 (2016). 170. E. Trombetta, T. Flores-Sahagun, K.G. Satyanarayana, Evaluation of polypropylene/saw dust composites prepared with maleated polypropylene (mapp) produced by reactive extrusion. Rev. Mate´r. 15(2), 309–318 (2010). 171. L. Barbes¸, C. Ra˘dulescu, C. Stihi, ATR-FTIR spectrometry characterisation of polymeric materials. Rom. Rep. Phys. 66(3), 765–777 (2014). 172. D.B. Edwards, L.J. Broutman, A.I. Kasner, Failure analysis of polypropylene used in a hot water environment—effect of different stabilizer systems, ‘‘failure analysis of polypropylene used in a hot water environment—effects of different stabilizer systems’’, in Proceedings of the Society of Plastics Engineers Annual Technical Conference (Cincinnati; 6–10 May 2007). 173. C. Longo, M. Savaris, M. Zeni, R.N. Brandalise, A.M.C. Grisa, Degradation study of polypropylene (PP) and bioriented polypropylene (BOPP) in the environment, Mat. Res. 14(4) (2011). Sa˜o Carlos Oct/Dec 2011 Epub Nov 01. 174. J.P. Luongo, Infrared study of polypropylene. J. Appl. Polym. Sci. 9(3), 302–309 (1960).

171

175. R.G. Kraus, E.D. Emmons, J.S. Thompson, A.M. Covington, Infrared absorption spectroscopy of polycarbonate at high pressure. J. Polym. Sci. Part B: Polym. Phys. 46(7), 734–742 (2008). 176. W. Urbaniak-Domagala, in The Use of the Spectrometric Technique FTIR-ATR to Examine the Polymers Surface, Advanced Aspects of Spectroscopy, ed. by M.A. Farrukh, Rijeka, Croatia (InTech, 2012). 177. L. Kazicina, N. Kupletska, Metody Spektroskopowe Wyznaczania Struktury Zwia˛zko´w Organicznych, 2nd edn. (PWN,Warsaw, 1976). 178. R. Paukkeri, A. Lehtine, Thermal behaviour of polypropylene fractions: 1. Influence of tacticity and molecular weight on crystallization and melting behaviour. Polymer 34, 4075–4082 (1993). 179. B.N. Jang, C.A. Wilkie, A TGA/FTIR and mass spectral study on the thermal degradation of bisphenol A polycarbonate. Polym. Degrad. Stab. 86(3), 419–430 (2004). 180. G.F. Tjandraatmadja, L.S Burn Csiro, M.J. Jollands, The Effects of Ultraviolet Radiation on Polycarbonate Glazing, Durability of Building Materials and Components 8, Ed. by M.A. Lacasse, D.J. Vanier (Institute for Research in Construction, Ottawa; 1999), pp. 884–898. 181. T. Artham, M. Doble, Fouling and degradation of polycarbonate in seawater: field and lab studies. J. Polym. Environ. 17(3), 170 (2009). 182. Z. Wen, X. Hu, D. Shen, The FTIR studies of photo-oxidative degradation of polypropylene. Chin. J. Polym. Sci. 6, 285–288 (1988).

172

Chapter 2

NON-INVASIVE ANALYSIS OF BIOLOGICAL MATERIALS

2.1 Introduction The recent revolution in mass spectrometry technology with the introduction of high throughput instruments and techniques has led to the widespread expansion of advanced analytical methods in many areas of science and technology. But today, the main target of modern mass spectrometry analysis in biomedical research can be summarize as the development of effective and reliable approaches able of discriminating healthy and diseased conditions at their earliest stage, in a non-invasive or minimally-invasive manner.

The aim of this part of PhD research was the development and application of non-invasive or minimally invasive methods for the qualitative and quantitative analysis of biological materials.

In the following subchapter the development of a new method for the non-invasive analysis and characterization of adenoma in colon rectal cancer will be presented. In particular, the method allows the analysis of small molecules produced by the microbiota adhered to the tumor in individuals with advanced colon adenomas. The developed workflow, together with the application of the method to real samples, will be presented.

173

In a second subchapter, a combined bi-dimensional/mono-dimensional gas chromatography mass spectrometry approach for the identification of new biomarkers for prostate cancer in serum will be presented. The discovery and validation phase, together with the obtained results will be shown.

2.1.1 Development of a new method for the non-invasive analysis of adenoma in CRC

According to GLOBACON estimates in 2018, colorectal cancer (CRC) was the third cancer most frequently diagnosed, globally characterized by a high rate of death1,2. While during the past decades the prognosis of patients with colorectal cancer has slowly improved, the 5-year relative survival is still the 50-65%, even if the mortality has recently decreased due to screening aimed at early detection of lesions. The morbidity and mortality induced by this tumor have a significant impact on the national economies. For this reason, the identification of new therapies and the development of effective secondary prevention measures are today an important goal for the scientific community.

Most colon cancers are sporadic, but about 35% are familiar. The role of the microbiota in modulating the risk of sporadic colon cancer is known: some microorganisms are able to directly favor the tumor development, producing inflammation, and/or indirectly through the production of metabolites able to accelerate the proliferation of enterocytes. Information on tumors associated with hereditary predisposition is much

174 less detailed, but some data show a certain difference in the role played by the microbiota in these tumors, compared with the sporadic ones. In recent years, the intestinal microbiota has become an increasingly popular topic due to its numerous important functions. The gut microbiota facilitates the development of the human immune system, and the immune system in turn shapes the composition of the microbiota3. Recent publications have revealed that a disruption in this balanced interaction may generate health disorders including cancer4. These findings suggest that the gut microbiome contributes to inter- individual variability in all aspects of disease situations, and thus its study it’s very important5.

Metabolomics analysis may have a central role in the study of CRC, because it can be used to improve the biology knowledge of the disease and its relation with microbiota. State-of-the-art metabolomic technologies allow measuring thousands of metabolites in biological fluids or biopsies, providing a metabolic fingerprint of individual patients and tissue. Metabolomics techniques can be used also to define the functional status of host–microbial relationships in biological specimens such as feces and tissues. However, most studies on the gut microbiome aim to explore disease-related metabolites or dysregulated metabolic pathways. Microbiome could also influence CRC development through their metabolites6, e.g. short chain fatty acids, such as butyrate, which is produced by bacterial fermentation of undigested fibers in the colon7. The analysis of gut metabolome is usually performed on feces3, but this analysis may not completely reflect the small molecules present on the intestinal mucosa, and in particular on the surface of cancer tissues.

175

Current metabolites biopsy extraction protocols involve sample destruction, precluding additional uses of the tissue, and in presence of small or high value samples with limited availability, this is particularly problematic. The sample is conserved at -80°C until the analysis, then the metabolites are extracted by a mix of solvents and after a few steps the tissue is ready to be analyzed, loosing forever the sample. Using this protocol, metabolomics of mucosal biopsies of patients with ulcerative colitis, collected during colonoscopies, were compared to biopsies from healthy subjects, revealing changes in the abundance of lysophosphatidylcholine, acyl carnitine, and amino acid proﬁles8. Another study showed that the metabolomic profile of routine needle biopsies was able to identify tumor type specific metabolic signatures for breast cancer stratification9.

Brown et al. developed an interesting method for cancer detection and biopsy classification using concurrent histopathological and metabolomic analysis of core biopsies. Biopsies were immediately placed into 80:20 methanol:water (v/v) and incubated for 24 hours at room temperature and then the solvent extract was used for metabolomics analysis. Following incubation, biopsies were removed from the solution and processed for histology. While they have shown that this method is compatible with one antibody cocktail for histopathology, the fixation in alcohol is not the standard method and more validation is still needed for additional antibodies10,11. Therefore, it would be interesting to find a preparation protocol that does not require destruction of the tissue.

The aim of this research is to develop a method that allows the analysis of small molecules produced by the microbiota and which are adhered to

176 the tumor in individuals with advanced colon adenomas. To overcome the limitations of the existing methods, that require sample destruction or are not completely suitable for a subsequent histological analysis, and to increase the amount of information obtained from patient biopsies, we developed a new workflow to perform metabolomic analysis and histological evaluation on the same biopsy sample. Colon adenomas are removed by the surgeon and then small molecules, in particular short chain fatty acids (SCFAs), are sampled using a swab. The swab is then conserved at -80°C until the analysis. SCFAs and other small molecules are extracted from the swab using two different liquid-liquid extractions and analyzed with GCxGC-TOFMS. The method was fully validated for the absolute quantification of acetic acid, propanoic acid, butanoic acid, isobutyric (propanoic, 2-methyl acid) acid, isovaleric (butanoic, 3-methyl acid) acid and valeric acid (pentanoic acid). Our results showed that the developed method is very robust and reliable. The analysis of samples from patients is still ongoing but the preliminary data showed that this new method is able to analyze SCFAs and small molecules released by the microbiota on the surface of adenoma.

2.1.2 Method development

Washing of the swab

Prior to the use, swabs were washed in order to eliminate and reduce contaminations. The washing was performed by leaving the swab for one hour in water. The swab was then placed in oven at 60°C for two hours

177 and then conserved in a methanol-water cleaned falcon filled with nitrogen gas.

Swab sample collection

After the surgical removing of colon adenomas, a sterile swab was passed over the surface of the biopsy, in order to adsorb the small molecules adhered on the adenoma. The swab was then conserved at -80°C until the analysis.

SCFAs extraction

SCFAs were extracted from the swab using water and sonication followed by a liquid-liquid extraction with methyl tert-butyl ether (MTBE). Briefly, the swab without the wood was placed in a tube, then 300 µL of water and 15 µL of the internal standard propanoic acid d2 (20,4 ppm) were added, and the sample was vortexed for 20 seconds followed by centrifugation for 15 seconds. The tube was then placed in a cold sonicator bath for 30 minutes. The swab was squeezed with cleaned tweezers and 200 µL of water extract were placed in a new tube, bringing the pH to 2 using 6 M HCl. For the extraction of SCFAs 140 µL of MTBE were added and the tube was placed on a rotator for 15 minutes, followed by a centrifugation for 10 minutes at 4°C and 21.1 g. Then, 100 µL of the organic phase, that contains the SCFAs, was analyzed with GCxGC-TOFMS. The remaining water phase solution was then subjected to a second extraction described in the following subchapter.

178

Metabolites extraction from the aqueous solution

The remaining water phase, that contains other small molecules such as amino-acids, sugars, medium and long fatty acids, was extracted again using methanol-isopropanol-acetonitrile. Briefly, 200 µL of water from the first liquid-liquid extraction were subjected to a second extraction using 1 mL mixture of acetonitrile, methanol and water 3:3:2, with 5 µL of tridecanoic acid (0,5 ug/mL) as internal standard. The sample was vortexed for 15 seconds and centrifuged for 15 minutes at 20 °C at 14.5 xg. One mL of the supernatant was placed in a new tube and dried in a speed-vacuum and then placed at -20°C until derivatization. The samples were derivatized with methoximation (20 µL of methoxamine, 80°C, 20 min) and silylation (30 µL of N,O-Bis(trimethylsilyl)trifluoroacetamide, 80 °C, 20 min) prior the GCxGC-TOFMS analysis.

GCxGC-TOFMS analysis

For the analysis, a LECO Pegasus BT 4D GCXGC/TOFMS instrument (Leco Corp., St. Josef, MI, USA) equipped with a LECO dual stage quad jet thermal modulator was used. The GC part of the instrument was an Agilent 7890 gas chromatograph (Agilent Technologies, Palo Alto, CA), equipped with a split/splitless injector. The first dimension column was a 30 m Rxi-5Sil (Restek Corp., Bellefonte, PA) MS capillary column with an internal diameter of 0.25 mm and a stationary phase film thickness of 0.25 μm for metabolite from aqueous solution, while for SCFAs the column was a 30 m DB-FATWAX-UI (Agilent Technologies, Santa Clara, CA) with a diameter of 0.25 mm and a film thickness of 0.25 μm, and the second dimension chromatographic columns was a 2 m Rxi-17Sil MS

179

(Restek Corp., Bellefonte, PA) with a diameter of 0.25 mm and a film thickness of 0.25 μm. High-purity helium (99,9999%) was used as the carrier gas with a flow rate of 1.4 mL/min. 1 μL of sample was injected in splitless mode at 250°C. The temperature program for metabolites analysis was as follows: the initial temperature was 70°C for 2 minutes, then ramped 6°C/min up to 160°C, 10°C/min up to 240°C, 20°C/min to 300 and then held at this value for 6 minutes. The secondary column was maintained at +5°C relative to the GC oven temperature of the first column. The temperature program for SCFAs was as follows: the initial temperature was 40°C for 2 minutes, then ramped 7°C/min up to 165°C, 25°C/min up to 240°C, maintained for 5 minutes. The secondary column was maintained at +5°C relative to the GC oven temperature of the first column. Electron impact ionization was applied (70 eV). The ion source temperature was set at 250°C, the mass range was 40 to 300 m/z with an extraction frequency of 32 kHz, for the SCFAs analysis, while for the metabolites the mass range was 25 to 550 m/z. The acquisition rates were 200 spectra/s. The modulation periods for both programs were 4s for the entire run. The modulator temperature offset was set at +15°C relative to the secondary oven temperature, while the transfer line was set at 280°C.

Data analysis

The chromatograms were acquired in TIC (total ion current) mode. Peaks with signal-to-noise (S/N) value lower than 500.0 were rejected. ChromaTOF version 5.31 was used for raw data processing. Mass spectral assignment was performed by matching with NIST MS Search 2.3 libraries adding Fiehn Library. Commercial mix standard of free fatty acids (Restek Corp., Bellefonte, PA), composed by acetic acid, propanoic acid,

180 propanoic acid 2-methyl, butanoic acid, butanoic acid 3-methyl and pentanoic acid was run individually and EI spectra were matched against the NIST library.

Statistical analysis

The calibration curves of the SCFAs were obtained using excels, while the analytical results were then processed and compared with the open source MetaboAnalyst software (www.metaboanalyst.org).

In figure 1 it is shown the workflow of the developed method.

Figure 1: workflow of the developed method.

2.1.3 Validation of the method

One of the most important issues in the analysis of biopsies is the sample destruction that is necessary to extract small molecules for the analysis.

181

To overcome this limitation, we developed a non-invasive method that is able to sample the small molecules from the surface of a tissue sample. A very efficient method of analysis should enable the extraction, the identification, and the reliable quantitative estimation of major and trace components while leaving unchanged the original sample for subsequent histological analysis. The developed method not only owns all this features, but it allows the analysis of metabolites that are directly produced by the microbiota of adenoma and that are present on the surface of a specific tissue. The protocol enables the analysis of SCFAs, which are the most important class of molecules for microbiome analysis, but also sugars, medium and long chain fatty acids, aminoacids, etc.

Here, the suitability of the new method is evaluated. The validation of the method was carried out on main SCFAs (acetic acid, propanoic acid, propanoic acid 2 methyl, butanoic acid, butanoic acid 3 methyl and pentanoic acid). Mixed standard solutions were deposited on sterile and cleaned swab and were extracted with the developed method. SCFAs were then analyzed with GCxGC-TOFMS. For each analyte, a calibration plot reporting the peak area of the “quantifier” transition signal (y) versus standard concentration (x) was built. Concentration levels in the following ranges were considered: from 5 ppm to 80 ppm for acetic acid, from 5 ppb to 2 ppm for propanoic acid and 2-methyl propanoic acid, from 5 ppb to 0.2 ppm for butanoic acid, from 5 ppb to 0.1 ppm for 3-methyl butanoic acid and from 5 ppb to 0.5 ppm for pentanoic acid. Moreover, to overcome possible memory effects, the standard solutions were injected in randomized order. For all the analytes, a linear regression fit with a weighting factor 1/x was used, and a good linearity was obtained. The

182 regression coefficients, reported in table 1, were the following: 0.98 for acetic acid, 0.99 for propanoic acid, 0.99 for propanoic acid 2 methyl, 0.96 for butanoic acid, 0.99 for butanoic acid 3 methyl and 0.99 for pentanoic acid. Table 1 summarizes also intercept, slope and linearity range.

The limit of detection (LOD) was calculated as the concentration of the analyte that gives a signal (peak area) equal to the average background (Sblank) plus 3 times the standard deviation sblank of the blank (LOD = Sblank + 3sblank), while the limit of quantification LOQ was calculated as LOQ = Sblank + 10sblank. LOD and the LOQ values were the following: 6 and 8 ppm for acetic acid, 5.6 and 7 ppb for propanoic acid, 6.2 and 9 ppb for 2-methyl propanoic acid, 5.6 and 7 ppb for butanoic acid, 5.6 and 7 ppb for 3-methyl butanoic acid and 5.5 and 6.5 ppb for pentanoic acid. The LODs and LOQs of the method are very good. Table 2 reports the inter-day precisions on concentration evaluated by analyzing the analytes every day (three replicates) for 3 days. The results show that the inter- day precisions are 9.94% for acetic acid, 3.07% for propanoic acid, 7.87 for 2-methyl propanoic acid, 6.40% for butanoic acid, 6.63% for 3-methyl butanoic acid and 2.25% for pentanoic acid. To evaluate the recovery R (%) of each analyte and to verify its possible dependence on the concentration, the analyte standard solutions at different concentration levels used for the calibration curve were deposited on a new swab and then extracted and analyzed. The recovery values were calculated as

Cobs/Cref, where Cobs is the difference between the concentration determined for the spiked sample and the native concentration in the same sample, and Cref is the spiked concentration. For all the analytes, a percentage of recovery R (%) was therefore calculated at low, medium

183 and high concentration (reported in table 3). As it can be observed, the R (%) values at 95% C.I. for acetic acid range from 91.44% to 117.44%, for propanoic acid range from 81.49% to 114.71%, for 2-methyl propanoic acid range from 45.96% to 100.65%, for butanoic acid range from 53.89% to 80.94%, for 3 -methyl butanoic acid range from 23.32% to 119.78% and for pentanoic acid range from 94.82% to 90.93%.

Table 1: Coefficient of correlation, slope, intercept and linearity range for the SCFAs standards used for the validation of the method.

Molecule Coefficient of Slope Intercept Linearity correlation range Acetic acid 0.98 1*108 3*108 8ppm- 80ppm Propanoic acid 0.99 4*108 -9*105 7ppb- 2ppm Propanoic acid, 2 0.99 8*108 1*108 9ppb- methyl 2ppm Butanoic acid 0.96 4*108 1*107 7ppb- 0.2ppm Butanoic acid, 3 0.99 1*109 6*106 7ppb- methyl 0.1ppm Pentanoic acid 0.99 3*108 -4*104 6.5ppb- 0.5ppm

184

Table 2: Intra-day and inter-day precision of the method calculated as percentage of coefficient of variation (CV%) for each analyte.

Molecule Intra-Day (CV %) n=3 Inter-Day (CV %) n=3 Acetic acid 5.55 % 9.94 % Propanoic acid 6.69 % 3.07 % Propanoic acid, 2 7.88 % methyl 7.31 % Butanoic acid 4.31 % 6.40 % Butanoic acid, 3 methyl 3.64 % 6.64 % Pentanoic acid 1.05 % 2.25 %

Table 3: Percentage of recovery R (%) calculated at low, medium and high concentration for each analyte.

Molecule Low n=3 Medium n=3 High n=3 Acetic acid 91.44±0.72 114.15±1.19 117.44±1.34 Propanoic acid 81.49±0.98 103.63±1.08 114.71±1.26 Propanoic acid, 2 methyl 45.96±0.94 89.59±0.97 100.65±1.25 Butanoic acid 53.89±0.39 73.13±0.95 80.94±0.99 Butanoic acid, 3 methyl 23.32±0.58 79.53±0.91 119.78±1.56 Pentanoic acid 94.82±0.83 101.43±1.23 90.93±1.05

Since we used an untargeted approach for SCFAs quantification, we had also the possibility to identify and quantify other molecules in the same sample. The analysis allowed the identification of more than 360

185 molecules from different classes as short chain carboxylic acids, essential fatty acids, methylated fatty acids and omega-7.

We have further investigated the identity of metabolites extracted with the second liquid-liquid extraction. In this case, we selected some molecules and we calculate the intraday and inter-day reproducibility as reported in table 4. More than 370 molecules were identified: medium (8-10 C-atoms), long (14-20 C-atoms) and very long (22 or more) chain fatty acids, including subclasses of monounsaturated with omega-9, polyunsaturated fatty acids (PUFAs) with omega-3 and omega-6; saturated fatty acids; essential fatty acids (EFAs); amino acids; phenols, sugars and sterols.

Table 4: Intra-day and inter-day precision of the method calculated as percentage of coefficient of variation (CV%) for each analyte. Molecule Intra-Day (CV %) Inter-Day (CV %) n=3 n=3 L-Alanine 7.67 % 8.88 % D-Fructofuranose 8.00 % 9.52 % Malic acid 5.67 % 8.87% Palmitic acid 3.76 % 6.30 % Tyrosine 4.59 % 5.61 %

2.1.4 Application of the method to colon rectal cancer

The developed method was then applied for the analysis of the metabolites present in the adherent metabolome in colon adenoma. 48 patients carrying at least one colon adenoma with size ≥ 1 cm were

186

recruited by professor Irma Dianzani (University of Piemonte Orientale) for the MIMEC project. Table 5 reports the clinical features of the patients. In this pilot study we only analyzed the samples for the absolute quantification of SCFAs.

Table 5: clinical features of the patients included in the pilot application of the new method. Clinical features Patients (n=48) Gender Male 25 (52.08%) Female 23 (47.92%) Age Mean ± SD 58.65 ± 9.35 Familiarity No 31 (64.58%) Yes 17 (35.42%) BMI Normal weight 24 (50%) Overweight 18 (37.5%) Obese 6 (12.5%) Histology Low-grade dysplasia 31 (64.6%) High-grade dysplasia 13 (27.08%) Adenocarcinoma 4 (8.33%) Red meat intake g/day ± SD 62.99± 42.6 Fruit and vegetable intake g/day ± SD 364.46± 159.5

Colon adenomas were sampled using the swab method and the metabolites were then analyzed using the previously described method.

187

Figure 3 reports the main results obtained from the metabolomics analysis. The dendrogram of hierarchical clustering of SCFA abundance in patients with low grade dysplasia (red) and with high grade dysplasia or adenocarcinoma (green) is reported (a). The cluster analysis showed that the SCFAs present on the surface of the polyp tissue are correlated with the tumor grade. In particular, the concentrations of 3-methyl butanoic acid and 2-methyl propanoic acid are higher in patients with a high grade level of carcinoma (b). The box plots (c) and heat map (d) are related to patients classified on the basis of the cancer score risk calculated using BMI, grade level of dysplasia and hereditary. Patients with low BMI, low grade dysplasia and no familiarity were classified with a score risk 0, patients with high BMI, high grade dysplasia and no familiarity were classified with a score risk 1 and patients with high BMI, high grade dysplasia and hereditary were classified with score risk 2. 3 methyl- butanoic and propanoic acids resulted more abundant in patients with a higher cancer score risk (high BMI, high grade dysplasia and hereditary) with respect to patients with a lower cancer risk. 3-methyl butanoic (isovaleric acid), 2-methyl propanoic (isobutyric acid) and propanoic acids were already founded higher in stool of patients with colorectal cancer patients12.

188

a b High grade dysplasia + adenocarcinoma FC > 1.5 & p-value < 0.05

Low grade dysplasia

H H

FC > 1.5 & p-value < 0.05

c d Score risk Score risk

Figure 3: The main results obtained from metabolomics analysis. A) the dendrogram of the two groups of patients (in red the low grade dysplasia, in green the high grade and adenocarcinoma) shows minimal separation; b) The cluster analysis of the two up- regulated metabolites in high grade dysplasia and adenocarcinoma: the butanoic acid, 3-methyl (left) and the propanoic acid, 2-methyl (right); c) the box plots related to the cancer score risk indicate that Propanoic and Butanoic, 3-methyl are more abundant in high risk patients; d) the heat map generated by average normalized peak areas shows progressive elevation with the progression of the risk: butanoic acid, 3-methyl and propanoic acid are more abundant in the two groups of high risk compare to the low.

We have further investigated whether the aqueous extract was able to provide more information from the swab. We then analyzed and compared the aqueous extract from three patients with low grade and three patients with high-grade adenocarcinoma. We obtained a different

189 metabolomic signature between the two groups as shown in Venn diagram reported in figure 2. The main differences are related to medium chain fatty acids, amino acids, sugars, phenols and sterols. Although these preliminary data are mainly qualitative, the proposed approach can be considered very promising.

Figure 2. Venn diagrams of identified metabolites from aqueous extraction from swab sampled from three patients with low grade cancer and from three patients with high grade cancer.

190

2.2 Identification of new biomarkers for the diagnosis of prostate cancer

Prostate Cancer (PCa) is the second most common cancer in men and it represents the ﬁfth leading cause of death13. The mechanisms of prostate carcinogenesis depend on genetic and environmental factors, with risk of the disease increasing with age and positive family history14. In PCa, which is a hormone-dependent tumor, androgens affect proliferation and differentiation of prostate luminal epithelium and drive prostate cancer cell growth15. In fact, the adult prostate contains luminal epithelial cells, which produce secretory proteins such as human PSA (prostate-specific antigen), are androgen-responsive16. The clinical course of PCa is variable, ranging from an indolent state to a rapid grown into an aggressive phenotype that spreads and metastasizes to the lymph nodes and bones. The symptoms are almost absent in the early stages because PCa originates, in 70% of cases, from the peripheral portion of the gland17. The most common clinically used tools for the early detection are digital rectal examination (DRE) and testing for PSA. In the early 90s, when PSA was introduced, it revolutionized the clinical practice, leading to an increasing number of PCa cases diagnosed at early stages, with a consequent decrease of PCa mortality. However, the use of PSA has some limitations: in fact, the presence of benign conditions, like prostatitis, chronic inflammation of the prostate parenchyma and benign prostatic hyperplasia (BPH), which are very common in the elderly, may cause an elevated serum PSA level. Although PSA is still used as central test in the diagnosis, PCa is present in up to 25% of patients with PSA levels below the current thresholds (3-4 ng/mL)18,19. The usefulness of PCa screening

191 with PSA test has been questioned for years due to contradictory results20, to its limited sensitivity and speciﬁcity, and because of its inability to discriminate aggressive from indolent PCa. In consequence, today, a trans-rectal ultrasound (TRUS)-guided prostate biopsy is always performed to conﬁrm the diagnosis. The biopsy involves the sampling of several areas of the prostatic gland in order to obtain an accurate and clear diagnosis. However, the biopsy suffers from some limitations and does not provide enough information to enable a precise discrimination between indolent and aggressive tumors, although biopsy has been associated with high false negative rates due to the high degree of PCa heterogeneity21.

Metabolomics, which is the study of the small molecules present in a biological system22, has been often used for the discovery of new biomarkers. The study of metabolome requires interdisciplinary platform techniques, which include analytical chemistry, biochemistry and bioinformatics. There are two main strategies to metabolomics studies: untargeted and targeted approaches. The untargeted approach has the aim of determining as many metabolites as possible. The targeted approach is mainly used for studying a panel of metabolites or the pathways that characterize a particular disease, that are undoubtedly associated with the disease or condition of interest23, or to validate specific biomarkers discovered using untargeted approaches. Metabolomics has been widely employed for the discovery of new potential biomarkers for PCa, nevertheless the reported results were not always very satisfactory and they did not lead to the development of a new screening test for the non-invasive diagnosis of PCa. Blood, urine,

192 tissue and seminal fluid were analyzed using several techniques, such as gas chromatography (GC), liquid chromatography (LC), nuclear magnetic resonance (NMR) and MALDI-TOF (for tissues). For example, serum samples from PCa patients and healthy subjects were analyzed with GC- MS by Huang et al.24,25, Mondul et al.26,27, and de Vogel and colleagues28, discovering some potential candidate biomarkers and assessing PCa risk, but they did not considered chronic conditions, such as prostate inflammation. Although Giskeødegård et al. included in the study PCa patients and BHP patients29, the study was performed on a limited number of samples. An ideal biomarker should permit to discriminate patients with PCa, reducing as much as possible the currently high false positive rates.

In the present study we performed metabolomics analysis of serum samples from PCa patients with different cancer aggressiveness (Gleason Score 6, 7, 8 and 9) and patients affected by chronic inflammation of the prostate parenchyma. All patients were subjected to diagnostic prostate biopsy according to clinical protocols that included PSA level measurement and, in some cases, digital rectal examination (DRE) and repeated measurements. We firstly performed an untargeted analysis on 30 patients using comprehensive bi-dimensional gas chromatography mass spectrometry (GCxGC-MS), in order to identify potential molecules biomarkers of PCa. We then validated the results on a bigger cohort (200 patients) using a monodimensional GC-MS and standards, in order to identify specific thresholds for the most interesting biomarkers. Furthermore, using the Gleason score as a prognostic factor for PCa outcome, we assessed whether the serum levels of this new set of

193 biomarkers could identify the patients with high-grade PCa, who are usually not eligible for a delayed clinical treatment.

2.2.1 Study population, materials and methods

Study population

This research focus on a two-centric study: 213 patients from the Department of Urology, San Giacomo Hospital, Novi (AL), Italy and from the Department of Urology, Galliera Hospital, Genoa, Italy (Tab. 1). All patients included in the study were subjected to a prostate biopsy according to the hospital protocols, prescribing PSA >10 ng/mL as first measurement, or DRE and repeated PSA measurements in suspect cases (PSA 4 ng/mL or lower but increased from preceding measurement, and/or DRE suspect). Most patients (172) included in the study had level of PSA ≥4 ng/mL at pre-biopsy measurement, while 11 patients had low PSA (<3 ng/mL) and suspect DRE. All patients in the study were considered PSA/DRE positive30. The hospitals’ ethics committees approved the study protocol. All patients were informed about inclusion in the study and gave their consent. Plasma samples were collected before biopsy and kept frozen at -80 °C until the analysis. The histological analysis showed that 76 patients were PCa positive and 137 were PCa negative (Tab. 1)30. PCa Gleason scores were also defined according to the WHO histological and scoring criteria31.

194

Table 1: Diagnosis, age, PSA and Gleason score of patients included in the study.

Hospital PCa patients Chronic Inflammation patients

Discovery phase 15 15 Ages 50-78 Ages 47-80 (median 68) (median 64)

Validation phase 61 122 Ages 52-86 Ages 53-81 (median 65.5) (median 68.5)

Average PSA 14.70 ng/mL 7.80 ng/mL

Sample preparation for metabolomics analysis

1 mL of a Acetonitrile (ACN)/Isopropyl alcohol (IPA)/water (3:3:2) solution, with tridecanoic acid at 1 ppm as internal standard, was added to 30 µL of serum. After vortexing for 10 s, the sample was centrifuged at room temperature for 15 minutes at 14500 g. The supernatant was then dried in a speed-vacuum. The sample was then derivatized performing methoximation (20 µL of Methoxamine, 80°C, 20 min) and sylilation (30 µL of BSTFA, 80 °C, 20 min). After derivatization the sample was ready for the GC-MS analysis.

GCxGC/TOFMS and GC/TOFMS Analysis Parameters

For the metabolomics analyses, a LECO Pegasus BT 4D GCXGC/TOFMS instrument (Leco Corp., St. Josef, MI, USA), equipped with a LECO dual stage quad jet thermal modulator was used. The GC part of the instrument was an Agilent 7890 gas chromatograph (Agilent Technologies, Palo Alto, CA), equipped with a split/splitless injector. The

195 first dimension column was a 30 m Rxi-5Sil (Restek Corp., Bellefonte, PA) MS capillary column with an internal diameter of 0.25 mm and a stationary phase film thickness of 0.25 μm, and the second dimension chromatographic column was a 2 m Rxi-17Sil MS (Restek Corp., Bellefonte, PA) with a diameter of 0.25 mm and a film thickness of 0.25 μm. High-purity helium (99,9999%) was used as the carrier gas with a flow rate of 1.4 mL/min and 1.0 mL/min for bi- and mono- dimensional analysis respectively. 1 µL of sample was injected in splitless mode at 250°C. The temperature program was as follows: the initial temperature was 70°C for 2 minutes, then ramped 6°C/min up to 160°C, 10°C/min up to 240°C, 20°C/min to 300 and then held at this value for 6 minutes. The secondary column was maintained at +5°C relative to the GC oven temperature of the first column. The programming rate was the same for the two columns. Electron impact ionization was applied (70 eV). The ion source temperature was set at 250°C, the mass range was 35 to 550 m/z with an extraction frequency of 32 kHz for the bi-dimensional analysis and 30 kHz for the mono-dimensional one. The acquisition rates were 200 spectra/s for 2D analysis and 10 spectra/s for 1D. The modulation period for the bi- dimensional analysis was 4s for the entire run. The modulator temperature offset was set at +15°C relative to the secondary oven temperature, while the transfer line was set at 280°C.

Data Analysis

The chromatograms were acquired in TIC (total ion current) mode. Peaks with signal-to-noise (S/N) value lower than 500.0 were rejected.

196

ChromaTOF version 5.31 was used for raw data processing. Mass spectral assignment was performed by matching with NIST MS Search 2.3 libraries adding FiehnLib. The statistical analysis was performed with Metaboanalyst software (www.metaboanalyst.org).

2.2.2 Results

In this study, a non-targeted metabolomics approach was used to discover the global serum metabolomic differences between patients with prostatic chronic inflammation (PCI) and with prostate cancer (PCa). We focused our analysis on the metabolic alterations between PCI and PCa, both subjected to prostate biopsy and with high levels of PSA. PCa and PCI patients were divided into two groups; the first group was used for the discovery of the biomarkers and the second for the validation of the diagnostic model. The discovery analyses were performed on a cohort of 30 subjects composed by 15 patients with PCa and by 15 patients with PCI, using bi-dimensional gas chromatography coupled to mass spectrometry. The validation phase was carried out using mono- dimensional gas chromatography on a larger cohort composed by 183 patients divided in two principal groups: PCI (n=61) and PCa patients with adenocarcinoma diagnosis and Gleason score (GS) ≥ 6 (n=122) (Table 1). The PCa patients were also divided into two subgroups according to the histopathological evaluation, which attributed the grades of the tumor: 63 patients with Gleason score 6 and 59 patients with Gleason score ≥ 7.

The choice to use a mono-dimensional analysis is due to several reasons which are related to the cost and to the easiness of the analyses. The main advantages are discussed in detail in the validation paragraph.

197

Discovery of potential biomarkers with GCxGC-MS

More than 1200 small molecules were identified in the serum samples analyzed in the discovery phase. Among these molecules there are amino acids, fatty acids, sugars, organic acids, steroids, etc. The similarity threshold for analytes identification was set to 750, and all the mass spectra and the retention index were manually checked and compared to the one present in the NIST library and in literature. The curated results reported the presence of 875 relevant analytes: the area of all these analytes was determined for each sample.

The relative abundance of the identified molecules was then elaborated using the Metaboanalyst software, in order to perform the statistical analysis and to identify the modulated molecules and the potential biomarkers. No data filtering was applied during this phase of processing. Log transformation and auto scaling normalization were used as preprocessing of the data. Monovariate statistical analysis was used to identify the regulated molecules and potential biomarkers. A total of 29 molecules resulted modulated with a fold change > of 1.5 and a p-value < 0.05. Table 2 reports the complete list of the 29 regulated metabolites (down regulated in PCa patients from n. 1 to 18; up regulated from n.19 to 29) found in discovery phase.

198

Table 2: fold change and p-value of modulated molecules obtained in the discovery phase.

N° Molecule Fold change p-value 1 2(3H)-Furanone 5-methyl- 4.26 6.25*10-5 2 Hippuric acid, TMS derivative 3.91 6.12*10-6 3 Phenylpropanolamine 2TMS derivative 2.45 4.10*10-2 4 D-Arabinose 4TMS derivative 2.38 2.06*10-2 5 Oleamide TMS derivative 2.29 2.75*10-2 6 26-Bistert-butylphenol TMS derivative 2.25 3.37*10-3 7 Aminocaproic acid 2.15 1.50*10-2 8 Nonanoic acid TMS derivative 1.70 2.41*10-5 9 Cholecalciferol TMS derivative 1.59 7.89*10-3 10 D-Mannopyranose 5TMS derivative 1.57 2.71*10-3 11 9H-Purin-6-ol, 2TMS derivative 1.54 6.99*10-3 12 Isoquinoline 1-methyl- 1.51 2.11*10-2 13 D-Mannose 5TMS derivative 1.51 2.35*10-3 14 2-Ethyl-3-hydroxypropionic acid di-TMS 1.50 2.92*10-2 15 Formic acid butyl ester 1.49 6.80*10-3 16 Propanedioic acid, 2TMS derivative 1.45 4.91*10-4 17 D-Glucopyranose 5TMS derivative 1.42 4.26*10-3 18 4-Hydroxybutanoic acid 2TMS derivative 1.40 3.09*10-2 19 Heptadecane 0.698 2.94*10-2 20 3,4-Dimethoxymandelic acid di-TMS 0.671 2.67*10-2 21 alpha-Tocospiro B 0.668 3.81*10-2 22 Linoelaidic acid methyl ester 0.626 3.36*10-4 23 Methyl stearate 0.591 6.98*10-6 24 Erythrono-14-lactone Z- 2TMS derivative 0.582 4.96*10-3 25 L-Proline trimethylsilyl ester 0.579 2.59*10-2

199

26 17a-Aza-D-homoandrostan-17-one (5- 0.392 2.50*10-2 alpha) 27 Vinyl trans-cinnamate 0.392 1.06*10-3 28 13-Octadecenoic acid E- TMS derivative 0.137 1.18*10-3 29 Pyrrole 7.12*10-4 3.48*10-2

Figure 1. shows the heat map of the top 70 areas of untargeted metabolites distributed in the serum sample of the two classes of patients (PCi and PCa) during the discovery step with bi-dimensional analysis. Higher concentrations of metabolites are colored in red, while in blue are reported molecules present at low levels of concentration. As reported in the heat-map, structured information related to the disease is present in the serum samples.

200

Figure 1: heat map of top 70 metabolites distributed in the serum sample of the two classes of patients (PCI and PCA) from the discovery step.

Figure 2 a and b shows the box plots of the most regulated molecules. Serum samples from the PCa patients exhibited lower level of Hippuric acid, Phenylpropanolamine, D-Arabinopyranose, Aminocaproic acid, Pelargonic acid, Cholecalciferol, D-Mannopyranose, Hypoxanthine, D- Mannose, D-Glucopyranose, Malonic acid, 4-Hydroxybutanoic acid and Butyl formate. Higher levels of Heptadecane, 3,4-Dimethoxymandelic

201 acid, alpha-Tocospiro B, Methyl stearate, L-Proline trimethylsilyl ester, Vinyl trans-cinnamate, Linoelaidic acid methyl ester, 17a-Aza-D- homoandrostan-17-one 5-, Pyrrole-2-carboxylic acid, 13-Octadecenoic acid and γ-Caprolactone were also quantified in these patients.

These altered serum metabolite levels reflect differences in the metabolomic profile of patients affected by PCa, and they could be used to provide a specific metabolomic signature and more information on the mechanism of prostate cancer formation. These metabolites were further used as candidate in the following validation step, performed by mono- dimensional gas chromatography.

202

Figure 2. Box plots of the most regulated molecules found in serum samples during the discovery step: down-regulated (a) and up-regulated (b) metabolites in patients with PCa (blue) compared to the patients with PCI (pink).

Multivariate statistical analysis was applied in order to classify the two groups of samples and to explore the metabolic differences among them. Principal component analysis (PCA) was firstly used. As reported in figure 3 A, B and C, PCA was able to clearly separate PCa and PCI patients, indicating a indicating a significant difference between the two groups of samples.

203

Figure 3. Multivariate data analysis based on data from GCXGC-MS of PCa (orange in the score plot) and PCi patients (blue). A) PCA score plot of PC3 vs PC5; B) PCA score plot of PC1 vs PC3; C) PCA score plot of PC5 vs PC3; D) explained variance, the first 6 Pcs explained about the 60.5% of information.

PLS-DA was also performed in order to separate the two groups of samples and to identify the most discriminant variables. The following figure reports the VIP plot that shows the molecules that highly contribute to the separation of the two classes of samples reported on the right of figure 4.

204

Figure 4: variable importance in the projection (VIP) obtained from the PLS-DA model (left), the metabolites features are ranked based on their contributions to classification accuracy; representation of PLS-DA latent variables (right), in green are reported PCi while in red PCa.

Metabolic pathway analysis

Today, it is widely recognized that metabolic alterations may play a crucial role in the development and progression of prostate cancer. Thus, a better understanding of the metabolomic profile can have a great

205 advantage in determining potential new screening methods and risk, both for initial diagnosis and for follow-up treatments32,33.

To explore the underlying molecular functions of these serum metabolite biomarkers, a metabolic pathway analysis was performed. Regulated metabolites were used to perform pathway analysis using MataboAnalyst software. The analysis revealed a wide range of changes, mainly related to Purine metabolism (A), Amino sugar and nucleotide sugar metabolism (B), Arginine and proline metabolism (C), Aminoacyl-tRNA biosynthesis (D), sugars metabolism (E), Phenylalanine metabolism (F), Propanoate metabolism (G), Pentose phosphate pathway (H), Glycolysis or Gluconeogenesis (I), as reported in the image displayed in figure 5. As confirmed by our results, the prostate tumorigenesis may be associated with dysregulation of key metabolic signal pathways.

Figure 5: Summary of the pathway analysis of patients affected by PCa. Each circle represents a different metabolism. Color gradient and circle size indicate the significance of the pathways, ranked by p-value and pathway impact score respectively.

206

Validation of biomarkers with GC-MS

In order to validate the results obtained using a restricted cohort of patients with the bi-dimensional gas chromatography, a second analysis on a larger cohort of patients (n=183) composed by prostate cancer patients (n=122) and prostatic chronic inflammation patients (n=61) was then performed. For the validation phase a mono-dimensional gas chromatography coupled to mass spectrometry (GC-MS) was used, instead of the GCxGC-MS platform employed in the discovery phase. Among the advantages of using a mono-dimensional GC-MS instrument there are: (i) the analysis is faster and cheaper because there is no need to use the modulator of the secondary oven and the flow of the mobile phase is reduced in GC-MS, (ii) the management of the data is faster and easier (bi-dimensional files are much bigger than mono-dimensional) and (iii) GC-MS are more diffuse than GCxGC-MS, so the method can be employed by more laboratories.

We decided to focus our attention on six potential biomarkers, 3 up- regulated (L-proline, a-D-Talopyranose and 2,5-Dimethoxymandelic acid) and 3 down-regulated (Hippuric acid, Propanedioic acid and 9H-Purin-6- ol), that were fully validated (on an external cohort) for the diagnoses of prostate cancer and for an accurate discrimination of prostate cancer from prostatic chronic inflammation despite of high levels of PSA. Hippuric acid, Propanedioc acid and 9H-Purin-6-ol quantified using standards while for L-proline, a-D-Talopyranose and 2,5- Dimethoxymandelic acid their relative abundances were used.

207

The calibration curves of standards Hippuric acid, 9H-Purin-6-ol and Propanedioic acid were built by injecting in GC-MS 1 μL of standard solution at 0.01, 0.05, 0.1, 0.5, 1, 2, 5 and 10 ppm for hippuric, at 0.02, 0.1, 0.2, 0.5, 1, 2, and 5 ppm for 9H-Purin-6-ol, and at 0.02, 0.05, 0.1, 0.5, 1, 2, 5, 10 and 20 ppm for propanedioic acid.

For all the analytes a linear regression ﬁt with a weighting factor 1/x was used and a good linearity was obtained: the regression coeﬃcients were 0.999 for Hippuric acid, 0.9988 for 9H-Purin-6-ol and 0.9999 for Propanedioic acid. Figure 6 shows the calibration curves of the standards.

Figure 6. the calibration curves of the 3 standards. b) Hippuric acid, TMS derivative; b) Purin-6-ol, 2TMS derivative; c) Propanedioic acid, 2TMS derivative.

The limit of detection (LOD) was calculated as the concentration of the analyte that gives a signal (peak intensity) equal to the average background (Sblank) plus 3 times the standard deviation sblank of the blank

(LOD = Sblank + 3sblank), while the limit of quantiﬁcation LOQ is given as LOQ

208

= Sblank + 9sblank. LOD and LOQ were calculated as 0.02 μg/mL and 0.07 μg/mL for Hippuric acid, 0.04 μg/mL and 0.15 μg/mL for Propanedioic acid, and 0.05 μg/mL and 0.16 μg/mL for 9H-Purin-6-ol, as reported in table 3.

Table 3 Calibration Curve Results for the Quantiﬁcation of derivatized standard metabolites: Hippuric acid, TMS derivative, Propanedioic acid, 2TMS derivative, 9H-Purin-6-ol, 2TMS derivative. Molecules Linearity range R2 LOD (ug/mL) LOQ (ug/mL) (ug/mL) Hippuric acid 0.01-10 0.999 0.02 0.070 Propanedioic acid 0.02-20 0.999 0.04 0.15 9H-Purin-6-ol 0.02-5 0.999 0.05 0.16

The selected molecules were then quantified for all the available samples and successively, in order to assess the capability of the potential marker metabolites to discriminate the two diseases, the area under the ROC curve (AUC) as well as its sensitivity and specificity were separately calculated for each of the 6 metabolites. The results are listed in table 4.

Table 4: Main m/z, the chemical formula, the name of marker, area under the receiver operating characteristic (ROC) curve with the 95% confidence interval (CI) range in parenthesis, sensitivity and specificity for each identified biomarker.

No. m/z Formula Metabolites AUC Sensitivity Specificity

1 206 C9H9NO3 Hippuric acid 0.91 (0.85-0.95) 0.84 0.94

2 280 C5H4N4O 9H-Purin-6-ol 0.87 (0.81-0.95) 0.83 0.92

3 147 C3H4O4 Propanedioic acid 0.88 (0.82-0.92) 0.81 0.80

4 204 C6H12O6 α-D-Talopyranose 0.87 (0.81-0.92) 0.80 0.85

5 70 C5H9NO2 L-Proline 0.82 (0.75-0.88) 0.78 0.76

209

6 239 C10H12O5 2,5- 0.85 (0.79-0.91) 0.73 0.83 Dimethoxymandelic acid

Figure 7 reports the ROC curves of the 6 most promising biomarkers identified during the validation step. As shown, hippuric acid resulted the best biomarker, with an AUC of 0.91, followed by propanedioic acid and 9H-purin-6-ol, both with an AUC of 0.88. L-proline, a-D-Talopyranose and 2,5-Dimethoxymandelic acid are characterized by an AUC of 0.82, 0.88 and 0.85 respectively. In the same figure in panel c it is reported also the AUC of PSA (0.67).

Figure 7: ROC curves of the 6 best markers for the diagnosis of PCA; the 3 up regulated markers (a, b, c), the 3 down regulated (d, e, f) compared with the ROC curve of PSA levels measured in the same samples (g).

210

The ROC curves obtained by the absolute quantification of 3 standards were also compared with the ROC curve performed in the discovery phase (figure 8). The ROC curves performed on the validation set resulted improved respect with the ones from the discovery phase.

Figure 8: ROC curves obtained from concentration of 3 down regulated markers with 95% confidence interval range in parentheses, in both modalities (mono and bidimensional analysis) found respectively in discovery and the validation step.

Multivariate ROC analysis of the 3 down-regulated markers combined with the levels of PSA was also performed. The analysis showed a better AUC when the three markers are combined together including also PSA level (figure 9).

211

Figure 9: multivariate ROC curves of the 3 markers with the PSA. a, b, c) the curves of individual metabolites combined with the PSA levels, d) the ROC curve of 3 down regulated biomarkers in PCA patients combined with PSA obtained in the validation step.

2.2.3 Relevance of our findings

In addition to the precise detection of prostate cancer and diseases, the discrimination between PCa and PCi is crucial for determining the most appropriate pharmacological or chirurgical treatment; however, as for several types of solid-cancer, the consistency of pathological reports still remains an important problem. Here, we investigated the possibility of using serum metabolomic profiles of a large cohort of PCa and PCi for correctly discriminate the two groups of patients, improving the specificity of PCa diagnosis and avoiding unnecessary biopsies. A heterogeneous panel of potential biomarkers was identified as valid diagnostic tool using a discovery approach, while the validation confirmed

212 and supported the selected biomarkers. Interestingly, the markers are primarily involved in purine and phenylalanine metabolism. 9H-Purin-6-ol (Hypoxanthine) was significantly decreased in patients affected by prostate cancer, which suggests that purine metabolism is down regulated in PCa patients. As already shown in other cancer researches, we hypothesize that the level of hypoxanthine might decrease due to consumption by tumor cells34. Likewise, a significantly decrease of Hippuric acid was observed in PCa patients compared to PCi, suggesting that the phenylalanine metabolism is perturbed in PCa subjects. It is well known that phenylalanine is an important energy metabolism precursor, thus the decreasing of serum hippuric acid may be due to the fact that cancer cells require more energy to proliferate35. Moreover, the decreasing of malonate (propanedioic acid) in prostate cancer patients compared to those with chronic inflammation, confirmed that a dysregulation of the fatty acid metabolism is occurred.

On the other hand, the validation phase confirmed that three metabolites were significantly increased in serum from PCa patients. The levels of α- D-Talopyranose, L-Proline trimethylsilyl ester and 2,5- Dimethoxymandelic acid were increased in patients with prostate cancer, which suggests that Proline metabolism is implicated in prostate cancer, as already found in previous cancer researches36,37. Moreover, an up regulation of α-D-Talopyranose, which is involved in the Mannose metabolism, was already identified for pancreatic cancer38. As regards 2,5-Dimethoxymandelic acid, although there are no previous data associated to this molecule, high levels of this potential biomarker have been linked with low renal function and selected for the prediction of

213 renal disease in diabetic patients39. This interesting data led us to hypothesize that although the symptoms of kidney disorders appear only in advanced stage of the prostate cancer, a first warning could be provided with the identification of this marker.

In conclusion, the present study showed that the combined GCxGC-MS and GC-MS analytical approach can be a considered a good strategy for the discovery and validation of new biomarkers for the diagnosis of diseases. Here, we used this scheme to identify and validate six biomarkers to discriminate prostate cancer from prostatic chronic inflammation. In fact, the widely used PSA marker is not a specific PCa biomarker, as it can be high in other circumstances (such as acute prostatitis, benign prostatic hyperplasia, after catheter manipulations, etc). Moreover, there is no universally accepted threshold value for PCA diagnosis, although normal values <4mg/L are often used.

Our biomarkers permitted to reach a very reliable result (high sensitivity and specificity), improving the specificity of PCa diagnosis and avoiding unnecessary biopsies. Finally, the pathway analysis allowed the identification of a dysregulation of key metabolic signal pathways.

214

References

1. F. Bray, J. Ferlay, I. Soerjomataram, R. L. Siegel, L. A. Torre, A. Jemal; ”Global Cancer Statistics 2018: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries”; CA: Cancer J Clin. 68:394-424. (2018);

2. H. Brenner, M. Kloor, C. P. Pox; “Colorectal cancer”; Lancet 383: 1490– 502 (2014);

3. Michael X.Chen San-YuanWang Ching-HuaKuo LinTsai, Metabolome analysis for investigating host-gut microbiota interactions, Journal of the Formosan Medical Association, Volume 118, Supplement 1, March 2019, Pages S10-S22.

4. B. Routy, V. Gopalakrishnan, R. Daillere, L. Zitvogel, J.A. Wargo, G. Kroemer, The gut microbiota influences anticancer immunosurveillance and general health, Nat Rev Clin Oncol, 15 (2018), pp. 382-396.

5. P.C. Kashyap, N. Chia, H. Nelson, E. Segal, E. Elinav, Microbiome at the frontier of personalized medicine, Mayo Clin Proc, 92 (2017), pp. 1855- 1864.

6. Louis P, Hold GL, Flint HJ. The gut microbiota, bacterial metabolites and colorectal cancer. Nat Rev Microbiol. 2014;12:661–72.

7. Hamer HM, Jonkers D, Venema K, Vanhoutvin S, Troost FJ, Brummer R- J. Review article: the role of butyrate on colonic function. Aliment Pharmacol Ther. 2008;27:104–19.

8. J. Diab, T. Hansen, R. Goll, H. Stenlund, E. Jensen, T. Moritz, J. Florholmen, G. Forsdahl; Mucosal “Metabolomic Proﬁling and Pathway

215

Analysis Reveal the Metabolic Signature of Ulcerative Colitis”; Metabolites, 9, 291(2019);

9. N. Harada Shoji, T. Soga, H. Tada, M. Miyashita, M. Harada, G. Watanabe, Y. Hamanaka, A. Sato, T. Suzuki, A. Suzuki, T. Ishida; “A Metabolic Profile of Routine Needle Biopsies Identified Tumor Type Specific Metabolic Signatures for Breast Cancer Stratification: a Pilot Study”; Metabolomics 15:147 (2019);

10. Meredith V Brown, Jonathan E McDunn, Philip R Gunst, Elizabeth M Smith, Michael V Milburn, Dean A Troyer & Kay A Lawton, Cancer detection and biopsy classification using concurrent histopathological and metabolomic analysis of core biopsies, Genome Medicine volume 4, Article number: 33 (2012).

11. Tao Huan, Dean A. Troyer, Liang Li, Metabolite Analysis and Histology on the Exact Same Tissue: Comprehensive Metabolomic Profiling and Metabolic Classification of Prostate Cancer, Scientific RepoRts | 6:32272 | DOI: 10.1038/srep32272

12. Weir TL, Manter DK, Sheflin AM, Barnett BA, Heuberger AL, et al. (2013) Stool Microbiome and Metabolome Differences between Colorectal Cancer Patients and Healthy Adults. PLoS ONE 8(8): e70803. doi:10.1371/journal.pone.0070803.

13. F. Bray, J. Ferlay, I. Soerjomataram, R. L. Siegel, L. A. Torre, A. Jemal; ”Global Cancer Statistics 2018: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries”; CA: Cancer J Clin. 68:394-424. (2018);

216

14. M. Kdadra, S. Höckner, H. Leung, W. Kremer, E. Schiffer; “Metabolomics Biomarkers of Prostate Cancer: a Systematic Review”; Diagnostics 9(1), 21 (2019);

15. N. Gómez-Cebrián, A. Rojas-Benedicto, A. Albors-Vaquer, J. A. López- Guerrero, A. Pineda-Lucena, L. Puchades-Carrasco; “Metabolomics Contributions to the Discovery of Prostate Cancer Biomarkers”; Metabolites 9(3), 48 (2019);

16. Dingxiao Zhang, Shuhong Zhao, Xinyun Li, Jason S. Kirk, and Dean G. Tang, Prostate Luminal Progenitor Cells in Development and Cancer, Trends Cancer. 2018 November; 4(11): 769–783

17. R. H. Young, J. R. Srigley, M. B. Amin, T.M. Ulbright, A. Cubilla; “Atlas of Tumor Pathology Tumors of the Prostate Gland, Seminal Vesicles, Male Urethra and Penis”; AFIP Washington D.C. (2000);

18. Thompson IM, Pauler DK, Goodman PJ, et al. Prevalence of prostate cancer among men with a prostate-specific antigen level < or =4.0 ng per milliliter. N Engl J Med 2004; 350: 2239-46

19. Wolf A, Wender R, Etzioni R, et al. American Cancer Society guideline for the early detection of prostate cancer: update 2010. CA Cancer J Clin 2010; 60: 70-98

20. Stark JR, Mucci L, Rothman KJ, et al. Screening for prostate cancer remains controversial. BMJ 2009;339:b3601]

21. R. Siegel, D. Naishadham, A. Jemal; “Cancer Statistics 2012”; CA Cancer J Clin. 62:10-29 (2012);

217

22. A. Klassen, A. T. Faccio, G. A. B. Canuto, P. L. R. da Cruz, H. C. Ribeiro, M. F. M. Tavares, A. Sussulini, “Metabolomics: Definitions and Significance in Systems Biology”; Advance in Experimental Medicine and Biology 965:3-17 (2017);

23. P. K. Cheu, M. H. Ma, H. F. Tse, K. F. Yeung, H. F. Tsang, M. K. M. Chu, C. M. Kan, W. C. S. Cho, L. B. Wah, L. W. C. Chan, S. C. C. Wong; “The applications of metabolomics in the molecular diagnostics of cancer, Expert Review of Molecular Diagnostics”; Expert Review of Molecular Diagnostics Sep;19(9):785-793 (2019);

24. J. Huang, A. M. Mondul,S. J. Weinstein, E. D. Karoly, J. N. Sampson, D. Albanes; “Prospective Serum Metabolomic Proﬁle of Prostate Cancer by Size and Extent of Primary Tumor”; Oncotarget, 8(28), 45190 (2017);

25. J. Huang, A. M. Mondul, S. J. Weinstein, S. Koutros, A. Derkach, E. Karoly, J. N. Sampson, S.C. Moore, S. I. Berndt, D. Albanes, “Serum Metabolomic Proﬁling of Prostate Cancer Risk in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial”; Br. J. Cancer, 115, 1087– 1095 (2016);

26. A. M. Mondul, S. C. Moore, S. J. Weinstein, S. Männistö, J. N. Sampson, D. Albanes; “1-Stearoylglycerol is Associated with Risk of Prostate Cancer: Results From a Serum Metabolomic Proﬁling Analysis”; Metabolomics, 10, 1036–1041 (2014);

27. A. M. Mondul, S. C. Moore, S. J. Weinstein, E. D. Karoly, J. N. Sampson, D. Albanes; ”Metabolomic Analysis of Prostate Cancer Risk in a Prospective Cohort: The Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) study” Int. J. Cancer, 137, 2124–2132 (2015)

218

28. S. de Vogel, A. Ulvik, K. Meyer, P. M. Ueland, O. Nygård, S. E. Vollset, G. S. Tell, J. F. 3rd Gregory, S. Tretli, T. Bjørge; “Sarcosine and Other Metabolites Along the Choline Oxidation Pathway in Relation to Prostate Cancer—A Large Nested Case–Control Study Within the JANUS Cohort in Norway”; Int. J. Cancer, 134, 197–206 (2014);

29. G. F. Giskeødegård, A. F. Hansen, H. Bertilsson, S. V. Gonzalez, K. A. Kristiansen, P. Bruheim, S. A. Mjøs, A. Angelsen, T. F. Bathen, M. B. Tessem; “Metabolic Markers in Blood Can Separate Prostate Cancer From Benign Prostatic Hyperplasia”; Br. J. Cancer, 113(12), 1712 (2015);

30. Roberto Puzone, Laura Paleari, Franco Montefiore, Luca Ruggiero, Matteo Puntoni, Massimo Maffezzini, Barbara Bobbio, Paola Marroni, Roberta Libener, Pier Giacomo Betta, Osteopontin plasma level does not detect prostate cancer in patients referred for diagnostic prostate biopsy, Int J Biol Markers 2010; 25 ( 4 ): 200-206.

31. Eble JN, Sauter G, Epstein JI, Sesterhenn IA, eds. World Health Organization Classification of Tumors. Pathology and genetics of tumours of the urinary system and male genital Organs. Lyon: IARC Press, 2004.

32. E. Eidelman, J. Twum-Ampofo, J. Ansari, M. Minhaj Siddiqui, The Metabolic Phenotype of Prostate Cancer, Front Oncol. 2017; 7: 131.

33. S. Srihari, P. Tattam, R. Simpson, E. Smith, Metabolic deregulation in prostate cancer, Molecular & Cellular Proteomics, 2015, 15 (1) 154-163.

34. Byong Chul Yoo, Sun-Young Kong,Sang-Geun Jang, Kyung-Hee Kim, Sun-A Ahn, Weon-Seo Park, Sohee Park, Tak Yun, and Hyeon-Seok,

219

Identification of hypoxanthine as a urine marker for non-Hodgkin lymphoma by low-mass-ion profiling, BMC Cancer. 2010; 10: 55.

35. Tan, G., Wang, H., Yuan, J. et al. Three serum metabolite signatures for diagnosing low-grade and high-grade bladder cancer. Sci Rep 7, 46176 (2017).

36. Su Liang, Beatriz Sanchez‐Espiridion, Huan Xie, Jing Ma, Xifeng Wu, Dong Liang, Determination of proline in human serum by a robust LC‐ MS/MS method: application to identification of human metabolites as candidate biomarkers for esophageal cancer early detection and risk stratification, Biomed Chromatogr. 2015 Apr; 29(4): 570–577.

37. Lima A.R., Bastos M. de L., Carvalho M., Guedes de Pinho P., Biomarker Discovery in Human Prostate Cancer: an Update in Metabolomics Studies, Translational Oncology, Volume 9, Issue 4, August 2016, Pages 357-370.

38. Guoxiang Xie, Lingeng Lu, Yunping Qiu, Quanxing Ni, Wei Zhang, Yu- Tang Gao, Harvey A. Risch, Herbert Yu, and Wei Jia, Plasma Metabolite Biomarkers for the Detection of Pancreatic Cancer, Journal of Proteome Research 2015 14 (2), 1195-1202.

39. Ng, D.P.K., Salim, A., Liu, Y. et al. A metabolomic study of low estimated GFR in non-proteinuric type 2 diabetes mellitus. Diabetologia 55, 499–508 (2012).

220

List of publications (a.a. 2016 /17-2018/19):

• Manfredi M, Barberis E, Marengo E, Prediction and classification of the degradation state of plastic materials used in modern and contemporary art, Appl. Phys. A (2017) 123:35. • Manfredi M, Barberis E, Aceto M, Marengo E, Non-invasive characterization of colorants by portable diffuse reflectance infrared Fourier transform (DRIFT) spectroscopy and chemometrics, Spectrochim. Acta A Mol. Biomol. Spectrosc. 181 (2017) 171–179. • Manfredi M, Barberis E, Gosetti F, Conte E, Gatti G, Mattu C, Robotti E, Zilberstein G, Koman I, Zilberstein S, Marengo E, Righetti PG, Method for Noninvasive Analysis of Proteins and Small Molecules from Ancient Objects, Anal. Chem. 2017, 89, 6, 3310–3317. • Barberis E, Baiocco S, Conte E, Gosetti F, Rava A, Zilberstein G, Righetti PG, Marengo E, Manfredi M. Towards the non-invasive proteomic analysis of cultural heritage objects. Microchemical Journal Volume 139, June 2018, Pages 450-457. • Manfredi M, Brandi J, Di Carlo C, Vanella VV, Barberis E, Marengo E, Patrone M, Cecconi D, Mining Cancer Biology Through Bioinformatic Analysis of Proteomic Data, Expert Rev Proteomics, 2019 Sep,16(9):733-747. • Manfredi M, Conte E, Barberis E, Buzzi, A, Robotti E, Caneparo V, Cecconi D, Brandi J, Vanni E, Finocchiaro M, Astegiano M, Gariglio, M, Marengo E, De Andrea M, Integrated serum proteins and fatty acids analysis for putative biomarker discovery in inflammatory

221

bowel disease, Integrated serum proteins and fatty acids analysis for putative biomarker discovery in inflammatory bowel disease, Journal of Proteomics, Volume 195, 20 March 2019, Pages 138- 149. • Barberis E, Manfredi M, Marengo E, Zilbersteinb G, Zilbersteinb S, Kossolapov A, Righetti PG, Leonardo's Donna Nuda unveiled, Journal of Proteomics 207 (2019) 103450.

Patents (a.a. 2016 /17-2018/19): • Industrial patent application 102019000005208, "Monitoring method of artefacts and related monitoring system", April5, 2019

222

Acknowledgements A conclusione di questo lavoro di tesi, vorrei ringraziare sentitamente tutte le persone che hanno dato il loro contribuito in questo lungo percorso, che ha visto nascere numerose collaborazioni e grandi soddisfazioni. Ringrazio innanzitutto il Professor Emilio Marengo, mio tutor, per avermi fornito grandi opportunità in questi anni da studente e borsista. Ringrazio tutti i responsabili dei dipartimenti del mio ateneo e di altre università con cui ho avuto ed ho ancora il piacere di collaborare in vari ambiti: Prof. Marco Malagodi, Prof. Pier Giorgio Righetti, Dr.ssa Raffaella Bianucci, Prof.ssa Irma Dianzani, Dr.ssa Roberta Libener. Ringrazio inoltre tutte le istituzioni culturali che hanno accettato di prestare i loro preziosi beni a questo studio con estrema fiducia nel mio operato: il Museo Egizio e la Fondazione dei Musei Civici di Torino, il Museo Herimatge di San Pietroburgo, La Soprintendenza per i Beni Storici del Piemonte e della provincia di Parma. Ringrazio ancora i miei colleghi di ricerca, con cui ho condiviso momenti preziosi e anche indimenticabili della mia vita: Eleonora M., Eleonora C., Arianna, Beppe, Virginia, Sara, Elia. Ringrazio infine la mia famiglia, che ha sempre creduto in me, e soprattutto mi ha sostenuta e aiutata anche nei momenti più difficili, venendomi incontro in ogni situazione. Un grazie di cuore a mio marito Marcello, colonna portante della mia vita e al nostro piccolo cuoricino Achille, la gioia più grande che rende ogni giorno la mia vita migliore.

223