Novel Markers to Map and Quantify Degradation on Cellulose Nitrate-Based Heritage: at the Submicrometer Level Using Synchrotron UV-Visible Multispectral Luminescence

Cellulose nitrate (CN) is an intrinsically unstable material that puts at risk the preservation of a great variety of objects in heritage collections, also posing threats to human health. For this reason, a detailed investigation of its degradation mechanisms is necessary to develop sustainable conservation strategies. To investigate novel probes of degradation, we implemented deep UV photoluminescence micro spectral-imaging, for the rst time, to characterize a corpus of historical systems composed of cellulose nitrate. The analysis of cinematographic lms and everyday objects dated from the 19th c. / early 20th c. (Perlov's collection), as well as of photo-aged CN and celluloid references allowed the identication of novel markers that correlate with different phases of CN degradation in artworks, providing insight into the role played by plasticizers, llers, and other additives instability. By comparison with photoaged references of CN and celluloid (70% CN and 30% camphor), it was possible to correlate camphor concentration with a higher rate of degradation of the cinematographic lms. Furthermore, the present study investigates, at the sub-microscale, materials heterogeneity that correlates to the artworks' history, associating the different emission proles of zinc oxide to specic color formulations used in the late 19th and early 20th centuries.


Introduction
A pioneer plastic in peril. Cellulose nitrate is considered the rst semisynthetic polymer and was trademarked as Parkesine and, later in the USA, as Celluloid by John Wesley Hyatt 1,2 . Its exibility and dimensional stability led to its extensive use as a new photographic and cinematographic medium for lms, being present in image heritage collections in archives and museums 3,4 . It was also widely used between the 1890s and 1920s for design pieces and everyday objects, all testimonies of our material culture, Figs. 1 and 2. Celluloid also attracted artists like Naum Gabo and Antoine Pevsner to creating sculptures that are now preserved in museums 5 . This heritage may be at risk, as we do not yet know what factors trigger an irreversible degradation leading to the total loss of these precious objects. It is for this reason that cellulose nitrate is considered an intrinsically unstable material in heritage collections, posing also risks to human health. To slow down degradation, low-temperature storage is accepted by the conservation community 6 , but these solutions prohibit public access, are price-sensitive, have high energy costs and there are concerns about its effects on the physical stability and material lifetime 7,8 . Important insights on its degradation were achieved over the last decade 7-32 , but a full elucidation of its mechanisms is still required to develop sustainable conservation strategies as in the NEMOSINE project, by developing a smart modular package with the main goal of energy-saving and extent conservation time [33][34][35] . Figure 2.
Celluloid in private and public collections of plastics. During its heyday (roughly from 1880 to the 1920s) celluloid was used for an astonishing array of toiletries, novelties, and "fancy goods". It was prized particularly for the ease with which it could be made to imitate semi-precious materials like ivory, tortoiseshell, coral, and mother-of-pearl 36 . Almost every article of commerce that could be made of these materials was fabricated from celluloid, and generally (but not always) sold more cheaply than they. For the most part, celluloid was not in any way functionally inferior to the materials it replaced, and it was frequently aesthetically equivalent 36 . In some forms, its cost was so low that it could be used in large volume for ephemeral articles-often advertising novelties (such as pocket calendars, rulers, book-marks, trade cards, and the like) 1 . Because these pieces were both more durable and more handsome than the card or paper items they replaced, they were often kept and stashed away rather than disposed of. Larger objects, such as combs, brushes, and other toiletry items, were often passed down as heirlooms, much as their ivory or tortoiseshell models would have been.
In the 20th century, celluloid was almost everywhere eclipsed by newer and generally less expensive plastics. Even in photography and cinema, the material was replaced by the less ammable acetate lms.
In consumer goods, from toys to toiletries to tools, celluloid gave way to the products of chemical industry, led at rst by phenol-formaldehyde (introduced as Bakelite) and then in even greater volume by a host of synthetic polymers, mostly made from petrochemicals. Celluloid became almost quaint as its uses fell away, until by mid-century not much more than table tennis balls and guitar picks remained 36 .
But celluloid remains well represented in private and public collections of plastics, and the objects found in those collections raise a host of issues for understanding both the behavior of the material and the ways in which we use the material and the objects for historic preservation and representation.
The study here attempts to demonstrate methods for investigating the historical and physical experience of this rst plastic in cinematographic lms (dated from 1925 to 1960) and objects, Fig. 2 and Table 1. The objects are part of a collection that originated with the private efforts of Dadie Perlov, who collected a wide range of small celluloid articles, largely from the northeastern United States during the last decades of the 20th century. Items from Perlov's collection constitute the primary celluloid holding in the museums of the Smithsonian Institution, in Washington, D.C 36 . Other items have ended up in other museums, and about 304 have been donated to Portuguese institutions, and these have been the basis of the study here. Synchrotron deep UV photoluminescence micro spectral-imaging to safeguard celluloid heritage. Celluloid nitrate absorbs and can be excited in the UV ( Supplementary Fig. S1), emitting in the UV-VIS region 37 ; early degradation products that are formed in extremely low concentrations are also prone to show speci c luminescence signals, making photoluminescence techniques very adapted to identify chemical changes 38 . In this work, we propose to assess the level of degradation by the use of luminescent chemical markers that are excited in the deep UV (DUV) using two instruments operating between 200 and 800 nm. The rst endstation is dedicated to raster micro-scanning photoluminescence spectroscopy using a confocal con guration (POLYPHEME) 39,40 . The second one is a multispectral imaging system based on a full-eld con guration (TELEMOS) 39,40 . In this speci c con guration, using synchrotron beam as a source of light to generate photoluminescence imaging, the limit of detection can be approached to 100 nM, enabling to detect intermediates and products that will be invisible to most spectroscopy techniques, due to their very low concentration 38 .
In addition, the possibility to retrieve centimetric elds of view at sub-micrometer lateral resolution has shown to be extremely e cient for tackling the intrinsic heterogeneity of the historical objects 39,40 , such as those studied in this work, which can be described as intrinsically heterogeneous systems, Table 1 and Figs. 1 and 2. Cinematographic lms can be described as multilayer systems in which CN-based polymers are used as image support: pigments or llers are not expected in this type of matrix. In contrast, in the celluloid artworks from the Perlov´s collection, llers and pigments are present, admixed with celluloid. The analysis of these complex historical samples is supported by highly characterized reference materials, arti cially aged, which will be studied both using DUV micro-imaging and UV-VIS spectro uorimetry. Historical lms were also arti cially aged to simulate longer periods of natural ageing.
In this work, the level of degradation assessed by the luminescent markers is compared with the degree of substitution (DS) of CN calculated by infrared spectroscopy, since DS decreases with ageing 20,26,28,29,33,41 . This decrease in DS in CN-based polymers, is a consequence of the main degradation mechanism, which occurs through homolytic scission of the nitrate groups and the glycosidic main chain 19,26 . In a pristine matrix used as support for cinematographic lms, the DS is expected to be ca 2.26 33,42 , and its decrease is a consequence of the scission of the nitrate group and its substitution by hydroxyl groups as depicted in Fig. 3. This leads to an increase and broadening of the carbonyl band at 1740 cm − 1 alongside the increase of the OH absorption between 3700 and 3100 cm − 1 , in the infrared spectrum 33 .
The data acquired by synchrotron deep UV photoluminescence micro spectral-imaging (DUV-µPL), will be for the rst time used to identify alteration markers in historical objects made from CN. This knowledge can support the development of early warning tools to monitor CN degradation in heritage collections.

Results And Discussion
Alteration markers in reference samples. Emission and excitation spectra collected on arti cially aged reference samples show that different stages of CN degradation correlate with the detection of luminescent intermediates that display speci c spectra, Fig. 4. An excitation spectrum mimics the UV-VIS absorption spectrum of the chromophore that emits, and these were acquired using spectro uorimetry (spot analyses,10x2mm 2 ). Originally, an unreacted CN lm is characterized by a broad excitation spectrum centered at 320 nm, which during the rst 20h of irradiation shifts to shorter wavelengths (290 nm) and only then it will evolve continuously to longer wavelengths, starting at 50h of irradiation with a maximum at circa 300 nm, Fig. 4 (left). The rst shift to 290 nm can result from the degradation of chromophores formed in the dark. In the lms that have suffered extensive degradation, a strong shift to higher wavelengths is observed in the excitation spectra; for example, at 130h of irradiation, several oxidized functions are identi ed by the presence of emission bands whose maxima are at 266, 325, 366 and 400 nm Fig. 4 (right). This agrees with the yellowing observed in the cellulose nitrate lms. Based on the excitation spectra collected on the various samples, we selected the excitation wavelength of 290 nm to acquire maps of emission spectra with the DUV micro-imaging set-up of the DISCO beamline, Fig. 5. The overall signal collected at the microscale is consistent with the averaged signal collected using spectro uorimetry, Supplementary Fig. S2. With an important advantage, the emission spectra obtained having a better resolution, it was possible to discriminate two bands in aged samples, in what was seen by spectro uorimetry as an unresolved broad band, Supplementary Fig. S2. For this reason, we will only discuss the emission spectra acquired using synchrotron radiation. An important rst result was to prove that degradation is homogeneous in the reference lms, which was achieved by showing that the average spectrum obtained with POLYPHEME mapping is representative of the luminescence of each pixel, Fig. 5B. POLYPHEME shows that CN at t0 is characterized by a weak emission at 425 nm, this band increases during irradiation (I max ca. 2600 cps). Also, a second band is formed and its intensity increases over time in comparison to the rst band at 420 nm, being visible at 150h of irradiation with maximum at 510-520 nm. This is an important result that was tested to assess the conservation condition of CN by plotting this ratio I 425nm /I 510nm versus irradiation time. We observed a linear correlation between those two parameters showing that quanti cation of speci c degradation state of CN can be assessed through the ratio between these two bands (I 425nm /I 510nm ), Supplementary Fig. S3: the higher the intensity of the CN2 band at 510-520 nm over the rst band CN1 (ca. 420 nm), the higher the extent of degradation when compared with the DS values obtained by infrared spectroscopy. In future work, more irradiation times will be measured as well as longer irradiation times to assess whether different mechanisms are at play.
When exciting at 290 nm, the spectrum of celluloid is dominated by the emission of camphor (I max ca. 1900 cps). To avoid it and to collect only the CN emission, excitation was performed at 330 nm for celluloid references. As observed for CN, degraded celluloid is also characterized by a rst band at 420 nm and a second one at 510-520 nm, Fig. 5C. However, the I 425nm /I 510nm ratios indicate that comparing to CN, celluloid has degraded more rapidly: celluloid value of 3.25 at 50h of irradiation is similar to CN at 100h of irradiation, 3.07.
In summary, the studies on reference samples identi ed relevant markers to probe alteration in CN lms.
To assess the relevance of those markers in naturally aged samples, we performed similar analytical procedures in the cinematographic lms and celluloid artworks.
Cinematographic lms dated from 1925 to 1960, naturally and arti cially aged. The observation of the stratigraphy of the cinematographic lms studied shows that they are composed of the celluloid support and the image layer, which is a proteinaceous matrix with dispersed silver particles 43 . On naturally aged DIF-50500 cross-section, the spatial distribution of luminescent species collected using POLYPHEME allows to distinguish three regions by their speci c emission spectra, Fig. 6: i) in the proteinaceous material, the image layer was characterized by a band at 412 nm. This emission can be associated with collagen as it was made from animal sources, principally calfskin 44 ; ii) in the celluloid support, the emission spectra presented a band at 425 nm and a shoulder at higher wavelengths (ca. 510 nm) that compare well with the 150 h irradiated celluloid reference (Fig. 5C) and, agree with the degree of substitution of 1.71 calculated for this lm by infrared spectroscopy (indicative of a degraded CN) 33 , Table 1; iii) at the support interfaces, with air and with the image layer, the spectra presented two bands at 420 nm and 538 nm; this second band is more shifted and intense than the second band produced during degradation of celluloid references (centered at ca 510-520 nm), and can represent other types of degradation products. This emission at 538 nm can be related to a more extensive degradation or a different mechanism, Fig. 6. This degradation trend was con rmed in the cinematographic lms S4, S5, S6 and 50509, where a more intense emission in the region between 500-550 nm at the interfaces was also observed, Supplementary Fig. S4.
DIF-50500 samples arti cially aged (λ ≥ 280 nm, 40 ºC, 150h) were analyzed in situ since it was not possible to prepare a cross-section due to the lm brittleness, Fig. 2. After 150h of irradiation, the emission spectrum acquired was very similar to that of the 150h arti cial aged cellulose nitrate reference emission, Supplementary Fig. S5. This behavior comparable to cellulose nitrate is probably due to the low amount of camphor in DIF-50-500, ca. 6% w/w, estimated by infrared spectroscopy 26 , Supplementary Fig.  S6. This further strengthens the results obtained in the reference samples, proving that the concentration of camphor in uences the degradation rate of celluloid.
19th c./early 20t h c. celluloid everyday objects collection. Zinc oxide (ZnO) was the main pigment used to produce the ivory-like color and it was identi ed by its emission at 380 nm, in all the objects studied 39,40,45−47 . Zinc oxide is a semiconductor having a rst narrow UV emission band at ca. 380 nm and one or more emission bands in the visible region that tend to be broad and centered at ca. 425 nm and 510 nm 40,46 . The latter can be attributed to defects or impurities that are related to levels located within the bandgap (trapped states). The ratio between the UV and visible bands can vary enormously in historical paints and may be used to distinguish between them 40,46 , Supplementary Fig. S7.
In the case of the celluloid objects of the Perlov collection, following excitation at 290 nm, the three bands described for ZnO were observed, Fig. 7 and Supplementary Figs. S7-S13. These bands were identi ed with maxima at ca 380 nm (ZnO1), at ca. 418 nm (ZnO2) and at ca. 507 nm (ZnO3), Supplementary Fig.  S7. Every pixel of the POLYPHEME maps performed has a signi cant contribution of the 380 nm emission, even for those with higher intensity emissions above 400 nm, con rming the ubiquitous presence of this pigment in the samples. The ratios between the three bands varied from pixel to pixel, in the same sample and within samples, showing the heterogeneity of this ZnO-polymer systems, Fig. 7 and Supplementary Figs. S7-S14. However, for each of the objects studied, characteristic trends were observed, which will be discussed below.
The ZnO band at ca. 380 nm does not overlap with the CN emission, Fig. 5C. On the other hand, considering that these objects have an estimated camphor concentration between 20 − 15% w/w, Supplementary Fig. S11, the emission of this plasticizer at ca. 420 nm can contribute to an intensity increase in this region.
Overall, the four objects analyzed showed different emission pro les: in the 1901 postcard the emission at 510 nm (ZnO3) dominates over the other two bands; the 1899 calendar is characterized by a highly heterogeneous system with high-intensity emissions from the three bands; for the Holy Bible Pin and the American Flag Pin, the band at 380 nm displays the highest intensity, Fig. 7 and Supplementary Figs. S12-S14.
The American ag advertisement pin mappings were characterized by the strong emission of zinc oxide particles, Fig. 7. The heterogeneity of the system was observed on images collected with the full eld endstation TELEMOS, by designating colors to each emission bandpass lter used: blue for the 352-388 nm range to image zinc oxide (ZnO); green for the 412-438 nm range to image ZnO trapped states, cellulose nitrate degradation, camphor, or other admixed materials; and red for the 535-607 nm range to image the ZnO trapped states and cellulose nitrate degradation, Fig. 7. In future work, we will investigate the origin of these differences by studying a set of aged and unaged reference samples. In this work, we will examine in more detail the American Flag Pin to better understand the information it is possible to extract based on the luminescent properties of ZnO, CN and camphor. The results correlated with POLYPHEME mapping in which we observed, spatially, spectral variations related to the main emission features, Fig. 7C. These emission features can be represented by three average spectra (calculated from 10 selected pixels) and are characterized by 1) 380 nm emission, accounting for 67% of the mapped area, 2) emissions between 400-450 nm (10%) and 3) ZnO trapped state emission above 500nm (23%), Fig. 7D, Supplementary Figs. S8 and S9. To better understand the causes of these variations, preliminary Raman and infrared analyzes were carried out, which will be further developed in future work. The main results are summarized in Supplementary Table S1. µRaman analysis of whitish particles in the American ag pin, studied by TELEMOS and depicted with a strong blue luminescence in Fig. 7 7. Emissions between 400 and 450 can be due the ZnO defect-associated emissions, camphor, or localized cellulose nitrate degradation. Future work will investigate more in-depth the origin of these emissions and if they can be related to speci c formulations of ZnO that could be used as a signature to a technological process (for colour production).
In summary, it was possible to visualize at the macro and sub microscale, the heterogeneity of these complex mixture of ZnO with celluloid. Three spectral regions of interest were identi ed: the ZnO1 (385nm) and ZnO3 (500-550nm) emissions whose intensity variations can be in uenced by the presence of organic functionalized ZnO and other inorganic additives or impurities; and the 400-450 nm regions whose contributions can be due to ZnO crystal defects (ZnO2) or to the polymer associated emissions, either camphor or cellulose nitrate degradation. Comparing POLYPHEME results for the four objects analyzed it was possible to observe differences in their spectral signatures and spatial distribution at the submicrometric scale, Fig. 7 and Supplementary Fig. S12-S14. The different emission features of these two objects can be due to formulation differences, shown by Raman analysis, which identi ed massicot/litharge (PbO) and zinc stearate for the 1901 postcard and anatase (TiO 2 ) and azurite for the holy bible pin, Supplementary table S1. These results will be further explored in future work, but already show the potentiality of this technique to distinguish color formulations that can also be linked with the three different ZnO manufacturing methods employed in the late 19th early 20th centuries 40,50 .

Conclusions
The present study here describes the implementation of DUV photoluminescence micro spectral-imaging for investigating, at the sub-microscale, cellulose nitrate-based objects' material heterogeneity that correlates to its history (of alteration and manufacture). The investigation of cellulose nitrate and celluloid luminescence properties using DUV excitation has revealed the presence of markers that can be used for the detection of the degradation process at the early stage. By using excitation between 250 and 300nm, DUV-µPL has allowed visualizing the distribution of degradation products in cinematographic lms, with very high resolution, providing new insights into their degradation mechanisms. During degradation, the emission spectra of cellulose nitrate in the UV-VIS modi es markedly; the initial very low emission at 420 nm increases and a new band at 510-520 nm appears. The higher the intensity of the band at 510-520 nm over the rst band (ca. 420 nm), the higher the extent of degradation. Importantly, we proved that the evolution of the ratio between the two bands (I 425nm /I 510nm ) correlates with the decrease in DS calculated by infrared spectroscopy, showing that it is possible to monitor the level of degradation of CN by DUV-µPL. This tool allowed another important discovery: the higher the camphor concentration the higher the degradation rate both in reference samples and on cinematographic lms. This crucial information will be next tested within an extended set of historical objects possibly using a chemometrics approach.
We also demonstrate the asset of these techniques to characterize and map, by combining spectral and spatial information, the heterogeneous microenvironments of the zinc oxide-based pigment in celluloid heritage. Due to the intrinsically heterogeneous nature of the objects in the Perlov collection, the variations observed in the luminescence of zinc oxide could be linked to the presence of lead carbonate or withe lead (identi ed by Raman microscopy), and as such to a speci c color formulation. Overall, this study shows that DUV-µPL, due to its unique detection limit and high spatial resolution, is a fundamental technique for the study of cellulose nitrate-based heritage and plastic heritage, in general.
Finally, this work shows how the collection of celluloid objects donated by Perlov is a valuable historical source. Each object offers a biography that represents a valuable way of informing its past. The object's biography is a necessary research tool to understand and contextualize the relationships between people and the meaning, purpose, and use of an object. It is also an effective didactic tool used in museums for teaching history. To write these biographies, POLYPHEME and TELEMOS have the potential to provide important data on manufacturing particularities. Knowing more about the object's materiality is also necessary to better preserve it in a historical and material sense.

Cinematographic lms and objects from Perlov's collection and sample preparation
The cinematographic lms were provided by the Austrian Academy of Sciences and the German Film Institute, partners of the NEMOSINE project. The selected celluloid objects belong to the Perlov´s collection, a collection that consists of 300 everyday celluloid objects, donated by Amy Schenkein and Dadie Perlov, American collectors and majority donors of the Smithsonian Institute's celluloid collection.
Cross-sections of the cinematographic lms were prepared by using a Leica RM 2155 rotary microtome equipped with low pro le blades Leica DB 80 LX. To do so, a small sample was removed from the border of the cinematographic lms (about 2 x 2 mm). This fragment was taped over a small piece of polymethylmethacrylate (PMMA) to hold it steady (the glue did not enter in contact with the sampling area). The PMMA was previously cleaned with ethanol. This assembly was xed in the microtome clamping system. 15 µm slices were cut by using a clean portion of the blade for each cut and making a quick but relatively gentle motion (to get a clean cut and avoid ridges and fractures). The cuts were controlled by using a stereomicroscope. The sample cross-sections were placed over a microscope slide. This method allows the proper observation of the stratigraphic layers of the cinematographic lms without the use of embedding resins.
Cross-sections of the celluloid objects were prepared by microsampling using Ted Pella µ-tools and a Leica MZ16 stereomicroscope. Since the objects studied have thicknesses below 1mm, the micro samples were cut from the corners of the objects from one end to the other (cross-section). The crosssections were embedded in a polyester resin (Clear Casting Polyester Resin AM) so that the areas in contact with the resin corresponded to the surface of the object and the interior of the cross-section to the object's bulk. The samples were wet ground on a polishing wheel to expose the cross-section. Micro-Mesh® sheets with grit 8000 were used for dry polishing.

Reference samples preparation
Cellulose nitrate lms were obtained from a solution of 4% (w/v) in methanol, prepared at room temperature (20°C) and allowing cellulose nitrate to dissolve through the night (approx. 12 hours). This solution was homogeneously poured over a porcelain vessel using a Pasteur pipette and placed inside a desiccator with silica gel. The solution was left drying for 3 hours. After drying, a transparent cellulose nitrate lm was peeled off the vessel with Ted-Pella micro tweezers. Samples with an area of 2.5x1 cm 2 were cut with a scalpel. The thickness of these reference lms was 150 µm (approaching the thickness of the polymeric support of a cinematographic lm). Celluloid lms were obtained by adding camphor to the previous solution, in a ratio of 70/30 (cellulose nitrate/camphor) in weight. They were prepared following the same methodology used for the cellulose nitrate lms.

Arti cial Ageing
The irradiation of the cellulose nitrate and celluloid reference lms and the cinematographic lm DIF 50 500 (cut in pieces with dimensions of ca. 2x1cm 2 ) was carried out in a CO.FO.ME.GRA accelerated aging apparatus (SolarBox 3000e) equipped with a Xenon-arc light source, an outdoor lter λ ≥ 280 nm, with constant irradiation of 800 W/m 2 and a temperature of 40°C. The lms were irradiated for a maximum period of 150h (total irradiance = 365 MJ/m 2 ). To avoid sample movement inside the irradiation apparatus, the samples were placed inside "homemade" glass boxes with a quartz plate over (100mm x 100mm x 3mm thick, UQG Optics Limited).

Optical microscopy
Microsampling of the celluloid objects was performed with Ted Pella µ-tools using a Leica MZ16 stereomicroscope. This microscope has a 7.1x to 115x zoom range lens, equipped with an integrated Leica ICD digital camera and a Leica KL 1500 LCD external cold light source with two exible optic bre cables. Microphotographs of the cinematographic cross-section were acquired using a Axioplan 2ie Zeiss microscope equipped with 10x ocular lenses and a 20x Epiplan objective, an incident halogen light illuminator (tungsten light source, HAL 100) and a digital Nikon camera DXM1200F, with Nikon ACT-1 software.

Spectro uorimetry
Cellulose nitrate excitation spectra were acquired using a Jovin-Yvon/Horiba SPEX Fluorog 3-2.2 spectro uorometer with a 450W xenon lamp and a double-grating monochromator. Measurements with the spectro uorometer were performed at front-face (ff). In this technique, the incident excitation beam is focused on the front surface of the samples and the uorescence emission is acquired from the same region at an angle of 22.5°, which minimizes re ected and scattered light. To ensure that different samples were analysed in the same area, the reference lms were placed on quartz demountable cells (Lightpath Optical (UK) Ltd.) and mounted in a cell holder for short path length cells (Lightpath Optical (UK) Ltd.), with the lm surface directly facing the beam. Corrected emission and excitation spectra were collected using 3 mm entrance and exit slits, an integration time of 0.2 seconds, an increment of 2 nm.
Excitation spectra were recorded collecting the signal at 390, 450 and 480 nm. To ensure the reproducibility of the results, 2 sets of four samples of irradiated cellulose nitrate and celluloid were analyzed.

UV-Vis spectroscopy
Ultraviolet-visible absorption spectra were recorded on a Varian-Cary 100 Bio spectrophotometer, between 200 and 800 nm with air as a reference and a constant temperature of 20 ± 1°C. To ensure that different samples were analysed in the same area, the reference lms were placed on quartz demountable cells (Lightpath Optical (UK) Ltd.) and mounted in a cell holder for short path length cells (Lightpath Optical (UK) Ltd. To obtain absorbances < 2, the lms were prepared with thickness < 10µm using solutions of 1.85% (w/v) in methanol applying the methodology described in Reference samples preparation. Micro-energy dispersive X-ray uorescence X-ray uorescence of the celluloid objects was performed in situ with a Bruker ArtTAX Pro spectrometer equipped with a Molybdenum (Mo) ampule, Peltier effect cooled X ash 3001® semiconductor detector and a movable arm. The experimental parameters used were voltage of 40 kV, current of 300 µA, acquisition time of 180 s.

Micrometer
Sample thickness was measured using a TOPEX 31C629 micrometer, which has a length measurement from 0 to 25 mm and an accuracy of 10 ± 5 µm.

Synchrotron deep UV-excited photoluminescence
Samples were analyzed at the POLYPHEME and TELEMOS end-stations of the DISCO beamline (Synchrotron SOLEIL, Gif-sur-Yvette, France). POLYPHEME was used to perform hyperspectral photoluminescence micro-imaging and acquire full emission spectra with high spectral resolution; TELEMOS for full-eld luminescence microscopy 39,51 . POLYPHEME Emission spectra were recorded on an Olympus IX71 inverted microscope with lens replacement to be transparent in the deep UV range, a 40× Zeiss Ultra uar UV-transmitting immersion objective, with monochromatic excitation wavelengths of 290 and 330nm, an emission wavelength range of 320-640, using integration times between 5-20s per spectrum. We tested our experimental conditions and found that no degradation is induced in the samples, with the excitation beam; thus, we did not observe changes in the uorescence of the references after repeated irradiation at 290 nm and 330 nm in the same point of analysis. Raster scanning maps were performed with integration times of 5s (2 accumulations) and 10s (1 accumulation), in areas between 280µm 2 -1.25 mm 2 using 2 to 15 µm steps.
References and cinematographic cross-sections were analysed between two 170µm quartz coverslips.
Embedded celluloid cross-sections were placed on top of a 170µm quartz coverslip. During the experiments, the in-situ analysis of the 1899 calendar was tested with success.
Data acquired were pretreated in Labspec for the removal of cosmic spikes using a top-hat lter and map colour assignments. For the quanti cation of the emission contributions for each pixel in the American ag pin POLYPHEME mapping, Fig. 7C, a direct classical least squares (DCLS) modeling using Labspec 5 was performed. Three average spectra, each one calculated from 10 spectra characterized by 1) strong near band edge emission (blue loading), 2) ZnO crystal defect emissions between 400-450 nm (green loading), and 3) strong green band emission (red loading), were used as reference component spectra (loadings) Fig. 7D. For each pixel, the model nds a linear combination of the reference component spectra which best ts the raw data. Using three loadings (blue, green, and red) with scores (x, y and z) the sum, S, of the linear combination is represented by: S = [ x * blue] + [ y * green] + [ z * red]. The model provides the scores in percentage and data was normalized so that the combination of all scores adds to 100%. The bigger the score the bigger the similarity with the loading. An example of the model output for an emission spectrum is given in Supplementary Fig. S8. This procedure was used to identify the distribution of the reference component spectra within the spectral array to create a pro le based on each component distribution, Supplementary Fig. S9. Based on the scores obtained for each pixel (448 pixels out of a total of 560 due to resin emission, not showed in Fig. SX2) the total averages were calculated for the entire map. Blue loading accounted for 67% of the mapped area, the green loading accounted for 10%, and the red loading for 23%. The error map for this model is showed in Supplementary Fig. S15.

TELEMOS
Images acquired were collected using an Axio ObserverZ1 microscope (Carl Zeiss MicroImaging) with a 40x Zeiss Ultra uar UV-transmitting immersion objective, with a monochromatic excitation wavelength of 290nm and a dichroic mirror with a cut-off wavelength at 300 nm. Fluorescence was collected using four emission bandpass lters: 352-388 nm; 412-438 nm; 535-607 nm. Acquisition time was set at 10s for all channels. Images obtained were treated with ImageJ software: a BASIC corrected DW processing was used to remove the artefacts; the false-color images were created by associating a designated color to each channel (blue, green, and red) and manipulating the color balance. This method allowed localizing areas with the highest uorescent signals.  Cinematographic lm samples S4, S5, S6, 50509 and DIF 50 500. DIF 50 500 was arti cially aged (λ ≥ 280 nm, 40 ºC): at 50 hours of irradiation the lm was yellowed and fragile; at 100h presented cracking; and at 150h the lm lost its integrity. B) NEMOSINE modular package, designed to extend the conservation time of cellulose nitrate and acetate cultural heritage (https://nemosineproject.eu).

Figure 3
Cellulose nitrate side-chain scission starts with the homolysis of a nitrate group in C2 or C3, the release of •NO2 and the formation of an alkoxy radical, which can be converted into a hydroxyl by hydrogen abstraction (inducing a DS decrease). Hydroperoxides formed in C1, lead to main chain scission through cleavage of the glycosidic bond, with the formation of a gluconolactone, a macroradical and a hydroxyl radical. The mechanisms triggered my light produce excited states that can simulate the natural ageing.

Figure 4
Irradiation of cellulose nitrate lms followed by luminescence. Left, unreacted CN lm is characterized by a broad excitation spectrum centred at 320 nm (λem = 390 nm), and in the rst 20h of irradiation the maximum is shifted to shorter wavelengths (290 nm) and then evolve to a maximum at circa 300 nm at 50h of irradiation, which will continue to shift to longer wavelengths. Right, at 130h of irradiation several oxidized functions are observed described by maxima at 266, 325, 366 and 400 nm (λem=390, 450 and 480 nm). This agrees with the yellowing observed in the lm.

Figure 5
A) Microscope images (7.1x magni cation) of cellulose nitrate homogeneous thin lms (150μm) used as reference lms before (t0) and after 150h of irradiation. B) POLYPHEME raster scanning map of a 150h aged celluloid reference (λexc = 330nm, 20 μm step, 5s acquisition time), together with the average, representative, emission spectrum. Intensity variations of the probed region between 415 and 425nm are due to the topography of the sample. C) Normalized emission spectra (λexc = 290nm) of arti cially aged cellulose nitrate (top) and celluloid (bottom) irradiated during 50h, 100h and 150h.